Browse the corpus

Hepatopulmonary syndrome (HPS) is a serious complication frequently observed in patients with chronic liver disease (1, 2). Previous studies have shown that pathological changes occurring in the lungs of animals with HPS included nonspecific pneumonia and epithelial cells injury (3). The tidal volume, minute ventilation, and mean inspiratory flow were significantly decreased, and chest wall pressure dissipation against the resistive and viscoelastic components and elasticity were reduced (4). All of these changes result in ventilation-perfusion (V/Q) mismatch, diffusion limitation of oxygen, and, less commonly, arteriovenous shunt. Orthotropic liver transplantation was considered the favorable method to improve the survival rate of patients with HPS (5). Many other operations have been performed on patients with HPS in clinics to improve hepatic function, such as endoscopic sphincterotomy, multiple bile duct stone extraction, and hilar tumor resection.

ropic liver transplantation was considered the favorable method to improve the survival rate of patients with HPS (5). Many other operations have been performed on patients with HPS in clinics to improve hepatic function, such as endoscopic sphincterotomy, multiple bile duct stone extraction, and hilar tumor resection. Hyperbilirubinemia is the typical feature of most severe liver disease and was considered the main cause of HPS (6). Previous studies have identified that high concentration of bilirubin leads to release of inflammatory cytokines from glial cells or neuronal cell apoptosis in the brain (7). In the lung, pulmonary vascular dilatation and high permeability of the pulmonary vascular barrier were found in rats undergoing common bile duct ligation (CBDL) surgery (8). Following destruction of the vascular endothelial barrier, high concentrations of bilirubin can enter interstitial tissue and, therefore, could be in direct contact with pulmonary alveolar epithelial cells, but the potential effect of bilirubin on the lung epithelial cells remains incompletely understood.

surgery (8). Following destruction of the vascular endothelial barrier, high concentrations of bilirubin can enter interstitial tissue and, therefore, could be in direct contact with pulmonary alveolar epithelial cells, but the potential effect of bilirubin on the lung epithelial cells remains incompletely understood. Dexmedetomidine, a potent α2 adrenergic agonist, has been commonly used in operating room and ICU for its sparing effect of other anesthetics or sedation (9, 10). Recent studies have demonstrated that dexmedetomidine could reduce systemic inflammation in animals and improves the gas exchange in patients (11, 12). An elegant previous study has shown that dexmedetomidine confers a remote lung acute injury following kidney ischemia reperfusion injury in mice (13). In addition, it has been found that dexmedetomidine has potent renoprotective effects against renal ischemia reperfusion injuries, which is very likely associated with p-Akt and Janus kinase/signal transducer and activator of transcription signaling activation in mice and rats (14, 15).Taken together, these studies indicate that dexmedetomidine possesses cytoprotective effects. Thus, we hypothesized that dexmedetomidine might alleviate the epithelial cell injury in HPS. In this study, we aim to investigate whether dexmedetomidine protects the lung alveolar epithelium against injury associated with hyperbilirubinemia both in vivo and in vitro.

dexmedetomidine possesses cytoprotective effects. Thus, we hypothesized that dexmedetomidine might alleviate the epithelial cell injury in HPS. In this study, we aim to investigate whether dexmedetomidine protects the lung alveolar epithelium against injury associated with hyperbilirubinemia both in vivo and in vitro. MATERIAL AND METHODS Cell Culture and Treatments The human alveolar epithelial cell line A549 (European Cell Culture Collection, Public Health English, Porton Down, Salisbury, UK) was used for this study. A549 cells were cultured in Roswell Park Memorial Institute 1640 (RPMI 1640) medium supplemented with 10% fetal bovine serum and cultured at 37° in 5% Co2. Cells were challenged with a gradient concentration of bilirubin (from 0 to 160 μM) in fresh RPMI 1640 medium free from fetal bovine serum for 24 hours. The concentration of bilirubin was chosen based on the blood test of unconjugated bilirubin of CBDL rats (16). Live cells counted from 10 random microscopic fields from each 6-cm culture dishes were analyzed. The average cell survival ratio relative to the naïve controls (NCs) was used for further data analysis. Dexmedetomidine concentrations were chosen from our previous study (14, 15), and the dose of atipamezole, an α2 adrenergic antagonist, was based on a binding affinity ratio 10:1 for agonist-antagonist (17). Briefly, the cells were pretreated with 1 nM dexmedetomidine for 15 minutes first in the presence or absence of 10 nM atipamezole and then exposed to 80 μM bilirubin for additional 24 hours. Cells were grouped as NC: no drug challenge; bilirubin alone (B80): 80 μM bilirubin; dexmedetomidine: 1 nM dexmedetomidine; dexmedetomidine + bilirubin (DB): pretreat with 1 nM dexmedetomidine for 15 minutes and then expose to 80 μM bilirubin; atipamezole: 10 nM atipamezole; atipamezole + dexmedetomidine + bilirubin: pretreat with atipamezole for 15 minutes and then treated with dexmedetomidine for another 15 minutes, then exposed to 80 μM bilirubin.

idine + bilirubin (DB): pretreat with 1 nM dexmedetomidine for 15 minutes and then expose to 80 μM bilirubin; atipamezole: 10 nM atipamezole; atipamezole + dexmedetomidine + bilirubin: pretreat with atipamezole for 15 minutes and then treated with dexmedetomidine for another 15 minutes, then exposed to 80 μM bilirubin. Determination of Δψm In Vitro A549 cells were labeled with lipophilic cationic probe 5, 5′, 6,6′-tetrachloro-1,1′,3,3′-tetraethylbenzimidazolcarbocyanineiodide (JC-1) and detected by flow cytometry and microscope separately (18, 19). Briefly, for flow cytometry detection, cells were detached using 0.25% trypsin and then transferred to 5 mL polystyrene tubes. After washing with fluorescence-activated cell sorting (FACS) buffer (10% fetal calf serum, 0.5 M EDTA in 0.1 M phosphate-buffered saline [PBS]) once, the cells were incubated with 0.2 μM JC-1 in FACS for 30 minutes at 37°C and protected from light. The cells were analyzed by flow cytometry (FACS Calibur; Becton Dickinson, Sunnyvale, CA) after washing with warm FACS buffer twice. Fluorescence was measured in the FL-1 (green fluorescence) and FL-2 (red fluorescence) channels, gating only on live cells. To detect the direct fluorescence changes of JC-1 staining, the cells cultured on cover slips were stained with JC-1 directly as previously stated with some modifications (19). Briefly, the cells were incubated with 10 μM JC-1 in FACS buffer for 30 minutes at 37°C after washing with warm 0.1 M PBS once. The cells were washed with warm 0.1 M PBS three times for 5 minutes each while carefully protecting from potent light, and then, the nuclei were stained using 4,6-diamidino-2-phenylindole (DAPI). The fluorescence stained on the cells was examined under rhodamine (red), fluorescein (green), and cyan (blue) spectral filter with Olympus (Watford, United Kingdom) BX40 microphotography system. The geometric mean of red fluorescence and green fluorescence of JC-1 staining on the cells was analyzed using FlowJo 7.6.1 software (TreeStar, San Carlos, CA) from eight independent flow cytometry detections. The mean ratio of red/green was calculated as an indicator of Δψm.

ited Kingdom) BX40 microphotography system. The geometric mean of red fluorescence and green fluorescence of JC-1 staining on the cells was analyzed using FlowJo 7.6.1 software (TreeStar, San Carlos, CA) from eight independent flow cytometry detections. The mean ratio of red/green was calculated as an indicator of Δψm. Immunohistochemistry For in vitro fluorescence staining, cells were fixed in paraformaldehyde in 0.1 M PBS solution. Cells were then incubated in 10% normal donkey serum and then incubated overnight with Rabbit anticytochrome C or B cell leukemia 2 associated X protein, B cell leukemia-2, cleaved-caspase 3 and 9, transforming growth factorβ (TGFβ), phosphorylated mammalian target of rapamycin (p-mTOR), and p42/44 mitogen-activated protein kinase (MAPK) (1:200; Santa Cruz, Dallas, TX) followed by secondary antibody for 1 hour. For in vivo fluorescence staining, 5-mm-thick paraffin sections were first dewaxed and subjected to heat-mediated antigen retrieval. Sections were incubated with donkey serum followed by the cleaved-caspase 3 antibody (1:200; Santa Cruz). After washing with PBS-Tween 20, the slides were incubated with fluorochrome-conjugated secondary antibodies (Millipore, Beeston, United Kingdom) for 1 hour. The slides were counterstained with nuclear dye DAPI and then examined by using an Olympus BX40 microscope under constant exposure level. Immunofluorescence was quantified using ImageJ (National Institutes of Health, Bethesda, MD), and the background was subtracted. Ten representative fields were randomly selected by an assessor blinded to the treatment groups. Values were then calculated as percentages of the mean value for NCs and expressed as percentage fluorescence. The proportion of positive cells was calculated as the number of positive cells relative to the number of DAPI-positive cells.

