Browse the corpus

introduction Colon cancer affects as many as 1.2 million people worldwide and more than 100 000 people in Japan every year [1]. Surgery is the mainstay of treatment of colon cancer, and postoperative adjuvant chemotherapy may be used to prevent postoperative recurrence, although not all patients benefit from this treatment. The benefit of adjuvant chemotherapy for stage II colon cancer is small [2], and routine use for all patients with stage II is not supported. For high-risk stage II colon cancer, however, adjuvant chemotherapy is considered a reasonable approach [3]. Adjuvant chemotherapy is standard treatment of stage III colon cancer and has also been recommended for the management of high-risk stage II colon cancer [3]. To date, only 5-fluorouracil (5-FU) alone and 5-FU plus oxaliplatin have been shown to be useful as postoperative adjuvant chemotherapy regimens for colon cancer [4–9].

t chemotherapy is standard treatment of stage III colon cancer and has also been recommended for the management of high-risk stage II colon cancer [3]. To date, only 5-fluorouracil (5-FU) alone and 5-FU plus oxaliplatin have been shown to be useful as postoperative adjuvant chemotherapy regimens for colon cancer [4–9]. To further improve the outcome of postoperative adjuvant chemotherapy for colon cancer, it is necessary to increase dose intensity, and extension of the treatment duration should be considered because 6 months may be insufficient. In studies comparing the usefulness of intravenous 5-FU//leucovorin (LV), oral capecitabine, oral uracil and tegafur (UFT)/LV, and oxaliplatin-based regimen as adjuvant chemotherapy, the duration of treatment was 6 months. Since the rate of recurrence reaches a peak between 1 and 2 years after surgery [10, 11], treatment duration for more than 1 year may suppress recurrences in a larger number of patients. To date, no trial has compared the effectiveness of 6-month to ≥1-year treatment with 5-FU/LV, and the optimum duration of treatment of each regimen remains unknown. Retrospective studies have suggested that long-term FU-based chemotherapy leads to favorable survival outcomes [12]. For long-term treatment, it is necessary to choose therapeutic regimens that are less likely to cause adverse effects and are superior in durability.

uration of treatment of each regimen remains unknown. Retrospective studies have suggested that long-term FU-based chemotherapy leads to favorable survival outcomes [12]. For long-term treatment, it is necessary to choose therapeutic regimens that are less likely to cause adverse effects and are superior in durability. A 5-day treatment +2-day rest regimen of UFT that lasts 12 months was shown to be useful in the NSAS-CC study [13]. Sadahiro et al. reported that a 5-day treatment +2-day rest regimen of UFT was superior with respect to patient compliance and tolerability, and the concentration of 5-FU was maintained at relatively high levels in tumors during the 2-day rest period [14]. UFT/LV is an oral therapy that was shown to be noninferior to intravenous 5-FU/LV in JCOG0205 [15]. The present phase III study compared 6-month and 18-month durations using UFT/LV to determine whether long-term adjuvant chemotherapy is beneficial for patients with high-risk stage II and stage III colon cancer.

d [14]. UFT/LV is an oral therapy that was shown to be noninferior to intravenous 5-FU/LV in JCOG0205 [15]. The present phase III study compared 6-month and 18-month durations using UFT/LV to determine whether long-term adjuvant chemotherapy is beneficial for patients with high-risk stage II and stage III colon cancer. methods patients Patients were eligible if they had undergone radical resection of colorectal cancer with extended (D2 or more) lymph-node dissection and had pathologically complete resection of stage IIB (T4, N0, M0) or stage III cancer of the colon or rectosigmoid according to the tumor–node–metastasis classification of the International Union against Cancer, sixth edition. Other eligibility criteria included age 20–75 years, an Eastern Cooperative Oncology Group performance status (PS) of 0 or 1, no previous chemotherapy or radiotherapy, could take drugs orally, adequate organ function, and ability to start postoperative adjuvant chemotherapy within 6 weeks after surgery. The study was approved by the institutional review board or ethics committee at each participating hospital and was conducted in accordance with the provisions of the Declaration of Helsinki and Ethical Guidelines for Clinical Research (overall revision dated 28 December 2004). All patients provided written informed consent.

udy was approved by the institutional review board or ethics committee at each participating hospital and was conducted in accordance with the provisions of the Declaration of Helsinki and Ethical Guidelines for Clinical Research (overall revision dated 28 December 2004). All patients provided written informed consent. randomization and masking The treatment assignments were randomized at the registration office in the Japanese Foundation for Multidisciplinary Treatment of cancer (JFMC) data center. Patients were randomly assigned (1:1) to receive UFT/LV either for 6 months (control group) or for 18 months (study group). A minimization method was used to balance assignments according to the following stratifying factors: TNM T category (T1–2, T3, T4), N category (N0, N1, N2), surgical procedure (laparoscopic surgery, open surgery), and institution. The study investigators and patients were not blinded to the treatment assignments. A data and safety monitoring board reviewed safety data during the study. The study was funded and conducted by JFMC. No commercial support was involved in the study.

A minimization method was used to balance assignments according to the following stratifying factors: TNM T category (T1–2, T3, T4), N category (N0, N1, N2), surgical procedure (laparoscopic surgery, open surgery), and institution. The study investigators and patients were not blinded to the treatment assignments. A data and safety monitoring board reviewed safety data during the study. The study was funded and conducted by JFMC. No commercial support was involved in the study. treatment The control group received UFT (300 mg/m2/day as tegafur) orally in three divided doses per day (every ∼8 h), avoiding 1 h before and after meals. LV (75 mg/day) was given orally in three divided doses per day at the same times as UFT. Drugs were administered for 28 consecutive days, followed by a 7-day rest (consecutive-day treatment), and this was defined as one course of treatment. Five courses of treatment (6 months) were administered. The study group received UFT plus LV at the same dose level as the control group. The drugs were administered orally for 5 consecutive days, followed by a 2-day rest. Five weeks of this regimen (5 days of treatment followed by a 2-day rest on Saturday and Sunday) were defined as one course of treatment, and 15 courses (18 months) were administered. After completing the scheduled treatment, patients were followed up with no further treatment until confirmation of metastasis or recurrence. The assigned treatment was started within 6 weeks after surgery. Additional details, i.e. dose modifications, have been previously reported [16].

s (18 months) were administered. After completing the scheduled treatment, patients were followed up with no further treatment until confirmation of metastasis or recurrence. The assigned treatment was started within 6 weeks after surgery. Additional details, i.e. dose modifications, have been previously reported [16]. follow-up During protocol treatment, clinical findings and laboratory data were evaluated every 2 weeks during the first two courses of treatment and then on the starting day of each subsequent course. After completion of the protocol treatment, patients were followed-up according to a predefined surveillance schedule until recurrence or death was confirmed for 5 years after surgery. Recurrence was assessed based on CT scans. These tests were carried out every 4 months during the first 2 years after surgery and once every 6 months from the third year onward. statistical analysis The primary end point was disease-free survival (DFS) and secondary end points were overall survival (OS) and safety. In this study, to achieve a power of 80% and an α of 0.05 (two-sided) by the Schoenfeld and Richter method, we required 398 patients for each group, 796 in total, with 2 years of accrual and 5 years of follow-up, assuming a 5-year DFS rate of 75% in the control group and hazard ratio (HR) of 0.667 for the investigational group over the control group. Therefore, the target sample size was set at 840 patients, assuming that the percentage of ineligible patients would be ∼5%.

otal, with 2 years of accrual and 5 years of follow-up, assuming a 5-year DFS rate of 75% in the control group and hazard ratio (HR) of 0.667 for the investigational group over the control group. Therefore, the target sample size was set at 840 patients, assuming that the percentage of ineligible patients would be ∼5%. DFS was defined as the period starting from the day of enrollment and ending on the day of recurrence; the day a cancerous, non-recurrent lesion (either synchronous or metachronous) was detected for the first time after the day of enrollment; or the day of death from any cause, whichever was earlier. The superiority in DFS was verified in all eligible patients using a log-rank test stratified by assignment factors (excluding study site; two-sided, significance level of 5%). For time-to-event analyses, the Kaplan–Meier method was used to estimate the survival rate for each group at each time point, and Greenwood's formula was used to calculate the confidence interval (CI). In addition, the Cox proportional hazards model was used to estimate the HR between treatment groups. OS was defined as the period from the day of enrollment to the day of death from any cause. OS was analyzed in the same manner as for DFS. Safety was evaluated using Common Terminology Criteria for Adverse Events v3.0, by compiling adverse events in the treatment group and comparing standard and investigational treatment groups in terms of the incidence of adverse events. Data were analyzed using SAS version 9.2 software (SAS Institute, Cary, NC, USA).

S. Safety was evaluated using Common Terminology Criteria for Adverse Events v3.0, by compiling adverse events in the treatment group and comparing standard and investigational treatment groups in terms of the incidence of adverse events. Data were analyzed using SAS version 9.2 software (SAS Institute, Cary, NC, USA). This trial is registered with UMIN-CTR [http://www.umin.ac.jp/ctr/] (C000000245). results study population A total of 1071 patients were enrolled at 233 hospitals in Japan between October 2005 and September 2007. Although the original target sample size was 840, the recruitment was continued throughout the scheduled registration period for 2 years to implement results of deliberation by the JFMC Clinical Trial Committee and finally 1071 patients were registered for the study. After excluding eight cases for reasons shown in Figure 1, 1063 patients were included in the safety analysis set and the safety results have been reported previously [16]. A total of 1050 patients (control group, 529 patients; study group, 531 patients) were included in the efficacy analysis set after excluding 11 ineligible cases. Patient demographics were well balanced in the two groups (Table 1).Table 1. Patient characteristics Control Study Total

After excluding eight cases for reasons shown in Figure 1, 1063 patients were included in the safety analysis set and the safety results have been reported previously [16]. A total of 1050 patients (control group, 529 patients; study group, 531 patients) were included in the efficacy analysis set after excluding 11 ineligible cases. Patient demographics were well balanced in the two groups (Table 1).Table 1. Patient characteristics Control Study Total N = 534 % N = 537 % N = 1071 % Sex Male 294 55.1 264 49.2 558 52.1 Female 240 44.9 273 50.8 513 47.9 Age (years) ≤50 51 9.6 51 9.5 102 9.5 51–60 140 26.2 154 28.7 294 27.5 61–70 231 43.3 228 42.5 459 42.9 71–80 112 21 104 19.4 216 20.2 Median 64 (23–75) 64 (24–75) 64 (23–75) PS 0 503 94.2 517 96.3 1020 95.2 1 31 5.8 20 3.7 51 4.8 Tumor location Right colon (C, A, T) 199 37.3 218 40.6 417 39 Left colon (D, S) 221 41.4 211 39.3 432 40.3 Rs 114 21.3 108 20.1 222 20.7 Operative procedure Laparoscopic 109 20.4 110 20.5 219 20.4 Laparotomy 425 79.6 427 79.5 852 79.6 Histological type Well 187 35 190 35.4 377 35.2 Mod 308 57.7 307 57.2 615 57.4 Poor 19 3.6 20 3.7 39 3.6 Muc 20 3.7 18 3.4 38 3.5 Sig 0 0 2 0.4 2 0.2 T (TNM 6th) T1 16 3 16 3 32 3 T2 51 9.6 45 8.4 96 9 T3 283 53 272 50.7 555 51.8 T4 184 34.5 204 38 388 36.2 N (TNM 6th) N0 69 12.9 75 14 144 13.4 N1 347 65 352 65.5 699 65.3 N2 118 22.1 110 20.5 228 21.3 Stage (TNM 6th) I 1 0.2 0 0 1 0.1 IIA 2 0.4 1 0.2 3 0.3 IIB 66 12.4 74 13.8 140 13.1 IIIA 59 11 57 10.6 116 10.8 IIIB 288 53.9 295 54.9 583 54.4 IIIC 118 22.1 110 20.5 228 21.3 Extent of LN dissection D2 147 27.5 136 25.3 283 26.4 D3 387 72.5 391 72.8 778 72.6 No. of LN examined <12 165 30.9 151 28.1 316 29.5 ≥12 369 69.1 386 71.9 755 70.5

.2 0 0 1 0.1 IIA 2 0.4 1 0.2 3 0.3 IIB 66 12.4 74 13.8 140 13.1 IIIA 59 11 57 10.6 116 10.8 IIIB 288 53.9 295 54.9 583 54.4 IIIC 118 22.1 110 20.5 228 21.3 Extent of LN dissection D2 147 27.5 136 25.3 283 26.4 D3 387 72.5 391 72.8 778 72.6 No. of LN examined <12 165 30.9 151 28.1 316 29.5 ≥12 369 69.1 386 71.9 755 70.5 Figure 1. Consort diagram. In the control group, 12.4% of patients were stage IIB and 87.0% were stage III. In the study group, 13.8% were stage IIB and 86.0% were stage III. A total of 11 patients did not meet the eligibility criteria related to baseline characteristics. In the control group, three patients were stage I or IIA and the resection of the primary tumor was incomplete in two patients. In the study group, the resection of the primary tumor was incomplete in four patients, primary tumor was located in the middle rectum in one patient, and one patient was stage IIA.

teristics. In the control group, three patients were stage I or IIA and the resection of the primary tumor was incomplete in two patients. In the study group, the resection of the primary tumor was incomplete in four patients, primary tumor was located in the middle rectum in one patient, and one patient was stage IIA. disease-free survival DFS was analyzed based on 334 events (31.5%); i.e. 167 events (31.6%) in the control group and 167 events (31.5%) in the study group. Five-year survival was 68.8% (95% CI 64.6–72.6) in the control group and 68.9% (95% CI 64.7–72.7) in the study group. The study group did not show statistical superiority over the control group (HR = 1.00; 95% CI 0.80–1.24; stratified log-rank test, P = 0.98; Figure 2A). The first relapse was observed in 135 patients (25.5%) in the control group and 132 patients (24.9%) in the study group. In the control and study groups, the major sites of recurrence were as follows: local recurrence in 12 (8.9%) and 15 patients (11.4%), the liver in 55 (40.7%) and 45 patients (34.1%), lungs in 33 (24.4%) and 28 patients (21.2%), lymph nodes in 14 (10.4%) and 16 patients (12.1%), and the peritoneum in 9 (6.7%) and 22 patients (16.7%), respectively. Year-to-year changes in DFS in this study (supplementary Table S1, available at Annals of Oncology online) show that recurrence occurred most frequently between 1 and 2 years in both groups.Figure 2. (A) Disease-free survival: The hazard ratio in the study group when compared with the control group was 1.00 (95% confidence interval 0.80–1.24, stratified log-rank test, P = 0.98). (B) Overall survival: The hazard ratio in the study group when compared with the control group was 1.05 (95% confidence interval 0.78–1.42, stratified log-rank test, P = 0.73).

ratio in the study group when compared with the control group was 1.00 (95% confidence interval 0.80–1.24, stratified log-rank test, P = 0.98). (B) Overall survival: The hazard ratio in the study group when compared with the control group was 1.05 (95% confidence interval 0.78–1.42, stratified log-rank test, P = 0.73). overall survival OS was analyzed based on 177 events (16.7%); i.e. 86 events (16.3%) in the control group and 91 events (17.1%) in the study group. Five-year survival was 84.9% (95% CI 81.5–87.7) in the control group and 84.5% (95% CI 81.1–87.4) in the study group. The HR for the study group over the control group was 1.05 (95% CI 0.78–1.42; stratified log-rank test, P = 0.73; Figure 2B). chemotherapy and safety Six-month treatment completion was 74.0% in the control group and 75.8% in the study group, and 18-month treatment completion was 56.0% in the study group. Details of the safety analysis have been reported previously [16]. In brief, the overall incidence of adverse events was 75.3% in the control group and 77.6% in the study group. The incidence of grade 3 or higher adverse events was low in both groups. Diarrhea was the only grade 3 or higher adverse event with an incidence of >5%, and occurred in 7.2 and 2.4% in the control and study groups, respectively. There was no significant difference in safety between the groups, and treatment was tolerated in both groups.

ade 3 or higher adverse events was low in both groups. Diarrhea was the only grade 3 or higher adverse event with an incidence of >5%, and occurred in 7.2 and 2.4% in the control and study groups, respectively. There was no significant difference in safety between the groups, and treatment was tolerated in both groups. At the completion of five courses of treatment, adverse events of any grade had been reported in 69.2% of patients in the study group. During the first 6 months, the incidence of subjective adverse events was significantly lower in the study group. discussion To further improve the outcome of postoperative adjuvant chemotherapy for colon cancer, we claim that it is necessary to increase the dose intensity and consider extending treatment duration. In Japan, oxaliplatin as adjuvant therapy became available in August 2009; i.e., oxaliplatin was not available at the initiation of this study. Moreover, for long-term treatment ≥6 months, convenience of use and safety are important factors for patients, and 5-FU is preferred due to its convenience and ease of use in routine practice. Therefore, we decided to use UFT/LV, for which the efficacy and safety have been shown to be equivalent to 5-FU/LV in 6-month treatment regimens in the NSABP C06 study, as the investigational drug. None of the clinical studies conducted to date have led to the conclusion that the optimum duration of treatment with fluorinated pyrimidines as postoperative adjuvant chemotherapy for colon cancer is 6 months. Therefore, it is considered necessary extending the treatment duration to 1–2 years.

he investigational drug. None of the clinical studies conducted to date have led to the conclusion that the optimum duration of treatment with fluorinated pyrimidines as postoperative adjuvant chemotherapy for colon cancer is 6 months. Therefore, it is considered necessary extending the treatment duration to 1–2 years. In clinical trials carried out in Japan, 1 year or 2 years of postoperative adjuvant chemotherapy with UFT alone, the key drug in this study, significantly improved survival rates compared with surgery alone in patients with rectal or colorectal cancer [13]. In patients with stage I lung adenocarcinoma, 2 years of UFT monotherapy revealed a significant impact on survival [17]. Moreover, most cases of recurrent colon cancer occur in the first 2 years after surgery [10]. Although the rate of recurrence reaches a peak between 1 and 2 years after surgery [11], adjuvant chemotherapy is discontinued within 6 months in many cases, and many recurrent lesions are found after discontinuation of adjuvant chemotherapy. It is estimated that continuing adjuvant chemotherapy for at least 1 year decreases recurrent events, delaying the timing of recurrence and thereby making the cumulative recurrence rate different from that in patients receiving 6-month treatment.

current lesions are found after discontinuation of adjuvant chemotherapy. It is estimated that continuing adjuvant chemotherapy for at least 1 year decreases recurrent events, delaying the timing of recurrence and thereby making the cumulative recurrence rate different from that in patients receiving 6-month treatment. Since the 5-year DFS rate was 68.8% in the control group and 68.9% in the study group, statistical superiority was not verified. Although the recurrence and mortality rates at each time point in the study group were thought to be delayed compared with those in the control group, there were no differences in results, suggesting that treatment during the first 6 months determines whether and when recurrence occurs in the study group. Year-to-year changes in DFS in this study (supplementary Table S1, available at Annals of Oncology online) show that the rate of recurrence peaks between 1 and 2 years after surgery in both groups, irrespective of the duration of postoperative adjuvant chemotherapy.

hether and when recurrence occurs in the study group. Year-to-year changes in DFS in this study (supplementary Table S1, available at Annals of Oncology online) show that the rate of recurrence peaks between 1 and 2 years after surgery in both groups, irrespective of the duration of postoperative adjuvant chemotherapy. Currently, 5-FU (5-FU/LV, capecitabine, UFT/LV, and S-1) and oxaliplatin are the only key drugs for postoperative adjuvant chemotherapy for colon cancer. The results of this study suggest that it is difficult to achieve better treatment results even if the same treatment is continued for a long time. Currently, short-course postoperative adjuvant chemotherapy is also being tested. The International Duration Evaluation of Adjuvant Chemotherapy is evaluating whether a 3-month adjuvant therapy is noninferior for the primary parameter to the 6-month identical therapy in patients with colon cancer. If shortened therapy could provide a noninferior treatment outcome, patients would be substantially relieved of the cumulative toxicity burden of oxaliplatin (e.g. peripheral neuropathy and allergic reactions). In conclusion, this study, which compared 18 and 6 months of oral UFT/LV treatment, failed to verify the superiority of 18-month treatment over 6-month treatment in either DFS or OS. The important finding from this study is that not 18 months but 6 months of treatment is enough for postoperative UFT/LV for stage IIB/III colon cancer. funding The study was funded and conducted by JFMC. The grant number is JFMC 33-0502. No commercial support was involved in the study.

In conclusion, this study, which compared 18 and 6 months of oral UFT/LV treatment, failed to verify the superiority of 18-month treatment over 6-month treatment in either DFS or OS. The important finding from this study is that not 18 months but 6 months of treatment is enough for postoperative UFT/LV for stage IIB/III colon cancer. funding The study was funded and conducted by JFMC. The grant number is JFMC 33-0502. No commercial support was involved in the study. disclosure YK has received honoraria from Chugai Pharmaceutical and Takeda Pharmaceutical. HB has received grants from Taiho Pharmaceutical and Japanese Foundation for Multidisciplinary Treatment of Cancer. CH has received grants and honoraria from Taiho Pharmaceutical. All remaining authors have declared no conflicts of interest. Supplementary Material Supplementary Data

Highlights from ASCO 2015 demonstrate the impasse we face in solid tumour oncology: the compelling novel immune and targeted therapies are often associated with cost–benefit ratios significantly above the thresholds for reimbursement. This is at least in part a consequence of our incomplete understanding of the mechanisms of response and resistance to these agents. For example, ipilimumab is associated with durable clinical benefit in 15%–20% of unselected advanced melanoma patients (∼£75 000 per patient treated), and while the responses to single-agent targeted therapies such as vemurafenib are higher, they are often relatively short-lived (∼£42 000 per median PFS of 6–7 months). New trial design strategies such as basket and umbrella studies have improved upon patient selection, but have not yielded detailed biological understanding of the drug targets, nor polygenic mechanisms of resistance within or between patients. Academically led studies have the opportunity and the responsibility to prioritize biological insights as trial end points, maximising research gain, increasing patient benefit/safety and ultimately, improving cost-effectiveness. Collection of tumour material is fundamental to these aims but the timing, handling and sample analysis are of critical importance (Figure 1).

ss trials using a fixed-effect model. Heterogeneity was assessed using the χ2 test and the I2 statistic. A random-effects model (DerSimonian and Laird) [37] was used to assess whether the results were robust to the choice of model. Probability values were two-sided, with P < 0.05 considered of statistical significance. We also preplanned analyses to explore whether the size or the direction of the effect of metformin therapy varied according to specific study or patient characteristics, including: DM status of the comparator group (with and without non-DM patients in the comparator group), prostate cancer primary treatment type (prostatectomy or radical radiotherapy) and study design. The resulting HR estimates from study group analyses were compared using the χ2 test for interaction. We also planned to explore the impact of metformin dose/exposure on the outcomes described above where available. We also conducted unplanned sensitivity analyses for the primary outcome of RFS where at least two studies were available after restrictions. This was carried out according to study quality (restricted to studies with an NOS score ≥ the median); publication type (restricted to studies where a full publication was available); setting (restricted to hospital-based studies); follow-up (restriction of follow-up <3 years); and by the potential confounding factors accounted for (restricted to studies that adjusted for BMI, age, gender, cancer-specific prognostic factors and other DM medications). An additional unplanned exploratory analysis was also conducted according to whether the study was from a Western (North America or Europe) or non-Western population after a wide geographical distribution of studies was noted. Study group and sensitivity analyses were only conducted where study numbers were sufficient to be meaningful. Statistical analyses were carried out using STATA version 14.

e responsibility to prioritize biological insights as trial end points, maximising research gain, increasing patient benefit/safety and ultimately, improving cost-effectiveness. Collection of tumour material is fundamental to these aims but the timing, handling and sample analysis are of critical importance (Figure 1). Resistance to targeted therapies can be mediated by pre-existing rather than de novo alterations. High resolution tracking of cancer cells in vitro demonstrated that only 10% of resistant clones arise de novo [1], while mathematical models of tumour growth suggest that radiographically detectable lesions harbour at least 10 resistant sub-clones [2]. Thus, comprehensive upfront tumour profiling could anticipate the genetic composition of such clone(s), while taking into account spatial and temporal tumour heterogeneity. Extensive sampling of metastatic sites at autopsy revealed 10 distinct PTEN alterations emerging under the selective pressure of PI(3)Kα inhibition [3], and five independent reversion events in a germline BRCA2 mutant carrier who progressed on olaparib and carboplatin [4]. Distinct mechanisms of BRAF and EGFR inhibitor resistance were detected across multiple metastases within individual patients with melanoma [5] and colorectal cancer [6], respectively. Figure 1. A schematic for biological sample collection throughout the course of disease and treatment. TILs, tumour-infiltrating lymphocytes; cfDNA, cell-free tumour DNA; PBMCs, peripheral mononuclear blood cells; PK, pharmacokinetic; PD, pharmacodynamic; PDX, patient-derived xenograft.

rectal cancer [6], respectively. Figure 1. A schematic for biological sample collection throughout the course of disease and treatment. TILs, tumour-infiltrating lymphocytes; cfDNA, cell-free tumour DNA; PBMCs, peripheral mononuclear blood cells; PK, pharmacokinetic; PD, pharmacodynamic; PDX, patient-derived xenograft. The benefit of combination strategies can be limited by excess toxicity (combined targeting of the PI3K and MAPK pathways [7]), cross-resistance (BRAF and MEK inhibitors in melanoma [8]) and the persistent role of intra-tumour heterogeneity (targeting of the T790M EGFR mutation in lung cancer [9]). Informed by pre-clinical models, such as discontinuous dosing in BRAF-mutant melanoma [10], academically led trials can address more finely tuned ways of managing treatment resistance. In colorectal cancer cell-free tumour DNA (cfDNA) shows pulsatile levels of mutant KRAS in response to intermittent EGFR inhibition [11], providing the molecular rationale for re-challenge with targeted therapy. Similar frameworks are required to prospectively evaluate alternative or sequential scheduling as well as the role of cfDNA in tracking tumour progression.

A) shows pulsatile levels of mutant KRAS in response to intermittent EGFR inhibition [11], providing the molecular rationale for re-challenge with targeted therapy. Similar frameworks are required to prospectively evaluate alternative or sequential scheduling as well as the role of cfDNA in tracking tumour progression. PD-L1 expression, a putative predictive marker for PD1/PDL1 inhibition, is also spatially heterogeneous [12]. Genomic data are a promising alternative biomarker in this area [13]. Mutational data, integrated with HLA typing, and tumour and peripheral T-cell profiling can define individual neo-antigenic repertoires. Academically led studies of immunotherapeutic agents must evaluate the ability of this approach to predict responses, inform immunotherapy/targeted combinations, and ultimately, facilitate adoptive T-cell therapy. Non-genetic causes of treatment resistance have been largely overlooked but studies that incorporate longitudinal biological sample collection and novel imaging techniques are well placed to examine tumour drug exposure (including heterogeneity of drug distribution [14]) and individual variation in drug metabolising enzymes, receptors, and transporters. Patient-derived xenografts can provide a useful platform for investigating personalised therapy in co-clinical trials [15], but only if robustly characterised and used in the full knowledge of their limitations (e.g. immunosuppressed host, mouse stroma and disparities in tumour burden between mouse and patient).

rters. Patient-derived xenografts can provide a useful platform for investigating personalised therapy in co-clinical trials [15], but only if robustly characterised and used in the full knowledge of their limitations (e.g. immunosuppressed host, mouse stroma and disparities in tumour burden between mouse and patient). There clearly are challenges to implementation of such complex studies but they can be overcome through close interdisciplinary work of academic/clinical consortia as illustrated by the Lung TRACERx programme [16], the use of measures such as one-time consent [17], post-mortem studies and stakeholder engagement (patient and public). In summary, we argue for a change of emphasis in drug development from learning little from many patients towards biologically rich clinical studies focussed on gleaning the maximum amount of biological information that might inform drug response and resistance for every patient entered into academic trial protocols. disclosure The authors have declared no conflicts of interest.

ether the study was from a Western (North America or Europe) or non-Western population after a wide geographical distribution of studies was noted. Study group and sensitivity analyses were only conducted where study numbers were sufficient to be meaningful. Statistical analyses were carried out using STATA version 14. results After screening 7670 reports and conference abstracts, we identified 23 full publications and 4 conference abstracts that met our eligibility criteria, comprising 24 178 participants [38–64]. All were retrospective cohort studies except for one prospective cohort study embedded in a clinical trial [41]. The PRISMA study selection diagram is shown in Figure 1. The majority of identified studies examined the effect of metformin in one of four tumour types: prostate, colorectal, breast and urothelial cancer, which, therefore, represent the main focus of this analysis. A summary of the main characteristics for studies of breast, colorectal and prostate cancer is presented in Table 1, and a table of study characteristics for other cancer types is presented in Table 2. Figure 1. PRISMA study selection diagram. Table 1 Main study characteristics: colorectal, prostate and breast cancer

results After screening 7670 reports and conference abstracts, we identified 23 full publications and 4 conference abstracts that met our eligibility criteria, comprising 24 178 participants [38–64]. All were retrospective cohort studies except for one prospective cohort study embedded in a clinical trial [41]. The PRISMA study selection diagram is shown in Figure 1. The majority of identified studies examined the effect of metformin in one of four tumour types: prostate, colorectal, breast and urothelial cancer, which, therefore, represent the main focus of this analysis. A summary of the main characteristics for studies of breast, colorectal and prostate cancer is presented in Table 1, and a table of study characteristics for other cancer types is presented in Table 2. Figure 1. PRISMA study selection diagram. Table 1 Main study characteristics: colorectal, prostate and breast cancer Tumour group Study author Patient characteristics Study characteristics Comparator DM status Outcomes Definition of metformin exposure Median follow-up (months) Potential confounders (R = reported and not significant, M = included in multivariate model, x = not assessed, or significant but not adjusted for) NOS score Treatment Stage/other restrictions Sample size (met/total) Article type Study location Setting (H = Hospital, P = Population) DM Non-DM RFS OS CSS BMI Age Sex Cancer- specific variables Other DM meds Colorectal adenocarcinoma Spillane [38] Not specified I–III 207/315 Full Ireland P ✓ X X ✓ ✓ In year before diagnosis 46 X M M M M 7 Lee, GE [39] Not specified II–III 223/356 Abstract Singapore H ✓ X ✓ ✓ X At diagnosis 78 X M X M X 5 Lee, JH [40] Not specified IIIa 96/220 Full Korea H ✓ X X ✓ ✓ >6 m exposure 41 Mb Mb Mb Mb Mb 8 Singh [41] Not specified III /colon only 115/267 Abstract USA and Canada H ✓ X ✓ ✓ X Before randomisation Not given X M M M X 5 Zanders [42] Not specified I–III 512/778 Full The Netherlands P ✓ X X ✓ X Cumulative exposure 41 X M M M M 7 Prostate adenocarcinoma Allott [43] Prostatectomy Localised 155/369 Full USA H ✓ X ✓ X ✓ At surgery 59/73c M M n/a M X 8 Kaushik [44] Prostatectomy Localised 323/885 Full USA H ✓ X ✓ ✓ X In 3 months before surgery 61 M M n/a M R 7 Rieken WJU [45] Prostatectomy Localised 287/6486 Full USA and Europe H X ✓ ✓ X X At surgery 25 X M n/a M n/a 6 Spratt [46] Radical radiotherapy Localised 157/319 Full USA H ✓ X ✓ ✓ ✓ At diagnosis or after radiotherapy 104 R M n/a M R 8 Margel [47] Prostatectomy or radical radiotherapy Localiseda/ ≥66 years old Total 955 Full Canada P ✓ X X ✓ ✓ Cumulative exposure 56 X M n/a M M 8 Zannella [48] Radical radiotherapy Localised 114/504 Full Canada H ✓ ✓ ✓ X X At the time of radiotherapy 82 X R n/a M X 5 Danzig [49] Prostatectomy Localised 98/767 Full USA H ✓ X ✓ X X At surgery 27 X M n/a M X 6 Taira [50] Brachytherapy Localised 126/2298 Full USA H ✓ ✓ X ✓ X Diagnosis to 3 months after brachytherapy 100 M M n/a M X 7 Breast adenocarcinoma Oppong [51] Adjuvant chemo I–III 76/141 Full USA H ✓ X ✓ ✓ X Diagnosis to 6 months after 87 R M n/a M M 8 Bayraktar [52] Adjuvant chemo I–I

✓ X X At surgery 27 X M n/a M X 6 Taira [50] Brachytherapy Localised 126/2298 Full USA H ✓ ✓ X ✓ X Diagnosis to 3 months after brachytherapy 100 M M n/a M X 7 Breast adenocarcinoma Oppong [51] Adjuvant chemo I–III 76/141 Full USA H ✓ X ✓ ✓ X Diagnosis to 6 months after 87 R M n/a M M 8 Bayraktar [52] Adjuvant chemo I–I II/triple negative 63/130 Full USA H ✓ X ✓ ✓ X During adjuvant chemo 62 Md M n/a M R 8 Lega [53] Breast cancer surgery Infer I–III/≥66 years 868/1774 Full Canada P ✓ X X ✓ ✓ Cumulative exposure 54 X M n/a M M 6 NOS, Newcastle–Ottawa Quality Assessment Scale for Cohort Studies; BMI, body mass index; met, metformin; N/A, not applicable; RFS, recurrence-free survival; OS, overall survival; CSS, cancer-specific survival. aData from subanalysis. bMain analysis only. cMetformin/non-metformin. dAdjustment for body weight. Table 2 Main study characteristics: other cancer types

II/triple negative 63/130 Full USA H ✓ X ✓ ✓ X During adjuvant chemo 62 Md M n/a M R 8 Lega [53] Breast cancer surgery Infer I–III/≥66 years 868/1774 Full Canada P ✓ X X ✓ ✓ Cumulative exposure 54 X M n/a M M 6 NOS, Newcastle–Ottawa Quality Assessment Scale for Cohort Studies; BMI, body mass index; met, metformin; N/A, not applicable; RFS, recurrence-free survival; OS, overall survival; CSS, cancer-specific survival. aData from subanalysis. bMain analysis only. cMetformin/non-metformin. dAdjustment for body weight. Table 2 Main study characteristics: other cancer types Tumour group Study author Patient characteristics Study characteristics Comparator DM status Outcomes Definition of metformin exposure Median follow-up (months) Potential confounders (R = reported & not significant, M = included in multivariate model X = not assessed, or significant but not adjusted for) NOS Score Treatment Stage/other restriction Sample size (met/total) Article type Study location Setting (H = hospital, P = population) DM Non-DM RFS OS CSS BMI Age Sex Cancer specific Other DM meds Urothelial carcinoma Rieken BJU [54] TURBT pTa–pT1 N0 M0 /urothelial carcinoma of bladder (NMI) 43/1035 Full USA and Europe H X ✓ ✓ ✓ X At surgery 64 X M R M n/a 8 Rieken UO [55] Radical surgery M0 /invasive urothelial carcinoma of bladder 80/1382 Full USA and Europe H X ✓ ✓ ✓ ✓ At diagnosis 34 M M M M n/a 8 Rieken EJS [56] Radical surgery M0/upper tract urothelial carcinoma 194/2330 Full USA, Europe and Japan H X ✓ ✓ ✓ ✓ At surgery 36 X M M M n/a 6 Head and neck (squamous cell carcinoma) Kwon [57] Curative surgery or radiotherapy No distant metastases 99/1072 Full Korea H X ✓ ✓ ✓ ✓ Ever exposure 65 M M R M n/a 8 Thompson [58] Not specified Disease-free at 3 months/oral-oropharynx 33/78 Full USA H ✓ X ✓ X X Diagnosis to relapse 44 X R R R X 5 Renal cell carcinoma Hakimi [59] Partial/radical nephrectomy T2–T3 N0 M0 55/784 Full USA H ✓ ✓ ✓ X ✓ At surgery 41 M M R M X 6 Psutka [60] Partial/radical nephrectomy Localised 83/200 Full USA H ✓ X ✓ ✓ ✓ In 90 days before surgery 97 R M R M X 8 Pancreatic adenocarcinoma Ambe [61] Radical surgery Resectable 19/44 Abstract USA H ✓ X X ✓ X At surgery Not given R R R R X 7 Non-small-cell lung carcinoma Fortune-Greeley [62] Not specified data on stage I–II Not given Abstract USA H ✓ X X ✓ X Not given Not given M M X M X 6 Endometrial cancer Ko [63] Not specified I–IV (RFS data extracted) 200/363 Full USA H ✓ X ✓ X X At diagnosis 33 R M n/a M R 8 Gastric cancer Lee, CK [64] Gastrectomy I–III 132/326 Full Korea H ✓ X ✓ ✓ ✓ Cumulative exposure 74 M M M M M 9 NOS, Newcastle–Ottawa Quality Assessment Scale for Cohort Studies; BMI, body mass index; met, metformin; N/A, not applicable; NMI, non-muscle invasive; TURBT, transurethral resectio

diagnosis 33 R M n/a M R 8 Gastric cancer Lee, CK [64] Gastrectomy I–III 132/326 Full Korea H ✓ X ✓ ✓ ✓ Cumulative exposure 74 M M M M M 9 NOS, Newcastle–Ottawa Quality Assessment Scale for Cohort Studies; BMI, body mass index; met, metformin; N/A, not applicable; NMI, non-muscle invasive; TURBT, transurethral resectio n of bladder tumour; RFS, recurrence-free survival; OS, overall survival; CSS, cancer-specific survival.

introduction Although cancer survival rates in the UK have doubled in the last 40 years, half of those diagnosed with cancer still die from their disease within 10 years [1, 2]. Adjuvant treatment after potentially curative cancer therapy improves survival rates, but relapse rates remain high in some tumour types, and for others, there are no proven adjuvant treatments. In the quest to improve cancer outcomes, a number of established medications with known anti-cancer properties have been considered as adjuvant anti-cancer therapies. Examples include aspirin [3], vitamin D [4], bisphosphonates [5], statins [6] and metformin. Metformin exhibits a number of attributes that make it appealing for repurposing as an anti-cancer therapy. It has been in use for over half a century and is the most widely prescribed anti-diabetic medication in the world [7]. Consequently, it has been administered alongside most cancer treatments without the emergence of any important interactions. Additionally, data on the toxicity profile of metformin in those without type II diabetes mellitus (DM) are already available from clinical trials investigating its role as a treatment for polycystic ovarian syndrome [8]. Metformin is also generically available worldwide at low cost.

ence of any important interactions. Additionally, data on the toxicity profile of metformin in those without type II diabetes mellitus (DM) are already available from clinical trials investigating its role as a treatment for polycystic ovarian syndrome [8]. Metformin is also generically available worldwide at low cost. Metformin has been shown to have anti-cancer activity both in vivo and in vitro [9], with the underlying mechanism subject to ongoing investigation. It has been proposed that the anti-cancer properties of metformin result from both direct effects on cancer cells, particularly through inhibition of the AMPK/mTOR pathway [10], and indirect effects on the host, by virtue of its blood glucose-lowering properties and anti-inflammatory effects [11, 12]. Both mechanisms are anticipated to be important, although their relative contribution may differ according to cancer stage. In vivo evidence has emerged from window studies showing an anti-proliferative effect in breast cancer [13, 14] and a reduction in precancerous changes in the colorectum [15]. Meta-analyses have examined the role of metformin in the primary prevention of cancer, where it was found to significantly reduce overall cancer incidence; however, findings were inconsistent when individual tumour types were considered [16–20]. Meta-analyses have also investigated the effect of metformin use across all stages of disease and have found that it reduces overall cancer mortality rates, but, again findings are conflicting for individual tumour types [21–28], suggesting analyses are best conduced for individual tumour types separately. Most recently, a randomised phase III trial of non-DM patients showed that low-dose metformin was effective in the chemoprevention of metachronous colorectal adenomas or polyps when compared with placebo [29].

individual tumour types [21–28], suggesting analyses are best conduced for individual tumour types separately. Most recently, a randomised phase III trial of non-DM patients showed that low-dose metformin was effective in the chemoprevention of metachronous colorectal adenomas or polyps when compared with placebo [29]. Benefits in the primary prevention, or advanced setting, do not necessarily translate to utility in the adjuvant setting as the mechanism of action may be different. Our objective was to conduct a systematic review and meta-analysis of randomised and non-randomised studies to investigate the effect of metformin use compared with non-use on recurrence-free survival (RFS), overall survival (OS) and cancer-specific survival (CSS) in adults who have potentially curable solid tumours. There have been a number of calls for systematic reviews and meta-analyses to be conducted as part of the scientific justification, and to inform the design, of new clinical trials [30, 31]. This is particularly relevant in the field of drug repurposing. The aim of this analysis was to advise further clinical investigation of metformin in the adjuvant setting. methods All methods for this systematic review and meta-analysis are outlined in a prospectively registered protocol available online [32] (PROSPERO identifier CRD42015020519), and reporting follows PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines.

Benefits in the primary prevention, or advanced setting, do not necessarily translate to utility in the adjuvant setting as the mechanism of action may be different. Our objective was to conduct a systematic review and meta-analysis of randomised and non-randomised studies to investigate the effect of metformin use compared with non-use on recurrence-free survival (RFS), overall survival (OS) and cancer-specific survival (CSS) in adults who have potentially curable solid tumours. There have been a number of calls for systematic reviews and meta-analyses to be conducted as part of the scientific justification, and to inform the design, of new clinical trials [30, 31]. This is particularly relevant in the field of drug repurposing. The aim of this analysis was to advise further clinical investigation of metformin in the adjuvant setting. methods All methods for this systematic review and meta-analysis are outlined in a prospectively registered protocol available online [32] (PROSPERO identifier CRD42015020519), and reporting follows PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. eligibility criteria Eligible studies include randomised, controlled trials and non-randomised studies (observational, cohort and case–control) that have investigated the use of metformin, with a comparator of no metformin, in participants over 16 years old with potentially curable solid tumours (defined as those either undergoing radical therapy with curative intent or those with an early-stage cancer where cure is normally the objective of standard treatment). Studies must have reported data on at least one of RFS, CSS or OS for individual tumour types.

ver 16 years old with potentially curable solid tumours (defined as those either undergoing radical therapy with curative intent or those with an early-stage cancer where cure is normally the objective of standard treatment). Studies must have reported data on at least one of RFS, CSS or OS for individual tumour types. search strategy Electronic searches of databases (Medline, EMBASE, Cochrane Central Register of Controlled Trials), clinical trial registries (clinicaltrials.gov, ISRCTN and EU Clinical Trials Register) and conference proceedings (American Society of Clinical Oncology, and European Society of Medical Oncology) were conducted. All sources were searched from inception until 31 May 2015 (conference abstracts 2005–2015). Bibliographies of the reports of all identified studies and review articles were hand-searched for further potentially eligible studies. Further details of the search strategy are available in supplementary data S1, available at Annals of Oncology online.

om inception until 31 May 2015 (conference abstracts 2005–2015). Bibliographies of the reports of all identified studies and review articles were hand-searched for further potentially eligible studies. Further details of the search strategy are available in supplementary data S1, available at Annals of Oncology online. study selection All retrieved studies were assessed for eligibility and, when sufficient information was not available from the title and/or abstract, the full-text publication or (for conference abstracts) the associated poster or presentation was acquired and where this was not available, we contacted the study author. For studies with multiple publications, or where there was overlap in the patients studied, the most recent publication was chosen. Any queries were checked by a second reviewer and resolved by consensus. No study was excluded for weakness of study design or quality. For the purpose of this analysis, studies presenting data separately by tumour type were treated as separate studies. Articles were grouped by cancer type according to the site of origin and histology.

checked by a second reviewer and resolved by consensus. No study was excluded for weakness of study design or quality. For the purpose of this analysis, studies presenting data separately by tumour type were treated as separate studies. Articles were grouped by cancer type according to the site of origin and histology. data items and collection Data on patient characteristics, interventions and outcomes were extracted for all studies into a predesigned table. These were cross-checked by a second independent reviewer and any disagreements were resolved by consensus. A list of data extracted is available in supplementary data S2, available at Annals of Oncology online. Studies were evaluated to determine whether they accounted for potential confounding factors [body mass index (BMI), age, gender, cancer-specific prognostic factors and the use of other anti-DM medications], either by demonstrating that there was no significant difference in their distribution between treatment groups or by inclusion in multivariable analyses. In order to minimise the potential for confounding by DM status, where the comparator included both non-DM patients and DM non-metformin users, we extracted data based on a DM non-metformin comparator in preference. Where a time-varying covariate was used to model treatment effect, the most conservative HR was selected. Where reported, the HR after adjustment for potential confounding factors was extracted in preference to an unadjusted value. Since all eligible studies were of cohort design, the Newcastle–Ottawa quality assessment scale for cohort studies (NOS) [33] was used to evaluate methodological quality.

vative HR was selected. Where reported, the HR after adjustment for potential confounding factors was extracted in preference to an unadjusted value. Since all eligible studies were of cohort design, the Newcastle–Ottawa quality assessment scale for cohort studies (NOS) [33] was used to evaluate methodological quality. statistical analysis HRs and associated statistics were either extracted directly from the study reports or estimated from the Kaplan–Meier curves or summary statistics using published methods [34–36]. Where sufficient data were available on outcomes for individual cancer types, a meta-analysis was conducted with a primary outcome of RFS and secondary outcomes of OS and CSS. HRs were combined across trials using a fixed-effect model. Heterogeneity was assessed using the χ2 test and the I2 statistic. A random-effects model (DerSimonian and Laird) [37] was used to assess whether the results were robust to the choice of model. Probability values were two-sided, with P < 0.05 considered of statistical significance.

diagnosis 33 R M n/a M R 8 Gastric cancer Lee, CK [64] Gastrectomy I–III 132/326 Full Korea H ✓ X ✓ ✓ ✓ Cumulative exposure 74 M M M M M 9 NOS, Newcastle–Ottawa Quality Assessment Scale for Cohort Studies; BMI, body mass index; met, metformin; N/A, not applicable; NMI, non-muscle invasive; TURBT, transurethral resectio n of bladder tumour; RFS, recurrence-free survival; OS, overall survival; CSS, cancer-specific survival. colorectal cancer RFS was assessed in two studies (623 patients), OS in five studies (1936 patients) and CSS in two studies (535 patients). Overall, metformin use appeared to demonstrate significant improvements in RFS (HR 0.63, CI 0.47–0.85), OS (HR 0.69, CI 0.58–0.83) and CSS (HR 0.58, CI 0.39–0.86) (Figure 2), although there was variation between the results of the individual studies for RFS (I2 = 83.1%, P = 0.015) and OS (I2 = 82.3 P < 0.001). When the random-effects model was applied, the benefits seen for both OS (HR 0.62, CI 0.40–0.97) and CSS (HR 0.58, CI 0.39–0.86) remained, but there was no longer a significant benefit of metformin on RFS (HR 0.62, CI 0.30–1.29). In an unplanned exploratory analysis that grouped studies with Western and non-Western populations separately, we found there was a significant interaction between the effect of metformin on OS and the population studied (χ2 = 14.31, P < 0.001). In studies in non-Western populations, there was a highly significant benefit of metformin on OS (HR 0.36, CI 0.25–0.53); however, there was evidence of heterogeneity (I2 = 85.8%, P = 0.013). In studies with Western populations, only a trend towards a significant effect was identified (OS HR 0.84, CI 0.68–1.03) with no clear evidence of heterogeneity (I2 = 4.6%, P = 0.350). In unplanned sensitivity analyses, there appeared to be a larger relative benefit of metformin on OS when analyses were restricted to studies that had follow-up of >3 years (HR 0.64, CI 0.52–0.78). Further details of study group and sensitivity analyses for all tumour types are available in supplementary Table S1, available at Annals of Oncology online. Figure 2. Colorectal cancer outcomes according to metformin use.

when analyses were restricted to studies that had follow-up of >3 years (HR 0.64, CI 0.52–0.78). Further details of study group and sensitivity analyses for all tumour types are available in supplementary Table S1, available at Annals of Oncology online. Figure 2. Colorectal cancer outcomes according to metformin use. prostate cancer RFS was assessed in six studies (9330 patients), OS in four studies (4457 patients) and CSS in three studies (1643 patients). Metformin use demonstrated a borderline significant improvement in RFS (HR 0.83, CI 0.69–1.00), and significant improvements in OS (HR 0.82, CI 0.73–0.93) and CSS (HR 0.58, CI 0.37–0.93) (Figure 3); however, the relationship was inconsistent across studies (RFS I2 = 64.8%, P = 0.014; OS I2 = 87.3%, P < 0.001; CSS I2 = 75.3%, P = 0.017), which was reflected when the random-effects model was applied (RFS HR 0.80, CI 0.57–1.13; OS 0.69, CI 0.44–1.10; CSS 0.64, CI 0.19–2.12). Figure 3. Prostate cancer outcomes according to metformin use.

nship was inconsistent across studies (RFS I2 = 64.8%, P = 0.014; OS I2 = 87.3%, P < 0.001; CSS I2 = 75.3%, P = 0.017), which was reflected when the random-effects model was applied (RFS HR 0.80, CI 0.57–1.13; OS 0.69, CI 0.44–1.10; CSS 0.64, CI 0.19–2.12). Figure 3. Prostate cancer outcomes according to metformin use. In a pre-specified analysis, there was significant interaction between the effect of metformin and the primary treatment type on RFS (χ2 test for interaction 9.03, P = 0.003). For patients receiving radical radiotherapy [46, 48], there was a significant benefit from metformin (HR 0.45, CI 0.29–0.70), whereas no significant benefit was seen for patients who underwent radical prostatectomy (HR 0.94, CI 0.77–1.15) (Figure 4). Only a single study was able to provide data on OS and CSS in those having radical radiotherapy; however, significant improvements were seen in both (OS 0.44, CI 0.27–0.72; CSS 0.19, CI 0.06–0.63) [46]. We found no evidence of an interaction between the effect of metformin on RFS and the presence or absence of non-DM patients in the comparator group (χ2 0.49, P = 0.48). Figure 4. Prostate cancer recurrence-free survival according to metformin use for different treatment groups. In unplanned sensitivity analyses, there appeared to be a larger relative benefit of metformin on RFS when analyses were restricted to studies that had a follow-up of >3 years (HR 0.77, CI 0.62–0.96) or considered other DM medications in their analysis (HR 0.79, CI 0.64–0.98).

In a pre-specified analysis, there was significant interaction between the effect of metformin and the primary treatment type on RFS (χ2 test for interaction 9.03, P = 0.003). For patients receiving radical radiotherapy [46, 48], there was a significant benefit from metformin (HR 0.45, CI 0.29–0.70), whereas no significant benefit was seen for patients who underwent radical prostatectomy (HR 0.94, CI 0.77–1.15) (Figure 4). Only a single study was able to provide data on OS and CSS in those having radical radiotherapy; however, significant improvements were seen in both (OS 0.44, CI 0.27–0.72; CSS 0.19, CI 0.06–0.63) [46]. We found no evidence of an interaction between the effect of metformin on RFS and the presence or absence of non-DM patients in the comparator group (χ2 0.49, P = 0.48). Figure 4. Prostate cancer recurrence-free survival according to metformin use for different treatment groups. In unplanned sensitivity analyses, there appeared to be a larger relative benefit of metformin on RFS when analyses were restricted to studies that had a follow-up of >3 years (HR 0.77, CI 0.62–0.96) or considered other DM medications in their analysis (HR 0.79, CI 0.64–0.98). breast cancer RFS was assessed in 2 studies containing 271 patients and OS in 3 studies including 2045 patients. Metformin demonstrated a trend towards improvement in RFS (HR 0.77, CI 0.49–1.22) (Figure 5); however, no effect was seen in OS (HR 0.99, CI 0.92–1.05). There was no evidence of variation between the results of the studies either for RFS (I2 = 0.0%, P = 0.74) or OS (I2 = 0.0%, P = 0.75). As CSS was only available for one study containing 1774 patients, no meta-analysis was possible for this outcome; however in this study, metformin did not appear to have an impact on CSS (HR 1.01, CI 0.86–1.19). There were insufficient study numbers for any meaningful study group or sensitivity analyses. Figure 5. Breast cancer outcomes according to metformin use.

ning 1774 patients, no meta-analysis was possible for this outcome; however in this study, metformin did not appear to have an impact on CSS (HR 1.01, CI 0.86–1.19). There were insufficient study numbers for any meaningful study group or sensitivity analyses. Figure 5. Breast cancer outcomes according to metformin use. urothelial cancer Studies included patients with upper tract urothelial carcinoma and urothelial carcinoma of the bladder. RFS was assessed in 3 studies including 4747 patients, and OS in 3 studies including 4747 patients, of which 2 also assessed CSS including 3712 patients. There was no clear evidence that metformin improved either RFS (HR 0.91, CI 0.73–1.14), OS (HR 0.94, CI 0.76–1.16) or CSS (HR 0.88, CI 0.66–1.17) (Figure 6). Although there was some evidence of inconsistency between the results of studies for both RFS (I2 = 59.0%, P = 0.087) and OS (I2–51.5%, P = 0.127), the results did not change significantly when the random-effects model was applied (RFS HR 0.84, CI 0.57–1.24; OS HR 1.00, CI 0.72–1.39; CSS HR 0.88, CI 0.66–1.17). There were insufficient study numbers for any meaningful study group or sensitivity analyses. Figure 6. Urothelial cancer outcomes according to metformin use.

results did not change significantly when the random-effects model was applied (RFS HR 0.84, CI 0.57–1.24; OS HR 1.00, CI 0.72–1.39; CSS HR 0.88, CI 0.66–1.17). There were insufficient study numbers for any meaningful study group or sensitivity analyses. Figure 6. Urothelial cancer outcomes according to metformin use. other cancer types There were insufficient studies identified to warrant meta-analyses for other cancer types, the findings of which are presented in Table 3. In head and neck cancer, a positive trend towards improved RFS and CSS was seen in one study [57], but there was no effect on OS. However, the second study identified showed a potential detriment of metformin use on RFS [58]. In renal cell carcinoma, two studies were identified, both showing a non-significant inverse relationship with metformin use and RFS, and no significant benefit in OS or CSS. Single studies were identified showing a significant improvement in OS in lung cancer, RFS and OS in endometrial cancer and RFS, OS and CSS in gastric cancer. A small single study in pancreatic cancer did not suggest any effect of metformin; however, this study had a very small sample size. Table 3 Cancer outcomes by metformin use for tumour types with limited numbers of studies

ent in OS in lung cancer, RFS and OS in endometrial cancer and RFS, OS and CSS in gastric cancer. A small single study in pancreatic cancer did not suggest any effect of metformin; however, this study had a very small sample size. Table 3 Cancer outcomes by metformin use for tumour types with limited numbers of studies Tumour group Study author Sample size Recurrence-free survival HR (95% CI) Overall survival HR (95% CI) Cancer-specific survival HR (95% CI) Head and neck Kwon [57] 1072 0.76 (0.49–1.21) 0.95 (0.59–1.50) 0.79 (0.42–1.50) Thompson [58] 78 1.26 (0.62–2.56)  —  — Renal cell carcinoma Hakimi [59] 784 1.22 (0.66–2.27)  — 0.76 (0.21–2.70) Psutka [60] 200 1.07 (0.61–1.88) 0.74 (0.48–1.15) 0.83 (0.41–1.67) Pancreas Ambe [61] 44  — 0.54 (0.16–1.68)  — Lung Fortune [62] Not given by stage  — 0.85 (0.77–0.93)  — Endometrial Ko [63] 363 0.56 (0.34–0.91) 0.43 (0.24–0..77)  — Gastric Lee, CK [64]a 326 0.86 (0.80–0.94) 0.87 (0.80–0.95) 0.87 (0.78–0.96) aHR for each 6 months of metformin use.

8) 0.74 (0.48–1.15) 0.83 (0.41–1.67) Pancreas Ambe [61] 44  — 0.54 (0.16–1.68)  — Lung Fortune [62] Not given by stage  — 0.85 (0.77–0.93)  — Endometrial Ko [63] 363 0.56 (0.34–0.91) 0.43 (0.24–0..77)  — Gastric Lee, CK [64]a 326 0.86 (0.80–0.94) 0.87 (0.80–0.95) 0.87 (0.78–0.96) aHR for each 6 months of metformin use. duration and dose The impact of different exposures to metformin on early-stage cancer outcomes is examined in some of the identified studies; however, limited data and differences in the methods used to investigate exposure preclude any study-group analyses. In colorectal cancer, Spillane et al. [38] conducted additional analyses on dose intensity and found survival benefits for high-intensity metformin users not using other diabetic therapies (CSS HR 0.44, CI 0.20–0.95; OS HR 0.41, CI 0.24–0.70), but no significant benefits were identified in other subgroups. In gastric cancer, Lee et al. [64] found that increased cumulative duration of metformin use improved cancer-specific and all-cause mortality. Single studies in colorectal [42] and prostate cancer [43] also investigated the impact of different exposures to metformin but found no significant associations. discussion Our analysis suggests that metformin could be a useful adjuvant agent, particularly in colorectal and prostate cancer. The number of studies identified for each tumour type is likely to reflect the incidence and demographics of the disease, particularly the likelihood of presentation with early-stage disease and a diagnosis of DM.

suggests that metformin could be a useful adjuvant agent, particularly in colorectal and prostate cancer. The number of studies identified for each tumour type is likely to reflect the incidence and demographics of the disease, particularly the likelihood of presentation with early-stage disease and a diagnosis of DM. The variation in the adjuvant effects of metformin according to tumour type could be explained by differences in both patient characteristics and tumour biology. The effect of metformin on AMPK signalling has been hypothesised to be a major pathway through which metformin exerts its anti-cancer effects [10]. AMPK signalling dysregulation is also associated with metabolic syndrome [65], a cluster of conditions which include raised fasting glucose, dyslipidaemia, high blood pressure and central obesity [66]. Metabolic syndrome is also known to increase the risk of developing some cancers, particularly colorectal cancer [67], where it is also associated with poorer recurrence and survival outcomes [68]. In addition, it is known to develop as a consequence of androgen deprivation therapy in men with prostate cancer [69]. Metformin may improve OS by reducing the number of cardiovascular deaths associated with metabolic syndrome; however, the improvements in RFS and CSS identified suggest a direct anti-cancer effect. In prostate cancer, our study group analysis suggests that the beneficial effects of metformin use could be limited to those undergoing radical radiotherapy. The AMPK pathway is known to play a role in regulating cellular responses to radiotherapy, [70] and studies in xenograft mice models suggest that metformin can improve tumour oxygenation and therefore radiation response [71].

beneficial effects of metformin use could be limited to those undergoing radical radiotherapy. The AMPK pathway is known to play a role in regulating cellular responses to radiotherapy, [70] and studies in xenograft mice models suggest that metformin can improve tumour oxygenation and therefore radiation response [71]. The limitations of our meta-analysis include the inherent weaknesses of observational data, particularly potential measurement errors in the exposure to metformin, and variation in the definition of metformin use, and the risk of time-related biases [72]. A high degree of variation between the results of studies was observed for a number of the outcomes investigated in most of the cancer types. Our sensitivity analyses were designed to explore possible reasons for this to inform future observational and clinical trial design; however, only a small number of analyses were possible due to insufficient study numbers. For both prostate and colorectal cancer, the relative effect size appeared to increase for studies with follow-up of 3 years or greater, highlighting the importance of ensuring adequate duration of follow-up in future studies. Similarities have been seen in studies of aspirin, where greater benefits have been seen with longer follow-up [73–75]. A limited number of studies investigated the relation with frequency, dose and duration of metformin in early-stage cancer; however, findings are inconsistent and further research is required to better understand this relationship.

studies of aspirin, where greater benefits have been seen with longer follow-up [73–75]. A limited number of studies investigated the relation with frequency, dose and duration of metformin in early-stage cancer; however, findings are inconsistent and further research is required to better understand this relationship. Previous studies have suggested that a diagnosis of DM has a negative impact on cancer outcomes [76, 77]; therefore, inclusion of non-DM patients in comparator groups could underestimate the beneficial effect of metformin. Owing to insufficient study numbers, it was only possible to analyse the effect of the presence or absence of non-DM patients in the comparator group for RFS in prostate cancer, where no evidence for an effect was found.

ion of non-DM patients in comparator groups could underestimate the beneficial effect of metformin. Owing to insufficient study numbers, it was only possible to analyse the effect of the presence or absence of non-DM patients in the comparator group for RFS in prostate cancer, where no evidence for an effect was found. Other meta-analyses have investigated the effect of metformin on survival outcomes, across all stages of cancer, in individual tumour types, the findings of which are presented in supplemen-tary Table S2, available at Annals of Oncology online. In colorectal cancer, four meta-analyses have examined the effect on OS [21–24], two of which also investigated colorectal CSS [23, 24]. All meta-analyses identified significant improvements in these end points, which is consistent with the findings of this study. For prostate cancer, findings are less consistent. Five meta-analyses have examined the effect of metformin on OS [22, 23, 25–27], two of which also investigated prostate CSS [25, 26]. Only two meta-analyses identified a significant benefit in OS [23, 25], with no benefit identified in prostate CSS. This differs from the findings of this study where significant benefits in OS and prostate CSS were identified, which could suggest that metformin is better suited to the adjuvant setting for prostate cancer. In breast cancer, four meta-analyses examined OS [21–23, 28], two of which investigated breast CSS. Two meta-analyses identified a significant benefit in OS [21, 23, 28], the other approached significance (HR 0.81, CI 0..64–1.04) [22] and the two meta-analyses investigating breast CSS also showed significant improvements [23, 28]. This differs from the findings of this study where no significant benefit in OS and breast CSS was identified. This could suggest that metformin may be effective in those with established breast cancer, which is consistent with the findings of breast cancer window studies where direct anti-tumour effects have been identified [13, 14].

m the findings of this study where no significant benefit in OS and breast CSS was identified. This could suggest that metformin may be effective in those with established breast cancer, which is consistent with the findings of breast cancer window studies where direct anti-tumour effects have been identified [13, 14]. Investigation of metformin in the primary prevention setting presents a number of challenges, where the balance between adverse effects and benefits is likely to be less favourable and difficult to detect in a clinical trial because of the low event rate. While the advanced setting can provide a sufficient event rate, there is evidence to suggest that metformin requires long-term use to exert its anti-cancer effect [78], and therefore, patients with established cancer with more limited prognoses may not be able to receive metformin long enough for a therapeutic benefit to emerge. Therefore, the adjuvant setting could be most suitable for investigating the anti-cancer effects of metformin.

-term use to exert its anti-cancer effect [78], and therefore, patients with established cancer with more limited prognoses may not be able to receive metformin long enough for a therapeutic benefit to emerge. Therefore, the adjuvant setting could be most suitable for investigating the anti-cancer effects of metformin. current trial activity In colorectal cancer, a phase III trial of metformin versus standard care assessing recurrence and survival in stage III disease is now in set-up phase in South Korea (NCT02614339). In prostate cancer, the Metformin Active Surveillance Trial (NCT01864096), an ongoing randomised phase III trial of metformin versus placebo given before primary therapy is assessing time to progression in men with low-risk prostate cancer. The STAMPEDE trial (NCT00268476), a multi-arm multi-stage randomised, controlled trial investigating a number of agents in the treatment of hormone-naïve, high-risk, localised and metastatic prostate cancer, aims to evaluate whether the addition of metformin improves survival in this group. Recruitment to this comparison is due to open in autumn 2016. In breast cancer, our results did not identify any meaningful benefit of metformin use in the adjuvant setting; however, this could be due to the limited number of studies identified. Additional supporting data are available in the primary prevention and treatment setting (across all stages), where meta-analyses have shown a beneficial effect [21, 23, 28, 79]. A randomised phase III trial of metformin versus placebo assessing recurrence and survival in early-stage breast cancer has recently completed recruitment (MA-32, NCT01101438) and the results are awaited.

ion and treatment setting (across all stages), where meta-analyses have shown a beneficial effect [21, 23, 28, 79]. A randomised phase III trial of metformin versus placebo assessing recurrence and survival in early-stage breast cancer has recently completed recruitment (MA-32, NCT01101438) and the results are awaited. conclusions The findings of this meta-analysis support the concept of randomised clinical trials using metformin in the adjuvant setting, with the strongest supporting evidence in colorectal and prostate cancer, particularly those treated with radical radiotherapy. Such trials could also further our understanding of the relationships between cancer outcomes and the dose and duration of metformin. The authors are not aware of any ongoing adjuvant phase III trials of metformin in prostate cancer, or colorectal cancer in Western populations. In other tumour types, where there is currently less evidence, further observational studies are needed to advise suitability for investigation in any future randomised, controlled trials. Supplementary Material Supplementary Data acknowledgements We thank Larysa Rydzewska for assisting with the search strategy design. funding This work was supported by the UK Medical Research Council. No grant numbers apply. disclosure The authors have declared no conflicts of interest.

introduction Although microscopic examination of formalin-fixed paraffin-embedded (FFPE) material remains crucial in cancer diagnosis, next-generation sequencing (NGS) of tumour DNA has emerged as a powerful diagnostic tool [1] and is a central component of personalised medicine initiatives. NGS relies heavily on high-quality DNA, and snap-frozen (SF) samples are preferred because formalin fixation induces chemical modifications and degradation of DNA [2, 3]. Comprehensive diagnostic strategies and translational research protocols therefore currently demand two samples, one SF for molecular analysis and the other FFPE for routine haematoxylin and eosin staining (H&E) and immunohistochemistry (IHC). Processing of SF samples for NGS has several disadvantages, including reduced ability to microdissect tumour material and significantly increased costs [4, 5]. In particular, there are significant barriers to obtaining SF material in large-scale clinical trials, where samples are typically collected from multiple hospitals in different countries. Therefore, alternatives to formalin-based fixation are required to circumvent the need for fresh-frozen sampling.

sed costs [4, 5]. In particular, there are significant barriers to obtaining SF material in large-scale clinical trials, where samples are typically collected from multiple hospitals in different countries. Therefore, alternatives to formalin-based fixation are required to circumvent the need for fresh-frozen sampling. Methanol-based fixation has emerged as a promising such alternative [5–7] (supplementary Table S5, available at Annals of Oncology online). Universal molecular fixative (UMFIX) has been shown to be superior for IHC to neutral-buffered formalin (NBF), and gives higher yield and molecular weight of extracted DNA and RNA [5, 6, 8]. In addition, prolonged exposure to methanol fixatives may have fewer deleterious effects on DNA/RNA quantity and quality than NBF [3, 5]. However, potential NGS sequencing artefacts from methanol fixation have not been studied. Here, we have tested the suitability of DNA extracted after methanol-based fixation for NGS assays compared with DNA from matched NBF and fresh-frozen tissues. We studied high-grade serous ovarian cancer (HGSOC) samples because they have ubiquitous TP53 mutation and TP53 sequences have been extensively studied for fixation artefacts [9, 10]. HGSOC also has marked genomic rearrangement and copy-number abnormalities (CNAs), which allow stringent inspection of the effects of DNA fragment length size on CNA profiling.

ncer (HGSOC) samples because they have ubiquitous TP53 mutation and TP53 sequences have been extensively studied for fixation artefacts [9, 10]. HGSOC also has marked genomic rearrangement and copy-number abnormalities (CNAs), which allow stringent inspection of the effects of DNA fragment length size on CNA profiling. patients and methods sample acquisition and processing Three equal fragments were macrodissected from tumour specimens removed from 16 patients, median age 62, with HGSOC undergoing debulking surgery. In addition, mock biopsies of the tumour were taken from 12 cases with a 16G core biopsy gun. All samples were reviewed by at least two pathologists and fixed in 10% NBF (Genta Medical, York, UK)), UMFIX (Sakura Finetek, Thatcham, UK) or SF (liquid nitrogen). Matched normal tissue controls were processed in parallel. Full clinical details are given in supplementary Table S1, available at Annals of Oncology online. immunohistochemistry 5 µm sections of NBF and UMFIX fixed material were stained for CK7, p53, PAX8, WT1 and CK20 using established clinical protocols in the Department of Pathology, Queen Elizabeth University Hospital, Glasgow, with additional optimization for WT1 staining of UMFIX tissues. Staining and image analysis protocols, as well as all histoscore data, are described in supplementary material, available at Annals of Oncology online.

blished clinical protocols in the Department of Pathology, Queen Elizabeth University Hospital, Glasgow, with additional optimization for WT1 staining of UMFIX tissues. Staining and image analysis protocols, as well as all histoscore data, are described in supplementary material, available at Annals of Oncology online. DNA extraction and quantification DNA was extracted using QIAmp DNA Micro and AllPrep DNA/RNA Micro Kit for UMFIX/NBF-fixed and SF tumours, respectively. DNA size distribution and quality were assessed by qPCR with Illumina FFPE QC Kit and KAPA hgDNA Quantification and QC Kit, respectively. tagged-amplicon sequencing (TAm-seq) The coding regions of TP53, PTEN, EGFR, PIK3CA, KRAS and BRAF were sequenced by TAm-Seq as described previously [11] on an Illumina MiSeq using PE-125 bp protocols. Data analysis is described in supplementary material, available at Annals of Oncology online. shallow whole-genome sequencing (sWGS) WGS libraries were prepared from 100 ng DNA using modified TruSeq Nano DNA LT Sample Prep Kit protocol. Library quality and quantity were assessed with DNA-7500 kit on 2100 Bioanalyzer and with Kapa Library Quantification kit according to the original protocols, respectively. Eighteen barcoded libraries were pooled together in equimolar amounts and each pool was sequenced on HiSeq2500 in SE-50 bp mode. Analysis methods are described in supplementary material, available at Annals of Oncology online.

r and with Kapa Library Quantification kit according to the original protocols, respectively. Eighteen barcoded libraries were pooled together in equimolar amounts and each pool was sequenced on HiSeq2500 in SE-50 bp mode. Analysis methods are described in supplementary material, available at Annals of Oncology online. mutation signature analysis Non-negative matrix factorisation was carried out to identify mutation signatures [12] in relation to different fixation (supplementary material, available at Annals of Oncology online). All non-reference base changes observed across the sequencing data were interrogated from both TAm-Seq and sWGS data. results Figure 1 summarises the study design and the flow of samples through the study. Additional REMARK data are provided in supplementary material, available at Annals of Oncology online.Figure 1. Study design. (A) Operative specimens from women undergoing surgery for HGSOC were sampled with a scalpel to acquire three surgical tumour samples and a 16G needle was used to obtain three mock biopsies. Matched surgical and biopsy samples from each case, with matched control tissue, were processed in parallel with fixation in NBF or UMFIX, or SF before downstream analysis. (B) Sample workflow: numbers of patients (P) and samples (S) used for analysis. Bx, biopsy; Tu, surgical tumour fragment; Ctrl, control tissue.

es. Matched surgical and biopsy samples from each case, with matched control tissue, were processed in parallel with fixation in NBF or UMFIX, or SF before downstream analysis. (B) Sample workflow: numbers of patients (P) and samples (S) used for analysis. Bx, biopsy; Tu, surgical tumour fragment; Ctrl, control tissue. methanol fixation yields higher yield and size of DNA fragments than buffered formalin There was no significant difference in tumour cellularity and TP53 allele fraction between UMFIX and NBF samples, thus allowing a direct comparison of DNA metrics (supplementary Figure S1, available at Annals of Oncology online). Quantification of extracted DNA showed similar yields of small (90 bp) fragments from UMFIX and SF samples, both of which were significantly higher than from NBF (Figure 2A). As expected, SF samples showed the highest yields of large fragments (129 bp, 305 bp), but yields from UMFIX samples were still significantly higher than NBF (Figure 2B).Figure 2. DNA yield and copy-number calling performance. Box plots show results of PCR assays for DNA size after extraction from SF, NBF and methanol (UMFIX) fixation from matched biopsy and surgical samples from 11 HGSOC patients. (A) Observed ΔCq values for DNA yield of 90 bp fragments (negative ΔCq values are shown for convenience). *P < 0.05, **P < 0.005, ***P < 0.0005. (B) Observed Q ratios for 129 bp/41 bp (top) and 305 bp/41 bp (bottom) fragments. Vertical brackets indicate Wilcoxon rank-sum test for difference in means: *P < 0.05, **P < 0.005, ***P < 0.0005. (C) Scatter plots show correlation between median normalised copy-number profiles from shallow WGS of SF compared with NBF or UMFIX biopsy and surgical samples from 12 patients. Spearman's rank-sum correlation rho is shown. Gray background (orange online) indicates plots with the highest correlation between UMFIX and NBF for each patient sample (biopsy or surgery). (D) Boxplots show an observed variance for each copy-number segment (n = 90 312) in 69 samples from 12 patients.

from 12 patients. Spearman's rank-sum correlation rho is shown. Gray background (orange online) indicates plots with the highest correlation between UMFIX and NBF for each patient sample (biopsy or surgery). (D) Boxplots show an observed variance for each copy-number segment (n = 90 312) in 69 samples from 12 patients. copy-number calling in methanol-fixed material is superior to formalin Copy-number profiles from sWGS were compared for correlation and variance of copy-number abnormality (CNA) estimation, using SF as gold standard. UMFIX showed superior copy-number profiles compared with NBF, with 9 of 11 biopsies and 10 of 12 surgical samples showing higher correlation with the matched SF (Figure 2C). UMFIX also had lower noise for segmental copy-number estimation than NBF (Figure 2D).

ber abnormality (CNA) estimation, using SF as gold standard. UMFIX showed superior copy-number profiles compared with NBF, with 9 of 11 biopsies and 10 of 12 surgical samples showing higher correlation with the matched SF (Figure 2C). UMFIX also had lower noise for segmental copy-number estimation than NBF (Figure 2D). single-nucleotide sequencing noise from methanol-fixed material is comparable with SF and NBF We analysed low-level sequence noise using 255 376 observed non-reference bases in the sWGS and TAm-Seq data. All analysed mutations were filtered using dbSNP specifically to exclude germline SNPs. Analysis of the flanking bases around each base change revealed three mutation signatures (Figure 3A): signature 1 was dominated by non-CpG C>A transversions and C>T transitions; Signature 2 had high rates of T>A, C>A, T>C and C>T transitions, with the latter enriched in the trinucleotide context NCA (where N indicates any base); Signature 3 showed T>C and CpG-related C>T transitions. A breakdown of the contribution of each signature across four categories (base changes common to all samples, changes unique to SF, UMFIX and NBF) showed that the common changes (containing a collection of both true SNVs and typical errors) were dominated by Signature 3, whereas the other categories were a mix of Signatures 1 and 2 (Figure 3B).Figure 3. Single-nucleotide noise profiles and variant calling performance. (A) Bar plots of the three somatic mutation signatures (S1–S3) identified by non-negative matrix factorisation using all non-reference bases observed in sWGS and TAm-SEQ sequencing data in 69 samples from 12 patients (n = 255 376). Bar plots are grouped by the observed base change with individual bars showing the proportion observed at different trinucleotide sequences. (B) Stacked bar plots show the proportion of the three mutation signatures observed only in SF and NBF or UMFIX fixed samples compared with signatures present in all samples from an individual patient (common). (C) Sensitivity (top) and specificity (bottom) for manually curated SNV calls (n = 546) from TAm-SEQ of biopsy and surgical samples from 11 patients. Bars indicate the 95% confidence interval around the indicted mean.