in the DB group (Fig. 1C). Cytochrome C released from the mitochondria in A549 cells was detected by immunofluorescence. Enhanced expression of cytochrome C was found after exposure to 80 μM bilirubin for 24 hours (p < 0.01 compared to NC), which was partly reversed by 1 nM dexmedetomidine (p < 0.01) (Fig. 1, D and E). Figure 1. Effect of dexmedetomidine (Dex) on the bilirubin-induced Δψm collapse and release of cytochrome C from mitochondria in A549 cells. A549 cells were pretreated with 1 nM dexmedetomidine for 15 min and then exposed to 80 μM bilirubin for additional 24 hr. A, Δψm was assessed by flow cytometry under fluorescein isothiocyanate and phycoerythrin channel after stained with 5, 5′, 6, 6′-tetrachloro-1,1′,3,3′-tetraethylbenzimidazolcarbocyanine iodide (JC-1). B, A549 cells were stained with JC-1, assessed through immunofluorescence microscope. Nuclear was counter stained with 4,6-diamidino-2-phenylindole. C, JC-1 fluorescence intensity ratio (red/green) compared with each other was from data of eight independent experiments of flow cytometry. D, Expression of cytochrome C in A549 cells detected with immunofluorescence. E, Fluorescence intensity of cytochrome C. Data are mean ± sd. n = 8. *p < 0.05 and **p< 0.01 and ***p < 0.001, scale bar: 50 μm. B80 = final concentration of bilirubin was 80 μM, MFI = mean fluorescence intensity, NC = naïve controls.

tive fields were randomly selected by an assessor blinded to the treatment groups. Values were then calculated as percentages of the mean value for NCs and expressed as percentage fluorescence. The proportion of positive cells was calculated as the number of positive cells relative to the number of DAPI-positive cells. Cell Cycle Analysis by Flow Cytometry The cell cycle was analyzed by flow cytometry as described previously (20). The cells were detached from the 24-well culture plate with 0.25% trypsin and then transferred to 5 mL polystyrene tubes specifically designed for flow cytometry. After washing twice with 0.1 M PBS, the cells were fixed with 70% ethanol at 4° overnight. After centrifuging at 2,500 rpm for 10 minutes and resuspending in 500 μL freshly prepared FACS buffer, 10 μL of 40 μg/L propidium iodide (PI) and 10 μL of 500 ng/L ribonuclease were added to the cell suspension and kept in a dark place for 10 minutes. Fluorescence of PI stained on the cells was detected with flow cytometry and analyzed with FlowJo 7.6.1 software (FACS Calibur, Becton Dickinson, Sunnyvale, CA). For the cell cycle analysis, a minimum of 10,000 cells per sample were analyzed with flow cytometry (TreeStar, San Carlos, CA; BioRad, Hemel Hempstead, UK). Data were analyzed by FlowJo software (TreeStar; BioRad), which showed basic statistics such as the fraction of cells in G0/1, S, and G2, the positions of the G0/1 and G2 peaks, and their widths. The percentage of cells in different phases of the cell cycle was therefore determined.

s, CA; BioRad, Hemel Hempstead, UK). Data were analyzed by FlowJo software (TreeStar; BioRad), which showed basic statistics such as the fraction of cells in G0/1, S, and G2, the positions of the G0/1 and G2 peaks, and their widths. The percentage of cells in different phases of the cell cycle was therefore determined. Animals and Surgical Procedure This study was approved by the Ethics Committee of Animal Experiments of Third Military Medical University. Every effort was made to minimize animal suffering and the number of animals used. Sprague-Dawley rats (220–250 g) were used for experiments and were kept under a 12-hour light/dark cycle with free access to food and water. Hyperbilirubinemia was induced by modified CBDL as we reported before (21, 22). Aseptic laparotomy was made in Sprague-Dawley rats (220–250 g) under 3.5% chloral hydrate anesthesia (10 mL/kg, IP). The common bile duct was identified and double ligated with 4-0 cotton sutures (CBDL). Just laparotomy without bile duct ligation or without any surgery served as the Sham controls and NCs, respectively. They were allowed to recover in individual cages as reported previously (13, 14); 25 μg/kg dexmedetomidine or the same volume saline (as vehicle control) was intraperitoneal (IP) injected after 3 hours of CBDL surgery on the 1st day and then for 6 consecutive days. Dexmedetomidine-controlled rats only received IP injection of 25 μg/kg dexmedetomidine daily and without any surgery. At the end of the experiments, the rats received terminal anesthesia (chloral hydrate 350 mg/kg, IP), and 2 mL blood was immediately collected through a needle punctured in left ventricle of heart. Blood gas and unconjugated bilirubin were measured with routine clinical laboratory apparatuses. The lungs were, subsequently, perfused with 4% paraformaldehyde under constant pressure and then embedded into paraffin and sectioned into 5 μm for further histological analysis.

needle punctured in left ventricle of heart. Blood gas and unconjugated bilirubin were measured with routine clinical laboratory apparatuses. The lungs were, subsequently, perfused with 4% paraformaldehyde under constant pressure and then embedded into paraffin and sectioned into 5 μm for further histological analysis. Histological Analysis The sections were stained with hematoxylin and eosin (H&E) staining, and the morphology in each lung (10 fields at ×20 magnifications) was evaluated by an observer who was blinded to the treatments under an Olympus BX40 microscope. The score for each field was calculated from the sum score of 10 areas chosen randomly. The injury of the lung alveolar epithelial cells was categorized (23) into grade 0: normal appearance, negligible damage; grade 1: mild-moderate interstitial congestion and neutrophil leukocyte infiltrations; grade 2: perivascular edema formation, partial destruction of pulmonary architecture, and moderate cell infiltration; grade 3: moderate lung alveolar damage and intensive cell infiltration; grade 4: severe cell infiltration and severe destruction of the pulmonary architecture. Lung Wet/Dry Ratio The lungs harvested from other cohorts after experiments under terminal anesthesia were weighed to obtain wet weight and then dried for 48 hours at 80°C oven to obtain the dry weight. The wet-to-dry ratio was calculated as an indicator of pulmonary edema.

Histological Analysis The sections were stained with hematoxylin and eosin (H&E) staining, and the morphology in each lung (10 fields at ×20 magnifications) was evaluated by an observer who was blinded to the treatments under an Olympus BX40 microscope. The score for each field was calculated from the sum score of 10 areas chosen randomly. The injury of the lung alveolar epithelial cells was categorized (23) into grade 0: normal appearance, negligible damage; grade 1: mild-moderate interstitial congestion and neutrophil leukocyte infiltrations; grade 2: perivascular edema formation, partial destruction of pulmonary architecture, and moderate cell infiltration; grade 3: moderate lung alveolar damage and intensive cell infiltration; grade 4: severe cell infiltration and severe destruction of the pulmonary architecture. Lung Wet/Dry Ratio The lungs harvested from other cohorts after experiments under terminal anesthesia were weighed to obtain wet weight and then dried for 48 hours at 80°C oven to obtain the dry weight. The wet-to-dry ratio was calculated as an indicator of pulmonary edema. Western Blotting Lung samples were mechanically homogenized in lysis buffer. The cell lysates were centrifuged and then supernatant was collected, and total protein concentration in the supernatant was quantified by the Bradford protein assay (BioRad). The protein extracts (40 μg/sample) were heated, denatured, and loaded on a NuPAGE 4–12% Bis-Tris gel (Invitrogen, Paisley, UK) for electrophoresis and then transferred to a polyvinylidene difluoride membrane. The membrane was treated with blocking solution (5% dry milk in Tris buffered saline with 0.1% Tween-20) for 2 hours and probed with cleaved-caspase 3 antibody (1:1,000; Santa Cruz) overnight at 4°C, followed by horseradish peroxidase-conjugated secondary antibody for 1 hour. The loading control was the constitutively expressed protein β-actin (1:10,000; Sigma-Aldrich, St. Louis, MO). The blots were visualized with enhanced chemiluminescence system (Santa Cruz) and analyzed with GeneSnap (Syngene, Cambridge, United Kingdom).

y horseradish peroxidase-conjugated secondary antibody for 1 hour. The loading control was the constitutively expressed protein β-actin (1:10,000; Sigma-Aldrich, St. Louis, MO). The blots were visualized with enhanced chemiluminescence system (Santa Cruz) and analyzed with GeneSnap (Syngene, Cambridge, United Kingdom). Terminal Deoxynucleotidyl Transferase–Mediated dUTP Nick-End Labeling Assay Lung alveolar cell apoptosis was detected using the one-step terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) apoptosis assay kit (Beyotime, Shanghai, China) according to the manufacturer’s instructions. In brief, the fixed, paraffin-embedded sections were deparaffinized and washed with PBS, labeled with terminal-deoxynucleotidyl transferase enzyme at 37°C for 0.5 hour, and then incubated with 5-bromo-4-chloro-3-indolyl-phosphate substrate labeled with Cy-3 for 40 minutes and DAPI for 5 minutes at room temperature. Ten randomly chosen fields were examined under ×40 view, and the proportion of TUNEL-positive cells was calculated as the number of TUNEL-positive cells relative to the number of DAPI-positive cells. Statistical Analysis All numerical data were presented as mean ± sd or otherwise box-whisker plot. Comparison between the treatment groups was analyzed by one-way analysis of variance followed by Tukey multiple group comparisons (equal variances) or Kruskal-Wallis (nonparametric) test followed by Dunn multiple group comparisons (unequal variances or scoring) (GraphPad Prism 5, San Diego, CA). A p value of 0.05 was considered as statistically significant.

was analyzed by one-way analysis of variance followed by Tukey multiple group comparisons (equal variances) or Kruskal-Wallis (nonparametric) test followed by Dunn multiple group comparisons (unequal variances or scoring) (GraphPad Prism 5, San Diego, CA). A p value of 0.05 was considered as statistically significant. RESULTS Effects of Dexmedetomidine on Δψm, Cytochrome C Release in A549 Cells After Bilirubin Challenge Δψm was measured with flow cytometry and microscope separately after JC-1 staining. In cells in NC and dexmedetomidine groups, the intensity of red fluorescence was about two times green (no significant difference in the ratio of red/green between these two groups; p > 0.05). In bilirubin challenged cells (the B80 group), there were 33.3% ± 2.71% cells presenting with low Δψm and this proportion of cells decreased to 28.7% ± 1.98% in the dexmedetomidine pretreated cells (DB group) (p < 0.01) (Fig. 1A). Meanwhile, the mean ratio of red/green of all cells in the B80 group was 77.96 ± 5.18% and was increased to 107.74 ± 13.00% in the DB group (Fig. 1C). Cytochrome C released from the mitochondria in A549 cells was detected by immunofluorescence. Enhanced expression of cytochrome C was found after exposure to 80 μM bilirubin for 24 hours (p < 0.01 compared to NC), which was partly reversed by 1 nM dexmedetomidine (p < 0.01) (Fig. 1, D and E).

metry. D, Expression of cytochrome C in A549 cells detected with immunofluorescence. E, Fluorescence intensity of cytochrome C. Data are mean ± sd. n = 8. *p < 0.05 and **p< 0.01 and ***p < 0.001, scale bar: 50 μm. B80 = final concentration of bilirubin was 80 μM, MFI = mean fluorescence intensity, NC = naïve controls. Effects of Dexmedetomidine on BAX, BCL-2, and Cleaved-Caspase 3 and 9 Expression in A549 Cells After Being Challenged by Bilirubin Immunofluorescence study revealed dexmedetomidine significantly inhibited upregulation of BAX expression in A549 cells induced by bilirubin (p < 0.001) (Fig. 2, A and B). Decreased fluorescence intensity of BCL-2 was found in bilirubin- and dexmedetomidine-treated A549 cells. No significant difference of mean fluorescence intensity (MFI) was found among bilirubin, dexmedetomidine, and combined treated cells (p > 0.05) (Fig. 2, C and D). Upregulation of cleaved-caspase 3 and 9 expression in A549 cells was detected after 80 μM bilirubin treatment for 24 hours (p < 0.05 vs NC). Dexmedetomidine completely reversed the upregulation of cleaved-caspase 9 but only had a partial effect on that of cleaved-caspase 3 induced by bilirubin in A549 cells (Fig. 2E–H).

pregulation of cleaved-caspase 3 and 9 expression in A549 cells was detected after 80 μM bilirubin treatment for 24 hours (p < 0.05 vs NC). Dexmedetomidine completely reversed the upregulation of cleaved-caspase 9 but only had a partial effect on that of cleaved-caspase 3 induced by bilirubin in A549 cells (Fig. 2E–H). Figure 2. Dexmedetomidine (Dex) reversed the bilirubin-induced BAX, cleaved-caspase 3, and caspase 9 overexpression but had no effect on BCL-2 downregulation. A549 cells were challenged with 80 μM bilirubin for 24 hr with or without pretreatment with 1 nM dexmedetomidine. A, Green fluorescence showed the overexpression of BAX in the cells exposed to bilirubin and which was reversed by pretreat with dexmedetomidine. B, The mean fluorescence intensity of BAX increased in bilirubin-challenged A549 cells and which was downregulated by pretreat with 1 nM dexmedetomidine. C, The expression of BCL-2 (green) in A549 cells. D, BCL-2 expression in A549 cells after bilirubin challenge with or without dexmedetomidine pretreat. E, Green fluorescence showed the expression of cleaved-caspase 3 in A549 cells. Overexpression of cleaved-caspase 3 induced by bilirubin was inhibited by pretreat with dexmedetomidine; histogram of cleaved-caspase 3 fluorescence intensity (F). Dexmedetomidine inhibited the overexpression of cleaved-caspase 9 in A549 cells; green fluorescence showed the expression of cleaved-caspase 9 in A549 cells (G); histogram of cleaved-caspase 9 fluorescence intensity (H). Dexmedetomidine inhibited the overexpression of cleaved-caspase 9 in A549 cells. Data are mean ± sd. n = 8. *p < 0.05 and **p < 0.01 and ***p < 0.001, scale bar: 50 μm. B80 = final concentration of bilirubin was 80 μM, DB = dexmedetomidine + bilirubin, MFI = mean fluorescence intensity, NC = naïve controls.

ensity (H). Dexmedetomidine inhibited the overexpression of cleaved-caspase 9 in A549 cells. Data are mean ± sd. n = 8. *p < 0.05 and **p < 0.01 and ***p < 0.001, scale bar: 50 μm. B80 = final concentration of bilirubin was 80 μM, DB = dexmedetomidine + bilirubin, MFI = mean fluorescence intensity, NC = naïve controls. Effects of Dexmedetomidine on Activation of TGFβ, p-mTOR, and p44/42MAPK Pathways in A549 Cells After Being Challenged by Bilirubin TGFβ expression in A549 cells was determined by immunofluorescence to explore the protective effect of dexmedetomidine on bilirubin-induced cell apoptosis. The result indicated that 80 μM bilirubin and 1 nM dexmedetomidine significantly inhibited the expression of TGFβ (p < 0.001). Pretreatment with dexmedetomidine for 15 minutes partly reversed the downregulation of TGFβ expression induced by bilirubin (MFI, 15.06 ± 3.00 in B80 group vs 25.32 ± 5.94 in DB group; p < 0.001). There was no significant difference in MFI between dexmedetomidine and DB groups (p > 0.05) (Fig. 3, A and B). The percentage of p-mTOR and p44/42MAPK-positive cells was reduced by either 80 μM bilirubin or 1 nM dexmedetomidine (p < 0.001 vs NC). The percentage of p-mTOR-positive cells increased from 13.25% ± 2.01% in the B80 group to 29.35% ± 2.96% in the DB group (p < 0.001). The percentage of p44/42MAPK-positive cells also increased in the dexmedetomidine pretreated cells (18.37% ± 6.51% in the B80 group vs 35.32% ± 7.28% in the DB group; p < 0.001) (Fig. 3C–F).

percentage of p-mTOR-positive cells increased from 13.25% ± 2.01% in the B80 group to 29.35% ± 2.96% in the DB group (p < 0.001). The percentage of p44/42MAPK-positive cells also increased in the dexmedetomidine pretreated cells (18.37% ± 6.51% in the B80 group vs 35.32% ± 7.28% in the DB group; p < 0.001) (Fig. 3C–F). Figure 3. Dexmedetomidine (Dex) reversed the bilirubin-induced downregulation of TGFβ, phosphorylated mammalian target of rapamycin (p-mTOR), and p42/44 mitogen-activated protein kinase (MAPK) in A549 cells. A549 cells were challenged with 80 μM bilirubin for 24 hr with or without pretreatment with 1 nM dexmedetomidine. A, Expression of TGFβ (green) was assessed by immunofluorescence staining. Nuclear was counterstained with 4,6-diamidino-2-phenylindole (DAPI); the mean green fluorescence intensity of TGFβ (B); expression of p-mTOR (green fluorescence) in A549 cells was assessed by immunofluorescence staining (C). Nuclear was counterstained with DAPI; percentage of p-mTOR-positive cells (D); expression of p42/44MAPK (green) in A549 cells was assessed by immunofluorescence staining (E). Nuclear was counterstained with DAPI; the percentage of p42/44MAPK-positive cells (F). Data are mean ± sd. n =8. *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm. White arrow, negative cells; yellow arrow, positive cells. B80 = final concentration of bilirubin was 80 μM, DB = dexmedetomidine + bilirubin, MFI = mean fluorescence intensity, NC = naïve controls.

ntage of p42/44MAPK-positive cells (F). Data are mean ± sd. n =8. *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm. White arrow, negative cells; yellow arrow, positive cells. B80 = final concentration of bilirubin was 80 μM, DB = dexmedetomidine + bilirubin, MFI = mean fluorescence intensity, NC = naïve controls. Effects of Dexmedetomidine on Cell Cycle Progression of A549 Cells Following Bilirubin Challenge The concentration-dependent effect of bilirubin on cell morphology and survivability of A549 cells was observed under the microscope. Bilirubin induced a concentration-dependent decrease of cell survival ratio after 24-hour treatment (Fig. 4, A and B). The cell cycle of A549 cells were investigated using flow cytometry. The percentage of the cells in G0/G1 phase gradually increased with the increasing of bilirubin concentration. There was no significant difference in S phase cells detected between each group, except B40 (p < 0.001 vs NC), and accordingly, the percentage of cells in G2/M phase decreased gradually (Fig. 4, C and D). Dexmedetomidine at 1 nM concentration was used to explore its effect on 80 μM bilirubin-induced cell death and G0/G1 arrest. Dexmedetomidine was found to increase the survivability of the A549 cells after bilirubin challenge. Blockage of the conjugation between dexmedetomidine and α2-adrenoceptor with high concentration of atipamezole (10 nM) partially abolished the effect of dexmedetomidine (Fig. 5, A and B). The percentage increase in G0/G1 and decrease in G2/M of cell cycle was found after A549 cells being challenged with 80 μM bilirubin and which was partly revised by pretreatment with 1 nM dexmedetomidine (from 77.35% ± 3.04% decreased to 51.73% ± 4.79% in G0/G1 phase and from 20.17% ± 2.89% increased to 36.22% ± 3.12% in G2/M phase; p < 0.05). There was no significant difference about the S phase cells between B80 and DB groups (p > 0.05). The revising effect of dexmedetomidine on bilirubin-induced G0/G1 phase percentage increase could be partly abolished by 10 nM atipamezole (from 51.73% ± 4.79% to 59.28% ± 3.92%; p < 0.01) (Fig. 5C–G).