FIX fixed samples compared with signatures present in all samples from an individual patient (common). (C) Sensitivity (top) and specificity (bottom) for manually curated SNV calls (n = 546) from TAm-SEQ of biopsy and surgical samples from 11 patients. Bars indicate the 95% confidence interval around the indicted mean. single-nucleotide variant calling from methanol-fixed material is comparable with fresh-frozen SNVs were called using TAm-Seq of 66 samples yielding 546 variants. Manual curation of these variants revealed lower average sensitivity and specificity for NBF compared with SF and UMFIX, albeit not significantly (Figure 3C).

single-nucleotide variant calling from methanol-fixed material is comparable with fresh-frozen SNVs were called using TAm-Seq of 66 samples yielding 546 variants. Manual curation of these variants revealed lower average sensitivity and specificity for NBF compared with SF and UMFIX, albeit not significantly (Figure 3C). methanol fixation permits high-quality H&E and IHC analyses Tissue morphology (H&E staining) of UMFIX samples was comparable with NBF fixation. Overall, differences between UMFIX and NBF were not diagnostically significant (Figure 4A). Statistically significant correlation was found between quantitative IHC histoscores in UMFIX and NBF-fixed samples for key HGSOC markers (p53, CK7, PAX8, WT1). CK20 was uniformly negative in all tumour samples, regardless of fixative (data not shown). There was no significant difference in median histoscore between the two sample sets for p53, CK7 and PAX8 (Figure 4B).Figure 4. H&E staining and IHC scoring. (A) H&E staining of tumour fragments (left) and biopsies (right) from matched tissues fixed in NBF and UMFIX. Bars represent 100 µm. (B) Representative images (left) show IHC staining for p53, PAX8, CK7, WT1 on matched NBF and UMFIX tissues. Quantitative histoscores (middle) of intensity of staining for each IHC marker on tumour fragments [dark gray points (pink online)], biopsies [light gray points (orange online)] and control tissue [black points (blue online)] samples. Spearman's rank-sum correlation rho is shown (P < 0.001 for all analyses). Paired data plots (right) show comparison of median histoscores from paired UMF- and NBF-fixed tissues for each IHC marker. Median scores were only significantly different for WT staining (P = 0.011).

tissue [black points (blue online)] samples. Spearman's rank-sum correlation rho is shown (P < 0.001 for all analyses). Paired data plots (right) show comparison of median histoscores from paired UMF- and NBF-fixed tissues for each IHC marker. Median scores were only significantly different for WT staining (P = 0.011). discussion The most important variables for NGS assays are DNA quality and yield. Formalin fixation can induce severe effects on the structure and integrity of DNA causing C>T, A>G, G>T, G>C and A>T base changes, methylene bridge formation, DNA denaturation and DNA fragmentation [6, 13–15]. After NGS, these chemical modifications result in greater SNV artefacts, higher sequence duplication rates, smaller insert sizes and lower fractions of mappable reads [16, 17]. We evaluated whether methanol-based fixation can reduce these detrimental artefacts when attempting to identify true somatic SNVs and accurate copy-number from clinical material. We show that UMFIX fixation yields longer amplifiable DNA fragments, in agreement with previous reports [3, 5, 8], which improves our ability to call DNA copy-number accurately. We show that SNV calling from UMFIX DNA has similar performance to DNA from SF tissues and that traditional H&E staining and IHC scoring can be carried out on UMFIX-embedded samples with minimal optimisation.

s, in agreement with previous reports [3, 5, 8], which improves our ability to call DNA copy-number accurately. We show that SNV calling from UMFIX DNA has similar performance to DNA from SF tissues and that traditional H&E staining and IHC scoring can be carried out on UMFIX-embedded samples with minimal optimisation. These findings are clinically highly important: although attempts have been made to reduce noise induced by formalin fixation (e.g. increasing targeted sequencing coverage or reducing C>T transitions with UDG treatment), these methods only mitigate some sources of noise when calling SNVs and do not improve the ability to call CNA [18, 19]. CNA detection is more challenging than SNV detection and remains the major clinical need for personalised treatment approaches in HGSOC.

cing coverage or reducing C>T transitions with UDG treatment), these methods only mitigate some sources of noise when calling SNVs and do not improve the ability to call CNA [18, 19]. CNA detection is more challenging than SNV detection and remains the major clinical need for personalised treatment approaches in HGSOC. In addition, we have used a state-of-the-art computational approach to perform in-depth exploration of the low-level sequence noise introduced by fixation and sample processing. In an advance over previous approaches, we modelled the trinucleotide context of each base change and de-convolved distinct trinucleotide noise signatures. This computational approach has previously been used to identify signatures in collections of SNVs observed across thousands of tumours, and these signatures used to infer underlying mutational processes [12]. In our data, we identified three distinct trinucleotide signatures. Signature 3 has high similarity to a previously identified CpG-age-related cytosine deamination (C>T) signature (signature 1B [12]), and a recently uncovered sequence error signature [20]. However, signatures 1 and 2 are novel and have no similarity to previously described signatures. In particular, they show high rates of C>T transition but not in CpG dinucleotide contexts. As expected, signature 3 contributed only to the set of base changes common to all samples across a patient. In contrast, signatures 1 and 2 contributed only to the base changes exclusive to SF, UMFIX or NBF samples. This suggests that the sequencing noise represented by these two signatures (C>T not at CpG) is induced through sample processing. The two fixative conditions showed a slightly increased contribution to signature 1 compared with SF, suggesting that fixation may have a specific effect. However, larger studies are required to achieve the power to discern this. This approach to modelling sequence noise provides powerful tools to explore sequencing artefacts and an analytical framework to understand the mechanisms behind their creation. Further studies with high coverage WGS are now underway to refine these data.

arger studies are required to achieve the power to discern this. This approach to modelling sequence noise provides powerful tools to explore sequencing artefacts and an analytical framework to understand the mechanisms behind their creation. Further studies with high coverage WGS are now underway to refine these data. There are no data on the effects of long-term methanol fixation on DNA quality or quantity, and this study utilised samples collected no more than 6 months before analysis. With FFPE material, it is possible to isolate DNA from long-term archived samples [21], although factors such as duration of fixation, age of the sample, exposure to heat and light, as well as the concentration, buffering and age of the formalin, can all influence DNA quality and extent of sequence artefact [22]. Careful longitudinal analyses will be required to ascertain whether similar problems emerge in UMFIX samples. We specifically did not examine RNA in this study. There are several previous publications on the utility of RNA extracted from methanol-fixed specimens in PCR and microarray assays, including from samples stored at room temperature for up to 8 weeks [5, 23]. However, we are not aware of any study assessing RNA sequencing or RNA profiling of samples extracted from methanol-fixed tissue—again, future studies will be required to confirm whether RNA extracted from methanol can be reliably used in such assays.

g from samples stored at room temperature for up to 8 weeks [5, 23]. However, we are not aware of any study assessing RNA sequencing or RNA profiling of samples extracted from methanol-fixed tissue—again, future studies will be required to confirm whether RNA extracted from methanol can be reliably used in such assays. In summary, whilst SF samples remain the gold standard for nucleic acid extraction from tumour material at present, there are significant costs associated with such samples in clinical trials and NGS-based personalised medicine studies. A key advantage of methanol fixation is that it allows easy collection and embedding of tumour material with associated economies for pathological verification and microdissection. Based on our findings of superior DNA quality, we recommend that UMFIX be routinely adopted for collection and storage of clinical cancer specimens for large-scale genomic analysis. Supplementary Material Supplementary Data acknowledgements We acknowledge the NIHR Cambridge Biomedical Research Centre for their support of this work. The authors would also like to thank Colin Nixon (CRUK Beatson Institute, Glasgow) for assistance with IHC optimisation. authors' contributions Experimental design: I.M.c.N., J.D.B., A.P., D.E., G.M., T.G. Pathological assessment: D.M., A.H., L.M., M.J.L. Data acquisition: A.P., D.E., G.M., T.G., N.S.G., M.V. Quantitative IHC: C.O. Data analysis: G.M., A.P., M.E., D.E., T.G., I.M.c.N., J.D.B. Manuscript preparation: I.M.c.N., J.D.B., A.P., G.M., T.G., D.E. All authors reviewed the manuscript before submission.

G.M., T.G. Pathological assessment: D.M., A.H., L.M., M.J.L. Data acquisition: A.P., D.E., G.M., T.G., N.S.G., M.V. Quantitative IHC: C.O. Data analysis: G.M., A.P., M.E., D.E., T.G., I.M.c.N., J.D.B. Manuscript preparation: I.M.c.N., J.D.B., A.P., G.M., T.G., D.E. All authors reviewed the manuscript before submission. funding This work was supported by Ovarian Cancer Action [BriTROC project grant: IMcN, JDB], Cancer Research UK [grant numbers A15973, A15601, A18072]; NHS Greater Glasgow and Clyde Biorepository; the Universities of Cambridge and Glasgow; National Institute for Health Research Cambridge Biomedical Research Centre; National Cancer Research Network; Cambridge and Glasgow Experimental Cancer Medicine Centres and Hutchison Whampoa Limited. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. disclosure The authors have declared no conflicts of interest.

introduction Associated with the substantial and welcome increase in survival for the majority of cancer patients over the past 40 years [1–3] has been an increased interest in the statistical estimation of population ‘cure’ [4, 5]. ‘Cure’ in this context is the point at which a group of cancer patients is observed to have no excess mortality (due to their cancer) in comparison with the population from which they were drawn (Figure 1). At the point of ‘cure’, the group of cancer patients are no longer more likely to die than if they had never been diagnosed with cancer.Figure 1. ‘Cure’ in a hypothetical group of cancer patients. We previously found little evidence that this point of ‘cure’ was reached after 23 years of follow-up among two populations of women diagnosed with invasive breast cancer in England and Australia during the 1980–1995 [6]. Subsequent analyses have supported this conclusion [7, 8]. ‘Cure’ among breast cancer patients has not yet been examined in the context of screen-detection. It is possible that micro-metastases, a likely candidate for the continued excess mortality seen in breast cancer patients overall [9], may be absent in the subpopulation diagnosed asymptomatically via screening. This subgroup would not then experience any long-term excess (cancer-related) mortality. This question is of great interest in the context of the recent review of the benefits and harms of mammographic screening [10] and the expansion of the screening age range in the UK to women aged 49–73 years [11].

y via screening. This subgroup would not then experience any long-term excess (cancer-related) mortality. This question is of great interest in the context of the recent review of the benefits and harms of mammographic screening [10] and the expansion of the screening age range in the UK to women aged 49–73 years [11]. Contrasting with survival, population ‘cure’ is independent of lead-time bias [12]. Indeed, the additional time afforded by early detection inflates cancer survival estimates at a given point in time after diagnosis, while it does not affect the proportion of patients who eventually display no long-term excess mortality. We aim here to establish whether women who are diagnosed asymptomatically via screening display long-term excess mortality. We also analyse patterns by socioeconomic status and ethnicity to investigate whether these impact ‘cure’. materials and methods cohort selection We examined women aged 50–70, diagnosed with a primary, invasive, non-metastatic breast cancer between 1 April 1989 and 31 March 2011 in the West Midlands region of England. Only those who had been continuously eligible for screening from the age of 50 onwards were included (described in detail elsewhere [13]). Cancer registry data on these individuals were obtained from the West Midlands Cancer Intelligence Unit and Breast Screening Quality Assurance Reference Centre [14]. Additional information was provided by Hospital Episode Statistics (HES) records individually linked to National Breast Screening Service (NBSS) data. Follow-up was complete on all women up to 31 July 2012.

ned from the West Midlands Cancer Intelligence Unit and Breast Screening Quality Assurance Reference Centre [14]. Additional information was provided by Hospital Episode Statistics (HES) records individually linked to National Breast Screening Service (NBSS) data. Follow-up was complete on all women up to 31 July 2012. tumour stage Information on tumour size, nodal involvement and presence of metastases was used to establish each woman's extent of disease at diagnosis, either localised (confined to the organ of origin) or regional (spread to adjacent muscle, organ, fat, connective tissue or regional lymph nodes). Those with distant metastases were excluded from all analyses a priori, since ‘cure’ was not a reasonable expectation for these women. deprivation Deprivation was measured using the income domain of the English indices of deprivation for 2004, 2007 or 2010 [15–17]. These scores are derived from routine administrative data, pertaining to the years 2001, 2005 and 2008, respectively, for each of the 32 482 Lower Super Output Areas as defined at the 2001 census (LSOAs, ∼1500 people). The scores are categorised according to the quintiles of their national distribution. Each woman was assigned to one of five deprivation levels on the basis of her address of residence when diagnosed.

008, respectively, for each of the 32 482 Lower Super Output Areas as defined at the 2001 census (LSOAs, ∼1500 people). The scores are categorised according to the quintiles of their national distribution. Each woman was assigned to one of five deprivation levels on the basis of her address of residence when diagnosed. Our approach for deriving ethnicity information for this cohort has been described [13]. Briefly, data on each woman's ethnicity were gathered from self-reports given on admittance to hospital (from HES data, 83% of women), or where this was missing, on presentation for breast screening (from NBSS data, 7%). We imputed the remaining 10% of ethnicity data using name recognition software, Onomap [18]. This software matches the first and last names of the cohort patients with databases of names from different ethnicities. estimation of net survival and ‘cure’ We estimated net survival using the non-parametric Pohar Perme estimator [19] implemented in stns: software available for Stata 13. Net survival provides an estimate of survival from the cancer itself, adjusting for expected mortality from other causes, which was obtained from ethnic-specific life tables for England and Wales adjusted for deprivation [20].

n-parametric Pohar Perme estimator [19] implemented in stns: software available for Stata 13. Net survival provides an estimate of survival from the cancer itself, adjusting for expected mortality from other causes, which was obtained from ethnic-specific life tables for England and Wales adjusted for deprivation [20]. We fitted flexible parametric log-cumulative excess hazard regression models [21] to estimate the age-adjusted excess hazard of breast cancer death. Models were fitted to follow-up times up to the 95th centile of (all) deaths. We assessed the linearity and time-dependence of age at diagnosis by the inclusion of restricted cubic splines, with the knots placed within the range of the data. Population ‘cure’ was then evaluated from the most parsimonious age-adjusted model by assuming that the excess hazard became equal to zero from a given time (as implemented in the software stpm2) [22, 23]. The final model was selected based on the lowest Akaike's information criterion (AIC) with a reduction of 3 or more in the AIC between successive models [24]. Where the difference between the ‘cure’ model and the age-adjusted model showed a reduction of 3 or more in the AIC, there was taken to be evidence of ‘cure’. The presence of ‘cure’ was also assessed by visual inspection of the survival curves.

C) with a reduction of 3 or more in the AIC between successive models [24]. Where the difference between the ‘cure’ model and the age-adjusted model showed a reduction of 3 or more in the AIC, there was taken to be evidence of ‘cure’. The presence of ‘cure’ was also assessed by visual inspection of the survival curves. co-variables examined Analyses were stratified by screening status (screen-detected/not-screen-detected). Additionally, we examined ‘cure’ by age (50–59/60–70 years), tumour stage (localised/regional), deprivation quintile [less deprived (quintiles 1 and 2)/more deprived (quintiles 3–5)] and ethnicity (White/Asian/Black). We also conducted a restricted analysis of localised cases only, by both age and deprivation. results The analysis included 19 800 women who had a first primary malignant breast tumour which was not classified as distant at diagnosis (mean age 57.5 years, standard deviation = 5.0).

co-variables examined Analyses were stratified by screening status (screen-detected/not-screen-detected). Additionally, we examined ‘cure’ by age (50–59/60–70 years), tumour stage (localised/regional), deprivation quintile [less deprived (quintiles 1 and 2)/more deprived (quintiles 3–5)] and ethnicity (White/Asian/Black). We also conducted a restricted analysis of localised cases only, by both age and deprivation. results The analysis included 19 800 women who had a first primary malignant breast tumour which was not classified as distant at diagnosis (mean age 57.5 years, standard deviation = 5.0). There was an overwhelming lack of evidence for ‘cure’. Despite high survival at 1, 5 and 10 years across the subgroups examined (defined by screening status, tumour stage, age, deprivation and ethnicity), there was a general pattern of a continuous decrease in net survival through time, with no obvious asymptotic tendency within 12 years (Figure 2, supplementary Figures S1–S3, available at Annals of Oncology online). The model-based analyses confirmed this observation; no ‘cure’ models were found to fit well for any subgroup examined (Table 1, supplementary Tables S1–S3, available at Annals of Oncology online).Table 1. Evidence of ‘cure’ by screen-detection status: women diagnosed in the West Midlands region of England 1989–2011 All Screen-detected women Non-screen-detected women n (%) Deaths (% of n) Evidence of ‘cure’?a n (%) Deaths (% of n) Evidence of ‘cure’? n (%) Deaths (% of n) Evidence of ‘cure’?

There was an overwhelming lack of evidence for ‘cure’. Despite high survival at 1, 5 and 10 years across the subgroups examined (defined by screening status, tumour stage, age, deprivation and ethnicity), there was a general pattern of a continuous decrease in net survival through time, with no obvious asymptotic tendency within 12 years (Figure 2, supplementary Figures S1–S3, available at Annals of Oncology online). The model-based analyses confirmed this observation; no ‘cure’ models were found to fit well for any subgroup examined (Table 1, supplementary Tables S1–S3, available at Annals of Oncology online).Table 1. Evidence of ‘cure’ by screen-detection status: women diagnosed in the West Midlands region of England 1989–2011 All Screen-detected women Non-screen-detected women n (%) Deaths (% of n) Evidence of ‘cure’?a n (%) Deaths (% of n) Evidence of ‘cure’? n (%) Deaths (% of n) Evidence of ‘cure’? All women 19 800 (100.0) 3153 (15.9) No evidence 10 466 (100.0) 984 (9.4) No evidence 9334 (100.0) 2169 (23.2) No evidence Age at diagnosis 50–59 years 12 933 (65.3) 2316 (17.9) No evidence 6563 (62.7) 699 (10.7) No evidence 6370 (68.2) 1617 (25.4) No evidence 60–69 years 6867 (34.7) 837 (12.2) No evidence 3903 (37.3) 285 (7.3) No evidence 2964 (31.8) 552 (18.6) No evidence Extent of disease at diagnosisb Localised 12 176 (61.5) 1121 (9.2) No evidence 7548 (72.1) 499 (6.6) No evidence 4628 (49.6) 622 (13.4) No evidence Regional 6364 (32.1) 1721 (27.0) No convergence 2385 (22.8) 422 (17.7) No evidence 3979 (42.6) 1299 (32.6) No evidence Ethnicityc White 19 040 (96.2) 3030 (15.9) No evidence 10 087 (96.4) 949 (9.4) No evidence 8953 (95.9) 2081 (23.2) No evidence Asian 572 (2.9) 85 (14.9) No evidence 293 (2.8) 25 (8.5) No convergence 279 (3.0) 60 (21.5) No evidence Black 188 (0.9) 38 (20.2) No evidence 86 (0.8) 10 (11.6) No convergence 102 (1.1) 28 (27.5) No evidence Deprivation quintiled Less deprived (1 and 2) 8592 (43.4) 1186 (13.8) No evidence 4519 (43.2) 345 (7.6) No evidence 4073 (43.6) 841 (20.6) No evidence More deprived (3–5) 11 190 (56.5) 1964 (17.6) No evidence 5940 (56.8) 639 (10.8) No evidence 5250 (56.2) 1325 (25.2) No evidence Among localised cases only n = 12 176 (100.0) n = 7548 (100.0) n = 4628 (100.0) Age at diagnosis 50–59 years 7701 (63.2) 796 (10.3) No evidence 4576 (60.6) 335 (7.3) No evidence 3125 (67.5) 461 (14.8) No evidence 60–69 years 4475 (36.8) 325 (7.3) No evidence 2972 (39.4) 164 (5.5) No evidence 1503 (32.5) 161 (10.7) No evidence Deprivation quintile Less deprived (1 and 2) 5379 (44.2) 410 (7.6) No evidence 3276 (43.4) 159 (4.9) No evidence 2103 (45.4) 251 (11.9) No evidence More deprived (3–5) 6791 (55.8) 711 (10.5) No evidence 4267 (56.5) 340 (8.0) No evidence 2524 (54.5) 371 (14.7) No evidence aAs determined by the difference in the AIC: reduction of 3 or more = evidence of ‘cure’; increase or a reduction of <3 = no evidence of ‘cure’; ‘cure’ model unable to converge = ‘no convergence’.

More deprived (3–5) 6791 (55.8) 711 (10.5) No evidence 4267 (56.5) 340 (8.0) No evidence 2524 (54.5) 371 (14.7) No evidence aAs determined by the difference in the AIC: reduction of 3 or more = evidence of ‘cure’; increase or a reduction of <3 = no evidence of ‘cure’; ‘cure’ model unable to converge = ‘no convergence’. bUnstaged cancers (n = 1260) were excluded from extent-specific analyses. cIndividual ethnicity: White includes all categories other than Asian and Black (see text). dQuintile of the IMD income domain score of the woman's LSOA of residence at diagnosis (see text). Women with missing data were excluded (n = 18). Figure 2. Non-parametric and modelled estimates of net survival up to 12 years following diagnosis. (A) All women. (B) All women, localised disease. (C) Screen-detected women. (D) Screen-detected women, localised disease. (E) Non-screen-detected women. (F) Non-screen-detected women, localised disease. (G) Women aged 50–59 years at diagnosis. (H) Women aged 60–70 years at diagnosis. (I) Less deprived women (quintiles 1 and 2). (J) More deprived women (quintiles 3–5). (K) Asian women. (L) Black women. Models did not always converge. Among the screen-detected group, parametric survival models could not be fitted for either Black or Asian women due to small numbers of patients and deaths in these groups. Fitting an asymptote to the age-adjusted model for women with regional disease also proved unachievable. For these women, ‘cure’ was not assessed using the modelling approach.

oup, parametric survival models could not be fitted for either Black or Asian women due to small numbers of patients and deaths in these groups. Fitting an asymptote to the age-adjusted model for women with regional disease also proved unachievable. For these women, ‘cure’ was not assessed using the modelling approach. The one subgroup which displayed a different pattern was affluent women screen-detected with localised disease (Figure 3). Here, survival was very high, in excess of 98% after 10 years. The net survival curve tended slightly towards an asymptote, and the model also confirmed a flattening of the curve. The ‘cure’ model did not, however, display a better fit than the age-adjusted model alone.Figure 3. Non-parametric and modelled estimates of net survival up to 12 years following diagnosis: less deprived women with localised disease whose tumour was screen-detected. discussion We have shown that there is a persistent lack of ‘cure’ among this cohort of middle-aged women diagnosed with breast cancer for all sociodemographic groups, even if their cancer is localised and/or detected via screening. Elevated mortality for all groups persists beyond the 10th anniversary of diagnosis. There was suggestive, but weak, evidence of ‘cure’ around 12 years after diagnosis for less deprived women with localised disease whose cancer was detected via screening. Although the net survival curve tended to level from the 11th year following diagnosis, the model-based analysis did not support the hypothesis that ‘cure’ was present, however.

ak, evidence of ‘cure’ around 12 years after diagnosis for less deprived women with localised disease whose cancer was detected via screening. Although the net survival curve tended to level from the 11th year following diagnosis, the model-based analysis did not support the hypothesis that ‘cure’ was present, however. strengths and limitations Our approach has several strengths in comparison with previous studies. Life tables specific, not just to the deprivation profile of this population, but also its ethnic mix, were applied to obtain the most accurate estimates of expected mortality in this setting. Screening status was established on the basis of individually linked data, and we restricted the cohort to women whom we know to have been invited for screening from their 50th birthday onwards. The influence of screen-detection upon ‘cure’ is not thus obscured by older women attending screening for the first time at ages over 50 years. We used flexible models to test the existence of ‘cure’, rather than one which assumes its existence, as necessitated by other methods [5, 7]. This means that the presence of the ‘cured’ proportion can therefore be formally evaluated. As the AIC assesses the whole curve, however, while for ‘cure’ the tail of the curve (where there are more sparse data) is most important, caution must be exercised in relying solely on this evaluation. To this end, the need for visual inspection of the net survival curves continues to be emphasised [7], which we did, with the same conclusions.

ole curve, however, while for ‘cure’ the tail of the curve (where there are more sparse data) is most important, caution must be exercised in relying solely on this evaluation. To this end, the need for visual inspection of the net survival curves continues to be emphasised [7], which we did, with the same conclusions. There are limitations of our analysis. Breast cancer survival is high, thus there were a relatively small number of deaths in our data. We therefore restricted all analyses to the first 95% of deaths to reduce poor model fit in particular at the end of follow-up. A related concern is the inappropriateness of the AIC for evaluating ‘cure’ models [7], because the AIC is less sensitive to the portion of follow-up where ‘cure’ occurs. However, deaths here occurred at a steadily decreasing rate throughout follow-up, with a not-so-skewed distribution of times to death (mean time to death 4.34 years, median 3.24 years, inter-quartile range 1.55–6.06). We have previously evaluated cure up to 23 years after diagnosis. Although the maximum follow-up of the present cohort was similar, our cautious restriction of examining ‘cure’ only up to the 95th centile deaths meant that the effective follow-up was much shorter: 12.3 years. It is possible that ‘cure’ could be reached by survivors remaining after this time, although the trajectory of the survival curves suggests that this is very unlikely.

imilar, our cautious restriction of examining ‘cure’ only up to the 95th centile deaths meant that the effective follow-up was much shorter: 12.3 years. It is possible that ‘cure’ could be reached by survivors remaining after this time, although the trajectory of the survival curves suggests that this is very unlikely. We excluded confirmed distant (metastatic) cancers from our analyses since we did not reasonably expect ‘cure’ to be attained for these patients. However, because we also included unstaged tumours in the overall analyses, survival reported here for all patients is a slight underestimate of the survival of women with localised or regional tumours. possible causal explanations Persistent excess mortality due to cancer among screen-detected women into the second decade following their diagnosis seems unlikely to be due to treatment inadequacies at the time of the initial diagnosis, but rather more likely to be due to long-term effects of either the cancer itself or of its treatment, and, or the distinctive natural history of this malignancy. For example, some women whose disease is apparently localised at diagnosis harbour micro-metastatic disease: it is possible that this is also the case among women who are asymptomatic and screen-detected. The data available did not allow us to investigate ‘cure’ by molecular subtype of breast cancer (e.g. luminal A or B, triple negative, HER2). Certain subtypes may have already metastasised even when the tumour itself is localised [25, 26], which could partly explain the lack of cure in the cohort overall.

possible causal explanations Persistent excess mortality due to cancer among screen-detected women into the second decade following their diagnosis seems unlikely to be due to treatment inadequacies at the time of the initial diagnosis, but rather more likely to be due to long-term effects of either the cancer itself or of its treatment, and, or the distinctive natural history of this malignancy. For example, some women whose disease is apparently localised at diagnosis harbour micro-metastatic disease: it is possible that this is also the case among women who are asymptomatic and screen-detected. The data available did not allow us to investigate ‘cure’ by molecular subtype of breast cancer (e.g. luminal A or B, triple negative, HER2). Certain subtypes may have already metastasised even when the tumour itself is localised [25, 26], which could partly explain the lack of cure in the cohort overall. A further hypothesis has been proposed that the act of breast cancer surgery itself provokes the activation of latent micro-metastases [27, 28]. However, this mechanism has been suggested only among pre-menopausal women, whereas women under 50 were not included in this study.

The data available did not allow us to investigate ‘cure’ by molecular subtype of breast cancer (e.g. luminal A or B, triple negative, HER2). Certain subtypes may have already metastasised even when the tumour itself is localised [25, 26], which could partly explain the lack of cure in the cohort overall. A further hypothesis has been proposed that the act of breast cancer surgery itself provokes the activation of latent micro-metastases [27, 28]. However, this mechanism has been suggested only among pre-menopausal women, whereas women under 50 were not included in this study. public health considerations The public health implications of these findings are twofold. First, our analysis strongly suggests that despite very high survival overall, women diagnosed with breast cancer experience a continuing risk of death from cancer beyond the 10th anniversary of their diagnosis, and that this occurs irrespective of their extent of disease at diagnosis. This has implications for the way in which clinicians, policy makers and public health professionals communicate with patients regarding the long-term prognosis to women newly diagnosed with breast cancer. In particular, data such as these question whether a woman diagnosed once with breast cancer can be considered to be disease-free, and increases the importance of using the correct language when communicating with those who have previously been treated for breast cancer [29, 30]. Second, since the pattern is consistent for both screen-detected and non-screen-detected women, our data suggest that screening does not afford protection from long-term excess mortality, even though it is associated with an important and significant survival advantage at all times since diagnosis, independent of lead-time bias [13]. Communication of this important and unique feature of breast cancer to those women considering screening and diagnosed via screening should also be carefully considered.

y, even though it is associated with an important and significant survival advantage at all times since diagnosis, independent of lead-time bias [13]. Communication of this important and unique feature of breast cancer to those women considering screening and diagnosed via screening should also be carefully considered. conclusion Our analyses have shown an overwhelming lack of evidence for ‘cure’ in our cohort of breast cancer patients. We have demonstrated continued excess mortality up to 12 years after diagnosis, irrespective of age, screening status, stage of disease, ethnicity or deprivation status. These findings are unlikely to be due to methodological inadequacies. Despite high and continually increasing survival among middle-aged women diagnosed in the UK, breast cancer leads to a tiny, but persistent, increased risk of death for all groups of women, including those whose cancer is detected asymptomatically. Communication of the long-term consequences of breast cancer among women recently diagnosed and to those considering undergoing screening should take due consideration of these patterns. funding This work was supported by the National Awareness and Early Diagnosis Initiative (NAEDI) (C23409/A14031 to MM) and by Cancer Research UK (C23409/A11415 to LMW and C1336/A11700 to BR). disclosure The authors have declared no conflicts of interest. Supplementary Material Supplementary Data

introduction Sorafenib is currently acknowledged as a standard therapy for advanced hepatocellular carcinoma (HCC), and is available worldwide [1]. After the introduction of sorafenib, a number of phase III trials of various molecular-targeted agents versus sorafenib as first-line chemotherapy have been conducted, but none of the agents examined so far has shown superior survival benefit to sorafenib [1].

tocellular carcinoma (HCC), and is available worldwide [1]. After the introduction of sorafenib, a number of phase III trials of various molecular-targeted agents versus sorafenib as first-line chemotherapy have been conducted, but none of the agents examined so far has shown superior survival benefit to sorafenib [1]. Hepatic arterial infusion chemotherapy (HAIC) is employed to treat patients with advanced HCC [2, 3]. This treatment modality is associated with increased local concentrations of the anticancer agents in the tumor and reduced systemic distribution of the drugs, and a stronger antitumor effect and lower incidence of systemic adverse reactions may be expected when compared with systemic chemotherapy. In fact, high response rates, favorable long-term outcomes, and acceptable toxicities with some chemotherapeutic regimens of HAIC have been reported [2, 3]. However, no consensus has been reached as to its place as a standard treatment of advanced HCC. Among HAI regimens, cisplatin alone can be easily administered using the Seldinger technique, without the need for an indwelling reservoir system [4]. In addition, sorafenib has been shown to interact with platinum transporter proteins [5], and to exert a synergistic anticancer effect with cisplatin in preclinical research [6]. Clinical trials of sorafenib used in combination with cisplatin have been carried out for various cancers [7–9], and favorable outcomes have been reported. Herein, we report the results of a randomized phase II trial of sorafenib plus HAIC with cisplatin (SorCDDP) versus sorafenib alone (Sor). The primary end point was the overall survival, while the secondary end points were the time to progression, response rate, and adverse events.

avorable outcomes have been reported. Herein, we report the results of a randomized phase II trial of sorafenib plus HAIC with cisplatin (SorCDDP) versus sorafenib alone (Sor). The primary end point was the overall survival, while the secondary end points were the time to progression, response rate, and adverse events. methods-patients and methods patient eligibility The patient inclusion criteria were as follows: advanced HCC confirmed histologically or by typical findings of hypervascular tumor on computed tomography (CT) or angiography and elevated serum alpha-fetoprotein (AFP), or protein induced by vitamin K absence or antagonist-II level; unsuitable for surgical resection, liver transplantation, local ablative therapy or transarterial chemoembolization (TACE); no prior history of chemotherapy; age 20–79 years old; presence of intrahepatic tumors affecting the prognosis irrespective of the presence of extrahepatic tumors; Eastern Cooperative Oncology Group Performance Status 0–1; adequate organ function [neutrophil count ≥1500 /mm3, hemoglobin ≥8.5 g/dl, platelet count ≥60 000 /mm3, serum total bilirubin ≤2.0 mg/dl, serum albumin ≥2.8 g/dl, aspartate aminotransferase (AST) and alanine aminotransferase (ALT) ≤5 times the upper limits of normal, serum creatinine ≤1.2 mg/dl, creatinine clearance ≥60 ml/min]; Child-Pugh score 5–7; HAIC technically feasible; written informed consent.

ount ≥60 000 /mm3, serum total bilirubin ≤2.0 mg/dl, serum albumin ≥2.8 g/dl, aspartate aminotransferase (AST) and alanine aminotransferase (ALT) ≤5 times the upper limits of normal, serum creatinine ≤1.2 mg/dl, creatinine clearance ≥60 ml/min]; Child-Pugh score 5–7; HAIC technically feasible; written informed consent. The main exclusion criteria were as follows: refractory pleural effusion or ascites; hepatic encephalopathy; severe and active co-morbidity or concomitant malignancy; allergic reaction to iodine contrast medium precluding angiography; pregnant and lactating females; females of childbearing age unless using effective contraception; and unsatisfactory general condition. Patients with hepatitis B or C virus infection were eligible for enrollment in this trial, provided they fulfilled the eligibility criterion pertaining to hepatic reserve.

angiography; pregnant and lactating females; females of childbearing age unless using effective contraception; and unsatisfactory general condition. Patients with hepatitis B or C virus infection were eligible for enrollment in this trial, provided they fulfilled the eligibility criterion pertaining to hepatic reserve. treatments The enrolled patients were randomly assigned 2:1 to the SorCDDP arm or the Sor arm. Randomization was done centrally using a minimization method with biased-coin assignment [10]. The dynamic allocation factors were the presence of portal vein tumor thrombosis and extrahepatic metastasis. In patients of the SorCDDP arm, based on the results of a phase I trial [11], sorafenib (Nexavar®, Bayer Health Care Pharmaceuticals; West Haven, CT, USA) was administered orally at a dose of 400 mg bid, and cisplatin (IA call®, Nippon Kayaku Co., Ltd; Tokyo, Japan) was administered concurrently at 65 mg/m2/cycle via a catheter placed in the proper, right, or left hepatic artery, or another feeding artery, every 4–6 weeks. In patients of the Sor arm, sorafenib was administered orally at a dose of 400 mg bid. The sorafenib treatment in both arms was continued until tumor progression or unacceptable toxicity, and the HAIC with cisplatin was administered up to a maximum of six cycles until radiological or symptomatic tumor progression, unacceptable toxicity, or technical difficulty in repeating the HAIC. If the protocol therapies were discontinued, the patient was allowed to receive other anticancer treatment at the physician's discretion.

th cisplatin was administered up to a maximum of six cycles until radiological or symptomatic tumor progression, unacceptable toxicity, or technical difficulty in repeating the HAIC. If the protocol therapies were discontinued, the patient was allowed to receive other anticancer treatment at the physician's discretion. The occurrence of grade 4 hematological toxicity, grade 3 non-hematological toxicity was generally considered as indication for suspending the sorafenib administration. When the toxicities improved by at least one grade when compared with the suspension criteria, the treatment was resumed at a reduced dose of 400 mg daily. If additional dose reduction was required, the dose was reduced further to a single administration of 400 mg every other day. The criteria for administering HAIC with cisplatin were as follows: neutrophil count ≥1200/mm3, platelet count ≥50 000/mm3, serum total bilirubin ≤3.0 mg/dl, serum AST or ALT levels ≤5 times the upper limit of normal, and a serum creatinine level ≤1.5 mg/dl. If the above parameters did not fall within the starting criteria, the HAIC with cisplatin was postponed until the criteria were fulfilled.

count ≥1200/mm3, platelet count ≥50 000/mm3, serum total bilirubin ≤3.0 mg/dl, serum AST or ALT levels ≤5 times the upper limit of normal, and a serum creatinine level ≤1.5 mg/dl. If the above parameters did not fall within the starting criteria, the HAIC with cisplatin was postponed until the criteria were fulfilled. response and toxicity assessment Evaluation of the tumor response by dynamic CT or MRI was carried out every 6 weeks using the modified Response Evaluation Criteria in Solid Tumors (RECIST) [12]. The responses were evaluated centrally by three independent reviewers. Overall survival was measured from the date of enrollment to the date of death or the date of the last follow-up. Time to progression was defined as the time from the date of enrollment to the first documentation of disease progression or death. Assessment of adverse events was based on the National Cancer Institute Common Toxicity Criteria, version 4.0.

he date of enrollment to the date of death or the date of the last follow-up. Time to progression was defined as the time from the date of enrollment to the first documentation of disease progression or death. Assessment of adverse events was based on the National Cancer Institute Common Toxicity Criteria, version 4.0. statistical analysis This was a multicenter open-labeled randomized phase II trial. The primary end point was overall survival stratified by the allocation factors, including the presence/absence of portal vein tumor thrombosis and extrahepatic metastases. If the median survival associated with Sor were assumed as 7.0 months and that of SorCDDP as 9.5 months, the hazard ratio (HR) was 0.74. SorCDDP would be judged as being favorable if the HR is 0.74 or lower. A total of 105 patients were needed to estimate the 1-year survival rate with an accuracy of ±10%. This study did not have sufficient statistical power to permit formal statistical comparison between the two arms.

hs, the hazard ratio (HR) was 0.74. SorCDDP would be judged as being favorable if the HR is 0.74 or lower. A total of 105 patients were needed to estimate the 1-year survival rate with an accuracy of ±10%. This study did not have sufficient statistical power to permit formal statistical comparison between the two arms. The differences in the categorical data between the two groups were analyzed by Wilcoxon's test. The overall survival time and time to progression were estimated by using the Kaplan–Meier method and the curves were compared using the log-rank test. HRs of the treatment effects were estimated using a Cox regression model, and stratified results by dynamic allocation factors, including the presence/absence of portal vein tumor thrombosis and extrahepatic metastasis, as well as unstratified results, were presented. This clinical trial was conducted with the approval of the review board of each participating institution and in accordance with the Declaration of Helsinki. This trial is registered with UMIN-CTR (http://www.umin.ac.jp/ctr/index-j.htm), identification number (UMIN000005703). Patient registration, random treatment allocation, and data collection were managed by the Japan Clinical Research Support Unit data center. The integrity of the data was ensured through careful review by the staff of the data center, the coordinating investigators (MI and SS), and the trial statistician (TS). All the data were fixed on 28 December 2014, and all the analyses of efficacy were carried out based on the full analysis set (FAS) by the TS using SAS 9.4 and JMP Pro 11.

ata was ensured through careful review by the staff of the data center, the coordinating investigators (MI and SS), and the trial statistician (TS). All the data were fixed on 28 December 2014, and all the analyses of efficacy were carried out based on the full analysis set (FAS) by the TS using SAS 9.4 and JMP Pro 11. results patient characteristics From June 2011 to December 2013, a total of 108 patients were enrolled and randomized into the two treatment arms (Figure 1). Forty-two patients were assigned to the Sor arm and 66 patients to the SorCDDP arm. While the planned random assignment was 2:1, the actual randomization ratio was 1.6:1, which was within random error. One patient from each of the arms could not receive the chemotherapy (development of paraplegia due to disease progression in one patient of the Sor arm, and withdrawal of informed consent in one patient of the SorCDDP arm). Therefore, the FAS included 41 patients in the Sor arm and 65 patients in the SorCDDP arm.Figure 1. Consort diagram. The patient characteristics of the 106 patients of the FAS are presented in Table 1. Seropositivity for hepatitis C viral antibody was more frequent in the Sor arm (n = 20, 48.8%) than in the SorCDDP arm (n = 18, 27.7%), and portal vein tumor thrombosis was less frequent in the Sor arm (n = 17, 41.5%) than in the SorCDDP arm (n = 40, 61.5%). In terms of all other variables, the patient characteristics were well-balanced.Table 1. Baseline patient characteristics

ent in the Sor arm (n = 20, 48.8%) than in the SorCDDP arm (n = 18, 27.7%), and portal vein tumor thrombosis was less frequent in the Sor arm (n = 17, 41.5%) than in the SorCDDP arm (n = 40, 61.5%). In terms of all other variables, the patient characteristics were well-balanced.Table 1. Baseline patient characteristics Characteristics Sorafenib alone (n = 41) Sorafenib + HAIC (cisplatin) (n = 65) Number of patients % Number of patients % Age, years Median 64 66 Range 42–78 25–79 Sex Male 32 78.1 56 86.2 Female 9 22.0 9 13.8 ECOG performance status 0 33 80.5 50 76.9 1 8 19.5 15 23.1 Etiology Hepatitis B 9 22.0 22 33.8 Hepatitis C 20 48.8 18 27.7 Child-Pugh score 5 27 65.9 38 58.5 6 12 29.3 19 29.2 7 2 4.9 8 12.3 Ascites 4 9.8 10 15.4 Previous therapy 21 51.2 33 50.8 Resection 6 17 PEI/RFA 7 8 TACE 14 23 Radiation 1 1 Other 2 1 BCLC stage B 16 39.0 19 29.2 C 25 61.0 46 70.8 Portal vein tumor thrombosis 17 41.5 40 61.5 Vp1 0 4 10.0 Vp2 4 23.5 9 22.5 Vp3 7 41.4 14 35.0 Vp4 6 35.3 13 32.5 Extrahepatic spread 13 31.7 19 29.2 Lung 6 8 Bone 3 1 Lymph node 6 10 Adrenal 1 1 Other 2 4 Number of tumors 1 4 9.8 8 12.3 2 3 7.3 5 7.7 3 1 2.4 1 1.5 4 3 7.3 5 7.7 ≥5 30 73.2 46 70.8 Maximum tumor size, cm Median 5.2 5.1 Range 1.1–17.5 1.0–20.0 Serum α-fetoprotein, ng/ml Median 188 223.5 Range 2–749 412 1.2–394 944 PIVKA II, mAU/ml Median 1790 1772 Range 9–1 410,000 10–261 920 ECOG, Eastern Cooperative Oncology Group; PEI/RFA, percutaneous ethanol injection/radiofrequency ablation; TACE, transarterial chemoembolization; BCLC, Barcelona Clinic Liver Cancer Group; Vp1, tumor thrombosis distal to the second branches of the portal vein; Vp2, tumor thrombosis in the second branches of the portal vein; Vp3, tumor thrombosis in the first branches of the portal vein; Vp4, tumor thrombosis in the main trunk of the portal vein or the opposite side branch of the portal vein; PIVKA II, protein induced by vitamin K absence or antagonist-II.

of the portal vein; Vp2, tumor thrombosis in the second branches of the portal vein; Vp3, tumor thrombosis in the first branches of the portal vein; Vp4, tumor thrombosis in the main trunk of the portal vein or the opposite side branch of the portal vein; PIVKA II, protein induced by vitamin K absence or antagonist-II. treatments By the data cutoff point, the protocol treatment had been discontinued in 41 patients of the Sor arm and 62 patients of the SorCDDP arm. The median number of cisplatin administrations and the median total dose of cisplatin in the SorCDDP arm were two times (range, 1–6 times) and 222 mg (range, 70–709 mg), respectively. The median dose intensity (range) was 488 mg/day (146–800 mg) in the Sor arm and 540 mg/day (193–800 mg) in the SorCDDP arm (P = 0.70). The proportion of patients in whom dose reduction of sorafenib was necessitated was 49.2% in the Sor arm and 63.4% in the SorCDDP arm. The median treatment duration (range) was 86 days (16–449 days) in the Sor arm and 75 days (4–881 days) in the SorCDDP arm (P = 0.58). After termination of the protocol treatment, 24 patients (59%) in the Sor arm and 40 patients (61.5%) in the SorCDDP arm received subsequent therapies, as follows: HAIC (8 and 19 patients, respectively), TACE (8 and 14 patients, respectively), local ablation (1 and 2 patients, respectively), other systemic chemotherapy (11 and 32 patients, respectively), palliative resection (2 and 5 patients, respectively), and radiotherapy (0 and 9 patients, respectively).

herapies, as follows: HAIC (8 and 19 patients, respectively), TACE (8 and 14 patients, respectively), local ablation (1 and 2 patients, respectively), other systemic chemotherapy (11 and 32 patients, respectively), palliative resection (2 and 5 patients, respectively), and radiotherapy (0 and 9 patients, respectively). efficacy At the final analysis, 37 patients of the Sor arm and 49 patients of the SorCDDP arm had died. The median survivals in the Sor and SorCDDP arms were 8.7 and 10.6 months, respectively (Figure 2A). The HR stratified by the allocation factors, including the presence/absence of portal vein tumor thrombosis and extrahepatic metastases (95% CI), was 0.60 (0.38–0.96), and P-value was 0.031. The crude HR [95% confidence interval (CI)] was 0.68 (0.44–1.049) (P = 0.073). The forest plot showing the pre-specified subgroup analyses of overall survival is shown in Figure 3. The patient subgroup with serum AFP <400 ng/ml showed a better overall survival in the SorCDDP arm (median 14.8 months) than in the Sor arm (median 8.7 months) (P = 0.042). At the data cutoff point, disease progression was observed in 39 patients in the Sor arm and 61 patients in the SorCDDP arm. The median time to progression was 2.8 months in the Sor arm and 3.1 months in the SorCDDP arm (Figure 2B). The crude HR was 0.78 (95% CI, 0.52–1.16, P = 0.212) and the HR stratified by the allocation factors was 0.78 (95% CI, 0.50–1.21, P = 0.257).Figure 2. Kaplan–Meier curves of overall survival (A) and time to progression (B) in the sorafenib arm (blue line) and sorafenib plus hepatic arterial infusion chemotherapy with cisplatin arm (green line). The tick marks indicate censored cases.

ratified by the allocation factors was 0.78 (95% CI, 0.50–1.21, P = 0.257).Figure 2. Kaplan–Meier curves of overall survival (A) and time to progression (B) in the sorafenib arm (blue line) and sorafenib plus hepatic arterial infusion chemotherapy with cisplatin arm (green line). The tick marks indicate censored cases. Figure 3. Forest plots showing subgroup analyses of the overall survival. TACE, transarterial chemoembolization; ECOG, Eastern Cooperative Oncology Group; OS, overall survival; HR, hazard ratio; CI, confidence interval. In the judgment by the central review, the number of patients evaluable by the modified RECIST criteria was 41 in the Sor arm and 60 patients in the SorCDDP arm. The response rate (95% CI) was 7.3% (1.5–19.9%) in the Sor arm and 21.7% (12.1–34.2%) in the SorCDDP arm (P = 0.09) (supplementary Figure 1, available at Annals of Oncology online). adverse events The adverse events in both the arms during the entire treatment period until the final analysis are presented in Table 2. Neutropenia, leukocytopenia, decreased hemoglobin, thrombocytopenia, hyponatremia, nausea, and hiccups of all grades were more frequent in the SorCDDP arm than in the Sor arm. There were two treatment-related deaths in this series: one developed liver failure 9 months after the initiation of SorCDDP therapy, and the other developed pulmonary infection 2 months after the initiation of Sor therapy.Table 2. Adverse events

of all grades were more frequent in the SorCDDP arm than in the Sor arm. There were two treatment-related deaths in this series: one developed liver failure 9 months after the initiation of SorCDDP therapy, and the other developed pulmonary infection 2 months after the initiation of Sor therapy.Table 2. Adverse events Sorafenib alone arm Sorafenib + HAIC (cisplatin) arm All grades Grade 3 Grade 4 All grades Grade 3 Grade 4 No. of pts (%) No. of pts (%) No. of pts (%) No. of pts (%) No. of pts (%) No. of pts (%) WBC decreased 18 43.9 0 0 0 0 49 75.4 12 18.5 0 0 Neu decreased 18 43.9 0 0 0 0 39 60 8 12.3 2 3.1 Hb decreased 30 73.2 2 4.9 1 2.4 58 89.2 6 9.2 2 3.1 Plt decreased 33 80.5 1 2.4 0 0 58 89.2 19 29.2 0 0 Bilirubin increased 29 70.7 5 12.2 0 0 48 73.8 6 9.2 2 3.1 AST increased 41 100 8 19.5 3 7.3 65 100 21 32.3 1 1.5 ALT increased 37 90.2 7 17.1 1 2.4 61 93.8 12 18.5 1 1.5 γGTP increased 37 90.2 14 34.1 2 4.9 63 96.9 21 32.3 3 4.6 Hypoalbuminemia 32 78 3 7.3 0 0 63 96.9 2 3.1 0 0 Cr increased 11 26.8 0 0 0 0 25 38.5 1 1.5 0 0 Hyponatremia 22 53.7 5 12.2 0 0 53 81.5 18 27.7 0 0 Amylase increased 21 52.5 0 0 0 0 41 64.1 10 15.6 1 1.6 Fatigue 16 39 1 2.4 – – 28 43.1 7 10.8 – – Malaise 17 41.5 – – – – 31 47.7 – – – – Appetite loss 22 53.7 1 2.4 0 0 45 69.2 4 6.2 2 3.1 Nausea 8 19.5 0 0 – – 27 41.5 0 0 – – Vomiting 5 12.2 0 0 0 0 12 18.5 0 0 0 0 Diarrhea 17 41.5 1 2.4 0 0 23 35.4 4 6.2 0 0 Hand–foot synd 26 63.4 7 17.1 – – 41 63.1 9 13.8 – – Skin rash 11 26.8 2 4.9 0 0 12 18.5 3 4.6 0 0 Hypertension 24 58.5 9 22 0 0 32 49.2 19 29.2 0 0 Hiccups 0 0 0 0 – – 6 9.2 0 0 – – WBC, white blood count; Neu, neutrophils; Hb, hemoglobin; Plt, platelets; AST, aspartate aminotransferase; ALT, alanine aminotransferase; γGTP, γ-glutamyl transpeptidase; Cr, creatinine; synd, syndrome; pts, patients.

12 18.5 3 4.6 0 0 Hypertension 24 58.5 9 22 0 0 32 49.2 19 29.2 0 0 Hiccups 0 0 0 0 – – 6 9.2 0 0 – – WBC, white blood count; Neu, neutrophils; Hb, hemoglobin; Plt, platelets; AST, aspartate aminotransferase; ALT, alanine aminotransferase; γGTP, γ-glutamyl transpeptidase; Cr, creatinine; synd, syndrome; pts, patients. discussion In this study, SorCDDP yielded favorable overall survival when compared with Sor in patients with advanced HCC. The pre-specified HR stratified by the allocation factors (95% CI) was 0.60 (0.38–0.96), and P-value was 0.031. Because we had set the condition that SorCDDP would be judged as favorable if the HR for overall survival was 0.74 or lower, the primary end point of this study was met. In the pre-specified subgroup analysis of overall survival, the SorCDDP arm showed more favorable overall survival than the Sor arm in all the subgroups, and the efficacy of SorCDDP can be anticipated in almost all subjects who are suitable candidates for sorafenib treatment. In this trial, patients with hepatitis C viral infection showed a more favorable overall survival following sorafenib treatment than those with hepatitis B viral infection. However, it remains unknown whether patients with hepatitis C viral infection actually benefitted more from this treatment or not, because of the small sample size of this study. Furthermore, the overall survival in the SorCDDP arm was better than that in the Sor arm among the patients with serum AFP <400 ng/ml [crude HR, 0.53 (95% CI, 0.28–0.99)], whereas no difference was observed between the SorCDDP arm and the Sor arm among the patients with serum AFP ≥400 ng/ml. AFP may be one of the predictive biomarkers in patients receiving SorCDDP therapy, although the reason remains unknown.

ng the patients with serum AFP <400 ng/ml [crude HR, 0.53 (95% CI, 0.28–0.99)], whereas no difference was observed between the SorCDDP arm and the Sor arm among the patients with serum AFP ≥400 ng/ml. AFP may be one of the predictive biomarkers in patients receiving SorCDDP therapy, although the reason remains unknown. Recently, immuno-oncology agents, such as tremelimumab [13] and nivolumab [14], have been introduced as promising agents for advanced HCC. The characteristics of these agents are a high response rate and long-lasting antitumor efficacy. In our study also, the response rate in the SorCDDP arm (21.7%) was threefold higher than that in the Sor arm (7.3%), and some patients in the SorCDDP arm showed long-lasting survival over 2 years. With regard to time to progression, the stratified HR by the allocation factors was 0.78 (95% CI, 0.59–1.21), and it was slightly worse than that of overall survival. In some phase III trials conducted for HCC, significant difference was observed in the time to progression or progression-free survival, but not in the overall survival [1]. Eventually, a negative result was concluded. However, in this study, the results were completely opposite. The most important difference between this study and these aforementioned trials may be in the anticancer treatments used: in this study, sorafenib was combined with a cytotoxic agent, while in the aforementioned phase III trials, it was used in combination with other molecular-targeted agents. Among patients showing marked tumor shrinkage on account of the favorable tumor shrinkage effect of SorCDDP, even a slight increase in the tumor size could result in their being classified as showing disease progression, whereas these patients may also show a prolonged overall survival because of the smaller tumor burden. This might also be the reason for the more favorable improvement of the overall survival than the time to progression.

increase in the tumor size could result in their being classified as showing disease progression, whereas these patients may also show a prolonged overall survival because of the smaller tumor burden. This might also be the reason for the more favorable improvement of the overall survival than the time to progression. The frequencies of the adverse events in the SorCDDP arm, except for those of neutropenia, leukocytopenia, hypohemoglobinemia, thrombocytopenia, hyponatremia, nausea and hiccups, were similar to those in the Sor arm. These adverse events were not severe. HAIC with cisplatin had only a mild toxicity profile [4] and the toxicities were not overlapped with the adverse effects of sorafenib. Therefore, SorCDDP therapy was also considered to be well-tolerated. Intra-arterial administration of cisplatin was generally thought to be troublesome, requiring the insertion of a catheter into the tumor-feeding arteries. Recently, a phase III trial of sorafenib plus intra-arterial cisplatin and 5-fluorouracil versus sorafenib alone demonstrated no survival benefit [15]. One of the reasons could be the difficulty in placing the indwelling reservoir system. However, cisplatin is easily administered without the need for an indwelling reservoir system. Furthermore, this combined treatment is medico-economically very viable, because the additional cost of the angiographic procedure and cisplatin is approximately $2000 per session, which is less than the cost of the recently administered molecular-targeted agents or immuno-oncology agents.

indwelling reservoir system. Furthermore, this combined treatment is medico-economically very viable, because the additional cost of the angiographic procedure and cisplatin is approximately $2000 per session, which is less than the cost of the recently administered molecular-targeted agents or immuno-oncology agents. In conclusion, this study demonstrated favorable overall survival in the SorCDDP arm when compared with that in the Sor arm in patients with advanced HCC, suggesting the effectiveness of HAIC against advanced HCC. However, since this study was only a randomized phase II trial, we could not arrive at any definitive conclusion with regard to the usefulness of sorafenib plus HAIC with cisplatin in the treatment of advanced HCC. A further phase III trial is being planned to confirm these results. funding This work was supported in part by the National Cancer Center Research and Development Fund (23-A-22).

In conclusion, this study demonstrated favorable overall survival in the SorCDDP arm when compared with that in the Sor arm in patients with advanced HCC, suggesting the effectiveness of HAIC against advanced HCC. However, since this study was only a randomized phase II trial, we could not arrive at any definitive conclusion with regard to the usefulness of sorafenib plus HAIC with cisplatin in the treatment of advanced HCC. A further phase III trial is being planned to confirm these results. funding This work was supported in part by the National Cancer Center Research and Development Fund (23-A-22). disclosure MI has received payment for lectures from Bayer Yakuhin and Nippon Kayaku; has been a consultant for Bayer Yakuhin; and has received research funding from Bayer Yakuhin. SS has received payment for lectures from Nippon Kayaku. YI has received payment for lectures from Bayer Yakuhin and Nippon Kayaku; and has received research funding from Bayer Yakuhin and Nippon Kayaku. AH has been on the speakers' bureau for Bayer Yakuhin. TO has received payment for lectures from Bayer Yakuhin. JF has received payment for lectures from Bayer Yakuhin and Nippon Kayaku; has been on advisory arrangements for Bayer Yakuhin; and received research funding from Bayer Yakuhin and Nippon Kayaku. TO has received research funding from Bayer Yakuhin and has received payment for lectures from Nippon Kayaku. All remaining authors have declared no conflicts of interest.

Yakuhin and Nippon Kayaku; has been on advisory arrangements for Bayer Yakuhin; and received research funding from Bayer Yakuhin and Nippon Kayaku. TO has received research funding from Bayer Yakuhin and has received payment for lectures from Nippon Kayaku. All remaining authors have declared no conflicts of interest. Supplementary Material Supplementary Data acknowledgements We thank all the patients and their families for their participation in this study. We thank the members of the Japan Clinical Research Support Unit Data Center for their support with the data management (Ms Youko Yoshimoto, Ms Rei Aida, and Mr Takashi Ando), the members of the study committee for safety and efficacy monitoring (Dr Yasuhiro Matsumura, Dr Kei Muro, and Dr Keigo Osuga), the members of the study committee for response evaluation (Dr Tatsushi Kobayashi, Dr Nobumasa Mizuno, and Dr Eiichiro Suzuki), and all the investigators of this trial.

Introduction In the emerging era of precision medicine, genomic analysis has become an integral component of the diagnostic work-up of cancer patients. Where initially DNA sequencing approaches tested individual cancer ‘hotspot’ loci (e.g. KRAS mutational status in colorectal cancer; EGFR mutational status in lung cancer), a more precise understanding of the biological basis of malignancy subsequently led to identification and deployment of specific ‘cancer gene panels’ as prognostic or treatment prediction tools. Additionally, the increased interrogative capacity afforded by next generation sequencing (NGS), allied to its decreasing cost, has empowered many institutions worldwide to perform whole exome sequencing (WES) or whole genome sequencing (WGS) on significant numbers of tumour samples. Primary data outputs from these initiatives are increasing exponentially, thus challenging scientific and clinical communities to develop workable solutions for optimal analysis, usage and storage of these datasets. Further complexity is introduced by the need to integrate this genomic data with associated clinical information.

ata outputs from these initiatives are increasing exponentially, thus challenging scientific and clinical communities to develop workable solutions for optimal analysis, usage and storage of these datasets. Further complexity is introduced by the need to integrate this genomic data with associated clinical information. Previously, on behalf of the Clinical Working Group of the Global Alliance for Genomics and Health (GA4GH) (a coalition of researchers, clinicians, patient advocates and life sciences/information technology industries dedicated to implementing worldwide data sharing solutions), we have highlighted the data challenges in cancer genomics [1], emphasized the currently siloed nature of the clinical, pathological and genomic datasets and proposed a blueprint solution that is predicated on a culture of responsible data sharing [2]. However, there is a lack of collective intelligence on current practices in cancer clinical sample sequencing initiatives worldwide. There is a paucity of information on the types of technical NGS platforms/parameters employed and choice of bioinformatics algorithms for analysis. Uniform approaches for collecting matched clinical and-genomic data on outcomes and treatment toxicities are lacking [3]. Information is limited on both institutional enthusiasm for sharing their data and the technical ability to facilitate a data sharing culture. Costs and resources required to establish multi-institutional/international data sharing programs are considerable. From ethical and legal perspectives, data protection legislation/privacy concerns are also challenging, particularly as they can vary significantly according to geographic region [4]. These issues pose significant challenges for effective data harmonization and sharing. Thus, a detailed assessment of the current global cancer clinical sample sequencing landscape is required to inform and enhance present and future data sharing efforts.

they can vary significantly according to geographic region [4]. These issues pose significant challenges for effective data harmonization and sharing. Thus, a detailed assessment of the current global cancer clinical sample sequencing landscape is required to inform and enhance present and future data sharing efforts. Recognizing these information deficits, we performed an international survey of cancer clinical sample sequencing initiatives. This survey was designed to provide an informative snapshot of current activities worldwide and identify potential barriers that may limit data sharing activities, thus informing creation of a global informatics ecosystem that facilitates the sharing of clinical and genomic cancer data at scale. Methods Recruitment of respondents, survey development and data collection Methodology for respondent recruitment, survey development and data collection is outlined in the supplementary Appendices S1 and S2, available at Annals of Oncology online.

Recognizing these information deficits, we performed an international survey of cancer clinical sample sequencing initiatives. This survey was designed to provide an informative snapshot of current activities worldwide and identify potential barriers that may limit data sharing activities, thus informing creation of a global informatics ecosystem that facilitates the sharing of clinical and genomic cancer data at scale. Methods Recruitment of respondents, survey development and data collection Methodology for respondent recruitment, survey development and data collection is outlined in the supplementary Appendices S1 and S2, available at Annals of Oncology online. Statistical analysis Data collected from Google Forms were exported to the R statistical package for analysis. Descriptive statistics were used to summarize survey responses. All analyses were performed using χ2 testing unless otherwise indicated. Likert scales were used to capture the extent of perceived barriers to data sharing (1 = minor barrier, 6 = major barrier). Given that not all questions were mandatory, sample size varied according to the particular question; thus responses have been displayed with the numerator (n) and denominator (N) (largest possible number of available responses). The denominator is reported for each section once, unless it changes.

6 = major barrier). Given that not all questions were mandatory, sample size varied according to the particular question; thus responses have been displayed with the numerator (n) and denominator (N) (largest possible number of available responses). The denominator is reported for each section once, unless it changes. Results The survey collected responses from July to October 2015. Out of the 107 initiatives invited, 59 responses were received (response rate = 55%). Of the non-responders, 9 initiatives indicated that their activities did not match the survey’s scope or had already been captured in our survey, thus giving a true response rate of 60%. Of the remaining non-responders, the majority resided in the US (n = 23) and Australia (n = 8). None of the Chinese initiatives responded (n = 3). Survey completion rates varied across sections, ranging from 81% [Privacy and Ethics (n = 48, N = 59)] to 88% [Barriers (n = 52, N = 59)]. Completed surveys included respondents from diverse locations and initiative size, with the majority coming from North America and Europe (supplementary Tables S1, S3 and Figure S1, available at Annals of Oncology online). The primary intention of the initiatives were: research [37% (n = 22)], clinical diagnostic [15% (n = 9)], combination [34% (n = 20)] and unknown [14% (n = 8)] as self-nominated by the individual initiative (supplementary Table S1, available at Annals of Oncology online). Relevant inter-institutional initiatives in Eastern Europe, Africa or India were not identified.

search [37% (n = 22)], clinical diagnostic [15% (n = 9)], combination [34% (n = 20)] and unknown [14% (n = 8)] as self-nominated by the individual initiative (supplementary Table S1, available at Annals of Oncology online). Relevant inter-institutional initiatives in Eastern Europe, Africa or India were not identified. Sequencing Platforms Wide variation in the type of sequencing platforms employed was observed. WES was the most frequently used (n = 28, N = 51, 55%) while WGS was also employed in a high number of initiatives (n = 22, 43%). A total of 35% (n = 18) used both WES/WGS, whereas 37% (n = 19) employed neither platform.

search [37% (n = 22)], clinical diagnostic [15% (n = 9)], combination [34% (n = 20)] and unknown [14% (n = 8)] as self-nominated by the individual initiative (supplementary Table S1, available at Annals of Oncology online). Relevant inter-institutional initiatives in Eastern Europe, Africa or India were not identified. Sequencing Platforms Wide variation in the type of sequencing platforms employed was observed. WES was the most frequently used (n = 28, N = 51, 55%) while WGS was also employed in a high number of initiatives (n = 22, 43%). A total of 35% (n = 18) used both WES/WGS, whereas 37% (n = 19) employed neither platform. Gene-panels were frequently used, with 55% and 51% of responding initiatives indicating that they employ a gene-panel with 51–250 genes (n = 28) or 251–1000 genes (n = 26), respectively. Gene-panels of fewer than 50 genes were also commonly utilized (n = 23, 45%). Very large gene-panels (1001–5000 genes) were used rarely (n = 7, 14%) (Table 1). WES/WGS was employed less frequently in clinical diagnostic initiatives compared with research initiatives (WES: n = 2, N = 9, P = 0.03; WGS n = 0, N = 9, P < 0.01). RNAseq/other transcriptomics techniques were employed in 59% (n = 30) and 41% (n = 21), of initiatives, respectively. The use of germline sequencing (as a filter to distinguish somatic from germline single nucleotide polymorphisms) was included in 59% of initiatives (n = 30, N = 51). Its use was associated with sequencing intent (P = 0.02), with only 22% of the Clinical Diagnostics using it reflecting their increased use of small (hotspot) panels. Table 1 Responses to the technical aspects of the survey

m germline single nucleotide polymorphisms) was included in 59% of initiatives (n = 30, N = 51). Its use was associated with sequencing intent (P = 0.02), with only 22% of the Clinical Diagnostics using it reflecting their increased use of small (hotspot) panels. Table 1 Responses to the technical aspects of the survey Sequencing platforms n % Diagnostic N = 9 Research N = 22 Diagnostic/research, N = 20 P-valuea Panel size  (genes) Small < 50 23 45 6  (67%) 7  (32%) 10  (50%) 0.17 Medium 51–250 28 55 5  (56%) 14  (64%) 9  (45%) 0.48 Large 251–1000 26 51 4  (44%) 11  (50%) 11  (55%) 0.86 Very large 1001–5000 7 14 0  (0%) 4  (18%) 3  (15%) 0.40 WES 28 55 2  (22%) 16  (73%) 10  (50%) 0.03 WGS 22 43 0  (0%) 14  (64%) 8  (40%) 0.01 RNAseq 30 59 3  (33%) 16  (73%) 11  (55%) 0.12 Transcriptomics 21 41 3  (33%) 13  (59%) 5  (25%) 0.07 Sequencing depth <25 1 2 0  (0%) 0  (0%) 1  (5%) 0.37 25–50 2 4 0  (0%) 0  (0%) 2  (10%) 0.15 51–100 15 29 0  (0%) 7  (32%) 8  (40%) 0.02 101–250 10 20 0  (0%) 5  (23%) 5  (25%) 0.08 251–1000 20 39 9  (100%) 8  (36%) 3  (15%) 0.21 >1000 3 6 0  (0%) 2  (9%) 1  (5%) 0.37 Certification ISO 11 22 2  (22%) 4  (18%) 5  (25%) 0.53 CLIA 18 35 5  (56%) 5  (23%) 8  (40%) 0.61 NEN/similar 15b 27 4  (44%) 5  (23%) 5  (25%) 0.93 None 20b 39 1  (5%) 12  (60%) 6  (30%) 0.01 a P-value represents χ2 testing comparisons between the intent of the particular initiatives. b One initiative did not indicate their intent. ISO, international organization for standardization; CLIA, clinical laboratory improvement amendments; NEN, Netherlands Standardization Institute.

Sequencing platforms n % Diagnostic N = 9 Research N = 22 Diagnostic/research, N = 20 P-valuea Panel size  (genes) Small < 50 23 45 6  (67%) 7  (32%) 10  (50%) 0.17 Medium 51–250 28 55 5  (56%) 14  (64%) 9  (45%) 0.48 Large 251–1000 26 51 4  (44%) 11  (50%) 11  (55%) 0.86 Very large 1001–5000 7 14 0  (0%) 4  (18%) 3  (15%) 0.40 WES 28 55 2  (22%) 16  (73%) 10  (50%) 0.03 WGS 22 43 0  (0%) 14  (64%) 8  (40%) 0.01 RNAseq 30 59 3  (33%) 16  (73%) 11  (55%) 0.12 Transcriptomics 21 41 3  (33%) 13  (59%) 5  (25%) 0.07 Sequencing depth <25 1 2 0  (0%) 0  (0%) 1  (5%) 0.37 25–50 2 4 0  (0%) 0  (0%) 2  (10%) 0.15 51–100 15 29 0  (0%) 7  (32%) 8  (40%) 0.02 101–250 10 20 0  (0%) 5  (23%) 5  (25%) 0.08 251–1000 20 39 9  (100%) 8  (36%) 3  (15%) 0.21 >1000 3 6 0  (0%) 2  (9%) 1  (5%) 0.37 Certification ISO 11 22 2  (22%) 4  (18%) 5  (25%) 0.53 CLIA 18 35 5  (56%) 5  (23%) 8  (40%) 0.61 NEN/similar 15b 27 4  (44%) 5  (23%) 5  (25%) 0.93 None 20b 39 1  (5%) 12  (60%) 6  (30%) 0.01 a P-value represents χ2 testing comparisons between the intent of the particular initiatives. b One initiative did not indicate their intent. ISO, international organization for standardization; CLIA, clinical laboratory improvement amendments; NEN, Netherlands Standardization Institute. Sequencing read depth Questions concerning sequencing read depth employed were completed by 86% of initiatives (n = 51, N = 59) (Table 1). Median reported tumour-sequencing depth was 101–250×, with one initiative (2%) indicating depths lower than 25×, whilst three initiatives (6%) indicated depths greater than 1000×. Clinical diagnostic initiatives (n = 9) all reported use of greater sequencing read depth (between 251 and 1000×) compared with research-based initiatives (P = 0.01). Of the research initiatives, 10 reported read depths greater than 251 (n = 10, N = 22, 46%), while four of the combined initiatives reported such depth (n = 4, N = 20, 20%).

initiatives (n = 9) all reported use of greater sequencing read depth (between 251 and 1000×) compared with research-based initiatives (P = 0.01). Of the research initiatives, 10 reported read depths greater than 251 (n = 10, N = 22, 46%), while four of the combined initiatives reported such depth (n = 4, N = 20, 20%). Nucleic acid and protein extraction The majority of initiatives (n = 29, N = 52, 56%) performed sequencing analysis from formalin fixed paraffin embedded (FFPE) and fresh frozen (FF) samples, whereas 27% (n = 14) only employed FFPE as source material and 17% (n = 9) only tested FF samples. Of the initiatives performing the extractions in-house (N = 44), the majority indicated extraction of DNA and RNA from the same sample (n = 34, 77%); protein extraction was relatively uncommon (n = 8, 18%). Laboratory certification/accreditation The majority of Clinical Diagnostic initiatives (95%) held laboratory certification/accreditation compared with research (40%) and combination (70%) initiatives (P = 0.01) (Table 1). The most common certification was Clinical Laboratory Improvement Amendments (CLIA) (n = 18, N = 51, 35%).

ry certification/accreditation The majority of Clinical Diagnostic initiatives (95%) held laboratory certification/accreditation compared with research (40%) and combination (70%) initiatives (P = 0.01) (Table 1). The most common certification was Clinical Laboratory Improvement Amendments (CLIA) (n = 18, N = 51, 35%). Bioinformatics tools Mutation calling The most commonly reported bioinformatics tools were GATK (n = 27, N = 51, 53%), Samtools (n = 25, 49%), VarScan2 (n = 23, 46%), and Mutect (n = 20, 39%) (supplementary Appendix S3, available at Annals of Oncology online). The frequency with which these tools were employed is shown in Figure 1A. Logistic regression analysis addressing the type of data employed (prospective, retrospective, combination), indicated that GATK is less likely to be used in a prospective study (P = 0.02). No other significant relationships concerning data type, geographic location or size were identified. Figure 1. Venn diagrams demonstrating the frequency of bioinformatics pipelines used either in isolation or in combination. Representative images of mutations callers (A) and variant annotation (B) reveal significant heterogeneity. Numbers are expressed as percentages. Variant annotation Variant annotation was most commonly performed using COSMIC (n = 37, N = 52, 71%). PolyPhen2 (n = 33, 63%), dbSNP (n = 33, 63%), and SIFT (n = 29, 56%) were also frequently used (supplementary Appendix S3, available at Annals of Oncology online). The frequency at which these methods were used (in isolation or in combination) is shown in Figure 1B.

ommonly performed using COSMIC (n = 37, N = 52, 71%). PolyPhen2 (n = 33, 63%), dbSNP (n = 33, 63%), and SIFT (n = 29, 56%) were also frequently used (supplementary Appendix S3, available at Annals of Oncology online). The frequency at which these methods were used (in isolation or in combination) is shown in Figure 1B. Copy number alterations Of the respondents, 85% (n = 44, N = 52) indicated that they estimated copy number alterations (CNAs) from their sequencing data, while 10% indicated not doing so. One initiative reported inference of CNA from targeted panel data. Versioning of pipelines The majority of initiatives indicated keeping records of which version of their computational procedures (also referred to as software pipelines) to analyse sequencing data that they employed (n = 44, N = 52, 85%). Seven initiatives (13%) indicated that they were unsure as to whether the versions of their pipelines were tracked and one initiative (2%) did not track pipeline versions.

sion of their computational procedures (also referred to as software pipelines) to analyse sequencing data that they employed (n = 44, N = 52, 85%). Seven initiatives (13%) indicated that they were unsure as to whether the versions of their pipelines were tracked and one initiative (2%) did not track pipeline versions. Clinical parameters Merged clinical and genomic data Of responding initiatives, 47 of 51 (92%, P < 0.01) attempted to link clinical information to genomic data. No differences in the initiatives’ intent (clinical diagnostic versus research versus combination) and linking of clinical and genomic data were identified (supplementary Table S3, available at Annals of Oncology online). Data extraction exclusively employing manual extraction of records was most commonly utilized (n = 23, 45%). Direct deposition of electronic health records (n = 9, 18%, versus manual extraction P = 0.01), a combination of manual and direct deposition (n = 9, 18%, versus manual extraction P = 0.01) and other approaches (e.g. in-house direct hospital data feeds) were less frequently utilized (n = 10, 20%, versus manual extraction P = 0.02) (supplementary Table S3, available at Annals of Oncology online). The majority of initiatives used a customized case report form for data collection (n = 34, N = 47, 72%).

0.01) and other approaches (e.g. in-house direct hospital data feeds) were less frequently utilized (n = 10, 20%, versus manual extraction P = 0.02) (supplementary Table S3, available at Annals of Oncology online). The majority of initiatives used a customized case report form for data collection (n = 34, N = 47, 72%). Genotype-drug matching Of the 51 responding initiatives, 39 (76%) were undertaking genomic-based patient-drug matching for subsequent clinical intervention. Of these, 77% were opportunistically matched (n = 30, N = 39) (e.g., phase I studies, off-label use). Nine (23%) initiatives reported that they performed clinical sample sequencing as part of biomarker-driven clinical trials. For measuring treatment efficacy, almost half of responding initiatives used combinations of efficacy endpoints (n = 25, N = 51, 49%). The most common endpoints were time on treatment (n = 29, 57%) and response evaluation criteria in solid tumours (RECIST) (n = 27, 53%); however, other parameters such as clinical assessment (n = 14, 27%) were also utilized. Toxicity data were collected in the majority of responding initiatives (n = 28, N = 48, 59%).