/M phase; p < 0.05). There was no significant difference about the S phase cells between B80 and DB groups (p > 0.05). The revising effect of dexmedetomidine on bilirubin-induced G0/G1 phase percentage increase could be partly abolished by 10 nM atipamezole (from 51.73% ± 4.79% to 59.28% ± 3.92%; p < 0.01) (Fig. 5C–G). Figure 4. Bilirubin-induced cell death and arrest of G0/G1 phase of live A549 cells in a concentration-dependent manner. A549 cells were challenged with gradually increased concentration of bilirubin for 24 hr. A, Cell morphology of A549 cells after treated with gradually increased bilirubin for 24 hr. B, Cell survival ratios after being challenged with a gradient concentration of bilirubin. ***p < 0.001; n.s. = no significant difference. C, Cell cycle of A549 cells after bilirubin challenge, assessed by propidium iodide flow cytometry. D, Percentage of cells at G0/G1, S, and G2/M after bilirubin challenge. *p < 0.001 compared with the cells in G0/G1 phase of naïve control (NC); #p < 0.001 compared with the cells in S phase of NC; Δp < 0.001 compared with the cells in G2/M phase of NC. B0 = final concentration of bilirubin was 0 μM, B20 = final concentration of bilirubin was 20 μM, B40 = final concentration of bilirubin was 40 μM, B80 = final concentration of bilirubin was 80 μM, B160 = final concentration of bilirubin was 160 μM. Data are mean ± sd. n = 8. *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 100 μm.

ncentration of bilirubin was 0 μM, B20 = final concentration of bilirubin was 20 μM, B40 = final concentration of bilirubin was 40 μM, B80 = final concentration of bilirubin was 80 μM, B160 = final concentration of bilirubin was 160 μM. Data are mean ± sd. n = 8. *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 100 μm. Figure 5. Dexmedetomidine (Dex) reversed the arrest of G0/G1 phase of A549 cells independent on α2 receptor. A549 cells were pretreated with 1 nM dexmedetomidine (or combined with its antagonist, 10 nM atipamezole [Ati]) for 15 min, then treated with 80 μM bilirubin for additional 24 hr. A, Dexmedetomidine attenuated the cell death induced by 80 μM bilirubin. Fewer cells dispatched from the bottom of plate in dexmedetomidine + bilirubin group cells compared with that in bilirubin treated only (B80 group). This effect was reversed by 10 nM atipamezole, scale bar = 100 μm. B, Summary of the cell survival ratio changes, ***p < 0.001. C, Dexmedetomidine reversed the G0/G1 arrest in A549 cells induced by 80 μM bilirubin, but it could not be completely reversed by the α2 antagonist (10 nM atipamezole). D, Statistical analysis of cell cycle after dexmedetomidine (associated with atipamezole or not) and bilirubin treatment and compared with that in naïve control (NC). *p < 0.001 compared with the cells in G0/G1 phase of NC; #p < 0.001 compared with the cells in S phase of NC; Δp < 0.001 compared with the cells in G2/M phase of NC. E, Comparison among the treatment groups about the percent of G0/1 phase cells. (F, Comparison among the treatment groups about the percent of S phase cells. G, Comparison among the treatment groups about the percent of G2/M phase cells. Data are mean ± sd in bar graph E, F, and G. n = 8. *p < 0.05, **p < 0.01, ***p < 0.001. DB = dexmedetomidine + bilirubin, ADB = atipamezole + dexmedetomidine + bilirubin.

rison among the treatment groups about the percent of S phase cells. G, Comparison among the treatment groups about the percent of G2/M phase cells. Data are mean ± sd in bar graph E, F, and G. n = 8. *p < 0.05, **p < 0.01, ***p < 0.001. DB = dexmedetomidine + bilirubin, ADB = atipamezole + dexmedetomidine + bilirubin. CBDL Operation Induced Lung Edema, Alveolar Epithelial Cell Apoptosis, and Respiratory Failure, Which Were Attenuated by Dexmedetomidine All rats survived until euthanize. After CBDL operation, the lung alveolar epithelial cells were flattened and the size of alveoli became more variable compared with naïve and sham rats when observed with H&E staining and the lung injury score increased significantly when compared with that in naïve or sham control rats (p < 0.001) (Fig. 6A). The level of blood unconjugated bilirubin increased significantly on the 7th day and remained at a relatively higher level on the 14th and 21st day (remained at 80.31–156.71 μM; p < 0.001 vs sham). The Pao2 of CBDL rats decreased gradually as the time extension and adverse tendency was observed on the Paco2 (Fig. 6, B–D). Compared with the naïve and dexmedetomidine control rats, the pulmonary edema and inflammation were significantly reduced in those dexmedetomidine-treated CBDL rats when observed with H&E staining at day 7 after operation. Meanwhile, IP injection of dexmedetomidine attenuated the increased ratio of lung wet/dry weight, increased blood Paco2 and decreased blood Pao2 in CBDL rats (p < 0.05, compared with CBDL rats and saline-treated CBDL rats) (Fig. 7A–D). Cleaved-caspase 3 expression in the lungs of CBDL rats at day 7 after operation was significantly attenuated by dexmedetomidine when detected with both immunochemistry and western blotting (p < 0.05, compared with CBDL rats and saline-treated CBDL rats) (Fig. 8A–D). Dexmedetomidine also inhibited the increase of TUNEL-positive cells in the lungs of CBDL rats at day 7 after operation (p < 0.05, compared with CBDL rats and saline-treated CBDL rats) (Fig. 8, E and F).

th immunochemistry and western blotting (p < 0.05, compared with CBDL rats and saline-treated CBDL rats) (Fig. 8A–D). Dexmedetomidine also inhibited the increase of TUNEL-positive cells in the lungs of CBDL rats at day 7 after operation (p < 0.05, compared with CBDL rats and saline-treated CBDL rats) (Fig. 8, E and F). Figure 6. Common bile duct ligation (CBDL) surgery induced a persistent hyperbilirubinemia and pulmonary injury in Sprague Dawley (SD) rats. SD rats were subjected to CBDL operation, and the blood and lung samples were harvested at 7, 14, and 21 d after operation. Just laparotomy without bile duct ligation or without any surgery served as the sham controls and naïve controls (NCs), respectively. A, Histology assessment of lung in rats after CBDL surgery. B, Blood direct bilirubin (DBIL) (also known as unconjugated bilirubin) concentration increased in SD rats. Blood DBIL concentration increased to a very high level at day 7 after CBDL operation and remained at relative high level at days 14 and 21 (p < 0.001 relative to sham and NC). C, Pao2 gradually decreased after CBDL operation on SD rats. D, Paco2 gradually increased after CBDL operation on SD rats. n.s. = no significant difference. n = 10, *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm.

peration and remained at relative high level at days 14 and 21 (p < 0.001 relative to sham and NC). C, Pao2 gradually decreased after CBDL operation on SD rats. D, Paco2 gradually increased after CBDL operation on SD rats. n.s. = no significant difference. n = 10, *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm. Figure 7. Intraperitoneal injection of dexmedetomidine (Dex) attenuated the pulmonary injury and respiratory failure of common bile duct ligation (CBDL) rats. A 25 μg/kg dexmedetomidine or the same volume saline (as vehicle control) was administered (intraperitoneally, IP) daily for 7 d after CBDL surgery. Dexmedetomidine-controlled rats only received IP injection of 25 μg/kg dexmedetomidine daily and without any surgery for 7 d. Hematoxylin and eosin staining of the lung sample (A); wet-to-dry ratio in lung samples of CBDL rats (B); decrease of Pao2 in CBDL rats was attenuated by dexmedetomidine on the day 7 after CBDL operation (C); increase of Paco2 in CBDL rats was inhibited by dexmedetomidine on the day 7 after CBDL operation (D). Dex7d = dexmedetomidine control rats, rats received injection of 25 μg/kg dexmedetomidine intraperitoneally for 7 d, NC = naïve control, n.s. = no significant difference. n = 10, *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm.

in CBDL rats was inhibited by dexmedetomidine on the day 7 after CBDL operation (D). Dex7d = dexmedetomidine control rats, rats received injection of 25 μg/kg dexmedetomidine intraperitoneally for 7 d, NC = naïve control, n.s. = no significant difference. n = 10, *p < 0.05, **p < 0.01, ***p < 0.001. Scale bar = 50 μm. Figure 8. Dexmedetomidine (Dex) attenuated the apoptosis of lung cells of the common bile duct ligation (CBDL) rats after 7 d of operation. A 25 μg/kg dexmedetomidine or the same volume saline (as vehicle control) was administered (IP) daily for 7 d after CBDL surgery. Samples of lungs were harvested for immunochemistry and terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay. A, Expression of cleaved-caspase 3 in the lungs of controlled and 7th day CBDL rats detected using immunochemistry. B, Percentage of cleaved-caspase 3 positive staining cells in rat lungs. C, Expression of cleaved-caspase 3 in the lungs of control and 7th day CBDL rats detected using western blotting. D, Expression of cleaved-caspase 3 in the lungs of control and 7th day CBDL rats. E, TUNEL-positive cells in the lungs of control and 7th day CBDL rats. F, Percentage of TUNEL-positive cells in the lungs. n = 10, ***p < 0.001. NC = naïve control, n.s. = no significant difference. Scale bar = 50 μm.

ng western blotting. D, Expression of cleaved-caspase 3 in the lungs of control and 7th day CBDL rats. E, TUNEL-positive cells in the lungs of control and 7th day CBDL rats. F, Percentage of TUNEL-positive cells in the lungs. n = 10, ***p < 0.001. NC = naïve control, n.s. = no significant difference. Scale bar = 50 μm. DISCUSSION The present study, for the first time, explored the effect of dexmedetomidine, an α2 receptor agonist, on bilirubin-induced lung alveolar epithelial cell injury both in vivo and in vitro. Our results demonstrated that 1) cell death and G0/G1 cell cycle arrest of A549 cells were detected after bilirubin treatment for 24 hours in a concentration-dependent manner; 2) dexmedetomidine attenuated bilirubin-induced cell death and enhanced the proliferation of alveolar epithelial cells; 3) persistent hyperbilirubinemia was observed after CBDL operation, which may contribute to the lung alveolar epithelial injury and respiratory failure in rats; and 4) dexmedetomidine attenuated the pulmonary epithelial cell injury and respiratory failure of CBDL rats.

ed the proliferation of alveolar epithelial cells; 3) persistent hyperbilirubinemia was observed after CBDL operation, which may contribute to the lung alveolar epithelial injury and respiratory failure in rats; and 4) dexmedetomidine attenuated the pulmonary epithelial cell injury and respiratory failure of CBDL rats. In our study, bilirubin at high concentrations induced epithelial cell apoptosis albeit in vitro. This is in line with a previous study in which bilirubin induced rat brain neuronal apoptosis via the mitochondrial pathway (24). Mitochondrial dysfunction has been shown to be involved in the induction of apoptosis (18). The opening of the mitochondrial permeability transition pore has been demonstrated to induce the depolarization of Δψm, release of apoptogenic factors such as cytochrome C, and disruption of adenosine triphosphate production (25). Mounting evidence indicates that disruption of the Δψm initiates the caspase cascade, leading to downstream activation of apoptosis (26, 27). Proapoptotic molecule BAX is essential in mitochondrial apoptotic pathway (28). In healthy mammalian cells, BAX is essentially cytosolic and inactive. Following a death signal, BAX protein is translocated to the outer mitochondrial membrane, where it promotes a permeabilization that favors the release of many apoptogenic factors, such as cytochrome C. Furthermore, BAX is a main member of the BCL-2 proapoptotic protein family. BCL-2, another member of BCL-2 family, counteracts BAX in the permeabilization of outer mitochondrial membrane and cell apoptosis (29). Increased BAX/BCL-2 ratio contributes to the initiation of caspase-induced apoptosis (30). In this study, bilirubin upregulated BAX and downregulated BCL-2 in A549 cells; in addition, dexmedetomidine treatment reversed the overexpression of BAX. Although dexmedetomidine did not stop the downregulation of BCL-2, the ratio of BAX/BCL-2 increased significantly (Fig. 2 A–D). Overexpression of initiator cleaved-caspase 9 and executioner cleaved-caspase 3 after bilirubin treatment was observed, which was also inhibited by pretreatment with dexmedetomidine (Fig. 2 E–H). These results indicated dexmedetomidine inhibited caspase-executed cell apoptosis not only at the early stage but also in the late phase.

n of initiator cleaved-caspase 9 and executioner cleaved-caspase 3 after bilirubin treatment was observed, which was also inhibited by pretreatment with dexmedetomidine (Fig. 2 E–H). These results indicated dexmedetomidine inhibited caspase-executed cell apoptosis not only at the early stage but also in the late phase. TGFβ has been shown to play a key role in tissue repair after damage (31). TGFβ regulate protein phosphorylation in the cytoplasm and then transfer the proliferation signal to mitochondria (32). In this study, TGFβ was found to be downregulated after bilirubin treatment, which could be partially reversed by dexmedetomidine (Fig. 3, A and B). These results indicated that dexmedetomidine protected the epithelial cells from the apoptotic lesion through reversing the inhibition of TGFβ, which may facilitate the cell proliferation. Both p-mTOR and p42/44MAPK were found to be downregulated after bilirubin treatment, which could be reversed by pretreatment with dexmedetomidine (Fig. 4 C–F). Increasing evidence shows that mammalian target of rapamycin (mTOR) plays an important role in TGFβ-mediated cell proliferation, protein phosphorylation, and epithelial mesenchymal transition (33, 34). The downregulated p-mTOR indicated that proliferation of cells was inhibited, which was considered to contribute to the repair deficiency of alveoli after the insult of bilirubin. p42/44MAPK contributed to the proliferation of cells, dependent on mitochondria, just like mTOR (35, 36). Dexmedetomidine recovered bilirubin-induced inhibition of p-mTOR and p42/44MAPK indicated that dexmedetomidine may be helpful to repair the injured alveoli after bilirubin insult through increasing the proliferation of epithelial cells. Although the molecular mechanisms of bilirubin-induced lung cell injury is far more complicated, some molecular entities that may be responsible together with the interaction of dexmedetomidine have been found in our study but warrant further studies.

bin insult through increasing the proliferation of epithelial cells. Although the molecular mechanisms of bilirubin-induced lung cell injury is far more complicated, some molecular entities that may be responsible together with the interaction of dexmedetomidine have been found in our study but warrant further studies. HPS was reported to be present in 4–32% of adult patients with end-stage liver disease (37). HPS was characterized as intrapulmonary vascular dilatation and arterial hypoxemia. A high concentration of bilirubin was considered to cause HPS. In this study, HPS induced by hyperbilirubinemia via CBDL in rats. It is a widely accepted model in this area of research (38). Indeed, it was found in this study that damage to lung alveoli was aggravated following the increase in blood bilirubin concentration. Meanwhile, Paco2 was increased and Pao2 was decreased gradually after CBDL operation, and all of these indicated that high concentration bilirubin is very likely contributes to the injury of lung alveolar epithelial cells and respiratory failure as seen in human (2).

owing the increase in blood bilirubin concentration. Meanwhile, Paco2 was increased and Pao2 was decreased gradually after CBDL operation, and all of these indicated that high concentration bilirubin is very likely contributes to the injury of lung alveolar epithelial cells and respiratory failure as seen in human (2). Dexmedetomidine is widely used during the perioperative period due to its sedative, analgesic, sympatholytic, and hemodynamic effects on the patients undergoing surgery or under critical care. In addition, dexmedetomidine has anti-inflammatory and antioxidative effects on vital organs, such as lung and kidney (39, 40). The inhibitory effect of dexmedetomidine on the bilirubin-induced cell cycle arrest of A549 cells indicated that dexmedetomidine alleviates the injury of lung alveoli, possibly through promoting epithelial cell proliferation and initiating prompt repair. Dexmedetomidine, similar to other α2 adrenoceptor agonists, has a structure bearing similarity to imidazoline, and activation of imidazoline receptors may also contribute to its antiapoptotic effect (41). Pretreatment with α2 adrenoceptor antagonist atipamezole did not abolish the effect of dexmedetomidine on bilirubin-induced cell cycle arrest of A549 cells, which may indicate that the antiapoptotic effect of dexmedetomidine was not only due to an α2 adrenoceptor-dependent signaling pathway.

antiapoptotic effect (41). Pretreatment with α2 adrenoceptor antagonist atipamezole did not abolish the effect of dexmedetomidine on bilirubin-induced cell cycle arrest of A549 cells, which may indicate that the antiapoptotic effect of dexmedetomidine was not only due to an α2 adrenoceptor-dependent signaling pathway. The benefits of dexmedetomidine for patients with pulmonary injury in the operation room and ICU have been observed in clinical trials (42, 43). In this study, dexmedetomidine was given to CBDL rats IP daily during the experimental period attenuated the lung injury and prevented respiratory failure development in the CBDL rat (Fig. 7). In addition, dexmedetomidine significantly reduced the apoptosis of lung cells of CBDL rats, and these cells not only limited to alveolar epithelial cells but other types as well (Fig. 8). However, owing to its short action, dexmedetomidine is normally administered by a continuous IV infusion clinically, but a long-term infusion of dexmedetomidine can be not done in our study. Hence, the translational value of our study can be questionable. One can appreciate that this is always the case in a preclinical study. Nevertheless, our data clearly showed that dexmedetomidine could directly protect the lung injury in our model study because α2 adrenoceptors distribute widely including the cell line that was used in our study (data not shown), but as reported previously (44), the protection could also be due to its indirect effects of a decrease of sympathetic tone and an increase of vagal tone. Collectively, our work reported here could call further large animal study and clinical trials in this area of research.

that was used in our study (data not shown), but as reported previously (44), the protection could also be due to its indirect effects of a decrease of sympathetic tone and an increase of vagal tone. Collectively, our work reported here could call further large animal study and clinical trials in this area of research. In summary, we demonstrated that dexmedetomidine attenuated the bilirubin-induced injury of epithelial alveolar cells both in vitro and in vivo. Although further studies especially in a large animal model and clinical trials are needed to further validate the protective effects of dexmedetomidine on HPS, its inhibitory effect on cell apoptosis and promoting effect on cell survival and proliferation represent a promising anesthetic/sedative choice in treating the patients with chronic lung injury following severe liver disease perioperatively. ACKNOWLEDGMENTS We thank Yazhou Wu from the Third Military Medical University, China, for his advices on statistical analysis and thank James J. Sun from Cambridge University, United Kingdom, for his critical comment during manuscript preparation. *See also p. 2043. Supported, in part, by a grant from the Natural Science Foundation of Chongqing, China (No. cstc2013jcyjA1150). Dr. Cui received support for article research from Scholarship of Chinese Society of Anesthesiology, Beijing, China. The remaining authors have disclosed that they do not have any potential conflicts of interest.