. The most common endpoints were time on treatment (n = 29, 57%) and response evaluation criteria in solid tumours (RECIST) (n = 27, 53%); however, other parameters such as clinical assessment (n = 14, 27%) were also utilized. Toxicity data were collected in the majority of responding initiatives (n = 28, N = 48, 59%). Privacy and ethics Written consent was obtained in 34 initiatives (N = 48, 71%), seven initiatives had implied consent/consent waivers (15%). The majority (n = 36, 75%) of initiatives allowed re-contacting of patients for follow-up information. A protocol for communicating somatic genetic results was in place in the majority of initiatives (n = 32, 67%) and a trend to association with the initiative’s intent (clinical diagnostic, research or combination) was identified (P = 0.06). A policy for incidental germline findings was in place in 23 of initiatives (48%), but no association was identified with the initiative’s intent (P = 0.55). Data warehousing Most of the respondents (n = 43, N = 50, 86%) indicated that their data storage/warehouse was centralized, for mutation data (n = 47, 94%), copy number alteration estimates (n = 45, 90%), clinical data (n = 42, 84%) and sequencing (BAM) files (n = 41, 82%). Histological data was the least likely to be stored centrally (n = 38, 76%).

spondents (n = 43, N = 50, 86%) indicated that their data storage/warehouse was centralized, for mutation data (n = 47, 94%), copy number alteration estimates (n = 45, 90%), clinical data (n = 42, 84%) and sequencing (BAM) files (n = 41, 82%). Histological data was the least likely to be stored centrally (n = 38, 76%). Data sharing The majority of respondents (n = 36, N = 50, 72%, P < 0.01) indicated that they allow sharing of their data (supplementary Figure S2, available at Annals of Oncology online). Fourteen percent indicated not intending to share, while another 14% indicated that they are in the process of developing data sharing policies. No association was identified between data sharing and purpose of the initiative (P = 0.14). Data sharing typically came with a varied set of restrictions such as regional legislation (e.g. European data that cannot leave the Eurozone, intellectual property (IP) concerns and material transfer agreement restrictions). Certain initiatives remarked that there were significant limitations in transferring raw data between institutions.

y came with a varied set of restrictions such as regional legislation (e.g. European data that cannot leave the Eurozone, intellectual property (IP) concerns and material transfer agreement restrictions). Certain initiatives remarked that there were significant limitations in transferring raw data between institutions. Perceived barriers The greatest barriers identified (defined as responses >4 on the Likert scale, N = 52) were: financial support for data sharing (77%, P < 0.01), bioinformatics concerns such as lack of conformity and interoperability of bioinformatics pipelines (69%, P = 0.02), and clinical data capture (60%, P = 0.19) (supplementary Figure S2, available at Annals of Oncology online). Initiatives with 1000 or more patients were more likely to perceive clinical data capture as a barrier compared with smaller initiatives (P < 0.01). Lack of expertise in the context of rapidly evolving technology (50%) and legal issues (37%) were also raised as potential barriers, whereas privacy/ethics issues (35%) and international legislation (33%) were not considered significant barriers. Of note, bioinformatics and financial concerns did not differ between size of initiative or whether the initiative was clinical diagnostic, research or combination. The free-text commentary of perceived barriers is shown in supplementary Table S4, available at Annals of Oncology online.

considered significant barriers. Of note, bioinformatics and financial concerns did not differ between size of initiative or whether the initiative was clinical diagnostic, research or combination. The free-text commentary of perceived barriers is shown in supplementary Table S4, available at Annals of Oncology online. Funding The most frequent source of funding was governmental (n = 11, N = 50, 22%), charities (n = 9, 18%), government combined with academic/professional (e.g. AACR/ASCO) sources or charities (n = 5, N = 50, 10%), industry (n = 4, 8%), charities combined with industry (n = 3, 6%) and academic/professional societies (n = 2, 4%). The remainder (n = 16, 32%) consisted of hybrid combinations with a frequency of one. Concerns were expressed in relation to funding of international data sharing initiatives (supplementary Table S4, available at Annals of Oncology online).

ined with industry (n = 3, 6%) and academic/professional societies (n = 2, 4%). The remainder (n = 16, 32%) consisted of hybrid combinations with a frequency of one. Concerns were expressed in relation to funding of international data sharing initiatives (supplementary Table S4, available at Annals of Oncology online). Discussion Molecular technologies such as NGS have revolutionized cancer biology discovery. Their successful clinical application depends on the sequencing platform and its robustness, the associated bioinformatics pipeline(s), and the availability of clinically annotated data from patients undergoing therapeutic interventions. Linking clinical and genomic data can justify molecular stratification of patients to specific interventions, but there is a realization that matched data must be available from sufficient numbers of patients to allow statistically robust, clinically meaningful conclusions to be drawn. Collaborative sharing of this information between initiatives increases the value and relevance of the data, for the scientist, the pharmaceutical industry, the clinician, the payer (insurance/taxpayer) and ultimately for the patient. However, effective data sharing is challenging, from technical, clinical, ethical, logistical and regulatory perspectives.

tion between initiatives increases the value and relevance of the data, for the scientist, the pharmaceutical industry, the clinician, the payer (insurance/taxpayer) and ultimately for the patient. However, effective data sharing is challenging, from technical, clinical, ethical, logistical and regulatory perspectives. From a technical perspective, respondents to the survey employed a number of sequencing platforms and methodologies. Of these platforms, WES (P = 0.03) and WGS (P < 0.01) were more relevant to research application, with low adoption rates in clinical diagnostic initiatives. Conversely, clinical diagnostic initiatives employed greater sequencing depths than research initiatives (P = 0.012). Surprisingly, nearly 40% of initiatives surveyed did not have clinical molecular diagnostic laboratory certification/accreditation, highlighting a deficiency that must be addressed in order for NGS to be routinely incorporated into mainstream clinical diagnostics.

equencing depths than research initiatives (P = 0.012). Surprisingly, nearly 40% of initiatives surveyed did not have clinical molecular diagnostic laboratory certification/accreditation, highlighting a deficiency that must be addressed in order for NGS to be routinely incorporated into mainstream clinical diagnostics. A key finding was the heterogeneity in variant/mutation calling and variant-annotation tools. Use of a single variant caller was rare and tended to involve products from the sequencing vendor or bespoke in-house algorithms. However, the employment of a suite of variant callers was the preferred approach. This heterogeneity in pipelines compromises the ability to compare results between different clinical sample sequencing initiatives [5]. Efforts to address this lack of harmonization are ongoing, through initiatives such as NCI’s Genome Data Commons [6] and the Somatic Mutation Calling Challenge (SMCC) [7]. The recent development of the GA4GH Application Programming Interface (API) [8] provides an easy-to-use web-based algorithm for improved identification of mutations and rearrangements in sequencing data, and is gaining traction in the translational bioinformatics community.

omatic Mutation Calling Challenge (SMCC) [7]. The recent development of the GA4GH Application Programming Interface (API) [8] provides an easy-to-use web-based algorithm for improved identification of mutations and rearrangements in sequencing data, and is gaining traction in the translational bioinformatics community. Over 90% of respondents indicated that they had mechanisms in place to capture linked clinical and genomic data. However, uniformity was lacking for the collection and aggregation of this information. While the majority of institutes employed electronic case report forms, nearly half of the initiatives surveyed were manually extracting clinical data. In order to address this, initiatives such as the ASCO’s CancerLinQ project are developing custom-built electronic feeds from community oncology practices to maximize collection of clinical data [9]. A second challenge relates to the quality of the clinical data collected. Incomplete data sets reduce the value of the information collected, while lack of a cancer specific ontology compromises the ability to aggregate and compare clinical and genomic data from different sources. Building a cancer specific Human Phenotype Ontology (which has been an invaluable asset to the rare diseases community) [10], would significantly enhance phenotype–genotype correlations in the study of malignancy.

ntology compromises the ability to aggregate and compare clinical and genomic data from different sources. Building a cancer specific Human Phenotype Ontology (which has been an invaluable asset to the rare diseases community) [10], would significantly enhance phenotype–genotype correlations in the study of malignancy. This survey also highlighted that longitudinal outcome/toxicity data are not captured through a standardized approach outside of clinical trials. Facilitating routine access to these data (e.g. through development of a minimum dataset) is necessary, in order to maximize the collective learning that can be achieved by aggregating clinical/genomic data, especially when analysing rare variants. It was encouraging that 75% of initiatives indicated that their protocol included the permission to re-contact patients, emphasizing the importance that clinical cancer sample sequencing initiatives place on the capture of follow-up patient outcome and toxicity data.

nomic data, especially when analysing rare variants. It was encouraging that 75% of initiatives indicated that their protocol included the permission to re-contact patients, emphasizing the importance that clinical cancer sample sequencing initiatives place on the capture of follow-up patient outcome and toxicity data. Over 70% of initiatives were in favor of sharing clinical and genomic data. However, a more detailed evaluation of both quantitative and qualitative responses revealed a number of barriers that exist and must be addressed. Lack of dedicated funding was perceived as the most significant barrier to data sharing activities. Collection of even a minimum clinical dataset has major human and technical resource requirements, leading to significant costs. Dedicated funding streams that actively promote data sharing should be encouraged. In this regard, the recent launch of the Innovative Medicines Initiative Big Data for Better Outcomes [11] incentivizes both the scientific and pharmaceutical communities to work together in large-scale data sharing activities. The second most commonly highlighted perceived barrier was lack of interoperability of bioinformatics pipelines, and we have already highlighted how initiatives/activities such as GDC [6], SMCC [7] and the GA4GH API [8] are addressing this challenge.

communities to work together in large-scale data sharing activities. The second most commonly highlighted perceived barrier was lack of interoperability of bioinformatics pipelines, and we have already highlighted how initiatives/activities such as GDC [6], SMCC [7] and the GA4GH API [8] are addressing this challenge. Issues with consent and data privacy were also raised in the free text narrative, with concerns relating to data protection legislation barriers in particular regions e.g. Europe, and harmonization of consent procedures. It is hoped that the recent decision of the European Commission on the EU-US Privacy Shield will help address inter-continental data privacy issues [12]. These regulatory challenges limit the effectiveness of global cancer knowledge networking. From an ethics perspective, ethics harmonization has been a key theme of GA4GH’s Framework for Responsible Sharing of Genomic and Health-Related Data [13] that we suggest should serve as an overarching ethical framework for clinical and genomic data sharing. Additionally, introduction of a federated data sharing approach, where data does not leave the particular legal jurisdiction but can be mined efficiently in situ, represents a potential solution for regions that are sensitive to primary data transfer. Concerns were also raised in relation to how data sharing might adversely affect publications and IP issues. Improving the quality of publications through effective data sharing, and developing micro-attribution based rewards where the work of data providers is acknowledged [14] should help allay these fears.

transfer. Concerns were also raised in relation to how data sharing might adversely affect publications and IP issues. Improving the quality of publications through effective data sharing, and developing micro-attribution based rewards where the work of data providers is acknowledged [14] should help allay these fears. The benefits of data sharing become increasingly relevant as we collectively realize that our current catalogue of actionable cancer mutations is limited, and even there, consensus is lacking. Molecular stratification approaches have identified distinct disease subtypes, some of which may be relatively rare. Thus, a collective approach employing information from data repositories worldwide is increasingly required to identify/verify relevant mutations that can inform improved diagnosis or identify novel targets. Such an approach has already been employed by GA4GH in the BRCA challenge [15], which convened BRCA experts from around the world to work together to share BRCA variants publicly, thus allowing expert review of variant interpretations to determine the pathogenicity of an increased number of variants in the BRCA1/BRCA2 genes. This work has resulted in BRCA exchange [16], a curated web portal that allows the BRCA community to query the current evidence of any BRCA1/2 variant present in the aggregated dataset. Extending the BRCA Challenge approach to other genes and cancers would allow a more granular understanding of variant relevance, thereby informing clinical actionability.

nge [16], a curated web portal that allows the BRCA community to query the current evidence of any BRCA1/2 variant present in the aggregated dataset. Extending the BRCA Challenge approach to other genes and cancers would allow a more granular understanding of variant relevance, thereby informing clinical actionability. We acknowledge that this study has several limitations. By its nature, it is a snapshot at a particular moment in time, in a rapidly advancing field. While our aspiration was to capture responses from cancer sample sequencing initiatives worldwide, there is an enrichment towards initiatives in North America and Europe, due to a combination of an inability to identify cancer clinical sample sequencing collaborative initiatives and/or a lack of response from such initiatives in particular countries/regions (e.g. India, China). Nonetheless, this survey is a first attempt to catalogue cancer clinical sample sequencing activity worldwide and represents a useful benchmark to inform cancer data sharing activities going forward.

orative initiatives and/or a lack of response from such initiatives in particular countries/regions (e.g. India, China). Nonetheless, this survey is a first attempt to catalogue cancer clinical sample sequencing activity worldwide and represents a useful benchmark to inform cancer data sharing activities going forward. Conclusions This is the first comprehensive global survey of cancer clinical sample sequencing initiatives. It provides an evidence-based perspective informed by responses from experts worldwide concerning the key barriers to data sharing. It emphasizes the need to break down individual data silos and underscores the requirement to provide robust approaches for clinical and genomic data collection and analysis. It highlights how limited dedicated funding, a dearth of standardized methodologies and a lack of thoughtful integration are hampering clinically relevant data sharing efforts. Developing a bioinformatics ecosystem that delivers open source interoperable solutions to overcome the barriers we highlight, would maximize the potential for responsible but effective sharing of clinical and genomic data for the benefit of cancer patients globally. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements This international survey is conducted on behalf of the GA4GH Clinical Working Group.

Conclusions This is the first comprehensive global survey of cancer clinical sample sequencing initiatives. It provides an evidence-based perspective informed by responses from experts worldwide concerning the key barriers to data sharing. It emphasizes the need to break down individual data silos and underscores the requirement to provide robust approaches for clinical and genomic data collection and analysis. It highlights how limited dedicated funding, a dearth of standardized methodologies and a lack of thoughtful integration are hampering clinically relevant data sharing efforts. Developing a bioinformatics ecosystem that delivers open source interoperable solutions to overcome the barriers we highlight, would maximize the potential for responsible but effective sharing of clinical and genomic data for the benefit of cancer patients globally. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements This international survey is conducted on behalf of the GA4GH Clinical Working Group. We thank all the respondents for their time and efforts to enable completion of this survey project: Dara L. Aisner (University of Colorado); Fabrice Andre (SAFIR); Wilson Araujo da Silva Junior (University of Sao Paulo); Philippe Bedard (IMPACT/COMPACT, Canada); Melissa Brammer (MyPathway); James D. Brenton (University of Cambridge); Benedikt Brors (DKFZ); Fabien Calvo (Cancer Core Europe); Paulo V Campregher (Hospital Israelita Albert Einstein); Mark Caulfield (Genomics England); George Chong (Jewish General Hospital); Wee Joo Chng (National University Cancer Institute, Singapore); Barbara Conley (NCI-Match); Christopher Corless (Oregon Health & Science University); William S. Dalton (ORIEN); Emmanuel Dias-Neto (AC Camargo Cancer Center); Stephen Fox (Cancer 2015); Laura Esserman (I-SPY); Tom Hudson (ICGC); Katherine Janeway (GAIN/iCAT); Ludovic Lacroix (Institut Gustave Roussy); Janessa Laskin (Personalized OncoGenomics); Vladimir Lazar (WIN Consortium); Christophe Le Tourneau (SHIVA); Subha Madhavan (COTA, G-DOC); Tim Maughan (FOCUS 4); Funda Meric-Bernstam (Clearinghouse, MD Anderson Precision Oncology Decision Support Core); Olena Morozova (Treehouse Childhood Cancer Project); Petra Nederlof (NKI); Keunchil Park (Korea Actionable Genome Consortium, NEXT-1); Peta Phillips (AGHA); Sun Young RHA (Genomic Initiative of Gastric Cancer Treatment); Richard Schilsky (CANCERLINQ, TAPUR); Richard Sinnott (University of Melbourne); Tatsuhiro Shibata (University of Tokyo); David Solit (MSK-IMPACT); Richie Soong (Centre for Translational Research and Diagnostics); Jean Charles Soria (SAFIR02); Amanda Spurdle (BRCA Exchange, ENGIMA); Dominiek Staelens (CREATE); Shawn M Sweeney (GENIE); Edit Szepessy (EORTC-SPECTA); Daniel Tan (IMPACT, POLARIS, Singapore); Ian Tomlinson (S-CORT); Anthony Tolcher (The San Antonio 1000 Cancer Genome Project);Vanessa Tyrrell (Children's Cancer Institute); Ana Vivancos (Val d’Hebron); Michael Watson (Clinical Genome Resource); Wendy Winckler (Novartis Institutes for Biomedical Research); Valtteri Wirta (Karolinska Institute); Jean Claude Zenklusen (TCGA, Cancer Driver Discovery Program, Clinical Trial Sequencing Program).

t);Vanessa Tyrrell (Children's Cancer Institute); Ana Vivancos (Val d’Hebron); Michael Watson (Clinical Genome Resource); Wendy Winckler (Novartis Institutes for Biomedical Research); Valtteri Wirta (Karolinska Institute); Jean Claude Zenklusen (TCGA, Cancer Driver Discovery Program, Clinical Trial Sequencing Program). Key Message In this first global study of cancer sequencing initiatives, we highlight the main barriers to effective data sharing and th significant variability in data capture processes. Addressing these issues will support effective and responsible sharing of cancer genomic and clinical data for the benefit of patients. Funding CLS was supported by Howard Hughes Medical Institute and US National Cancer Institute (Grant #CA008748); ML was funded by the Medical Research Council - Cancer Research UK Stratified Medicine in Colorectal Cancer (S:CORT) Programme grant; LLS was supported by Cancer Care Ontario Research Chair and Applied Cancer Research Units Grant; EEV was supported by the Barcode for Life Foundation and the Hartwig Medical Foundation; DJV was supported by Barcode for Life. No grant number is applicable. Disclosure The authors have declared no conflicts of interest.

Introduction Elderly patients with diffuse large B-cell lymphoma (DLBCL) have a worse prognosis compared to the younger patient population. This is partly explained by lower treatment tolerability in elderly patients with difficulties to administer adequate doses of chemotherapy. However, even when receiving comparable treatment intensities, elderly DLBCL patients have inferior outcome, potentially indicating more aggressive disease biology. Therefore, dose-intensified administration of R-CHOP immunochemotherapy might be of particular benefit for elderly DLBCL patients to overcome these high-risk factors. Treatment of patients >60 years (y) with 6× R-CHOP-14 plus 2× rituximab in the German RICOVER-60 trial has achieved the best long-term outcome in elderly DLBCL patients published to date [1]. However, superiority of dose-intensified R-CHOP-14 compared to the 3-weekly administration in elderly DLBCL patients could not be demonstrated in randomized trials. The GELA LNH03-6B trial comparing R-CHOP-14 and R-CHOP-21 in DLBCL patients aged 60–80y showed no difference of either regimen [2], but results were criticized due to high treatment-related mortality and low dose intensities in the R-CHOP-14 arm. The UK NCRI R-CHOP14v21 trial compared the 2- and 3-weekly R-CHOP regimens in DLBCL patients aged 18–88y and similarly did not observe a difference in outcome across age groups [3]. However, outcomes of the elderly R-CHOP14v21 trial cohort have not been reported separately and it remained unclear whether particular subgroups of elderly patients benefit from intensified treatment.

HOP regimens in DLBCL patients aged 18–88y and similarly did not observe a difference in outcome across age groups [3]. However, outcomes of the elderly R-CHOP14v21 trial cohort have not been reported separately and it remained unclear whether particular subgroups of elderly patients benefit from intensified treatment. The International Prognostic Index (IPI) is widely used for prognostication of younger and elderly DLBCL patients. Due to differences in disease biology and outcomes it has been proposed to use separate prognostic scores for the elderly patient group [4, 5], but these have not yet been validated in large independent cohorts.

ternational Prognostic Index (IPI) is widely used for prognostication of younger and elderly DLBCL patients. Due to differences in disease biology and outcomes it has been proposed to use separate prognostic scores for the elderly patient group [4, 5], but these have not yet been validated in large independent cohorts. Several molecular high-risk markers have been identified in DLBCL that could potentially refine clinical prognostic models. Cell-of-origin (COO) assessment of DLBCL according to gene-expression-profiling separates the germinal center B-cell (GCB) and the poor prognostic activated B-cell (ABC) subtypes, but these analyses lack prospective validation and methodological problems currently limit their use in standard practice. The negative prognostic impact of MYC rearrangements (MYC-R) as well as MYC- and concomitant BCL2- or BCL6 rearrangements (double-hit lymphoma; DHL) has been shown in several DLBCL cohorts [6, 7]. The prognostic significance of MYC-R seems to be particularly high in older DLBCL patients [6]. Due to the low incidence of MYC-R and DHL and possibly due to their age-dependent relevance, an independent prognostic significance of these markers in multivariate models has not yet been demonstrated in prospective trial cohorts of R-CHOP-treated patients. The aim of this subgroup analysis was to provide detailed outcomes and toxicity data on elderly patients treated within the R-CHOP14v21 trial and to investigate the impact of clinical and molecular factors on outcome in this age group.

Several molecular high-risk markers have been identified in DLBCL that could potentially refine clinical prognostic models. Cell-of-origin (COO) assessment of DLBCL according to gene-expression-profiling separates the germinal center B-cell (GCB) and the poor prognostic activated B-cell (ABC) subtypes, but these analyses lack prospective validation and methodological problems currently limit their use in standard practice. The negative prognostic impact of MYC rearrangements (MYC-R) as well as MYC- and concomitant BCL2- or BCL6 rearrangements (double-hit lymphoma; DHL) has been shown in several DLBCL cohorts [6, 7]. The prognostic significance of MYC-R seems to be particularly high in older DLBCL patients [6]. Due to the low incidence of MYC-R and DHL and possibly due to their age-dependent relevance, an independent prognostic significance of these markers in multivariate models has not yet been demonstrated in prospective trial cohorts of R-CHOP-treated patients. The aim of this subgroup analysis was to provide detailed outcomes and toxicity data on elderly patients treated within the R-CHOP14v21 trial and to investigate the impact of clinical and molecular factors on outcome in this age group. Patients and methods Patient characteristics in the R-CHOP14v21 trial have been published in detail [3]. A brief description of the trial is given in the Supplement (available at Annals of Oncology online). Of 1080 R-CHOP14v21 patients, 604 were ≥60y and included in the current analysis. Details of statistical analyses are provided in the Supplement.

Patients and methods Patient characteristics in the R-CHOP14v21 trial have been published in detail [3]. A brief description of the trial is given in the Supplement (available at Annals of Oncology online). Of 1080 R-CHOP14v21 patients, 604 were ≥60y and included in the current analysis. Details of statistical analyses are provided in the Supplement. COO was assessed by the immunohistochemistry (IHC)-based Hans algorithm. Assessment of MYC-, BCL2- and BCL6-rearrangements was done with fluorescence insitu hybridization (FISH; N = 217). DHL was defined as presence of MYC- and either BCL2- or BCL6-rearrangements. In order to increase the sample size to assess the impact of MYC-R and DHL on outcome in elderly DLBCL patients, we performed a joint analysis with data from 204 elderly DLBCL patients treated on the RICOVER-60 trial who had molecular results available (supplementary Table S1, available at Annals of Oncology online). Details of the German high-grade non-Hodgkin lymphoma study group (DSHNHL) RICOVER-60 trial and methods of molecular analyses within the trial have been previously described [1, 7]. Results We included 604 elderly patients from the R-CHOP14v21 trial in this subgroup analysis. Patients’ median age was 67y (range 60–88). Baseline characteristics were well balanced between treatment arms (Table 1). There was a trend towards a higher rate of BCL6 rearrangements and DHL in R-CHOP-14 (P = 0.10 and P = 0.06, respectively). Table 1 Baseline characteristics

CHOP14v21 trial in this subgroup analysis. Patients’ median age was 67y (range 60–88). Baseline characteristics were well balanced between treatment arms (Table 1). There was a trend towards a higher rate of BCL6 rearrangements and DHL in R-CHOP-14 (P = 0.10 and P = 0.06, respectively). Table 1 Baseline characteristics Characteristics R-CHOP-21 R-CHOP-14 (N=301) (N=303) n (%) n (%) Age (years) 60–69 192 (64) 196 (65) ≥70 109 (36) 107 (35) Sex Female 148 (49) 150 (50) Male 153 (51) 153 (50) WHO performance status 0 120 (40) 143 (47) 1 132 (44) 118 (39) 2 49 (16) 42 (14) Stage (N=596) IA 9 (3) 9 (3) IB 6 (2) 7 (2) II 90 (30) 83 (28) III 91 (31) 104 (35) IV 102 (34) 95 (32) Bulk (N=601) 139 (47) 126 (42) B symptoms 121 (40) 134 (44) Elevated LDH 200 (66) 197 (65) >1 extranodal sites 94 (31) 82 (27) IPI score 1 48 (16) 44 (15) 2 75 (25) 90 (30) 3 98 (33) 104 (34) 4 66 (22) 56 (18) 5 14 (5) 9 (3) Subtype (N=317) GCB 76 (50) 82 (50) Non-GCB 77 (50) 82 (50) β2-microglobulin ≥3mg/L (N=371) 88 (51) 102 (52) Albumin ≤35g/L (N=598) 100 (34) 86 (29) MYC rearrangement (N=217) 9 (9) 14 (12) BCL2 translocation (N=220) 26 (25) 33 (28) BCL6 rearrangement (N=218) 17 (16) 30 (26) Double-hit abnormality (N=215) 5 (5) 9 (8)

3) Subtype (N=317) GCB 76 (50) 82 (50) Non-GCB 77 (50) 82 (50) β2-microglobulin ≥3mg/L (N=371) 88 (51) 102 (52) Albumin ≤35g/L (N=598) 100 (34) 86 (29) MYC rearrangement (N=217) 9 (9) 14 (12) BCL2 translocation (N=220) 26 (25) 33 (28) BCL6 rearrangement (N=218) 17 (16) 30 (26) Double-hit abnormality (N=215) 5 (5) 9 (8) Dose intensities were high in both trial arms. Median total doses of cyclosphosphamide, doxorubicin, vincristine, prednisolone and rituximab received were 98% versus 99%, 98% versus 99%, 100% versus 100%, 98% versus 100%, and 98% versus 98% in R-CHOP-21 and R-CHOP-14, respectively. Seventy-one (24%) patients on R-CHOP-21 and 46 (15%) patients on R-CHOP-14 did not complete all treatment cycles (P = 0.01). Reasons for early treatment termination are listed in supplementary Table S2, available at Annals of Oncology online, with treatment-related toxicity being the most common cause. Frequency of dose reductions was similar in both arms (15% for R-CHOP-21 versus 16% for R-CHOP-14; P = 0.73).

ll treatment cycles (P = 0.01). Reasons for early treatment termination are listed in supplementary Table S2, available at Annals of Oncology online, with treatment-related toxicity being the most common cause. Frequency of dose reductions was similar in both arms (15% for R-CHOP-21 versus 16% for R-CHOP-14; P = 0.73). Treatment toxicities are given in Table 2. There was evidence of more grade ≥3 neutropenia (62% versus 36%; P < 0.0001) and less grade ≥3 thrombocytopenia (7% versus 12%; P = 0.05) in R-CHOP-21 compared to R-CHOP-14. Patients on R-CHOP-21 had lower incidence of anemia (20% versus 31%; P = 0.001), with a similar trend for grade ≥3 anemia (2% versus 5%; P = 0.11). No significant difference in the incidence of fever and infections or any other toxicity was observed. The incidence of treatment-related deaths, fatal cardiac events and secondary malignancies were similar in both arms (supplementary Table S3, available at Annals of Oncology online). Table 2 Most common grade ≥3 toxicities and cause of treatment-related deaths

of fever and infections or any other toxicity was observed. The incidence of treatment-related deaths, fatal cardiac events and secondary malignancies were similar in both arms (supplementary Table S3, available at Annals of Oncology online). Table 2 Most common grade ≥3 toxicities and cause of treatment-related deaths R-CHOP-21 (N=301) R-CHOP-14 (N=303) Any grade Grade ≥3 Any grade Grade ≥3 All toxicities 292 (97%) 216 (72%) 299 (99%) 182 (60%) Neutropenia 224 (74%) 185 (61%) 138 (46%) 109 (36%) Thrombocytopenia 73 (24%) 22 (7%) 112 (37%) 37 (12%) Anemia 60 (20%) 6 (2%) 95 (31%) 14 (5%) Infection 145 (48%) 71 (24%) 146 (48%) 71 (23%) Fever 70 (23%) 16 (5%) 56 (18%) 16 (5%) Mucositis 143 (48%) 4 (1%) 167 (55%) 8 (3%) Nausea 188 (62%) 7 (2%) 151 (50%) 12 (4%) Vomiting 98 (33%) 7 (2%) 82 (27%) 9 (3%) Diarrhoea 109 (36%) 12 (4%) 113 (37%) 16 (5%) Constipation 185 (61%) 7 (2%) 160 (53%) 8 (3%) Neurological 167 (55%) 23 (8%) 183 (60%) 36 (12%) Fatigue 240 (80%) 31 (10%) 252 (83%) 40 (13%) Bone pain 68 (23%) 7 (2%) 102 (34%) 6 (2%) Cardiac 29 (10%) 2 (1%) 29 (10%) 9 (3%) Treatment-related deaths: 3 in R-CHOP-21: 2 nonneutropenic sepsis and 1 neutropenic sepsis; and 7 in R-CHOP-14: 2 nonneutropenic sepsis, 1 neutropenic sepsis, 1 renal failure and 3 not specified.

mphocyte. Finally, based on these classes descriptive cellular metrics are computed, including cellular density. Here lymphocyte density is calculated as follows: for every detected lymphocyte in a section, the average distance R to the 50 nearest lymphocytes (N = 50) is calculated using a K-nearest-neighbour approach. For each lymphocyte, density is estimated as N/(πR2) and the median of this value for all detected lymphocytes is taken as the summary statistic for a given section. The computational pathology approach has been described in detail previously [3] and the analysis code is available at http://www.ast.cam.ac.uk/∼adariush/files/codes/.

gue 240 (80%) 31 (10%) 252 (83%) 40 (13%) Bone pain 68 (23%) 7 (2%) 102 (34%) 6 (2%) Cardiac 29 (10%) 2 (1%) 29 (10%) 9 (3%) Treatment-related deaths: 3 in R-CHOP-21: 2 nonneutropenic sepsis and 1 neutropenic sepsis; and 7 in R-CHOP-14: 2 nonneutropenic sepsis, 1 neutropenic sepsis, 1 renal failure and 3 not specified. Response was assessable in 274 patients in each arm. There was no evidence of a difference in response rates between R-CHOP-21 and R-CHOP-14 [complete response (CR)/unconfirmed CR (CRu): 67% versus 62%, P = 0.21; overall response rate (ORR) both 91%; Table 3]. CR/CRu rates after four cycles of therapy were 39% and 33%, respectively (P = 0.15). 61% and 60% of patients are still alive without progression (supplementary Table S3, available at Annals of Oncology online). Four patients on R-CHOP-21 and seven on R-CHOP-14 presented with central nervous system relapse (P = 0.55). Table 3 Response to treatment End of treatment response R-CHOP-21 R-CHOP-14 (N=274) (N=274) n (%) n (%) Complete response (CR) 145 (53) 119 (43) Unconfirmed complete response (CRu) 39 (14) 50 (18) Partial response 64 (23) 80 (29) Stable disease 16 (6) 16 (6) Progressive disease or relapse 10 (4) 9 (3) CR/Cru 184 (67) 169 (62) Overall response rate 248 (91) 249 (91)

response R-CHOP-21 R-CHOP-14 (N=274) (N=274) n (%) n (%) Complete response (CR) 145 (53) 119 (43) Unconfirmed complete response (CRu) 39 (14) 50 (18) Partial response 64 (23) 80 (29) Stable disease 16 (6) 16 (6) Progressive disease or relapse 10 (4) 9 (3) CR/Cru 184 (67) 169 (62) Overall response rate 248 (91) 249 (91) After a median follow-up of 77.7 months, there was no evidence of a difference in progression-free survival (PFS) and overall survival (OS) between treatment arms in patients ≥60y or ≥70y (Figure 1A-D). No difference in survival between R-CHOP-21 and R-CHOP-14 was observed in patients who only achieved partial response (PR) after four cycles (P = 0.79 for PFS; P = 0.68 for OS). There was also no difference between treatment arms with respect to gender (P = 0.54 for PFS; P = 0.67 for OS) or IPI (P = 0.64 for PFS; P = 0.50 for OS). 5y-PFS was 64% (95% CI: 60-68) in patients ≥60y and 58% (95% CI: 51-65) in patients ≥70y. 5y-OS was 69% (95% CI: 66–73) and 61% (95% CI: 54–68), respectively. Figure 1. Kaplan–Meier curves of PFS and OS in (A-B) patients over 60 years and (C-D) patients over 70 years. 63/280 (23%) patients with available data received consolidation radiotherapy. Of those, 36 had initial bulk, 20 extranodal disease, and 10 had both. Disease status before radiotherapy was available for 61 patients: 23 (37%) CR/CRu, 31 (51%) PR and 7 (12%) SD. In patients with PR or SD who are supposed to benefit most from radiotherapy, the use of radiotherapy was not associated with OS (supplementary Figure S3, available at Annals of Oncology online).

10 had both. Disease status before radiotherapy was available for 61 patients: 23 (37%) CR/CRu, 31 (51%) PR and 7 (12%) SD. In patients with PR or SD who are supposed to benefit most from radiotherapy, the use of radiotherapy was not associated with OS (supplementary Figure S3, available at Annals of Oncology online). In multivariable analysis, only age and B2M levels were of independent prognostic significance for OS (supplementary Table S4, available at Annals of Oncology online). There was no significant impact of COO subtypes on outcomes (supplementary Figure S1, available at Annals of Oncology online). When comparing prognostic scores IPI, R-IPI, E-IPI and ABE4 (supplementary Table S5 and Figure S2, available at Annals of Oncology online), ABE4 achieved the best fit and discrimination for predicting OS, followed by the IPI. Similar results were obtained for PFS (data not shown).

ailable at Annals of Oncology online). When comparing prognostic scores IPI, R-IPI, E-IPI and ABE4 (supplementary Table S5 and Figure S2, available at Annals of Oncology online), ABE4 achieved the best fit and discrimination for predicting OS, followed by the IPI. Similar results were obtained for PFS (data not shown). To assess the impact of MYC-R and DHL on outcome we performed a joint analysis with cases from RICOVER-60. 23/217 (11%) patients from our cohort and 19/204 (9%) patients from RICOVER-60 had MYC-R as determined by FISH. 14/215 (7%) and 9/182 (5%) had DHL, respectively. MYC-R and DHL cases had significantly worse OS compared to cases without these abnormalities [HR = 1.96 (95% CI: 1.22–3.16); P = 0.01 and HR = 2.21 (95% CI: 1.18–4.11); P = 0.01, respectively); Figure 2]. Similar effect sizes were observed after adjusting for individual IPI factors and trial arms [HR = 1.76 (95% CI: 1.09–2.85); P = 0.02 and HR = 2.08 (95% CI: 1.11–3.90); P = 0.02, respectively)]. The difference in OS between DHL and MYC-R was not significant (HR = 1.38 (95% CI: 0.55–3.43; P = 0.49). There was no significant impact of BCL2- or BCL6-rearrangements on OS (P = 0.34 and P = 0.99, respectively). Figure 2. Kaplan–Meier curves of OS according to MYC- and BCL2 rearrangements and double-hit abnormality in R-CHOP treated elderly patients from R-CHOP14v21 (N = 215) and RICOVER-60 (N = 182). Discussion With a median follow-up of 6.5y, we provide a detailed analysis of outcome and toxicities from patients with newly diagnosed DLBCL aged ≥60y treated on the phase 3 R-CHOP14v21 trial.

Figure 2. Kaplan–Meier curves of OS according to MYC- and BCL2 rearrangements and double-hit abnormality in R-CHOP treated elderly patients from R-CHOP14v21 (N = 215) and RICOVER-60 (N = 182). Discussion With a median follow-up of 6.5y, we provide a detailed analysis of outcome and toxicities from patients with newly diagnosed DLBCL aged ≥60y treated on the phase 3 R-CHOP14v21 trial. Elderly DLBCL patients in our cohort had an excellent long-term outcome with 5y-OS of 69% (3y-PFS 71%; 3y-OS 76%). These results are similar to data from elderly DLBCL patients treated with 6× R-CHOP-14 on RICOVER-60 (3y-PFS 73%; 3y-OS 78%) and better than outcomes in the GELA LNH03-6B trial (3y-PFS 61%; 3y-OS 73%) [1, 2]. Of note, patients’ median age was higher in LNH03-6B (70y) compared to our cohort (67y) and RICOVER-60 (68y). In addition, there were more cases presenting with high IPI (3–5) in the LNH03-6B trial (75% versus 57% in our subgroup versus 43% in RICOVER-60), which might have contributed to inferior outcome seen in this trial population.

s’ median age was higher in LNH03-6B (70y) compared to our cohort (67y) and RICOVER-60 (68y). In addition, there were more cases presenting with high IPI (3–5) in the LNH03-6B trial (75% versus 57% in our subgroup versus 43% in RICOVER-60), which might have contributed to inferior outcome seen in this trial population. Toxicity profiles in our cohort of elderly DLBCL patients were favorable in both treatment arms. As expected, patients on R-CHOP-21 had a higher incidence of neutropenia probably due to reduced use of G-CSF, but less thrombocytopenia. Importantly, there was no difference in infectious complications or treatment-related deaths. The incidence of deaths during chemotherapy was very low at 1.7%, suggesting adequate management of elderly patients in participating centers. In the LNH03-6B trial, a high treatment-related mortality of 9% was observed in the initial recruitment period, which improved towards the end of the study, indicating gain of clinical experience with dose-intensified treatment in elderly patients. With 6.5y median follow-up, there was no difference in long-term toxicity, specifically cardiac events and secondary malignancies, between arms.

ved in the initial recruitment period, which improved towards the end of the study, indicating gain of clinical experience with dose-intensified treatment in elderly patients. With 6.5y median follow-up, there was no difference in long-term toxicity, specifically cardiac events and secondary malignancies, between arms. Dose intensities were high in both arms and as seen in the entire R-CHOP14v21 trial cohort [3]. The low dose intensity of 88% for R-CHOP-14 in the LNH03-6B trial could have potentially underestimated efficacy of the 2-weekly regimen. Our results support equivalence of both regimens in elderly DLBCL patients when adequate doses are achieved. However, the study was not powered for this posthoc subgroup analysis in elderly patients. We did not identify any subgroup of elderly DLBCL patients that showed differential response to either regimen, including gender and IPI groups. No difference between treatment arms could be seen in patients ≥70y. An analysis of patients ≥80y was not feasible due to low numbers (N = 20). Moreover, there was no benefit of dose-intense treatment in late responders who had not achieved CR/Cru after four cycles.

e to either regimen, including gender and IPI groups. No difference between treatment arms could be seen in patients ≥70y. An analysis of patients ≥80y was not feasible due to low numbers (N = 20). Moreover, there was no benefit of dose-intense treatment in late responders who had not achieved CR/Cru after four cycles. Consolidation radiotherapy was at the discretion of the investigators and performed in 23% of elderly patients with available data. The main indication for radiotherapy was initial bulky or extranodal disease. The benefit of radiotherapy to initial bulk in elderly DLBCL patients is reported to be greatest for patients who are not in CR/CRu after induction therapy [8]. Accordingly, most patients in our analysis received radiotherapy to PR or SD at the end of treatment, without evidence of a survival benefit for this strategy. However, these data have significant limitations (nonrandomized approach, small numbers). In addition, no PET-CT data were recorded. The on-going DSHNHL OPTIMAL > 60 trial will investigate whether consolidation radiotherapy can be safely omitted in elderly DLBCL patients who are PET-negative at the end of treatment. Remarkably, PFS of elderly patients was only 8 percentage points worse at 5y compared to younger patients (5y-PFS 64% versus 72%), supporting the concept of treating elderly patients with full doses of chemotherapy whenever possible. Toxicities were also similar between elderly and younger patients (data not shown), besides a significantly higher rate of grade ≥3 neutropenia in elderly (P≤0.001).

mpared to younger patients (5y-PFS 64% versus 72%), supporting the concept of treating elderly patients with full doses of chemotherapy whenever possible. Toxicities were also similar between elderly and younger patients (data not shown), besides a significantly higher rate of grade ≥3 neutropenia in elderly (P≤0.001). Differences between DLBCL of elderly and younger patients have been described on the molecular level, with higher frequencies of ABC subtypes, BCL6 rearrangements, gains in 1q21, 18q21, and 7q21, and a higher genetic complexity associated with increasing age [9]. We did not observe material differences in the frequency of MYC-R, BCL6- and BCL2-rearrangements between age groups, nor in the incidence of IHC-based cell-of-origin subtypes (data not shown). We found lower frequency of bulky disease and higher B2M levels in elderly compared to younger patients, implying differences in disease biology between both groups.

s in the frequency of MYC-R, BCL6- and BCL2-rearrangements between age groups, nor in the incidence of IHC-based cell-of-origin subtypes (data not shown). We found lower frequency of bulky disease and higher B2M levels in elderly compared to younger patients, implying differences in disease biology between both groups. Age-specific clinical and molecular features suggest the need for a separate prognostic scoring system for elderly patients. We compared performances of two recently proposed prognostic scores for elderly DLBCL (ABE4 [5] and E-IPI [4]) with the standard IPI and R-IPI in our cohort. Both scores use an age cut-off of 70y. ABE4 further incorporates bulky disease and separates PS ≥ 1 instead of ≥ 2. The ABE4 performed best in our cohort, despite bulky disease not being significantly associated with patient outcomes. Therefore, separating patients with PS 0 from those with PS ≥ 1 could be a more appropriate cut-off in an elderly patient group. Both ABE4 and IPI distinguished meaningful prognostic groups for PFS and OS. However, clinical utility of the ABE4 score might be limited by the fact that only 9% of patients from our cohort were in the high-risk group compared with 14% in the original Czech Lymphoma Registry [5]. As discussed by Ziepert et al. [10], introduction of new scores have to be seen with caution and should only be considered if properly validated and if changing patients’ management. The main use of the IPI has been in the context of clinical trials, allowing risk-stratification of patients and facilitating comparison of results across trials. A NCCN-IPI has recently been proposed which separates three different age groups as risk factors [11]. The great disadvantage of this score is that it cannot be used for elderly and young patient groups separately and is therefore unsuitable for age-specific DLBCL trials. In contrast, the IPI as age-adjusted IPI has been validated in both young and elderly DLBCL.

eparates three different age groups as risk factors [11]. The great disadvantage of this score is that it cannot be used for elderly and young patient groups separately and is therefore unsuitable for age-specific DLBCL trials. In contrast, the IPI as age-adjusted IPI has been validated in both young and elderly DLBCL. In line with previous findings, IHC-based cell-of-origin classification did not impact on outcomes, further underscoring limitations of this method. However, final analyses of the REMoDL-B trial will reveal if the concept of cell-of-origin classification as prognostic marker holds true when assessed prospectively [12]. Our combined analysis of FISH data from R-CHOP14v21 and RICOVER-60 demonstrates for the first time independent prognostic significance of both MYC-R and DHL in patients treated with R-CHOP within prospective cohorts. A negative prognostic impact of MYC-R and DHL has been reported in several heterogeneous DLBCL populations, but did not reach independent significance in trial cohorts due to small numbers [3, 7]. On-going prospective trials will reveal if these patients benefit from upfront treatment intensification. In conclusion, our data demonstrate excellent short and long-term results with both R-CHOP-14 and R-CHOP-21 in elderly DLBCL patients. This analysis contributes important information to the longstanding discussion about optimal management of the elderly DLBCL patient population and provides a detailed analysis of molecular and clinical prognostic factors in this age group.

long-term results with both R-CHOP-14 and R-CHOP-21 in elderly DLBCL patients. This analysis contributes important information to the longstanding discussion about optimal management of the elderly DLBCL patient population and provides a detailed analysis of molecular and clinical prognostic factors in this age group. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements We would like to thank participating centers of the R-CHOP14v21 and RICOVER-60 trials, as well as patients and their families involved. Funding Cancer Research UK provided endorsement and funding of the R-CHOP14v21 trial (CRUKE/03/019). The National Health Service (NHS) provided funding to the National Institute for Health Research (NIHR) Biomedical Research centers at both University College London and the Royal Marsden Hospital/Institute of Cancer Research, London, UK (no grant numbers apply). Chugai Pharmaceuticals provided an educational grant and lenograstim within the R-CHOP14v21 trial (no grant numbers apply). Key Message We provide a detailed outcome analysis of elderly patients with DLBCL treated with 2- or 3-weekly R-CHOP within the UK R-CHOP14v21 trial, indicating equivalence of both regimens in the elderly patient population. In a joint analysis of R-CHOP14v21 and RICOVER-60 molecular data, we demonstrate that MYC rearrangements and double-hit-lymphoma are independent poor prognostic factors in elderly DLBCL.

ith 2- or 3-weekly R-CHOP within the UK R-CHOP14v21 trial, indicating equivalence of both regimens in the elderly patient population. In a joint analysis of R-CHOP14v21 and RICOVER-60 molecular data, we demonstrate that MYC rearrangements and double-hit-lymphoma are independent poor prognostic factors in elderly DLBCL. Disclosure DC has received research funding from Amgen, Astra Zeneca, Bayer, Celgene, Medimmune, Merrimack, Merck Serono and Sanofi. EAH has received travel expenses from Takeda and Bristol-Myers Squibb. CP has received travel expenses from Gilead and Speaker fees from Janssen. KMA has received research funding, conference expenses and honoraria for attending or chairing advisory boards from Roche. GO was supported by the Robert-Bosch-Stiftung, Stuttgart, Germany. All other authors have no conflicts of interest to disclose.

Introduction Prostate cancer (PC) is the second most common malignancy in men with an incidence of 1.1 million men per year leading to an estimated 307,000 deaths worldwide [1]. While the prognosis of clinically low-risk PC is excellent [2], there is significant mortality associated with clinically high-risk disease, with approximately 20%–25% 10-year cancer-specific mortality despite radical treatments [3]. A key challenge in PC is to identify patients with potentially lethal disease while avoiding the morbidity of overtreatment in patients with indolent disease. Accurate risk stratification has been confounded by underestimation of disease burden using standard transrectal ultrasound-guided biopsies and extensive tumour heterogeneity of primary PC [4–6]. A recent improvement on transrectal ultrasound-guided biopsies is targeted magnetic resonance imaging (MRI)-guided biopsies that increase the likelihood of sampling clinically significant disease [7]. Multi-regional sampling of tumours allows measurement of intratumoural heterogeneity, a prognostic entity in PC [8]. To date, in-depth exploration of genomic ITH with multiregion sequencing (M-Seq) in the primary PC has relied on prostatectomy patient series [9, 10] to provide good-quality tissue; however, this has enriched for clinically low- and intermediate-risk disease [5, 6].

umoural heterogeneity, a prognostic entity in PC [8]. To date, in-depth exploration of genomic ITH with multiregion sequencing (M-Seq) in the primary PC has relied on prostatectomy patient series [9, 10] to provide good-quality tissue; however, this has enriched for clinically low- and intermediate-risk disease [5, 6]. Tumour infiltrating lymphocyte density has been shown, albeit inconsistently, to be prognostic in PC [11–13]. The impact of tumour genetics on prostate immunobiology is unclear and deciphering this could improve risk stratification, prognostication and immunotherapeutic approaches. We conducted the PROGENY study (PROstate cancer GENomic heterogeneitY) to attain high-quality multi-regional prostate biopsies to determine the driver and evolutionary events of clinically high-risk PC at the time of diagnosis and to correlate genomic and immune parameters. Methods and materials Patient selection Between September 2013 and December 2015, 49 men with a prostate-specific antigen ≥15, a multi-parametric MRI detectable lesion in the prostate and no prior prostate-directed biopsies or treatments were enrolled into the PROGENY study, with local ethics committee approval. Of these, 23 patients and a further 2 contemporaneous patients from the institutional biobank met the criteria for the planned genetic and T-cell analysis (supplementary Table S1 and Figure S1, available at Annals of Oncology online).

reatments were enrolled into the PROGENY study, with local ethics committee approval. Of these, 23 patients and a further 2 contemporaneous patients from the institutional biobank met the criteria for the planned genetic and T-cell analysis (supplementary Table S1 and Figure S1, available at Annals of Oncology online). Tissue procurement Multi-regional PC biopsies were obtained using multi-parametric MRI, image-fusion transperineal template targeting as described previously [7] (supplementary Figure S2, available at Annals of Oncology online). Blood samples were obtained before the biopsy for isolation of germline DNA. Sequencing studies Tumour DNA was extracted using the Allprep Micro Kit (Qiagen, CA) and germline DNA extracted with the DNeasy Blood & Tissue Kit (Qiagen, MD) following the manufacturer’s instructions. Further details are contained in the supplementary data, available at Annals of Oncology online. Immunohistochemistry Single and multiplexed IHC was carried out as described previously [14]. Antibody details are in supplementary Table S2, available at Annals of Oncology online. TM and ML jointly carried out quantification of inflammatory infiltrate (INIF), blinded to patient characteristics. Samples with ≥8% (median of all samples) INIF (15/25 patients) in any one region were subjected to digital image analysis. These correlated well with the manual estimation (R2 = 0.71) (supplementary Figures S3 and S4, available at Annals of Oncology online).

matory infiltrate (INIF), blinded to patient characteristics. Samples with ≥8% (median of all samples) INIF (15/25 patients) in any one region were subjected to digital image analysis. These correlated well with the manual estimation (R2 = 0.71) (supplementary Figures S3 and S4, available at Annals of Oncology online). Results The extent of intratumoural heterogeneity in high-risk PC Across 25 prospectively recruited patients, M-Seq from 79 tumour regions identified a total of 4484 exonic somatic nucleotide variations (SNV) (3382 non-silent), of which 1962 were ubiquitous, 495 were shared and 2027 were private (Figure 1A). The overall estimation of exonic SNV burden was 0.93 mutations per megabase (median, range, 0.18–33 per megabase), consistent with prior studies in PC [15].

ied a total of 4484 exonic somatic nucleotide variations (SNV) (3382 non-silent), of which 1962 were ubiquitous, 495 were shared and 2027 were private (Figure 1A). The overall estimation of exonic SNV burden was 0.93 mutations per megabase (median, range, 0.18–33 per megabase), consistent with prior studies in PC [15]. Figure 1. Intratumoural heterogeneity in prostate cancer at the somatic nucleotide variation (SNV) and somatic copy number alteration (SCNA) levels. (A) Number of somatic exonic mutations identified in each tumour region, fraction of SNVs and SCNAs that were ubiquitous (present in every tumour region of a given patient) (blue), shared (present in more than one tumour region, but not all) (light orange) or private (present in only one tumour region) (dark orange). Data tracks below indicate if patient was metastatic on presentation (red), Gleason grade (shades of green), level of tumoural inflammatory infiltrate (shades of brown), and if the tumour had undergone whole-genome doubling (purple, triangle indicating heterogeneous genome doubling). (B) Scatterplot showing correlation between degree of SNV and SCNA heterogeneity. (C and D) Boxplots comparing the fraction of genome affected by SCNA and SNV mutational burden in metastatic hormone naive prostate cancer (mHNPC) versus high-risk localised prostate cancer (hrlPC).

icating heterogeneous genome doubling). (B) Scatterplot showing correlation between degree of SNV and SCNA heterogeneity. (C and D) Boxplots comparing the fraction of genome affected by SCNA and SNV mutational burden in metastatic hormone naive prostate cancer (mHNPC) versus high-risk localised prostate cancer (hrlPC). The overall fraction of the genome subject to somatic copy number alterations (SCNAs) was 23.1% (median, range 1.9%–41.6%). Of this fraction, a median of 52.3% (range 2.1%–95.3%) was heterogeneous (Figure 1A). The degree of SNV and SCNA heterogeneity among the tumours was positively correlated (Figure 1B) (r = 0.49, P = 0.013, Pearson’s). Two patients, BP0001 and PR0103, had markedly elevated SNV rates. BP0001 had a previous diagnosis of Lynch Syndrome and was found to harbour a germline mutation in MSH6 (p.G39E, rs1042821) and a somatic heterozygous deletion encompassing the region encoding for MSH2 and MSH6, resulting in a hemizygous variant in MSH6. PR0103 had a somatic 10 Mb deletion overlapping MSH2 and MSH6 and a 5 kb somatic deletion across MSH2, leading to biallelic loss of MSH2. IHC of MSH2 and MSH6 in both of these patients showed complete loss of protein expression in the tumours (supplementary Figures S5 and S6, available at Annals of Oncology online).

6. PR0103 had a somatic 10 Mb deletion overlapping MSH2 and MSH6 and a 5 kb somatic deletion across MSH2, leading to biallelic loss of MSH2. IHC of MSH2 and MSH6 in both of these patients showed complete loss of protein expression in the tumours (supplementary Figures S5 and S6, available at Annals of Oncology online). Genomic events enriched in patients presenting with metastatic disease After the diagnostic biopsy, 12/25 patients were found to have metastatic disease on imaging (mHNPC) and 13 patients had localised PC with high risk for metastatic disease (hrlPC). mHNPC primary tumours had significantly higher burden of SCNAs compared with hrlPC tumours (29.6% ± 10.6% versus 12.5% ± 8.9%, P = 7.57 × 10−4, Mann–Whitney U test) (Figure 1C) and this was independent of Gleason grade. Comparing mHNPC and hrlPC patients, there was no significant difference in the proportion of heterogeneous SCNAs (P = 0.89, Mann–Whitney U test), overall mutational burden (P = 0.74, Mann–Whitney U test ) (Figure 1D), or proportion of heterogeneous mutations (P = 0.11, Mann–Whitney U test). To explore the relative frequency of SNVs and SCNAs in mHNPC and hrlPC, we focused on driver genes identified in previous PC series (Figure 2) [16, 17]. We found no significant differences between mHNPC and hrlPC tumours. However, there was a significant enrichment of 3q26.2 and 3q21.3 gains in mHNPC compared to hrlPC tumours (5/12 versus 1/13 and 3/12 versus 1/13, respectively) (Figure 3), which remained significantly enriched after controlling for the differing levels of SCNAs.

significant differences between mHNPC and hrlPC tumours. However, there was a significant enrichment of 3q26.2 and 3q21.3 gains in mHNPC compared to hrlPC tumours (5/12 versus 1/13 and 3/12 versus 1/13, respectively) (Figure 3), which remained significantly enriched after controlling for the differing levels of SCNAs. Figure 2. Clonal and subclonal driver events in prostate cancer. List of driver genes previously reported as significantly mutated in primary prostate cancer (blue), metastatic castrate-resistant prostate cancer (mCRPC) (grey), or both (black). Ubiquitous ETS fusion (purple), homozygous loss (dark blue), heterozygous deletion (blue), amplification (red) in each tumour is depicted by a coloured square, and heterozygous events are indicated with a triangle. Nonsynonymous mutations are depicted as smaller squares, whether missense (green), frameshift (yellow) or nonsense (dark blue). Clonal and subclonal mutations are indicated by a purple and orange outline, respectively. Known recurrent mutations in TP53, PIK3CA, CTNNB1 and BRAF are indicated with a red star. The barplots on the right are an aggregate of clonal/ubiquitous or subclonal/heterogenous events in each gene across all samples. Metastatic hormone naive prostate cancer (mHNPC); high-risk localised prostate cancer (hrlPC).

vely. Known recurrent mutations in TP53, PIK3CA, CTNNB1 and BRAF are indicated with a red star. The barplots on the right are an aggregate of clonal/ubiquitous or subclonal/heterogenous events in each gene across all samples. Metastatic hormone naive prostate cancer (mHNPC); high-risk localised prostate cancer (hrlPC). Figure 3. Recurrent somatic copy number alterations (SCNAs) in prostate cancer. (A) An overview of the SCNA landscape across all 25 tumours: fraction of cohort (y-axis) with ubiquitous gains (red), heterogeneous gains (pink), ubiquitous loss (dark blue) and heterogeneous loss (light blue) are shaded across the genome (x-axis). (B) Frequencies of occurrence of previously identified GISTIC focal and arm-level SCNAs across all tumours, metastatic on presentation (mHNPC) and non-metastatic on presentation (hrlPC) tumours. Shades of colours as in A. Parallel evolution of wnt/β-catenin pathway We observed one tumour (PR0139) with three distinct CTNNB1 mutations, all previously described gain-of-function mutations in exon 3 of CTNNB1 leading to stabilisation of β-catenin and activation of Wnt/β-catenin signalling (Figure 4A). Phylogenetic analysis of the clonal structure in this tumour revealed that all 3 CTNNB1 mutations were in three separate subclones (Figure 4B), providing strong evidence for parallel evolution leading to activation of the Wnt/β-catenin pathway in this tumour.

enin and activation of Wnt/β-catenin signalling (Figure 4A). Phylogenetic analysis of the clonal structure in this tumour revealed that all 3 CTNNB1 mutations were in three separate subclones (Figure 4B), providing strong evidence for parallel evolution leading to activation of the Wnt/β-catenin pathway in this tumour. Figure 4. Parallel evolution in Wnt pathway in PR0139 and association with CD8+/FOXP3+ ratio across 15 tumours. (A) Left, fraction of cancer cells in sequenced tumour regions R1, R2 and R3 harbouring different CTNNB1 mutations, p.S33P (pink), p.T41A (blue) and p.S33C (orange). Right, schematic showing different compositions of subclones in each sequenced tumour region, colours correspond to left panel. (B) Phylogenetic tree showing evolutionary history of PR0139 and acquisition of various driver mutations. Relative sizes of circles correspond to number of SNV mutations in that mutational cluster. Temporal order of driver events in clinically high-risk PC To explore the relative timing of driver events in PC, we utilised a modified version of Pyclone to cluster the mutations (supplementary Methods, available at Annals of Oncology online ). Consistent with previous reports about PC tumourigenesis [18, 19], we observed ETS fusions and mutations or loss of TP53 to be early (clonal) events (Figure 2 and supplementary Table S3, available at Annals of Oncology online), PTEN a later event (60% clonal), and mutations or deletions of chromatin modifiers (KMT2C, KMT2D and CHD1) as a later (subclonal) event (Figure 2).

s [18, 19], we observed ETS fusions and mutations or loss of TP53 to be early (clonal) events (Figure 2 and supplementary Table S3, available at Annals of Oncology online), PTEN a later event (60% clonal), and mutations or deletions of chromatin modifiers (KMT2C, KMT2D and CHD1) as a later (subclonal) event (Figure 2). The landscape of SCNAs was also highly consistent with previous studies [15, 17], (Figure 3). In general, we observed that the majority of recurrent SCNA peaks were early events across most tumours in the cohort, aside from 8q and 7p gains, which occurred heterogeneously in 7/13 and 4/6 tumours. Next, we investigated the mutational processes in the two patients with defective MMR (supplementary Figure S7, available at Annals of Oncology online). BP0001, who had germline MSH2 and MSH6 aberrations, had a high proportion of ubiquitous mutations associated with Signature 6 (DNA repair) compared with PR0103 (41.3% versus 8.6%) in keeping with loss of MMR as an early tumourigenic process. Conversely, ubiquitous mutations in PR0103 were mainly associated with Signature 1 (age), suggesting that MMRD was not an initial driver of this tumour, but rather the acquired biallelic loss of MSH2 was a later event that provided a selective advantage, possibly through an accelerated mutation rate.

ourigenic process. Conversely, ubiquitous mutations in PR0103 were mainly associated with Signature 1 (age), suggesting that MMRD was not an initial driver of this tumour, but rather the acquired biallelic loss of MSH2 was a later event that provided a selective advantage, possibly through an accelerated mutation rate. T-cell infiltrate heterogeneity and neoantigen burden There was considerable variation in the total inflammatory infiltrate (INIF) (CD8+ or CD4+ and/or FoxP3+ cells in tumour region) between patients (Figure 5A), as well as between different regions within each patient. This intratumoural heterogeneity of INIF is well illustrated by PR0123, where 4 separate core biopsies have different levels of INIF (mean 15%, range 5%–25%) (Figure 5B). Figure 5. T-cell heterogeneity in prostate tumours. (A) Manual quantification of inflammatory infiltrate. Mean is represented by horizontal lines, box and whiskers show the 95% confidence interval and range, respectively. The dotted line marks the threshold for high inflammatory infiltrate. (B) Multiplex immunohistochemistry (IHC) analysis of four different prostate core biopsies (R1–4) from a patient, PR0123, showing heterogeneity in T-cell infiltration. CD8 staining in red, CD4 in brown and FoxP3 in blue. (C) Boxplot comparing CD8+/FOXP3+ ratios between tumours with and without somatic activation of Wnt pathway (gain-of-function mutation in CTNNB1, amplification in RSPO2, loss in APC, RNF43 and ZNRF3) across 15 tumours with digital image analysis.

T-cell infiltration. CD8 staining in red, CD4 in brown and FoxP3 in blue. (C) Boxplot comparing CD8+/FOXP3+ ratios between tumours with and without somatic activation of Wnt pathway (gain-of-function mutation in CTNNB1, amplification in RSPO2, loss in APC, RNF43 and ZNRF3) across 15 tumours with digital image analysis. We noted that both PR0103 and BP0001 had extensive INIF (maximal infiltrate >20% of all nucleated cells per biopsy) (2/2) compared with patients without MMR deficiency, where only 6/23 had extensive INIF. Patients PR0112 and PR0129 had ubiquitous and heterozygous loss of MLH1 and MSH2 respectively, but this was not associated with high mutational burden or high INIF. As mutational load has been reported to correlate with neoantigen load and neoantigens can elicit a clonal expansion of neoantigen reactive T- (NART) cells [20–22], we hypothesised that the abundant INIF in these MMRD deficient tumours might be related to a high neoantigenic burden. Consistent with this, PR0103 and BP0001 displayed a high neoantigen burden. However, extending this analysis to all 25 patients in this cohort, there was no association between neoantigen burden nor clonal neoantigen burden and INIF (supplementary Figure S8, available at Annals of Oncology online).

neoantigenic burden. Consistent with this, PR0103 and BP0001 displayed a high neoantigen burden. However, extending this analysis to all 25 patients in this cohort, there was no association between neoantigen burden nor clonal neoantigen burden and INIF (supplementary Figure S8, available at Annals of Oncology online). Wnt signalling and modulation of immune response Activation of tumour intrinsic Wnt/β-catenin signalling in melanoma has recently been reported to lead to T-cell exclusion from the tumour preventing anti-tumour immunity [23]. However, PR0139 who had parallel evolution of activated β-catenin, had high levels of CD8+ infiltrate (Figure 5A), but was noted to also have high FOXP3+ levels giving a low CD8+/FOXP3+ ratio. A low ratio of tumour-infiltrating CD8+ and FOXP3+ lymphocytes is increasingly being recognised as a measure of immune suppression and as a potential prognostic indicator [24–26]. In our cohort, the 15 patients with levels of INIF at or above the median underwent digital pathology analysis. Of these patients, 7/15 had activating mutations in the Wnt pathway (gain-of-function CTNNB1 mutations, RSPO2 amplification, and deletion of APC, RNF43 and ZNRF3 [17, 27]). We observed a significantly lower CD8+/FOXP3+ ratio in patients with tumours containing activating mutations of the Wnt pathway compared with wild-type tumours (2.65 ± 1.2 versus 6.08 ± 5.0, P = 0.043, Mann–Whitney U test) (Figure 5C).

NNB1 mutations, RSPO2 amplification, and deletion of APC, RNF43 and ZNRF3 [17, 27]). We observed a significantly lower CD8+/FOXP3+ ratio in patients with tumours containing activating mutations of the Wnt pathway compared with wild-type tumours (2.65 ± 1.2 versus 6.08 ± 5.0, P = 0.043, Mann–Whitney U test) (Figure 5C). Discussion We have conducted the largest prospective clinical cohort study of M-Seq in high-risk PC patients and carried out an integrated genomic and tumour immune infiltrate analysis. Uniquely, we have compared M-Seq of diagnostic prostate biopsies from mHNPC and hrlPC and demonstrated increased SCNA in mHNPC patients, consistent with previous reports correlating biochemical recurrence following prostatectomy with high SCNA in localised disease [28]. We observed no differences in SNV frequency between mHNPC and hrlPC patients, which is surprising given the large differences seen in other studies between localised PC and pre-treated metastatic castrate-resistant prostate cancer [17, 29]. This may be a consequence of the small sample size or may suggest that unlike SCNA changes, SNV accumulation is a later evolutionary event, possibly as a result of the selective pressure of treatment. In this study, there was enrichment for gains of 3q26.2 and 3q21.3 in mHNPC patients. Both amplicons contain genes previously implicated in PC, e.g. 3q26.2 contains PRKCI, expression of which is associated with biochemical relapse following prostatectomy [30]. Interestingly, these gains in copy number are early evolutionary events, and the fact that these focal gains are enriched in patients presenting with metastatic disease suggests that some PCs are hard-wired to be aggressive.

PRKCI, expression of which is associated with biochemical relapse following prostatectomy [30]. Interestingly, these gains in copy number are early evolutionary events, and the fact that these focal gains are enriched in patients presenting with metastatic disease suggests that some PCs are hard-wired to be aggressive. We describe the first report of parallel evolution of Wnt signalling in PC, where 3 separate gain-of-function mutations of β-catenin (CTNNB1) were identified in a single tumour. This is similar to the distinct TMPRSS-ERG fusions identified in several regions of the primary prostate tumour [5] and alterations of SETD2, PTEN and KDM5C in renal cancer [31]. Parallel evolution of the Wnt pathway, a pathway already implicated in PC cell growth, proliferation and epidermal to mesenchymal transition [32], points to its biological importance in PC. Unlike mouse melanoma models, where tumour intrinsic Wnt/β-catenin signalling led to T-cell exclusion from the tumour [23], we observed that patients with activated Wnt/β-catenin signalling can have normal or high levels of INIF, but that this is predominantly CD8+/FOXP3+ low, consistent with a dysfunctional T-cell response. Future studies will be needed to further elucidate the role and mechanism of Wnt/β-catenin signalling in immune modulation in human PC, which is of particular interest given the number of potential novel drugs targeting this pathway.

his is predominantly CD8+/FOXP3+ low, consistent with a dysfunctional T-cell response. Future studies will be needed to further elucidate the role and mechanism of Wnt/β-catenin signalling in immune modulation in human PC, which is of particular interest given the number of potential novel drugs targeting this pathway. We identified two patients with hypermutation associated with MMR deficiency and high INIF, the latter being similar to a report of 12/16 (75%) men at risk of Lynch syndrome and diagnosed with PC having significant INIF [33]. Similar to reports in advanced PC [34], our hrlPC patients with MMRD had complex structural rearrangements of DNA repair genes MSH2 and MSH6 leading to inactivation. Overall however, we did not demonstrate an association with INIF and neoepitope burden, but given the small number of patients with DNA repair aberrations in this series, this analysis is underpowered. MMRD deficiency has been associated with response to immune checkpoint inhibition in a number of tumour types including PC [35, 36]. The finding of high INIF and neoepitope burden in some PC patients in this study supports current attempts to evaluate the role of mutational burden and neoepitopes in prospective therapeutic clinical trials (NCT02113657 and NCT03061539). We have demonstrated extensive intratumoural heterogeneity of INIF in primary PC. The impact of this on prognosis and predicting treatment response is unknown, but future studies testing INIF as a potential biomarker will need to consider testing multiple tumour regions or developing a liquid biopsy strategy.

We identified two patients with hypermutation associated with MMR deficiency and high INIF, the latter being similar to a report of 12/16 (75%) men at risk of Lynch syndrome and diagnosed with PC having significant INIF [33]. Similar to reports in advanced PC [34], our hrlPC patients with MMRD had complex structural rearrangements of DNA repair genes MSH2 and MSH6 leading to inactivation. Overall however, we did not demonstrate an association with INIF and neoepitope burden, but given the small number of patients with DNA repair aberrations in this series, this analysis is underpowered. MMRD deficiency has been associated with response to immune checkpoint inhibition in a number of tumour types including PC [35, 36]. The finding of high INIF and neoepitope burden in some PC patients in this study supports current attempts to evaluate the role of mutational burden and neoepitopes in prospective therapeutic clinical trials (NCT02113657 and NCT03061539). We have demonstrated extensive intratumoural heterogeneity of INIF in primary PC. The impact of this on prognosis and predicting treatment response is unknown, but future studies testing INIF as a potential biomarker will need to consider testing multiple tumour regions or developing a liquid biopsy strategy. In conclusion, our findings reveal how mutational and SCNA changes may drive aggressive metastatic PC. We show that activated Wnt signalling is correlated with immune suppression in primary PC, and suggest that activated Wnt/β-catenin, MMR, high INIF and the CD8+/FOXP3+ ratio should be explored as predictive biomarkers for immunotherapeutics in prostate cancer.

tional and SCNA changes may drive aggressive metastatic PC. We show that activated Wnt signalling is correlated with immune suppression in primary PC, and suggest that activated Wnt/β-catenin, MMR, high INIF and the CD8+/FOXP3+ ratio should be explored as predictive biomarkers for immunotherapeutics in prostate cancer. Funding This work was supported by Prostate Cancer Foundation. CS, ML, ME, SAQ and TS are supported by the National Institute for Health Research, the University College London Hospitals Biomedical Research Centre (no grant numbers apply). ML has received support from a BMS II-ON grant (no grant numbers apply). ME is a UK National Institute of Health Research (NIHR) Senior Investigator. CS, HUA and ML are supported by the Cancer Research UK University College London Experimental Cancer Medicine Centre (no grant numbers apply). CS is Royal Society Napier Research Professor. This work was supported by the Francis Crick Institute (no grant numbers apply) which receives its core funding from Cancer Research UK (FC001169), the UK Medical Research Council (FC001169), and the Wellcome Trust (FC001169); by the UK Medical Research Council (grant reference MR/FC001169/1); CS is funded by Cancer Research UK (TRACERx), the CRUK Lung Cancer Centre of Excellence, Stand Up 2 Cancer (SU2C), the Rosetrees Trust, NovoNordisk Foundation (ID 16584), the Prostate Cancer Foundation, the Breast Cancer Research Foundation (BCRF), the European Research Council (THESEUS) (no grant numbers apply). HUA acknowledges funding from the Medical Research Council (UK), the Pelican Cancer Foundation Charity, Prostate Cancer UK, St Peters Trust Charity, Prostate Cancer Research Centre the Wellcome Trust, National Institute of Health Research-Health Technology Assessment Programme and the US National Institute of Health-National Cancer Institute (no grant numbers apply). SAQ is funded by a CRUK Career Development Fellowship, CRUK Biotherapeutic Programme Grant, World Wide Cancer Research, and a Cancer Research Institute Investigator Award (no grant numbers apply). ZS is supported by the Breast Cancer Research Foundation, Basser Foundation, Mazzone Foundation, EU FP7 project PREDICT, the Széchenyi Progam, Hungary (KTIA_NAP_13-2014-0021) and the NovoNordisk Foundation (ID 16854). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.

he Breast Cancer Research Foundation, Basser Foundation, Mazzone Foundation, EU FP7 project PREDICT, the Széchenyi Progam, Hungary (KTIA_NAP_13-2014-0021) and the NovoNordisk Foundation (ID 16854). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article. Disclosure The authors have declared no conflicts of interest. Key Message Multiregion analysis of high-risk prostate cancer showed intratumoural heterogeneity of INIF, mutations and copy number. INIF was associated with hypermutation due to mismatch repair (MMR) deficiency and low CD8+/FoxP3+ correlated with activating mutations in the Wnt pathway. INIF, Wnt/βcatenin and MMR should be explored as immunotherapy predictive biomarkers. Supplementary Material Supplementary Figure S1 Click here for additional data file. Supplementary Figure S2 Click here for additional data file. Supplementary Figure S3 Click here for additional data file. Supplementary Figure S4 Click here for additional data file. Supplementary Figure S5 Click here for additional data file. Supplementary Figure S6 Click here for additional data file. Supplementary Figure S7 Click here for additional data file. Supplementary Figure S8 Click here for additional data file. Supplementary Table S1 Click here for additional data file. Supplementary Table S2 Click here for additional data file. Supplementary Table S3 Click here for additional data file. Supplementary Methods Click here for additional data file.

Key Message We show that fully automated computational pathology can accurately determine lymphocyte density in digital slides of pre- and post-treatment breast tumours. Using the ARTemis randomized trial, we validate our previous observations that higher lymphocyte density is associated with pCR and that a paradoxical increase in pre- to post-treatment lymphocyte density is associated with residual disease. Introduction Tumour-infiltrating lymphocytes (TILs) have been widely investigated as a prognostic and predictive biomarker in breast cancer [1]. However, routine assessment of TILs in the clinical setting is hindered by poor reproducibility of their manual pathological evaluation. We previously conducted a systematic analysis of quantitative pathology metrics in the Neo-tAnGo trial [2] and found that pre-treatment tumour lymphocyte density was independently associated with pathological complete response (pCR) [3]. Our observations suggest that computational pathology performs as well as pathologist read scores. Moreover, it is automated, objective and quantitative and may, therefore, facilitate clinical implementation. In addition, we found that a relative increase in lymphocyte density after treatment was inversely associated with pCR and that this relationship significantly differed by taxane sequencing [3], suggesting that in a subset of patients chemotherapy modulates the post-treatment immune microenvironment.

al implementation. In addition, we found that a relative increase in lymphocyte density after treatment was inversely associated with pCR and that this relationship significantly differed by taxane sequencing [3], suggesting that in a subset of patients chemotherapy modulates the post-treatment immune microenvironment. The ARTemis trial showed that the addition of bevacizumab to standard neoadjuvant chemotherapy significantly increased the proportion of patients with a pCR [4], but this has not impacted on disease-free and overall survival [5]. Here, we tested whether our original findings could be validated in this independent study, and have also conducted exploratory analyses of associations with disease-free and overall survival.

cantly increased the proportion of patients with a pCR [4], but this has not impacted on disease-free and overall survival [5]. Here, we tested whether our original findings could be validated in this independent study, and have also conducted exploratory analyses of associations with disease-free and overall survival. Methods Study design ARTemis was a multicentre phase III randomized controlled trial conducted to test whether the addition of bevacizumab to three cycles of docetaxel, followed by three cycles of fluorouracil, epirubicin and cyclophosphamide increased the proportion of patients with a pCR [4]. Women with human epidermal growth factor receptor 2 (HER2)-negative early breast cancer were recruited from May 2009 until January 2013. Of the 800 patients randomized, 781 were available for the primary endpoint analysis. The primary endpoint was pCR (absence of invasive cancer in both the breast and lymph nodes). Here, whether a pCR had occurred was either determined based on central pathology review or, where central review was not possible, on histopathology reports [6]. Details of eligibility and follow-up procedures are provided in the main trial report [4]. The trial was approved by the multicentre and local research ethics committees. All patients provided written, informed consent. The trial was registered at ClinicalTrials.gov (NCT01093235). Supplementary Table S1, available at Annals of Oncology online, details characteristics of patients included in this analysis.

]. The trial was approved by the multicentre and local research ethics committees. All patients provided written, informed consent. The trial was registered at ClinicalTrials.gov (NCT01093235). Supplementary Table S1, available at Annals of Oncology online, details characteristics of patients included in this analysis. Computational pathology Digital whole slide images of haematoxylin and eosin (H&E) stained tissue sections both before and after treatment, were captured using a Hamamatsu Nanozoomer (Hamamatsu City, Shizuoka Pref., Japan). Blinded to all pathological and clinical parameters, we used our computational pathology analysis pipeline to compute cellular metrics from these images. Supplementary Figure S1, available at Annals of Oncology online, summarizes the computational pathology workflow. Briefly, the algorithm segments cell nuclei and, based on a training set of approximately 1000 objects per category, uses machine learning (support-vector-machine) to classify cells into three categories: cancer, stromal and lymphocyte. Finally, based on these classes descriptive cellular metrics are computed, including cellular density. Here lymphocyte density is calculated as follows: for every detected lymphocyte in a section, the average distance R to the 50 nearest lymphocytes (N = 50) is calculated using a K-nearest-neighbour approach.

hocyte, density is estimated as N/(πR2) and the median of this value for all detected lymphocytes is taken as the summary statistic for a given section. The computational pathology approach has been described in detail previously [3] and the analysis code is available at http://www.ast.cam.ac.uk/∼adariush/files/codes/. Statistical analyses We tested for associations between lymphocyte density and pCR using logistic regression, reporting odds ratios (OR) and 95% confidence intervals (95% CI). Lymphocyte density and change in lymphocyte density were modelled as continuous variables. Multivariable models were adjusted for age, randomization arm, histological grade, estrogen receptor (ER) status, tumour size and lymph node status at randomization. Age and histological grade were modelled as continuous variables. Tumour size (<51 mm versus >50 mm) and lymph node status (negative versus positive) were modelled as categorical variables. Associations with categorical clinical variables were tested using Kruskal–Wallis tests. Associations with overall survival (OS) defined as all-cause mortality, and disease-free survival (DFS) were tested using Cox proportional-hazards models, where follow-up commenced from day of surgery. DFS was calculated to date of first relapse (loco-regional or distant, not including DCIS); to date of death in women dying without invasive relapse; or to date of censoring in women alive and disease free. Survival analyses were conducted separately by ER-status to account for known violations of the proportional-hazards assumption [7]. Statistical analyses were conducted using Stata SE version 14.2 (Stata Corp, College Station, TX).

dying without invasive relapse; or to date of censoring in women alive and disease free. Survival analyses were conducted separately by ER-status to account for known violations of the proportional-hazards assumption [7]. Statistical analyses were conducted using Stata SE version 14.2 (Stata Corp, College Station, TX). Results Of the 781 patients included in the ARTemis primary analysis, 609 (78%) had computational pathology and baseline outcome data (Figure 1), where 109 (18%) experienced pCR, a similar proportion to the entire group of 781 patients where 20% experienced pCR. Of these 609, 383 patients had matched pre- and post-treatment samples to calculate change in lymphocyte density; of which 17 (4%) achieved pCR (supplementary Table S1, available at Annals of Oncology online). Median time at risk for OS was 3.1 years (range 0.07–6.3 years). Among the 609 patients, there were 140 DFS events and 98 OS events. Figure 1. Flowchart of patients and samples through analytic stages. Pre-treatment lymphocyte density was associated with ER status (P < 0.001), tumour size (P = 0.003), and histological grade (P < 0.001) (supplementary Figure S2, available at Annals of Oncology online).