Subarachnoid hemorrhage (SAH) is usually caused by rupture of a cerebral aneurysm located in the circle of Willis (1). It disproportionately affects a younger population (2) and is often fatal. The main mechanisms of brain injury after SAH include early brain injury (EBI), which occurs in the first 72 hours following ictus, and delayed cerebral ischemia (DCI), in 30% of patients unpredictably 3–14 days later (3). It is difficult to predict the evolution of cerebral injury following SAH using currently available methods. Clinical scoring systems have poor accuracy, and bedside clinical monitoring requires specialist interpretation. The focus of SAH research has been to investigate the mechanisms behind DCI, as this remains the most important cause of morbidity and mortality in patients who survive initial aneurysm rupture (4). Historically, DCI was attributed to spasm of the cerebral blood vessels; however, the relationship between angiographic evidence of vessel spasm and DCI is weak (5, 6). In addition, treatment of cerebral arterial vasoconstriction does not improve clinical outcome (7). There is now a growing body of evidence suggesting that changes occurring during EBI set the scene for the development of DCI (3, 7). An improved understanding and measurement of the processes associated with EBI could offer an opportunity to better predict DCI and improve patient outcomes (8).

ve clinical outcome (7). There is now a growing body of evidence suggesting that changes occurring during EBI set the scene for the development of DCI (3, 7). An improved understanding and measurement of the processes associated with EBI could offer an opportunity to better predict DCI and improve patient outcomes (8). Emerging evidence suggests that disruption to the nitric oxide (NO) signaling pathway may play a critical role in EBI (9–11). Endogenous NO has been proposed to exert a protective action after brain injury through a number of different pathways, including promoting cerebral blood flow, attenuating mitochondrial damage, and preventing cellular apoptosis (12). Following SAH, the integrity of these pathways becomes disrupted, resulting in a cascade of cellular injury resulting in cell death (13). The level of disruption of the NO signaling pathways has been shown to correlate with eventual outcome after SAH (10). Sodium nitrite (a prodrug) is suited as an exogenous NO donor in SAH patients because it is converted to NO under conditions of hypoxia or acidosis. As sodium nitrite has a relatively rapid onset of action, its effects are measurable in a brief time frame and, when combined with a measurement of cerebral injury, it may represent an ideal drug probe of EBI severity.

enous NO donor in SAH patients because it is converted to NO under conditions of hypoxia or acidosis. As sodium nitrite has a relatively rapid onset of action, its effects are measurable in a brief time frame and, when combined with a measurement of cerebral injury, it may represent an ideal drug probe of EBI severity. Quantitative electroencephalography (qEEG) uses power spectral analysis to obtain measures of the different components of the electroencephalography, sensitively detecting disturbed neuronal activity during the development of ischemia (14). SAH results in a variety of abnormalities in qEEG variables, specifically decreases in the alpha/delta frequency ratio (ADR), and a decrease in relative alpha power variability (15–18). These patterns have been shown to correlate with subsequent cerebral ischemia (19). Based on the evidence that disruption of the cerebral NO pathway is a main driver of the pathophysiologic processes occurring during EBI, we hypothesized that the qEEG response to a sodium nitrite, an NO donor, could be used as a dynamic probe of EBI severity and that this response would be linked to the subsequent development of DCI. MATERIALS AND METHODS Patients 18–80 years old admitted to the neuroscience ICU (NICU) at the John Radcliffe Hospital, Oxford, after having suffered severe aneurysmal SAH (World Federation of Neurosurgeons [WFNS] grade 3, 4, or 5 at the time of presentation) were eligible for inclusion in the study. No patient showed clinical evidence of DCI or angiographic cerebral arterial constriction at the time of the study.

hn Radcliffe Hospital, Oxford, after having suffered severe aneurysmal SAH (World Federation of Neurosurgeons [WFNS] grade 3, 4, or 5 at the time of presentation) were eligible for inclusion in the study. No patient showed clinical evidence of DCI or angiographic cerebral arterial constriction at the time of the study. Written informed consent was obtained from the next of kin of all participants and from participants if they regained capacity to consent. The study was approved by the South Central—Oxford C NHS Health Research Authority Ethics Committee 12/SC/0366. Exclusion criteria included contraindications to sodium nitrite, specifically severe cardiovascular compromise and preexisting methemoglobinemia. Next of kin provided information regarding smoking, medication, hypertension, and family history. All patients underwent standard clinical care that was not influenced by inclusion in this study. All were given nimodipine for 14 days. Computed tomography (CT) of the brain was performed in the event of lack of wakening in sedated patients or worsening of neurologic signs in awake patients, according to the clinical policy of the neuro-ICU. Patients who did not demonstrate neurologic deterioration did not undergo CT scanning in the acute period; however, they did later receive follow-up brain magnetic resonance imaging at 6 months as per local protocol, which confirmed the lack of new infarction.

atients, according to the clinical policy of the neuro-ICU. Patients who did not demonstrate neurologic deterioration did not undergo CT scanning in the acute period; however, they did later receive follow-up brain magnetic resonance imaging at 6 months as per local protocol, which confirmed the lack of new infarction. DCI was diagnosed based on consensus guidelines (20). In those patients who remained intubated and sedated, this was done by CT. Two patients (patients 2 and 4) had treatment withdrawn and subsequently died after CT evidence of widespread infarction secondary to DCI. The study investigators were not responsible for the clinical care of these patients. Treatment of DCI was via a standardized protocol involving hypertension, maintenance of euvolemia, and maintenance of a hemoglobin level above 8 g/dL.

and subsequently died after CT evidence of widespread infarction secondary to DCI. The study investigators were not responsible for the clinical care of these patients. Treatment of DCI was via a standardized protocol involving hypertension, maintenance of euvolemia, and maintenance of a hemoglobin level above 8 g/dL. Study Design Electroencephalography. Following definitive endovascular aneurysm treatment, each patient underwent a 2-hour period of continuous electroencephalographic monitoring (Porti 7 system, Twente Medical Systems International, Oldenzaal, The Netherlands) on one occasion, as soon as possible after endovascular securing of the aneurysm. We used a simplified electroencephalographic montage (16), a compromise between stable maintenance of recording and full coverage of all vascular territories. Seven to 13 unipolar electroencephalographic electrodes were used at the following positions defined according to the international 10–20 system: Cz, Fz, Pz, Fp1, Fp2, F3, F4, P3, P4, T3, T4, O1, and O2. Electroencephalographic data were digitized at a sampling rate of 2,048 Hz, with a high pass filter of 0.5 Hz and a low pass filter of 30 Hz.

troencephalographic electrodes were used at the following positions defined according to the international 10–20 system: Cz, Fz, Pz, Fp1, Fp2, F3, F4, P3, P4, T3, T4, O1, and O2. Electroencephalographic data were digitized at a sampling rate of 2,048 Hz, with a high pass filter of 0.5 Hz and a low pass filter of 30 Hz. Sodium Nitrite Infusion. An infusion of sodium nitrite at 10 μg/kg/min was commenced at the start of the second hour of electroencephalographic recording and continued for 1 hour. The dosing schedule was developed as a compromise between ensuring adequate delivery of cerebral NO and minimization of cardiovascular effects. Changes in infusion rates of sedative drug levels and vasopressors were minimized for the duration of the recording. Physiologic Measurements. Participants underwent simultaneous transcranial Doppler (TCD) monitoring. Insonation of the middle cerebral artery (MCA) M1 segment was performed unilaterally on the side with the best window using color-coded duplex ultrasound (EZ-Dop, DWL; 2-Mz probe; EZ-Dop GmbH, Singen, Germany). End-tidal Co2, end-tidal O2, arterial blood pressure, and pulse oximetry were recorded continuously and collected on a Power-1401 data acquisition interface (Cambridge Electronic Design, Cambridge, UK). Arterial Co2 values were collected once during the duration of the recording as part of routine clinical care. Missing values are as follows: patient 10 did not have an adequate TCD window and we were unable to record the intracranial pressure (ICP) waveform for patients 1 and 13 and end-tidal gases in patient 7.

UK). Arterial Co2 values were collected once during the duration of the recording as part of routine clinical care. Missing values are as follows: patient 10 did not have an adequate TCD window and we were unable to record the intracranial pressure (ICP) waveform for patients 1 and 13 and end-tidal gases in patient 7. Follow-Up. Each surviving patient was followed up at 3–6 months post rupture. Telephone follow-up was performed for patients who were unable to attend hospital. Primary outcome was defined as the presence or absence of DCI. Secondary outcome was assessed by modified Rankin scale (21) at 3 months, via structured standardized questions in person or by telephone. Quantitative Electroencephalographic Analysis Preprocessing was carried out using custom-written MATLAB (MathWorks, Natick, MA) code and the EEGLAB version 13.4.3b analysis toolbox (EEGLAB, San Diego, CA) (22). Datasets were rereferenced to the average of mastoid reference electrodes and band pass filtered from 0.5 to 15 Hz using a linear finite impulse response filter. Each electroencephalographic recording was visually inspected, and artifacts were manually removed.

ion 13.4.3b analysis toolbox (EEGLAB, San Diego, CA) (22). Datasets were rereferenced to the average of mastoid reference electrodes and band pass filtered from 0.5 to 15 Hz using a linear finite impulse response filter. Each electroencephalographic recording was visually inspected, and artifacts were manually removed. Spectral analysis was carried out using FieldTrip (23), a MATLAB software toolbox for electroencephalographic analysis. Data were windowed into 30-second segments that overlapped by 50%. Time-frequency analysis was performed using a multitaper spectral estimation using discrete prolate spheroidal (Slepian) sequences with 14 tapers and fast Fourier transform algorithm for each electrode channel. Five 60-second epochs were selected randomly from the first (baseline) and last 30 minutes of the recording (during infusion). The epochs were separated by at least 60 seconds to avoid autocorrelation. The corresponding frequency distribution in each epoch was identified, which enabled determination of power values in the following frequency bands: delta, 1–4 Hz; alpha, 8–12 Hz; and total low frequency power, 1–15 Hz.

ording (during infusion). The epochs were separated by at least 60 seconds to avoid autocorrelation. The corresponding frequency distribution in each epoch was identified, which enabled determination of power values in the following frequency bands: delta, 1–4 Hz; alpha, 8–12 Hz; and total low frequency power, 1–15 Hz. Two multivariate analyses were used to investigate the effects of sodium nitrite on ADR, including potential confounding factors as covariates (age, propofol, midazolam, and WFNS grade, using the R statistical package; R Foundation for Statistical Computing, Vienna, Austria). The square root of the ADR () was used as the response variable to achieve the required normality and homoscedasticity in the residuals. The goodness of fit was assessed via Shapiro-Wilk normality tests on fixed and random effect residuals and by calculation of correlation coefficient (R2).

Computing, Vienna, Austria). The square root of the ADR () was used as the response variable to achieve the required normality and homoscedasticity in the residuals. The goodness of fit was assessed via Shapiro-Wilk normality tests on fixed and random effect residuals and by calculation of correlation coefficient (R2). A multilevel linear regression model incorporated both baseline ADR and the ADR response to sodium nitrite as response variables. This model takes into account the repeated measures taken on each patient. A second model incorporated baseline as a covariate, with the ADR response to sodium nitrite as a response. This model used a Bayesian approach, allowing the baseline to be incorporated as a normally distributed random variable. This enabled direct investigation of the effect of sodium nitrite on the baseline and allowed us to explore whether the dependence of the sodium nitrite effect on baseline for patients who did not develop DCI differed significantly from those that subsequently developed DCI. Further description of these approaches can be found in the supplementary information (Supplemental Digital Content 1, http://links.lww.com/CCM/B945). Physiologic Data Analysis Waveform analysis was performed using custom-written MATLAB code, enabling calculation of average baseline and nitrite infusion values for TCD MCA velocity (MCAV), arterial blood pressure, end-tidal Co2, end-tidal O2, and ICP. Values were compared for each subject using paired t tests. p values less than 0.05 were considered significant.

as performed using custom-written MATLAB code, enabling calculation of average baseline and nitrite infusion values for TCD MCA velocity (MCAV), arterial blood pressure, end-tidal Co2, end-tidal O2, and ICP. Values were compared for each subject using paired t tests. p values less than 0.05 were considered significant. Sample Size Calculation As a study such as this has not been performed before, a formal power calculation was not possible. We anticipated a powerful effect on electroencephalographic power based on a previous resting electroencephalographic study, which showed a 24% lower ADR in nine patients who developed DCI (15). Animal studies have demonstrated that sodium nitrite seems to have a strong effect on the subsequent development of ischemia (24). We therefore expected 10–14 patients to demonstrate enough change to allow an effect of the drug to be detected. RESULTS Demographics, Treatment and Clinical Outcomes Fourteen patients (mean age, 52.8 yr [range, 41–69 yr]; 11 women) with spontaneous SAH successfully treated with endovascular coiling were recruited over a total study period of 13 months. All patients admitted to the NICU at the John Radcliffe Hospital were eligible for inclusion in the study.

nt and Clinical Outcomes Fourteen patients (mean age, 52.8 yr [range, 41–69 yr]; 11 women) with spontaneous SAH successfully treated with endovascular coiling were recruited over a total study period of 13 months. All patients admitted to the NICU at the John Radcliffe Hospital were eligible for inclusion in the study. All patients had modified Fisher grade 4 (thick SAH with intraventricular hemorrhage) and WFNS grades 3–5 on initial presentation. The WFNS grade 3 patients (patients 5, 7, 9, 10, and 12) were sedated and intubated either because of a subsequent drop in their Glasgow Coma Score secondary to seizures or subsequent episodes of vomiting. Detailed information on patient demographics and complications and levels of sedative drugs and vasopressors can be found in the supplementary information (Supplemental Digital Content 1, http://links.lww.com/CCM/B945). Data were collected between 2 and 4 days (mean, 3.5 d) following primary SAH. Because of cardiovascular instability and unknown behavior of sodium nitrite in this population at that time, it was not possible to collect data sooner than day 4 in patients 2 and 6. Seven of the study patients (50%) developed DCI as defined by consensus guidelines (20). This is in keeping with a higher incidence of DCI reported in previous studies of high-grade (WFNS grade, 3–5) SAH patients (25) (Table 1). TABLE 1. Demographics of Patients Recruited to the Study, Showing Age, World Federation of Neurosurgeons Grade, Aneurysm Location, and the Development of Delayed Cerebral Ischemia

Data were collected between 2 and 4 days (mean, 3.5 d) following primary SAH. Because of cardiovascular instability and unknown behavior of sodium nitrite in this population at that time, it was not possible to collect data sooner than day 4 in patients 2 and 6. Seven of the study patients (50%) developed DCI as defined by consensus guidelines (20). This is in keeping with a higher incidence of DCI reported in previous studies of high-grade (WFNS grade, 3–5) SAH patients (25) (Table 1). TABLE 1. Demographics of Patients Recruited to the Study, Showing Age, World Federation of Neurosurgeons Grade, Aneurysm Location, and the Development of Delayed Cerebral Ischemia All patients were diagnosed with hydrocephalus and were treated with external ventricular drainage immediately at admission to the neurosurgical centre. There was no rebleeding. All patients were treated with endovascular embolization. Three patients died (patients 2, 4, and 7), two from complications following severe DCI and one from cardiovascular instability. Three patients (patients 1, 5, and 14) developed sepsis secondary to chest infection, which were treated with IV antibiotics. qEEG Results Visual inspection of raw electroencephalographic data did not reveal any ictal or preictal activity in any of the recruited patients.

All patients were diagnosed with hydrocephalus and were treated with external ventricular drainage immediately at admission to the neurosurgical centre. There was no rebleeding. All patients were treated with endovascular embolization. Three patients died (patients 2, 4, and 7), two from complications following severe DCI and one from cardiovascular instability. Three patients (patients 1, 5, and 14) developed sepsis secondary to chest infection, which were treated with IV antibiotics. qEEG Results Visual inspection of raw electroencephalographic data did not reveal any ictal or preictal activity in any of the recruited patients. Spectrograms of the entire recording for two patients are shown in Figure 1. Figure 2 illustrates the raw data from each patient represented as an average of the five values before and during infusion, which are converted into a percentage change from baseline. Figure 3 demonstrates the percentage change from baseline ADR over time for the two groups. Figure 1. Single-channel spectrograms from a patient that did not develop delayed cerebral ischemia (DCI) (top) and from a patient who developed DCI (bottom).

Spectrograms of the entire recording for two patients are shown in Figure 1. Figure 2 illustrates the raw data from each patient represented as an average of the five values before and during infusion, which are converted into a percentage change from baseline. Figure 3 demonstrates the percentage change from baseline ADR over time for the two groups. Figure 1. Single-channel spectrograms from a patient that did not develop delayed cerebral ischemia (DCI) (top) and from a patient who developed DCI (bottom). Figure 2. Scatter plot showing the percentage change in alpha/delta frequency ratio (ADR) for each patient, grouped by the presence or absence of subsequent delayed cerebral ischemia (DCI). The horizontal bar represents the mean for each group. It can be seen that there is greater variability in the response to the drug in the group that did not develop DCI, suggesting that there may be a spectrum of neuronal damage that is unmasked by the drug, raising the possibility that there may be differing degrees of neuronal disruption after severe subarachnoid hemorrhage. Figure 3. Percentage change from baseline alpha/delta frequency ratio (ADR) over time, calculated by dividing the ADR at each time point by the baseline for the patients who subsequently did not develop delayed cerebral ischemia (DCI) (top) and for those who did develop DCI (bottom). Error bars represent the sem. The dashed vertical line represents the start of the sodium nitrite infusion.

DR) over time, calculated by dividing the ADR at each time point by the baseline for the patients who subsequently did not develop delayed cerebral ischemia (DCI) (top) and for those who did develop DCI (bottom). Error bars represent the sem. The dashed vertical line represents the start of the sodium nitrite infusion. Results from the linear regression model showed an increase in ADR from a mean of 0.033 (se = 0.008) to a mean of 0.055 (se = 0.010) in response to sodium nitrite in the no-DCI group (p < 0.0001) and a decrease in ADR from mean baseline of 0.056 (se = 0.010) to a mean of 0.050 (se = 0.009) in response to sodium nitrite in the DCI group (p = 0.006). There was a trend for a higher baseline ADR in the DCI group than the no-DCI group (p = 0.072). There was no evidence of an effect of propofol, midazolam, age, or WFNS grade on the ADR response. Comparing the change in ADR for the DCI group with the change in ADR in the no-DCI group demonstrated that the estimated mean for the DCI group was 0.028 (se = 0.003) less than that of the no-DCI group (p < 0.0001). The Bayesian model confirmed the results described above and revealed that the baseline ADR had a significant effect on the ADR response to the drug. More specifically, it showed that in the no-DCI group, the baseline ADR increases by 0.273 , whereas in the DCI group, the baseline ADR decreases by in response to the drug. This model therefore further clarified that the ADR response to the drug was significantly affected by whether the patient subsequently developed DCI.

specifically, it showed that in the no-DCI group, the baseline ADR increases by 0.273 , whereas in the DCI group, the baseline ADR decreases by in response to the drug. This model therefore further clarified that the ADR response to the drug was significantly affected by whether the patient subsequently developed DCI. Physiologic Data In response to sodium nitrite infusion, there was a significant decrease (p = 0.026) in mean arterial pressure (MAP) from a mean of 87 mm Hg (sd, 13 mm Hg) to 84 mm Hg (sd, 13 mm Hg). There were no significant changes in MCAV, ICP, end-tidal Co2, or end-tidal O2 values (Table 2). TABLE 2. Mean Physiologic Values Pre Versus During Sodium Nitrite Infusion DISCUSSION This pilot study investigated the qEEG response to IV sodium nitrite as a potential way to predict the development of DCI in patients with severe SAH. Patients who did not develop DCI showed a strong increase in the ADR in response to sodium nitrite, whereas patients who went on to develop DCI showed a small decrease or no change in the ADR. Our findings suggest mechanistic differences in the way the brain responds to increasing cerebral NO, depending on the severity of the injury. The no-DCI group responded as expected, demonstrating a move toward a less ischemic picture as demonstrated by increased ADR in response to NO repletion. However, the opposite was seen in the DCI group.