Results Of the 781 patients included in the ARTemis primary analysis, 609 (78%) had computational pathology and baseline outcome data (Figure 1), where 109 (18%) experienced pCR, a similar proportion to the entire group of 781 patients where 20% experienced pCR. Of these 609, 383 patients had matched pre- and post-treatment samples to calculate change in lymphocyte density; of which 17 (4%) achieved pCR (supplementary Table S1, available at Annals of Oncology online). Median time at risk for OS was 3.1 years (range 0.07–6.3 years). Among the 609 patients, there were 140 DFS events and 98 OS events. Figure 1. Flowchart of patients and samples through analytic stages. Pre-treatment lymphocyte density was associated with ER status (P < 0.001), tumour size (P = 0.003), and histological grade (P < 0.001) (supplementary Figure S2, available at Annals of Oncology online). Higher pre-treatment lymphocyte density was associated with a greater chance of pCR in unadjusted (OR, 2.93; 95% CI, 1.77–4.85; P < 0.001) and adjusted (OR, 2.13; 95% CI, 1.24–3.67; P = 0.006) analyses (Table 1 and Figure 2). However, there was no association between pre-treatment lymphocyte density and survival (OS or DFS) in either ER-positive or ER-negative disease (supplementary Table S2, available at Annals of Oncology online). Consistent with our previous observations [3], an increase in lymphocyte density between pre- and post-treatment was associated with residual disease (adjusted OR for pCR, 0.1; 95% CI, 0.033–0.31; P < 0.001; Figure 2 and supplementary Table S3, available at Annals of Oncology online). Change in lymphocyte density was not associated with OS or DFS in either ER-positive or ER-negative disease (supplementary Table S2, available at Annals of Oncology online). Table 1 Univariable and multivariable logistic regression of lymphocyte density and clinical covariates against pCR

at Annals of Oncology online). Change in lymphocyte density was not associated with OS or DFS in either ER-positive or ER-negative disease (supplementary Table S2, available at Annals of Oncology online). Table 1 Univariable and multivariable logistic regression of lymphocyte density and clinical covariates against pCR Variable Categories Univariate Multivariate Odds ratio 95% CI P value Observations Odds ratio 95% CI P value Observations Median lymphocyte density Continuous 2.93 1.77–4.85 0.00003 609 2.13 1.24–3.67 0.006 557 Grade 1,2,3 4.82 2.80–8.29 <0.00001 557 2.80 1.58–4.96 0.0004 ER status Negative, Positive 0.19 0.12–0.30 <0.00001 609 0.29 0.18–0.47 <0.00001 Age Continuous 0.97 0.94–0.99 0.007 609 0.98 0.95–1.00 0.06 Node status Negative, Positive 0.69 0.45–1.04 0.08 609 0.65 0.41–1.05 0.08 Chemotherapy BEV+D FEC, D FEC 0.72 0.48–1.10 0.13 609 0.60 0.38–0.97 0.04 Tumour size <51 mm, >50 mm 0.73 0.42–1.26 0.25 609 1.05 0.56–1.97 0.87 a.u., arbitrary units, FEC, fluorouracil, epirubicin and cyclophosphamide; BEV, bevacizumab; pCR, pathological complete response. Figure 2. Association between lymphocyte density, change in lymphocyte density, cellular proportions and chemotherapy response. Observations are ranked by pre-treatment lymphocyte density scores. Lymphocyte density has been rescaled to between zero and one for illustration. a.u., arbitrary units; pCR, pathological complete response; RD, residual disease.

cyte density, change in lymphocyte density, cellular proportions and chemotherapy response. Observations are ranked by pre-treatment lymphocyte density scores. Lymphocyte density has been rescaled to between zero and one for illustration. a.u., arbitrary units; pCR, pathological complete response; RD, residual disease. Discussion In this computational pathology analysis of the ARTemis trial, we have validated our previous observation that higher pre-treatment lymphocyte density is associated with pCR and that an increase in lymphocyte density after treatment is seen in a subset of surgical resection samples with residual disease. Pre-treatment lymphocyte density, while predicting pCR independent of clinical variables, was not associated with survival. Although this contrasts with the findings of past studies [8–10], it should be noted that in these published reports lymphocyte density was not quantified using the approach described here. Our finding should also be interpreted cautiously since analyses were modestly powered due to small sample sizes and limited follow-up time.

his contrasts with the findings of past studies [8–10], it should be noted that in these published reports lymphocyte density was not quantified using the approach described here. Our finding should also be interpreted cautiously since analyses were modestly powered due to small sample sizes and limited follow-up time. Our analyses were limited to tissue morphology in H&E slides. While this is a pragmatic and therefore clinically feasible approach, it overlooks functional differences in infiltrating lymphocytes, which have been shown to influence clinical outcome [11–13]. A second limitation was the incomplete representation of post-treatment specimens. A possible explanation for this, and for the lower proportion of patients with pCR in this subset, is that slides from surgical samples in which a pCR is observed are less likely to be digitized since they do not contain cancer cells. Similarly, we were not able to include all patients recruited to the trial because some slides were not available for digitization. Importantly, the findings validate those of our previous independent study and therefore are more likely to be generalizable.

likely to be digitized since they do not contain cancer cells. Similarly, we were not able to include all patients recruited to the trial because some slides were not available for digitization. Importantly, the findings validate those of our previous independent study and therefore are more likely to be generalizable. Our findings validate pre-treatment lymphocyte density—a computational pathology metric—as a predictor of pCR. This highlights that automated quantitative pathology can perform at a level comparable to pathologist-read scores and may therefore improve the standard histopathological evaluation of tumour samples. Such approaches have the additional advantage of being objective and reproducible. Moreover, our finding that an increase in pre- to post-treatment lymphocyte density is associated with residual disease again highlights perturbations in the immune microenvironment secondary to, and presumably caused by, treatment. We speculate that such a comparative metric could serve as a biomarker to identify patients likely to respond to post-neoadjuvant immunotherapy. Higher pre-treatment lymphocyte density is validated as a predictor of pCR among women with early stage breast cancer. In addition, an increase in lymphocyte density following chemotherapy is again observed to be associated with residual disease. Patients with low pre-treatment lymphocyte density may benefit from more aggressive therapies or enrolment into clinical trials. In addition, immunotherapies may prove more effective following an increase in lymphocyte density following neoadjuvant chemotherapy.

is again observed to be associated with residual disease. Patients with low pre-treatment lymphocyte density may benefit from more aggressive therapies or enrolment into clinical trials. In addition, immunotherapies may prove more effective following an increase in lymphocyte density following neoadjuvant chemotherapy. Funding Cancer Research UK (CRUK/08/037), Roche, Sanofi-Aventis. Disclosure The authors have declared no conflicts of interest. Supplementary Material Supplementary Table S1 Click here for additional data file. Supplementary Table S2 Click here for additional data file. Supplementary Table S3 Click here for additional data file. Supplementary Figure 1 Click here for additional data file. Supplementary Figure 2 Click here for additional data file.

Key Message Detectable circulating tumor DNA postsurgery for high-risk stage II/III melanoma is predictive of relapse and survival, independent of standard prognostic indices. Future adjuvant trials should analyze for treatment effect in this poor prognostic group of patients. Introduction Many patients with loco-regional melanoma will subsequently develop distant metastases; however, current predictors of relapse are relatively crude. Currently, American Joint Committee on Cancer (AJCC) staging of loco-regional melanoma at the time of surgery is used to identify different risk groups and can inform decisions on intensity of follow-up, potential adjuvant therapy and inclusion in clinical trials. However, there are significant limitations because staging is based on a single snapshot of anatomical or histological features and then used as a surrogate for the biological behavior of the tumor over time. Patients are divided into broad prognostic groups without sufficient information to accurately predict the likely outcome on an individual patient basis. This is particularly relevant in light of the recent approval of adjuvant ipilimumab in melanoma. Ipilimumab (10 mg/kg) was associated with 11% improvement in overall survival (OS) at 5 years, from 54.4% to 65.4%, but at the expense of grade 3/4 toxicity in >54% of patients, including five (1.1%) treatment-related deaths [1]. This is in a patient population that is potentially cured by surgery alone.

b in melanoma. Ipilimumab (10 mg/kg) was associated with 11% improvement in overall survival (OS) at 5 years, from 54.4% to 65.4%, but at the expense of grade 3/4 toxicity in >54% of patients, including five (1.1%) treatment-related deaths [1]. This is in a patient population that is potentially cured by surgery alone. It is, therefore, important to develop tools that can accurately identify patients who are at highest risk of progression to stage IV disease. Circulating tumor DNA (ctDNA, the tumor-derived fraction of circulating free DNA or cfDNA) is emerging as a useful measure of tumor burden and prognostic marker in stage IV melanoma [2, 3]. This study aims to determine whether having detectable ctDNA within 12 weeks of surgery carried out with curative intent for high-risk stage II/III disease was associated with worse survival in a subgroup of patients whose tumors were known to have either a BRAF or NRAS mutation. Methods Study design Samples were collected as part of the AVAST-M trial (ISRCTN 81261306), which compared bevacizumab versus placebo in 1343 patients with resected high-risk stage II/III melanoma [4]. This study reported a difference in disease-free interval (DFI) between trial arms [hazard ratio (HR) 0.83; 95% confidence interval (CI) 0.70–0.98, P = 0.03] but no impact on distant metastasis-free interval (DMFI) or OS [4]. Patients with a confirmed BRAF or NRAS mutation were randomly selected from both arms of AVAST-M. Work was carried out in accordance with the Declaration of Helsinki (supplementary Methods, available at Annals of Oncology online).

0.70–0.98, P = 0.03] but no impact on distant metastasis-free interval (DMFI) or OS [4]. Patients with a confirmed BRAF or NRAS mutation were randomly selected from both arms of AVAST-M. Work was carried out in accordance with the Declaration of Helsinki (supplementary Methods, available at Annals of Oncology online). Sample size In this retrospective analysis, 150 patients were determined to provide 80% power to detect a HR of at least 3.5 between patients with undetectable and detectable ctDNA for DFI with a 5% significance level, assuming a 10% marker prevalence and an event rate of 40%. Analysis of ctDNA Mutational status was determined using several different methods including pyro-sequencing of formalin fixed paraffin embedded tissue from the resected primary lesion or involved lymph node, or both where available. Discordant results were repeated in triplicate to provide a consensus result. Baseline plasma samples were taken within 12 weeks (median 8.3 weeks from surgery to blood draw; range 2.4–12 weeks) of surgical clearance for stage IIB, IIC or III melanoma. cfDNA was isolated from up to 2 ml of plasma (individual patient details in supplementary Table S1, available at Annals of Oncology online) using QIAamp Circulating Nucleic Acid kits according to the manufacturer’s instructions (Qiagen, Hilden, Germany) and droplet digital PCR (ddPCR) carried out using a QX200 ddPCR system (Bio-Rad, details in supplementary Methods, available at Annals of Oncology online). ctDNA was defined as detectable if there was  ≥1 copy of mutant DNA detected.

cleic Acid kits according to the manufacturer’s instructions (Qiagen, Hilden, Germany) and droplet digital PCR (ddPCR) carried out using a QX200 ddPCR system (Bio-Rad, details in supplementary Methods, available at Annals of Oncology online). ctDNA was defined as detectable if there was  ≥1 copy of mutant DNA detected. Statistical analysis The baseline ctDNA result (undetectable/detectable) was compared against patient and tumor characteristics [age, gender, AJCC stage, nodal classification, primary melanoma Breslow and ulceration, as well as Eastern Cooperative Oncology Group performance status (ECOG PS)] collected at the time of trial entry using Wilcoxon rank sum test for continuous factors and a chi-square test or Fisher’s exact test with small number for categorical factors. A stepwise logistic regression model was used to identify the independent factors for predicting detectable ctDNA, with a P-value of 0.05 for inclusion and exclusion.

f trial entry using Wilcoxon rank sum test for continuous factors and a chi-square test or Fisher’s exact test with small number for categorical factors. A stepwise logistic regression model was used to identify the independent factors for predicting detectable ctDNA, with a P-value of 0.05 for inclusion and exclusion. DFI, DMFI and OS were calculated from the date of randomization to the trial until date of first recurrence, date of distant metastases and date of death, respectively. The Kaplan–Meier method was used to construct survival curves for differences between DFI, DMFI and OS in patients with detectable ctDNA levels versus undetectable levels and compared using a Cox proportional hazards model to obtain HRs and 95% CIs. Baseline ctDNA (detectable or undetectable) and other factors associated with prognosis (Breslow, ulceration, stage, nodal classification and ECOG PS) were analyzed using univariate and multivariate Cox proportional hazards regression models for DFI, DMFI and OS. Details regarding internal validation of ctDNA and performance modeling can be found in supplementary Methods, available at Annals of Oncology online. All analyses were carried out using the SAS statistical package (version 9.4).

ivariate and multivariate Cox proportional hazards regression models for DFI, DMFI and OS. Details regarding internal validation of ctDNA and performance modeling can be found in supplementary Methods, available at Annals of Oncology online. All analyses were carried out using the SAS statistical package (version 9.4). Results Patient demographics and detection of baseline ctDNA To evaluate the potential for ctDNA to identify melanoma patients at high risk of relapse following surgery with curative intent, we analyzed ctDNA in the plasma from 161 patients carrying either a BRAF or NRAS mutation in their baseline resected tumor (Table 1 and supplementary Tables S1 and S2, available at Annals of Oncology online). Patient demographics are presented in Table 1. Within the cohort, 132 tumors had a p.V600E BRAF mutation and 29 tumors had a p.Q61L/K NRAS mutation. CtDNA was detected in 19 (12%) of the plasma samples (10 from the treatment arm and 9 from the observation arm). Of the 19 positive plasma samples, 15 had a p.V600E BRAF mutation and 4 had a p.Q61L/K NRAS mutation. The Poisson-corrected ctDNA levels ranged from 1.4 to 1608 copies, with a median of 2.8 (supplementary Tables S1 and S2, available at Annals of Oncology online). Table 1. Demographics of patients with detectable or undetectable ctDNA

ive plasma samples, 15 had a p.V600E BRAF mutation and 4 had a p.Q61L/K NRAS mutation. The Poisson-corrected ctDNA levels ranged from 1.4 to 1608 copies, with a median of 2.8 (supplementary Tables S1 and S2, available at Annals of Oncology online). Table 1. Demographics of patients with detectable or undetectable ctDNA Characteristic Total Undetectable ctDNA Detectable ctDNA N (%) N (%) N (%) Age in years, median (range) 52 (19–87) 52 (19–79) 59 (22–87) P value 0.29 Gender Male 77 (48) 70 (49) 7 (37) Female 84 (52) 72 (51) 12 (63) P value 0.31 Breslow of primary tumor ≤2.0 mm 61 (38) 53 (37) 8 (42) >2–4.0 mm 49 (30) 43 (30) 6 (32) >4.0 mm 42 (26) 38 (27) 4 (21) Unknown 9 (6) 8 (6) 1 (5) P value 0.96 Ulceration of primary tumor Present 63 (39) 57 (40) 6 (32) Absent 77 (48) 69 (49) 8 (42) Unknown 21 (13) 16 (11) 5 (26) P value 0.19 Disease stage II 36 (22) 33 (23) 3 (16) IIIA 29 (18) 27 (19) 2 (11) IIIB 59 (37) 51 (36) 8 (42) IIIC 37 (23) 31 (22) 6 (32) P value 0.61 Nodal classification II (No or N/A) 36 (22) 33 (23) 3 (16) III (N1a and N2a) 41 (26) 36 (25) 5 (26) III (other N) 84 (52) 73 (52) 11 (58) P value 0.81 ECOG PS 0 138 (86) 125 (89) 13 (68) 1 22 (14) 16 (11) 6 (32) P value 0.03 Mutation status BRAF V600E 132 (82) 117 (82) 15 (79) NRAS Q61K/L 29 (18) 25 (18) 4 (21) P value 0.75 Trial arm Bevacizumab 81 (50) 71 (50) 10 (53) Observation 80 (50) 71 (50) 9 (47) P value 0.83 Total 161 (100) 142 (88) 19 (12) ctDNA, circulating tumor DNA; ECOG PS, Eastern Cooperative Oncology Group performance status; N, number.

tus BRAF V600E 132 (82) 117 (82) 15 (79) NRAS Q61K/L 29 (18) 25 (18) 4 (21) P value 0.75 Trial arm Bevacizumab 81 (50) 71 (50) 10 (53) Observation 80 (50) 71 (50) 9 (47) P value 0.83 Total 161 (100) 142 (88) 19 (12) ctDNA, circulating tumor DNA; ECOG PS, Eastern Cooperative Oncology Group performance status; N, number. P values were obtained using the Wilcoxon rank sum test for continuous factors and a chi-square test or Fisher's exact test with small number for categorical factors. In univariate analyses of known prognostic factors, only PS was identified as significantly associated with detectable ctDNA (P = 0.03) (Table 1). This was confirmed using a multivariate logistic regression. There was a significantly increased chance of having positive ctDNA in patients with PS 1 compared with 0 (odds ratio = 3.61; 95% CI 1.20–10.82, P = 0.02).

factors, only PS was identified as significantly associated with detectable ctDNA (P = 0.03) (Table 1). This was confirmed using a multivariate logistic regression. There was a significantly increased chance of having positive ctDNA in patients with PS 1 compared with 0 (odds ratio = 3.61; 95% CI 1.20–10.82, P = 0.02). Patient outcomes At a median of 5 years, 21% (95% CI 7–41%) of patients with detectable ctDNA were alive and recurrence free compared with 49% (95% CI 40%–57%) for those with undetectable ctDNA. Of the four patients who did not recur, one patient had stage II disease, one patient stage IIIA and two patients had stage IIIB disease, three patients were on the treatment arm and one patient on the observation arm. All nine patients with >3 mutant copies have recurred (one patient had regional lymph nodes metastases; two patients recurred distantly only and six patients had both loco-regional recurrence and distant metastases). Fifty-two percent (74/142) of patients with undetectable ctDNA have recurred (patterns of the relapses/outcomes are presented in supplementary Tables S3 and S4, available at Annals of Oncology online). Twelve (63%) of the 19 patients with detectable ctDNA are known to have died compared with 49 (35%) of the 142 patients without detectable ctDNA.

of patients with undetectable ctDNA have recurred (patterns of the relapses/outcomes are presented in supplementary Tables S3 and S4, available at Annals of Oncology online). Twelve (63%) of the 19 patients with detectable ctDNA are known to have died compared with 49 (35%) of the 142 patients without detectable ctDNA. Prognostic significance of detectable ctDNA Median DFI was 0.3 years (95% CI 0.1–1.0) in patients with detectable ctDNA compared with 4.2 years (95% CI 2.5–limit not reached) in those where ctDNA was not detected (Figure 1A). There was no significant interaction between trial arm and the ctDNA in predicting DFI (P = 0.60) (supplementary Analysis S5, available at Annals of Oncology online). Patients with detectable ctDNA had significantly increased risk of recurrence compared with those with undetectable ctDNA [HR for detectable ctDNA 3.12; 95% CI 1.79–5.47; P < 0.0001; prognostic separation D statistics (PSDS) = 0.97; standard error (SE) = 0.24; Table 2]. Bootstrapping provided internal validation of ctDNA, with ctDNA being a significant predictor of DFI in 92% of the bootstrapped samples (PSDS = 0.99, SE = 0.24). At 1 year, 26% (95% CI 10%–47%) of the patients with detectable ctDNA were disease free compared with 74% (95% CI 66%–81%) for patients with undetectable ctDNA (Table 2). Sensitivity for predicting relapse was 18% and specificity 95%, with a positive predictive value of 79% and negative predictive value of 51%. Table 2. Univariate Cox proportional hazards regression analysis for prediction of disease-free interval (DFI), distant metastasis-free interval (DMFI) and overall survival (OS)

Sensitivity for predicting relapse was 18% and specificity 95%, with a positive predictive value of 79% and negative predictive value of 51%. Table 2. Univariate Cox proportional hazards regression analysis for prediction of disease-free interval (DFI), distant metastasis-free interval (DMFI) and overall survival (OS) Parameter DFI DMFI OS % DF % DF at 1 year (95% CI) Univariate analysis % DMF % DMF at 1 year (95% CI) Univariate analysis % alive % alive at 1 year (95% CI) Univariate analysis P HR (95% CI) P HR (95% CI) P HR (95% CI) ctDNA <0.0001 <0.0001 0.003 Undetectable 48 74 (66–81) 1.00 58 84 (77–89) 1.00 65 94 (89–97) 1.00 Detectable 21 26 (10–47) 3.12 (1.79–5.47) 26 37 (17–57) 3.22 (1.80–5.79) 37 72 (46–88) 2.63 (1.40–4.96) Breslow 0.51 0.67 0.42 ≤2.0 mm 48 65 (52–76) 1.00 51 75 (62–84) 1.00 57 90 (79–95) 1.00 >2–4.0 mm 49 69 (54–80) 0.94 (0.56–1.58) 59 79 (65–88) 0.79 (0.45–1.40) 67 91 (79–97) 0.76 (0.41–1.42) >4.0 mm 33 69 (53–81) 1.27 (0.77–2.11) 50 81 (65–90) 0.96 (0.55–1.68) 57 95 (82–99) 0.98 (0.54–1.79) Unknown 56 89 (43–98) ND 67 89 (43–98) ND 89 89 (43–98) ND Ulceration 0.94 0.93 0.57 Present 43 71 (58–81) 1.00 54 77 (65–86) 1.00 59 92 (81–96) 1.00 Absent 45 67 (56–77) 0.96 (0.62–1.50) 53 82 (71–89) 0.97 (0.60–1.58) 62 95 (86–98) 0.86 (0.50–1.45) Unknown 48 67 (43–83) ND 57 71 (47–86) ND 71 81 (57–92) ND Disease stage 0.03 0.03 0.14 II 56 86 (69–94) 0.47 (0.25–0.88) 64 91 (76–97) 0.45 (0.23–0.89) 67 97 (81–100) 0.60 (0.28–1.25) IIIA 59 79 (60–90) 0.42 (0.21–0.84) 69 93 (75–98) 0.37 (0.17–0.80) 79 96 (77–99) 0.37 (0.15–0.94) IIIB 41 62 (49–73) 0.75 (0.45–1.25) 51 74 (61–84) 0.67 (0.39–1.16) 56 91 (81–96) 0.86 (0.47–1.58) IIIC 30 54 (37–68) 1.00 38 62 (45–76) 1.00 54 83 (67–92) 1.00 Nodal classification 0.06 0.12 0.38 II (No or N/A) 56 86 (69–94) 0.56 (0.32–0.99) 64 91 (76–97) 0.58 (0.32–1.08) 67 97 (81–100) 0.69 (0.36–1.33) III (N1a and N2a) 51 73 (57–84) 0.64 (0.38–1.07) 61 83 (67–91) 0.64 (0.36–1.14) 68 93 (79–98) 0.70 (0.37–1.32) III (other N) 37 59 (48–69) 1.00 46 71 (60–80) 1.00 57 89 (80–94) 1.00 ECOG PS 0.02 0.01 0.01 0 48 73 (65–80) 0.51 (0.30–0.88) 57 83 (76–88) 0.46 (0.26–0.83) 66 96 (91–98) 0.43 (0.23–0.80) 1 27 45 (24–64) 1.00 36 54 (32–72) 1.00 41 68 (44–83) 1.00

0.38–1.07) 61 83 (67–91) 0.64 (0.36–1.14) 68 93 (79–98) 0.70 (0.37–1.32) III (other N) 37 59 (48–69) 1.00 46 71 (60–80) 1.00 57 89 (80–94) 1.00 ECOG PS 0.02 0.01 0.01 0 48 73 (65–80) 0.51 (0.30–0.88) 57 83 (76–88) 0.46 (0.26–0.83) 66 96 (91–98) 0.43 (0.23–0.80) 1 27 45 (24–64) 1.00 36 54 (32–72) 1.00 41 68 (44–83) 1.00 Figure 1. Kaplan-Meier curves for (A) Disease-free interval (DFI). Median DFI was 0.3 years (95% CI 0.1–1.0) in patients with detectable ctDNA compared with 4.2 years (2.5–limit not reached) in those with undetectable ctDNA. (B) Distant metastasis-free interval (DMFI). Median DMFI was 0.6 years (95% CI 0.2–2.8) with detectable ctDNA compared with the median not reached (95% CI 5.0–limit not reached) for those with undetectable ctDNA. (C) Overall survival (OS). Median OS was 2.9 years (95% CI 0.9–limit not reached) with detectable ctDNA compared with the median not reached for those with undetectable ctDNA (95% CI 6.0–limit not reached).

able ctDNA compared with the median not reached (95% CI 5.0–limit not reached) for those with undetectable ctDNA. (C) Overall survival (OS). Median OS was 2.9 years (95% CI 0.9–limit not reached) with detectable ctDNA compared with the median not reached for those with undetectable ctDNA (95% CI 6.0–limit not reached). Median DMFI was 0.6 years (95% CI 0.2–2.8) with detectable ctDNA, but was not reached even with 5-year follow-up (95% CI 5.0–limit not reached) for those with undetectable ctDNA (Figure 1B). Patients with detectable ctDNA had a significantly increased risk of distant metastatic recurrence compared with those with undetectable ctDNA (Table 2, HR 3.22; 95% CI 1.80–5.79; P < 0.0001, PSDS = 0.99, SE = 0.25). Bootstrapping confirmed ctDNA as a significant predictor of DMFI in 92% of samples (PSDS =1.03, SE = 0.26). At 1 year, 37% (95% CI 17%–57%) of the patients with detectable ctDNA were free of distant metastases compared with 84% (95% CI 77%–89%) for patients with undetectable ctDNA (Table 2). Sensitivity for predicting distant relapse was 20% and specificity 95% with a positive predictive value of 74% and negative predictive value of 61%.

% (95% CI 17%–57%) of the patients with detectable ctDNA were free of distant metastases compared with 84% (95% CI 77%–89%) for patients with undetectable ctDNA (Table 2). Sensitivity for predicting distant relapse was 20% and specificity 95% with a positive predictive value of 74% and negative predictive value of 61%. OS was significantly worse for the 19 patients that had detectable ctDNA compared with the 142 with undetectable ctDNA (Table 2, HR 2.63; 95% CI 1.40–4.96; P = 0.003, PSDS = 0.82, SE = 0.27). Bootstrapping confirmed ctDNA as a significant predictor of OS in 81% of samples (PSDS = 0.83, SE = 0.26). Median OS was 2.9 years (95% CI 0.9–limit not reached) with detectable ctDNA compared with median not reached with 5-year follow-up for those with undetectable ctDNA (95% CI 6.0–limit not reached, Figure 1C). At 1 year, 72% (95% CI 46%–88%) of patients with detectable ctDNA were alive compared with 94% (95% CI 89%–97%) for patients with undetectable ctDNA (Table 2). At 5 years, 33% (95% CI 14%–55%) of patients with detectable ctDNA were alive compared with 65% (95% CI 56%–72%) for those with undetectable ctDNA. Of note, only 12 patients (none in the ctDNA detectable group) received targeted or immune therapy on relapse due to limited availability of these treatments at the time of the study (supplementary Table S3, available at Annals of Oncology online). Results were similar within the BRAF and NRAS mutant subgroups for all above perimeters (data not shown).

the ctDNA detectable group) received targeted or immune therapy on relapse due to limited availability of these treatments at the time of the study (supplementary Table S3, available at Annals of Oncology online). Results were similar within the BRAF and NRAS mutant subgroups for all above perimeters (data not shown). Association of prognostic factors and ctDNA on outcome In univariate analysis ctDNA (P < 0.0001) was significantly more predictive of DFI than either PS (P = 0.02) or disease stage (P = 0.03) (Table 2), and none of the other known prognostic factors (Breslow, ulceration, nodal classification) were significant. Similarly, ctDNA (P ≤ 0.0001) was significantly more predictive of DMFI than PS (P = 0.01) or disease stage (P = 0.03) (Table 2). Critically, in a multivariate Cox proportional hazards regression model, ctDNA remained a significant predictor for DFI (HR 3.26, 95% CI 1.83–5.83, P < 0.0001) and DMFI (HR 3.45, 95% CI 1.88–6.34, P < 0.0001) after adjustment for PS and disease stage (Table 3). For OS, in univariate analyses, ctDNA (P = 0.003) was significantly more predictive than PS (P = 0.01), and disease stage was not predictive, nor were other factors associated with AJCC staging (Table 2). In multivariate analysis, ctDNA remained a significant predictor of OS after adjustment for PS (HR 2.50, 95% CI 1.32–4.74, P = 0.005, Table 3). Finally, to compare the performance of ctDNA in addition to standard prognostic factors, we modeled the prognostic ability of variables associated with AJCC staging (stage, nodal classification, ulceration, Breslow) and then adjusted for ctDNA (Table 4). When adjusted for ctDNA, all indices (PSDS, Nagelkerke’s R2, Calibration shrinkage measure) showed significantly improved prognostic value for DFI, DMFI and OS (Table 4). Table 3. Multivariate Cox proportional hazards regression analysis for prediction of DFI, DMFI and OS

Breslow) and then adjusted for ctDNA (Table 4). When adjusted for ctDNA, all indices (PSDS, Nagelkerke’s R2, Calibration shrinkage measure) showed significantly improved prognostic value for DFI, DMFI and OS (Table 4). Table 3. Multivariate Cox proportional hazards regression analysis for prediction of DFI, DMFI and OS Parameter DFI DMFI OS P HR (95% CI) P HR (95% CI) P HR (95% CI) ctDNA <0.0001 <0.0001 0.005 Undetectable 1.00 1.00 1.00 Detectable 3.26 (1.83–5.83) 3.45 (1.88–6.34) 2.50 (1.32–4.74) ECOG 0.02 0.01 0.02 0 0.52 (0.30–0.89) 0.46 (0.25–0.82) 0.47 (0.25–0.87) 1 1.00 1.00 1.00 Disease stage 0.02 0.02 II 0.46 (0.25–0.87) 0.45 (0.23–0.89) IIIa 0.38 (0.19–0.76) 0.34 (0.16–0.74) IIIb 0.66 (0.39–1.12) 0.59 (0.34–1.04) IIIc 1.00 1.00 Table 4. Model performance measures for the staging variables associated with AJCC classification (stage, Nodal classification, ulceration and Breslow) and the model adjusted for ctDNA Model Outcome Measure AJCC staging variables Adjusted for ctDNA DFI OS DMFI DFI OS DMFI Prognostic separation measure D statistic 0.63 (SE = 0.17) 0.70 (SE = 0.21) 0.53 (SE = 0.18) 0.96 (SE = 0.20) 0.98 (SE = 0.23) 1.01 (SE = 0.22) Predictive ability measure Nagelkerke’s R2 0.093 0.085 0.077 0.17 0.13 0.15 Calibration shrinkage measure 0.43 0.36 0.29 0.65 0.53 0.63 SE, standard error.

OS DMFI DFI OS DMFI Prognostic separation measure D statistic 0.63 (SE = 0.17) 0.70 (SE = 0.21) 0.53 (SE = 0.18) 0.96 (SE = 0.20) 0.98 (SE = 0.23) 1.01 (SE = 0.22) Predictive ability measure Nagelkerke’s R2 0.093 0.085 0.077 0.17 0.13 0.15 Calibration shrinkage measure 0.43 0.36 0.29 0.65 0.53 0.63 SE, standard error. Discussion In the evolving paradigm of effective adjuvant therapy in melanoma, it is essential to develop biomarkers identifying patients at high risk of relapse. Currently, features of the primary tumor such as ulceration, Breslow and number of mitoses in addition to nodal classification and disease stage are standard measures to predict melanoma progression [5]. Furthermore, gene expression profiling has identified subsets that are associated with a poor outcome in stages I–III melanoma; however, patient numbers in these studies were small and have yet to be confirmed in larger cohorts [6, 7].

ication and disease stage are standard measures to predict melanoma progression [5]. Furthermore, gene expression profiling has identified subsets that are associated with a poor outcome in stages I–III melanoma; however, patient numbers in these studies were small and have yet to be confirmed in larger cohorts [6, 7]. In this study, we showed that detecting ctDNA in plasma taken within 12 weeks of curative intent surgery is highly predictive of relapse in patients with stage II/III melanoma. The majority of patients with detectable ctDNA relapsed within 1 year of surgery suggesting that ctDNA in the plasma can reveal occult metastatic disease that is not evident on radiological imaging. Notably, we were able to identify melanoma patients at high risk of both distant metastatic relapse and local recurrence, which is consistent with studies showing that ctDNA can signal micrometastatic disease after neoadjuvant chemotherapy postsurgical resection in breast cancer, and following surgery for stage II colorectal cancer [8, 9].

to identify melanoma patients at high risk of both distant metastatic relapse and local recurrence, which is consistent with studies showing that ctDNA can signal micrometastatic disease after neoadjuvant chemotherapy postsurgical resection in breast cancer, and following surgery for stage II colorectal cancer [8, 9]. Critically, our findings were independent of standard staging indices, demonstrating the value of this approach in melanoma. It is reported that PS is an independent predictor for DFI, DMFI and OS in the AVAST-M study population [4]. Although the reasons for this intriguing observation are not known, even when adjusted for PS in multivariate analysis, ctDNA was significant in predicting DFI, DMFI and OS. Moreover, in this cohort, AJCC variables were a poor predictor of relapse, but when we created a model in which the standard AJCC variables were adjusted for ctDNA, the performance improved significantly. The test was specific, but not sensitive and therefore should be seen as an adjunct to current AJCC staging when discussing risk of relapse and adjuvant options for the individual patient.

, but when we created a model in which the standard AJCC variables were adjusted for ctDNA, the performance improved significantly. The test was specific, but not sensitive and therefore should be seen as an adjunct to current AJCC staging when discussing risk of relapse and adjuvant options for the individual patient. The patients evaluated in this study were treated in an era where access to immune and targeted therapies was limited. It will be important to determine whether outcome can be improved in this extremely poor prognostic subgroup using targeted and immune therapies in future clinical trials. In stage IV melanoma, baseline ctDNA levels in patients treated with both targeted and immune therapies have been shown to correlate with inferior survival and disease burden [2, 3]. Taken together, these data show that ctDNA levels during the course of disease reflect disease biology and are associated with patient outcome.

ge IV melanoma, baseline ctDNA levels in patients treated with both targeted and immune therapies have been shown to correlate with inferior survival and disease burden [2, 3]. Taken together, these data show that ctDNA levels during the course of disease reflect disease biology and are associated with patient outcome. A minimally invasive, blood test based on ddPCR, which is simple, relatively inexpensive and could be carried out within 5 days, is particularly advantageous in the clinical setting especially when compared with next generation sequencing of tumor material, which is time-consuming, costly and requires specialist skills to perform and analyze. ddPCR is extremely sensitive and can reach detection sensitivities of approximately 0.01% [10] and we were able to demonstrate significance with only 2 ml of patient plasma. Moreover, our proof-of-principle study focused on the driver mutations BRAF and NRAS, which account for up to 70% of melanomas, and are well suited to this purpose because of the low likelihood of clonal diversity with trunk mutations such as these in melanoma. Furthermore, driver mutations usually have the highest variant allele frequency, which improves the sensitivity of the test. To analyze ctDNA in patients without BRAF or NRAS mutations, next generation sequencing of the primary tumor/lymphadenectomy can be used to identify the driver/trunk mutations in individual patients and create bespoke panels.

usually have the highest variant allele frequency, which improves the sensitivity of the test. To analyze ctDNA in patients without BRAF or NRAS mutations, next generation sequencing of the primary tumor/lymphadenectomy can be used to identify the driver/trunk mutations in individual patients and create bespoke panels. Critically, the strength of this retrospective study is that it allows sufficient follow-up to identify the patients who relapsed, using small amounts of sample. Clearly, prospective studies in the adjuvant setting will be needed to validate these findings and examine treatment effect of new agents. Based on the findings of others [8] and the low sensitivity seen in this assay, it is unlikely that a single time-point following surgery will identify all patients who are going to relapse, but we propose that longitudinal sampling will resolve this issue and improve the sensitivity. Longitudinal sampling has identified treatment relapse before radiological imaging in stage IV melanoma providing a rationale for such an approach [3]. Furthermore, longitudinal sampling will reduce the likelihood of false-positive results. For one of the patients with detectable ctDNA within 12 weeks who did not relapse within the follow-up period, subsequent analysis was found to be negative for ctDNA at 3, 6, 12, 18, 24, 36, 48 and 60 months, and therefore confirmatory testing should be mandatory for this assay to be part of clinical decision-making.

. For one of the patients with detectable ctDNA within 12 weeks who did not relapse within the follow-up period, subsequent analysis was found to be negative for ctDNA at 3, 6, 12, 18, 24, 36, 48 and 60 months, and therefore confirmatory testing should be mandatory for this assay to be part of clinical decision-making. The ability to predict progression to stage IV disease is extremely important in light of recent findings that immune checkpoint inhibition improves OS in stage III melanoma [1]. Detection of ctDNA allows identification of a subgroup of patients at high risk of early relapse and inferior survival, allowing stratification of patients to adjuvant regimens associated with higher toxicity but greater potential for efficacy [11]. Taken at a single time-point following surgery, it can add to AJCC staging in informing individual prognosis and therefore discussion regarding risks and benefits of adjuvant therapy, while longitudinal sampling will likely improve the ability to detect disease progression before radiological imaging. We advocate that our findings be confirmed in clinical trials investigating treatment responses in this population in order to evaluate whether it is also a predictive biomarker for response to immune or targeted therapy. Supplementary Material Supplementary Analysis Click here for additional data file. Supplementary Methods Click here for additional data file. Supplementary Tables Click here for additional data file. Acknowledgement The authors acknowledge to the patients for donating the samples.

The ability to predict progression to stage IV disease is extremely important in light of recent findings that immune checkpoint inhibition improves OS in stage III melanoma [1]. Detection of ctDNA allows identification of a subgroup of patients at high risk of early relapse and inferior survival, allowing stratification of patients to adjuvant regimens associated with higher toxicity but greater potential for efficacy [11]. Taken at a single time-point following surgery, it can add to AJCC staging in informing individual prognosis and therefore discussion regarding risks and benefits of adjuvant therapy, while longitudinal sampling will likely improve the ability to detect disease progression before radiological imaging. We advocate that our findings be confirmed in clinical trials investigating treatment responses in this population in order to evaluate whether it is also a predictive biomarker for response to immune or targeted therapy. Supplementary Material Supplementary Analysis Click here for additional data file. Supplementary Methods Click here for additional data file. Supplementary Tables Click here for additional data file. Acknowledgement The authors acknowledge to the patients for donating the samples. Funding Cancer Research UK Manchester Institute (C5759/A20971); the Wellcome Trust (100282/Z/12/Z). AVASTM trial and PROM grant funded by Cancer Research UK (C7535/A6408 and C2195/A8466). Disclosure The authors have declared no conflicts of interest.

Introduction Colorectal cancer (CRC) is the third most commonly diagnosed neoplasm and the third leading cause of cancer death in the United States [1], whereas it is the most common cause of cancer death after lung cancer in Europe [2]. Approximately one-quarter of patients with CRC present with metastatic disease (mCRC) at diagnosis (synchronous disease), and ∼40% of patients develop metachronous metastases after treatment, contributing to the high mortality rate associated with this disease. Systemic therapy is the mainstay of treatment for mCRC, and there are currently several approved drugs for the management of mCRC, which include chemotherapy agents, small molecules, and monoclonal antibodies.

metachronous metastases after treatment, contributing to the high mortality rate associated with this disease. Systemic therapy is the mainstay of treatment for mCRC, and there are currently several approved drugs for the management of mCRC, which include chemotherapy agents, small molecules, and monoclonal antibodies. Anti-epidermal growth factor receptor (EGFR) targeted therapies, such as cetuximab and panitumumab, do not have universal efficacy. In mCRC, there is a high frequency of Kirsten rat sarcoma (KRAS) viral oncogene homolog mutations, which are responsible for resistance to these monoclonal antibodies [3]. Traditionally, prescribing anti-EGFR therapy required assessment of mutations in KRAS Exon 2, which occur in ∼40% of patients. Additional research has expanded the spectrum mutations in KRAS and neuroblastoma RAS (NRAS) viral oncogene homolog genes that predict a lack of efficacy to these treatments, including mutations in KRAS exon 3 (codons 59 and 61) and exon 4 (codons 117 and 146) and mutations in NRAS exons 2, 3, and 4 [4–8]. Accordingly, clinical practice guidelines in the United States and Europe have been updated to reflect the need for broader RAS testing, in order to refine the most appropriate patient population to receive anti-EGFR therapy [9–11].

and 61) and exon 4 (codons 117 and 146) and mutations in NRAS exons 2, 3, and 4 [4–8]. Accordingly, clinical practice guidelines in the United States and Europe have been updated to reflect the need for broader RAS testing, in order to refine the most appropriate patient population to receive anti-EGFR therapy [9–11]. Molecular analysis of genomic alterations has been traditionally performed on archival formalin-fixed paraffin-embedded (FFPE), but this procedure entails limitations in terms of poor quality of the extracted DNA and lack of standardization of testing methods. Moreover, tumor heterogeneity and clonal molecular evolution throughout therapy confound tissue sampling and there is no consensus as to whether analysis of the primary tumor is sufficient or whether a metastatic lesion should be studied in patients with metastases [12–14]. As recurring tissue biopsies are not routinely performed in patients with advanced, the use of circulating tumor DNA (ctDNA) to detect tumor-specific mutations has emerged as an important tool to inform clinical decisions according to tumor dynamics. This procedure is often referred to as a ‘liquid biopsy’ and represents an attractive strategy for better patient selection and treatment individualization throughout multiple lines of therapy [15].

etect tumor-specific mutations has emerged as an important tool to inform clinical decisions according to tumor dynamics. This procedure is often referred to as a ‘liquid biopsy’ and represents an attractive strategy for better patient selection and treatment individualization throughout multiple lines of therapy [15]. Concept and methodological features of liquid biopsy Liquid biopsy allows the analysis of several blood-based biomarkers, i.e. circulating tumor cells, protein molecules, mRNA, microRNA, and cell-free DNA. Methodological limitations seem to be more important for CTCs than for cell-free circulating nucleic acids [16]. In fact, circulating DNA has been investigated for decades as a potential marker for screening, diagnosis, prognosis, and monitoring treatment response. Indeed, the release of DNA into the bloodstream from apoptotic or necrotic tumor cells has been reported in different neoplasms [17–19]. Bettegowda et al. showed that the frequency of cases with detectable ctDNA was proportional to the stage of the disease. Although ctDNA was detected in 100% of patients with stage IV malignancies including colorectal, bladder, and ovarian cancer the rate of detection of ctDNA in early-stage malignancies was ∼50% [20]. They also compared the quantities of ctDNA and CTCs in the same blood sample from patients with solid tumors and showed that the level of ctDNA was always higher than that of CTCs. In 13 of 16 patients, ctDNA levels were relatively high, whereas no CTCs at all could be detected [20].

y-stage malignancies was ∼50% [20]. They also compared the quantities of ctDNA and CTCs in the same blood sample from patients with solid tumors and showed that the level of ctDNA was always higher than that of CTCs. In 13 of 16 patients, ctDNA levels were relatively high, whereas no CTCs at all could be detected [20]. One of the main advantages of ctDNA analyses is its high degree of specificity, since somatic mutations found in ctDNA are not present in normal cell-free DNA. However, the analysis of ctDNA is challenging and requires highly sensitive techniques. Advances in the pre-analytical stage can improve the performance of ctDNA detection, for example ctDNA should be extracted from plasma rather than serum due to its lower concentration of background wild-type DNA in plasma [21]. After years of investigation there are now several methods for detecting ctDNA, and platforms primarily based on digital PCR and next-generation sequencing (NGS) are widely used, though each has inherent advantages and disadvantages in terms of sensitivity, specificity, throughput, and breadth of mutational coverage (Table 1). Readers interested in further details on the principles and different characteristics of these techniques are invited to consult the following recent review [22]. Among them, an emulsion PCR-based technology platform known as BEAMing (Beads, Emulsions, Amplification, and Magnetics) is the first liquid biopsy test that has been clinically validated [20, 23–26] and is CE Mark’d (OncoBEAM RAS CRC test) for testing the RAS-mutation status in CRC patients (OncoBEAM RAS CRC IVD IfU). Table 1. Overview of technologies used for detection of circulating tumor DNA (ctDNA)

ons, Amplification, and Magnetics) is the first liquid biopsy test that has been clinically validated [20, 23–26] and is CE Mark’d (OncoBEAM RAS CRC test) for testing the RAS-mutation status in CRC patients (OncoBEAM RAS CRC IVD IfU). Table 1. Overview of technologies used for detection of circulating tumor DNA (ctDNA) Method of detection Test Detection limit (% ctDNA) Advantage(s) Limitation(s) Digital PCR Droplet-based digital PCR ∼0.01% High sensitivity Detection of limited genomic loci/single-nucleotide variants BEAMing Ease of use (available kits) Microfluidic digital PCR Clinically validated Targeted deep sequencing (with NGS) SafeSeq/TamSeq/Ion-AmpliSeq/Ontarget/CAPP-Seq ∼0.01–2.0% High sensitivity For selected alterations across targeted regions Relatively inexpensive Need of assay personalization (except for CAPP-Seq) Whole-genome sequencing (with NGS) Digital karyotyping/PARE ∼1% Broad application without personalization Expensive Low sensitivity Whole-exome sequencing (with NGS) Currently for research purposes ∼5% Broad application without personalization Expensive Low sensitivity Lack of standarization PCR, polymerase chain reaction; BEAMing, beads, emulsion, amplification, and magnetics; NGS, next-generation sequencing; SafeSeq, safe sequencing system; TamSeq, tagged amplicon deep sequencing; CAPP-Seq, cancer personalized profiling by deep sequencing; PARE, personalized analysis of rearranged end.

of standarization PCR, polymerase chain reaction; BEAMing, beads, emulsion, amplification, and magnetics; NGS, next-generation sequencing; SafeSeq, safe sequencing system; TamSeq, tagged amplicon deep sequencing; CAPP-Seq, cancer personalized profiling by deep sequencing; PARE, personalized analysis of rearranged end. OncoBEAM platform: analytical and clinical validity in CRC The OncoBEAM platform addresses the technical challenge of identifying rare DNA molecules with enhancements to the PCR process known as BEAMing. DNA amplification is used to increase the quantity of DNA species of interest and facilitate measurement. Conventional PCR carries out one reaction per single sample and provides one signal. BEAMing involves partitioning of the PCR process into many individual reactions to provide high resolution detection of rare DNA sequences (e.g. ones having a RAS mutation). In performing BEAMing, each DNA molecule in a sample is amplified and converted into a single magnetic particle to which thousands of copies of DNA identical in sequence to the original DNA molecule are bound. The resulting magnetic particles (beads) are a one-to-one representation of the starting DNA molecules in a given sample and are compartmentalized into water-in-oil microemulsions such that each emulsion contains only one DNA molecule template. A subsequent amplification step is performed within each emulsion. Flow cytometry is then applied to assess the variation within the original population of DNA molecules. BEAMing is designed to assess hundreds of millions of individual DNA molecules with standard laboratory equipment. BEAMing can be used for the identification and subsequent quantification of cell-free tumor DNA molecules, which are differentiated from normal (wild-type) DNA by somatic mutations [27]. As a result, this technology is highly sensitive and reliably detects mutated ctDNA even when the mutation exists as a rare event, at a level as low as 0.01% [28].

entification and subsequent quantification of cell-free tumor DNA molecules, which are differentiated from normal (wild-type) DNA by somatic mutations [27]. As a result, this technology is highly sensitive and reliably detects mutated ctDNA even when the mutation exists as a rare event, at a level as low as 0.01% [28]. The OncoBEAM assays are able to analyze hotspot mutations and the number of mutations detected varies per tumor type analyzed. The OncoBEAM RAS CRC Kit detects 34 mutations in codons 12, 13, 59, 61, 117, and 146 of the KRAS and NRAS oncogenes. For other relevant oncogenes, CRC guidelines recommend BRAF mutational analysis to be performed solely for prognostic stratification [29]. A second-generation version of the OncoBEAM CRC kit will incorporate BRAF in order to ensure patients with BRAF mutations are segregated from the RAS wild-type patients.

oncogenes. For other relevant oncogenes, CRC guidelines recommend BRAF mutational analysis to be performed solely for prognostic stratification [29]. A second-generation version of the OncoBEAM CRC kit will incorporate BRAF in order to ensure patients with BRAF mutations are segregated from the RAS wild-type patients. OncoBEAM RAS testing has been investigated in several head to head studies and these have demonstrated high concordance between blood-based testing versus standard tissue-based RAS testing methods [26, 30–32]. In particular, OncoBEAM was utilized to evaluate the concordance of RAS status between plasma and tissue in a large cross-section of mCRC patients [26]. In this evaluation, two geographically separated cohorts of CRC patients with metastatic disease from Germany and Australia were selected to represent the intended use population for anti-EGFR therapy, with plasma taken at the time of tissue biopsy or surgical resection. Tissue RAS status was evaluated according to the local standard of care and compared with plasma OncoBEAM RAS testing results. This initial study showed that the concordance between plasma RAS status determined by BEAMing versus that obtained by FFPE testing was 91.8% for stage IV CRC patients. A larger performance evaluation from six centers in Europe showed that the overall concordance between OncoBEAM RAS CRC testing and standard of care tissue testing for 238 patients was 93.3% [33]. The RAS mutation status determined by OncoBEAM from plasma versus the tissue reference method is summarized in Table 2. Plasma RAS mutations were found in 112/121 KRAS mutant cases determined by tissue-based testing (92.6% of positive percent agreement) and no RAS mutations were found in 110/117 not mutant cases according to tissue testing (94% of negative percent agreement. Overall, RAS mutations were detected in 51% of tumor-tissue samples and in 50% of plasma samples [33]. The frequency of RAS mutations in patients investigated in this study was in agreement with the results of other groups performing expanded RAS analysis [34]. All this evidence provides a high level of confidence that the clinical performance of plasma RAS testing using OncoBEAM RAS is comparable with FFPE tissue testing and can be useful in a clinical setting to select advanced CRC patients for anti-EGFR therapy. Table 2. Concordance of RAS mutation status: plasma ctDNA versus tumor tissue analyses

s a high level of confidence that the clinical performance of plasma RAS testing using OncoBEAM RAS is comparable with FFPE tissue testing and can be useful in a clinical setting to select advanced CRC patients for anti-EGFR therapy. Table 2. Concordance of RAS mutation status: plasma ctDNA versus tumor tissue analyses Tumor-tissue RAS result RAS Mutant WT Total PPA (95% CI) NPA (95% CI) OPA (95% CI) Plasma ctDNARAS result Mutant 112 7 119 100×112/121=92.6% (86%, 96%) 100×110/117=94.0% (88%, 97%) 100×222/238=93.3% (89%, 96%) WT 9 110 119 Total 121 117 238 RAS, Fisher’s exact test was used to test for a relationship between RAS mutation-positive results in plasma versus tissue samples (positive percent agreement, PPA), WT results in plasma versus tissue samples (negative percent agreement, NPA), and the combination of RAS mutation-positive and WT results in plasma versus tissue samples (overall percent agreement/concordance, OPA). Data from reference [33].

ive results in plasma versus tissue samples (positive percent agreement, PPA), WT results in plasma versus tissue samples (negative percent agreement, NPA), and the combination of RAS mutation-positive and WT results in plasma versus tissue samples (overall percent agreement/concordance, OPA). Data from reference [33]. The minimal differences in RAS mutation status between plasma and tissue may be attributed to intra- or inter-tumor molecular heterogeneity or variability in tissue techniques. While a false negative result (no mutation detected in a patient with a RAS mutant tumor) in either tissue or plasma may lead to the inappropriate assignment of first-line treatment anti-EGFR therapy (with the risk of detrimental outcome with anti-EGFR exposure, the risk of toxicity and/or allergic reaction and the high expenses associated with these agents), this risk is largely mitigated by the high frequency of radiographic surveillance in mCRC patients undergoing therapy. A false positive result (mutation detected in a patient that does not have a RAS mutation) may lead to a patient forgoing first-line anti-EGFR therapy and receiving a less effective chemotherapy. This risk may be mitigated by the longitudinal evaluation of plasma RAS status during therapy in order to inform subsequent treatment lines [32]. The evaluation of the outcomes of false positive and negative cases with OncoBEAM RAS testing is an area of active investigation. As a more comprehensive elucidation of ever-changing tumor dynamics and host biological influences emerges, this information will lead to a better understanding of biological factors influencing ctDNA testing and subsequent implementation into clinical practice.

OncoBEAM RAS testing is an area of active investigation. As a more comprehensive elucidation of ever-changing tumor dynamics and host biological influences emerges, this information will lead to a better understanding of biological factors influencing ctDNA testing and subsequent implementation into clinical practice. BEAMing advantages and current usefulness The current standard of care for CRC patients involves tissue mutation testing to determine whether a patient will benefit from the administration of anti-EGFR therapy. Blood-based RAS mutation testing should address the following unmet needs associated with tissue testing: Delayed processing: Plasma testing is favorable in newly diagnosed patients or when a patient’s mutation status is unknown and expedited results are required for therapy selection (reduction of turnaround time to <7 days). Recently, a quality survey found that half of all European participating laboratories exceeded the required turnaround time of 14 days for complete RAS tissue-based testing [35]. These delays in processing tissue may prohibit the timing of molecular testing results being available at the time of a patient’s visit, thus delaying the prescription of optimal therapy.

all European participating laboratories exceeded the required turnaround time of 14 days for complete RAS tissue-based testing [35]. These delays in processing tissue may prohibit the timing of molecular testing results being available at the time of a patient’s visit, thus delaying the prescription of optimal therapy. Invasive biopsies: Obtaining a blood sample is minimally invasive to the patient compared with obtaining a tissue sample and provides an alternative when tumor tissue is insufficient for molecular testing. Need of repeated biopsies in case of metastatic disease: In a patient with recurrent disease, archival tissue may not reflect current mutational status and it is common practice to not perform a tissue biopsy at time of recurrence. A liquid biopsy test can provide real-time information about the current mutational status of the primary tumor and can serve as an alternative to archival tissue, which suffers from degradation and may reflect an ‘archival’ mutation profile that may not be not representative of the current status [36]. Additionally, a systemic assessment of tumor mutational status is not practical or feasible in a patient with metastatic disease. Evidence of molecular heterogeneity among primary and metastatic sites [14, 37] suggest that a ctDNA assessment of the mutation status of systemic tumor burden can more precisely guide targeted systemic therapy [14].

ssessment of tumor mutational status is not practical or feasible in a patient with metastatic disease. Evidence of molecular heterogeneity among primary and metastatic sites [14, 37] suggest that a ctDNA assessment of the mutation status of systemic tumor burden can more precisely guide targeted systemic therapy [14]. When compared with other platforms including NGS, OncoBEAM assays detect mutations at a higher level of sensitivity with a focus on clinically actionable gene mutations. While NGS does have the capability to detect additional gene mutations, the clinical utility of these additional variants is often not well established or aligned to an approved therapeutic; an additional limitation is its low sensitivity. OncoBEAM blood-based assays offer rapidity of results compared with NGS, with OncoBEAM RAS testing delivering results within 2–3 days, a critical advantage when treating patients with advanced or progressing disease. Moreover, a recent review focused on different technologies for molecular classification of cancers established that BEAMing is associated with the lowest costs comparing with other digital PCR or NGS techniques [22]. Since OncoBEAM RAS test analyzes an expanded panel of 34 RAS mutations in KRAS and NRAS, it provides RAS testing in accordance with the current standard of care molecular testing guidelines as defined as the ESMO and NCCN committees.

ociated with the lowest costs comparing with other digital PCR or NGS techniques [22]. Since OncoBEAM RAS test analyzes an expanded panel of 34 RAS mutations in KRAS and NRAS, it provides RAS testing in accordance with the current standard of care molecular testing guidelines as defined as the ESMO and NCCN committees. BEAMing future applications Real-time monitoring of treatment Minimally invasive diagnostic assays performed by BEAMing are ideally suited for disease monitoring throughout therapy. Various studies have demonstrated that ctDNA measurements can be used to reliably monitor tumor dynamics in subjects undergoing surgery or chemotherapy [38, 39]. In these studies, ctDNA levels were compared with imaging and biomarkers such as carcinoembryonic antigen (CEA) in predicting response to treatment. Results demonstrated that BEAMing was more sensitive than imaging, CEA, or total DNA to provide an earlier indication of response to treatment. This is further supporting evidence to show that not only does ctDNA levels correlate with changes in tumor burden, they provide a more immediate and sensitive measure of response than either imaging or CEA [28, 39]. Vidal et al. [32] examined the utility of OncoBEAM RAS CRC ctDNA testing to monitor the efficacy of response to treatment taking serial blood draws from 21 patients with baseline RAS mutations undergoing systemic therapy. Analysis of RAS ctDNA at the time of a first CT scan (8–12 weeks of treatment) revealed a dramatic decrease in plasma RAS mutant allele fraction (MAF) in responding patients with a median of 100%. MAF percentage of change was significantly lower in patients that progressed at first evaluation of response compared with patients with clinical benefit (132% increase versus 99% reduction, respectively, P = 0.027). The authors concluded that RAS plasma mirrored clinical and radiological response to chemotherapy drugs and was an early predictor of response [32].

wer in patients that progressed at first evaluation of response compared with patients with clinical benefit (132% increase versus 99% reduction, respectively, P = 0.027). The authors concluded that RAS plasma mirrored clinical and radiological response to chemotherapy drugs and was an early predictor of response [32]. Detection of resistance mechanisms A blood-based approach enables detection of emergence or disappearance of genetic mutations linked to resistance or susceptibility to targeted therapies [24]. Emergence of RAS mutations is a frequent mechanism of resistance in mCRC patients treated with anti-EGFR therapy. A recent study utilizing the OncoBEAM RAS assay provides preliminary evidence to support the role of monitoring emerging RAS in CRC patients receiving anti-EGFR therapy [40]. In this study, 62 of 70 (89%) of mCRC patients who initially responded to anti-EGFR therapy and chemotherapy were found to develop resistance. At the time of resistance, acquired mutations in KRAS were detected by BEAMing in the plasma of 27/62 (44%) patients. In order to evaluate whether newly detected KRAS mutations were already present in treatment-naïve primary tumors as undetectable low frequency clones or were truly acquired mutations, the investigators utilized the OncoBEAM RAS assay on the original tumor tissue to re-analyze archival samples from 20 of 27 patients for traces of KRAS clones. This analysis revealed that seven (35%) patients had low-frequency KRAS mutations and that overall, these seven patients had a poorer prognosis than those determined to be truly KRAS wild type (median progression-free survival: 3.0 versus 8.0 months, P = 0.0004) [40]. In another study [8], researchers conducted a post hoc investigation of patients enrolled in the phase III CRYSTAL study in order to determine the treatment effects of cetuximab plus FOLFIRI versus FOLFIRI alone for patients whose tumors had mutations in one of the less common RAS mutations (located in KRAS codons 61, 117 and 146, or NRAS codons 12, 13, and 61). Mutational status for an additional 26 RAS mutations was determined by the OncoBEAM RAS assay for these patients. Using a 5% mutant/wild-type cutoff, an additional 63 patients (14.7%) were classified as RAS mutant positive; 86 patients (20%) were identified when a less stringent cutoff ≥ 0.1% mutant/wild-type sequences was used.

status for an additional 26 RAS mutations was determined by the OncoBEAM RAS assay for these patients. Using a 5% mutant/wild-type cutoff, an additional 63 patients (14.7%) were classified as RAS mutant positive; 86 patients (20%) were identified when a less stringent cutoff ≥ 0.1% mutant/wild-type sequences was used. When considering efficacy outcomes between treatment groups, there was no clear benefit for the addition of cetuximab to FOLFIRI treatment in patients with either ≥5% RAS mutant/wild-type cutoff. However, patients with a RAS mutant allele fraction <5% were able to derive benefit from the addition of cetuximab to FOLFIRI [8]. The clinical application of OncoBEAM RAS CRC assay for monitoring of acquired resistance to anti-EGFR therapy in routine clinical practice has been recently evaluated by Vidal et al. [32]. Emergence of RAS mutations was detected in 7/18 patients (39%) showing disease progression after an initial complete response, partial response or stable disease for >16 weeks, and in three cases, different RAS mutations were concomitantly detected [32]. Moreover, Toledo et al. [41] performed a prospective validation of the BEAMing technique to monitor newly diagnosed KRAS wt mCRC patients who received a standard FOLFIRI-cetuximab regimen. They showed that the patients who initially responded to anti-EGFR therapy but later acquired resistance presented intermediate and gradually increasing levels of circulating mutant alleles, whereas patients with long-term responses maintained a wt circulating status throughout the anti-EGFR therapy. As data continues to emerge showing that the identification of RAS mutations in the plasma of relapsed patients indicates resistance to anti-EGFR therapy, regular analyses could inform clinical decision making and may offer patients the opportunity to benefit from therapies designed to overcome resistance, which in every instance is a more cost-effective approach compared with tissue-based clinical management.

relapsed patients indicates resistance to anti-EGFR therapy, regular analyses could inform clinical decision making and may offer patients the opportunity to benefit from therapies designed to overcome resistance, which in every instance is a more cost-effective approach compared with tissue-based clinical management. Early detection of relapse and/or residual disease The potential of BEAMing technology in this regard was demonstrated in a proof of concept study that provides insight into the utility of ctDNA for monitoring treatment response and predicting disease recurrence following surgery [23]. In this study, mCRC patients were followed throughout both surgery and systemic treatment. A total of 18 patients underwent 22 surgeries of which 17 were complete tumor resections and 5 were incomplete resections. A BEAMing assay specific to the molecular profile of each patient was designed in order to perform plasma mutation assessments. Plasma samples were collected from 18 patients on the date of surgery, 24-h post-surgery, and then at regular intervals between 13 and 56 days. ctDNA levels decreased dramatically in all patients after surgery but were detectable in the first follow-up visit (within 13–56 days) in 16/20 instances (plasma samples were only available in 20 instances—not 22). In all but one of these instances the patients’ diseases recurred. In contrast, in 4 patients in whom mutant DNA was undetectable, no recurrence occurred [23]. The pilot study of Misale et al. [39] showed that emergence of KRAS mutations was detectable in plasma as early as 10 months before the documentation of disease progression by radiological assessment. In line with these results, ctDNA-based detection preceded clinical detection of metastasis in 86% of breast cancer patients with an average lead time of 11 months (range 0–37 months), whereas patients with long-term disease-free survival had undetectable ctDNA postoperatively [42].

gression by radiological assessment. In line with these results, ctDNA-based detection preceded clinical detection of metastasis in 86% of breast cancer patients with an average lead time of 11 months (range 0–37 months), whereas patients with long-term disease-free survival had undetectable ctDNA postoperatively [42]. A pressing question is how often to perform a liquid biopsy [8], while this is not yet established, the most practical frequency appears to be every 4–6 weeks interspersed and/or in conjunction with radiological scans. Serial ctDNA measurements could complement routine imaging-based assessments in evaluation of disease bulk [43], response to chemotherapeutic agents, re-challenge with anti-EGFR therapy [44], and detection of residual disease after surgical resection of the tumor [45]. In the last case, this approach could aid to select treatment strategies in patients with residual disease that could benefit from adjuvant chemotherapy and intensive surveillance. Early detection of cancer It is worthwhile to highlight the potential application of OncoBEAM RAS testing as a screening tool for pre-neoplastic lesions or localized neoplastic disease. Bettegowda et al. [20] showed that ctDNA is more readily detected in the blood of patients with more invasive tumors rather than earlier tumors, but even so the sensitivity for detection was 50% in early-stage disease. However, only RAS mutant patients would benefit from the use of this test as screening method.

sease. Bettegowda et al. [20] showed that ctDNA is more readily detected in the blood of patients with more invasive tumors rather than earlier tumors, but even so the sensitivity for detection was 50% in early-stage disease. However, only RAS mutant patients would benefit from the use of this test as screening method. Concluding remarks The clinical utility of a diagnostic test pertains to the ability of the test to provide new information that leads to a clinical benefit. The importance of RAS testing in mCRC patient selection for anti-EGFR therapy is well established, but due to known limitations of tissue-based testing and delays in clinical turnaround time, a significant proportion of patients are being treated or receiving delayed treatment without the information that RAS testing can provide. The clinical utility of the OncoBEAM RAS test allows patients to benefit from international guideline-recommended expanded RAS testing with rapid turnaround times.

turnaround time, a significant proportion of patients are being treated or receiving delayed treatment without the information that RAS testing can provide. The clinical utility of the OncoBEAM RAS test allows patients to benefit from international guideline-recommended expanded RAS testing with rapid turnaround times. OncoBEAM RAS testing has also potential utility in monitoring patients who have relapsed on treatment with anti-EGFR therapy. Identification by blood-based OncoBEAM testing of any RAS mutation in relapsed patients may be indicative of emergent resistance to anti-EGFR therapy, providing insight into the timing of subsequent lines of therapy. The high degree of concordance of RAS testing results generated by blood-based OncoBEAM RAS testing versus standard tissue testing methods supports the conclusion that detection of RAS mutations in the blood with BEAMing may be a useful replacement to tumor testing. OncoBEAM RAS testing also makes possible to examine a minimally invasive method for detecting early resistance to anti-EGFR therapy. All these features represent a clear benefit to patient care. Incorporation of the OncoBEAM RAS into clinical practice is therefore likely to add precision and provide cost-effective management by individualizing treatment plans for the CRC patient. Acknowledgements The writing assistance in the preparation of this manuscript was provided by Anabel Herrero, and editorial assistance was provided by Springer Healthcare. Acknowledgements to the Cellex Foundation for providing research facilities and equipment (Ana Vivancos).

OncoBEAM RAS testing has also potential utility in monitoring patients who have relapsed on treatment with anti-EGFR therapy. Identification by blood-based OncoBEAM testing of any RAS mutation in relapsed patients may be indicative of emergent resistance to anti-EGFR therapy, providing insight into the timing of subsequent lines of therapy. The high degree of concordance of RAS testing results generated by blood-based OncoBEAM RAS testing versus standard tissue testing methods supports the conclusion that detection of RAS mutations in the blood with BEAMing may be a useful replacement to tumor testing. OncoBEAM RAS testing also makes possible to examine a minimally invasive method for detecting early resistance to anti-EGFR therapy. All these features represent a clear benefit to patient care. Incorporation of the OncoBEAM RAS into clinical practice is therefore likely to add precision and provide cost-effective management by individualizing treatment plans for the CRC patient. Acknowledgements The writing assistance in the preparation of this manuscript was provided by Anabel Herrero, and editorial assistance was provided by Springer Healthcare. Acknowledgements to the Cellex Foundation for providing research facilities and equipment (Ana Vivancos). Funding Unrestricted grant from Sysmex-Inostics (no grant number applied).

Acknowledgements The writing assistance in the preparation of this manuscript was provided by Anabel Herrero, and editorial assistance was provided by Springer Healthcare. Acknowledgements to the Cellex Foundation for providing research facilities and equipment (Ana Vivancos). Funding Unrestricted grant from Sysmex-Inostics (no grant number applied). Disclosure JGF: Advisory role for Amgen, Astellas, AstraZeneca, Boehringer Ingelheim, BMS, Bayer, Celgene, Gilead, GSK, Janssen, Lilly, Merck Serono, MSD, Novartis, Pharmamar, Pfizer, Roche, Sanofi, and Sysmex-Inostics; EA: Advisory role for Roche, Pfizer, Celgene; EA: Consultant or advisory role for Roche, Amgen, Merck Serono, Bayer, Servier, and Sanofi; EDR: Consultant or advisory role for Merck, Amgen, Roche, Bayer, Servier, MSD, and Sanofi; RLL: No conflict of interest to declare; JT: Advisory role for Amgen, Bayer, Boehringer Ingelheim, Celgene, Chugai, Lilly, MSD, Merck Serono, Novartis, Pfizer, Roche, Sanofi, Symphogen, Taiho, and Takeda; AV: Consultant or advisory role for Sysmex-Inostics, Merck Serono, AstraZeneca, and Novartis.

Introduction Monoclonal antibodies (moAb) directed against EGFR—cetuximab and panitumumab—are standard components of treatment regimens for metastatic colorectal cancer (mCRC) patients, either alone or in combination with chemotherapy. The current standard of care is to determine mutations in RAS in all mCRC tumors before initiating treatment, as critical biomarkers of innate resistance to anti-EGFR [1]. Moreover, all mCRC patients that initially respond to anti-EGFR therapy eventually develop resistance, which in ∼50% of cases is due to the emergence of RAS mutations [2–5]. Currently, RAS mutation determination is carried out in formalin fixed paraffin-embedded samples from tumor tissue. Circulating DNA fragments carrying tumor specific sequence alterations (circulating tumor DNA, ctDNA) are found in the cell-free fraction of blood, representing a variable and generally small fraction of the total circulating cell-free DNA (cfDNA). Tumor genotyping using ctDNA offers potential advantages particularly in the metastatic setting as a safe minimally invasive alternative to tissue [3].

ing tumor DNA, ctDNA) are found in the cell-free fraction of blood, representing a variable and generally small fraction of the total circulating cell-free DNA (cfDNA). Tumor genotyping using ctDNA offers potential advantages particularly in the metastatic setting as a safe minimally invasive alternative to tissue [3]. Prior studies have demonstrated a high degree of concordance between somatic mutations detected in tumor tissue and those determined in ctDNA of patients with advanced tumors [6, 7]. The use of ctDNA has also demonstrated utility to predict treatment response to chemotherapy. Previous ctDNA studies utilized massively parallel (direct) sequencing of tumor tissue in order to identify somatic alterations specific to individual patients, which were subsequently incorporated into the development of a personalized gene panel to detect these mutations in blood samples. Although useful in a research setting, a personalized NGS panel approach is currently not amenable to routine clinical practice in that it requires significant dedicated resources in highly qualified research laboratories. Alternatively, blood-based tests that encompass a panel of the most frequently occurring mutations for a given tumor type and which can be used to interrogate the plasma of patients with high sensitivity present a practical approach for routine clinical care. The first and only test thus far for the determination of RAS mutations in ctDNA with European Conformity (CE-marked) in vitro diagnostic (CE-IVD) is the OncoBEAM RAS CRC assay, which detects 34 mutations in exons 2, 3, and 4 in the KRAS and NRAS genes as recommended by current clinical practice treatment guidelines (NCCN, ESMO, EMA).

ly test thus far for the determination of RAS mutations in ctDNA with European Conformity (CE-marked) in vitro diagnostic (CE-IVD) is the OncoBEAM RAS CRC assay, which detects 34 mutations in exons 2, 3, and 4 in the KRAS and NRAS genes as recommended by current clinical practice treatment guidelines (NCCN, ESMO, EMA). The aim of the present study was to evaluate the clinical applications of the OncoBEAM RAS CRC assay in routine clinical practice for the diagnosis, assessment of response to chemotherapy/antiangiogenic treatment and monitoring of acquired resistance to anti-EGFR therapy in mCRC patients. Materials and methods Study design and sample collection A retrospective-prospective study was carried out in two Spanish Institutions. Patients with histologically confirmed metastatic colorectal cancer and anti-EGFR treatment naïve were eligible for the study. Blood samples were collected in all patients before the administration of anti-EGFR treatment. For those patients undergoing monitoring, serial blood samples were collected every 4 weeks coinciding with the treatment visit and at the moment of progressive disease. See full inclusion criteria and regulatory aspects in supplementary material, available at Annals of Oncology online. OncoBEAM™ RAS CRC assay was used to detect RAS mutations in plasma, and RAS mutation detection in tissue samples were carried out according to standard-of-care (SoC) procedures validated by each hospital (details in supplementary material and Table S4, available at Annals of Oncology online).

Materials and methods Study design and sample collection A retrospective-prospective study was carried out in two Spanish Institutions. Patients with histologically confirmed metastatic colorectal cancer and anti-EGFR treatment naïve were eligible for the study. Blood samples were collected in all patients before the administration of anti-EGFR treatment. For those patients undergoing monitoring, serial blood samples were collected every 4 weeks coinciding with the treatment visit and at the moment of progressive disease. See full inclusion criteria and regulatory aspects in supplementary material, available at Annals of Oncology online. OncoBEAM™ RAS CRC assay was used to detect RAS mutations in plasma, and RAS mutation detection in tissue samples were carried out according to standard-of-care (SoC) procedures validated by each hospital (details in supplementary material and Table S4, available at Annals of Oncology online). Statistical analysis Variables were described using median and interquartile range (IQR) when continuous, and percentage when categorical. For mutant allele fraction (MAF) levels comparisons between different groups regarding clinical variables, we carried out Mann–Whitney U test for dichotomic variables and Kruskal–Wallis test for polycothomic variables. Tests were carried out under SPSS v.22 with a significance level of P < 0.05. Graphics were built using R 3.3.1.

allele fraction (MAF) levels comparisons between different groups regarding clinical variables, we carried out Mann–Whitney U test for dichotomic variables and Kruskal–Wallis test for polycothomic variables. Tests were carried out under SPSS v.22 with a significance level of P < 0.05. Graphics were built using R 3.3.1. Results Patient characteristics and concordance of extended RAS determination in plasma versus tissue From June 2009 to August 2016, 115 patients with mCRC were included, all of them had at least one baseline blood draw. Study flowchart is presented in Figure 1. Clinico-pathological characteristics of the patients are described in supplementary Table S1, available at Annals of Oncology online. At the time of basal ctDNA collection, all patients were naïve to anti-EGFR treatment and 82 patients (71%) had not received any therapy in the metastatic setting. The median time from tumor tissue specimen collection to ctDNA collection was 47.5 days (range 0–1783 days) in therapy-naïve patients. Figure 1. Study flowchart. Number of patients included in each of the analysis endpoints and reasons for exclusion are depicted.

Results Patient characteristics and concordance of extended RAS determination in plasma versus tissue From June 2009 to August 2016, 115 patients with mCRC were included, all of them had at least one baseline blood draw. Study flowchart is presented in Figure 1. Clinico-pathological characteristics of the patients are described in supplementary Table S1, available at Annals of Oncology online. At the time of basal ctDNA collection, all patients were naïve to anti-EGFR treatment and 82 patients (71%) had not received any therapy in the metastatic setting. The median time from tumor tissue specimen collection to ctDNA collection was 47.5 days (range 0–1783 days) in therapy-naïve patients. Figure 1. Study flowchart. Number of patients included in each of the analysis endpoints and reasons for exclusion are depicted. Of the 115 patients included in the study, 55 (47.8%) and 59 (51.3%) were shown to have RAS mutations in their tumor samples as detected by SoC RAS tissue testing and as detected in ctDNA by RAS OncoBEAM, respectively (supplementary Figure S1, available at Annals of Oncology online). The overall concordance of RAS results between ctDNA RAS OncoBEAM assay and standard techniques for tissue analysis was 93% (107/115 patients), kappa index 0.844 (95% CI 0.746–0.941) (Figure 2).

ing and as detected in ctDNA by RAS OncoBEAM, respectively (supplementary Figure S1, available at Annals of Oncology online). The overall concordance of RAS results between ctDNA RAS OncoBEAM assay and standard techniques for tissue analysis was 93% (107/115 patients), kappa index 0.844 (95% CI 0.746–0.941) (Figure 2). Figure 2. Comparison of RAS mutations detected in tissue versus plasma and analysis of discrepancies. Overall concordance analysis between RAS mutations detected in tumor by SoC and BEAMing plasma. Positive agreement (patients RAS mutated in plasma and tissue analysis) and negative agreement (patients wild-type in tissue and plasma). Clinico-pathological and treatment characteristics from eight discordant cases. Characterization of RAS tissue versus RAS plasma discordant cases Among 55 patients in whom a RAS mutation was detected in tissue, 53 also had a RAS mutation in plasma (positive percentage agreement, PPA of 96.4%) (Figure 2). There were only two cases with RAS mutated tissue in whom no RAS mutation was detected in ctDNA. Both had localized tumors at the time of diagnosis that were initially removed. At relapse, when ctDNA extraction was carried out, both patients had minimal tumor burden: one with only peritoneum metastasis and the other had one infra-centimetric lung metastasis and one single implant in the peritoneum.

ected in ctDNA. Both had localized tumors at the time of diagnosis that were initially removed. At relapse, when ctDNA extraction was carried out, both patients had minimal tumor burden: one with only peritoneum metastasis and the other had one infra-centimetric lung metastasis and one single implant in the peritoneum. Among 60 patients determined to be RAS wt in tissue, no RAS mutations were observed in ctDNA in 54 cases (negative percentage agreement, NPA of 90%). In the remaining 6 patients, plasma RAS BEAMing detected a RAS mutation that SoC tissue testing had not revealed. In all of these cases, the primary tumor served as the source for RAS mutational analysis and had been removed before ctDNA sampling. Notably, all six patients had, at least, liver metastasis when blood was drawn for ctDNA analysis. Interestingly, in five out of the six RAS tissue-/plasma+ discordant cases, the RAS MAF detected in ctDNA was under 1%. We compared the MAF of concordant and discordant cases. As shown in Figure 3A, there was a trend towards lower RAS plasma MAF in discordant cases compared with patients with concordant RAS in tissue and plasma (median RAS MAF 0.281% and 2.317%, respectively; P=0.193).

cases, the RAS MAF detected in ctDNA was under 1%. We compared the MAF of concordant and discordant cases. As shown in Figure 3A, there was a trend towards lower RAS plasma MAF in discordant cases compared with patients with concordant RAS in tissue and plasma (median RAS MAF 0.281% and 2.317%, respectively; P=0.193). Figure 3. Correlation of circulating RAS mutations and clinico-pathological characteristics. (A) Differences in RAS ctDNA mutant allele fraction (MAF) in patients with concordant RAS plasma and tissue determination compared with discordant cases (RAS wild-type in plasma and RAS mutated in tissue or RAS mutated in plasma and RAS wild-type in tissue). (B) Differences in RAS ctDNA MAF according to patients with liver metastasis versus patients without liver involvement. (C) Differences in RAS ctDNA MAF according to patients without peritoneum involvement, patients with only peritoneum metastasis and patients with peritoneum plus another metastatic site. (D) Differences in RAS ctDNA MAF according to treatment naïve patients versus patients with previous systemic treatment received within a month prior ctDNA blood extraction. Differences in RAS ctDNA MAF according to clinico-pathological characteristics The global median RAS plasma MAF was 1.84% (IQR: 0.284–11.290) (supplementary Table S1, available at Annals of Oncology online). No differences in MAFs were observed in relation to age, gender, initial stage at diagnosis or primary site of disease (right versus left).

F according to clinico-pathological characteristics The global median RAS plasma MAF was 1.84% (IQR: 0.284–11.290) (supplementary Table S1, available at Annals of Oncology online). No differences in MAFs were observed in relation to age, gender, initial stage at diagnosis or primary site of disease (right versus left). While no differences in RAS MAF were seen in relation to the number of metastatic sites, differences were observed depending on the site of metastasis location (Figure 3B and C). Patients with liver involvement had higher RAS ctDNA compared with those without liver metastases (4.806% versus 0.203%; P=0.001). In contrast, MAF from patients having only peritoneum metastases was lower (0.1%) than patients without peritoneum involvement (4.026%), than patients with peritoneum metastasis in addition to at least one other metastatic site (1.109%; P =0.056). MAF was lower in patients with only lung metastatic involvement (0.033%). Of note, in two patients that presented with tumors having mucinous histology MAF was below the median (0.451% and 0.161%, respectively; P < 0.05).

s with peritoneum metastasis in addition to at least one other metastatic site (1.109%; P =0.056). MAF was lower in patients with only lung metastatic involvement (0.033%). Of note, in two patients that presented with tumors having mucinous histology MAF was below the median (0.451% and 0.161%, respectively; P < 0.05). We then sought to analyze the impact of treatment on RAS mutation detection in plasma. No differences were observed in RAS plasma MAF between patients in whom the primary tumor had been removed before basal ctDNA extraction (4.026%) and patients in whom ctDNA was extracted from blood drawn at the time primary tumors had not been resected (1.558%; P=0.584). Regarding the relation between systemic treatment and plasma RAS mutation detection, 8 of 59 RAS mutant patients (13.6%) had received previous treatment with chemotherapy (comprising 5FU, oxaliplatin and/or irinotecan) ± anti-VEGF within a month prior ctDNA blood extraction. In all of these patients, lower RAS plasma MAFs were observed (0.173%; range 0.074–1.156) when compared with treatment-naïve patients (4.178%; range 0.451–12.620; P=0.007) (Figure 3D), emphasizing the relevance of blood draw timing for an accurate RAS determination.

nti-VEGF within a month prior ctDNA blood extraction. In all of these patients, lower RAS plasma MAFs were observed (0.173%; range 0.074–1.156) when compared with treatment-naïve patients (4.178%; range 0.451–12.620; P=0.007) (Figure 3D), emphasizing the relevance of blood draw timing for an accurate RAS determination. Response to anti-EGFR therapy and prognosis according to baseline RAS ctDNA determination We then determined the impact of RAS detection in predicting response to anti-EGFR based therapy. Among 54 patients having RAS wt tumor tissue, 34 were treated with anti-EGFR monoclonal antibodies (31 with cetuximab and 3 with panitumumab-based regimens; 30 plus chemotherapy and 4 in monotherapy). Twenty-three achieved a complete or partial response (68%) and 7 patients (20%) had stable disease for more than 16 weeks. Among tissue RAS wt patients treated with anti-EGFR therapy, four were found to have a plasma RAS positive result (Figure 2, discordant patients #1, #2, #4, #5). Three of them achieved a partial response, whereas one showed progressive disease after administration of anti-EGFR treatment. Despite the retrospective nature of our study, we determined the progression-free survival (PFS) according to RAS mutation determination in plasma and RAS mutation determination in tissue. PFS was 10.3 month (95% CI 7.7–25) for wt RAS tissue patients and 10.3 months (95% CI 7.7–19.8) for RAS wt plasma patients.

Response to anti-EGFR therapy and prognosis according to baseline RAS ctDNA determination We then determined the impact of RAS detection in predicting response to anti-EGFR based therapy. Among 54 patients having RAS wt tumor tissue, 34 were treated with anti-EGFR monoclonal antibodies (31 with cetuximab and 3 with panitumumab-based regimens; 30 plus chemotherapy and 4 in monotherapy). Twenty-three achieved a complete or partial response (68%) and 7 patients (20%) had stable disease for more than 16 weeks. Among tissue RAS wt patients treated with anti-EGFR therapy, four were found to have a plasma RAS positive result (Figure 2, discordant patients #1, #2, #4, #5). Three of them achieved a partial response, whereas one showed progressive disease after administration of anti-EGFR treatment. Despite the retrospective nature of our study, we determined the progression-free survival (PFS) according to RAS mutation determination in plasma and RAS mutation determination in tissue. PFS was 10.3 month (95% CI 7.7–25) for wt RAS tissue patients and 10.3 months (95% CI 7.7–19.8) for RAS wt plasma patients. In addition, baseline high RAS ctDNA MAF have been associated with low survival [8]. We analyzed the prognosis impact of basal MAF levels in a cohort of 22 patients with at least 3 year of follow-up. Patients with MAF levels ≥1% had significant lower PFS and OS than those with basal levels <1% (supplementary Figure S2, available at Annals of Oncology online). These data suggest that ctDNA levels could also provide valuable information to predict the disease evolution in RAS mutant patients before treatment onset.

. Patients with MAF levels ≥1% had significant lower PFS and OS than those with basal levels <1% (supplementary Figure S2, available at Annals of Oncology online). These data suggest that ctDNA levels could also provide valuable information to predict the disease evolution in RAS mutant patients before treatment onset. Longitudinal ctDNA RAS testing for assessing response to patients treated with systemic treatment Because ctDNA analysis has the capacity to reflect tumor load [2], we examined the utility of OncoBEAM RAS CRC ctDNA testing to monitor the efficacy of response of patients to treatment.

. Patients with MAF levels ≥1% had significant lower PFS and OS than those with basal levels <1% (supplementary Figure S2, available at Annals of Oncology online). These data suggest that ctDNA levels could also provide valuable information to predict the disease evolution in RAS mutant patients before treatment onset. Longitudinal ctDNA RAS testing for assessing response to patients treated with systemic treatment Because ctDNA analysis has the capacity to reflect tumor load [2], we examined the utility of OncoBEAM RAS CRC ctDNA testing to monitor the efficacy of response of patients to treatment. RAS was longitudinally monitored in serial blood draws from 21 patients with baseline RAS mutations undergoing systemic therapy. Seven patients were treated with combination chemotherapy + antiangiogenic therapy, 12 received chemotherapy alone and 2 anti-EGFR-based treatment (supplementary Table S2, available at Annals of Oncology online). A first CT-scan was carried out concurrent with the ctDNA RAS monitoring to evaluate tumor response after 8–12 weeks of treatment. Analysis of RAS ctDNA at the time of this first CT-scan revealed a dramatic decrease in plasma RAS MAF in responding patients with a median of 100%. For patients with clinical response to treatment, no differences in MAFs were observed in relation to the type of response achieved (median MAF reduction in patients with SD 99% versus 100% in patients with PR; P = 0.21). However, MAF percentage of change was significantly lower in patients that progressed at first evaluation of response compared with patients with clinical benefit (PR + SD) (132% increase versus 99% reduction, respectively, P = 0.027)

ian MAF reduction in patients with SD 99% versus 100% in patients with PR; P = 0.21). However, MAF percentage of change was significantly lower in patients that progressed at first evaluation of response compared with patients with clinical benefit (PR + SD) (132% increase versus 99% reduction, respectively, P = 0.027) In 10 out of 11 responding patients that subsequently progressed, RAS ctDNA MAF increased accordingly, though in most cases patients exhibited lower MAFs than at the time of diagnosis (Figure 4A; supplementary Table S2, available at Annals of Oncology online). Of note, in one patient having a basal KRAS mutation in codon 12 (basal RAS MAF 9.12%), a decrease in RAS MAF was initially observed that was quickly followed by an increase in RAS MAF at week 12 although the CT scan showed stable disease. This patient subsequently showed a rapid progression of disease and died 4 months later.

in one patient having a basal KRAS mutation in codon 12 (basal RAS MAF 9.12%), a decrease in RAS MAF was initially observed that was quickly followed by an increase in RAS MAF at week 12 although the CT scan showed stable disease. This patient subsequently showed a rapid progression of disease and died 4 months later. Figure 4 Longitudinal analysis of plasma RAS ctDNA to evaluate response to treatment with chemotherapy ± antiangiogenic. (A) RAS ctDNA dynamics in nine patients with RAS mutated tumors treated with chemotherapy ± antiangiogenic that initially respond to treatment. Frequency of circulating RAS mutant alleles at baseline, at time of first CT-scan to evaluate treatment response and at disease progression. Decline and increase in circulating RAS MAF correlate with response and progression to treatment, respectively. (B) RAS ctDNA dynamics in 5 patients with RAS mutated tumors that progressed at first CTscan at 8–12 weeks from beginning of treatment. (C) Patient diagnosed with stage IV rectal cancer with liver metastasis. An NRAS codon 61 mutation was detected in tissue and plasma. After 4 weeks of treatment with FOLFOX + bevacizumab, plasma MAF dramatically decreased correlating with a stable disease observed in the CT-scan at week 12. The patient underwent surgery of the primary tumor and liver metastasis, and plasma RAS became undetectable. Eight months later, the patient relapsed and RAS MAF increased accordingly. Three months after initiating second line treatment with FOLFIRI + aflibercept, the patient achieved a stable disease by CT scan, and no plasma ctDNA RAS mutations were detected. (D) Monitoring ctDNA KRAS codon 146 mutation during treatment with FOLFOX-Bevacizumab in patient #3, diagnosed with a stage IV colon cancer with lung and liver metastasis. The colonoscopic biopsy analysis was RAS wt but plasma ctDNA showed a KRAS codon 146 mutation. Following removal of the primary tumor, re-analysis of RAS in the surgical sample confirmed the plasma result. The patient received FOLFOX + bevacizumab with an early decrease in RAS ctDNA that became undetectable at 12 weeks, alongside at the first CT scan. Treatment was discontinued and a subsequently increase in KRAS codon 146 MAF was observed, which then rapidly decreased when the chemotherapy was reintroduced. Gray area indicates tumor load. Blue line indicates changes in ctDNA KRAS146 frequency.

NA that became undetectable at 12 weeks, alongside at the first CT scan. Treatment was discontinued and a subsequently increase in KRAS codon 146 MAF was observed, which then rapidly decreased when the chemotherapy was reintroduced. Gray area indicates tumor load. Blue line indicates changes in ctDNA KRAS146 frequency. Representative time courses of ctDNA along with clinical and radiologic data on two subjects are provided (Figure 4C and D), showing the high accuracy of RAS plasma ctDNA dynamics as a surrogate marker of tumor load and a potential tool to evaluate early response to treatment.

NA that became undetectable at 12 weeks, alongside at the first CT scan. Treatment was discontinued and a subsequently increase in KRAS codon 146 MAF was observed, which then rapidly decreased when the chemotherapy was reintroduced. Gray area indicates tumor load. Blue line indicates changes in ctDNA KRAS146 frequency. Representative time courses of ctDNA along with clinical and radiologic data on two subjects are provided (Figure 4C and D), showing the high accuracy of RAS plasma ctDNA dynamics as a surrogate marker of tumor load and a potential tool to evaluate early response to treatment. ctDNA extended RAS for monitoring RAS mutations during and after withdrawal of anti-EGFR therapy We and others previously reported that acquired resistance to anti-EGFR treatment is linked to the emergence of RAS mutations, that can be tracked in the blood of patients [2, 3, 8, 9]. In our study, we examined the value of OncoBEAM RAS CRC testing to detect the emergence of RAS mutations during anti-EGFR treatment. Plasma was available at the time of disease progression from 18 cases with acquired resistance to anti-EGFR therapy (i.e. disease progression after an initial complete response, partial response or stable disease for more than 16 weeks). Emergence of RAS mutations was detected in 7/18 patients (39%), 5 of them treated with chemotherapy + anti-EGFR and two with anti-EGFR monotherapy (supplementary Table S3, available at Annals of Oncology online). The most frequent mutations involved KRAS codon 12, KRAS codon 13 and NRAS codon 61. In three cases, different RAS mutations were concomitantly detected. Median RAS MAF detected at the time of anti-EGFR progression was 2.17% (range 0.024–24.957).

therapy (supplementary Table S3, available at Annals of Oncology online). The most frequent mutations involved KRAS codon 12, KRAS codon 13 and NRAS codon 61. In three cases, different RAS mutations were concomitantly detected. Median RAS MAF detected at the time of anti-EGFR progression was 2.17% (range 0.024–24.957). Discussion Our study demonstrates that OncoBEAM RAS CRC assay is an efficient and accurate tool to be used in routine clinical practice with several applications in mCRC patients, including determination of baseline RAS at diagnosis to decide anti-EGFR therapy, assessment of efficacy to treatment and monitoring of the emergence of RAS mutations as a mechanism of resistance to anti-EGFR therapy.

and accurate tool to be used in routine clinical practice with several applications in mCRC patients, including determination of baseline RAS at diagnosis to decide anti-EGFR therapy, assessment of efficacy to treatment and monitoring of the emergence of RAS mutations as a mechanism of resistance to anti-EGFR therapy. The high overall agreement between baseline plasma and tissue RAS mutation status demonstrated in more than 100 patients evaluated in our study supports the use of blood-based testing with OncoBEAM™ RAS CRC as a viable alternative to tissue SoC for determining RAS mutation status in mCRC patients treated in routine clinical practice. Previous studies have shown that ctDNA can be detected in patients with mCRC by using personalized research panels with dPCR [7, 10, 11]. Recent publications have also shown a very high sensitivity with BEAMing to detect ctDNA mutations [12, 13]. However, to the best of our knowledge, this is the first study that explores the clinical use of plasma RAS determination by using a CE-marked assay in a daily clinical routine setting and in a large real world cohort of patients. Moreover, the minimal level of discordance (6%) between RAS tissue and plasma detection shown in our study is acceptable from a clinical point of view. In fact, it is far lower than the 5%–20% discrepancy found in RAS mutation detection when comparing two different tissue RAS testing SoC techniques [14, 15].

patients. Moreover, the minimal level of discordance (6%) between RAS tissue and plasma detection shown in our study is acceptable from a clinical point of view. In fact, it is far lower than the 5%–20% discrepancy found in RAS mutation detection when comparing two different tissue RAS testing SoC techniques [14, 15]. In an effort to explain plasma/tissue discrepancies as well as to better understand the biology of circulating tumor DNA, we identified several clinico-pathological features linked to low RAS ctDNA detection, including peritoneal/lung metastases or mucinous histology. In contrast, no correlation was found between the number of metastasis and RAS ctDNA mutations in our study. This data suggests that intrinsic biological characteristics of the tumor rather than tumor burden may impact ctDNA release. In order to appreciate the utility and further optimize the routine evaluation of RAS ctDNA determination in daily clinical practice, we also studied external factors that may influence the result of RAS ctDNA determination. We found that the administration of recent systemic treatment had a clear negative impact on the ability to detect RAS mutations in the blood of patients, emphasizing the importance of collecting plasma for basal RAS analysis before the initiation of any systemic treatment. On the contrary removal of the primary tumor before blood draw for RAS analysis did not impact the RAS mutation results. In global, our study shows that the pattern of genetic alterations in cancer patients is dynamic and is affected by intrinsic and extrinsic factors.

s before the initiation of any systemic treatment. On the contrary removal of the primary tumor before blood draw for RAS analysis did not impact the RAS mutation results. In global, our study shows that the pattern of genetic alterations in cancer patients is dynamic and is affected by intrinsic and extrinsic factors. Importantly, we also found a potential role of OncoBEAM RAS ctDNA assay in monitoring response and resistance during treatment. In patients with RAS mutant tumors, RAS plasma mirrored clinical and radiological response to treatment with chemotherapy drugs and was an early predictor of response. Likewise, Tie et al. [10] reported changes in ctDNA for mCRC patients treated with chemotherapy, although a more complex research-based approach was used. Moreover, we showed a potential use of RAS ctDNA in evaluating response to antiangiogenic drugs, which could be complementary to RECIST. On the other hand, in patients with RAS wt tumors treated with anti-EGFR, OncoBEAM RAS CRC was a valid tool to detect RAS mutations of resistance. Despite the great value of the results presented, there are several limitations to our study. It is a retrospective analysis. Longitudinal blood extractions were only carried out in a limited number of patients. Additionally, given the low number of patients with specific clinico-pathological characteristics, our inferences from associations with P-values marginally <0.05% should be cautiously interpreted.

udy. It is a retrospective analysis. Longitudinal blood extractions were only carried out in a limited number of patients. Additionally, given the low number of patients with specific clinico-pathological characteristics, our inferences from associations with P-values marginally <0.05% should be cautiously interpreted. Overall, our data show that the OncoBEAM RAS CRC assay offers a minimally invasive and highly sensitive method for RAS assessment in plasma of mCRC patients which can be readily implemented into routine clinical practice to perform baseline diagnosis to select candidate patients to anti-EGR therapy. Moreover, we show a potential use of OncoBEAM RAS in assessing the dynamics of RAS to monitor response and resistance to treatment practice. Supplementary Material mdx125_supp Click here for additional data file. Acknowledgements The authors wish to thank biobank of I.D.I.S.-C.H.U.S. (PT13/0010/0068) for providing part of the tissue samples from colorectal cancer patients including at Complexo Hospitalario Universitario de Santiago de Compostela. The authors want to thank Mario Martin for his valuable contribution in the statistical analysis and Cristina Blanco for data management support. Funding This work was supported by Instituto de Salud Carlos III grants PI15/00457, INT 16/00106 and DTS15/00048 (CM), PIE15/00008 and PI15/00146 (JA); 2014SGR740 (JA); Xarxa de Banc de Tumors de Catalunya and Fundació Cellex. This work was supported in part by Merck KGaA, Darmstadt, Germany.

Acknowledgements The authors wish to thank biobank of I.D.I.S.-C.H.U.S. (PT13/0010/0068) for providing part of the tissue samples from colorectal cancer patients including at Complexo Hospitalario Universitario de Santiago de Compostela. The authors want to thank Mario Martin for his valuable contribution in the statistical analysis and Cristina Blanco for data management support. Funding This work was supported by Instituto de Salud Carlos III grants PI15/00457, INT 16/00106 and DTS15/00048 (CM), PIE15/00008 and PI15/00146 (JA); 2014SGR740 (JA); Xarxa de Banc de Tumors de Catalunya and Fundació Cellex. This work was supported in part by Merck KGaA, Darmstadt, Germany. Disclosure FJ, DE and VS are employees of Sysmex-Inostics Inc. All remaining authors have declared no conflicts of interest.

Key Message In the PORTEC-3 trial, central pathology review before randomisation by expert gynaeco-pathologists changed histological items in 43% of HR-EC patients. This led to ineligibility for the PORTEC-3 trial in 8% of patients. Upfront pathology review is recommended for future trials as well as in daily practice to ensure enrolment of the target trial-population and to avoid over- or undertreatment. Introduction Adjuvant treatment of women with endometrial cancer (EC) is based on clinicopathological risk factors, such as histological grade, myometrial invasion, age and lymph-vascular space invasion (LVSI) [1–3]. A minority of patients (15%) have high-risk disease features, which include endometrioid endometrial carcinoma (EEC) of FIGO stage I grade 3 with deep invasion or with substantial LVSI; stage II or III EEC; or non-endometrioid histologies (NEEC) stage I–III [1–4]. For these patients higher risks of distant metastases and EC-related death have been reported, and adjuvant chemotherapy may be considered [5–8]. As these high-risk criteria comprise different features of the pathology diagnosis, reproducibility is essential. Studies of pathology review by expert subspecialty pathologists, however, have shown that evaluation of female reproductive tract pathology had the highest rates of discrepancies between original and review pathology assessment including discrepancies with consequences for treatment [9]. Challenges for pre-treatment pathology review are that review is time-consuming and expensive, that timelines are tight and logistical procedures are complicated.

athology had the highest rates of discrepancies between original and review pathology assessment including discrepancies with consequences for treatment [9]. Challenges for pre-treatment pathology review are that review is time-consuming and expensive, that timelines are tight and logistical procedures are complicated. The PORTEC-3 trial is an international randomised phase III trial of adjuvant therapy in high-risk EC (HR-EC). Women with HR-EC were randomly allocated (1 : 1) to pelvic radiotherapy (RT) alone or RT plus concurrent and adjuvant chemotherapy. Primary end points are overall survival and failure-free survival. To select patients with true HR-EC and avoid unnecessary intensive treatment in lower-risk cases, upfront pathology review was carried out by expert gynaeco-pathologists of the participating groups to confirm HR-EC and eligibility for the study. The current analysis was done to establish the value of upfront pathology review. The aims were to explore the proportion of patients who were ineligible for the PORTEC-3 trial after pathology review, and to evaluate inter-observer variability between original and review pathology assessments.

The PORTEC-3 trial is an international randomised phase III trial of adjuvant therapy in high-risk EC (HR-EC). Women with HR-EC were randomly allocated (1 : 1) to pelvic radiotherapy (RT) alone or RT plus concurrent and adjuvant chemotherapy. Primary end points are overall survival and failure-free survival. To select patients with true HR-EC and avoid unnecessary intensive treatment in lower-risk cases, upfront pathology review was carried out by expert gynaeco-pathologists of the participating groups to confirm HR-EC and eligibility for the study. The current analysis was done to establish the value of upfront pathology review. The aims were to explore the proportion of patients who were ineligible for the PORTEC-3 trial after pathology review, and to evaluate inter-observer variability between original and review pathology assessments. Methods Study design and participants PORTEC-3 is a randomised Intergroup trial led by the Dutch Gynaecological Oncology Group, with participating groups MRC-NCRI (UK), ANZGOG (Australia and New Zealand), MaNGO (Italy), Fedegyn (France) and CCTG (Canada). Surgery comprised hysterectomy with salpingo-oophorectomy. Lymphadenectomy was at the discretion of the participating centres. For serous or clear cell cancers, surgical staging including omentectomy; peritoneal biopsies and lymphadenectomy was recommended. Details on patient selection and treatment have been described in a previous publication [10]. Eligible patients had EEC of FIGO 2009 stage 1A grade 3 with LVSI; IB grade 3; stage II, IIIA, IIIBparametrial or IIIC; or NEEC stage IA–III.

Methods Study design and participants PORTEC-3 is a randomised Intergroup trial led by the Dutch Gynaecological Oncology Group, with participating groups MRC-NCRI (UK), ANZGOG (Australia and New Zealand), MaNGO (Italy), Fedegyn (France) and CCTG (Canada). Surgery comprised hysterectomy with salpingo-oophorectomy. Lymphadenectomy was at the discretion of the participating centres. For serous or clear cell cancers, surgical staging including omentectomy; peritoneal biopsies and lymphadenectomy was recommended. Details on patient selection and treatment have been described in a previous publication [10]. Eligible patients had EEC of FIGO 2009 stage 1A grade 3 with LVSI; IB grade 3; stage II, IIIA, IIIBparametrial or IIIC; or NEEC stage IA–III. Patients were randomised (1 : 1) to RT (48.6 Gy) or RT plus adjuvant chemotherapy (two cycles of cisplatin 50 mg/m2 in weeks 1 and 4 of RT, followed by four cycles of carboplatin AUC5 and paclitaxel 175 mg/m2 every 3 weeks). Written informed consent (IC) was obtained from all patients. The protocol was approved by the Dutch Cancer Society and the Ethics committees. Participating groups obtained their own IRB and ethics approvals and were funded by separate grants.

Patients were randomised (1 : 1) to RT (48.6 Gy) or RT plus adjuvant chemotherapy (two cycles of cisplatin 50 mg/m2 in weeks 1 and 4 of RT, followed by four cycles of carboplatin AUC5 and paclitaxel 175 mg/m2 every 3 weeks). Written informed consent (IC) was obtained from all patients. The protocol was approved by the Dutch Cancer Society and the Ethics committees. Participating groups obtained their own IRB and ethics approvals and were funded by separate grants. Procedures Each participating group had appointed expert gynaeco-pathologists as reviewers for the study. After surgery, the pathology diagnosis was made by the regional pathologist. In case of HR-EC, all histopathology slides and a copy of the pathology report were sent for pathology review as part of patient management, to confirm HR-EC within 1 week, with the aim to ensure that only true HR-EC cases were informed and enrolled in the trial. If IC was given, pathology review for the PORTEC-3 trial was completed with trial-specific items. Upon consent for storage of tumour tissue for translational research a formalin-fixed paraffin-embedded (FFPE)-block was centrally stored. All other blocks and slides were sent back to the referring centre.

d in the trial. If IC was given, pathology review for the PORTEC-3 trial was completed with trial-specific items. Upon consent for storage of tumour tissue for translational research a formalin-fixed paraffin-embedded (FFPE)-block was centrally stored. All other blocks and slides were sent back to the referring centre. The items for original and review pathology included WHO histological type, grade, depth of myometrial invasion, distance to serosa or serosal breach, LVSI, cervical stromal involvement, involvement of the tubes and/or ovaries and lymph node involvement. Histological type was evaluated as endometrioid, serous, clear cell, mixed (endometrioid with serous/clear cell components), mucinous, or other histologies according to WHO-classification [11]. Mixed tumours were classified as serous or clear cell when this component was at least 25%, otherwise as mixed. Mucinous tumours were grouped with EEC for analysis. Histological grading was done according to WHO [11]. NEEC was considered high grade per definition (grade 3). The differences in histological grading between original and review pathology were evaluated for EEC. Immunohistochemistry (IHC) was carried out only incidentally, at the discretion of the review pathologist and only if FFPE-blocks were available at time of the central review process.

gh grade per definition (grade 3). The differences in histological grading between original and review pathology were evaluated for EEC. Immunohistochemistry (IHC) was carried out only incidentally, at the discretion of the review pathologist and only if FFPE-blocks were available at time of the central review process. For the current analysis, anonymised original and review pathology reports from both randomised and non-randomised patients in the Netherlands (NL) and the UK (UK) were assessed. These two countries were chosen as they had the largest number of patients in the trial (together 48%) and all pathology reviews had been done at two centres in each country. For the UK patients, the review pathologist provided a short confirmation of HR-EC and eligibility. For the randomised patients, the review report was completed after IC was given. Outcomes Discrepancies between original and central pathology review were assessed as discrepancies with and without change of eligibility for the PORTEC-3 trial. Reasons for non-eligibility were checked by two expert gynaeco-pathologists (TB and NS). Statistical analysis The data were collected in a SPSS database (version 23.0). For the comparison of the pathology items, Cohen’s kappa value (κ) was used [12]. For the interpretation of the κ values the scale proposed by Landis and Koch was used [13]. Differences between eligible women who were included or declined the study were analysed by the χ2 test. Items with P-values <0.05 were considered significant.

Statistical analysis The data were collected in a SPSS database (version 23.0). For the comparison of the pathology items, Cohen’s kappa value (κ) was used [12]. For the interpretation of the κ values the scale proposed by Landis and Koch was used [13]. Differences between eligible women who were included or declined the study were analysed by the χ2 test. Items with P-values <0.05 were considered significant. Results Population and compliance The PORTEC-3 trial included 686 patients (2006–2013), of whom 145 were recruited in NL and 184 in the UK. Slides from 1295 patients (395 NL, 900 UK) were sent for pathology review. Fifteen original pathology reports (9 NL, 6 UK) were not available for analysis. Fifty-four patients (18 NL, 36 UK) were ineligible based on the original pathology report, which was confirmed by pathology review and they were therefore excluded from the analysis. A total of 1226 patients (368 NL, 858 UK) were eligible based on local pathology and were included in this analysis (see Figure 1, Table 1 and supplementary Table S1, available at Annals of Oncology online). Table 1. Major pathology criteria of the eligible patients (n = 1226)

were therefore excluded from the analysis. A total of 1226 patients (368 NL, 858 UK) were eligible based on local pathology and were included in this analysis (see Figure 1, Table 1 and supplementary Table S1, available at Annals of Oncology online). Table 1. Major pathology criteria of the eligible patients (n = 1226) Major pathologic criteria NL patients (n = 368) UK patients (n = 858) n % n % Age <60 100 37% 239 28% 60–69 110 41% 373 44% ≥70 58 22% 243 28% Missing 100 3 FIGO stage (2009) IA 72 20% 138 16% IB 93 26% 178 21% II 99 27% 263 31% IIIA 43 12% 97 12% IIIB 18 5% 62 7% IIIC 40 11% 101 12% Missing 3 19 Histological type Endometrioid or mucinous 262 71% 501 59% Serous or mixed serous 66 18% 193 23% Clear cell or mixed clear cell 31 8% 111 13% Othera 9 2% 45 5% Missing 0 8 Histological grade 1 81 22% 155 18% 2 53 14% 135 16% 3 127 35% 201 24% NEEC 107 29% 354 42% Missing 0 13 Myometrial invasion <50% 135 37% 215 38% ≥50% 233 63% 346 62% Missing 0 297 Growth through serosa Yes 21 6% 31 4% No 346 94% 675 96% Missing 1 152 Cervical glandular involvement Yes 135 38% 172 43% No 224 62% 230 57% Missing 9 456 Cervical stromal involvement Yes 138 38% 339 47% No 225 62% 382 53% Missing 5 137 LVSI Yes 198 54% 287 60% No 169 46% 194 40% Missing 1 377 Involvement of the ovaries Yes 46 13% 67 9% No 322 87% 666 91% Missing 0 125 Lymph node involvement Not applicable 252 69% 553 66% No malignancy 73 20% 184 22% Metastasis 41 11% 101 12% Missing 2 20 Parametrial involvement Yes 24 13% 61 16% No 167 87% 326 84% Missing 177 471 Missing values were not taken into account to the percentages.

ovaries Yes 46 13% 67 9% No 322 87% 666 91% Missing 0 125 Lymph node involvement Not applicable 252 69% 553 66% No malignancy 73 20% 184 22% Metastasis 41 11% 101 12% Missing 2 20 Parametrial involvement Yes 24 13% 61 16% No 167 87% 326 84% Missing 177 471 Missing values were not taken into account to the percentages. The pathology criteria of the NL versus the UK patients were based on review pathology. a Other histology includes undifferentiated, carcinosarcoma or mixed combinations other than serous/clear cell with endometrioid. FIGO, International Federation of Gynecology and Obstetrics; LVSI, lymph-vascular space invasion; EEC, endometrioid endometrial cancer; NEEC, non-endometrioid endometrial cancer. Figure 1. CONSORT diagram. Discrepancies and inter-observer variability A total of 6356 pathology items were evaluable for both original and review pathology. For 679 items (11%) there was a discrepancy between original and review pathology. The highest agreement was found for serosal breach (98%) and cervical stromal involvement (94%), and most disagreement for histological type (15%) and grade (20%; see Table 2). Table 2. Inter-observer variability between original and review pathology report for the total cohort

cy between original and review pathology. The highest agreement was found for serosal breach (98%) and cervical stromal involvement (94%), and most disagreement for histological type (15%) and grade (20%; see Table 2). Table 2. Inter-observer variability between original and review pathology report for the total cohort Total cohort Pathology item Total number available for analysisa Missing items Total discrepancies Disagreement %b Leading to ineligibility Leading to ineligibility %c Not leading to ineligibility Not leading to ineligibility %d κ value Histological type 1217 9 185 15% 35 19% 150 81% 0.72 Histological grade (EEC only) 701 0 139 20% 19 14% 120 86% 0.70 Myometrial invasion 923 304 88 10% 7 8% 81 92% 0.79 Cervical glandular involvement 626 600 73 12% 0 0% 73 100% 0.73 Cervical stromal involvement 1063 163 69 6% 27 39% 42 61% 0.87 LVSI 762 464 101 13% 4 4% 97 96% 0.72 Growth through serosa 1064 162 24 2% 0 0% 24 100% 0.76 Total 6356 1702 679 11% 92 14% 587 86% NA a Total number of pathology items available for comparison between original and review pathology. b Total discrepancies/total number of pathology items available for analysis. c Number of pathology items leading to ineligibility/total discrepancies. d Number of pathology items not leading to ineligibility/total discrepancies. LVSI, lymph vascular space invasion; EEC, endometrioid endometrial cancer.

Total cohort Pathology item Total number available for analysisa Missing items Total discrepancies Disagreement %b Leading to ineligibility Leading to ineligibility %c Not leading to ineligibility Not leading to ineligibility %d κ value Histological type 1217 9 185 15% 35 19% 150 81% 0.72 Histological grade (EEC only) 701 0 139 20% 19 14% 120 86% 0.70 Myometrial invasion 923 304 88 10% 7 8% 81 92% 0.79 Cervical glandular involvement 626 600 73 12% 0 0% 73 100% 0.73 Cervical stromal involvement 1063 163 69 6% 27 39% 42 61% 0.87 LVSI 762 464 101 13% 4 4% 97 96% 0.72 Growth through serosa 1064 162 24 2% 0 0% 24 100% 0.76 Total 6356 1702 679 11% 92 14% 587 86% NA a Total number of pathology items available for comparison between original and review pathology. b Total discrepancies/total number of pathology items available for analysis. c Number of pathology items leading to ineligibility/total discrepancies. d Number of pathology items not leading to ineligibility/total discrepancies. LVSI, lymph vascular space invasion; EEC, endometrioid endometrial cancer. In 532 cases (43%) at least one pathology item changed after review, which led to ineligibility for the PORTEC-3 trial in 8% (n = 102; Table 3). Most frequent reasons were change of histological type (34%, n = 35), cervical stromal involvement (27%, n = 27) and change of histological grade in 19% (n = 19), which was similar between the NL and UK cohorts. Eighty-three of these 102 became low risk after central pathology review, while in 19 cases the histological type was reclassified as carcinosarcoma; these were therefore still high risk but were not eligible for the PORTEC-3 trial. Table 3. Reasons for ineligibility of 102 patients based on pathological review report

Eighty-three of these 102 became low risk after central pathology review, while in 19 cases the histological type was reclassified as carcinosarcoma; these were therefore still high risk but were not eligible for the PORTEC-3 trial. Table 3. Reasons for ineligibility of 102 patients based on pathological review report Pathology variables Cohort (n = 102) NL cohort (n = 42) UK cohort (n = 60) n % n % n % Histological type 35 34 14 33 21 35 Histologic gradea 19 19 7 17 12 20 Myometrial invasion 7 7 3 7 4 7 Cervical involvement 27 27 12 29 15 25 LVSI 4 4 2 5 2 3 Otherb 10 10 4 10 6 10 Total ineligible patients 102 100 42 100 60 100 Percentage of total cohort 102 8 42 11 60 7 a Grade shift for endometrioid endometrial carcinoma. b Other reasons included the absence of involvement of the ovaries, tube or parametrium, or other primary tumour site (cervix, tube or adnex). LVSI, lymph vascular space invasion; NL, Netherlands; UK, United Kingdom. Highest rates of inter-observer variability were found for histological type (κ = 0.72), LVSI (κ = 0.72) and histological grade (κ = 0.70; Table 2). See supplementary Table S2, available at Annals of Oncology online for results by country and supplementary Figure S1, available at Annals of Oncology online. Lowest inter-observer variability was found for cervical stromal invasion (κ = 0.87), with overall agreement of 94%. However, a discrepancy here led to ineligibility for the trial in 27/69 (39%) of cases.

nals of Oncology online for results by country and supplementary Figure S1, available at Annals of Oncology online. Lowest inter-observer variability was found for cervical stromal invasion (κ = 0.87), with overall agreement of 94%. However, a discrepancy here led to ineligibility for the trial in 27/69 (39%) of cases. Serosal breach was present in only 5% of cases. Although agreement was high for both countries (97% and 99%), κ values differed (NL κ = 0.83 versus UK κ = 0.63), showing that κ values are less reliable for items with few observations. Histological type and grade Figure 2 shows the agreement of histological classification and grade. Overall agreement of histological type was 85%; discrepancies led to ineligibility in 19% of cases (Table 2). Discrepancies were found for all histologies, although the agreement was highest for EEC. Figure 2. Histological type (A) and histological grade evaluation (B) in original and review pathology. The overall agreement for histological grade was 80%; 16% (n = 113) were downgraded at review pathology, with most frequent shifts (76 cases) from grade 2 to 1. In 4% (n = 26), the grade was higher at review.

Histological type and grade Figure 2 shows the agreement of histological classification and grade. Overall agreement of histological type was 85%; discrepancies led to ineligibility in 19% of cases (Table 2). Discrepancies were found for all histologies, although the agreement was highest for EEC. Figure 2. Histological type (A) and histological grade evaluation (B) in original and review pathology. The overall agreement for histological grade was 80%; 16% (n = 113) were downgraded at review pathology, with most frequent shifts (76 cases) from grade 2 to 1. In 4% (n = 26), the grade was higher at review. Discussion In the PORTEC-3 trial of adjuvant RT with or without chemotherapy for women with HR-EC, upfront pathology review was carried out before patient counselling to ensure that only true HR-EC patients were informed about the trial, and that the trial only enrolled true HR-EC cases. The expert gynaeco-pathology review changed the eligibility for 102 women (8%), most frequently due to changes in histological type or cervical stromal involvement. These lower-risk patients did therefore not risk receiving more intensive and potentially toxic treatment. Furthermore, a true HR-EC study population in the PORTEC-3 trial was ensured. For 19 patients the histological type changed to carcinosarcoma and although they were high risk, they were not eligible for the trial.

These lower-risk patients did therefore not risk receiving more intensive and potentially toxic treatment. Furthermore, a true HR-EC study population in the PORTEC-3 trial was ensured. For 19 patients the histological type changed to carcinosarcoma and although they were high risk, they were not eligible for the trial. The inter-observer agreement between original and review pathology was highest for cervical stromal invasion. The most frequent discrepancies were found for histological type, histological grade and presence of LVSI. While many of these discrepancies did not affect eligibility for the current study, they were important for prognosis and adjuvant treatment of patients in clinical practice. Discrepancies in gynaeco-pathology diagnosis between original and review pathology have been reported before. A Canadian study reported EC as the tumour site with most frequent differences in pathological assessment [14]. Another Canadian cohort reported major discrepancies in 8% of biopsies and hysterectomy specimens taken together, and in 12% of hysterectomy specimens. The most frequent diagnostic discrepancies were assessment of myometrial invasion and histological subtype [15].

t frequent differences in pathological assessment [14]. Another Canadian cohort reported major discrepancies in 8% of biopsies and hysterectomy specimens taken together, and in 12% of hysterectomy specimens. The most frequent diagnostic discrepancies were assessment of myometrial invasion and histological subtype [15]. In the PORTEC-1 and -2 trials pathology review showed that 24% and 14%, respectively, of patients were in retrospect ineligible, while this was 8% for the PORTEC-3 trial [1, 16, 17]. Eligibility in the PORTEC-1 and -2 studies was determined by grade, myometrial invasion and histological type. Differences in eligibility were often caused by shift of grade 2 to grade 1, while such grade shift did not affect the PORTEC-3 trial where patients had to have either grade 3 or NEEC or advanced stages. Minor discrepancies in grade or histology changed the eligibility for the PORTEC-3 trial in only a minority of patients. However, some shift of grade 2 to grade 1 was seen in the PORTEC-3 trial as well. Previous studies have shown that the intermediate grade is the least reproducible and that a two-tiered grading system assessing high versus low grade would be preferable [18–20]. The lower inter-observer variation in the current study could also reflect the increasing standardisation of pathology criteria and subspecialty training.

es have shown that the intermediate grade is the least reproducible and that a two-tiered grading system assessing high versus low grade would be preferable [18–20]. The lower inter-observer variation in the current study could also reflect the increasing standardisation of pathology criteria and subspecialty training. Frequent causes of discrepancies were assessment of histological type and cervical involvement. Several studies have addressed challenges in diagnosing serous, clear cell and mixed cancers, the level of agreement varying from 62% to 83% [21–23]. In the study by Han et al. [21], there was consensus on histological type in 72% of cases. With a panel of three IHC markers the agreement increased to 96% [21]. The use of IHC was not routine practice in the period of the PORTEC-3 trial and was only carried out in incidental cases. Variations in defining cervical stromal involvement have also been reported in a study of 76 cases reviewed by 6 expert gynaeco-pathologists with agreement in only 50%. Difficulties comprised the definition of the junction between the lower uterine segment and the endocervix, and the distinction between unattached tumour components or true cervical stromal involvement [24].

also been reported in a study of 76 cases reviewed by 6 expert gynaeco-pathologists with agreement in only 50%. Difficulties comprised the definition of the junction between the lower uterine segment and the endocervix, and the distinction between unattached tumour components or true cervical stromal involvement [24]. A limitation of this study could be that the pathology reviews took place at four university centres, and inter-observer variations between these gynaeco-pathologists were not assessed. The percentages of major discrepancies were, however, quite similar between the two countries. In the PORTEC-2 trial, higher risk of distant metastasis and lower survival were found for patients who were considered ‘high-risk’ after central review pathology, suggesting that the review pathology was more reliable to predict prognosis when compared with the original pathology [16].

milar between the two countries. In the PORTEC-2 trial, higher risk of distant metastasis and lower survival were found for patients who were considered ‘high-risk’ after central review pathology, suggesting that the review pathology was more reliable to predict prognosis when compared with the original pathology [16]. Creating a well-defined trial population with confirmed eligibility by upfront pathology review should be considered the standard for future scientific studies. Expert consultation is being increasingly used, but pathology review might not be part of the standard procedure, because it is time consuming and expensive. To this purpose, further standardisation of pathology criteria, expert education and subspecialisation in gynaeco-pathology are essential, as well as rapid access to expert consultation. The transition to digital pathology will greatly facilitate rapid consultation. Introduction of IHC and molecular analysis using the TCGA molecular subgroup classification will further improve risk assignment [25, 26]. A substantial proportion of eligible women declined participation in the trial, mostly because they did not want to receive chemotherapy. Younger patients and those with a more advanced stage of disease more often consented to participate in the trial (supplementary Table S1, available at Annals of Oncology online). The potential treatment consequences for patients should be the main reason to incorporate pathology review in daily practice. In the current study, most patients with discrepancies were downgraded and were spared unnecessary treatment.

pate in the trial (supplementary Table S1, available at Annals of Oncology online). The potential treatment consequences for patients should be the main reason to incorporate pathology review in daily practice. In the current study, most patients with discrepancies were downgraded and were spared unnecessary treatment. In conclusion, upfront pathology review by expert gynaeco-pathologists identified changes in histological type, grade or other items in 43% of patients. Of these, 8% of patients were found ineligible for the trial. This resulted in a true HR-EC population and reliable pathology assessment in the PORTEC-3 trial, which ensures the quality of future translational research. Upfront pathology review is to be preferred in future gynaecological oncology trials and in daily practice. The transition to digital pathology will strongly facilitate rapid expert pathology consultation. Supplementary Material Supplementary Figure S1 Click here for additional data file. Supplementary Table S1 Click here for additional data file. Supplementary Table S2 Click here for additional data file. Supplementary Data Click here for additional data file.

In conclusion, upfront pathology review by expert gynaeco-pathologists identified changes in histological type, grade or other items in 43% of patients. Of these, 8% of patients were found ineligible for the trial. This resulted in a true HR-EC population and reliable pathology assessment in the PORTEC-3 trial, which ensures the quality of future translational research. Upfront pathology review is to be preferred in future gynaecological oncology trials and in daily practice. The transition to digital pathology will strongly facilitate rapid expert pathology consultation. Supplementary Material Supplementary Figure S1 Click here for additional data file. Supplementary Table S1 Click here for additional data file. Supplementary Table S2 Click here for additional data file. Supplementary Data Click here for additional data file. Acknowledgments We thank all the participating groups: Dutch Gynaecology Oncology Group (the Netherlands), the National Cancer Research Institute (UK), Australian and New Zealand Gynaecologic Oncology Group (Australia and New Zealand), MaNGO (Italy), Fedegyn (France) and Canadian Cancer Trials Group (Canada); their coordinating staff, principal investigators and clinical research teams at the participating centres for all their work, and the patients who participated in the trial. We acknowledge the regional and central trial pathologists and the members of the Data and Safety Monitoring Board listed in the supplementary Appendix S1, available at Annals of Oncology online.

and clinical research teams at the participating centres for all their work, and the patients who participated in the trial. We acknowledge the regional and central trial pathologists and the members of the Data and Safety Monitoring Board listed in the supplementary Appendix S1, available at Annals of Oncology online. Funding This work was supported by a grant from the Dutch Cancer Society (UL2006-4168/CKTO 2006-04), The Netherlands. The PORTEC-3 trial was supported in the UK by Cancer Research UK (C7925/A8659). This study is registered with ISRCTN (ISRCTN14387080, www.controlled-trials.com) and with ClinicalTrials.gov (NCT00411138). The travel and stay in the UK for this project has been sponsored by the Leiden University Fund/van Steeden. Disclosure The authors have declared no conflicts of interest.

T cell checkpoint-targeted immunotherapy is effective in multiple cancers, but only in subsets of patients [1]. Failure of immunotherapy may be secondary to tumour intrinsic and/or systemic factors that impair immune response. Glucocorticoid administration has known systemic immunosuppressive effects [2] with potential to impair immunotherapy outcome [3], and should therefore be regulated at patient enrolment. We performed a cross-sectional analysis of T cell checkpoint-targeted cancer immunotherapy trials in solid malignancies registered on the U.S. National Institutes of Health (NIH) trial registry (clinicaltrials.gov) by October 7, 2016. Trials were searched by study type, condition, and interventions targeting the T cell checkpoint proteins CTLA-4, PD-1, PD-L1, PD-L2, LAG3, B7-H3, CD137, OX40, CD27 and GITR. Trials were reviewed manually and independently by two clinicians and registered data on glucocorticoid administration within enrolment criteria recorded.

re searched by study type, condition, and interventions targeting the T cell checkpoint proteins CTLA-4, PD-1, PD-L1, PD-L2, LAG3, B7-H3, CD137, OX40, CD27 and GITR. Trials were reviewed manually and independently by two clinicians and registered data on glucocorticoid administration within enrolment criteria recorded. We identified 1017 registered T cell checkpoint-targeted cancer immunotherapy trials. The number of registrations has progressively increased, exponentially between 2010 and 2015 (R2 = 0.95) (Figure 1A). For the completed years, 2001–2015, chronic glucocorticoid administration was stated as an exclusion criterion in 40% (276/685), permitted in 29% (201/685) and not specified in 30% (208/685) of trial registration details. The proportion of trials that did not allow glucocorticoid use has decreased significantly (P < 0.001), while the proportion allowing glucocorticoid use has increased significantly (P < 0.001) (Figure 1B). Of the trials permitting glucocorticoid use, the maximum permitted dose of prednisolone equivalent per day was up to 10 mg in 57% of trials (115/201), over 10 mg in 4% of trials (9/201) and not specified in 14% of trials (28/201); 24% of trials (49/201) permitted chronic glucocorticoid use for physiological replacement.

f the trials permitting glucocorticoid use, the maximum permitted dose of prednisolone equivalent per day was up to 10 mg in 57% of trials (115/201), over 10 mg in 4% of trials (9/201) and not specified in 14% of trials (28/201); 24% of trials (49/201) permitted chronic glucocorticoid use for physiological replacement. Figure 1. Longitudinal registration count and glucocorticoid administration in T cell checkpoint-targeted cancer immunotherapy trials. (A) Annual registration count of T cell checkpoint-targeted cancer immunotherapy trials. Trials registered by October 7, 2016 on the U.S. National Institutes of Health (NIH) trial registry were categorized according to year of registration and checkpoint protein target. T cell checkpoint proteins targeted in fewer than 10 trials are grouped as ‘other’, and include single agent OX40, GITR, CD137, B7-H3, LAG3, PD-L2, CD27 and trials comparing multiple checkpoint-targeting agents. ‘Combination’ trials include all pre-defined T cell checkpoint-targeted agents used in combination with another agent. (B) Specification of glucocorticoid administration within enrolment criteria of T cell checkpoint-targeted cancer immunotherapy trials. Trials registered on the U.S. NIH trial registry between 2001 and 2015 were categorized according to the specification of chronic systemic glucocorticoid administration within registered patient enrolment criteria. Univariate analysis for data presented was performed using the Cochran–Armitage test for trend assuming monotonical change over time and expected frequencies were met (80% of expected frequencies >5). ***P < 0.001; NS= not significant.

systemic glucocorticoid administration within registered patient enrolment criteria. Univariate analysis for data presented was performed using the Cochran–Armitage test for trend assuming monotonical change over time and expected frequencies were met (80% of expected frequencies >5). ***P < 0.001; NS= not significant. These findings are concerning. The immunosuppressive effects of glucocorticoids are dose dependent, starting at less than 10 mg of prednisolone per day [2], and may be compounded by hypoalbuminaemia present in patients with cancer [4]. Moreover, our pre-clinical work has demonstrated that low dose glucocorticoid administration is sufficient to suppress response to cancer immunotherapy [3]. Therefore, unregulated glucocorticoid administration may result in treatment failure independent of the T cell checkpoint-targeted agent or tumour type. The use of glucocorticoids as appetite stimulants and anti-emetics, particularly relevant in combination trials with emetogenic chemotherapy or radiotherapy, may also be immunosuppressive and will require critical review. While the use of glucocorticoids for adrenal replacement or chronic immune illness may be unavoidable, stratification for their use at enrolment should be considered. In addition to glucocorticoid administration, endogenous glucocorticoid levels may also impact on response to immunotherapy. Monitoring and stratification according to baseline glucocorticoid levels, or clinical surrogates of these such as longitudinal weight measurements [3], may yield predictive and prognostic markers of response.

glucocorticoid administration, endogenous glucocorticoid levels may also impact on response to immunotherapy. Monitoring and stratification according to baseline glucocorticoid levels, or clinical surrogates of these such as longitudinal weight measurements [3], may yield predictive and prognostic markers of response. We note that the chronic use of glucocorticoids should be considered independently from the use of glucocorticoids in managing immune-related adverse events during immunotherapy. In fact, the positive correlation of autoimmune side effects and treatment efficacy [5] provides further rationale for considering the role of systemic immunomodulatory variables in determining response to immunotherapy. Our study is limited by the exclusive use of data from the U.S. NIH trial registry. However, this is the largest clinical trial registry, and the registration of key inclusion and exclusion criteria is international standard [6] and has been mandatory for consideration of publication by the International Committee of Medical Journal Editors member journals since 2005. In summary, we find glucocorticoid administration to be a neglected immunomodulatory variable in cancer immunotherapy trials, and suggest striving for greater harmony in the monitoring and regulation of systemic glucocorticoids to improve outcomes in cancer immunotherapy. Funding This work was supported by the Wellcome Trust Translational Medicine and Therapeutics Programme [RJAG/076 to TJ], Cancer Research UK and the Cambridge Translational Medicine and Therapeutics Academic Clinical Fellowship Programme (to CMC).

In summary, we find glucocorticoid administration to be a neglected immunomodulatory variable in cancer immunotherapy trials, and suggest striving for greater harmony in the monitoring and regulation of systemic glucocorticoids to improve outcomes in cancer immunotherapy. Funding This work was supported by the Wellcome Trust Translational Medicine and Therapeutics Programme [RJAG/076 to TJ], Cancer Research UK and the Cambridge Translational Medicine and Therapeutics Academic Clinical Fellowship Programme (to CMC). Disclosure The authors have declared no conflicts of interest.

Introduction In metastatic colorectal cancer (mCRC), treatment with anti-epidermal growth factor receptor (EGFR) monoclonal antibodies cetuximab or panitumumab has demonstrated efficacy in wild-type (WT) RAS mutations and it is now considered imperative this determination at the time of diagnosis [1, 2]. Formalin-fixed, paraffin-embedded (FFPE) tumor tissue with PCR analysis is currently used as standard of care (SoC) for RAS testing and is considered the gold standard [3]. Circulating-free DNA (cfDNA) is natural DNA present in the cell-free fraction of blood. Recent studies have suggested that genomic alterations in solid tumors may be characterized by studying the circulating tumor DNA (ctDNA) released from cancer cells into the plasma [4]. In mCRC, ctDNA is detected in almost all patients but the low abundance requires highly sensitive techniques to study mutations present at low frequencies. This approach represents a liquid non-invasive biopsy with a potential for determining RAS status. The main benefits are based on the safety and convenience associated with minimally invasive procedures, accessibility at any time point—that favor dynamic/evolutive evaluation—and is not affected by sample selection bias, although accuracy and concordance with tumor-based techniques has not been fully elucidated in patients from clinical practice [5–7].

afety and convenience associated with minimally invasive procedures, accessibility at any time point—that favor dynamic/evolutive evaluation—and is not affected by sample selection bias, although accuracy and concordance with tumor-based techniques has not been fully elucidated in patients from clinical practice [5–7]. Here, we carried out a concordance biomarker analysis of 146 mCRC patients using plasma and tissue-based RAS mutation testing with BEAMing and SoC techniques in both specimens. Discordant results were analyzed in-depth taking into consideration both technical and clinical conditions. We investigated the value of this determination in terms of progression-free survival (PFS) in patients who had received anti-EGFR as well as overall survival (OS) and mutant allele fraction (MAF) analysis. Materials and methods Study design This prospective-retrospective study recruited patients candidate for therapy from three Spanish hospitals as well as from a phase II multicentric TTD ULTRA clinical trial (NCT01704703) for prospective biomarker investigation. It was approved by the ethics committees of each hospital and all patients provided written informed consent. Patients were required to have a diagnosis of mCRC with available tumor tissue for mutational analysis, have not received anti-EGFR agents before plasma collection, and have evidence of measurable disease according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 [8].

vided written informed consent. Patients were required to have a diagnosis of mCRC with available tumor tissue for mutational analysis, have not received anti-EGFR agents before plasma collection, and have evidence of measurable disease according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 [8]. Plasma was obtained from 10 ml of blood and all patients had FFPE tissue (either primary tumor or metastasis) with >15% tumor area. Tumor tissue area was evaluated by the pathologist taking into consideration the amount of sample occupied by the tumor in a standardized procedure. All samples were analyzed blinded to the study endpoints. Full description in supplementary methods, available at Annals of Oncology online. RAS mutational analysis RAS status determination was carried out with available plasma and tumor tissue using BEAMing and Real-Time PCR as SoC technique. The DNA extracted from FFPE tissue sections was partitioned and used for both determinations (BEAMing and real-time PCR). The panel of RAS mutations evaluated with BEAMing was identical to that previously validated (supplementary Table S1, available at Annals of Oncology online) [2]. Each plasma and tumor sample was independently processed (using an 8-step workflow, supplementary Figure S1, available at Annals of Oncology online). In discordant cases the historical RAS reports were reviewed and further RAS determinations were carried out when metastases tissue was available, using SoC techniques (supplementary Table S2, available at Annals of Oncology online).

an 8-step workflow, supplementary Figure S1, available at Annals of Oncology online). In discordant cases the historical RAS reports were reviewed and further RAS determinations were carried out when metastases tissue was available, using SoC techniques (supplementary Table S2, available at Annals of Oncology online). Depending on the specific assay, samples with a detectable mutation rate above 0.02%–0.04% were considered positive using BEAMing in ctDNA and 1% in tumor tissue. CtDNA testing was carried out with the commercially available CE-IVD BEAMing RAS plasma kit with the same thresholds for the specific mutations. The sensitivity for Real-Time PCR as SoC analysis in tumor tissue is ∼1%–5%. Full description in supplementary methods and Table S3, available at Annals of Oncology online. Statistics Full description in supplementary methods, available at Annals of Oncology online. Results Patient characteristics A total of 157 mCRC patients were initially included, 11 of whom were excluded because of specific pre-analytical requirements or lack of tumor tissue availability (supplementary Figure S2, available at Annals of Oncology online). Patient baseline characteristics, number and location of metastasis, and number and description of previous lines of therapy are summarized in supplementary Table S4, available at Annals of Oncology online.

Results Patient characteristics A total of 157 mCRC patients were initially included, 11 of whom were excluded because of specific pre-analytical requirements or lack of tumor tissue availability (supplementary Figure S2, available at Annals of Oncology online). Patient baseline characteristics, number and location of metastasis, and number and description of previous lines of therapy are summarized in supplementary Table S4, available at Annals of Oncology online. Overall, 61 (42%) patients were naïve for therapy in the metastatic setting at the time of ctDNA collection, while the remaining 85 (58%) patients had received a range of treatments but all were anti-EGFR therapies naive. The median time from tumor tissue specimen to ctDNA collection was 1.2 months (range 0–34) in therapy-naive patients. The range in previously exposed patients was wide, with a median of 20.2 months (range 0.4–282). A group of 67 (46%) patients received anti-EGFR immediately after ctDNA collection mainly in second and third line (supplementary Table S4, available at Annals of Oncology online). Median PFS and median OS were described in supplementary Table S5, available at Annals of Oncology online.

of 20.2 months (range 0.4–282). A group of 67 (46%) patients received anti-EGFR immediately after ctDNA collection mainly in second and third line (supplementary Table S4, available at Annals of Oncology online). Median PFS and median OS were described in supplementary Table S5, available at Annals of Oncology online. Correlation between RAS status in tissue and plasma Using qPCR, we found tumor tissue samples positive for KRAS mutations in 44/146 samples (30%) and NRAS mutations in 10/146 (7%) (Table 1; supplementary Table S6, available at Annals of Oncology online). Using BEAMing in tissue samples, KRAS mutations were detected in 49/130 (38%) available samples and NRAS mutations in 11/130 (8%). For ctDNA analysis, 46/146 (31%) and 11/146 (8%) plasma samples harbored KRAS and NRAS mutations, respectively. Table 1 Concordance between tumor-tissue and ctDNA analysis (N = 146) ctDNA analysis Tumor-tissue analysis SoC Sensitivity (%) Specificity (%) PPV (%) NPV (%) KRAS mut NRAS mut WT BEAMing KRAS mut 40 0 6 89 90 84 93 NRAS mut 0 8 3 WT 4 2 83 Total 44 10 92 Tumor-tissue analysis BEAMinga 85 91 89 88 BEAMing KRAS mut 42 0 4 NRAS mut 0 9 2 WT 7 2 64 Total 49 11 70 Tumor-tissue analysis Tumor-tissue analysis SoC 94 88 85 96 BEAMing KRAS mut 42 0 7 NRAS mut 0 9 2 WT 2 1 67 Total 44 10 76 a Tumor-tissue analysis with BEAMing was carried out in 130 samples. WT, wild type; SoC, standard of care; ctDNA, circulating tumor DNA; PPV, positive predictive value; NPV, negative predictive value.

ctDNA analysis Tumor-tissue analysis SoC Sensitivity (%) Specificity (%) PPV (%) NPV (%) KRAS mut NRAS mut WT BEAMing KRAS mut 40 0 6 89 90 84 93 NRAS mut 0 8 3 WT 4 2 83 Total 44 10 92 Tumor-tissue analysis BEAMinga 85 91 89 88 BEAMing KRAS mut 42 0 4 NRAS mut 0 9 2 WT 7 2 64 Total 49 11 70 Tumor-tissue analysis Tumor-tissue analysis SoC 94 88 85 96 BEAMing KRAS mut 42 0 7 NRAS mut 0 9 2 WT 2 1 67 Total 44 10 76 a Tumor-tissue analysis with BEAMing was carried out in 130 samples. WT, wild type; SoC, standard of care; ctDNA, circulating tumor DNA; PPV, positive predictive value; NPV, negative predictive value. Figure 1 shows concordance of RAS status between the three methods. ctDNA analysis showed a Cohen's Kappa estimate of 0.80 (95% CI 0.71–0.90) compared with tumor tissue evaluated by SoC reflecting almost perfect agreement according to Landis and Koch classification [9]. Results were similar for RAS status in plasma and tissue using BEAMing with a Kappa index of 0.79 (95% CI 0.69–0.89), and in tumor tissue using SoC and BEAMing a Kappa index of 0.83 (95% CI 0.74–0.92). Figure 1. Concordance analysis. SoC tumor and BEAMing plasma analysis was carried out in 146 samples, BEAMing tumor was carried out in 130 samples. mut, mutation; SoC, standard of care.

Figure 1 shows concordance of RAS status between the three methods. ctDNA analysis showed a Cohen's Kappa estimate of 0.80 (95% CI 0.71–0.90) compared with tumor tissue evaluated by SoC reflecting almost perfect agreement according to Landis and Koch classification [9]. Results were similar for RAS status in plasma and tissue using BEAMing with a Kappa index of 0.79 (95% CI 0.69–0.89), and in tumor tissue using SoC and BEAMing a Kappa index of 0.83 (95% CI 0.74–0.92). Figure 1. Concordance analysis. SoC tumor and BEAMing plasma analysis was carried out in 146 samples, BEAMing tumor was carried out in 130 samples. mut, mutation; SoC, standard of care. Discordant samples description In the population of samples with discordance between RAS status according to ctDNA BEAMing and tissue by SoC, two groups were identified, as detailed below (Table 2). To clarify these cases, the historical RAS testing was reviewed and additional RAS determinations were carried out by SoC in metastases whenever tissue was available (supplementary Table S2, available at Annals of Oncology online). Table 2 Discordant samples

SoC, two groups were identified, as detailed below (Table 2). To clarify these cases, the historical RAS testing was reviewed and additional RAS determinations were carried out by SoC in metastases whenever tissue was available (supplementary Table S2, available at Annals of Oncology online). Table 2 Discordant samples ID qPCR (SoC) tumor BEAMing plasma BEAMing tumor Additional (SoC) tumorc Historical (SoC) tumorc Codon MAF (BEAMing plasma) adjMAF (BEAMing tumor) Tissue source Tissue tumor area (%) Time tissue- plasma (month) Previous chemo lines Previous treatment receivedd Anti-EGFR after plasma collection and best response Possible explanation Group Aa 1 WT MUT MUT MUT WT NRAS Q61 0.43 0.072 Primary 15 8 1 Capox adyuvant FOLFIRI+ Panitumumab 1L (PR) SoC sensitivity 2 WT MUT MUT MUT WT KRAS A146 0.0065 0.25 Primary 95 1 0 FOLFOX+ Cetuximab 1L (SD) 3 WT MUT MUT NA NA KRAS A146 0.0061 0.29 Primary 50 7 0 No 4 WT MUT MUT NA MUT KRAS G12 0.0006 0.058 Primary 20 108 2 5FU adyuvant, FOLFIRI 1L No 5 WT MUT MUT NA MUT NRAS Q61 0.0005 0.085 Primary 75 2 0 No 6 WT MUT WT NA WT KRAS G12 0.0005 Metastasis 45 4 1 FOLFIRI+BVZ 1L FOLFIRI+ Panitumumab 2L (PR) Molecular heterogeneity 7 WT MUT WT NA WT KRAS G12 0.0008 Primary 70 3 0 No 8 WT MUT WT WT WT KRAS Q61 0.0015 Primary 70 1 0 FOLFIRI+ Cetuximab 1L (PR) 9 WT MUT WT NA WT NRAS Q61 0.0005 Primary 100 10 1 FOLFIRI 1L FOLFIRI+ Panitumumab 2L (PD) Group Bb 10 MUT WT MUT MUTe MUT KRAS G12 0,27 Primary 50 1 0 No Low tumor burden? 11 MUT WT MUT MUT MUT KRAS G12 0,14 Primary 40 33 2 FOLFOX 1L, FOLFIRI 2L No 12 MUT WT MUT NA MUT KRAS G12 0,12 Primary 95 3 0 In course FOLFOX 1L (PR) No Chemotherapy effect? 13 NA WT MUT NA MUT NRAS G13 NA Primary 70 8 0 In course FOLFOX + BVZ 1L (SD) No 14 NA WT MUT NA MUT NRAS Q61 NA Primary 70 4 0 In course FOLFOX 1L (PD) No 15 MUT WT WT NA MUTf KRAS Q61 Primary 60 3 0 No Technical issues? a Group A: mutations detected in plasma but not in tissue by SoC.

1L (PR) No Chemotherapy effect? 13 NA WT MUT NA MUT NRAS G13 NA Primary 70 8 0 In course FOLFOX + BVZ 1L (SD) No 14 NA WT MUT NA MUT NRAS Q61 NA Primary 70 4 0 In course FOLFOX 1L (PD) No 15 MUT WT WT NA MUTf KRAS Q61 Primary 60 3 0 No Technical issues? a Group A: mutations detected in plasma but not in tissue by SoC. b Group B: mutation detected in tissue by SoC but not in plasma. c Supplementary Table S2, available at Annals of Oncology online. d In those patients with plasma extraction during chemotherapy immediate response after extraction is reported between brackets. e Codon NRAS A59. f Codon KRAS G13. SoC, standard of care; MAF, mutant allele fraction; adjMAF, adjusted mutant allele fraction; Chemo, chemotherapy (adyuvant and/or metastatic setting); MUT, mutation; NA, not available; PR, partial response; SD, stable disease; PD, progression disease; 1L, frontline metastatic therapy; 2L, second line metastatic therapy; Capox, Capecitabine + oxaliplatin; 5FU, 5-fluorouracil; BVZ, Bevacizumab; FOLFIRI, 5FU + leucovorin + irinotecan; FOLFOX, 5FU + leucovorin + oxaliplatin. Group A includes patients with evidence of mutations detected in plasma but not in tissue by SoC techniques. In the first five cases the SoC tissue technique failed to detect mutations that were detected in the same tumor sample by BEAMing (Table 2).

SoC, standard of care; MAF, mutant allele fraction; adjMAF, adjusted mutant allele fraction; Chemo, chemotherapy (adyuvant and/or metastatic setting); MUT, mutation; NA, not available; PR, partial response; SD, stable disease; PD, progression disease; 1L, frontline metastatic therapy; 2L, second line metastatic therapy; Capox, Capecitabine + oxaliplatin; 5FU, 5-fluorouracil; BVZ, Bevacizumab; FOLFIRI, 5FU + leucovorin + irinotecan; FOLFOX, 5FU + leucovorin + oxaliplatin. Group A includes patients with evidence of mutations detected in plasma but not in tissue by SoC techniques. In the first five cases the SoC tissue technique failed to detect mutations that were detected in the same tumor sample by BEAMing (Table 2). Interestingly, in cases 1 and 2, SoC analysis of additional metastatic samples showed the same mutations as those found in plasma supporting the concept that plasma can be used to capture tumor heterogeneity. Likewise, in cases 4 and 5 the historical reports showed identical mutated as plasma BEAMing but the new qPCR result was WT. On the remaining four cases (ID 6–9) of this group the mutation detected by plasma BEAMing could not be identified by any other tumor sampling test. These cases appeared not to have specific clinicopathologic features or differential tissue sampling timing.

Interestingly, in cases 1 and 2, SoC analysis of additional metastatic samples showed the same mutations as those found in plasma supporting the concept that plasma can be used to capture tumor heterogeneity. Likewise, in cases 4 and 5 the historical reports showed identical mutated as plasma BEAMing but the new qPCR result was WT. On the remaining four cases (ID 6–9) of this group the mutation detected by plasma BEAMing could not be identified by any other tumor sampling test. These cases appeared not to have specific clinicopathologic features or differential tissue sampling timing. In group B, mutations were detected in tissue but not in plasma in six patients (Table 2). In this group, we also reviewed the CT scan carried out closest to the blood extraction to calculate tumor burden. Patient 10 had small hepatic lesions (<1.5 cm) and patient 11 had only three peritoneal lesions, both of which reflect low tumor burden. For three patients (ID 12–14), plasma extraction was carried out during the course of chemotherapy, which may have altered ctDNA detection. The immediate RECIST 1.1 response after plasma extraction was also reviewed. The last case (ID 15) had discordant results between tissue BEAMing and SoC evaluations even though the DNA for this analysis originated from the same tumoral tissue block. Again, these cases did not have any other particular clinic-pathologic features or differential time to tumor sampling. MAF analysis: distribution and median values

In group B, mutations were detected in tissue but not in plasma in six patients (Table 2). In this group, we also reviewed the CT scan carried out closest to the blood extraction to calculate tumor burden. Patient 10 had small hepatic lesions (<1.5 cm) and patient 11 had only three peritoneal lesions, both of which reflect low tumor burden. For three patients (ID 12–14), plasma extraction was carried out during the course of chemotherapy, which may have altered ctDNA detection. The immediate RECIST 1.1 response after plasma extraction was also reviewed. The last case (ID 15) had discordant results between tissue BEAMing and SoC evaluations even though the DNA for this analysis originated from the same tumoral tissue block. Again, these cases did not have any other particular clinic-pathologic features or differential time to tumor sampling. MAF analysis: distribution and median values RAS MAFs had a median of 0.02 (range 0.0001–0.43) in plasma and were found in a wide distribution, 48% showed <1% (MAF <0.01) mutant alleles in their cfDNA (Figure 2A). RAS-adjusted MAFs had a median of 0.25 (range 0.03–0.99) in tumor tissue.

In group B, mutations were detected in tissue but not in plasma in six patients (Table 2). In this group, we also reviewed the CT scan carried out closest to the blood extraction to calculate tumor burden. Patient 10 had small hepatic lesions (<1.5 cm) and patient 11 had only three peritoneal lesions, both of which reflect low tumor burden. For three patients (ID 12–14), plasma extraction was carried out during the course of chemotherapy, which may have altered ctDNA detection. The immediate RECIST 1.1 response after plasma extraction was also reviewed. The last case (ID 15) had discordant results between tissue BEAMing and SoC evaluations even though the DNA for this analysis originated from the same tumoral tissue block. Again, these cases did not have any other particular clinic-pathologic features or differential time to tumor sampling. MAF analysis: distribution and median values RAS MAFs had a median of 0.02 (range 0.0001–0.43) in plasma and were found in a wide distribution, 48% showed <1% (MAF <0.01) mutant alleles in their cfDNA (Figure 2A). RAS-adjusted MAFs had a median of 0.25 (range 0.03–0.99) in tumor tissue. Figure 2. Mutant allele fraction analysis. (A) RAS mutant allele fractions in ctDNA BEAMing, a MAF of 0.01 corresponds to a percentage of mutant alleles of 1%. (B) Comparison of RAS mutant allele fractions in ctDNA and positivity for RAS mut tumor by SoC testing. (C) Correlation of RAS mutant allele fractions with BEAMing carried out in tumor (adjusted for purity) and ctDNA, according to prior systemic therapy exposure. Samples with RAS wild-type by SoC were excluded. (D) Correlation of RAS mutant allele fractions with BEAMing carried out in tumor (adjusted for purity) and ctDNA, according to number of metastatic sites. Samples with RAS wild-type by SoC were excluded. mut, mutation SoC, standard of care.

ior systemic therapy exposure. Samples with RAS wild-type by SoC were excluded. (D) Correlation of RAS mutant allele fractions with BEAMing carried out in tumor (adjusted for purity) and ctDNA, according to number of metastatic sites. Samples with RAS wild-type by SoC were excluded. mut, mutation SoC, standard of care. In the group of patients with concordant mutant samples in ctDNA and tissue by SoC (N = 48), median MAF in plasma was 0.04 (range 0.0001–0.37) (Figure 2B). In the discordant cases (n = 9) median MAF was 0.0008 (range 0.0004–0.43) (P = 0.069, Kruskal test). In concordant samples by BEAMing tested in both tumor and plasma (N = 48), median adjusted MAF was 0.26 (95% CI 0.04–0.99) in tumor and 0.14 (95% CI 0.05–0.99) (P = 0.16, Kruskal test) in discordant samples (N = 7). Overall, there was a tendency for lower MAFs both in tumor and plasma for the samples with discordant results. The median MAF in ctDNA was also described according to prior chemotherapy exposure and number of metastatic sites. In the first case, median MAF was 0.07 (95% CI 0.002–0.16) and 0.04 (95% CI 0.006–0.15) in those with no prior therapy and those exposed, respectively (P = 0.69, Kruskal test). In the second case, median MAF was 0.05 (95% CI 0.002–0.13) in those with one or two metastatic sites and 0.15 (95% CI 0.009–0.18) in those with three or more (P = 0.24, Kruskal test).

5% CI 0.002–0.16) and 0.04 (95% CI 0.006–0.15) in those with no prior therapy and those exposed, respectively (P = 0.69, Kruskal test). In the second case, median MAF was 0.05 (95% CI 0.002–0.13) in those with one or two metastatic sites and 0.15 (95% CI 0.009–0.18) in those with three or more (P = 0.24, Kruskal test). Correlation of MAF in concordant mutant samples in plasma and tissue We carried out a RAS-adjusted MAF correlation analysis with BEAMing carried out in tumor and ctDNA in the same patient according to prior systemic therapy exposure or number of metastatic sites (Figure 2C and D). Mutational load showed very high heterogeneity and poor correlation, with a Pearson correlation coefficient in the overall population (N = 43) of 0.10 (95% CI −0.21 to 0.39, P = 0.54). RAS status and correlation with anti-EGFR treatment benefit The predictive value of RAS WT status from plasma and tumor determination was analyzed in the subset of patients who received anti-EGFR plus the irinotecan backbone in second- or third-line therapy (N = 52). RAS WT patients detected by SoC (N = 50) had a median PFS of 8.9 months (95% CI 6.8–11.3). RAS WT patients detected by ctDNA (N = 47) showed a median PFS of 8.7 months (95% CI 6.8–11.3) (Figure 3A).

in the subset of patients who received anti-EGFR plus the irinotecan backbone in second- or third-line therapy (N = 52). RAS WT patients detected by SoC (N = 50) had a median PFS of 8.9 months (95% CI 6.8–11.3). RAS WT patients detected by ctDNA (N = 47) showed a median PFS of 8.7 months (95% CI 6.8–11.3) (Figure 3A). Figure 3. (A) Progression-free survival after anti-EGFR plus irinotecan-based therapy in the second or third-line setting in RAS wild-type metastatic colorectal cancer patients according to method of RAS mutation detection (SoC tumor tissue at baseline N =50 or ctDNA plasma before therapy N =47). (B) Survival in metastatic setting according to RAS mutant allele fraction by ctDNA plasma. MAF of 0.1 corresponds to a percentage of mutant alleles of 10%. SoC, standard of care. Potential impact in OS We describe outcomes for OS according to RAS MAF detection by ctDNA (Figure 3B). In the group of patients with RAS mutant samples with MAF < 0.1 by ctDNA (N = 40), median OS was 27.8 months (95% CI 24.9–47.2), with an HR of 1.60 (95% CI 0.95–2.73; P = 0.08) when compared with RAS WT population. In the group with MAF ≥0.1 (n = 16) median OS was 16.4 months (95% CI 11.4–not reached), and the HR for this group was 2.87 (95% CI 1.46–5.67, P = 0.002) when compared with RAS WT population.

ian OS was 27.8 months (95% CI 24.9–47.2), with an HR of 1.60 (95% CI 0.95–2.73; P = 0.08) when compared with RAS WT population. In the group with MAF ≥0.1 (n = 16) median OS was 16.4 months (95% CI 11.4–not reached), and the HR for this group was 2.87 (95% CI 1.46–5.67, P = 0.002) when compared with RAS WT population. Relevant parameters were included in a multivariable Cox proportional hazards model on the entire cohort: mutation status and MAF in two ranges by ctDNA, tumor location and number of metastatic sites. RAS mutation with MAF ≥0.1 by ctDNA was shown to be a significant prognostic factor with a HR of 2.47 (95% CI 1.2–5.0, P = 0.01) (supplementary Table S7, available at Annals of Oncology online). Discussion This is the first clinical series showing the usefulness of detecting RAS point mutations by ctDNA in the largest cohort of patients published so far and carried out locally in a general hospital. Our data revealed a very high overall concordance, close to 90% compared with gold standard tumor tissue analysis techniques. This result is in accordance with previous reports, where RAS mutation detection in cfDNA has been directly compared with tumor tissue in CRC cohorts [4–6]. Siravegna et al. [7] focused on clonal evolution and resistance to EGFR blockade, also described excellent concordance in matched tissue and plasma samples using droplet digital PCR (N = 100). Our results prove the feasibility for implementing this technique in the day-by-day care.

th tumor tissue in CRC cohorts [4–6]. Siravegna et al. [7] focused on clonal evolution and resistance to EGFR blockade, also described excellent concordance in matched tissue and plasma samples using droplet digital PCR (N = 100). Our results prove the feasibility for implementing this technique in the day-by-day care. The detailed description of discordant samples reflected in Table 2 confirms the complexity of RAS genotyping in both tumor tissue and plasma samples. Translation of these new technologies to clinical practice reveal not only the technical limitations, but also bring to light conflicting data that provide information about the biological behavior of each tumor. Tumor tissue genotyping has inherent limitations the genomic profiles of primary tumors and metastases are not always concordant owing to the intrinsic molecular tumor heterogeneity [10, 11]. Likewise, several reports have shown differences ranging 3%–20% between different techniques to detect RAS mutations in tissue [12–14]. When analyzing tumor tissue by SoC and BEAMing analysis we detected a 9.1% rate of discordance, mostly justified by differences in sensitivity cut-off.

ar tumor heterogeneity [10, 11]. Likewise, several reports have shown differences ranging 3%–20% between different techniques to detect RAS mutations in tissue [12–14]. When analyzing tumor tissue by SoC and BEAMing analysis we detected a 9.1% rate of discordance, mostly justified by differences in sensitivity cut-off. To account for spatial and temporal changes, the genomic profiles of CRC patients should be evaluated repeatedly during the course of therapy and liquid biopsies could play a role for determinations that are more representative of the specific molecular scenario of a patient at the time of anti-EGFR therapy selection [7, 15]. The possibility of RAS testing at the time of decision-making is one of the strongest points arguing in favor of this minimally invasive technique. Furthermore, we consider several issues regarding RAS genotyping in plasma need to be highlighted. In our cohort, six patients had mutations in tissue that could not be detected in plasma. Lack of RAS mutations in plasma may be attributed to biological factors that impact ctDNA release and is an important matter that should be investigated. False negative results represent a major issue for RAS mutation testing on plasma because of the possible negative interaction of anti-EGFR agents with oxaliplatin-based regimens in RAS mutant patients.

be attributed to biological factors that impact ctDNA release and is an important matter that should be investigated. False negative results represent a major issue for RAS mutation testing on plasma because of the possible negative interaction of anti-EGFR agents with oxaliplatin-based regimens in RAS mutant patients. Commonly used chemotherapeutic agents as well as targeted drugs can alter the molecular landscape in these tumors. It is widely acknowledged that acquired KRAS mutations are associated with secondary resistance to EGFR blockade [15, 16]. However, the effect on the molecular profile derived from other therapies such as anti-angiogenics or cytostatic agents before anti-EGFR administration is yet to be determined [17, 18]. Patients 6 and 9 (Table 2) may be such cases. Tie et al. [19] reported changes in ctDNA for mCRC patients during the course of the chemotherapy, with significant reductions in ctDNA levels (median 5.7-fold) observed before cycle 2 in 41 of the 48 patients with concordant mutant samples in ctDNA and tissue by SoC. This could impact RAS status determination in patients exposed to therapy, we hypothesize that this could be the case for three patients in our cohort (ID 12–14 in Table 2), although we could not associate this with a homogeneous pattern of response. Taking this a step further, we detected a lower median MAF in the group of patients exposed to prior therapy.

on in patients exposed to therapy, we hypothesize that this could be the case for three patients in our cohort (ID 12–14 in Table 2), although we could not associate this with a homogeneous pattern of response. Taking this a step further, we detected a lower median MAF in the group of patients exposed to prior therapy. If ultimately we move towards routine RAS determination in plasma in clinical practice, there will likely be subgroups of patients in whom we should continue to perform determinations in tissue for possible alterations in ctDNA release after a negative liquid biopsy.

on in patients exposed to therapy, we hypothesize that this could be the case for three patients in our cohort (ID 12–14 in Table 2), although we could not associate this with a homogeneous pattern of response. Taking this a step further, we detected a lower median MAF in the group of patients exposed to prior therapy. If ultimately we move towards routine RAS determination in plasma in clinical practice, there will likely be subgroups of patients in whom we should continue to perform determinations in tissue for possible alterations in ctDNA release after a negative liquid biopsy. Although the cohort size of patients with mutations (N = 48) in our study is a somewhat limiting factor, we nonetheless could draw interesting conclusions from analyzing MAF, providing to our knowledge, the first published data in this field. When considering MAF distribution, a high proportion of patients showed mutant alleles in cfDNA between 0.0001 (0.01%) and 0.01 (1%). This highlights the importance of using an extremely sensitive technique when analyzing plasma samples and must be considered at the time of analysis to translate this into clinical practice. Interestingly, there is a tendency for lower MAFs both in tumor tissue and plasma for samples with discordant results, suggesting that sensitivity for mutation detection in tumor tissue is a real issue that needs to be addressed. We found no correlation of RAS MAF with BEAMing carried out in tumor and ctDNA, regardless of prior systemic therapies. The concept of a cut-off for plasma samples similar to that applied in tissue is complex and in our interpretation should not be equivalent.

in tumor tissue is a real issue that needs to be addressed. We found no correlation of RAS MAF with BEAMing carried out in tumor and ctDNA, regardless of prior systemic therapies. The concept of a cut-off for plasma samples similar to that applied in tissue is complex and in our interpretation should not be equivalent. Finally, in an exploratory analysis, and as an indirect way of confirming the possibility of selecting patients for anti-EGFR therapy with plasma, a PFS analysis was carried out in the most homogeneous group of our cohort, showing no relevant differences between detection methods. To our knowledge no other concordance studies have reported this, and this type of analysis is relevant to the implementation of liquid biopsies in clinical practice. We can conclude that ctDNA analysis in plasma can detect RAS mutations to an equivalent level as SoC techniques in tissue, and thus detecting potential mCRC patients who could benefit from anti-EGFR therapies. Supplementary Material mdx112_supp Click here for additional data file. Acknowledgements The authors acknowledge the excellent medical editing assistance of Sarah MacKenzie (PhD). We want to acknowledge the Cellex Foundation for providing facilities and equipment as well as the Tumor Biomarkers Research Program of the Banco Bilbao Vizcaya Argentaria (BBVA) Foundation for their financial support to the Cancer Genomics Lab, VHIO.

e the excellent medical editing assistance of Sarah MacKenzie (PhD). We want to acknowledge the Cellex Foundation for providing facilities and equipment as well as the Tumor Biomarkers Research Program of the Banco Bilbao Vizcaya Argentaria (BBVA) Foundation for their financial support to the Cancer Genomics Lab, VHIO. Funding This work was supported by Merck, S.L., Madrid, Spain, an affiliate of Merck KGaA, Darmstadt, Germany and partially by the Instituto de Salud Carlos III (Ministerio de Economía y Competitividad) and ‘Fondo Europeo de Desarrollo Regional (FEDER), una manera de hacer Europa’ grants [FIS PI12-01589 to RS] and RETICC Cancer: Grupo Cáncer Digestivo – Instituto de Salud Carlos III.TTD ULTRA study (EC11-050) was supported by the Ministerio de Sanidad y Politica Social [SPI/2885/2011]. To CM grants [PI15/00457 and DTS15/00048]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Funding This work was supported by Merck, S.L., Madrid, Spain, an affiliate of Merck KGaA, Darmstadt, Germany and partially by the Instituto de Salud Carlos III (Ministerio de Economía y Competitividad) and ‘Fondo Europeo de Desarrollo Regional (FEDER), una manera de hacer Europa’ grants [FIS PI12-01589 to RS] and RETICC Cancer: Grupo Cáncer Digestivo – Instituto de Salud Carlos III.TTD ULTRA study (EC11-050) was supported by the Ministerio de Sanidad y Politica Social [SPI/2885/2011]. To CM grants [PI15/00457 and DTS15/00048]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Disclosure JMV has received honoraria for advisory role from Merck, S.L., Madrid, Amgen and investigation from Merck, S.L., Madrid. CL has received honoraria for advisory role and investigation from Merck, S.L., Madrid, Amgen, Roche, Sanofi. EA has received honoraria for advisory role from Amgen, Bayer, Celgene, Merck, S.L., Madrid, Roche, Sanofi. FJ and VS are employees of Sysmex Inostics, Inc. JT has served in a consulting or advisory role for Amgen, Boehringer Ingelheim, Celgene, Chugai, Imclone, Lilly, Merck, S.L., Madrid, Merck Serono, Millennium Pharmaceuticals, Inc., Novartis, Roche, Sanofi, and Taiho. CM has served in a consultant or advisory role for Merck, S.L., Madrid, Roche and Sanofi. RS has served in a consultant or advisory role for Amgen, Merck, S.L., Madrid, Roche Dx and research funding for Roche Dx. AV has served in a consultant or advisory role for Merck, S.L., Madrid, Merck Serono and Sysmex. All remaining authors have declared no conflicts of interest.

Introduction Cancer mutations encode novel (neo-)peptides that can be presented to T cells by major histocompatibility complex (MHC) molecules. A subset of neopeptides are sufficiently self-dissimilar to be immune targets and are termed neoantigens. These can render individual tumours uniquely antigenic [1, 2] by forming a substrate for lymphocyte mediated anti-cancer immunity. We and others have identified neoantigen reactive T cells in non-small-cell lung cancer (NSCLC) [3], melanoma [4] and gastrointestinal tumours [5], with observational and experimental evidence to suggest their role in clinically relevant tumour control [6, 7]. Based on high-throughput tumour genomic analysis, each nonsynonymous mutation (referred to as mutation, hereafter) can potentially give rise to multiple neopeptides, resulting in a vast total number. Specific identification of immunogenic candidates is consequently a major challenge. Studies of immunity to infectious agents first suggested peptide immunogenicity is associated with affinity for MHC, with strong binders considered more likely to achieve cell surface presentation and thus the opportunity for immune recognition [8]. In support of a relationship between predicted neopeptide affinity and immunogenicity, recent clinical studies have shown a correlation between the burden of strong-binding neopeptides with affinity for MHC-I of <500 nM (referred to as neoantigens) and patient outcome in advanced melanoma and lung cancer [9–11].

Based on high-throughput tumour genomic analysis, each nonsynonymous mutation (referred to as mutation, hereafter) can potentially give rise to multiple neopeptides, resulting in a vast total number. Specific identification of immunogenic candidates is consequently a major challenge. Studies of immunity to infectious agents first suggested peptide immunogenicity is associated with affinity for MHC, with strong binders considered more likely to achieve cell surface presentation and thus the opportunity for immune recognition [8]. In support of a relationship between predicted neopeptide affinity and immunogenicity, recent clinical studies have shown a correlation between the burden of strong-binding neopeptides with affinity for MHC-I of <500 nM (referred to as neoantigens) and patient outcome in advanced melanoma and lung cancer [9–11]. However, strategies to improve neopeptide immunogenicity prediction are required, as illustrated by multiple studies revealing T-cell responses to only 0%–3% of predicted strong binders [3, 12, 13] and the recognition of neoantigens of low predicted affinity [5, 14]. Suboptimal performance of affinity prediction algorithms likely contributes to these findings, but the description of immunogenic neopeptides with very low invitro measured MHC affinity supports the concept that weak MHC binding alone is insufficient to exclude immunogenicity [15].

ns of low predicted affinity [5, 14]. Suboptimal performance of affinity prediction algorithms likely contributes to these findings, but the description of immunogenic neopeptides with very low invitro measured MHC affinity supports the concept that weak MHC binding alone is insufficient to exclude immunogenicity [15]. We have recently shown that less heterogenous cancers bearing a high burden of clonal neoantigens shared by all cells are more strongly associated with survival than simply those with high neoantigen burden, suggesting neopeptide features other than affinity contribute to patient outcomes [3]. Tumour reactive T cells can differentiate between self and mutant peptides that differ by a single amino acid. Mechanistically, preclinical work suggests the difference in predicted affinity for any given wild-type/mutant peptide pair (termed differential agretopicity index; DAI) is a broad indicator of neopeptide dissimilarity from self and a feature of immunogenicity. For individual peptides in a preclinical model, DAI was reportedly a better indicator of immunogenicity than mutant affinity [15]. Extending this to human tumours, we hypothesised that tumours enriched for high DAI neopeptides may be more susceptible to immune recognition and hence clinically relevant tumour control. As immunogenic strong binding neopeptides are described in both lung cancer [3] and melanoma [4], we additionally hypothesised DAI may be of particular relevance amongst this subset.

hat tumours enriched for high DAI neopeptides may be more susceptible to immune recognition and hence clinically relevant tumour control. As immunogenic strong binding neopeptides are described in both lung cancer [3] and melanoma [4], we additionally hypothesised DAI may be of particular relevance amongst this subset. Using sequencing data from the Cancer Genome Atlas (TCGA) and three published cohorts of patients with advanced melanoma and lung cancer re-analysed with our peptide affinity prediction pipeline, we investigated the relationship between patient survival, markers of immune activity and DAI, to define whether this measurement is relevant to the human anti-tumour immune response. Methods Clinical cohorts and outcome assessments Cohorts of patients with stage III/IV lung adenocarcinoma (LUAD; n = 106/522) and stage IIIC/IV cutaneous melanoma (SKCM; n = 145/470) were identified from TCGA and served as datasets for initial evaluation of the association between DAI on survival. Advanced stage TCGA patients were selected to match immunotherapy-treated cohorts described below. Patients with stage III and IV disease in the LUAD cohort had similar survival outcomes and the former group was therefore included in analyses. Further datasets of immunotherapy-treated patients comprised a cohort with stage IV NSCLC of predominantly adenocarcinoma subtype treated with pembrolizumab [11] and two cohorts of patients with advanced melanoma treated with anti-CTLA-4 directed immunotherapies [9, 10].

Methods Clinical cohorts and outcome assessments Cohorts of patients with stage III/IV lung adenocarcinoma (LUAD; n = 106/522) and stage IIIC/IV cutaneous melanoma (SKCM; n = 145/470) were identified from TCGA and served as datasets for initial evaluation of the association between DAI on survival. Advanced stage TCGA patients were selected to match immunotherapy-treated cohorts described below. Patients with stage III and IV disease in the LUAD cohort had similar survival outcomes and the former group was therefore included in analyses. Further datasets of immunotherapy-treated patients comprised a cohort with stage IV NSCLC of predominantly adenocarcinoma subtype treated with pembrolizumab [11] and two cohorts of patients with advanced melanoma treated with anti-CTLA-4 directed immunotherapies [9, 10]. Following filtering to retain high-quality samples (see supplementary Methods, available at Annals of Oncology online), final cohorts consisted of n = 66 LUAD, n = 75 SKCM, n = 78 Van Allen, n = 31 Rizvi and n = 51 Snyder patients. Patient survival was the primary outcome measure in this study. For TCGA datasets, Snyder [9] and Van Allen [10], overall survival data are available. For the Rizvi cohort [11], progression-free survival only is available.

Following filtering to retain high-quality samples (see supplementary Methods, available at Annals of Oncology online), final cohorts consisted of n = 66 LUAD, n = 75 SKCM, n = 78 Van Allen, n = 31 Rizvi and n = 51 Snyder patients. Patient survival was the primary outcome measure in this study. For TCGA datasets, Snyder [9] and Van Allen [10], overall survival data are available. For the Rizvi cohort [11], progression-free survival only is available. Neopeptide prediction and DAI analysis Full details of the informatics pipeline used to identify patient HLA status, nonsynonymous mutations and the predicted MHC-I binding affinity of mutant peptides have previously been published [3] and outlined in the supplementary Methods, available at Annals of Oncology online. Mutation clonality was inferred from single sample sequenced tumours using a modified version of PyClone as previously described. To calculate DAI, MHC-I affinity was predicted for mutant and wild-type peptide pairs arising from the same mutation and differing by a single amino acid. The DAI of each mutant peptide was calculated by subtraction of its predicted binding affinity from the value of the corresponding wild-type peptide. Further details are within the supplementary Methods, available at Annals of Oncology online.

e pairs arising from the same mutation and differing by a single amino acid. The DAI of each mutant peptide was calculated by subtraction of its predicted binding affinity from the value of the corresponding wild-type peptide. Further details are within the supplementary Methods, available at Annals of Oncology online. Results To assess the relationship between DAI and patient survival in advanced cancer, we selected TCGA and immunotherapy-treated cohorts of patients with advanced lung cancer and melanoma for whom high quality whole exome sequencing and outcome data were available, with demographics summarised in supplementary Table S1, available at Annals of Oncology online. Preclinical work has previously found high DAI peptides to be preferentially mutated at anchor residues [15] and we tested this relationship in human samples. Amongst all 9mer peptides from the LUAD cohort (n = 166 746), we found a strong correlation between probability of anchor residue mutation and DAI predicted for HLA-A, with close to 100% of the most positive and negative DAI peptides mutated at anchor residues P2 and P9 (supplementary Figure S1, available at Annals of Oncology online).

peptides from the LUAD cohort (n = 166 746), we found a strong correlation between probability of anchor residue mutation and DAI predicted for HLA-A, with close to 100% of the most positive and negative DAI peptides mutated at anchor residues P2 and P9 (supplementary Figure S1, available at Annals of Oncology online). Mean DAI was selected to summarise DAI values for each sample. For individual patients and across cohorts, mean DAI was found to associate with both maximum DAI and the proportion of peptides with DAI >0 nM (Figure 1). As an indicator of both DAI magnitude and positive skew, mean DAI therefore represents a suitable indicator of samples enriched for high DAI neopeptides. Whilst mean DAI distribution was similar across melanoma cohorts, LUAD patients had significantly higher values compared with Rizvi [11] (Figure 2; supplementary Table S2, available at Annals of Oncology online). Figure 1 (A) Distribution of DAI for all peptides in three LUAD samples (highest, average and lowest mean DAI, respectively). (B–D) Correlation between mean DAI and non-synonymous (NS) mutation load, proportion of peptides with a DAI >0 and maximum DAI across five cohorts was evaluated by linear regression. Figure 2 . Density plots representing the distribution of mean DAI across cohorts, with dotted lines indicating the first quartile cut point used to stratify patients for subsequent survival analysis in LUAD and Rizvi lung cancer cohorts. One way ANOVA P-values are shown.

Figure 1 (A) Distribution of DAI for all peptides in three LUAD samples (highest, average and lowest mean DAI, respectively). (B–D) Correlation between mean DAI and non-synonymous (NS) mutation load, proportion of peptides with a DAI >0 and maximum DAI across five cohorts was evaluated by linear regression. Figure 2 . Density plots representing the distribution of mean DAI across cohorts, with dotted lines indicating the first quartile cut point used to stratify patients for subsequent survival analysis in LUAD and Rizvi lung cancer cohorts. One way ANOVA P-values are shown. As the relationship between mean DAI and outcomes was non-linear, patients were stratified into mean DAI quartiles for survival analysis. Thresholds of low versus high mean DAI for each cancer type were defined in TCGA data. By subsequently applying these thresholds for outcome analysis in the corresponding immunotherapy-treated cohorts, we tested for a survival signal magnified by treatment. In LUAD, cut point analysis by inspection of survival curves and univariate Cox regression (supplementary Figure S2A, available at Annals of Oncology online) revealed low mean DAI (≤lower quartile) to significantly associate with worse overall survival (P = 0.004; Figure 3A). The correlation between mean DAI and survival at this threshold was replicated in the Rizvi cohort [11] (P = 0.002).

te Cox regression (supplementary Figure S2A, available at Annals of Oncology online) revealed low mean DAI (≤lower quartile) to significantly associate with worse overall survival (P = 0.004; Figure 3A). The correlation between mean DAI and survival at this threshold was replicated in the Rizvi cohort [11] (P = 0.002). Figure 3 . Kaplan–Meier survival curves for patients with advanced lung cancer (A; TCGA LUAD and Rizvi cohorts) and melanoma (B; TCGA SKCM and Van Allen cohorts), stratified into high and low comparator groups for each variable (columns). (A) Mean DAI was calculated for all predicted neopeptides. For each variable, patients were stratified into high (>first quartile) and low (<first quartile) subgroups, respectively. (B) Mean DAI was calculated for mutant peptides with a predicted MHC affinity of <500 nM (neoantigens). For each variable, patients were stratified into high (>median) and low (<median) subgroups respectively. Mean DAI (mDAI) was additionally calculated for peptides arising from clonal and subclonal mutations. Log rank P-values are shown. Mutational and neoantigen burden refers to the number of nonsynonymous mutations and neoantigens, respectively. OS, overall survival; PFS, progression free survival; NA, neoantigen. In comparison, neoantigen but not mutational load was found to predict survival in LUAD (P = 0.023 and 0.675, respectively) and neither were predictive in Rizvi [11] using the same threshold to define low versus high groups.

Figure 3 . Kaplan–Meier survival curves for patients with advanced lung cancer (A; TCGA LUAD and Rizvi cohorts) and melanoma (B; TCGA SKCM and Van Allen cohorts), stratified into high and low comparator groups for each variable (columns). (A) Mean DAI was calculated for all predicted neopeptides. For each variable, patients were stratified into high (>first quartile) and low (<first quartile) subgroups, respectively. (B) Mean DAI was calculated for mutant peptides with a predicted MHC affinity of <500 nM (neoantigens). For each variable, patients were stratified into high (>median) and low (<median) subgroups respectively. Mean DAI (mDAI) was additionally calculated for peptides arising from clonal and subclonal mutations. Log rank P-values are shown. Mutational and neoantigen burden refers to the number of nonsynonymous mutations and neoantigens, respectively. OS, overall survival; PFS, progression free survival; NA, neoantigen. In comparison, neoantigen but not mutational load was found to predict survival in LUAD (P = 0.023 and 0.675, respectively) and neither were predictive in Rizvi [11] using the same threshold to define low versus high groups. High-affinity neoantigens are targeted by T cells in both NSCLC [3] and melanoma [4], suggesting peptides with combined high affinity and DAI may be particularly immunogenic. We tested this hypothesis by calculating mean DAI amongst neoantigens of predicted affinity <500 nM. Neoantigen mean DAI was not associated with survival in LUAD (P = 0.66), but was in Rizvi [11] (P = 0.04, supplementary Figure S3, available at Annals of Oncology online).

finity and DAI may be particularly immunogenic. We tested this hypothesis by calculating mean DAI amongst neoantigens of predicted affinity <500 nM. Neoantigen mean DAI was not associated with survival in LUAD (P = 0.66), but was in Rizvi [11] (P = 0.04, supplementary Figure S3, available at Annals of Oncology online). Applying this approach to melanoma, mean DAI of all peptides was not associated with overall survival in SKCM. Excluding low-affinity peptides, we calculated the neoantigen mean DAI. Neoantigen mean DAI was similarly distributed across the melanoma cohorts (supplementary Figure S4 and Table S3, available at Annals of Oncology online). Cut point analysis revealed low neoantigen mean DAI (≤median; supplementary Figure S2B, available at Annals of Oncology online) to associate with a non-statistically significant trend to poor overall survival in SKCM (P = 0.068; Figure 3B). Applying the same threshold, we found low neoantigen mean DAI to correlate with poorer survival in the Van Allen [10] (P = 0.003, Figure 3B) but not Snyder cohorts [9] (supplementary Figure S5, available at Annals of Oncology online, P = 0.582). Neither neoantigen nor mutational burden correlated with survival in the melanoma cohorts, although tests of association at other thresholds were not carried out.

Women who self-reported medication cessation at a visit were classified as non-adherent. Each woman was assessed for persistent use of medication for at least 4.5 years (adherent) or stopping before 4.5 years (non-adherent). All women in the prevention and DCIS IBIS-II trials have finished 5 years of active treatment. Participant symptoms Symptoms were assessed at each follow-up visit using pre-defined items for arthralgia (arthritis, arthrosis, or joint disorder), hot flashes/night sweats, vaginal discharge, irregular vaginal bleeding, eye diseases/cataracts, and osteoporosis/fractures. Vaginal discharge and irregular vaginal bleeding were grouped together as gynecological symptoms because they are similar. All symptoms were classified as mild, moderate, or severe as judged by the women. The most severe gynecological symptom was used when computing this item. Statistical analysis Adherence to trial medication was calculated using the Kaplan–Meier method [18], both overall and by treatment group separately. Further details of statistical methods are provided in supplementary material, available at Annals of Oncology online.

urvival in the Van Allen [10] (P = 0.003, Figure 3B) but not Snyder cohorts [9] (supplementary Figure S5, available at Annals of Oncology online, P = 0.582). Neither neoantigen nor mutational burden correlated with survival in the melanoma cohorts, although tests of association at other thresholds were not carried out. As affinity prediction values may be most accurate for 9mer peptides, mean DAI was recalculated for 9mers in the LUAD and SKCM cohorts, and was found to correlate better with survival compared with all-mer mean DAI, particularly in the SKCM cohort (supplementary Figure S6, available at Annals of Oncology online; all-mer P = 0.069, 9mer P = 0.035). Low neoantigen intra-tumoural heterogeneity (defined as the proportion of neoantigens derived from subclonal mutations) combined with high neoantigen burden is a superior measure of patient outcome compared with the latter alone and we have additionally shown the immunogenicity of clonal neoantigens [3]. This subset may play an important role in anti-cancer immunity and we therefore next evaluated the association between survival and mean DAI of peptides according to clonality. For cohorts within which mean DAI was associated with survival, this was the case when calculated for neopeptides arising from clonal but not subclonal mutations (Figure 3).

ant role in anti-cancer immunity and we therefore next evaluated the association between survival and mean DAI of peptides according to clonality. For cohorts within which mean DAI was associated with survival, this was the case when calculated for neopeptides arising from clonal but not subclonal mutations (Figure 3). Mean DAI weakly correlated with mutational burden in melanoma but not lung cancer cohorts (Figure 1) so we carried out multivariate Cox regression to control for potential confounding effects. After correction for multiple factors, the correlation between mean DAI and survival remained significant in LUAD, Rizvi and Van Allen (Figure 4). As a continuous variable, mutational burden did not correlate with survival in the four cohorts tested. Figure 4 . Multivariate Cox regression modelling of survival in advanced lung cancer (A) and melanoma (B). NS, non-synonymous; NA, neoantigen; HR, hazard ratio; CI, confidence interval. Data on n = 74/75 SKCM patients available for analysis. In SKCM, lymphocyte density and distribution were previously measured to define a semi-quantitative lymphocyte score and an RNA expression profile of high immune infiltration was determined [16]. Whilst neoantigen mean DAI did not correlate with high lymphocyte infiltration (score >2 as originally defined) nor immune gene expression when tested individually (Figure 5A), patients with both factors were found to have a significantly higher neoantigen mean DAI (P = 0.027), with no difference in mutational nor neoantigen burden (Figure 5B and C).

ot correlate with high lymphocyte infiltration (score >2 as originally defined) nor immune gene expression when tested individually (Figure 5A), patients with both factors were found to have a significantly higher neoantigen mean DAI (P = 0.027), with no difference in mutational nor neoantigen burden (Figure 5B and C). Figure 5 . TCGA patients with advanced melanoma have previously been stratified into high and low immune-infiltrated groups based on unsupervised cluster analysis of transcriptomic data (RNAseq cluster) and histopathological assessment of lymphocyte density and distribution (lymphocyte score, LS). (A) Patients with immune-infiltrated tumours as defined by RNAseq cluster combined with a high LS have a significantly higher neoantigen mean DAI. (B, C) Mutational and neoantigen burden were not different between high- and low-infiltrated groups. Wilcoxon rank sum test P-values are shown. In LUAD, a 13 gene MHC-II expression signature has recently been shown to strongly correlate with the presence of multiple MHC-II expressing immune cell types, serving as a proxy measure of infiltration [17]. As tissue MHC-II expression is upregulated by IFN-γ produced during effector T cell activation, this signature may additionally represent a marker of immune activity. Having stratified the cohort by MHC-II expression score based on TCGA RNA-sequencing data, high expression (>median) significantly correlated with mean DAI (P = 0.024), but not mutational nor neoantigen burden (Figure 6A).

ffector T cell activation, this signature may additionally represent a marker of immune activity. Having stratified the cohort by MHC-II expression score based on TCGA RNA-sequencing data, high expression (>median) significantly correlated with mean DAI (P = 0.024), but not mutational nor neoantigen burden (Figure 6A). Figure 6 . (A) A 13-gene MHC-II expression signature has previously been shown to correlate with immune infiltration in LUAD. Patients with a high (above the median) MHC-II expression score have higher mean DAI but no difference in mutational/neoantigen burden. (B) Patients in the Rizvi cohort were stratified into high and low PD-L1 expression groups based on previously published histopathological evaluation (n = 29 available for analysis). There is a non-statistically significant trend of association between PD-L1 expression and mutation/neoantigen burden and mean DAI. Wilcoxon rank sum test P-values are shown. In the absence of RNA-sequencing data for the Rizvi dataset, we investigated the relationship between mean DAI and tumour PDL1 expression as an indicator of immune infiltration, and found a non-statistically significant association (Figure 6B).

mdx490_supplementary_table_s1_s3_s4 Click here for additional data file. mdx490_supplementary_table_s2 Click here for additional data file. mdx490_supplementary_methods_corrected Click here for additional data file. Acknowledgement The authors would like to recognize the important contributions from other researchers whose work could not be cited due to space constraints. Funding None declared. Disclosure JHC, DP, RH, LY, ABS, LMG, GMF, PJS, DL, TIM, VAM, JSR, and SMA are employees and hold equity stake in Foundation Medicine, Inc. BHP is a paid consultant for Foundation Medicine, Inc. and has research contracts with Foundation Medicine, Inc. All remaining authors have declared no conflicts of interest.

Figure 6 . (A) A 13-gene MHC-II expression signature has previously been shown to correlate with immune infiltration in LUAD. Patients with a high (above the median) MHC-II expression score have higher mean DAI but no difference in mutational/neoantigen burden. (B) Patients in the Rizvi cohort were stratified into high and low PD-L1 expression groups based on previously published histopathological evaluation (n = 29 available for analysis). There is a non-statistically significant trend of association between PD-L1 expression and mutation/neoantigen burden and mean DAI. Wilcoxon rank sum test P-values are shown. In the absence of RNA-sequencing data for the Rizvi dataset, we investigated the relationship between mean DAI and tumour PDL1 expression as an indicator of immune infiltration, and found a non-statistically significant association (Figure 6B). Discussion Multiple factors including peptide abundance, MHC affinity, stability and amino acid composition shape peptide immunogenicity [18]. Self-similar HIV peptides are less likely to be T cell targets [19] and we have recently demonstrated the importance of frameshift insertion and deletion (indel) cancer mutations in generating highly self-dissimilar neoantigens that correlate with immunotherapy efficacy [20]. The notion that peptide self-dissimilarity favours immune recognition was developed in pre-clinical work that defined DAI as an indicator of immunogenicity [15]. In this study, predicted affinity alone as an immunogenicity marker was challenged by the finding that 8/10 immunogenic high DAI peptides had a measured affinity of >500 nM, of which 6 had affinity of >50 000 nM indicative of weak/non-MHC binding by conventional criteria.

ned DAI as an indicator of immunogenicity [15]. In this study, predicted affinity alone as an immunogenicity marker was challenged by the finding that 8/10 immunogenic high DAI peptides had a measured affinity of >500 nM, of which 6 had affinity of >50 000 nM indicative of weak/non-MHC binding by conventional criteria. Should DAI mark peptide immunogenicity, we reasoned that samples enriched for high DAI peptides may be more susceptible to immune recognition, translating to a survival advantage. Mean DAI correlates with the proportion of peptides with DAI >0 nM and the sample maximum DAI, which are individually associated with survival in independent cohorts (data not shown). As mean DAI captures both metrics, it was chosen as a summary score to indicate tumours biased towards high DAI peptides. Assuming peptide synthesis rates are DAI independent, a greater intracellular representation of high DAI neopeptides would be expected in samples enriched for mutations engendering these species. Protein abundance correlates with the probability of MHC presentation [21] and as the capacity of peptide presentation pathways is finite, it is conceivable that presentation of high DAI neopeptides is favoured in samples with positively skewed DAI.

expected in samples enriched for mutations engendering these species. Protein abundance correlates with the probability of MHC presentation [21] and as the capacity of peptide presentation pathways is finite, it is conceivable that presentation of high DAI neopeptides is favoured in samples with positively skewed DAI. The demonstration that mean DAI correlates with survival in three cohorts of patients across two tumour types supports DAI as a potential contributor to peptide immunogenicity in cancer. Ongoing work is aimed at further refining DAI summary metrics to improve upon the performance of the sample mean and more precisely characterise features of immunogenicity amongst high DAI peptides. We have previously reported that neoantigen clonal architecture is a likely determinant of the anti-cancer immune response [3]. The finding that mean DAI of clonal but not subclonal mutations is associated with patient survival supports the hypothesis that effective immune responses are directed towards the latter.

The demonstration that mean DAI correlates with survival in three cohorts of patients across two tumour types supports DAI as a potential contributor to peptide immunogenicity in cancer. Ongoing work is aimed at further refining DAI summary metrics to improve upon the performance of the sample mean and more precisely characterise features of immunogenicity amongst high DAI peptides. We have previously reported that neoantigen clonal architecture is a likely determinant of the anti-cancer immune response [3]. The finding that mean DAI of clonal but not subclonal mutations is associated with patient survival supports the hypothesis that effective immune responses are directed towards the latter. Although recent landmark studies correlated mutational/neoantigen burden and clinical outcome, our results are divergent. Van Allen et al. [10] measured a composite clinical outcome, as opposed to overall survival used here. Snyder et al. [9] found overall survival was correlated with high-mutational burden in discovery but not validation sets. In a subsequent re-analysis, the association was limited to patients who underwent tissue biopsy before but not post therapy initiation [22]. In the Rizvi study [11], above median mutational/neoantigen burden correlated with survival, but this association is not apparent here using the lower quartile as a stratification threshold. Neoantigen but not mutational burden was associated with improved survival in LUAD (Figure 3A), arguing against these factors as determinants of survival in advanced NSCLC generally.

Key Message APOBEC mutagenesis creates diversity within viral and tumour cell populations. Sublethal APOBEC mutagenesis underlies the formation of HIV drug resistant and immune escape variants, and similar functions for APOBEC in cancer are expected. Incomplete inhibition may promote escape whereas, complete APOBEC inhibition may be attractive as adjuvant therapy. Introduction Apolipoprotein B mRNA-editing enzyme, catalytic polypeptide-like 3 (APOBEC3; A3) is the name of a seven-membered family of single-stranded DNA cytosine deaminases in humans. Independent approaches including analyses of next-generation sequencing data implicated APOBEC-catalysed DNA damage and mutagenesis in breast cancer [1, 2]. Subsequent studies confirmed and extended the involvement of APOBEC in mutating the cancer genome to at least 16 other cancer types [3–5]. APOBEC signature mutations (C-to-T and C-to-G in TCA and TCT trinucleotide motifs) are the most prevalent in cancer after those attributable to ageing (C-to-T in CG dinucleotide motifs, most likely due to water-mediated deamination of methyl-cytosine) [4]. Furthermore, the clinical relevance of APOBEC in cancer is underscored by associations with poor patient outcomes and treatment resistance [6, 7], activation of oncogenic drivers [8–10], tumour subclonal diversification [9, 11, 12], and increased prevalence in metastases in comparison with primary tumours [13].

ntigen burden correlated with survival, but this association is not apparent here using the lower quartile as a stratification threshold. Neoantigen but not mutational burden was associated with improved survival in LUAD (Figure 3A), arguing against these factors as determinants of survival in advanced NSCLC generally. Whilst no common threshold defined high mean DAI across the tumour sites studied, our approach of defining thresholds in TCGA data and discovering confirmation in secondary cohorts supports the validity of our findings. The observation that independently described markers of immune infiltration are associated with mean DAI in both SKCM and LUAD supports the hypothesis that DAI may be a marker of peptide immunogenicity. The suggestion that mean DAI is a better survival predictor in immunotherapy-treated cohorts further supports our conclusions. Although our study lacked power to discover an immunotherapy effect, whilst high neoantigen mean DAI is not a significant survival factor in LUAD nor SKCM, there was a clear significant association in the Rizvi and Van Allen cohorts [10, 11]. Many neopeptides verified to be immunogenic have high predicted MHC affinity, and we hypothesised that mean DAI may be of specific relevance amongst this subset. The finding that mean DAI of high affinity binders correlates with survival is in keeping with the role of this factor in characterising cancer peptide immunogenicity. This effect is more pronounced in melanoma than lung cancer, suggesting the potential immunogenicity of high DAI/low affinity peptides in the latter.

t. The finding that mean DAI of high affinity binders correlates with survival is in keeping with the role of this factor in characterising cancer peptide immunogenicity. This effect is more pronounced in melanoma than lung cancer, suggesting the potential immunogenicity of high DAI/low affinity peptides in the latter. This difference, along with the finding that no common mean DAI threshold separates good from poor prognostic categories across the two cancer types studied, indicates possible context dependency of rules governing peptide immunogenicity. Whilst mean DAI may indicate intra-tumoural differences in T cell antigen recognition, immunosuppressive mechanisms may differ by tumour type and site and are likely to differentially affect the overall immunogenicity of tumour cells and infiltrating T cell effector function. Different thresholds to define clinically significant values of mean DAI may therefore reflect differential activity of regulatory pathways, making it unlikely for the same threshold to apply across tumour types, as in some cases stronger antigenic stimulation may be required to bypass tumour immunosuppression. A number of clinical trials are investigating neoantigen-based vaccines and target selection is critical to development of successful therapeutics. In one recent phase I study, predicted neoantigens with mutations occurring at MHC anchor residues were prioritised [23]; such mutations generate high DAI neopeptides. Our study provides evidence in favour of this and more direct approaches to selection of high DAI peptides in future trials.

of successful therapeutics. In one recent phase I study, predicted neoantigens with mutations occurring at MHC anchor residues were prioritised [23]; such mutations generate high DAI neopeptides. Our study provides evidence in favour of this and more direct approaches to selection of high DAI peptides in future trials. In summary, we have shown that mean DAI is associated with clinical outcome in patients with advanced melanoma and lung cancer. Mean DAI is relevant only amongst clonal mutations and correlates with immunotherapy efficacy and indicators of immune infiltration. Our findings support the notion that DAI is a relevant predictor of neopeptide immunogenicity that should be considered in ongoing attempts to refine the selection of neoantigen targets for adoptive cell transfer and vaccine studies. Supplementary Material Supplementary Figures S1-6 Click here for additional data file. Supplementary Methods and Tables Click here for additional data file. Acknowledgements We thank Pramod Srivastava for many useful discussions and review of the manuscript, and Levi Garraway, Eli Van Allen, Alexandra Snyder, Matthew Hellman, Naiyer Rizvi and Timothy Chan for kindly permitted access to study data. The results published here are in part based upon data generated by The Cancer Genome Atlas pilot project established by the NCI and the National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov/.

ults published here are in part based upon data generated by The Cancer Genome Atlas pilot project established by the NCI and the National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov/. Funding This work was supported by the Francis Crick Institute which receives its core funding from Cancer Research UK (FC001169), the UK Medical Research Council (FC001169), and the Wellcome Trust (FC001169). EG is funded by a National Institute of Health Research Fellowship and a Wellcome Trust Research Training Fellowship (no grant number applies). NM receives funding from Cancer Research UK, the Rosetrees trust and the University College London Hospitals Biomedical Research Centre (no grant number applies). KSP receives funding from the NIHR BTRU for Stem Cells and Immunotherapies (167097), of which he is the Scientific Director. S.A.Q. holds a Cancer Research U.K. (CRUK) Senior Fellowship (C36463/A22246) and is funded by a CRUK Biotherapeutic Program Grant (C36463/A20764). CS is Royal Society Napier Research Professor and acknowledges support from the Cancer Research UK (TRACERx, CRUK-UCL Centre and CRUK Cancer Immunotherapy Catalyst Network), the CRUK Lung Cancer Centre of Excellence, Stand Up 2 Cancer (SU2C), the Rosetrees Trust, NovoNordisk Foundation (ID 16584), the Prostate Cancer Foundation, the Breast Cancer Research Foundation (BCRF), the European Research Council (THESEUS) and Marie Curie Network PloidyNet (no grant numbers apply). Support was also provided to SAQ, CS and KSP by the National Institute for Health Research, the University College London Hospitals Biomedical Research Centre and the Cancer Research UK University College London Experimental Cancer Medicine Centre (no grant numbers apply).

ie Network PloidyNet (no grant numbers apply). Support was also provided to SAQ, CS and KSP by the National Institute for Health Research, the University College London Hospitals Biomedical Research Centre and the Cancer Research UK University College London Experimental Cancer Medicine Centre (no grant numbers apply). Disclosure RR, NM, JLR, KSP and SAQ report personal fees from Achilles Therapeutics, outside of the submitted work. CS reports personal fees from Janssen, Boehringer Ingelheim, Ventana, Novartis, Roche, Sequenom, Natera, Achilles Therapeutics and Sarah Cannon Research Institute, and personal fees and other from Apogen Biotechnologies, Epic Sciences and GRAIL, outside of the submitted work. All remaining authors have declared no conflicts of interest.

Key Message Preventive therapy for women at high risk of developing breast cancer can reduce disease burden. Implementing preventive therapy in routine clinical practice for high-risk women has been proved difficult due to several factors, including side-effects. Interventions should be targeted at women within the first 18 months of preventive therapy, as this is when medication withdrawal is most likely. Introduction Breast cancer is the most frequent cancer among women worldwide, with an estimated 1.67 million new cancer cases diagnosed in 2012 [1] and hence the prevention of breast cancer is a recognized priority [2]. There have been increases in female breast cancer incidence rates since the 1970s, although rates appear to be stabilizing among younger women and in more economically developed countries [3, 4]. Preventive therapy for women at high risk of developing breast cancer or those with a diagnosis of ductal carcinoma in situ (DCIS) can reduce disease burden. Aromatase inhibitors (AIs) reduce breast cancer risk among high-risk postmenopausal women. Data from the International Breast Cancer Intervention Study II (IBIS-II) prevention trial show that women randomly assigned to receive anastrozole (1 mg/day) were over 50% less likely to be diagnosed with breast cancer compared with those taking a matching placebo [5]. A 65% relative risk reduction of invasive breast cancer was shown among women taking exemestane compared with placebo in the MAP.3 trial [6].

how that women randomly assigned to receive anastrozole (1 mg/day) were over 50% less likely to be diagnosed with breast cancer compared with those taking a matching placebo [5]. A 65% relative risk reduction of invasive breast cancer was shown among women taking exemestane compared with placebo in the MAP.3 trial [6]. Two major AI prevention trials have been done among women with DCIS. In the IBIS-II DCIS trial, anastrozole was non-inferior to tamoxifen in reducing breast cancer recurrence [7]. The National Surgical Adjuvant Breast and Bowel Project (NSABP) B-35 trial showed statistically significant improvements in breast cancer-free interval among women taking anastrozole compared with tamoxifen [8]. Implementing preventive therapy in routine clinical practice for high-risk women is difficult due to reluctance among clinicians to prescribe the medication [9], low patient uptake [10–12], and sub-optimal adherence to therapy [10, 12, 13]. A major factor affecting implementation is side-effects [14, 15]. Patients are less willing to initiate preventive therapy if they perceive it to be linked with side-effects [16]. In the IBIS-I prevention trial, rates of adherence were lower among women reporting side-effects, but this finding was consistent in both tamoxifen and placebo groups [13]. Little is known about the acceptance of anastrozole in women at high risk of developing breast cancer.

eive it to be linked with side-effects [16]. In the IBIS-I prevention trial, rates of adherence were lower among women reporting side-effects, but this finding was consistent in both tamoxifen and placebo groups [13]. Little is known about the acceptance of anastrozole in women at high risk of developing breast cancer. Data from the IBIS-II prevention trial indicate women taking anastrozole are more likely to experience musculoskeletal events, vasomotor symptoms, and hypertension compared with those taking placebo [5]. In the IBIS-II DCIS trial, women taking anastrozole were more likely to experience fractures, musculoskeletal events, hypercholesterolemia, and strokes compared with women taking tamoxifen [7]. Large proportions of women reported arthralgia and hot flashes with both placebo [5] and tamoxifen [7], indicating that anastrozole may not be solely responsible for these patient reported side-effects. Here, we assess the association between participant-reported symptoms on adherence to anastrozole in the IBIS-II prevention and DCIS trials.

men reported arthralgia and hot flashes with both placebo [5] and tamoxifen [7], indicating that anastrozole may not be solely responsible for these patient reported side-effects. Here, we assess the association between participant-reported symptoms on adherence to anastrozole in the IBIS-II prevention and DCIS trials. Methods Participants The IBIS-II prevention study is an international, randomized, double-blind, and placebo-controlled trial conducted in 18 countries [5]. Postmenopausal women (n = 3864) aged 40–70 years were randomly assigned to either 1 mg anastrozole or matching placebo daily for 5 years. The trial is registered, number ISRCTN31488319. The IBIS-II DCIS study recruited 2980 postmenopausal women with locally excised estrogen receptor positive or progesterone positive DCIS. Women were randomized to receive 1 mg/day oral anastrozole or 20 mg/day oral tamoxifen for 5 years [7]. The trial is registered, number ISRCTN37546358. Details of patient cohorts and characteristics are provided in supplementary material, available at Annals of Oncology online.

r positive or progesterone positive DCIS. Women were randomized to receive 1 mg/day oral anastrozole or 20 mg/day oral tamoxifen for 5 years [7]. The trial is registered, number ISRCTN37546358. Details of patient cohorts and characteristics are provided in supplementary material, available at Annals of Oncology online. Adherence Adherence was defined as the period between trial randomization date and the date of the final follow-up visit [17]. Adherence (full/deviation/holiday/stopped) and further details on non-adherence were recorded on each follow-up CRF at 6-monthly visits. Pre-defined rules for assessing adherence were developed and used by SS to review all CRFs (supplementary material, available at Annals of Oncology online). Women who self-reported medication cessation at a visit were classified as non-adherent. Each woman was assessed for persistent use of medication for at least 4.5 years (adherent) or stopping before 4.5 years (non-adherent). All women in the prevention and DCIS IBIS-II trials have finished 5 years of active treatment.

Participant symptoms Symptoms were assessed at each follow-up visit using pre-defined items for arthralgia (arthritis, arthrosis, or joint disorder), hot flashes/night sweats, vaginal discharge, irregular vaginal bleeding, eye diseases/cataracts, and osteoporosis/fractures. Vaginal discharge and irregular vaginal bleeding were grouped together as gynecological symptoms because they are similar. All symptoms were classified as mild, moderate, or severe as judged by the women. The most severe gynecological symptom was used when computing this item. Statistical analysis Adherence to trial medication was calculated using the Kaplan–Meier method [18], both overall and by treatment group separately. Further details of statistical methods are provided in supplementary material, available at Annals of Oncology online. Results IBIS-II prevention study Postmenopausal women (n = 3864) were randomized to receive 1 mg/day anastrozole versus matching placebo. Women were excluded from the current analysis if they did not start their allocated medication (n = 77) or if they were ineligible (n = 24) (supplementary Figure S1, available at Annals of Oncology online). Hence, 3763 women (97.4%) were included in this analysis (1868 anastrozole versus 1895 placebo). For the analyses investigating associations between early reported symptoms and adherence, those who did not reach the 6 month visit were excluded (n = 159). Baseline participant characteristics were balanced between treatment groups (supplementary Table S1, available at Annals of Oncology online).

ersus 1895 placebo). For the analyses investigating associations between early reported symptoms and adherence, those who did not reach the 6 month visit were excluded (n = 159). Baseline participant characteristics were balanced between treatment groups (supplementary Table S1, available at Annals of Oncology online). Overall, 1287 women (34.2%) were non-adherent. For women randomized to anastrozole, adherence was non-significantly lower compared with those on placebo [HR = 0.97 (0.87–1.09), P = 0.6] (Figure 1). Mean time on the treatment was similar in both treatment arms (anastrozole 3.90 years versus placebo 4.00 years). Overall, annual drop-out rates were highest within the first 12–18 months of follow-up (Figure 1) and declined sharply thereafter. The following predictors were significantly associated with adherence in the univariate model: 60 years or older, hysterectomy, oophorectomy, natural menopause, and previous participation in the IBIS-I prevention trial (supplementary Table S2, available at Annals of Oncology online). When adherence was investigated adjusted for all previous significant predictors, being older than 60 years of age [OR = 1.17 (1.01–1.34), P = 0.03], not having had a hysterectomy [OR = 0.75 (0.59–0.96), P = 0.03], and previous participation in the IBIS-I trial [OR = 1.38 (1.14–1.67), P = 0.001] remained significant predictors for adherence.

stigated adjusted for all previous significant predictors, being older than 60 years of age [OR = 1.17 (1.01–1.34), P = 0.03], not having had a hysterectomy [OR = 0.75 (0.59–0.96), P = 0.03], and previous participation in the IBIS-I trial [OR = 1.38 (1.14–1.67), P = 0.001] remained significant predictors for adherence. Figure 1. Kaplan–Meier plots for non-adherence and annual non-adherence rates (%) according to treatment arm for the IBIS-II prevention (A, B) and DCIS (C, D) studies. Kaplan–Meier curves were calculated and tested for equality using log-rank test. All statistical tests were two-sided. IBIS, International Breast cancer Intervention Study; HR, hazard ratio; CI, confidence interval. At 6 months of follow-up (n = 3604), significantly more women randomized to anastrozole compared with placebo reported arthralgia (31.5% versus 25.5%, P < 0.001), hot flashes/night sweats (42.6% versus 34.1%, P < 0.001), and gynecological symptoms (11.4% versus 9.0%, P = 0.02). Women reporting arthralgia [HR = 0.85 (0.75–0.97), P = 0.01] or gynecological symptoms [HR = 0.78 (0.65–0.94), P = 0.008] were significantly less likely to be adherent at 4.5 years than those not reporting these symptoms (Table 1). However, absolute differences in adherence were small. Table 1. Early reported symptoms at 6 months associated with non-adherence in the IBIS-II prevention and DCIS study

toms [HR = 0.78 (0.65–0.94), P = 0.008] were significantly less likely to be adherent at 4.5 years than those not reporting these symptoms (Table 1). However, absolute differences in adherence were small. Table 1. Early reported symptoms at 6 months associated with non-adherence in the IBIS-II prevention and DCIS study IBIS-II prevention IBIS-II DCIS Non-adherence (%) HR (95% CI)a P-value Non-adherence (%) HR (95% CI)b P-value Arthralgia No (n=2577) 30.0 – – No (n=2069) 28.9 – – Yes (n=1027) 24.6 0.85 (0.75–0.97) 0.01 Yes (n=702) 31.2 0.90 (0.77–1.05) 0.2 Hot flashes/night sweats No (n=2224) 30.8 – – No (n=1561) 31.3 – – Yes (n=1380) 32.0 1.02 (0.90–1.15) 0.8 Yes (n=1210) 27.3 1.18 (1.02–1.36) 0.02 Gynecological No (n=3238) 30.6 – – No (n=2496) 29.4 – – Yes (n=366) 37.4 0.78 (0.65–0.94) 0.008 Yes (n=275) 30.9 0.93 (0.74–1.16) 0.5 Eye disease No (n=3424) 31.1 – – No (n=2657) 29.4 – – Yes (n=180) 34.4 0.90 (0.69–1.16) 0.4 Yes (n=114) 32.5 0.87 (0.63–1.22) 0.4 Osteoporosis No (n=3534) 31.3 – – No (n=2717) 29.5 – – Yes (n=70) 30.0 1.00 (0.65–1.55) 0.9 Yes (n=54) 31.5 0.92 (0.57–1.48) 0.7 a HR adjusted for age, hysterectomy, and previous IBIS-1 participation. b HR adjusted for HRT. IBIS, International Breast cancer Intervention Study; HR, hazard ratio; CI, confidence intervals.

e) [4]. Furthermore, the clinical relevance of APOBEC in cancer is underscored by associations with poor patient outcomes and treatment resistance [6, 7], activation of oncogenic drivers [8–10], tumour subclonal diversification [9, 11, 12], and increased prevalence in metastases in comparison with primary tumours [13]. Although involvement of APOBEC mutagenesis in cancer has only recently come to light, these enzymes have been a focus of virology research for over a decade, beginning with the near simultaneous discoveries of APOBEC3G (A3G) as an HIV-1 restriction factor and as a DNA cytosine deaminase [14, 15] (reviewed elsewhere [16, 17]). We envision that many lessons learnt regarding APOBEC within virology will be applicable to oncology. For this reason, we explore the parallels between the role of APOBEC in HIV and cancer mutagenesis. We will especially focus on how APOBEC mutagenesis can promote intratumour heterogeneity, drug resistance, and immune escape.

IBIS-II prevention IBIS-II DCIS Non-adherence (%) HR (95% CI)a P-value Non-adherence (%) HR (95% CI)b P-value Arthralgia No (n=2577) 30.0 – – No (n=2069) 28.9 – – Yes (n=1027) 24.6 0.85 (0.75–0.97) 0.01 Yes (n=702) 31.2 0.90 (0.77–1.05) 0.2 Hot flashes/night sweats No (n=2224) 30.8 – – No (n=1561) 31.3 – – Yes (n=1380) 32.0 1.02 (0.90–1.15) 0.8 Yes (n=1210) 27.3 1.18 (1.02–1.36) 0.02 Gynecological No (n=3238) 30.6 – – No (n=2496) 29.4 – – Yes (n=366) 37.4 0.78 (0.65–0.94) 0.008 Yes (n=275) 30.9 0.93 (0.74–1.16) 0.5 Eye disease No (n=3424) 31.1 – – No (n=2657) 29.4 – – Yes (n=180) 34.4 0.90 (0.69–1.16) 0.4 Yes (n=114) 32.5 0.87 (0.63–1.22) 0.4 Osteoporosis No (n=3534) 31.3 – – No (n=2717) 29.5 – – Yes (n=70) 30.0 1.00 (0.65–1.55) 0.9 Yes (n=54) 31.5 0.92 (0.57–1.48) 0.7 a HR adjusted for age, hysterectomy, and previous IBIS-1 participation. b HR adjusted for HRT. IBIS, International Breast cancer Intervention Study; HR, hazard ratio; CI, confidence intervals. For women reporting arthralgia at 6 months, only those randomized to placebo were significantly less adherent [HR = 0.81 (0.67–0.97), P = 0.02]. There was no significant difference in non-adherence for those randomized to anastrozole [HR = 0.90 (0.75–1.07), P = 0.2] (Figure 2). Women randomized to anastrozole and reporting gynecological symptoms were 31% less likely to be adherent at 4.5 years [HR = 0.69 (0.55–0.88), P = 0.003] compared with those not reporting these symptoms (Figure 2). No difference in adherence was observed for those reporting gynecological symptoms in the placebo arm [HR = 0.91 (0.69–1.20), P = 0.5]. For all other reported symptoms at 6 months, no significant differences between treatment arms were observed with regard to adherence (Figure 2). The majority of symptoms reported at 6 months among both treatment groups were of mild or moderate severity. Non-adherence was similar between those not reporting a symptom and those reporting mild symptoms at 6 months (supplementary Figure S2, available at Annals of Oncology online). We observed significant trends for non-adherence with increasing severity for all reported symptoms (P < 0.001), except for eye diseases (P = 0.8) and osteoporosis (P = 0.9) (supplementary Figure S2, available at Annals of Oncology online).

at 6 months (supplementary Figure S2, available at Annals of Oncology online). We observed significant trends for non-adherence with increasing severity for all reported symptoms (P < 0.001), except for eye diseases (P = 0.8) and osteoporosis (P = 0.9) (supplementary Figure S2, available at Annals of Oncology online). Figure 2. Forest plots for non-adherence (hazard ratios) among women reporting symptoms at 6 months by treatment arm for the IBIS-II prevention (A) and DCIS (B) studies. The squares represent the point estimates. Sizes of the squares represent the number of events. The horizontal error bars show the 95% confidence intervals (CI) of each hazard ratio. IBIS, International Breast cancer Intervention Study; CI, confidence interval. IBIS-II DCIS study Postmenopausal women (n = 2980) diagnosed with DCIS within 6 months before randomization were allocated to receive 1 mg/day anastrozole versus 20 mg/day tamoxifen. Women were excluded from the current analysis if they did not start the allocated medication (n = 32) or if they were ineligible (n = 18). This left 2930 women (98.3%) for the adherence analysis (1486 tamoxifen versus 1444 anastrozole) (supplementary Figure S1, available at Annals of Oncology online). For the analysis of the association of early reported symptoms and adherence an additional 159 women who did not reach the 6 month follow-up point were excluded (Figure 1). Baseline demographics were balanced between treatment groups (supplementary Table S1, available at Annals of Oncology online).

of Oncology online). For the analysis of the association of early reported symptoms and adherence an additional 159 women who did not reach the 6 month follow-up point were excluded (Figure 1). Baseline demographics were balanced between treatment groups (supplementary Table S1, available at Annals of Oncology online). Overall, non-adherence was 33.3% and non-significantly different between anastrozole and tamoxifen [HR = 1.06 (0.94–1.20), P = 0.4] (Figure 1). Mean time on study was also similar between treatment arms (anastrozole: 3.99 years versus tamoxifen: 3.95 years). As with the prevention study, rates of non-adherence were greatest within the first 12–18 months and decrease thereafter (Figure 1). In the univariate analysis, previous HRT use [OR = 0.79 (0.67–0.92), P = 0.003], hysterectomy [OR = 0.83 (0.70–0.98), P = 0.03], and oophorectomy [OR = 0.75 (0.58–0.97), P = 0.03] were significantly associated with decreased likelihood of adherence. Women who had a natural menopause [OR = 1.25 (1.04–1.50), P = 0.02] were significantly more likely to be adherent than their counterparts (supplementary Table S1, available at Annals of Oncology online). In the multivariate analysis, only previous HRT use remained a significant predictor of non-adherence [OR = 0.81 (0.69–0.95), P = 0.009] in women with DCIS.

1.25 (1.04–1.50), P = 0.02] were significantly more likely to be adherent than their counterparts (supplementary Table S1, available at Annals of Oncology online). In the multivariate analysis, only previous HRT use remained a significant predictor of non-adherence [OR = 0.81 (0.69–0.95), P = 0.009] in women with DCIS. A total of 2770 women were included to investigate early reported symptoms and adherence. At 6 months, significantly more women randomized to anastrozole reported arthralgia compared with tamoxifen (30.4% versus 20.3%, P < 0.001). In contrast, significantly more women randomized to tamoxifen reported hot flashes/night sweats (40.6% versus 46.7%, P = 0.001) and gynecological symptoms (7.0% versus 12.8%, P < 0.001). Women reporting hot flashes were significantly more adherent than those not reporting these symptoms [HR = 1.18 (1.02–1.36), P = 0.02] (Table 1). Those on anastrozole reporting these symptoms were significantly more adherent than their counterparts [HR = 1.23 (1.00–1.52), P = 0.05], and a non-significant increase in non-adherence was also observed for those on tamoxifen (Figure 2). For all other symptoms, no association with adherence was observed either overall or by treatment arm (Table 1 and Figure 2). Non-adherence was similar between those not reporting a symptom and those reporting mild symptoms at 6 months (supplementary Figure S2, available at Annals of Oncology online). We observed significant trends for non-adherence with increasing severity for arthralgia (P < 0.001), hot flashes (P = 0.009), and gynecological symptoms (P = 0.03). There were no significant trends for any other symptoms (supplementary Figure S2, available at Annals of Oncology online).

at Annals of Oncology online). We observed significant trends for non-adherence with increasing severity for arthralgia (P < 0.001), hot flashes (P = 0.009), and gynecological symptoms (P = 0.03). There were no significant trends for any other symptoms (supplementary Figure S2, available at Annals of Oncology online). Discussion In the IBIS-II prevention and DCIS trials, over one-third of women were non-adherent for the full course of therapy. There were no overall significant differences in study drop-outs by treatment arm for either study. Arthralgia and gynecological symptoms at any severity significantly reduced the likelihood of adherence in the placebo and anastrozole groups of the prevention trial, respectively. In the DCIS trial, hot flashes were associated with a higher likelihood of adherence. Associations between symptoms and non-adherence strengthened with increasing severity. Participant-reported early symptoms may be partially responsible for non-adherence to anastrozole preventive therapy, however, other factors are likely to play an important role.

were associated with a higher likelihood of adherence. Associations between symptoms and non-adherence strengthened with increasing severity. Participant-reported early symptoms may be partially responsible for non-adherence to anastrozole preventive therapy, however, other factors are likely to play an important role. Similarities in non-adherence between the treatment groups suggest that some factors associated with non-adherence are likely to be unrelated to anastrozole. Similar non-adherence rates between treatment arms were also reported in the MAP.3 prevention trial (32.8% exemestane versus 28.7% placebo) [19] and the NSABP-B35 trial (35.6% anastrozole versus 35.8% tamoxifen) [8]. Identifying strategies to reduce the burden of moderate and severe symptoms could be one approach increase medication adherence. However, identifying modifiable factors other than medication induced side-effects that explain non-adherence could improve behavioral interventions. Our previous systematic review of uptake and adherence to breast cancer chemoprevention suggests women’s perceived risk of breast cancer is likely to play a role [10].

e. However, identifying modifiable factors other than medication induced side-effects that explain non-adherence could improve behavioral interventions. Our previous systematic review of uptake and adherence to breast cancer chemoprevention suggests women’s perceived risk of breast cancer is likely to play a role [10]. Non-adherence in the IBIS-I tamoxifen prevention trial showed higher levels of study drop-outs in the first 12–18 months of therapy [13] and a finding has been reported in the adjuvant setting [20, 21]. We observed the same pattern in both the IBIS-II prevention and DCIS trials. This consistency highlights that this time period may be the most appropriate point to deliver interventions supporting adherence. Identifying the optimal timing of interventions is important, but there is a paucity of strategies shown to effectively improve adherence to endocrine therapy. Identifying modifiable determinants of adherence to preventive therapy should be prioritized, so they can be incorporated into strategies to improve medication taking behavior.

optimal timing of interventions is important, but there is a paucity of strategies shown to effectively improve adherence to endocrine therapy. Identifying modifiable determinants of adherence to preventive therapy should be prioritized, so they can be incorporated into strategies to improve medication taking behavior. Participant-reported symptoms do not completely explain non-adherence to preventive therapy. Arthralgia, hot flashes/night sweats, and gynecological symptoms were more common among women taking anastrozole in the IBIS-II prevention trial. Fewer women taking anastrozole in the IBIS-II DCIS trial reported hot flashes/night sweats or gynecological symptoms, however arthralgia was more common compared with the tamoxifen treatment arm. The NSABP-B35 observed similar trends [22], but demonstrated no significant differences in quality of life between women taking anastrozole and tamoxifen. Strategies to manage these symptoms are required to ensure quality of life is not affected among women taking preventive therapy.

d with the tamoxifen treatment arm. The NSABP-B35 observed similar trends [22], but demonstrated no significant differences in quality of life between women taking anastrozole and tamoxifen. Strategies to manage these symptoms are required to ensure quality of life is not affected among women taking preventive therapy. This study has strengths and limitations. We are among the first to provide a detailed report of the relationship between symptoms and non-adherence among women taking anastrozole for preventive therapy. The data derive from two large international randomized studies that were carefully monitored throughout. However, because these data were from motivated participants willing to enroll in a clinical trial, we may have over-estimated the proportion of women who are able to complete the full course of therapy. In addition, we were not able to investigate concurrent medication associated with symptoms relieve, which may contribute to better adherence. There is no gold standard measure of medication adherence, our reported outcome was recorded during clinic visits and may be an inflated estimate [23]. Quality of life assessments that encapsulate the full impact of symptoms on everyday life may be more closely associated with adherence [24]. We noted a weaker than expected relationships between symptoms reported at 6 months and non-adherence. One possibility is that late-onset symptoms could be responsible for subsequent trial drop-out. However, this is unlikely to account for a large amount of drop-outs as the majority of treatment induced side-effects occur in the first year of therapy [25]. Our 6-month assessment may also be identifying symptoms that are transient in nature. Women experiencing symptoms that persist for longer or become more severe may be more likely to drop-out, and this would not be captured by our analysis.

sion that many lessons learnt regarding APOBEC within virology will be applicable to oncology. For this reason, we explore the parallels between the role of APOBEC in HIV and cancer mutagenesis. We will especially focus on how APOBEC mutagenesis can promote intratumour heterogeneity, drug resistance, and immune escape. The AID/APOBEC superfamily: a diverse set of cytosine deaminase enzymes implicated in cancer APOBEC3 belongs to the AID/APOBEC superfamily, consisting of activation induced deaminase (AID), APOBEC1 (A1), APOBEC2 (A2), APOBEC3A-H (A3A, A3B, A3C, A3D, A3F, A3G, and A3H), and APOBEC4 (A4). AID deaminates cytosines at the immunoglobulin locus, enabling antibody gene diversification via somatic hypermutation, and class switch recombination [18]. A1 was identified originally as an RNA editing enzyme, deaminating apolipoprotein B mRNA at a specific position to create an early stop codon [19], but it also has robust DNA deamination activity [14, 20]. The functions of A2 and A4 are still unclear and these proteins have yet to show enzymatic activity.

ority of treatment induced side-effects occur in the first year of therapy [25]. Our 6-month assessment may also be identifying symptoms that are transient in nature. Women experiencing symptoms that persist for longer or become more severe may be more likely to drop-out, and this would not be captured by our analysis. In conclusion, non-adherence with anastrozole was moderate in both the IBIS-II prevention and DCIS trials. Only women reporting moderate or severe symptoms were less likely to be adherent. Identifying factors other than medication induced side-effects that explain non-adherence could help to target future intervention strategies to support medication taking behavior. Once interventions have been developed, they should be targeted at women within the first 18 months of therapy, as this is when medication cessation is most likely. Supplementary Material Supplementary Figure S1 Click here for additional data file. Supplementary Figure S2 Click here for additional data file. Supplementary Table S1 Click here for additional data file. Supplementary Table S2 Click here for additional data file. Supplementary Material Click here for additional data file. Acknowledgements We would like to thank the women who participated in the IBIS-II prevention and DCIS trials, and the nurses and clinicians in the local centers for their continuing support. Funding This work was supported by Cancer Research UK (C569/A5032 to JC, C42785/A17965 to SGS); and AstraZeneca (no grant number applies). AstraZeneca provided anastrozole, tamoxifen, and matching placebo.

Acknowledgements We would like to thank the women who participated in the IBIS-II prevention and DCIS trials, and the nurses and clinicians in the local centers for their continuing support. Funding This work was supported by Cancer Research UK (C569/A5032 to JC, C42785/A17965 to SGS); and AstraZeneca (no grant number applies). AstraZeneca provided anastrozole, tamoxifen, and matching placebo. Disclosure JC received grant funding from AstraZeneca. All remaining authors have declared no conflicts of interest.

Introduction Genomic changes that characterize primary breast cancer (BC) have been elucidated by extensive genomic profiling studies [1]. Comparative analyses of estrogen receptor-positive (ER+) metastatic BC (mBC) have demonstrated genomic evolution during metastatic progression, and following treatment, such as the enrichment of HER2 and ESR1 genomic alterations (GAs) [2, 3]. Clonal evolution can arise due to independent primary lesions, expansion of subclonal populations, or acquisition of novel GAs. Genomic changes following therapy are exemplified by acquired activating ESR1 GAs that mediate aromatase inhibitor (AI) resistance [2–4]. Clonal evolution processes highlight the importance of profiling a contemporaneous sample that is representative of disease progression to guide further management. However, limitations in performing repeated prospective biopsies of metastatic lesions over the disease course for a patient can present challenges for clinical genomic analysis [5]. Liquid biopsy and sequencing of circulating tumor DNA (ctDNA) from blood could provide a complementary approach to tissue-based genomic testing for mBC.

tions in performing repeated prospective biopsies of metastatic lesions over the disease course for a patient can present challenges for clinical genomic analysis [5]. Liquid biopsy and sequencing of circulating tumor DNA (ctDNA) from blood could provide a complementary approach to tissue-based genomic testing for mBC. Research studies of BC identified genomic changes in ctDNA following therapy, however, limited numbers of ER+ BC have been profiled [6–8]. In larger studies of ctDNA from ER+ mBC, droplet digital PCR (ddPCR) identified select mutations in ESR1 or PIK3CA [9–11]. In phase 3 trials for ER+/HER2-negative (HER2−) BC, prospective ctDNA assessment identified patients with PIK3CA mutation who derived survival benefit from buparlisib [4]. Retrospective analyses of ctDNA in phase 3 trials suggest that ESR1 mutations are associated with resistance to AI but not selective ER down-regulators (SERDs), and can guide therapy selection [11]. In this study, we carried out hybrid capture-based genomic profiling to characterize GAs in ctDNA from 254 patients with ER+ BC during the course of their clinical care.

Research studies of BC identified genomic changes in ctDNA following therapy, however, limited numbers of ER+ BC have been profiled [6–8]. In larger studies of ctDNA from ER+ mBC, droplet digital PCR (ddPCR) identified select mutations in ESR1 or PIK3CA [9–11]. In phase 3 trials for ER+/HER2-negative (HER2−) BC, prospective ctDNA assessment identified patients with PIK3CA mutation who derived survival benefit from buparlisib [4]. Retrospective analyses of ctDNA in phase 3 trials suggest that ESR1 mutations are associated with resistance to AI but not selective ER down-regulators (SERDs), and can guide therapy selection [11]. In this study, we carried out hybrid capture-based genomic profiling to characterize GAs in ctDNA from 254 patients with ER+ BC during the course of their clinical care. Methods Detailed descriptions of patient samples/methods are presented in supplementary methods, available at Annals of Oncology online. Briefly, peripheral blood samples were collected from 254 patients with ER+ BC, plasma was isolated from 20 ml whole blood, ≥20 ng DNA was extracted, and hybrid capture-based genomic profiling of ctDNA was carried out in a CLIA-certified/CAP-accredited laboratory [Foundation Medicine (FM)] to identify substitutions, short insertions/deletions, rearrangements/fusions, and amplifications [12]. Sixty-two genes (supplementary Table S1, available at Annals of Oncology online) were sequenced (Illumina HiSeq 2500 or 4000) to a median unique coverage depth of 7503×. Maximum somatic allele frequency (MSAF) was used to estimate the ctDNA fraction in plasma.

deletions, rearrangements/fusions, and amplifications [12]. Sixty-two genes (supplementary Table S1, available at Annals of Oncology online) were sequenced (Illumina HiSeq 2500 or 4000) to a median unique coverage depth of 7503×. Maximum somatic allele frequency (MSAF) was used to estimate the ctDNA fraction in plasma. Results Patient characteristics This study of hybrid capture-based sequencing of ctDNA in blood included consecutive genomic profiling results from 254 female patients with an initial diagnosis of ER+ BC, determined by routine IHC. Patient characteristics are described in Table 1 and supplementary Table S2, available at Annals of Oncology online. For patients with available clinical information, 94% were stage IV, 88% had received prior chemotherapy, and 88% had received prior AI in the adjuvant and/or metastatic setting. Table 1. Patient characteristics

characteristics are described in Table 1 and supplementary Table S2, available at Annals of Oncology online. For patients with available clinical information, 94% were stage IV, 88% had received prior chemotherapy, and 88% had received prior AI in the adjuvant and/or metastatic setting. Table 1. Patient characteristics All ER +  ER+/HER2− ER+/HER2+ ER+/HER2 unk N 254 197 28 29 Median age, years (range) 58 (32–85) 58 (33–85) 57 (33–79) 62 (32–78) Stage, N (%) I 2 (0.8%) 2 (1.1%) 0 (0%) 0 (0%) II 5 (2.1%) 3 (1.6%) 1 (3.7%) 1 (3.8%) III 7 (2.9%) 7 (3.7%) 0 (0%) 0 (0%) IV 226 (94.2%) 175 (93.6%) 26 (96.3%) 25 (96.2%) Unknown 14 10 1 3 Previous chemotherapya Yes 120 (88.2%) 94 (87.0%) 18 (90.0%) 8 (100%) [adj/met/unk], N [24/91/5] [17/72/5] [4/14/0] [3/5/0] No 16 (11.8%) 14 (13.0%) 2 (10.0%) 0 (0%) Unknown 118 89 8 21 Previous aromatase inhibitora Yes 115 (88.5%) 95 (92.2%) 15 (75.0%) 5 (71.4%) [adj/met/unk], N [19/91/5] [15/75/5] [2/13/0] [2/3/0] No 15 (11.5%) 8 (7.8%) 5 (25.0%) 2 (28.6%) Unknown 124 94 8 22 Previous tamoxifena Yes 56 (43.8%) 41 (40.6%) 8 (40.0%) 7 (100%) [adj/met/unk], N [41/13/2] [28/11/2] [7/1/0] [6/1/0] No 72 (56.2%) 60 (59.4%) 12 (60.0%) 0 (0%) Unknown 126 96 8 22 Previous fulvestranta Yes 69 (54.3%) 57 (56.4%) 7 (36.8%) 5 (71.4%) No 58 (45.7%) 44 (43.6%) 12 (63.2%) 2 (28.6%) Unknown 127 96 9 22 a See supplementary Table S2, available at Annals of Oncology online, for detailed descriptions of treatments in adjuvant/metastatic settings, and for treatment/response status at the time of sample collection.

s 69 (54.3%) 57 (56.4%) 7 (36.8%) 5 (71.4%) No 58 (45.7%) 44 (43.6%) 12 (63.2%) 2 (28.6%) Unknown 127 96 9 22 a See supplementary Table S2, available at Annals of Oncology online, for detailed descriptions of treatments in adjuvant/metastatic settings, and for treatment/response status at the time of sample collection. adj, adjuvant only; met, metastatic or metastatic and adjuvant; unk, unknown. GAs identified in ctDNA from ER+ BC At least 1 GA was detected in 78% of cases with an average of 2.5 GA/sample (range 0–27). MSAF was calculated for each case and provided a median estimated ctDNA fraction of 1.7% (interquartile range 0.3%–9.2%). Eighty-four percent of cases have evidence of ctDNA in the blood (MSAF > 0). There was no evidence of ctDNA in the blood (MSAF = 0) in 15% (33/226) of stage IV cases and 43% (6/14) of stage I–III cases. The most frequently altered genes in ER+ BC were TP53 (38%), ESR1 (31%), PIK3CA (31%), CDH1 (10%), and ERBB2 (8%) (supplementary Figure S1, available at Annals of Oncology online).

d (MSAF > 0). There was no evidence of ctDNA in the blood (MSAF = 0) in 15% (33/226) of stage IV cases and 43% (6/14) of stage I–III cases. The most frequently altered genes in ER+ BC were TP53 (38%), ESR1 (31%), PIK3CA (31%), CDH1 (10%), and ERBB2 (8%) (supplementary Figure S1, available at Annals of Oncology online). For ER+/HER2− BC, the most frequently altered genes were TP53 (35%), ESR1 (34%), PIK3CA (31%), and CDH1 (12%) (Figure 1A). ERBB2 activating mutations were identified in 3% of cases. ERBB2 amplification was identified in one patient initially diagnosed with ER+/HER2− BC (IHC on a breast biopsy); gain of HER2 was subsequently confirmed by IHC (3+) on a metastatic biopsy. Activating kinase fusions (FGFR2-INA, FGFR3-TACC3, NCOA4-RET) were observed in three cases (2%). Frequently altered pathways included PI3K-AKT-mTOR (38%), RAS-RAF-MEK (15%), FGFR (14%), cell cycle (8%), and BRCA (6%).

(IHC on a breast biopsy); gain of HER2 was subsequently confirmed by IHC (3+) on a metastatic biopsy. Activating kinase fusions (FGFR2-INA, FGFR3-TACC3, NCOA4-RET) were observed in three cases (2%). Frequently altered pathways included PI3K-AKT-mTOR (38%), RAS-RAF-MEK (15%), FGFR (14%), cell cycle (8%), and BRCA (6%). Figure 1. Genomic alterations in ctDNA from patients with ER+ breast cancer and comparisons with tissue. (A) GAs identified in 197 cases of ER+/HER2− BC. Percent of cases altered is indicated. (B) GAs identified in 28 cases of ER+/HER2+ BC. (C) Comparison of the most frequently mutated (top panel) or amplified (bottom panel) genes observed in ctDNA in this study (N = 254) with tissue-based genomic profiling of ER+ BC. Datasets from ER+ BC tissue used for comparison were from the FM database (N = 851) and published studies of tissue from early BC (eBC, TCGA [1]: N = 594) and mBC (Lefebvre et al. [3]: N = 143; Fumagalli et al. [2]: N = 182). Data from [1, 3] were extracted from cBioPortal. Black dots represent genes that were not assessed in [2]. (D) Concordance between GAs found in ctDNA and matched tumor tissue from 14 patients. Days between ctDNA and tissue collection are shown. The ctDNA fraction was estimated using MSAF. Concordant/shared GAs are in blue, GAs found in tissue only are in grey, and GAs found in ctDNA only are in red. For samples with multiple unique mutations in a gene (patient-5 and patient-8), the number of mutations is shown.

s between ctDNA and tissue collection are shown. The ctDNA fraction was estimated using MSAF. Concordant/shared GAs are in blue, GAs found in tissue only are in grey, and GAs found in ctDNA only are in red. For samples with multiple unique mutations in a gene (patient-5 and patient-8), the number of mutations is shown. For ER+/HER2+ BC, the most frequently altered genes were TP53 (61%), ERBB2 (36%), PIK3CA (25%), and ESR1 (25%) (Figure 1B). ERBB2 amplification was observed in 29% (8/28) of HER2+ cases, consistent with next-generation sequencing (NGS) studies of ctDNA in HER2+ BC [7, 13]. The estimated ctDNA fraction was significantly higher for HER2+ cases with ERBB2 amplification compared with cases without ERBB2 amplification detected (supplementary Figure S2A, available at Annals of Oncology online), suggesting that the ability to detect ERBB2 amplification was associated with the quantity of ctDNA in the blood.

ctDNA fraction was significantly higher for HER2+ cases with ERBB2 amplification compared with cases without ERBB2 amplification detected (supplementary Figure S2A, available at Annals of Oncology online), suggesting that the ability to detect ERBB2 amplification was associated with the quantity of ctDNA in the blood. Comparison of GAs in ctDNA and tissue We compared frequently altered genes in ctDNA with ER+ BC tissue samples from the FM database and published studies [1–3] (Figure 1C). For the majority of genes, mutation frequencies in ctDNA were similar to the range observed in tissue; ESR1 was mutated at a higher frequency (greater than twofold) compared with tissue, as expected from our study population comprising mostly patients who had received or were receiving AI treatment (supplementary Table S2, available at Annals of Oncology online) [9]. Amplifications were observed at lower frequencies in ctDNA compared with tissue, consistent with other studies of amplification detection in ctDNA from BC [7, 13]. The estimated ctDNA fraction was significantly higher for cases with an amplification detected compared with cases without (supplementary Figure S2B, available at Annals of Oncology online).

wer frequencies in ctDNA compared with tissue, consistent with other studies of amplification detection in ctDNA from BC [7, 13]. The estimated ctDNA fraction was significantly higher for cases with an amplification detected compared with cases without (supplementary Figure S2B, available at Annals of Oncology online). Genomic profiles of matched blood and tissue samples collected within 60 days of each other were available for 14 cases. We compared GAs assessed in both ctDNA and tissue (Figure 1D; supplementary Table S3, available at Annals of Oncology online). For short variant mutations, 89% (17/19) that were detected in tissue were also detected in ctDNA. Six mutations were detected in ctDNA only and two mutations were in tissue only. One ctDNA only ESR1 mutation (patient-5) was found in a second tissue sample from a distinct metastatic site, collected 408 days before blood sampling. One case (patient-10) harbored one shared and one tissue only ESR1 mutation; the allele frequency (AF) for the shared mutation (AF = 34%) was 10-fold higher compared with the tissue only mutation (AF = 3%). Two cases harbored both shared and ctDNA only mutations for the same gene (patient-8: ESR1; patient-12: PIK3CA): the shared mutation had a higher AF than the ctDNA only mutations in both cases (ESR1: twofold; PIK3CA: threefold), suggesting that ctDNA only mutations occur in less represented clones that may not be detected in a single tumor biopsy, consistent with clonal heterogeneity. For amplifications, 27% (3/11) that were detected in tissue were also detected in ctDNA; no amplifications were detected in ctDNA only. The estimated ctDNA fraction was higher for two cases where at least one amplification was detected in both tissue and ctDNA than for cases where amplification was detected in tissue only.

amplifications, 27% (3/11) that were detected in tissue were also detected in ctDNA; no amplifications were detected in ctDNA only. The estimated ctDNA fraction was higher for two cases where at least one amplification was detected in both tissue and ctDNA than for cases where amplification was detected in tissue only. Landscape of ESR1 alterations in ctDNA A total of 131 ESR1 GA were observed in 80 ER+ cases (supplementary Figure S3A, available at Annals of Oncology online), including both ER+/HER2− and ER+/HER2+ cases (Figure 1A and B); whereas, only 1 ESR1 GA (amplification) was observed in the ctDNA of 74 ER- cases (P = 0.0001, Fisher’s exact test, two-tailed, supplementary Figure S3B, available at Annals of Oncology online). For the 130 ER+ patients with available clinical information regarding AI treatment, 35% (40/115) of all AI-treated patients had an ESR1 GA, and consistent with previous studies [9], ESR1 GAs were more frequent in patients treated with AI in the metastatic setting (40%, 36/91) versus patients treated with adjuvant AI only (11%, 2/19); all patients (40/40) with ESR1 GA had received prior AI (supplementary Table S2, available at Annals of Oncology online).

A, and consistent with previous studies [9], ESR1 GAs were more frequent in patients treated with AI in the metastatic setting (40%, 36/91) versus patients treated with adjuvant AI only (11%, 2/19); all patients (40/40) with ESR1 GA had received prior AI (supplementary Table S2, available at Annals of Oncology online). The most frequent ESR1 GAs were D538G, Y537S, Y537N, and E380Q. All observed ESR1 mutations are activating or occur at the L536 position where multiple activating mutations have been characterized (Figure 2A). Of the 80 ESR1-altered cases, 40% had >1 ESR1 GA (range 2–4) (Figure 2B). In comparison, 24% (19/79) of PIK3CA-altered cases harbored >1 PIK3CA GA (range 2–8) and 23% (22/96) of TP53-altered cases harbored >1 TP53 GA (range 2–11). In cases with >1 ESR1 mutation, no one ESR1 mutation had a consistently greater AF than co-occurring ESR1 mutations, suggesting that diverse ESR1 mutations could contribute to AI resistance (supplementary Figure S4, available at Annals of Oncology online).

(22/96) of TP53-altered cases harbored >1 TP53 GA (range 2–11). In cases with >1 ESR1 mutation, no one ESR1 mutation had a consistently greater AF than co-occurring ESR1 mutations, suggesting that diverse ESR1 mutations could contribute to AI resistance (supplementary Figure S4, available at Annals of Oncology online). Figure 2. Landscape of ESR1 alterations in ctDNA. (A) Graph represents the number of cases with each ESR1 GA. The percent indicate the frequency of each ESR1 GA relative to all 131 ESR1 GA identified. AMP, amplification; RE, rearrangement. (B) Percent of cases with 1, 2, 3, or 4 ESR1 GAs. (C) Compound mutations in ESR1: in case-1 all instances of L536F and D538G were observed on the same read in cis. In case-2 ESR1 c.1609T>A (Y537N) was observed alone in the majority of reads, and ESR1 c.1609T>A and c.1610A>G were seen together as compound mutations in a portion of reads to generate ESR1 Y537S. (D) ESR1 variants of unknown significance identified in this study. (E) ESR1 rearrangements identified in this study. Numbered boxes represent exons and numbers above indicate amino acid position. Rearrangement break points in ESR1 were in exon 4 or exon 5. (F) Assessment of GAs that co-occur with ESR1 GA: frequency of ESR1-altered cases (N = 80) with alterations in the genes indicated. ‘Multiple’ represents cases harboring multiple classes of genomic alteration.

ination [18]. A1 was identified originally as an RNA editing enzyme, deaminating apolipoprotein B mRNA at a specific position to create an early stop codon [19], but it also has robust DNA deamination activity [14, 20]. The functions of A2 and A4 are still unclear and these proteins have yet to show enzymatic activity. In general, the A3 family members are considered part of the innate immune system, forming overlapping barriers to virus and transposon replication. Consistent with such a physiological function, A3 genes show profound copy number and amino acid variation in mammals. For instance, most humans have seven A3 genes arranged in tandem, whereas rodents have only one at the same genomic location [21, 22], and each A3 gene in humans as well as several other mammals manifests high levels of amino acid variation due to positive selection [23].

s and numbers above indicate amino acid position. Rearrangement break points in ESR1 were in exon 4 or exon 5. (F) Assessment of GAs that co-occur with ESR1 GA: frequency of ESR1-altered cases (N = 80) with alterations in the genes indicated. ‘Multiple’ represents cases harboring multiple classes of genomic alteration. Multiple ESR1 GAs in the same sample are thought to be polyclonal in origin [9, 10]. We carried out a pairwise assessment of all co-occurring ESR1 mutations to determine whether any mutation pairs existed as compound mutations on the same allele (supplementary Table S4, available at Annals of Oncology online). Out of 67 mutation pairs, 49 were close enough to be evaluated on the same sequencing read; compound mutations were observed in 2/49 mutation pairs (Figure 2C). In case-1, ESR1 L536F/D538G were observed as compound mutations on all reads. In case-2, ESR1 Y537N occurred as a single mutation in most reads, but a subset of reads harbored a compound mutation at the Y537 codon that resulted in conversion of Y537N to Y537S; the existence of two subsets of reads suggests sequential mutational events.

536F/D538G were observed as compound mutations on all reads. In case-2, ESR1 Y537N occurred as a single mutation in most reads, but a subset of reads harbored a compound mutation at the Y537 codon that resulted in conversion of Y537N to Y537S; the existence of two subsets of reads suggests sequential mutational events. In addition to the GAs described above, 21 ESR1 variant of unknown significance (VUS) mutations were identified (Figure 2D) in 15 ER+ cases including 5 cases with no ESR1 GAs (supplementary Figure S3A, available at Annals of Oncology online); no ESR1 VUS was observed in 74 ER- cases. To evaluate compound mutations, we analyzed 51 co-occurring ESR1 mutation pairs that involved an ESR1 VUS, and 20 could be evaluated on the same sequencing read. Compound mutations were observed in 4/20 mutation pairs and 2 mutation pairs existed as compound mutations in only a subset of sequencing reads (supplementary Figure S5, available at Annals of Oncology online). ESR1 rearrangements were observed in three cases: two ESR1 rearrangements had potential 3’ fusion partners (AKAP12, NKAIN2) and one ESR1 rearrangement was fused to an intergenic region (Figure 2E). ESR1-AKAP12 is recurrent in BC and all three ESR1 rearrangements resulted in loss of the ligand-binding domain (LBD), which likely results in constitutive ER activation [14, 15]. Each ESR1-rearranged case harbored concurrent ESR1 mutation, suggesting prior AI exposure: we confirmed prior adjuvant AI and fulvestrant treatment of the patient with ESR1 fused to intergenic space.

ements resulted in loss of the ligand-binding domain (LBD), which likely results in constitutive ER activation [14, 15]. Each ESR1-rearranged case harbored concurrent ESR1 mutation, suggesting prior AI exposure: we confirmed prior adjuvant AI and fulvestrant treatment of the patient with ESR1 fused to intergenic space. Co-occurring GAs with ESR1 To inform therapeutic strategies for AI refractory patients, we evaluated co-occurring alterations with ESR1 GAs and identified concurrent GAs that have been associated with responses to targeted therapy in BC [4] including PIK3CA (35%), FGFR1 (16%), ERBB2 (8%), BRCA1/2 (5%), and AKT1 (4%) (Figure 2F). For cases with concurrent PIK3CA/ESR1 mutation, the PIK3CA:ESR1 AF ratio was ≥1 for 75% (21/28) of cases, consistent with PIK3CA being a truncal driver and ESR1 arising following AI (supplementary Figure S6, available at Annals of Oncology online). Concurrent ESR1/ERBB2 mutation was more frequent in ctDNA than tissue: in ctDNA, 4% (3/79) of ESR1-mutated cases had concurrent ERBB2 mutation; whereas, in the FM database, 0.6% (6/969) of ESR1-mutated BC tissue samples had concurrent ERBB2 mutation. Discussion Genomic profiling of ctDNA has the potential to capture GAs that drive recurrent disease or therapeutic resistance and may provide an alternative when tissue biopsy is challenging. However, genomic profiles of ctDNA from ER+ mBC have not been extensively studied. We describe GAs identified in ctDNA from the blood of 254 patients with ER+ mBC.

NA has the potential to capture GAs that drive recurrent disease or therapeutic resistance and may provide an alternative when tissue biopsy is challenging. However, genomic profiles of ctDNA from ER+ mBC have not been extensively studied. We describe GAs identified in ctDNA from the blood of 254 patients with ER+ mBC. Eighty-four percent of samples had evidence of ctDNA in the blood, consistent with a study of ctDNA release in mBC [8]. For cases with no evidence of ctDNA in the blood, lack of detectable somatic alterations is, in part, likely associated with insufficient ctDNA release into the blood at the time point of sampling that can be affected by clinical parameters including disease stage, tumor size, number of metastatic sites, albumin level, and number of lines of treatment [8, 16]; these parameters were variable in this study of unselected cases (supplementary Table S2, available at Annals of Oncology online).

time point of sampling that can be affected by clinical parameters including disease stage, tumor size, number of metastatic sites, albumin level, and number of lines of treatment [8, 16]; these parameters were variable in this study of unselected cases (supplementary Table S2, available at Annals of Oncology online). Alterations were identified in genes that have been associated with responses to targeted therapy in ER+ BC (PIK3CA, ESR1, ERBB2, FGFR1, BRCA1/2, AKT1) [4]. Compared with genomic studies of ER+ BC tissue biopsies, we identified similar mutation frequencies in ctDNA [1–3]. Tumor burden can be monitored by longitudinal assessment of variant AFs in ctDNA [8]. However, genomic profiling of large numbers of genes is best-suited for guiding therapy selection, but may be cost-prohibitive for serial testing. Instead, genomic profiling could establish GAs present in ctDNA at baseline for a patient, and guide design of personalized serial monitoring assays. GAs reported here could inform prioritization of genes to include for limited sequencing panels for longitudinal disease monitoring of ER+ mBC.

bitive for serial testing. Instead, genomic profiling could establish GAs present in ctDNA at baseline for a patient, and guide design of personalized serial monitoring assays. GAs reported here could inform prioritization of genes to include for limited sequencing panels for longitudinal disease monitoring of ER+ mBC. For a smaller subset of patients with temporally matched ctDNA and tissue samples, 89% of short variant mutations that were detected in tissue were also detected in ctDNA. Additional ESR1, TP53, and PIK3CA mutations were identified only in ctDNA; other studies have similarly observed additional mutations for each of these genes in ctDNA compared with matched tissue [5, 10, 17]. Additional mutations in ctDNA could reflect the utility of liquid biopsy to capture heterogeneity of metastatic sites in ER+ mBC [6]. Consistent with this idea, for paired cases with both shared and ctDNA only mutations in one gene, the shared mutation AF was two to threefold higher than the ctDNA only mutation AF. This hypothesis warrants confirmation in prospective trials and may be more relevant in clinical settings where targeted therapies are routinely employed.

his idea, for paired cases with both shared and ctDNA only mutations in one gene, the shared mutation AF was two to threefold higher than the ctDNA only mutation AF. This hypothesis warrants confirmation in prospective trials and may be more relevant in clinical settings where targeted therapies are routinely employed. In this study, genomic profiling was carried out as part of routine clinical care for unselected patients, including patients with low tumor burden; therefore, many samples had a low estimated ctDNA fraction (supplementary Table S2, available at Annals of Oncology online). The sensitivity for amplification detection in ctDNA was 27% for the 14 matched ctDNA-tissue pairs. Amplifications (including CCND1, FGFR1, ERBB2) were detected in ctDNA at lower frequencies than tissue; specifically ERBB2 amplification was identified in 29% of HER2+ cases. Detection of amplifications was associated with higher estimated ctDNA fraction. Our findings are consistent with NGS studies [using NGS panels or alternative approaches for amplification detection such as low coverage whole genome sequencing (plasma-Seq)] that highlighted the limitations for robust detection of amplifications in the context of low ctDNA fractions [18, 19]. Other studies identified similarly low frequencies of ERBB2 amplification (21%–32%) in ctDNA from HER2+ BC and detected other common BC amplifications (including CCND1, FGFR1) at significantly lower frequencies in ctDNA compared with matched tissue [7, 13]. Therefore, although amplifications may be detected in a subset of cases with sufficient ctDNA fraction, tissue-based genomic testing may be a more reliable method of detection. In BC, ERBB2 amplification remains the only established clinically utilized copy number biomarker, but amplifications including FGFR1 and 11q13 are being evaluated as biomarkers in trials [4]; tissue-based testing may be the preferable method for treatment selection based on copy number biomarkers.

method of detection. In BC, ERBB2 amplification remains the only established clinically utilized copy number biomarker, but amplifications including FGFR1 and 11q13 are being evaluated as biomarkers in trials [4]; tissue-based testing may be the preferable method for treatment selection based on copy number biomarkers. We observed a high frequency of ESR1 GAs that are associated with AI resistance, as expected for this patient population of mostly mBC with prior AI treatment [9]. The ESR1 mutation frequency reported here is consistent with studies of AI-treated, ER+ mBC that used ddPCR to assess selected ESR1 mutations in ctDNA, including frequencies reported in several phase 3 trials [4, 9–11]. We observed a similar distribution of ESR1 mutations compared with a study of common ESR1 mutations in ctDNA using ddPCR [10]. Consistent with other studies, we frequently observed cases harboring >1 ESR1 GA [9, 10]. Multiple ESR1 mutations are thought to reflect convergent evolution of distinct clones during AI resistance [6, 10]; for the few cases evaluated using dual mutation-specific ddPCR probes, different ESR1 mutations existed on separate alleles [17, 20]. We confirmed that most ESR1 mutation pairs occur on distinct sequencing reads, likely reflecting polyclonal origin; however, we also identified cases with ESR1 compound mutations on the same allele. Studies of ESR1 have focused on single mutations; ESR1 compound mutations in cis warrant further study, and such mutations might display differential therapeutic sensitivities compared with characterized single mutations.

yclonal origin; however, we also identified cases with ESR1 compound mutations on the same allele. Studies of ESR1 have focused on single mutations; ESR1 compound mutations in cis warrant further study, and such mutations might display differential therapeutic sensitivities compared with characterized single mutations. Diverse ESR1 alterations were observed, including rearrangements with break points resulting in loss of the LBD. Similar ESR1 rearrangements with variable 3’ fusion partners have been described and are activating [14, 15]. ESR1 rearrangements demonstrate preclinical resistance to AI and SERDs, therefore, detection of ESR1 rearrangements may be important for therapy selection [15]. ESR1 VUSs reported here could represent novel functional mutations that warrant characterization. We identify co-occurring alterations with ESR1, which could represent alternative targets or rational targets for combination therapy with SERDs [4]. Some of these GAs have been successfully targeted in the context of co-occurring ESR1 GA: responses have been observed for patients with concurrent ESR1/AKT1 mutation (to AZD5363) [21], ESR1/PIK3CA mutation (to alpelisib) [22], and for a patient from this study with ESR1/BRCA2 mutation (to olaparib; Dr S. Blau, personal communication).

have been successfully targeted in the context of co-occurring ESR1 GA: responses have been observed for patients with concurrent ESR1/AKT1 mutation (to AZD5363) [21], ESR1/PIK3CA mutation (to alpelisib) [22], and for a patient from this study with ESR1/BRCA2 mutation (to olaparib; Dr S. Blau, personal communication). Concurrent ESR1/ERBB2 activating mutations occurred in ER+/HER2− BC and were more frequently observed in ctDNA compared with tissue, suggesting that ERBB2 and ESR1 mutations may commonly reside on distinct clones that may not be detected in a single tissue biopsy; ESR1 mutations were also observed in 25% of ER+/HER2+ BC. Combinations of SERDs with HER2-targeted therapy could be relevant for such cases. Here, we demonstrate the clinical implementation of genomic profiling of ctDNA from patients with ER+ BC and identify clinically relevant GAs. Blood-based testing may provide an alternative or complementary approach to tissue-based genomic testing for patients with mBC. Supplementary Material mdx490_supplementary_figure_s1 Click here for additional data file. mdx490_supplementary_fig_s2 Click here for additional data file. mdx490_supplementary_figure_s3-s6 Click here for additional data file. mdx490_supplementary_table_s1_s3_s4 Click here for additional data file. mdx490_supplementary_table_s2 Click here for additional data file. mdx490_supplementary_methods_corrected Click here for additional data file. Acknowledgement The authors would like to recognize the important contributions from other researchers whose work could not be cited due to space constraints. Funding None declared.

opy number and amino acid variation in mammals. For instance, most humans have seven A3 genes arranged in tandem, whereas rodents have only one at the same genomic location [21, 22], and each A3 gene in humans as well as several other mammals manifests high levels of amino acid variation due to positive selection [23]. A3G has been studied intensely in the field of virology, as it was recognized early on to deaminate cytosines in cDNA reverse transcription intermediates of retroviruses including HIV-1 [24, 25]. Reverse transcriptase places an adenine opposite to the newly created uracil nucleobase, introducing a viral genomic strand G→A mutation [26]. This inhibits HIV replication by directly rendering the viral genome dysfunctional or by indirectly triggering viral cDNA degradation by subsequent uracil DNA glycosylase activity and endonuclease digestion [26, 27]. A3G can also directly bind to HIV-1 genomic RNA and interfere with viral cDNA synthesis [28]. A3D, A3F, and A3H also contribute to HIV-1 mutagenesis through similar mechanisms, and it is generally accepted that different subsets of A3 family members restrict the replication of different classes of viruses and transposons (reviewed elsewhere [16, 17]).

-1 genomic RNA and interfere with viral cDNA synthesis [28]. A3D, A3F, and A3H also contribute to HIV-1 mutagenesis through similar mechanisms, and it is generally accepted that different subsets of A3 family members restrict the replication of different classes of viruses and transposons (reviewed elsewhere [16, 17]). Adding to the complexity of seven A3 family members in humans, different subsets of A3 enzymes are expressed in different tissue types [29, 30]. Together with high levels of DNA sequence similarity (near perfect identity in many regions), determining which of these enzymes is responsible for mutagenesis of different cancer genomes has been challenging. Thus far, the greatest numbers of publications support A3B and A3A (reviewed more extensively elsewhere [31–33]). However, recent reports indicate that A3H is also important and, together with A3B, may account for the entire APOBEC mutation signature observed in breast and lung cancers [34, 35].

s has been challenging. Thus far, the greatest numbers of publications support A3B and A3A (reviewed more extensively elsewhere [31–33]). However, recent reports indicate that A3H is also important and, together with A3B, may account for the entire APOBEC mutation signature observed in breast and lung cancers [34, 35]. APOBEC mutagenesis increases subclonal diversity APOBEC mutagenesis occurs independently within single cancer cells and viruses, often resulting in strand-coordinated hypermutations (sometimes referred to as kataegis [1]). Evidence for A3B upregulation has been found in over half of all cancers [3] (reviewed elsewhere [31]). Additionally, our groups and others have identified APOBEC activity as contributing to branched evolution and the acquisition of subclonal mutations later in the evolutionary course of lung adenocarcinoma, estrogen receptor (ER)-negative breast cancer, head and neck squamous cell carcinoma, and esophageal adenocarcinomas [11, 35, 36]. A recent analysis of the intratumour heterogeneity present in 100 TRACERx patients with untreated surgically resected primary non-small-cell lung cancer revealed a significant correlation between frequencies of APOBEC signature mutation and the overall number of subclonal mutations [9]. Furthermore, in 19 patients, subclonal driver events were detected as occurring in the APOBEC mutational context [9] (Figure 1A). Moreover, tumours with the largest number of subclonal mutations had extensive evidence of APOBEC mutational signatures, suggesting that APOBEC activity is a strong mutagenic force late in cancer evolution (Figure 1B and C). Subclonal mutations generated from APOBEC activity could potentially drive cancer evolution by enabling the acquisition of late driver mutations. Across multiple types of cancer, there is evidence that APOBEC mutagenesis is responsible for creating driver mutations [5, 8, 9, 12, 31]. The most striking examples are two helical domain hot spot mutations in PIK3CA in papillomavirus-positive head and neck squamous cell carcinomas (E542K and E545K) [8].

er mutations. Across multiple types of cancer, there is evidence that APOBEC mutagenesis is responsible for creating driver mutations [5, 8, 9, 12, 31]. The most striking examples are two helical domain hot spot mutations in PIK3CA in papillomavirus-positive head and neck squamous cell carcinomas (E542K and E545K) [8]. Figure 1. Apolipoprotein B mRNA-editing enzyme, catalytic polypeptide-like (APOBEC) mutagenesis within the TRACERx 100 cohort. (A) Phylogenetic trees of TRACERx patients harbouring a subclonal driver mutation in an APOBEC preferred motif are shown. The mutation is indicated near the clone in which it occurs. Clonal clusters are shown in blue, subclonal clusters are shown in red, and subclonal clusters containing the driver mutation in a preferred APOBEC motif are shown in orange. (B) The total numbers of mutations in each patient are shown, with mutations in an APOBEC context shown in orange and all other mutational contexts shown in green. (C) The total number of driver mutations in each patient are shown, with mutations in an APOBEC context shown in orange and all other mutational contexts shown in green. (D) The fraction of subclonal mutations for each patient that could be attributed to APOBEC activity are shown, with mutations in an APOBEC context shown in orange and all other mutational contexts shown in green.

own, with mutations in an APOBEC context shown in orange and all other mutational contexts shown in green. (D) The fraction of subclonal mutations for each patient that could be attributed to APOBEC activity are shown, with mutations in an APOBEC context shown in orange and all other mutational contexts shown in green. APOBEC expression has also been reported to impact responses to cancer therapy. In ER-positive breast cancer, A3B mRNA expression levels inversely correlated with clinical benefit to tamoxifen, and A3B overexpression correlated with tamoxifen resistance in mouse xenograft models, implicating the enzyme in promoting drug resistance [6]. Similarly, an enrichment of A3B mutations was observed in chemotherapy-resistant urothelial carcinomas [37].

ion levels inversely correlated with clinical benefit to tamoxifen, and A3B overexpression correlated with tamoxifen resistance in mouse xenograft models, implicating the enzyme in promoting drug resistance [6]. Similarly, an enrichment of A3B mutations was observed in chemotherapy-resistant urothelial carcinomas [37]. APOBEC-induced mutagenesis is also prevalent in HIV-1 proviral DNA sequences [38], with some reports estimating the percentage of new mutations attributable to APOBEC activity to be as high as 98% [39]. However, other studies have estimated lower contributions for APOBEC in HIV-1 genetic variation, and definitive work has yet to be done to dissociate the contributions of APOBEC from mistakes made by the non-proofreading viral reverse transcriptase [38, 40]. Nevertheless, APOBEC activity has been linked to early diversification in newly infected individuals as the transmitted founder virus adapts in response to the immune pressure exerted within its new host [41]. Thus, APOBEC mutagenesis has the capability to increase subclonal diversity in both cancer and HIV, with the potential for a profound impact on the subsequent evolution of both the tumour and the virus (Figure 2).

the transmitted founder virus adapts in response to the immune pressure exerted within its new host [41]. Thus, APOBEC mutagenesis has the capability to increase subclonal diversity in both cancer and HIV, with the potential for a profound impact on the subsequent evolution of both the tumour and the virus (Figure 2). The fact that APOBEC increases diversity is likely to be important for the clinical course of both diseases, since genetic heterogeneity is a substrate for Darwinian evolution and is likely to confound treatments. Indeed, in some cancers, increased intratumour heterogeneity has been shown to correlate with a shorter progression free survival [42, 43]. Additionally, measures of genetic heterogeneity have been associated with poor progression free survival in cancer [44] and intermediate thresholds of copy number instability correlate with poorest clinical outcomes [31], as they presumably allow for diversity without exceeding toxic levels of genomic instability (Figure 2). Paralleling findings in cancer, among patients with HIV exhibiting rapid disease progression, there are fewer hypermutated viral sequences as well as fewer minimally edited sequences, suggesting that suboptimal APOBEC activity may enhance genetic diversity, allowing an increase in pathogenicity [39] (Figure 2). From a clinical standpoint, this suggests that successful and complete inhibition of APOBEC enzymatic activity may provide a way to limit subclone diversification and potentially enable normal adaptive immune responses to control or clear HIV-1 infection [45]. Conceivably, complete inhibition of APOBEC enzymatic activity may limit tumour evolution and boost the efficacy of a variety of targeted or immune therapies that often fail due to the acquisition of escape variants brought about by resistance mutations and/or subclonal neoantigens.

ontrol or clear HIV-1 infection [45]. Conceivably, complete inhibition of APOBEC enzymatic activity may limit tumour evolution and boost the efficacy of a variety of targeted or immune therapies that often fail due to the acquisition of escape variants brought about by resistance mutations and/or subclonal neoantigens. Figure 2. Parallels between APOBEC mutagenesis in HIV and cancer. (A) Within HIV, APOBEC mutagenesis is counteracted through Vif. As APOBEC mutagenesis increases, the chance of lethal mutagenesis and population variation increases. The trade-off between lethal mutagenesis and population variation creates an optimal range in which APOBEC mutagenesis increases population fitness. (B) Within cancer, the toxic effects of APOBEC mutagenesis are counteracted through DNA damage repair and DNA damage tolerance. As APOBEC mutagenesis increases, the chance of lethal mutagenesis and population variation increases. The trade-off between lethal mutagenesis and population variation creates an optimal range in which APOBEC mutagenesis increases population fitness.

are counteracted through DNA damage repair and DNA damage tolerance. As APOBEC mutagenesis increases, the chance of lethal mutagenesis and population variation increases. The trade-off between lethal mutagenesis and population variation creates an optimal range in which APOBEC mutagenesis increases population fitness. Additional lessons from HIV: APOBEC mutagenesis fuels drug resistance and immune escape In order for HIV and cancer to replicate effectively in human hosts, these pathogenic entities must constantly avoid detection and destruction by innate and adaptive immune responses. As a consequence of this predator–prey relationship, lentiviruses and cancer have evolved elegant mechanisms to avoid or actively counteract immune responses. One of these overlapping mechanisms between HIV and cancer appears to entail co-opting APOBEC mutagenesis as a means of immune evasion.

daptive immune responses. As a consequence of this predator–prey relationship, lentiviruses and cancer have evolved elegant mechanisms to avoid or actively counteract immune responses. One of these overlapping mechanisms between HIV and cancer appears to entail co-opting APOBEC mutagenesis as a means of immune evasion. Lentiviruses and host A3 genes have co-evolved for millions of years in non-human primate relatives [46] and more recently in humans [47]. HIV-1, as well as other lentiviruses, has adapted to suppress A3-mediated innate anti-viral activity through utilizing the viral infectivity factor (Vif) as a defensive measure [15, 25]. Vif counteracts A3 functionality by polyubiquitination and proteasomal degradation [48]. Additionally, Vif has been shown to affect A3G translation through mRNA binding [49] and A3G gene transcription through heterodimerization with the transcription co-factor CBF-β [50]. These discoveries indicate that targeting the Vif-APOBEC3 interaction may be a novel avenue to combat HIV infection. However, over a decade of virology research into the A3-mediated restriction mechanism has demonstrated that interventions in this pathway may be a double-edged sword. Several studies have found that the range of APOBEC mutagenesis within HIV virions can vary largely. Lethal HIV mutagenesis will be selected against in vivo, and moderate APOBEC mutagenesis appears to induce sublethal variation that fuels viral heterogeneity and immune escape [41, 51–53] (Figure 3). Similar to cancer [54], viral genetic heterogeneity together with an appropriate selective pressure can enable the emergence of resistant populations. Sublethal G→A mutations in the DNA sequence motif preferred by A3G have been observed in drug-resistant HIV variants [55–58] and immune escape variants [38, 41, 51, 53] (Figure 2). Mutations in the APOBEC motif are enriched in cytotoxic T lymphocyte epitopes of HIV, and can result in diminished CD8+T cell responses against previously antigenic epitopes, suggestive of selection as a consequence of immune escape [38] (Figure 4). Finally, within cancer there is early evidence that APOBEC correlates with the overexpression of the immune checkpoint molecule PD-L1, potentially contributing to the development of immune exhaustion [59] (Figure 4). Indeed, early clinical trial data suggest that reversing immune exhaustion with an anti-PD-L1 antibody enhances the immune response against HIV in a subset of participants [60].

the overexpression of the immune checkpoint molecule PD-L1, potentially contributing to the development of immune exhaustion [59] (Figure 4). Indeed, early clinical trial data suggest that reversing immune exhaustion with an anti-PD-L1 antibody enhances the immune response against HIV in a subset of participants [60]. Figure 3. Potential roles of APOBEC supporting the escape and progression of HIV and cancer. (A) Sublethal APOBEC mutagenesis promotes the formation of drug escape and immune escape variants in HIV that will be selected upon exposure to treatment or the immune system. (B) Sublethal APOBEC mutagenesis promotes the formation of drug escape and immune escape variants in cancer that will be selected upon exposure to treatment or the immune system. APOBEC mutagenesis also underlies the formation of driver gene mutations and potentially also replication stress-induced genomic instability, although the latter is speculative. Figure 4. APOBEC mutagenesis promotes HIV immune escape and potentially also that of cancer. APOBEC mutations have been identified in cytotoxic T lymphocyte epitopes of HIV. APOBEC hypermutation has been linked to PD-L1 ligand overexpression and potentially contributes to immune exhaustion of tumour infiltrating lymphocytes. Studies within the HIV literature are suggestive of sublethal APOBEC mutagenesis driving the expansion of escape variants.

d in cytotoxic T lymphocyte epitopes of HIV. APOBEC hypermutation has been linked to PD-L1 ligand overexpression and potentially contributes to immune exhaustion of tumour infiltrating lymphocytes. Studies within the HIV literature are suggestive of sublethal APOBEC mutagenesis driving the expansion of escape variants. Some of the evolutionary dynamics that have been identified in HIV in response to therapy resistance also appear to apply to cancer. For example, in HIV-1, the V3 loop of gp120 interacts with the host cellular coreceptor CCR5 in order to gain entry to the cell. CCR5 antagonists bind CCR5 and prevent the entry of CCR5-tropic HIV-1 [61]. A common route of CCR5 antagonist resistance is the emergence of HIV-1 variants using CXCR4 instead of CCR5 as a coreceptor for cellular entry [62, 63]. Independent studies detected minor CCR5 antagonist-resistant variants containing resistance mutations in the APOBEC context at baseline that were rapidly selected through therapeutic selective pressure [55, 56]. Complementary to these in vivo studies, in vitro experiments point towards APOBEC mutagenesis generating subclonal resistance mutations, which are under positive selection during drug exposure [58]. Similarly, in cancer, minor drug resistant variants are rapidly selected upon treatment (reviewed in Schmitt et al. [64]), although the contribution of APOBEC in this process remains to be quantified.

EC mutagenesis generating subclonal resistance mutations, which are under positive selection during drug exposure [58]. Similarly, in cancer, minor drug resistant variants are rapidly selected upon treatment (reviewed in Schmitt et al. [64]), although the contribution of APOBEC in this process remains to be quantified. An optimal range of APOBEC mutagenesis may exist for both HIV and tumour evolution (Figure 2). It has been shown that several Vif mutants are less potent in inhibiting A3G, with Vif-K22H incapable of fully neutralizing A3G and therefore appearing to enable sufficient levels of sublethal A3G mutagenesis for the emergence of antiretroviral resistance [57]. Thus, it appears that suboptimal A3G mutagenesis is less effective in inactivating the viral genome and in contrast may promote drug resistance and immune escape [38, 45, 58]. There are also parallels in the context of APOBEC mutagenesis of the cellular genome. Extensive A3B and A3A mutagenesis induces a DNA damage response (DDR) and at excessive levels it is toxic to the cell [2, 65–69]. Therefore, similar to HIV, cancer cells may have to attenuate APOBEC mutagenesis, enhance repair mechanisms, and/or dampen the DDR pathways to help ensure optimal cell survival (Figure 2). For instance, the loss of p53 enables DNA damage tolerance to A3B-mediated mutagenesis [65] and correlates positively with A3B overexpression in breast cancer [2]. Furthermore, a recent report focusing on Y-family polymerases and PrimPol, a translesion synthesis polymerase with re-priming properties, found that PrimPol as well as POLK and POLI may serve to limit the detrimental effects of APOBEC mutagenesis [70].

correlates positively with A3B overexpression in breast cancer [2]. Furthermore, a recent report focusing on Y-family polymerases and PrimPol, a translesion synthesis polymerase with re-priming properties, found that PrimPol as well as POLK and POLI may serve to limit the detrimental effects of APOBEC mutagenesis [70]. In addition to sublethal APOBEC mutagenesis driving tumour heterogeneity, APOBEC has recently also been shown to produce replication stress in cancer [65, 71, 72]. This complements our previous observation that replication stress itself can drive A3B expression and activity [73], which suggests a positive feedforward loop involving replication stress and APOBEC mutagenesis (Figure 5). We have previously implicated replication stress in inducing a widespread endogenous DDR as a biological barrier to tumour progression [74–76], as well as driving chromosome segregation errors and ensuing chromosomal instability (CIN) [76, 77]. It is therefore conceivable that oncogene-induced and APOBEC-induced replication stress itself is a driver of CIN in cancer. Furthermore, APOBEC may provide a cancer cell the necessary genomic plasticity to evade immune surveillance. Firstly, a large subclonal neoantigen burden has been associated with a poor response to immunotherapy [78] and conceivably APOBEC-induced subclonal neoantigens may contribute to this process. Although the mechanism of how subclonal neoantigens potentially confound the response to immunotherapy is still unclear, it is conceivable that subclonal neoantigens foster the outgrowth of T cells that are reactive towards subclonal neoantigens that outcompete T cells reactive towards clonal neoantigens. This may result in the outgrowth of T cells that are only reactive to a proportion of the tumour population, diminishing chances of total tumour control. Secondly, ongoing chromosome missegregations (i.e. CIN) may enable the dynamic loss of immunogenic clonal neoantigens [79]. In effect, CIN-mediated loss of clonal neoantigens (provided these mutations are not required for tumour fitness) as well as the acquisition of subclonal neoantigens (due to APOBEC activity) could both aid immune evasion. Interestingly, highly aneuploid (i.e. an abnormal state of chromosomal copy number) tumours have less immune infiltration and are associated with a worse patient survival than less aneuploid tumours [80, 81].

ss) as well as the acquisition of subclonal neoantigens (due to APOBEC activity) could both aid immune evasion. Interestingly, highly aneuploid (i.e. an abnormal state of chromosomal copy number) tumours have less immune infiltration and are associated with a worse patient survival than less aneuploid tumours [80, 81]. It is yet to be quantified to which extent APOBEC-induced replication stress and potential ensuing CIN contributes to the process of immune evasion. Figure 5. A feedforward replication stress-APOBEC loop potentially drives subclonal somatic point mutations and copy number variations. APOBEC induces subclonal somatic point mutations and has recently been shown to induce replication stress. APOBEC-mediated replication stress could potentially contribute to CIN. Therapeutic considerations By analogy to hypermutation and hypomutation within HIV [45], there are at least two general strategies for targeting APOBEC in cancer. The first strategy is therapy by hypermutation, by enhancing the mutagenic effects of APOBEC to the point where cancer cells suffer catastrophic levels of DNA damage and selectively die. Indeed, recent studies have suggested that DDR inhibitors, such as PARP and ATR inhibitors, may sensitize tumour cells with high levels of APOBEC to an APOBEC-dependent death [65, 71, 72].

ncing the mutagenic effects of APOBEC to the point where cancer cells suffer catastrophic levels of DNA damage and selectively die. Indeed, recent studies have suggested that DDR inhibitors, such as PARP and ATR inhibitors, may sensitize tumour cells with high levels of APOBEC to an APOBEC-dependent death [65, 71, 72]. The second strategy is therapy by hypomutation, by inhibiting APOBEC-dependent tumour evolution and potentially suppressing adverse outcomes including recurrence, metastasis, and drug resistance. Constraining cancer evolvability may be accomplished with drugs to inhibit APOBEC gene expression [73, 82] or with chemical inhibitors of DNA deaminase activity (for exemplary studies on A3G see [83–85]). Proof of principle has been achieved with a genetic knockdown of A3B causing an improvement in the durability of tamoxifen treatment of ER+ xenograft tumours in mice [6].

th drugs to inhibit APOBEC gene expression [73, 82] or with chemical inhibitors of DNA deaminase activity (for exemplary studies on A3G see [83–85]). Proof of principle has been achieved with a genetic knockdown of A3B causing an improvement in the durability of tamoxifen treatment of ER+ xenograft tumours in mice [6]. Besides harnessing APOBEC through a cancer cell intrinsic mechanism, a complementary strategy may be provoking immune responses to cancer cells with high levels of APOBEC-induced neoantigens. It is important to consider whether neoantigens are present in the trunk of the tumour (clonal) or in the branches (subclonal), since clonal neoantigens may improve response to immune checkpoint blockade in contrast to subclonal neoantigens [78]. Interestingly, one APOBEC-mediated mutational signature (signature 2) is primarily found in the branches of multiple different cancer types [12]. In contrast, the other APOBEC-mediated mutational signature (signature 13) is primarily found in the trunk of bladder cancer [12]. This could potentially explain paradoxical effects of APOBEC mutagenesis on immune surveillance and patient outcome. For example, in breast cancer, A3B expression is reported to worsen overall patient survival [6, 7]. In contrast, APOBEC mutagenesis within bladder cancer has been correlated with an improved patient outcome [86, 87], although the influence of APOBEC on survival in bladder cancer is still a matter of debate [88]. Therefore, it is conceivable that patients with tumours containing extensive clonal APOBEC mutagenesis, in contrast to subclonal APOBEC mutagenesis, are more suitable for immune checkpoint blockade.

atient outcome [86, 87], although the influence of APOBEC on survival in bladder cancer is still a matter of debate [88]. Therefore, it is conceivable that patients with tumours containing extensive clonal APOBEC mutagenesis, in contrast to subclonal APOBEC mutagenesis, are more suitable for immune checkpoint blockade. It is also important to note that several APOBEC family members including A3G and A3H are highly expressed in immune cells, including tumour infiltrating T cells, and correlated with improved outcomes [89, 90]. Thus, as studies advance, it will be crucial to not only consider APOBEC expression and mutation signature in the tumour itself, but also APOBEC expression in the larger tumour microenvironment. Immunohistochemistry approaches would be ideal to help address these relationships, but specific monoclonal antibodies for each human APOBEC family member have been challenging to develop due to extensive protein similarity.

gnature in the tumour itself, but also APOBEC expression in the larger tumour microenvironment. Immunohistochemistry approaches would be ideal to help address these relationships, but specific monoclonal antibodies for each human APOBEC family member have been challenging to develop due to extensive protein similarity. Discussion Conclusions Despite the numerous studies that have detected APOBEC-associated mutations by sequencing clinical cancer samples, it is still not fully clear why APOBEC mutagenesis is such a widespread and recurring mutational signature. In this review, we have postulated functions of APOBEC in cancer through exploring the past virology research focused on HIV and APOBEC. Within the field of virology, sublethal APOBEC mutagenesis of HIV virions has been linked to increasing HIV diversity and the creation of drug and immune escape variants. Conceivably, similar functions for APOBEC mutagenesis may be operating, and be selected for, during cancer evolution with clear opportunities for the development of novel therapeutic interventions. Funding This work was partially funded from Cancer Research UK, Rosetrees and the University College London Hospitals Biomedical Research Centre (NM); the Danish Cancer Society, Swedish Research Council, NovoNordisk Foundation (ID 16584), Danish National Research Foundation (project CARD) and Danish Council for Independent Research (JB);

Funding This work was partially funded from Cancer Research UK, Rosetrees and the University College London Hospitals Biomedical Research Centre (NM); the Danish Cancer Society, Swedish Research Council, NovoNordisk Foundation (ID 16584), Danish National Research Foundation (project CARD) and Danish Council for Independent Research (JB); IAVI with the generous support of USAID and other donors (JH) (a full list of IAVI donors is available at www.iavi.org); grants from the National Institutes of Health (NIAID R37 AI064046 and NCI R21 CA206309 (RSH). This work was supported by Cancer Research UK (TRACERx), the Rosetrees Trust, NovoNordisk Foundation (ID 16584), EU FP7 (projects PREDICT and RESPONSIFY, ID number 259303), the Prostate Cancer Foundation, the Breast Cancer Research Foundation, the European Research Council (THESEUS), and National Institute for Health Research University College London Hospitals Biomedical Research Centre (CS). The contents of this manuscript are the responsibility of the authors and do not necessarily reflect the views of USAID or the US Government. RSH is the Margaret Harvey Schering Land Grant Chair for Cancer Research, a Distinguished McKnight University Professor, and an Investigator of the Howard Hughes Medical Institute. This work was supported by the Francis Crick Institute which receives its core funding from Cancer Research UK (FC001169), the UK Medical Research Council (FC001169), and the Wellcome Trust (FC001169). Disclosure RSH is a co-founder, consultant, and shareholder of ApoGen Biotechnologies Inc.

This work was supported by Cancer Research UK (TRACERx), the Rosetrees Trust, NovoNordisk Foundation (ID 16584), EU FP7 (projects PREDICT and RESPONSIFY, ID number 259303), the Prostate Cancer Foundation, the Breast Cancer Research Foundation, the European Research Council (THESEUS), and National Institute for Health Research University College London Hospitals Biomedical Research Centre (CS). The contents of this manuscript are the responsibility of the authors and do not necessarily reflect the views of USAID or the US Government. RSH is the Margaret Harvey Schering Land Grant Chair for Cancer Research, a Distinguished McKnight University Professor, and an Investigator of the Howard Hughes Medical Institute. This work was supported by the Francis Crick Institute which receives its core funding from Cancer Research UK (FC001169), the UK Medical Research Council (FC001169), and the Wellcome Trust (FC001169). Disclosure RSH is a co-founder, consultant, and shareholder of ApoGen Biotechnologies Inc. CS is a member of the advisory board of ApoGen Biotechnologies Inc. All remaining authors have declared no conflicts of interest.

Cyclin-dependent kinase (CDK) 4/6 inhibition has been demonstrated to improve progression-free survival (PFS) in patients with human epidermal growth factor receptor 2 (HER2−), hormone receptor positive (HR+) in advanced breast cancer [1–3]. Palbociclib, ribociclib and abemaciclib are orally bioavailable selective CDK 4/6 inhibitors. These small molecules likely bind the ATP-binding pocket within the CDK4/6 protein kinases thereby inhibiting phosphorylation of retinoblastoma tumour suppressor protein (Rb). In its hypophosphorylated state Rb remains bound to E2F thereby preventing progression through the G1-S-cell cycle checkpoint [4]. The mechanism behind the observed efficacy of CDK inhibition in metastatic breast cancer may relate to a dependence of HR+ breast cancer on CDK4/6 activity to override Rb mediated repression of cell cycle progression (Figure 1) [5].

bound to E2F thereby preventing progression through the G1-S-cell cycle checkpoint [4]. The mechanism behind the observed efficacy of CDK inhibition in metastatic breast cancer may relate to a dependence of HR+ breast cancer on CDK4/6 activity to override Rb mediated repression of cell cycle progression (Figure 1) [5]. Figure 1. Cell cycle progression through E2F regulation, and the role of CDK and estrogen (ER) inhibitors. Transcriptional activation of cyclin-D1 (CCND1) through the estrogen receptor (ESR1), promotes dimerization of CCND1 and CDK4, and CCND1 and CDK6, escaping inhibition by p16. The cyclin-D/CDK complex phosphorylates Rb, releasing E2F to promote cell cycle progression through transcriptional activation of S-phase and G2/M gene sets. Additional transcriptional activation through E2F induction may affect genes involved in DNA methylation and PD-L1 expression. Pharmacological inhibition of ER and CDK4/6 synergistically affects downstream activation of E2F and inhibits cell cycle progression in the context of wild-type Rb. Mutational inactivation of Rb promotes therapeutic resistance.

ion through E2F induction may affect genes involved in DNA methylation and PD-L1 expression. Pharmacological inhibition of ER and CDK4/6 synergistically affects downstream activation of E2F and inhibits cell cycle progression in the context of wild-type Rb. Mutational inactivation of Rb promotes therapeutic resistance. CDK4/6 inhibitors have been approved by the US Food and Drug Administration (FDA) for initial endocrine therapy in postmenopausal women with metastatic or advanced HR+/HER2− breast cancer in combination with an aromatase inhibitor and for the treatment of endocrine therapy-resistant HR+/HER2− advanced or metastatic breast cancer in combination with Fulvesterant (a selective estrogen receptor degrader) [6]. In December 2017 the National Institute for Health and Care Excellence (NICE) has recommended CDK4/6 inhibitors in combination with aromatase inhibition as a first-line option for treating locally advanced or metastatic HR+/HER2− breast cancer [7]. Despite the success of the clinical studies that led to these recommendations, not all patients with HR+ breast cancer respond to CDK inhibition and a significant fraction progress within 2 years of initiation of treatment [1–3]. This underscores the need to identify mechanism of resistance to these targeted therapies to anticipate and target novel or subclonal resistance mechanisms driving breast cancer progression in these patients.

respond to CDK inhibition and a significant fraction progress within 2 years of initiation of treatment [1–3]. This underscores the need to identify mechanism of resistance to these targeted therapies to anticipate and target novel or subclonal resistance mechanisms driving breast cancer progression in these patients. Circulating tumour DNA (ctDNA) describes molecules of cell-free DNA circulating in plasma that originate from a patient’s tumour. ctDNA analyses by next-generation sequencing are demonstrating translational utility within clinical contexts ranging from non-invasive screening [8], tracking cancer burden and identifying residual disease in patients undergoing treatment of their disease [9–11] and identifying cancer associated mutations with therapeutic implications [12, 13]. In this edition of Annals of Oncology Condorelli et al. [14] leverage the ability of ctDNA analysis to interrogate the mutational landscape of progressive metastatic cancer to highlight loss of Rb function as a potential resistance mechanism to CDK4/6 inhibition. They provide a case-series of three patients treated at different institutions, by separate investigators, who developed progressive metastatic breast cancer following treatment with CDK4/6 inhibitors. In each case evidence of somatic alteration involving the RB1 gene was noted through plasma ctDNA analyses at the point of disease progression. In the first patient a frameshift event involving exon 8 of RB1 was observed that was predicted to result in a non-functioning truncated version of the protein. This event was not observed through NGS analysis of a liver biopsy acquired before CDK4/6 inhibition. In the second patient of the case-series four RB1 alterations were noted at progression on palbociclib that were not detectable before initiation of therapy. The variant with the highest allele frequency in plasma at progression (Chr13(GRCh37): g.48937094G>A) has been previously shown in lung cancer to result in loss of the Rb protein region responsible for the binding of Rb to E2F-transcription factor complexes [15]. The final patient was observed to have a p.His483Tyr RB1 variant following ribociclib that is predicted to be deleterious.

sion (Chr13(GRCh37): g.48937094G>A) has been previously shown in lung cancer to result in loss of the Rb protein region responsible for the binding of Rb to E2F-transcription factor complexes [15]. The final patient was observed to have a p.His483Tyr RB1 variant following ribociclib that is predicted to be deleterious. This study is of interest for the following reasons. Firstly, it provides observational evidence of deleterious RB1 alterations potentially being selected at disease progression following intervention with CDK4/6 inhibitors in patients with metastatic breast cancer. These observations build on a previous invivo investigation of CDK4/6 inhibitor resistance using patient-derived tumour xenograft models that suggested Rb1 inactivation as a resistance mechanism to chronic CDK4/6 inhibition [16]. Secondly, this study provides an early glimpse into the potential of ctDNA panels to detect acquisition of actionable alterations in patients who experience disease progression on anticancer therapy. Such a resource could inform mechanisms underlying resistance across a range of systemic therapies. There are advantages to ctDNA analyses as a research tool to understand the biology of heavily treated metastatic disease. The non-invasive nature of ctDNA examination overcomes barriers to tissue acquisition in late stage disease that include poor patient health, increased risk from biopsy procedures and cost.

rapies. There are advantages to ctDNA analyses as a research tool to understand the biology of heavily treated metastatic disease. The non-invasive nature of ctDNA examination overcomes barriers to tissue acquisition in late stage disease that include poor patient health, increased risk from biopsy procedures and cost. There are however caveats to consider regarding this case-series. The number of patients described within the manuscript is small and there is no indication as to the frequency by which Rb1 alterations are detected at progression on CDK4/6 inhibition in this patient population. Additionally, patients 1 and 3 in the case-series were treated with two lines of therapy in between the biopsies showing lack of RB1 alterations and ctDNA analyses demonstrating acquired RB1 alterations—patient 1 received everolimus and exemstane before palbociclib and patient 2 received capecitabine and paclitaxel following ribociclib. Therefore, we cannot be certain that the acquisition of Rb1 alterations solely associate with selective pressure induced by CDK4/6 inhibition. Advancing the findings reported in this case-series will require a larger cohort to determine the incidence of Rb1 alterations as resistance mechanisms in patients with metastatic breast cancer on CDK4/6 inhibitors. Furthermore, more frequent ctDNA monitoring is necessary to follow the dynamics by which RB1 alterations emerge and ascertain the association of their emergence with disease progression.

o determine the incidence of Rb1 alterations as resistance mechanisms in patients with metastatic breast cancer on CDK4/6 inhibitors. Furthermore, more frequent ctDNA monitoring is necessary to follow the dynamics by which RB1 alterations emerge and ascertain the association of their emergence with disease progression. Given this work, it is notable that CDK4/6 inhibition has recently been associated with increasing tumour cell antigen presentation through a mechanism involving downregulation of Rb1-E2F induced DNA methyltransferase 1 (DNMT1) activity, increased expression of endogenous retroviral elements and type III interferon production [17]. This response to CDK4/6 inhibition was ameliorated by silencing of RB1 and therefore could conceivably underlie an immune predatory selection pressure toward selection of Rb1 altered populations whilst undergoing treatment with CDK4/6 inhibitors. The fact that CDK4/6 inhibition has recently been shown to increase PD-L1 expression in mouse models of breast cancer provides a clear rationale for anti-PD1 treatment as a combination therapy with CDK4/6 inhibition before the emergence of Rb1 loss of function [18].

s whilst undergoing treatment with CDK4/6 inhibitors. The fact that CDK4/6 inhibition has recently been shown to increase PD-L1 expression in mouse models of breast cancer provides a clear rationale for anti-PD1 treatment as a combination therapy with CDK4/6 inhibition before the emergence of Rb1 loss of function [18]. Funding This work is supported by the Francis Crick Institute, which receives its core funding from Cancer Research UK (FC001169, FC001202), the UK Medical Research Council (FC001169, FC001202) and the Wellcome Trust (FC001169, FC001202). CS is funded by Cancer Research UK (TRACERx and CRUK Cancer Immunotherapy Catalyst Network), the CRUK Lung Cancer Centre of Excellence, Stand Up 2 Cancer (SU2C), the Rosetrees Trust, NovoNordisk Foundation (ID 16584), the Prostate Cancer Foundation, the Breast Cancer Research Foundation and the European Research Council (THESEUS), and support was provided to CS by the National Institute for Health Research, the University College London Hospitals Biomedical Research Centre and the Cancer Research UK University College London Experimental Cancer Medicine Centre. Disclosure The authors have declared no conflicts of interest.

Key Message Our results show that adding AAP to ADT is more effective than doc plus ADT in improving survival in mHNPC, however, the absolute difference in survival benefit between the two treatments remains uncertain. A STOPCAP network meta-analysis using individual participant data is underway to resolve uncertainty and to assess how trial or patient variability impacts on these results. Introduction Numerous randomised controlled trials (RCTs) have evaluated or are currently evaluating, the addition of other therapies to androgen-deprivation therapy (ADT) in men with metastatic hormone-naive prostate cancer (mHNPC). To determine reliably which are effective, we are conducting a series of systematic reviews under the auspices of the Systemic Treatment Options for Cancer of the Prostate (STOPCAP) collaboration. Our prior STOPCAP systematic reviews showed improved survival when abiraterone acetate plus prednisolone/prednisone (AAP) or docetaxel (Doc), but not zoledronic acid (ZA), were added to ADT [1, 2]. Trial evidence also suggests a benefit of combining celecoxib (Cel) with ZA and ADT [3].

Prostate (STOPCAP) collaboration. Our prior STOPCAP systematic reviews showed improved survival when abiraterone acetate plus prednisolone/prednisone (AAP) or docetaxel (Doc), but not zoledronic acid (ZA), were added to ADT [1, 2]. Trial evidence also suggests a benefit of combining celecoxib (Cel) with ZA and ADT [3]. However, only the reported comparisons of the STAMPEDE multi-arm, multi-stage (MAMS) platform and the ongoing Prostate Cancer Consortium in Europe (PEACE)-1 trial (NCT01957436) will be able to provide head-to head results comparing these therapies. Network meta-analysis [4, 5], which takes advantage of both direct and indirect comparisons, is therefore needed to determine reliably which is the optimal treatment(s), and so to inform patients, clinicians and policy makers. We have therefore conducted a systematic review and network meta-analysis of aggregate data (AD) to assess the optimal systemic treatments for men with mHNPC, making use of existing STOPCAP reviews [1, 2] and up-to-date results from individual trials, and also taking account of the MAMS platform design of the STAMPEDE trial protocol. Methods The full protocol for this review was registered in July 2017 (http://www.crd.york.ac.uk/PROSPERO/display_record.asp? ID=CRD42017071811, 1 March 2018, date last accessed).

However, only the reported comparisons of the STAMPEDE multi-arm, multi-stage (MAMS) platform and the ongoing Prostate Cancer Consortium in Europe (PEACE)-1 trial (NCT01957436) will be able to provide head-to head results comparing these therapies. Network meta-analysis [4, 5], which takes advantage of both direct and indirect comparisons, is therefore needed to determine reliably which is the optimal treatment(s), and so to inform patients, clinicians and policy makers. We have therefore conducted a systematic review and network meta-analysis of aggregate data (AD) to assess the optimal systemic treatments for men with mHNPC, making use of existing STOPCAP reviews [1, 2] and up-to-date results from individual trials, and also taking account of the MAMS platform design of the STAMPEDE trial protocol. Methods The full protocol for this review was registered in July 2017 (http://www.crd.york.ac.uk/PROSPERO/display_record.asp? ID=CRD42017071811, 1 March 2018, date last accessed). Eligibility criteria The eligibility criteria for inclusion in this network meta-analysis mirror those in prior systematic reviews [1, 2]. In brief, eligible trials should have been randomised in a way which precluded prior knowledge of the treatment assigned and compared ADT alone with ADT in combination with any of the agents (or combinations of agents) under consideration, namely celecoxib (Cel), zoledronic acid (ZA), celecoxib and zoledronic acid (ZA + Cel), docetaxel (Doc), zoledronic acid + docetaxel (ZA + Doc) or abiraterone acetate plus prednisolone (AAP). The men randomised were diagnosed with mHNPC, and either starting or responding to the first-line ADT for metastatic disease (they may have received prior treatments for early, localised disease). Trials were also eligible if they met the above criteria but additionally co-administered supportive treatments on the experimental arm only. Trials were excluded if they had randomised men who had castration-resistant prostate cancer or if they had included additional first-line treatments only on the control arm only.

rials were also eligible if they met the above criteria but additionally co-administered supportive treatments on the experimental arm only. Trials were excluded if they had randomised men who had castration-resistant prostate cancer or if they had included additional first-line treatments only on the control arm only. Trial identification As part of the wider STOPCAP project, we regularly and systematically search a number of sources to identify all published, unpublished and ongoing trials in mHNPC, providing a comprehensive and up-to-date database of all RCTs eligible for all of our STOPCAP systematic reviews. We also request regular updates from relevant trial teams on the status and reporting plans. Thus, all trials included in our previous STOPCAP reviews [1, 2] and any additional RCTs meeting the eligibility criteria were included. In summary, we searched MEDLINE, EMBASE, clinicaltrials.gov and the Cochrane Central Register of Controlled Trials (CENTRAL), using database-specific search strategies. We also searched proceedings from relevant conferences. In addition, reference lists of review articles and bibliographies of identified trial reports were screened for further eligible trials. Full search strategies have been previously reported [1, 2]. Outcomes The primary outcome was overall survival (OS), with failure-free survival (FFS) the secondary outcome.

In summary, we searched MEDLINE, EMBASE, clinicaltrials.gov and the Cochrane Central Register of Controlled Trials (CENTRAL), using database-specific search strategies. We also searched proceedings from relevant conferences. In addition, reference lists of review articles and bibliographies of identified trial reports were screened for further eligible trials. Full search strategies have been previously reported [1, 2]. Outcomes The primary outcome was overall survival (OS), with failure-free survival (FFS) the secondary outcome. Data extraction The principal data extracted or derived from included studies was the log hazard ratio and SE or the information to estimate them [e.g. a hazard ratio (HR) and confidence interval (CI) or P value [6] for OS and FFS]. Outcome definitions were also extracted for each trial to ensure their consistency and the appropriateness of combining results in a formal meta-analysis. Additional summary data including start and end dates of recruitment, details of treatment schedules on the control and experimental arms, numbers of patients and their demographics were also extracted, either directly from the trial publications or from prior systematic reviews.

ning results in a formal meta-analysis. Additional summary data including start and end dates of recruitment, details of treatment schedules on the control and experimental arms, numbers of patients and their demographics were also extracted, either directly from the trial publications or from prior systematic reviews. Assessing the risk of bias of included trials Assessment of study quality for all trials included in the prior STOPCAP reviews was previously carried out in the individual reviews, using the Cochrane risk of bias tool [7] and all included studies were assessed as having low risk of bias based on reported information and study protocols [8]. Risk of bias assessments for additional eligible studies identified for inclusion in the network meta-analysis was also carried out using the Cochrane tool. Analysis One of the trials identified as eligible for inclusion, the STAMPEDE trial, used a MAMS platform design. The nature of this type of design means that some patients randomised to the control arm contribute to multiple pairwise treatment comparisons within the trial. Therefore, to appropriately include data from this trial in an AD network meta-analysis, we needed to assess the correlations between the effect estimates arising due to periods of overlap in the common control arm. Correlations were estimated using the control-arm event counts within the periods of overlap for which data were obtained directly from the STAMPEDE investigators.

an AD network meta-analysis, we needed to assess the correlations between the effect estimates arising due to periods of overlap in the common control arm. Correlations were estimated using the control-arm event counts within the periods of overlap for which data were obtained directly from the STAMPEDE investigators. The primary analysis was carried out using a frequentist contrast-based network meta-analysis model and the network suite of commands [5] in Stata v14.1 (StataCorp, Texas, USA). Because all trials in this network, apart from the MAMS trial, are two-arm comparisons with the common control arm, any test for inconsistency would assess heterogeneity between the MAMS and other studies. Therefore, we fitted a consistency model and assessed heterogeneity, assuming a common heterogeneity variance across all comparisons. In the primary analysis, all comparisons of treatment combinations (e.g. ADT + ZA + Doc versus ADT) were assumed to be unrelated to the comparisons of their component treatments (e.g. ZA and Doc). A sensitivity analysis assumed additive treatment effects.

assuming a common heterogeneity variance across all comparisons. In the primary analysis, all comparisons of treatment combinations (e.g. ADT + ZA + Doc versus ADT) were assumed to be unrelated to the comparisons of their component treatments (e.g. ZA and Doc). A sensitivity analysis assumed additive treatment effects. A network map (Figure 1) was constructed to display all of the available relationships, with distinct treatments represented by nodes, and trials (or separate trial comparisons within the single MAMS design trial) by lines joining appropriate nodes. The thickness of the lines, representing the extent of available data for each comparison, was estimated from the combined number of events for all trials contributing to each individual comparison. Borrowing of strength statistics, which represent the proportion of the information for each treatment comparison that the indirect evidence from the network model has contributed, were calculated using the score decomposition method [9]. Figure 1. Network meta-analysis structure. AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid.

A network map (Figure 1) was constructed to display all of the available relationships, with distinct treatments represented by nodes, and trials (or separate trial comparisons within the single MAMS design trial) by lines joining appropriate nodes. The thickness of the lines, representing the extent of available data for each comparison, was estimated from the combined number of events for all trials contributing to each individual comparison. Borrowing of strength statistics, which represent the proportion of the information for each treatment comparison that the indirect evidence from the network model has contributed, were calculated using the score decomposition method [9]. Figure 1. Network meta-analysis structure. AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid. Estimates of relative effect for each pairwise treatment comparison from the primary consistency model were estimated on the HR scale along with corresponding 95% confidence limits and displayed in a network forest plot [5]. Treatment rankings were also calculated and summarised as a surface under the cumulative rank (SUCRA) value, representing the re-scaled mean ranking [10]. Further detailed methods relating to all the planned analyses may be found in the Statistical Analysis Plan (available on request).

yed in a network forest plot [5]. Treatment rankings were also calculated and summarised as a surface under the cumulative rank (SUCRA) value, representing the re-scaled mean ranking [10]. Further detailed methods relating to all the planned analyses may be found in the Statistical Analysis Plan (available on request). Finally, we conducted an indirect comparison of the two most effective treatments to estimate the relative difference in the size of the effects. We estimated the absolute benefit of treatment on OS at 3 years by applying the HR estimates to the approximate survival at 3 years. Role of the funding source The funders of the study (Medical Research Council and Prostate Cancer UK) had no role in study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

funders of the study (Medical Research Council and Prostate Cancer UK) had no role in study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Description of the included trials Searches undertaken for the STOPCAP project identified 11 trials that were eligible for inclusion in the network meta-analysis. Five of these 11 eligible trials could not be included in the network meta-analysis. Of these five, two trials that together randomised 72 men to receive ADT or ADT plus doc and two trials that together randomised 102 men to receive ADT versus ADT plus ZA identified as eligible for a previous STOPCAP review [2] could not be included here as they had not yet presented results for survival outcomes. The fifth trial, PEACE-1 (NCT01957436), comparing standard of care with or without AAP continues to recruit towards a target of 1168 men and no results are currently available for inclusion.

ligible for a previous STOPCAP review [2] could not be included here as they had not yet presented results for survival outcomes. The fifth trial, PEACE-1 (NCT01957436), comparing standard of care with or without AAP continues to recruit towards a target of 1168 men and no results are currently available for inclusion. Therefore, six RCTs that had reported results were included in the network meta-analysis (Figure 1, Table 1). Two trials compared ADT with ADT plus ZA; two compared ADT with ADT plus Doc and one trial compared ADT with ADT plus AAP. The final trial, STAMPEDE [11], contributed six separate treatment comparisons [3, 12, 13] to the network. Four comparisons were of ADT with Cel, ZA, Doc or AAP and two further comparisons were of ADT with combinations of ZA plus Cel or ZA plus Doc. Importantly, although each of the six comparisons shared a common control arm, there was some non-contemporaneous recruitment to individual treatment comparisons. In total, 6204 men were included in the network meta-analysis, representing 97% of men randomised in all completed eligible trials (at least 83% of men randomised in all 11 eligible trials). Accounting for the shared control arm patients in the STAMPEDE trial, 2615 men were randomised to receive ADT alone, and 3589 men were randomised to receive ADT in combination with one of the treatments being considered in the network. Table 1. Description of included trials (or treatment comparisons from the STAMPEDE trial) and FFS definition used in the trial. All trials had a control arm of ADT

ere randomised to receive ADT alone, and 3589 men were randomised to receive ADT in combination with one of the treatments being considered in the network. Table 1. Description of included trials (or treatment comparisons from the STAMPEDE trial) and FFS definition used in the trial. All trials had a control arm of ADT Trial Recruitment period Median follow-up (months) Treatment Treatment (N) Control (N) Definition of FFS CALGB 90202 [21] June 2004 to April 2012 Unknown ADT + ZA 323 322 Time to first bone progression, PSA progression, or death GETUG 15 [22] Oct 2004 to Dec 2008 84 ADT + Doc 192 193 Time to PSA progression, clinical progression or death STAMPEDE (Arms A versus D) [3] Oct 2005 to April 2011 69 ADT + Cel 188 377 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer STAMPEDE (Arms A versus F) [3] Oct 2005 to April 2011 69 ADT + ZA + Cel 190 377 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer STAMPEDE (Arms A versus B) [13] Oct 2005 to March 2013 43 ADT +ZA 366 724 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer STAMPEDE (Arms A versus C) [13] Oct 2005 to March 2013 43 ADT + Doc 362 724 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer STAMPEDE (Arms A versus E) [13] Oct 2005 to March 2013 43 ADT + ZA + Doc 365 724 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer CHAARTED [23] July 2006 to Dec 2012 54 ADT + Doc 397 393 Time to PSA rise or clinical progression ZAPCA (KYUH TRIG0705) [24] May 2008 to Dec 2010 42 ADT +ZA 109 110 Time to earliest date of PSA progression, clinical progression, first SRE, death for any reason, or cessation of protocol treatment for any reason STAMPEDE (Arms A versus G) [12] Nov 2011- Jan 2014 40 ADT + AAP 500 502 Time to PSA failure, progression of local, lymph-node, or distant metastases; or death from prostate cancer LATITUDE [14] Feb 2013 to Dec 2014 30 ADT + AAP 597 602 Time to radiographic progression or death from any cause AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; FFS, failure-free survival; SRE, skeletal related events; ZA, zoledronic acid.

cancer LATITUDE [14] Feb 2013 to Dec 2014 30 ADT + AAP 597 602 Time to radiographic progression or death from any cause AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; FFS, failure-free survival; SRE, skeletal related events; ZA, zoledronic acid. OS and FFS were reported for all of the treatment comparisons. The definition of FFS included time to prostate specific antigen (PSA) or clinical progressionor death for all of the trials, except LATITUDE [14], which did not include time to PSA progression and CHAARTED [15], which did not include time to death. Further details of the trials, including the definitions used for FFS, are given in Table 1. Borrowing of strength from the network Inclusion in the network led to a gain in information for each of the pairwise treatment comparisons (Table 2). For OS, this ranged from 0.9% (AAP) to 9.2% (ZA + Doc). For FFS, the gains were generally greater and ranged from 6.7% (Doc) to 21.7% (ZA + Doc). Table 2. Borrowing of strength statistics (% of information gained per pairwise analysis through inclusion in the network) Comparison OS, % FFS ADT versus ADT +Cel 5.4 17.3 ADT versus ADT + ZA 3.9 7.7 ADT versus ADT + ZA + Cel 5.2 17.1 ADT versus ADT + Doc 2.0 6.7 ADT versus ADT + ZA + Doc 9.2 21.7 ADT versus ADT + AAP 0.9 7.4 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; FFS, failure-free survival; OS, overall survival; ZA, zoledronic acid.

ADT versus ADT + ZA + Cel 5.2 17.1 ADT versus ADT + Doc 2.0 6.7 ADT versus ADT + ZA + Doc 9.2 21.7 ADT versus ADT + AAP 0.9 7.4 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; FFS, failure-free survival; OS, overall survival; ZA, zoledronic acid. Overall survival The network meta-analysis HR estimates suggested that compared with ADT alone each of AAP (HR = 0.61, 95% CI 0.53–0.71), Doc (HR = 0.77, 95% CI 0.68–0.87), ZA + Doc (HR = 0.79, 95% CI 0.66–0.94) and ZA + Cel (HR = 0.78, 95% CI 0.62–0.97) in combination with ADT improved survival. There was no survival advantage observed with ADT in combination with either ZA (HR = 0.90, 95% CI 0.79–1.03) or Cel (0.94, 95% CI 0.75–1.17) over ADT alone. For the comparisons of ADT versus ADT + Cel, ADT + ZA + Cel and ADT + ZA + Doc, the only data available were from single comparisons within the STAMPEDE trial [3, 13]. There was no evidence of heterogeneity between the effects of treatment within any of the individual treatment comparisons and all of the estimates from the network analysis were in keeping with those obtained in the previously reported pairwise meta-analyses where available (Figure 2). Figure 2. Overall survival. Forest plot of network and pairwise estimates of treatment effects [all treatments compared with androgen-deprivation therapy (ADT) alone]. AAP, abiraterone acetate plus prednisolone/prednisone; CI, confidence interval; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid.

Overall survival The network meta-analysis HR estimates suggested that compared with ADT alone each of AAP (HR = 0.61, 95% CI 0.53–0.71), Doc (HR = 0.77, 95% CI 0.68–0.87), ZA + Doc (HR = 0.79, 95% CI 0.66–0.94) and ZA + Cel (HR = 0.78, 95% CI 0.62–0.97) in combination with ADT improved survival. There was no survival advantage observed with ADT in combination with either ZA (HR = 0.90, 95% CI 0.79–1.03) or Cel (0.94, 95% CI 0.75–1.17) over ADT alone. For the comparisons of ADT versus ADT + Cel, ADT + ZA + Cel and ADT + ZA + Doc, the only data available were from single comparisons within the STAMPEDE trial [3, 13]. There was no evidence of heterogeneity between the effects of treatment within any of the individual treatment comparisons and all of the estimates from the network analysis were in keeping with those obtained in the previously reported pairwise meta-analyses where available (Figure 2). Figure 2. Overall survival. Forest plot of network and pairwise estimates of treatment effects [all treatments compared with androgen-deprivation therapy (ADT) alone]. AAP, abiraterone acetate plus prednisolone/prednisone; CI, confidence interval; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid. Treatment rankings When used in combination with ADT, AAP has the highest probability (94%) of being the most effective treatment, Doc has a 35% probability of being the second-best treatment and ADT alone has the highest probability of being the least effective treatment (67%, Table 3). Table 3. Treatment ranking (% probability) and SUCRA values based on overall survival results

the highest probability (94%) of being the most effective treatment, Doc has a 35% probability of being the second-best treatment and ADT alone has the highest probability of being the least effective treatment (67%, Table 3). Table 3. Treatment ranking (% probability) and SUCRA values based on overall survival results AAP Doc ZA + Doc ZA + Cel ZA Cel ADT alone Best 94.2 0.7 1.3 3.8 0.0 0.0 0.0 Second best 5.3 34.9 25.5 33.0 0.3 1.0 0.0 Third best 0.4 36.8 30.3 27.0 2.4 3.1 0.0 Fourth best 0.1 23.6 30.8 23.9 12.2 9.3 0.1 Fifth best 0.0 3.8 9.3 9.3 48.7 26.0 2.9 Sixth best 0.0 0.2 2.6 2.5 31.3 33.6 29.8 Worst 0.0 0.0 0.2 0.5 5.1 27.0 67.2 SUCRA 1.0 0.7 0.6 0.6 0.3 0.2 0.1 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; SUCRA, surface under the cumulative rank; ZA, zoledronic acid.

8 9.3 9.3 48.7 26.0 2.9 Sixth best 0.0 0.2 2.6 2.5 31.3 33.6 29.8 Worst 0.0 0.0 0.2 0.5 5.1 27.0 67.2 SUCRA 1.0 0.7 0.6 0.6 0.3 0.2 0.1 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; SUCRA, surface under the cumulative rank; ZA, zoledronic acid. Failure-free survival There was an FFS benefit associated with adding ADT to each of AAP (HR = 0.38 95% CI 0.31–0.46), Doc (HR + 0.64 95% CI 0.54–0.75) and ZA + Doc (HR = 0.63 95% CI 0.49–0.80) compared with ADT alone. No statistically significant benefit was seen with the addition of Cel (HR = 0.89 95% CI 0.67–1.17); ZA + Cel (HR = 0.80 95% CI 0.60–1.05) or ZA alone (HR = 0.88 95% CI 0.75–1.05). In all cases, the HR estimates obtained through the network were very similar to those obtained using a standard pairwise meta-analysis, providing confirmation that the network model is behaving as expected. There was evidence of variation or inconsistency between the effects of treatment within the individual treatment comparisons of ADT versus ADT plus AAP (I2=91%, heterogeneity P = 0.001) where there was a large variation between the size of the relative effects (but not the direction of the effect) observed between the two included trial comparisons. However, there was no evidence of variation or inconsistency between the effects of treatment within the remaining treatment comparisons, and all of the estimates from the network analysis were in keeping with those obtained in the previously reported pairwise meta-analyses where available (Figure 3).

rial comparisons. However, there was no evidence of variation or inconsistency between the effects of treatment within the remaining treatment comparisons, and all of the estimates from the network analysis were in keeping with those obtained in the previously reported pairwise meta-analyses where available (Figure 3). Figure 3. Failure-free survival. Forest plot of network and pairwise estimates of treatment effects [all treatments compared with androgen-deprivation therapy (ADT) alone]. AAP, abiraterone acetate plus prednisolone/prednisone; CI, confidence interval; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid. Therefore, we carried out a sensitivity analysis using the outcome of time to PSA failure as reported in LATITUDE to assess the robustness of our primary analysis. This analysis, whilst not changing our interpretation, did result in an HR estimate from the network analysis was slightly more in favour of treatment (HR = 0.30, 95% CI 0.27–0.34) with no evidence of variation of inconsistency (I2=0, heterogeneity P = 0.78). Based on the treatment rankings, when combined with ADT, AAP has the highest probability (100%) of being the most effective treatment in terms of FFS, whilst either Doc alone (45% probability) or in combination with ZA (52% probability) is most likely to be the second-best treatment. ADT alone has the highest probability of being the least effective treatment (73%, Table 4). Table 4. Treatment ranking (% probability) and SUCRA values based on failure-free survival results

t either Doc alone (45% probability) or in combination with ZA (52% probability) is most likely to be the second-best treatment. ADT alone has the highest probability of being the least effective treatment (73%, Table 4). Table 4. Treatment ranking (% probability) and SUCRA values based on failure-free survival results AAP ZA + Doc Doc ZA + Cel ZA Cel ADT alone Best 100.0 0.0 0.0 0.0 0.0 0.0 0.0 Second best 0.0 52.0 45.1 2.6 0.0 0.3 0.0 Third best 0.0 41.3 47.9 9.5 0.1 1.2 0.0 Fourth best 0.0 5.7 6.7 53.3 14.7 19.1 0.5 Fifth best 0.0 1.0 0.3 21.5 42.0 31.4 3.8 Sixth best 0.0 0.0 0.0 10.4 37.6 29.1 22.9 Worst 0.0 0.0 0.0 2.7 5.6 18.9 72.8 SUCRA 1.0 0.7 0.7 0.4 0.3 0.3 0.1 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; SUCRA, surface under the cumulative rank; ZA, zoledronic acid.

0.3 21.5 42.0 31.4 3.8 Sixth best 0.0 0.0 0.0 10.4 37.6 29.1 22.9 Worst 0.0 0.0 0.0 2.7 5.6 18.9 72.8 SUCRA 1.0 0.7 0.7 0.4 0.3 0.3 0.1 AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; Cel, celecoxib; Doc, docetaxel; SUCRA, surface under the cumulative rank; ZA, zoledronic acid. Indirect comparison of the two most effective treatments When used in combination with ADT, two treatments, AAP and Doc, emerged as being effective in terms of improving both OS and FFS relative to ADT alone, and with the greatest probabilities of being the top two most effective treatments; therefore, they were compared indirectly in a pairwise comparison. The HR estimate for the effect of ADT + AAP relative to the effect of ADT + Doc on OS is 0.80 (95% CI 0.66–0.96). Assuming a baseline OS of 60% at 3 years with ADT + Doc, this translates to an absolute survival benefit associated with AAP of 6% (95% CI = 1% to 11%), that is, to 66% at 3 years (95% CI 61% to 71%). For FFS, the HR for the effect of ADT + AAP relative to ADT + Doc is 0.59 (95% CI 0.46–0.75) (Figure 4). Figure 4. Indirect comparison of the two most effective treatment combinations (A) overall survival and (B) failure-free survival. AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; CI, confidence interval; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid.

Indirect comparison of the two most effective treatments When used in combination with ADT, two treatments, AAP and Doc, emerged as being effective in terms of improving both OS and FFS relative to ADT alone, and with the greatest probabilities of being the top two most effective treatments; therefore, they were compared indirectly in a pairwise comparison. The HR estimate for the effect of ADT + AAP relative to the effect of ADT + Doc on OS is 0.80 (95% CI 0.66–0.96). Assuming a baseline OS of 60% at 3 years with ADT + Doc, this translates to an absolute survival benefit associated with AAP of 6% (95% CI = 1% to 11%), that is, to 66% at 3 years (95% CI 61% to 71%). For FFS, the HR for the effect of ADT + AAP relative to ADT + Doc is 0.59 (95% CI 0.46–0.75) (Figure 4). Figure 4. Indirect comparison of the two most effective treatment combinations (A) overall survival and (B) failure-free survival. AAP, abiraterone acetate plus prednisolone/prednisone; ADT, androgen-deprivation therapy; CI, confidence interval; Cel, celecoxib; Doc, docetaxel; ZA, zoledronic acid. Discussion Summary of results Based on the current data, either AAP or Doc alongside ADT improves the survival of men with mHNPC. Although AAP has the highest probability of being the most effective treatment, with Doc + ADT the second most effective treatment, uncertainty remains about the difference in absolute magnitude of survival benefit between these two treatments (between 1% and 9% at 3 years). Adding ZA + Doc or ZA + Cel to ADT also improved survival relative to ADT alone.

lity of being the most effective treatment, with Doc + ADT the second most effective treatment, uncertainty remains about the difference in absolute magnitude of survival benefit between these two treatments (between 1% and 9% at 3 years). Adding ZA + Doc or ZA + Cel to ADT also improved survival relative to ADT alone. Strengths This is the first network meta-analysis to comprehensively assess and rank the effects of all the systemic treatments for mHNPC recently tested in RCTs. By working with trial investigators, we have been able to make use of the most up-to-date, and consistent AD from six trials and 6204 men, representing 97% of those randomised in reported, eligible trials. Thus, it provides the most reliable assessment of the relative effects of these agents to date. In particular, our analysis has been able to shed light on the comparison of AAP and Doc in conjunction with ADT and has also reinforced the observed survival benefit associated with the combination of Cel and ZA, although not with either of these agents given individually [3]. Furthermore, to the best of our knowledge, this is the first example of a network meta-analysis that has taken account of the complexities of including a MAMS platform trial, appropriately adjusting the analysis for the proportion of the control-arm patients common to different pairwise comparisons. Obtaining limited unpublished information direct from the investigators made this possible.

a network meta-analysis that has taken account of the complexities of including a MAMS platform trial, appropriately adjusting the analysis for the proportion of the control-arm patients common to different pairwise comparisons. Obtaining limited unpublished information direct from the investigators made this possible. Limitations Whilst we have been comprehensive in our approach to this review, relying on AD inevitably has limitations. For example, whilst most trials include PSA progression as part of the definition of FFS, the LATITUDE trial [14] presented results for PSA failure and clinical or radiological progression separately. Whilst our pre-specified primary analysis used time to clinical or radiological progression for the LATITUDE trial, a sensitivity analysis which used time to PSA failure data from LATITUDE showed that our results are robust to these different definitions. In hindsight, although this might have been the preferable primary analysis, it would have made little difference to our interpretation and conclusions.

ion for the LATITUDE trial, a sensitivity analysis which used time to PSA failure data from LATITUDE showed that our results are robust to these different definitions. In hindsight, although this might have been the preferable primary analysis, it would have made little difference to our interpretation and conclusions. There are differences in the patient characteristics, which may influence the effects of the various treatments either within or across trials. Despite these differences, however, there has been no clear evidence of heterogeneity in the effects of zoledronic acid, docetaxel or abiraterone across trials in the individual STOPCAP systematic reviews [1, 2]. Nevertheless, some of the individual trials have suggested treatment–covariate interactions, notably that the effects of Doc may be moderated by the extent of metastatic disease volume and that the effects of AAP may be associated with age [1, 12]. There are no well-defined methods to appropriately assess such interactions in an AD network meta-analysis. Therefore, they will be best investigated through the collection and re-analysis of individual participant data from all of the eligible trials, which the STOPCAP collaborators are currently working to undertake.

There are no well-defined methods to appropriately assess such interactions in an AD network meta-analysis. Therefore, they will be best investigated through the collection and re-analysis of individual participant data from all of the eligible trials, which the STOPCAP collaborators are currently working to undertake. Context Previously, both Doc and AAP have been shown to improve survival and delay progression in men with mHNPC [2, 16]. However, as no trials have set out to directly compare ADT plus docetaxel with ADT plus AAP, network meta-analysis provides the only comprehensive approach to this treatment comparison. However, the design of the STAMPEDE trial, in which treatments were simultaneously compared against ADT alone, has enabled an opportunistic analysis in which outcomes for men randomised within the same time frame to receive ADT plus Doc or ADT plus AAP were compared [17]. Whilst not a fully powered analysis, the results demonstrated an advantage of AAP over Doc with respect to FFS (HR = 0.51, 95% CI 0.39–0.67, P < 0.001), which did not translate to a survival benefit (HR = 1.16, 95% CI 0.82–1.65, P = 0.40). It is not clear why the findings of the STAMPEDE analysis differ with respect to OS from the findings of the network meta-analysis. The STAMPEDE analysis is a direct comparison of men who were randomised contemporaneously to receive either AAP or Doc in addition to ADT. The network meta-analysis has considerably more power to detect differences in treatment effects, due to both the inclusion of more trials and more men, and the additional strength that other treatment comparisons lend within the network. However, the inclusion of more trials, that randomised men to receive a variety of treatments over a longer time period inevitably brings with it greater variability, due to the broader case mix of patients, with different prognoses, and with differing access to treatments after progression of disease. It has not been possible in this analysis to account for changes in care over time, particularly whilst the number of new treatments available at relapse has rapidly increased, and once again this is best achieved through the collection and analysis of individual participant data.

eatments after progression of disease. It has not been possible in this analysis to account for changes in care over time, particularly whilst the number of new treatments available at relapse has rapidly increased, and once again this is best achieved through the collection and analysis of individual participant data. Other network meta-analyses have been reported. Feyerabend et al. [18] presented the results of an indirect comparison of AAP and Doc based on the results of just two trials. Wallis et al. [19] have also reported indirect comparisons of ADT plus AAP with ADT plus Doc. However, we have concerns regarding the methodological approaches used by the authors, in particular, the inclusion of patients with and without metastases; the double-counting of shared control arm patients, and therefore correlations, from the STAMPEDE trial and the use of flawed subgroup analysis methodology [20]. Finally, a key question that remains for both clinicians and patients is whether the two most effective treatments, AAP and Doc, could be safely combined and if so, what the impact of the combination may be on OS. This question will only be resolved when results from the ongoing PEACE-1 trial are available.

Other network meta-analyses have been reported. Feyerabend et al. [18] presented the results of an indirect comparison of AAP and Doc based on the results of just two trials. Wallis et al. [19] have also reported indirect comparisons of ADT plus AAP with ADT plus Doc. However, we have concerns regarding the methodological approaches used by the authors, in particular, the inclusion of patients with and without metastases; the double-counting of shared control arm patients, and therefore correlations, from the STAMPEDE trial and the use of flawed subgroup analysis methodology [20]. Finally, a key question that remains for both clinicians and patients is whether the two most effective treatments, AAP and Doc, could be safely combined and if so, what the impact of the combination may be on OS. This question will only be resolved when results from the ongoing PEACE-1 trial are available. Conclusions Our results support the use of either AAP or Doc alongside ADT in men with mHNPC. AAP appears to be the most effective treatment, but it is not clear to what extent and whether this is due to a true increased benefit with AAP or to the variable features of the individual trials. To fully account for patient variability across trials, changes in prognosis or treatment effects over time, and the potential impact of treatment on progression, a network meta-analysis based on individual participant data is currently in development.

efit with AAP or to the variable features of the individual trials. To fully account for patient variability across trials, changes in prognosis or treatment effects over time, and the potential impact of treatment on progression, a network meta-analysis based on individual participant data is currently in development. Acknowledgements The STOPCAP Collaborators thank all patients who participated in the trials and contributed to this research. The meta-analysis would not have been possible without their participation or without the institutions and collaborative groups that carried out the trials, including a great many trial sites. We thank STAMPEDE trial team for supplying additional data to facilitate the network meta-analysis. Funding UK Medical Research Council (MC_UU_12023/25); Prostate Cancer UK Research Innovation Award (RIA16-ST2-020).

Acknowledgements The STOPCAP Collaborators thank all patients who participated in the trials and contributed to this research. The meta-analysis would not have been possible without their participation or without the institutions and collaborative groups that carried out the trials, including a great many trial sites. We thank STAMPEDE trial team for supplying additional data to facilitate the network meta-analysis. Funding UK Medical Research Council (MC_UU_12023/25); Prostate Cancer UK Research Innovation Award (RIA16-ST2-020). Disclosure CLV, DJF, IWW, JFT, LHMR, SB and GG declare no competing interests. MKBP, MSp and MSy report grants from Astellas, Janssen, Novartis, Pfizer, Sanofi and Clovis Oncology outside the submitted work. JC reports grants from MRC as well as personal fees from Pfizer and GSK, and grants from Novartis outside the submitted work. NWC reports participation on advisory boards for Astellas, Janssen, Ferring, Bayer, AstraZeneca and corporate-sponsored Research for AstraZeneca. KF reports participation on advisory boards for Janssen and Sanofi outside the submitted work. MDM reports receiving personal fees from Sanofi, Bayer and Janssen, outside the submitted work. NDJ reports participation on advisory boards for Sanofi and Novartis, and grants, personal fees and non-financial support from Janssen, outside the submitted work. CJS reports Ownership BIND; Advisory board participation for Astellas/Medivation, Astra Zeneca, Sanofi, Janssen, BIND, Bayer; Grants from Janssen, Astellas/Medivation, Sanofi, Janssen, Soti, Exelixis; and consultancy for Astellas/Medivation, Pfizer, Sanofi, Janssen, BIND, Bayer, Genentech.

Key Message Abiraterone acetate and docetaxel, with predniso(lo)ne (AAP, DocP) separately improved survival when added to standard-of-care for hormone-sensitive prostate cancer. STAMPEDE randomised 566 patients to these treatment arms when both were accruing, the only head-to-head data available. No evidence of a difference in overall or prostate cancer-specific survival. Research in context Evidence before this study Abiraterone acetate plus prednisone/prednisolone (AAP) and docetaxel with prednisone/prednisolone (DocP) have separately been shown to improve survival when used in addition to the previous international standard-of-care (SOC) for hormone-sensitive prostate cancer of androgen deprivation therapy with further therapy such as AAP or DocP on relapse. This has been confirmed in a number of separate trials and on meta-analysis. The largest body of evidence for both AAP and DocP comes from the systemic therapy for advanced or metastatic prostate cancer: evaluation of drug efficacy (STAMPEDE) platform trial.

therapy with further therapy such as AAP or DocP on relapse. This has been confirmed in a number of separate trials and on meta-analysis. The largest body of evidence for both AAP and DocP comes from the systemic therapy for advanced or metastatic prostate cancer: evaluation of drug efficacy (STAMPEDE) platform trial. Added value of this study Recruitment to DocP and AAP overlapped in STAMPEDE giving the only head-to-head evidence comparing these two new standard treatment approaches. We report data from the 566 patients who were directly randomised between these two treatment approaches while the two research arms were both open to recruitment. The data show strong evidence favouring SOC + AAP on earlier, more biochemically driven outcome measures (OMs). For longer-term, more clinically driven OMs, including bone complications, prostate cancer-specific and overall survival, there is no evidence of a significant difference between AAP and DocP. Implications of all the available evidence The reported trials and meta-analyses showed a larger effect on survival for AAP over the previous SOC than did DocP over the standard SOC. These data show that the story may be more complicated. No other directly randomised data on survival of these treatments are available. Individual patient data network meta-analysis using all of the published trials are warranted, accounting for differences in patient characteristics, treating clinicians and centres and salvage treatment access. The STAMPEDE team is collaborating with the STOPCAP meta-analysis group to achieve this.

ents are available. Individual patient data network meta-analysis using all of the published trials are warranted, accounting for differences in patient characteristics, treating clinicians and centres and salvage treatment access. The STAMPEDE team is collaborating with the STOPCAP meta-analysis group to achieve this. Introduction For several decades, the standard-of-care (SOC) for most patients with high-risk locally advanced or metastatic prostate cancer has been long-term androgen deprivation therapy (ADT) alone. The past few years, there have been great changes, first with results from randomised controlled trials (RCTs) showing a survival advantage compared with ADT alone for adding radiotherapy to the prostate in men with non-metastatic disease and no known nodal involvement [1–3]; then with systemic treatments for all men starting long-term hormone therapy: docetaxel plus prednisolone/prednisone (DocP) [4–9] and, most recently, abiraterone acetate plus prednisolone/prednisone (AAP) [10, 11]. As both therapeutic combinations are effective, there are now two distinct standards-of-care with little information to guide clinicians as to which is the more effective; there are no prospective, powered, RCTs that will deliver direct comparative data.

abiraterone acetate plus prednisolone/prednisone (AAP) [10, 11]. As both therapeutic combinations are effective, there are now two distinct standards-of-care with little information to guide clinicians as to which is the more effective; there are no prospective, powered, RCTs that will deliver direct comparative data. Systemic therapy for advanced or metastatic prostate cancer: evaluation of drug efficacy (STAMPEDE) is a multi-arm, multi-stage platform protocol which assessed both of these treatment approaches, separately, against the previous SOC [12, 13]. The ‘docetaxel comparison’ of STAMPEDE recruited patients allocated to SOC + DocP between October 2005 and March 2013. The ‘abiraterone comparison’, the first comparison to be added to STAMPEDE, recruited patients allocated to SOC or SOC + AAP between November 2011 and January 2014. Each of those comparisons had primary outcome measure (OM) of overall survival (OS) for the patients randomised contemporaneously to the control arm and the relevant research arm. Consequently, between 15 November 2011 and 31 March 2013, patients were directly randomised contemporaneously between these two research arms (and other research arms) and we now present these data. Methods Trial design The STAMPEDE protocol and design have been described in detail elsewhere [7, 10, 12, 14]. Briefly, STAMPEDE comprises a series of multi-arm multi-stage (MAMS) comparisons that have overlapped in recruitment and follow-up time.

patterns for adverse events according to treatment. The prevalence of grade 3 or 4 toxicity in patients with assessments at 1 year without a prior FFS event was 11% SOC + DocP and 11% SOC + AAP; at 2 years this was 11% SOC + DocP and 11% SOC + AAP. Table 3. Worst adverse event (grade) reported over entire time on trial SOC + Doc (n = 189) SOC + AAP (n = 377) Safety population Number of patients included in analysisa 172 373 Patients with an adverse event—no. (%) Grade 1–5 adverse event 172 (100) 370 (99) Grade 3–5 adverse event 86 (50) 180 (48) Grade 3–5 adverse events—no. (%) Endocrine disorder 15 (9) 49 (13) Febrile neutropenia 29 (17) 3 (1) Neutropenia (neutrophils) 22 (13) 4 (1) General disorder 18 (10) 21 (6) Fatigue 7 (4) 8 (2) Oedema 1 (1) 2 (1) Musculoskeletal disorder 9 (5) 33 (9) Cardiovascular disorder 6 (3) 32 (9) Hypertension 0 (0) 12 (3) Myocardial infarction 2 (1) 4 (1) Cardiac dysrhythmia 1 (1) 5 (1) Gastrointestinal disorder 9 (5) 28 (8) Hepatic disorder 1 (1) 32 (9) Increased AST 0 (0) 6 (2) Increased ALT 1 (1) 23 (6) Respiratory disorder 12 (7) 11 (3) Dyspnoea 4 (2) 1 (1) Renal disorder 5 (3) 20 (5) Lab abnormalities 9 (5) 11 (3) Hypokalaemia 0 (0) 3 (1) a The safety population includes patients who started their allocated treatment.

Systemic therapy for advanced or metastatic prostate cancer: evaluation of drug efficacy (STAMPEDE) is a multi-arm, multi-stage platform protocol which assessed both of these treatment approaches, separately, against the previous SOC [12, 13]. The ‘docetaxel comparison’ of STAMPEDE recruited patients allocated to SOC + DocP between October 2005 and March 2013. The ‘abiraterone comparison’, the first comparison to be added to STAMPEDE, recruited patients allocated to SOC or SOC + AAP between November 2011 and January 2014. Each of those comparisons had primary outcome measure (OM) of overall survival (OS) for the patients randomised contemporaneously to the control arm and the relevant research arm. Consequently, between 15 November 2011 and 31 March 2013, patients were directly randomised contemporaneously between these two research arms (and other research arms) and we now present these data. Methods Trial design The STAMPEDE protocol and design have been described in detail elsewhere [7, 10, 12, 14]. Briefly, STAMPEDE comprises a series of multi-arm multi-stage (MAMS) comparisons that have overlapped in recruitment and follow-up time. Patient selection Eligible patients were those starting long-term ADT for the first time. This was defined as patients with metastatic disease, nodal involvement or node negative, non-metastatic disease with two or more of three high-risk features: T-category 3 or 4, Gleason sum score 8–10 or PSA > 40 ng/ml. Patients rapidly relapsing after previous local therapy were also permitted if they had PSA > 20 ng/ml or PSA > 4 ng/ml with a PSA doubling time <6 months or those who developed loco-regional or metastatic spread whilst not on hormone therapy.

-risk features: T-category 3 or 4, Gleason sum score 8–10 or PSA > 40 ng/ml. Patients rapidly relapsing after previous local therapy were also permitted if they had PSA > 20 ng/ml or PSA > 4 ng/ml with a PSA doubling time <6 months or those who developed loco-regional or metastatic spread whilst not on hormone therapy. As with all STAMPEDE comparisons, the primary OM of the two underpinning comparisons (against control) was OS. Failure-free survival (FFS) was an intermediate primary OM, defined as time from randomisation to the first of: rising PSA (where rising PSA was defined as a confirmed rise to >4 ng/ml, and >50% above the lowest value in the first 6 months after randomisation); new disease or progression of: distant metastases, lymph nodes or local disease; or death from prostate cancer. Progression-free survival (PFS) was defined as time from randomisation to the first of: new disease or progression of: distant metastases, lymph nodes or local disease; or death from prostate cancer [15]. Metastatic PFS (MPFS) was defined as time from randomisation to death from any cause, new metastases or progression of distant metastases. All patients provided written informed consent; all versions of the protocol have been reviewed by the relevant research ethics committees and the regulatory agencies; the original protocol and all subsequent versions involving the introduction of a new research arm and comparison were independently peer-reviewed by Cancer Research UK (CRUK).

itten informed consent; all versions of the protocol have been reviewed by the relevant research ethics committees and the regulatory agencies; the original protocol and all subsequent versions involving the introduction of a new research arm and comparison were independently peer-reviewed by Cancer Research UK (CRUK). Patients have been allocated across a number of research treatments as depicted in Figure 1. Here we focus on those patients randomised between 15 November 2011 and 31 March 2013, while both the ‘docetaxel comparison’ and the ‘abiraterone comparison’ were open to recruitment, and who were allocated to either SOC + DocP or SOC + AAP. Figure 1. Activity-by-time diagram: patients included in this comparison. SOC, standard-of-care; Doc, docetaxel; Abi, abiraterone acetate+prednisone/prednisolone. Boxes represents periods of recruitment (x-axis) to each of the trial arms (y-axis). The blue boxes represent recruitment periods contributing to this analysis; the green boxes other recruitment period, past and future, contributing to other aspects of the STAMPEDE. The squares represent the time point of the first key comparative analyses for each comparison in pink and for this comparison in blue.

). The blue boxes represent recruitment periods contributing to this analysis; the green boxes other recruitment period, past and future, contributing to other aspects of the STAMPEDE. The squares represent the time point of the first key comparative analyses for each comparison in pink and for this comparison in blue. Trial treatment, masking and follow-up The SOC was long-term hormone therapy with LHRH analogues (with short term antiandrogen if relevant) or orchidectomy. Unless contraindicated, radiotherapy to the prostate was mandated in all patients with N0M0 disease, encouraged in patient with N + M0 disease, and permitted in patients with M1 disease until the activation of the ‘M1|RT comparison’ in January 2013. On the DocP arm, docetaxel (75 mg/m2) was given once every 3 weeks for six cycles, with prednisolone/prednisone (10 mg) daily. On the AAP arm, abiraterone acetate (1000 mg) with prednisolone/prednisone (5 mg) daily was given until PSA, clinical and radiological progression or a change of treatment. AAP duration was capped after 2 years in M0 patients having radical radiotherapy. Modifications for toxicities were described in the protocol and previous papers [7, 10]. Treatment allocation was not masked for practical reasons. Patients were seen 6-weekly at first, dropping to 6-monthly after 2 years. Imaging scans after baseline were at the investigator’s discretion.

having radical radiotherapy. Modifications for toxicities were described in the protocol and previous papers [7, 10]. Treatment allocation was not masked for practical reasons. Patients were seen 6-weekly at first, dropping to 6-monthly after 2 years. Imaging scans after baseline were at the investigator’s discretion. Randomisation Patients were randomised centrally using minimisation with a random element across a number of stratification factors using unequal allocation (previously described) [7, 10]. The allocation ratio was initially 2 : 1 control : research; the ‘abiraterone comparison’ was brought in with an equal allocation (1 : 1) ratio to the control. Therefore the allocation ratio here is 1 : 2 for SOC + DocP : SOC + AAP. Statistical analysis The comparison presented here is of SOC + AAP against SOC + DocP because both of these arms have demonstrated better OS than their contemporaneous controls in the population of men starting long-term hormone therapy. The protocol specified that research arms which were better than the control arm could be compared, following a closed test approach. The maturity of the data used for SOC + AAP matches that recently reported [10] in the primary results and is updated to the same data freeze timepoint for SOC + DocP so is longer-term data than previously reported results for this arm [7].

etter than the control arm could be compared, following a closed test approach. The maturity of the data used for SOC + AAP matches that recently reported [10] in the primary results and is updated to the same data freeze timepoint for SOC + DocP so is longer-term data than previously reported results for this arm [7]. The previously-reported comparisons of SOC + DocP versus SOC and SOC + AAP versus SOC had formal sample size calculations; there is no formal sample size calculation for this comparison: it is an opportunistic comparison between the contemporaneously recruited research arm patients. Although the recruitment overlap is only 17 months, 566 patients were allocated to the 2 research arms of interest and thus contribute substantial information to inform this comparison. Standard survival analysis methods were used, following the approach for each of these underpinning comparisons; hazard ratios (HR) were estimated from adjusted Cox models, after checking that the proportional hazards assumption held, where an HR < 1 represents evidence in favour of SOC + AAP and HR > 1 represents evidence in favour of SOC + DocP. Nominal confidence intervals are presented at the 95% level. A P-value <0.1 was considered indicative of treatment-baseline characteristic interaction, recognising the limited power of the heterogeneity tests. Efficacy analyses were done in the intention-to-treatment basis, by allocated treatment. Safety analyses were done only in patients who started their allocated treatment.

A P-value <0.1 was considered indicative of treatment-baseline characteristic interaction, recognising the limited power of the heterogeneity tests. Efficacy analyses were done in the intention-to-treatment basis, by allocated treatment. Safety analyses were done only in patients who started their allocated treatment. Results Accrual and characteristics The dataset for this comparison was frozen on 10 February 2017. Between 15 November 2011 and 31 March 2013, 1348 patients joined all open arms STAMPEDE. Of the 566 randomised to the comparison reported here, 189 (14%) were allocated to SOC + DocP, 377 (28%) to SOC + AAP. The flow of patients to this comparison is shown in Figure 2. Table 1 shows the baseline characteristics of patients in this comparison which differ only slightly from the previous papers (summarised in supplementary Table S1, available at Annals of Oncology online). Median follow-up, calculated by reverse censoring on survival, was 48 months. Table 1. Baseline characteristics of patients allocated to SOC + DocP or SOC + AAP by whether contributing to the direct comparison

ghtly from the previous papers (summarised in supplementary Table S1, available at Annals of Oncology online). Median follow-up, calculated by reverse censoring on survival, was 48 months. Table 1. Baseline characteristics of patients allocated to SOC + DocP or SOC + AAP by whether contributing to the direct comparison SOC + DocP SOC + AAP Overall Characteristic N % N % N % Metastases M0 74 39 150 40 224 40 M1 115 61 227 60 342 60 Nodal stage N0 82 43 158 42 240 44 N+ 99 52 202 53 301 56 NX 8 4 17 5 25 n/a Combination N0 M0 43 23 84 22 127 22 N+M0 31 16 66 18 97 17 N0 M1 39 21 74 20 113 20 N+ M1 68 36 136 36 204 36 NX M1 8 4 17 5 25 4 Tumour category <T3 24 13 36 10 60 11 T3 123 65 249 66 372 69 T4 39 20 68 18 107 20 Tx 3 2 24 6 27 n/a Gleason category ≤7 35 19 91 25 126 23 8–10 153 81 276 75 429 76 Unknown 1 — 10 — 11 n/a Previous local therapy No 183 97 350 93 533 94 Yes 6 3 27 7 33 6 WHO performance status 0 149 79 300 80 449 79 1–2 40 21 77 20 117 21 Age (years) <70 134 71 267 71 401 71 70+ 55 29 110 29 165 29 Median (quartiles) 66 (62–71) 66 (61–70) 66 (62–70) Mean (SD) 66 (7) 66 (7) 66 (7) Use of NSAID or aspirin No use 141 75 280 74 421 74 Uses either 48 25 97 26 145 26 PSA (ng/ml) Median (quartiles) 58 (29–162) 55 (20–194) 56 (22–185) Mean (SD) 193 (421) 274 (631) 247 (571) Ln PSA (ng/ml) Median (quartiles) 4.1 (3.4–5.1) 4.0 (3.0–5.3) 4.0 (3.1–5.2) Mean (SD) 4.2 (1.4) 4.2 (1.6) 4.2 (1.5) RT planned M0, yes 57 77 118 79 175 78 M0, no 17 23 32 21 49 22 M1, yes 12 10 21 9 33 10 M1, no 103 89 206 91 309 90 Hypertension Yes (still fit for trial) 64 34 149 40 213 38 No 125 66 227 60 352 62 Year of randomisation 2011 15 8 27 7 42 7 2012 138 73 277 73 415 73 2013 36 19 73 19 109 19

.2 (1.4) 4.2 (1.6) 4.2 (1.5) RT planned M0, yes 57 77 118 79 175 78 M0, no 17 23 32 21 49 22 M1, yes 12 10 21 9 33 10 M1, no 103 89 206 91 309 90 Hypertension Yes (still fit for trial) 64 34 149 40 213 38 No 125 66 227 60 352 62 Year of randomisation 2011 15 8 27 7 42 7 2012 138 73 277 73 415 73 2013 36 19 73 19 109 19 Figure 2. CONSORT diagram. SOC, standard-of-care; DocP, docetaxel+prednisolone/prednisone; AAP, abiraterone acetate+prednisolone/prednisone. Selection of patients for this comparison. Overall survival There were 44/189 (23%) deaths on the SOC + DocP arm and 105/377 (28%) deaths on the SOC + AAP arm. The estimated HR = 1.16 (95% CI 0.82–1.65; P = 0.40) (Figure 3A). Estimates in patients with and without metastases are shown in Table 2, with HR = 1.51 (95% CI 0.58–3.93) in M0 patients and HR = 1.13 (95% CI 0.77–1.66) in M1 patients. There was no evidence of interaction in the treatment effect by baseline metastases (P = 0.69). Table 2. Hazard ratio for SOC + AAP relative to SOC + DocP from adjusted Cox models Outcome measure Patient group Events/Pts SOC + DocP Events/Pts SOC + AAP Hazard ratioa (95% CI) P-value Interaction by metastases P-value Failure-free survivalb All 97/189 122/377 0.51 (0.39–0.67) <0.001 M0 18/74 13/150 0.34 (0.16–0.69) 0.003 M1 79/115 109/227 0.56 (0.42–0.75) <0.001 0.169 Progression-free survivalb All 72/189 103/377 0.65 (0.48–0.88) 0.005 M0 10/74 9/150 0.42 (0.17–1.05) 0.064 M1 62/115 94/227 0.69 (0.50–0.95) 0.023 0.323 Metastatic progression-free survivalc All 71/189 118/377 0.77 (0.57–1.03) 0.079

Outcome measure Patient group Events/Pts SOC + DocP Events/Pts SOC + AAP Hazard ratioa (95% CI) P-value Interaction by metastases P-value Failure-free survivalb All 97/189 122/377 0.51 (0.39–0.67) <0.001 M0 18/74 13/150 0.34 (0.16–0.69) 0.003 M1 79/115 109/227 0.56 (0.42–0.75) <0.001 0.169 Progression-free survivalb All 72/189 103/377 0.65 (0.48–0.88) 0.005 M0 10/74 9/150 0.42 (0.17–1.05) 0.064 M1 62/115 94/227 0.69 (0.50–0.95) 0.023 0.323 Metastatic progression-free survivalc All 71/189 118/377 0.77 (0.57–1.03) 0.079 M0 10/74 18/150 0.91 (0.42–2.01) 0.824 M1 61/115 100/227 0.76 (0.55–1.04) 0.085 0.744 Freedom from symptomatic skeletal events All 36/189 63/377 0.83 (0.55–1.25) 0.375 M0 2/74 5/150 1.28 (0.24–6.67) 0.771 M1 34/115 58/227 0.82 (0.53–1.25) 0.351 0.648 Overall survival All 44/189 105/377 1.16 (0.82–1.65) 0.404 M0 6/74 16/150 1.51 (0.58–3.93) 0.395 M1 38/115 89/227 1.13 (0.77–1.66) 0.528 0.691 Outcome measure Patient group Events/Pts SOC+Doc Events/Pts SOC+AAP Sub-hazard ratiod(95% CI) P-value Interaction by metastases P-value Death from prostate cancere All 40/189 86/377 1.02 (0.70–1.49) 0.916 M0 4/74 6/150 0.82 (0.24–2.81) 0.751 M1 36/115 80/227 1.05 (0.71–1.56) 0.807 0.620 Death from other causesf All 4/189 19/377 2.33 (0.77–6.99) 0.131 M0 2/74 10/150 3.00 (0.66–13.66) 0.155 M1 2/115 9/227 1.91 (0.43–8.41) 0.393 0.771 a From Cox proportional hazards model, adjusted for stratification factors at randomisation (except hospital and choice of hormone therapy) and stratified by time period. b Includes death from prostate cancer. c Includes death from any cause.

M0 2/74 10/150 3.00 (0.66–13.66) 0.155 M1 2/115 9/227 1.91 (0.43–8.41) 0.393 0.771 a From Cox proportional hazards model, adjusted for stratification factors at randomisation (except hospital and choice of hormone therapy) and stratified by time period. b Includes death from prostate cancer. c Includes death from any cause. d From competing risks regression model, adjusted for stratification factors at randomisation (except hospital and choice of hormone therapy) and time period, and treating causes of death other than the focus as a competing event. e Cause attributed on central death review; prostate cancer death as event, other cause of death as competing event. f Cause attributed on central death review; other causes of death as event, prostate cancer as competing event. Figure 3. Efficacy analysis—survival, metastases-free survival, failure-free survival, skeletal-related events. Kaplan–Meier (survival) plots for the key efficacy outcome measures. Each step down the y-axis represents an event. The number of patients contributing information (at risk) over time since randomisation is shown under the table. The number of patients with an event between these points is shown in brackets. The number of patients censored in a time window is not shown, but is calculable as the difference between the number of patients at risk at two times points and the number of patients with events, e.g. in Figure 3E between 0 and 6 months on the SOC+AAP arm (377−362)−12=3 patients are censored.

ts is shown in brackets. The number of patients censored in a time window is not shown, but is calculable as the difference between the number of patients at risk at two times points and the number of patients with events, e.g. in Figure 3E between 0 and 6 months on the SOC+AAP arm (377−362)−12=3 patients are censored. Totally, 126/149 deaths were attributed to prostate cancer, comprising 10/22 and 116/127 deaths in patients with M0 and M1 disease at entry, respectively. Competing risks regression shows no evidence of a difference in prostate cancer-specific survival (sub-HR = 1.02, 95% CI 0.70–1.49). For non-prostate cancer-specific survival, with 23/149 deaths attributed to other causes, the sub-HR was 2.33 (95% CI 0.78–6.99). There was no evidence of heterogeneity of treatment effect by baseline metastases in either outcome. Other efficacy OMs Table 2 shows the effect size overall and by whether the patients had metastases at entry for FFS, PFS, MPFS and skeletal-related events. There is no evidence of heterogeneity of the treatment effect by baseline metastases in any of these OMs. Figure 4 summarises the effect for all OMs. Figure 4. Depiction of disease state over time.

Other efficacy OMs Table 2 shows the effect size overall and by whether the patients had metastases at entry for FFS, PFS, MPFS and skeletal-related events. There is no evidence of heterogeneity of the treatment effect by baseline metastases in any of these OMs. Figure 4 summarises the effect for all OMs. Figure 4. Depiction of disease state over time. Safety The safety population includes people who started their allocated treatment. While nearly all patients allocated to AAP started it, a proportion of those patients allocated to receive docetaxel declined to start it. Table 3 summarises the worst toxicity reported for patients over their time on trial in the safety population and shows differing patterns for adverse events according to treatment. The prevalence of grade 3 or 4 toxicity in patients with assessments at 1 year without a prior FFS event was 11% SOC + DocP and 11% SOC + AAP; at 2 years this was 11% SOC + DocP and 11% SOC + AAP. Table 3. Worst adverse event (grade) reported over entire time on trial

der 9 (5) 28 (8) Hepatic disorder 1 (1) 32 (9) Increased AST 0 (0) 6 (2) Increased ALT 1 (1) 23 (6) Respiratory disorder 12 (7) 11 (3) Dyspnoea 4 (2) 1 (1) Renal disorder 5 (3) 20 (5) Lab abnormalities 9 (5) 11 (3) Hypokalaemia 0 (0) 3 (1) a The safety population includes patients who started their allocated treatment. Second-line treatment Figure 5 shows time from randomisation to any subsequent exposure to docetaxel or AR-targeted therapy with AAP or enzalutamide. Figure 6 shows time from an FFS event to reported exposure to selected treatments that are licensed for CRPC: docetaxel, AAP, enzalutamide. There was limited reported use of cabazitaxel, radium and sipuleucel-T at this point (not shown). Figure 5. Time from randomisation to reported starting docetaxel, AAP, enzalutamide or AR-targeting therapy. Kaplan–Meier (survival) plots showing cumulative incidence of exposure to treatments after randomisation. Each step up the y-axis represents an event, namely starting that particular treatment. The number of patients contributing information (at risk) over time since randomisation is shown under the table. The number of patients with an event between these points is shown in brackets. For example, in Figure 4C between 24 and 36 months after randomisation, 4 patients on the SOC+DocP arm report starting abiraterone and (150−129)−4 are 17 are censored and may start in the future.

ince randomisation is shown under the table. The number of patients with an event between these points is shown in brackets. For example, in Figure 4C between 24 and 36 months after randomisation, 4 patients on the SOC+DocP arm report starting abiraterone and (150−129)−4 are 17 are censored and may start in the future. Figure 6. Time from failure-free survival event to subsequent treatment by allocated treatment. Kaplan–Meier (survival) plots showing cumulative incidence of exposure to treatments after a failure-free survival (FFS) event. Doc, docetaxel; AAP, abiraterone acetate + prednisolone; Enz, enzalutamide. Each step up the y-axis represents an event, namely starting that particular treatment.

ment by allocated treatment. Kaplan–Meier (survival) plots showing cumulative incidence of exposure to treatments after a failure-free survival (FFS) event. Doc, docetaxel; AAP, abiraterone acetate + prednisolone; Enz, enzalutamide. Each step up the y-axis represents an event, namely starting that particular treatment. Discussion We and others have previously shown a survival advantage for adding docetaxel (with or without prednisolone/prednisone) and for adding abiraterone acetate and prednisolone/prednisone, in patients starting long-term hormone therapy for the first time [4–11]. However, there is currently no direct evidence available to help clinicians or patients assess which combination might be better. Here, we reported a pre-specified (but not pre-powered) analysis using only patients who were randomised during a period of the study when recruitment to the two research arms overlapped. We used data collected prospectively from over 100 sites across two countries as part of a clinical trial protocol. The MAMS platform design of STAMPEDE, an approach sometimes referred to as a master protocol [16], facilitated this comparison. Separate, traditional, two-arm RCTs, would not have allowed any directly randomised comparative evidence to be available so soon.

tes across two countries as part of a clinical trial protocol. The MAMS platform design of STAMPEDE, an approach sometimes referred to as a master protocol [16], facilitated this comparison. Separate, traditional, two-arm RCTs, would not have allowed any directly randomised comparative evidence to be available so soon. Our recently reported overall treatment effect on survival, in STAMPEDE, for adding AAP compared with the SOC (HR = 0.63) [10] was larger than the previously-reported overall treatment effect, in STAMPEDE, on survival for adding DocP to the same SOC (HR = 0.78) [7]. The earlier secondary efficacy OMs favoured adding AAP over DocP, including FFS—perhaps unsurprising given the direct antiandrogenic action of AAP (around four in every five FFS events was driven only by a rise in PSA) and PFS (which excludes rising PSA). There was weak evidence favouring AAP for MPFS and no evidence of a difference in symptomatic skeletal events, prostate cancer-specific survival or OS. Comparing the results indirectly of these two therapies by readers extracting data from STAMPEDE’s AAP and docetaxel papers [7, 10] may not be the most appropriate way to compare the relative effectiveness: the patient cohorts were all not randomised contemporaneously and there may be confounding biases when comparing the two datasets, in particular, many DocP patients had very limited salvage CRPC options compared with AAP patients, simply due to the timing of licences of new therapies (see below).

elative effectiveness: the patient cohorts were all not randomised contemporaneously and there may be confounding biases when comparing the two datasets, in particular, many DocP patients had very limited salvage CRPC options compared with AAP patients, simply due to the timing of licences of new therapies (see below). Importantly, the two therapies are being used in different ways. AAP is used until the patient has castrate-resistant prostate cancer (CRPC), often lasting many years and consequently exhausting a major therapy option for CRPC. In contrast, DocP is given as an 18-week course thus all CRPC options should remain available. Our data reveal important differences in the pattern of treatment failure yet we do not see any differences in survival, suggesting that the relative time spent before and after first-line treatment failure are quite different by initial treatment. This may explain why the early, often biochemically driven OMs, favour AAP but the later post CRPC end points such as skeletal events, prostate cancer-specific survival and OS show no good evidence of a difference. Men receiving DocP will thus spend longer with CRPC than men receiving AAP but with a broader range of more effective options available. Supplementary Figure S1, available at Annals of Oncology online, shows the status of all patients at each moment in time after randomisation. That the DocP cohort had more durable survival after failure, perhaps longer than before failure, may be important in counselling patients’ biochemically failing after DocP.

ble. Supplementary Figure S1, available at Annals of Oncology online, shows the status of all patients at each moment in time after randomisation. That the DocP cohort had more durable survival after failure, perhaps longer than before failure, may be important in counselling patients’ biochemically failing after DocP. The number of events is an important consideration in time-to-event analyses. The number of patients with metastases at baseline was balanced by arm, but, particularly because of their poorer prognosis, these patients tend to predominate in this analysis. There is no evidence of heterogeneity in the treatment effect by baseline metastasis for any of the OMs, but power to detect any heterogeneity is very limited, especially in later OMs with fewer events.

ced by arm, but, particularly because of their poorer prognosis, these patients tend to predominate in this analysis. There is no evidence of heterogeneity in the treatment effect by baseline metastasis for any of the OMs, but power to detect any heterogeneity is very limited, especially in later OMs with fewer events. The patterns of toxicity are quite different for the two treatment approaches, consistent with the known effects of the drugs. The proportion of patients reporting at least one grade 3 or worse toxicity was similar and in line with previously reported toxicities for these agents (Table 3). In patients who started their allocated treatment and who are without disease progression at 1 year, the prevalence of grade 3 or worse toxicity was about 11% on both arms and very similar to our previous estimate for SOC. Nearly all patients started their allocated abiraterone, whereas about 1 in 12 patients did not start their allocated docetaxel. Our results may change future compliance with both treatments in routine practice; but the lack of compliance with allocated treatment of docetaxel is likely to have had some impact on our estimated effect sizes.

tarted their allocated abiraterone, whereas about 1 in 12 patients did not start their allocated docetaxel. Our results may change future compliance with both treatments in routine practice; but the lack of compliance with allocated treatment of docetaxel is likely to have had some impact on our estimated effect sizes. A key limitation is that the comparison was opportunistic and not designed in the usual way, hence power is limited to detect any realistic differences. The trigger for the analysis was the reporting of our ‘abiraterone comparison’ data [10]. The unequal allocation ratio reflects the planned design of the comparisons. The allocated treatment being given was not masked for practical reasons. This, of course, allowed for relapse therapies to be given at the investigator’s discretion. We observed that after relapse, many patients received the treatment class that they had not received up-front.

ts the planned design of the comparisons. The allocated treatment being given was not masked for practical reasons. This, of course, allowed for relapse therapies to be given at the investigator’s discretion. We observed that after relapse, many patients received the treatment class that they had not received up-front. Salvage options have changed over time: men recruited earlier on to DocP (2005–2013) will have had very different options to those recruited later to AAP (2011–2014) when there were more CRPC therapies likely available, including AAP [17, 18], cabazitaxel [19], docetaxel [20, 21], enzalutamide [22, 23], radium-223 [24] and sipuleucel-T [25] (although not widely accessible in Europe). For this analysis, we limited ourselves to patients contemporaneously randomised to either arm to make this comparison as fair as possible. However, FFS events generally happened sooner with DocP than with AAP in time from randomisation and, therefore, calendar year (Table 4) may partially influence outcomes. Furthermore, a FFS event was more of an indication to change treatments on DocP; AAP continued beyond this point. Table 4. Year of FFS event and death by arm Year of event FFS event Death SOC + DocP SOC + AAP SOC + DocP SOC + AAP N % N % N % N % 2012 14 7 25 6 1 1 5 1 2013 38 20 43 11 12 6 18 5 2014 25 13 33 9 9 5 33 9 2015 14 7 11 3 16 8 38 10 2016 6 3 10 3 6 3 11 3 No event 92 49 255 68 145 77 272 72

Salvage options have changed over time: men recruited earlier on to DocP (2005–2013) will have had very different options to those recruited later to AAP (2011–2014) when there were more CRPC therapies likely available, including AAP [17, 18], cabazitaxel [19], docetaxel [20, 21], enzalutamide [22, 23], radium-223 [24] and sipuleucel-T [25] (although not widely accessible in Europe). For this analysis, we limited ourselves to patients contemporaneously randomised to either arm to make this comparison as fair as possible. However, FFS events generally happened sooner with DocP than with AAP in time from randomisation and, therefore, calendar year (Table 4) may partially influence outcomes. Furthermore, a FFS event was more of an indication to change treatments on DocP; AAP continued beyond this point. Table 4. Year of FFS event and death by arm Year of event FFS event Death SOC + DocP SOC + AAP SOC + DocP SOC + AAP N % N % N % N % 2012 14 7 25 6 1 1 5 1 2013 38 20 43 11 12 6 18 5 2014 25 13 33 9 9 5 33 9 2015 14 7 11 3 16 8 38 10 2016 6 3 10 3 6 3 11 3 No event 92 49 255 68 145 77 272 72 As far as we are aware there are no ongoing randomised trials directly comparing adding AAP versus adding docetaxel for patients starting long-term ADT. All of our published STAMPEDE data have contributed to the STOpCaP aggregate data network meta-analysis that has used all of the reported RCTs in metastatic patients to perform indirect comparisons and allow some assessment of potential ranking of effective therapies. This aggregate data analysis (co-submitted) will be supplemented by a forthcoming individual patient data (IPD) network meta-analysis which will hopefully provide a more accurate reflection of the temporal interval between the application of the two different therapies, to which STAMPEDE will contribute all relevant data. We will continue to follow-up patients for long-term OMs.

ted by a forthcoming individual patient data (IPD) network meta-analysis which will hopefully provide a more accurate reflection of the temporal interval between the application of the two different therapies, to which STAMPEDE will contribute all relevant data. We will continue to follow-up patients for long-term OMs. Considering their mechanisms of action and their proven oncological benefits, the question is raised of whether a combination of AAP plus docetaxel might lead to an approximately additive benefit of using them both, further extending survival. Randomised data on docetaxel with or without abiraterone will emerge from a subset the PEACE-1 trial (https://clinicaltrials.gov/ct2/show/NCT01957436), as will non-randomised, time-stratified data on abiraterone with or without docetaxel. Similarly comparative data will also emerge for enzalutamide, another AR-targeted therapy, from the ENZAMET trial (https://clinicaltrials.gov/ct2/show/NCT02446405) and with the combination of enzalutamide and AAP in STAMPEDE (Figure 1). In conclusion, there are now two systemic therapies, DocP and AAP, which have shown a survival benefit from RCTs when added to treatment of patients starting long-term ADT for the first time. The evidence from our directly randomised data comparing these two therapies showed no evidence of a difference in overall or prostate cancer-specific survival, nor in other important outcomes such as symptomatic skeletal events, suggesting that both currently remain viable new standards-of-care. Supplementary Material Supplementary Data Click here for additional data file.

In conclusion, there are now two systemic therapies, DocP and AAP, which have shown a survival benefit from RCTs when added to treatment of patients starting long-term ADT for the first time. The evidence from our directly randomised data comparing these two therapies showed no evidence of a difference in overall or prostate cancer-specific survival, nor in other important outcomes such as symptomatic skeletal events, suggesting that both currently remain viable new standards-of-care. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements Independent oversight committee members Independent Data Monitoring Committee: John Yarnold (chair), Doug Altman, Ronald de Wit, Bertrand Tombal; Previous—Reg Hall, Chris Williams Trial Steering Committee: Jonathan Ledermann (chair), Jan Erik Damber, Richard Emsley, Alan Horwich; Previous—John Fitzpatrick, David Kirk, Jim Paul Participating site list Structure: City, Hospital (Number of patients by data freeze: site PI; other investigators) UK Aberystwyth, Bronglais General Hospital (4: Porfiri; Durrani) Ashford William Harvey Hospital (19: Thomas; Mithal) Aylesbury, Stoke Mandeville Hospital (14: Sabharwal; Camilleri) Ayr Hospital (54: Glen; Ansari) Barnet General Hospital (25: McGovern; Eichholz) Basingstoke & N Hampshire Hospital (21: Shaffer) Bath, Royal united Hospital (70: Frim; Beresford) Belfast City (191: O'Sullivan; Mitchell, Stewart, Shum) Birmingham, City Hospital (26: Sivoglo; Ford) Birmingham, Good Hope Hospital (18: Ford) Birmingham, Heartlands Hospital (38: Zarkar) Birmingham, QE (180: James; Porfiri, Ford)

Barnet General Hospital (25: McGovern; Eichholz) Basingstoke & N Hampshire Hospital (21: Shaffer) Bath, Royal united Hospital (70: Frim; Beresford) Belfast City (191: O'Sullivan; Mitchell, Stewart, Shum) Birmingham, City Hospital (26: Sivoglo; Ford) Birmingham, Good Hope Hospital (18: Ford) Birmingham, Heartlands Hospital (38: Zarkar) Birmingham, QE (180: James; Porfiri, Ford) Blackburn East Lancashire Trust (180: Parikh; Charnley) Bolton, Royal Bolton Hospital (30: Elliott, Maddineni) Boston, Pilgrim Hospital (38: Sreenivasan; Panades) Bournemouth, Royal Bournemouth Hospital (100: Brock) Bradford Royal Infirmary (36: Brown) Brighton, Royal Sussex County Hospital (92: Robinson; Robinson, Bloomfield) Bristol Haematology & Oncology Centre (106: Bahl; Herbert, Masson) Burton, Queen's Hospital (108: Smith-Howell; Chetiyawardana, Pattu) Bury St Edmunds, West Suffolk Hospital (21: Woodward) Cardiff, Velindre (341: Lester; Staffurth, Barber, Kumar, Palaniappan, Button, Tanguay) Chelmsford, Broomfield Hospital (88: Hamid; Panwar, Leone) Cheltenham General Hospital (54: Bowen) Chester, Countess of Chester Hospital (79: Ibrahim) Coventry & Warwickshire, University Hospital (40: Worlding; Stockdale) Crewe, Leighton Hospital (54: Wylie) Cumbria, Cumberland Infirmary (18: Kumar) Darlington Memorial Hospital (49: Kagzi; Hardman, Peedell) Derby, Royal Derby Hospital (130: Chakraborti; Pattu) Devon, North Devon District Hospital (33: Sheehan) Doncaster Royal Infirmary (35: Bowen; Ferguson) Dorset County Hospital (30: Crellin; Afzal, Andrews) Dudley, Russells Hall Hospital (81: Keng-Koh; Ramachandra) Durham University Hospital (17: Heath; McMenemin)

Darlington Memorial Hospital (49: Kagzi; Hardman, Peedell) Derby, Royal Derby Hospital (130: Chakraborti; Pattu) Devon, North Devon District Hospital (33: Sheehan) Doncaster Royal Infirmary (35: Bowen; Ferguson) Dorset County Hospital (30: Crellin; Afzal, Andrews) Dudley, Russells Hall Hospital (81: Keng-Koh; Ramachandra) Durham University Hospital (17: Heath; McMenemin) Eastbourne District General Hospital (63: McKinna) Edinburgh, Western General (112: McLaren) Essex County Hospital (58: Muthukumar; Sizer, Kumar) Exeter, Royal Devon & Exeter (189: Sheehan; Srinivasan) Gillingham, Medway Hospital (29: Kumar; Taylor) Glasgow, Beatson West of Scotland Cancer Centre (323: Graham; Venugopal, Wallace, Jones, Lamb, Glen, Russell) Guildford, Royal Surrey County Hospital (132: Laing; Khaksar, Wood, Money-Kyrle) Harlow, Princess Alexandra Hospital (54: Gupta; Melcher, Melcher) Hereford County Hospital (71: Grant; Cook) Huddersfield Royal Infirmary (105: Hofmann) Hull, Castle Hill Hospital (119: Simms; Hetherington) Inverness, Raigmore Hospital (88: McPhail; MacGregor) Ipswich Hospital (103: Brierly; Venkitaraman, Scrase) Keighley, Airedale Hospital (52: Brown; Crawford) Kent and Canterbury Hospital (79: Thomas; Raman, Mithal, Malde) Kent, Queen Elizabeth Queen Mother Hospital (27: Thomas; Raman) Kidderminster General Hospital (40: Capaldi; Churn) Larbert, Forth Valley Royal Hospital (36: Sidek) Leeds, St James University Hospital (94: Cross; Loughrey, Bottomley, Prescott) Lincoln County Hospital (50: Sreenivasan; Ballesteros-Quintail, Panades, Baria) Liverpool, Royal Liv University Hospital (88: Malik; Robson, Eswar)

Kent, Queen Elizabeth Queen Mother Hospital (27: Thomas; Raman) Kidderminster General Hospital (40: Capaldi; Churn) Larbert, Forth Valley Royal Hospital (36: Sidek) Leeds, St James University Hospital (94: Cross; Loughrey, Bottomley, Prescott) Lincoln County Hospital (50: Sreenivasan; Ballesteros-Quintail, Panades, Baria) Liverpool, Royal Liv University Hospital (88: Malik; Robson, Eswar) Liverpool, University Hospital Aintree (26: Robson) London, Charing Cross Hospital (38: Falconer; Mangar) London, Guy's Hospital (161: Chowdhury) London, Hammersmith Hospital (4: Falconer; Mangar) London, North Middlesex Hospital (24: Gupta; Newby, Thompson) London, Royal Free Hospital (44: Vilarino-Varela; Pigott) London, St Georges Hospital (35: Pickering) London, St Mary's Hospital (8: Falconer; Stewart) London, University College Hospital (46: McGovern) Maidstone, Kent Oncology Centre (114: Beesley) Manchester Christie Hospital (167: Clarke; Elliott, Livsey, Choudhury, Wylie) Manchester Hope Hospital (59: Clarke; Elliott, Lau, Tran) Manchester, Royal Oldham Hospital (54: Conroy; Livsey, Choudhury) Manchester, Withington Hospital (7: Sangar) Middlesbrough, James Cook UH (103: Peedell; Van der Voet, Hardman, Shakespeare) Newcastle, Freeman Hospital (92: Azzabi; McMenemin, Frew) North Staffordshire UH (80: Adab) Northwood, Mount Vernon Hospital (126: Hoskin; Anyamene, Ostler, Alonzi) Nottingham University Hospitals (City Campus) (141: Sundar; Mills) Nuneaton, George Eliot Hospital (14: Khan; Chan) Oxford, Churchill Hospital (165: Protheroe; Cole, Sabharwal, Sugden) Poole Hospital (62: Davies) Portsmouth, Q Alexandra Hospital (173: Gale)

Northwood, Mount Vernon Hospital (126: Hoskin; Anyamene, Ostler, Alonzi) Nottingham University Hospitals (City Campus) (141: Sundar; Mills) Nuneaton, George Eliot Hospital (14: Khan; Chan) Oxford, Churchill Hospital (165: Protheroe; Cole, Sabharwal, Sugden) Poole Hospital (62: Davies) Portsmouth, Q Alexandra Hospital (173: Gale) Preston, Royal Preston Hospital (221: Birtle; Parikh, Wise) Reading, Royal Berkshire Hospital (42: Rogers; O'Donnell, Brown, Brown) Redditch, Alexandra Hospital (15: Capaldi; Hamilton) Romford, Queen's Hospital (127: Gibbs; Subramaniam) Scarborough General Hospital (82: Hingorani) Sheffield, Weston Park (142: Ferguson) Shrewsbury, Royal Shrewsbury Hospital (192: Srihari) Somerset, Weston General Hospital (18: Hilman) Southampton General Hospital (75: Jones; Heath, Wheater, Crabb) Southend University Hospital (114: Tsang; Ahmed, Chan) Southport and Formby District GH (46: Bhalla; Sivapalasuntharam, Sivapalasuntharam) St Leonards-on-Sea, Conquest Hospital (42: McKinna; Beesley, Lees) Stevenage, Lister Hospital (35: Hughes) Stockport, Stepping Hill Hospital (106: Logue; Coyle) Stockton-on-Tees, UH North Tees (28: Leaning; Shakespeare) Sunderland Royal Hospital (45: Azzabi) Sutton-in-Ashford, King's Mill Hospital (64: Saunders) Sutton and London, Royal Marsden Hospital (162: Dearnaley; Parker, Selvadurai) Swansea, Singleton (188: Wagstaff; Phan, Phan) Swindon, Great Western Hospital (52: Khan; Cole) Taunton, Musgrove Park Hospital (137: Gray; Graham, Varughese, Plataniotis) Torbay District General Hospital (135: Lydon; Srinivasan) Tyne & Wear, S Tyneside District Hospital (6: Azzabi) Warrington Hospital (111: Syndikus; Tolan)

Sutton and London, Royal Marsden Hospital (162: Dearnaley; Parker, Selvadurai) Swansea, Singleton (188: Wagstaff; Phan, Phan) Swindon, Great Western Hospital (52: Khan; Cole) Taunton, Musgrove Park Hospital (137: Gray; Graham, Varughese, Plataniotis) Torbay District General Hospital (135: Lydon; Srinivasan) Tyne & Wear, S Tyneside District Hospital (6: Azzabi) Warrington Hospital (111: Syndikus; Tolan) Warwick Hospital (17: Chan; Stockdale) Wigan, Royal Albert Edward Infirmary (37: Tran) Wirral, The Clatterbridge Cancer Centre NHS Foundation Trust (128: Tolan; Syndikus, Ibrahim, Montazeri, Littler) Wolverhampton, New Cross Hospital (53: Gray; Sayers) Woolwich, Queen Elizabeth Hospital (18: Hughes) Worcestershire Royal Hospital (57: Capaldi; Bowen) Worthing Hospital (90: Nikapota) Wycombe Hospital (52: Sabharwal; Protheroe, Pwint) Switzerland Basel Universitatsspital (5: Rentsch) Berne University Hospital (Inselspital) (5: Thalmann) Chur Kantonsspital Graubunden (31: Strebel; Cathomas) Kantonsspital St Gallen (10: Engeler) Lausanne, Centre Hospital Univ Vaudois (7: Berthold; Jichlinski) Plus more than 3000 local site team staff across these hospitals. Trials Unit Staff (from 2011 onwards) MRC Clinical Trials Unit at UCL Statisticians—Matthew Sydes, Max Parmar, Melissa Spears, Chris Brawley; Previously—Gordana Jovic, Rachel Jinks, Patrick Royston, Sophie Barthel, Babak Choodari-Oskooei, Daniel Bratton, Andrew Embleton

Plus more than 3000 local site team staff across these hospitals. Trials Unit Staff (from 2011 onwards) MRC Clinical Trials Unit at UCL Statisticians—Matthew Sydes, Max Parmar, Melissa Spears, Chris Brawley; Previously—Gordana Jovic, Rachel Jinks, Patrick Royston, Sophie Barthel, Babak Choodari-Oskooei, Daniel Bratton, Andrew Embleton Project and Trial Managers—Claire Amos, Nafisah Atako; Claire Murphy, Joanna Calvert, Mazna Anjum, Chris Wanstall, Arlen Wilcox; Previously—Sharon Naylor, Neil Kelk, James Latham, Jacqui Nuttall, Karen Sanders, Tom Fairfield, Charlene Green, Francesca Schiavone, Katie Ward, Mazna Anjum, Anna Herasimtschuk, Jenny Petrie, Alanna Brown, Orla Prendiville Data Managers—Carly Au, Danielle Johnson, Lina Bergstrom, Tasmin Philips; Previously—Emma Donoghue, Tim Smith, Jacque Millett, Shama Hassan, Philip Pollock, Richard Gracie, Laura Van Dyck, Charlene Green, Elizabeth Clark, Sara Peres, Hannah Gardner, Dominic Hague, Katie Ward, Peter Vaughan, Eva Ades, Hannah Babiker, Zohrah Khan, Nargis Begum, Saba Khan, Jenna Grabey Data Scientists and Programmers—Nadine Van Looy, Zaheer Islam, Dominic Hague; Previously—Lindsey Masters, Will Cragg, Sajad Khan Clinicians—Clare Gilson, Alastair Ritchie; Previously—Sarah Meredith, Ruth Langley Trial Assistants—Stephanie Wetton, Amy Fiddament; Previously—Leigh Dobson, Alexandra Wadia, Nat Thorogood, Shanaz, Sohail, Tracey Fisher, Andrew Whitney Swiss Group for Cancer Clinical Research Project and Trial Managers—Corinne Schar; Previously—Estelle Cassolly Patient and Public Involvement representatives—David Matheson, Robin Millman

Trial Assistants—Stephanie Wetton, Amy Fiddament; Previously—Leigh Dobson, Alexandra Wadia, Nat Thorogood, Shanaz, Sohail, Tracey Fisher, Andrew Whitney Swiss Group for Cancer Clinical Research Project and Trial Managers—Corinne Schar; Previously—Estelle Cassolly Patient and Public Involvement representatives—David Matheson, Robin Millman Funding The trial was sponsored by the UK Medical Research Council (MRC) and conducted by the MRC Clinical Trials Unit at UCL. In the UK the trial was supported by the UK Clinical Research Network, and funded by CRUK and the MRC, and in Switzerland, by the Swiss Group for Cancer Clinical Research (SAKK). Industry collaboration and support has been provided to STAMPEDE by Astellas, Clovis Oncology, Janssen, Novartis, Pfizer and Sanofi-Genzyme. MRC employees were central to the conduct of the trial and the development of this manuscript. Authors MRSy and MRSp accessed raw data. The funding bodies had no role in determining this publication. Research support for trial: Cancer Research UK (CRUK_A12459), Medical Research Council (MRC_MC_UU_12023/25); Janssen, Sanofi-Aventis; Astellas, Clovis Oncology, Novartis, Pfizer. DPD, JSdeB, GA and CCP acknowledge NHS funding to the NIHR Biomedical Research Centre at the Royal Marsden NHS Foundation Trust and Institute of Cancer Research.

Research support for trial: Cancer Research UK (CRUK_A12459), Medical Research Council (MRC_MC_UU_12023/25); Janssen, Sanofi-Aventis; Astellas, Clovis Oncology, Novartis, Pfizer. DPD, JSdeB, GA and CCP acknowledge NHS funding to the NIHR Biomedical Research Centre at the Royal Marsden NHS Foundation Trust and Institute of Cancer Research. Disclosure CA reports grants and non-financial support from Sanofi-Aventis, Novartis, Pfizer, Janssen, Astellas and Clovis Oncology during the conduct of the study. GA reports personal fees, grants and/or travel support from Janssen during the conduct of the study; personal fees and/or travel support from Astellas, Pfizer, Janssen, Millennium Pharmaceuticals, Ipsen, Ventana, Veridex, Novartis, Abbott Laboratories, ESSA Pharmaceuticals, Bayer Healthcare Pharmaceuticals, Takeda and Sanofi-Aventis and grant support from AstraZeneca, Innocrin Pharma and Arno Therapeutics, outside the submitted work; in addition, GA’s former employer, The Institute of Cancer Research, receives royalty income from abiraterone and GA receives a share of this income through the ICR’s Rewards to Discoverers Scheme. AB reports other from Astellas, personal fees and other from Sanofi, from Janssen, during the conduct of the study; other from Bayer, other from Astra Zeneca, outside the submitted work. SB reports other from Janssen, outside the submitted work; and attendance at ESMO 2018 funded by Janssen. PC reports grants from Janssen, during the conduct of the study. AC reports funding from Prostate Cancer UK, Cancer Research UK, National Institute of Health Research, Medical Research Council and Astra Zeneca, outside the submitted work. SC reports grants and personal fees from Sanofi-Aventis and personal fees from Janssen Pharmaceutical, outside the submitted work. NWC reports personal fees from Janssen Pharmaceuticals, during the conduct of the study; personal fees from Janssen Pharmaceuticals, outside the submitted work; personal fees from Bayer and Astellas. WC reports personal fees from Janssen and Bayer, outside the submitted work. JSdB reports other from ICR and Janssen, during the conduct of the study; other from AstraZeneca, Pfizer, GlaxoSmithKline, Taiho, Daiichi, Novartis, Genmab, Merck Serano, Merck and Genentech/Roche, outside the submitted work.

yer and Astellas. WC reports personal fees from Janssen and Bayer, outside the submitted work. JSdB reports other from ICR and Janssen, during the conduct of the study; other from AstraZeneca, Pfizer, GlaxoSmithKline, Taiho, Daiichi, Novartis, Genmab, Merck Serano, Merck and Genentech/Roche, outside the submitted work. DPD reports other from UK National Institute for Health Research Clinical Research Network (NIHR CRN), during the conduct of the study; grants from Cancer Research UK; personal fees and other from Takeda, Amgen, Astellas and Sandoz, personal fees, non-financial support and other from Janssen, personal fees and other from Cadence Research, other from Clovis, personal fees and non-financial support from ISSECAM, outside the submitted work; in addition, DPD has a patent GB9305269-17-substituted steroids useful in cancer treatment with royalties paid to Janssen Pharmaceutical Company. DF reports honoraria from Janssen, Novartis and Sanofi, outside the submitted work. JDG reports other support as a local principal investigator for a study of radium-223 in prostate cancer funded by Bayer, and other support as a local principal investigator for a study of LHRH antagonist in prostate cancer funded by Millennium Pharmaceuticals, outside the submitted work. SG reports personal fees from Bayer, other from Bayer and CureVac, personal fees from Janssen Cilag, other·from Janssen Cilag, personal fees from Dendreon Corporation, other from Astellas, personal fees from Millennium Pharmaceuticals, personal fees from Orion, Sanofi and MaxiVax SA, other from AAA Advanced Accelerator Applications International, Bristol-Myers Squibb, Ferring, Roche, Orion, lnnocrin Pharmaceuticals, Sanofi, Novartis, Nektar Therapeutics and ProteoMedix, outside the submitted work. CG reports grants from Clovis Oncology, outside the submitted work. NDJ reports grants and personal fees from Sanofi and Novartis, during the conduct of the study; grants and personal fees from Janssen, Astellas and Bayer, outside the submitted work. RJJ reports grants from Sanofi, and grants and non-financial support from Novartis, during the conduct of the study; grants, personal fees and non-financial support from Sanofi and Novartis, grants and personal fees from Janssen, Astellas and Bayer, outside the submitted work. JL reports personal fees, non-financial support and other from Janssen, Astellas and Sanofi, outside the submitted work.

rtis, during the conduct of the study; grants, personal fees and non-financial support from Sanofi and Novartis, grants and personal fees from Janssen, Astellas and Bayer, outside the submitted work. JL reports personal fees, non-financial support and other from Janssen, Astellas and Sanofi, outside the submitted work. ZIM reports and Consultancy and advisory boards Janssen Consultancy and advisory boards Sanofi Advisory board Astellas Sponsorship to attend medical conferences Astellas, Bayer and Janssen. MDM reports personal fees from Sanofi, Bayer, Dendreon, Bristol-Myers and Janssen, outside the submitted work. DM reports support from Astellas and personal speaker fees from Bayer, outside the submitted work. JS reports support for travel and speakers fees for the following companies in the field of prostate cancer, not related to this study: Janssen Bayer and Astellas. CCP reports personal fees from AAA and Janssen, research funding and speaker's honoraria from Bayer, outside the submitted work. MKBP reports grants and non-financial support from Janssen, during the conduct of the study; grants and non-financial support from Astellas, Clovis Oncology, Novartis, Pfizer and Sanofi, outside the submitted work. AP reports personal fees from Ipsen, Bayer, Roche and BMS, grants from Merck, personal fees from Merck, outside the submitted work. JMR reports personal fees from Janssen (lecture fee), outside the submitted work. DS reports conference travel costs from Ipsen and Astellas, outside the submitted work. MRSp reports grants and non-financial support from Sanofi-Aventis, Novartis, Pfizer, Janssen and Astellas, during the conduct of the study. SS reports personal fees and non-financial support from Sanofi-Aventis, outside the submitted work. MRSy reports grants and non-financial support from Sanofi-Aventis, Novartis, Pfizer, Janssen and Astellas, during the conduct of the study; and personal fees from Eli-Lilly, outside the submitted work. ST reports other from Sanofi, other support from Astellas, personal fees from Astellas and other support from Janssen, outside the submitted work. MV reports travel grants from Janssen. JW reports a paid consultancy for Janssen. All remaining authors have declared no conflicts of interest.

Key Message Complex chromosomal rearrangements in breast cancer recur across patients in 21 genomic locations. These ‘hotspots’ contain known oncogenic drivers and putative new driver loci. Detailed analysis of rearrangements at these hotspots highlights chromosomal aberrations likely driven by selection and analysis reveals the underlying mutational processes. Background Extensive copy number characterisation using comparative genomic hybridisation (CGH) technology has led to remarkable insights into the somatic genetics of breast cancer, including identification of recurrent whole arm gains and losses, homozygous deletions (e.g. CDKN2A/B, PTEN) and large, common, recurrent driver amplifications (e.g. ERRB2, CCND1) [1–4]. Despite the increasing resolution provided by CGH technology, there remains a limit to the resolution of detection of copy number aberrations (CNAs) of several hundred kilobases (kb) (supplementary Figure S1, available at Annals of Oncology online) [5]. However, CNAs are demarcated by rearrangements that can be detected from whole-genome sequences even when the size of the abnormal copy number segment is as small as 1 kb.

detection of copy number aberrations (CNAs) of several hundred kilobases (kb) (supplementary Figure S1, available at Annals of Oncology online) [5]. However, CNAs are demarcated by rearrangements that can be detected from whole-genome sequences even when the size of the abnormal copy number segment is as small as 1 kb. Somatic rearrangements are extremely diverse. Inter-patient variation exists in the quantity, type and distribution of somatic rearrangements even in cancers of the same tissue type [6, 7] and the consequences of rearrangements can also vary considerably. Solitary or low numbers of rearrangement breakpoints may directly confer selective advantage; for example breakpoints that transect tumour suppressor genes or that generate in-frame gene fusion events, such as ETV6-NTRK3 in breast cancer and TMPRSS-ERG fusions in prostate cancer [8, 9]. Collections of breakpoints can reflect driver amplifications. They can also be markers of complex, stochastic chromosomal events (e.g. chromoplexy, chromothripsis) [7, 10, 11] and provide increased resolution in studying mechanisms underpinning CNAs, for example, revealing that breakage-fusion bridge sometimes underpins the formation of the ERBB2 amplicon [12].

er amplifications. They can also be markers of complex, stochastic chromosomal events (e.g. chromoplexy, chromothripsis) [7, 10, 11] and provide increased resolution in studying mechanisms underpinning CNAs, for example, revealing that breakage-fusion bridge sometimes underpins the formation of the ERBB2 amplicon [12]. Methods Recently, 560 whole-genome sequenced breast cancers were expansively curated for somatic mutations, including rearrangements [5]. We previously defined ‘clustered’ rearrangements as clusters of breakpoints that occurred at high density in individual cancer genomes (see supplementary Methods, available at Annals of Oncology online). In the current study, we focus on characteristics of clustered rearrangements in 560 breast cancers that so far remained unexplored. In order to assess the impact of clustered rearrangements on breast cancer, we identified chromosomal hotspots where clustered rearrangements recurred in samples from different patients. Using the Piecewise-Constant-Fitting (PCF) algorithm [13] (see supplementary Methods, available at Annals of Oncology online), we sought genomic segments where groups of rearrangements exhibited short inter-mutation distances, indicative of ‘hotspots’ that are more frequently rearranged than the background rate. Using this method, we identified highly rearranged genomic loci that recurred in breast cancers. These sites make important contributions to tumorigenesis and reveal mechanisms underpinning chromosomal instability.

utation distances, indicative of ‘hotspots’ that are more frequently rearranged than the background rate. Using this method, we identified highly rearranged genomic loci that recurred in breast cancers. These sites make important contributions to tumorigenesis and reveal mechanisms underpinning chromosomal instability. Results PCF-based method identifies 21 hotspots of clustered rearrangements across 560 breast cancers There were 624 clusters of rearrangements in individual breast cancer genomes, comprising 17 247 intra-chromosomal rearrangements, and 6509 inter-chromosomal translocations. Clusters of rearrangements were common: 372 of 560 samples had at least one and were almost as frequent in triple-negative breast cancers (0.96 rearrangement clusters per sample) as in oestrogen receptor (ER)-positive breast cancers (1.00 rearrangement clusters per sample). Among PAM50 subtypes, luminal A cancers had fewest rearrangement clusters per sample (0.6, 95% Poisson CI 0.5–0.9) compared with other subtypes (luminal B 1.2, CI 1.0–1.5 and basal 1.2, CI 1.0–1.5). To identify loci where clusters of rearrangements recur across multiple independent tumour samples, we pooled all breakpoints in the ‘clustered’ category and sorted them according to position in the reference genome. PCF was applied to find hotspot regions in the genome that are recurrently affected by clusters of breakpoints in multiple patients (Figure 1A and B for workflow).

ss multiple independent tumour samples, we pooled all breakpoints in the ‘clustered’ category and sorted them according to position in the reference genome. PCF was applied to find hotspot regions in the genome that are recurrently affected by clusters of breakpoints in multiple patients (Figure 1A and B for workflow). Figure 1. Identification of hotspots of clustered rearrangements in breast cancers. (A) Workflow. (B) Schematic of clusters of rearrangements in individual samples, some of which form hotspots (grey shading). (C) Chromosomal localisation of breakpoints of rearrangements across 560 breast cancer genomes shown as counts (1 Mb bins). Chromosomes are depicted around the outside of the circle. Twenty-one hotspots of clustered rearrangements are shown in red. Positions of genes within each hotspot are indicated. Supplementary Figure S2, available at Annals of Oncology online shows a weighted histogram which is robust with respect to extremely rearranged individual samples. In all, 21 such hotspots of clustered rearrangements were identified (Figure 1C, supplementary Table S1 and Figures S2 and S3, available at Annals of Oncology online), encompassing 8% of the genome, but involving 46% of all breakpoints of clustered rearrangements.

Figure 1. Identification of hotspots of clustered rearrangements in breast cancers. (A) Workflow. (B) Schematic of clusters of rearrangements in individual samples, some of which form hotspots (grey shading). (C) Chromosomal localisation of breakpoints of rearrangements across 560 breast cancer genomes shown as counts (1 Mb bins). Chromosomes are depicted around the outside of the circle. Twenty-one hotspots of clustered rearrangements are shown in red. Positions of genes within each hotspot are indicated. Supplementary Figure S2, available at Annals of Oncology online shows a weighted histogram which is robust with respect to extremely rearranged individual samples. In all, 21 such hotspots of clustered rearrangements were identified (Figure 1C, supplementary Table S1 and Figures S2 and S3, available at Annals of Oncology online), encompassing 8% of the genome, but involving 46% of all breakpoints of clustered rearrangements. Recurrent clustered rearrangements identify common, large driver amplicons as well as rare, smaller amplicons Breakpoint densities for each of the 21 hotspots of clustered rearrangements identified in chromosomes 1, 6, 8, 11, 12, 15, 17, 19, 20 and 21 ranged between 35 and 165 breakpoints per Mb. We expected to find common driver amplification regions such as CCND1, ERBB2, ZNF217, chr8:ZNF703/FGFR1, IGF1R, and MYC as sites of clustered rearrangements recurring across many patients (Figure 1C). These were identified without exception. Hotspots were also identified at GNAS, RUNX1, and MDM2, all recognised as breast cancer genes, even if less frequent.

fication regions such as CCND1, ERBB2, ZNF217, chr8:ZNF703/FGFR1, IGF1R, and MYC as sites of clustered rearrangements recurring across many patients (Figure 1C). These were identified without exception. Hotspots were also identified at GNAS, RUNX1, and MDM2, all recognised as breast cancer genes, even if less frequent. Interestingly, several hotspots of clustered rearrangements were found near oncogenes that are not typically associated with breast cancer. Curation revealed that a subset had focal copy number gains typical of driver amplicons, albeit on a smaller scale (supplementary Figure S4, available at Annals of Oncology online). These hotspots at or near MCL1 (5.7% samples, 2.7% resulting in MCL1 amplification), PTP4A1 (4.5% samples, 1.25% PTP4A1 amplification) and MYB (6.3%, 1.4% MYB amplification) occurred at lower frequencies than that of common breast cancer amplicons (supplementary Figures S5 and S6 and Note 1, available at Annals of Oncology online for gene expression analysis). Further experiments will be required to verify whether these rarer, smaller and more modest amplicons are indeed driver events.

on) occurred at lower frequencies than that of common breast cancer amplicons (supplementary Figures S5 and S6 and Note 1, available at Annals of Oncology online for gene expression analysis). Further experiments will be required to verify whether these rarer, smaller and more modest amplicons are indeed driver events. Co-occurring hotspots: Inferring co-evolution through detailed breakpoint analyses Apart from an increased resolution in identifying copy number changes, whole-genome sequencing provides information to base-pair level about direct, physical connections between disparate genomic locations. Each of the 21 hotspots was identified independently through an agnostic approach. If we find that different hotspots are co-occurring at a higher frequency than would be expected, and further are physically connected to each other, this would suggest co-evolution of those allegedly independent hotspots, regardless of their original location on chromosomes. Below we report on two observations—an intra-chromosomal and an inter-chromosomal example—that provide insights into putative drivers and mutational mechanisms.

sically connected to each other, this would suggest co-evolution of those allegedly independent hotspots, regardless of their original location on chromosomes. Below we report on two observations—an intra-chromosomal and an inter-chromosomal example—that provide insights into putative drivers and mutational mechanisms. Co-evolving clusters on chromosome 6: possible driver loci? Four distinct hotspots of clustered rearrangements were identified on chromosome 6; the small amplicon attributed to PTP4A1 (chr6: 63.3Mb) and three larger hotspots at chr6: 96.6Mb, chr6: 117.6Mb and chr6: 128.5Mb (Figure 2A, supplementary Figure S6, available at Annals of Oncology online). Although they are independently identified loci, first we found that the four hotspots occurred together in different combinations in 10 samples (1.7% of cohort, Figure 2C). Second, they were also frequently physically linked through intra-chromosomal rearrangements indicating that they arose or evolved together during tumorigenesis (Figure 2B).