anistic differences in the way the brain responds to increasing cerebral NO, depending on the severity of the injury. The no-DCI group responded as expected, demonstrating a move toward a less ischemic picture as demonstrated by increased ADR in response to NO repletion. However, the opposite was seen in the DCI group. There are several potential explanations for these findings. One possibility is that there may be more severe brain injury and greater cerebral NO pathway dysfunction in patients who subsequently develop DCI. The dose of sodium nitrite chosen may have been insufficient for eliciting a response in the electroencephalography. A longer duration of infusion or a higher dose may have demonstrated different electroencephalographic changes.

njury and greater cerebral NO pathway dysfunction in patients who subsequently develop DCI. The dose of sodium nitrite chosen may have been insufficient for eliciting a response in the electroencephalography. A longer duration of infusion or a higher dose may have demonstrated different electroencephalographic changes. A further possibility is that increased cerebral NO may selectively vasodilate in areas where there is less tissue damage, diverting blood away from the more ischemic areas, causing a “steal” phenomenon. This would cause a deterioration in perfusion to injured areas and a move toward a more ischemic pattern on electroencephalography. It also implies loss of local autoregulation, already known to occur after severe SAH and which is linked to poor outcomes (26). In addition, delivering NO to areas of the brain with increased levels of free radicals may encourage the production of neurotoxic peroxynitrite, further contributing to cell damage (27). It is important to note that the two groups of patients were indistinguishable at presentation in terms of clinical severity (WFNS score), Fisher grade, or baseline ADR. Therefore, increasing cerebral NO unmasks cerebral neuronal and metabolic dysfunction that is otherwise not detectable. Electroencephalographic changes have been previously linked to the development of DCI but have required several days of recording (15, 28). Using a drug to probe electroencephalographic responses dynamically enabled the duration of recording to be considerably shorter than previous studies (average of 5 d; range 1–60 d) (29).

A further possibility is that increased cerebral NO may selectively vasodilate in areas where there is less tissue damage, diverting blood away from the more ischemic areas, causing a “steal” phenomenon. This would cause a deterioration in perfusion to injured areas and a move toward a more ischemic pattern on electroencephalography. It also implies loss of local autoregulation, already known to occur after severe SAH and which is linked to poor outcomes (26). In addition, delivering NO to areas of the brain with increased levels of free radicals may encourage the production of neurotoxic peroxynitrite, further contributing to cell damage (27). It is important to note that the two groups of patients were indistinguishable at presentation in terms of clinical severity (WFNS score), Fisher grade, or baseline ADR. Therefore, increasing cerebral NO unmasks cerebral neuronal and metabolic dysfunction that is otherwise not detectable. Electroencephalographic changes have been previously linked to the development of DCI but have required several days of recording (15, 28). Using a drug to probe electroencephalographic responses dynamically enabled the duration of recording to be considerably shorter than previous studies (average of 5 d; range 1–60 d) (29). TCD and ICP recordings remained stable in response to sodium nitrite, confirming that the observed qEEG changes were not because of changes in global cerebral blood flow. The patients who developed DCI were not followed up with repeat TCD or other measurements outside routine clinical care to detect angiographic vasospasm, but this would be an interesting addition to any future studies. The small drop in MAP is very unlikely to have clinical implications.

in global cerebral blood flow. The patients who developed DCI were not followed up with repeat TCD or other measurements outside routine clinical care to detect angiographic vasospasm, but this would be an interesting addition to any future studies. The small drop in MAP is very unlikely to have clinical implications. Limitations The number of patients recruited was small, and there is risk of bias because of the unblinded nature of this study. A double-blinded and randomized validation study in a larger group of patients is necessary. As sodium nitrite was infused for only hour, it is possible that steady state might not have been achieved in all patients. Future studies might investigate longer recording durations and investigate the electroencephalographic changes during the offset of sodium nitrite. It would also be important to investigate patients with a less severe degree of EBI by including WFNS grades 1 and 2 in future work. Although there was no control group, comparing baseline to values collected during the infusion allowed each patient to act as their own control, minimizing the effects of metabolic alterations, ICP changes, or effects of sedation. Although sedation will have affected the electroencephalographic pattern, it is unavoidable when studying this cohort of patients. The limited number of electroencephalographic electrodes hinders interpretation of spatial resolution, but fewer electroencephalographic electrodes increase the practicability of using this method in an intensive care setting.

the electroencephalographic pattern, it is unavoidable when studying this cohort of patients. The limited number of electroencephalographic electrodes hinders interpretation of spatial resolution, but fewer electroencephalographic electrodes increase the practicability of using this method in an intensive care setting. CONCLUSIONS In conclusion, we have shown that a 1-hour infusion of IV sodium nitrite can induce measurable qEEG changes capable of discriminating which patients eventually develop DCI. Our findings emphasize the importance of EBI as a window of therapeutic opportunity to institute aggressive neuroprotective strategies. Measuring qEEG responses to an NO donor, such as sodium nitrite, might also represent a potentially useful method for patient stratification, which may be useful in clinical trials. Therefore, with further validation, these findings demonstrate the potential to develop an electroencephalography-based, patient-specific tool to predict DCI. ACKNOWLEDGMENTS We thank Dr. Daniel Lunn for his invaluable advice regarding the statistical methods used in this article, Dr. David Garry for his comments on a previous version of this article, and Dr. Hilary Madder for supporting the study. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

ACKNOWLEDGMENTS We thank Dr. Daniel Lunn for his invaluable advice regarding the statistical methods used in this article, Dr. David Garry for his comments on a previous version of this article, and Dr. Hilary Madder for supporting the study. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by the National Institute for Health Research Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and University of Oxford. Ezra and Rowland were supported by the Oxford University Clinical Academic Graduate School. Rowland was supported by the Medical Research Council, United Kingdom. Drs. Garry, Rowland, Ezra, Westbrook, and Pattinson are named as coinventors on a provisional U.K. patent application titled “Use of cerebral nitric oxide donors in the assessment of the extent of brain dysfunction following injury.”

Supported, in part, by the National Institute for Health Research Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and University of Oxford. Ezra and Rowland were supported by the Oxford University Clinical Academic Graduate School. Rowland was supported by the Medical Research Council, United Kingdom. Drs. Garry, Rowland, Ezra, Westbrook, and Pattinson are named as coinventors on a provisional U.K. patent application titled “Use of cerebral nitric oxide donors in the assessment of the extent of brain dysfunction following injury.” Dr. Garry received support for this article research from the National Institutes of Health (NIH). Drs. Rowland and Ezra received support for this article research from the Research Councils UK (RCUK). Dr. Herigstad received support for this article research from other. Dr. Hayen’s institution received funding from the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and University of Oxford. Dr. Westbrook disclosed off-label product use (sodium nitrite used as a physiologic probe, not therapeutic). His institution received funding from Biomedical Research Committee OUH Oxford. Dr. Pattinson received support for this article research from the NIHR Oxford Biomedical Research Centre and disclosed off-label product use (sodium nitrite used as a physiologic probe). The remaining authors have disclosed that they do not have any potential conflicts of interest.

Sepsis remains a common cause for hospital admission and is the commonest reason for admission to ICUs. Despite improvements in management of sepsis, the prevalence of sepsis continues to increase and is the leading cause of death in critically ill patients, affecting approximately 750,000 U.S. patients annually with a mortality rate of approximately 25% (1, 2). Sepsis describes a complex clinical syndrome that results from a harmful or damaging host response to infection. A significant proportion of patients with sepsis go on to develop severe sepsis or septic shock (1, 2). Despite considerable research, there still remains a lack of targeted pharmacologic interventions to treat and improve outcomes from sepsis (3, 4). Over 1 billion people worldwide are believed to have vitamin D deficiency (VDD) (5). Epidemiologic studies have demonstrated that 25-hydroxyvitamin D3 (25[OH]D3) concentrations are related to geography and season (6, 7). In a study of middle-aged adults in the United Kingdom, 40% had serum 25(OH)D3 concentrations above 30 ng/L (75 nmol/L) in the summer months but this decreased to less than 17% in the winter (8). The prevalence and mortality of sepsis is higher during the winter when 25(OH)D3 concentrations are lower (9).

eason (6, 7). In a study of middle-aged adults in the United Kingdom, 40% had serum 25(OH)D3 concentrations above 30 ng/L (75 nmol/L) in the summer months but this decreased to less than 17% in the winter (8). The prevalence and mortality of sepsis is higher during the winter when 25(OH)D3 concentrations are lower (9). Vitamin D metabolites have important pleiotropic effects on human immunity, acting as modulators of cells of the innate and adaptive system (10). Biologically active 1,25(OH)2D3 directly enhances signaling to increase antimicrobial peptides, cathelicidin (LL-37, its active form), and β-defensin by the innate immune system (11). Gram-positive bacteria, invasive pneumococcal disease, meningococcal disease, and group A streptococcal disease are more common when 25(OH)D3 concentrations are low (12). VDD is associated with an increased risk of ICU admission and mortality in patients with pneumonia (13). Studies suggest VDD is common in critically ill patients and associated with adverse outcome (14, 15). Patients with sepsis who are not vitamin D sufficient (VDS) have an increased risk of mortality after critical care initiation (16). Recent meta-analyses support the association of VDD with increased susceptibility for severe infection, sepsis, and mortality in the critically ill (17, 18).

ith adverse outcome (14, 15). Patients with sepsis who are not vitamin D sufficient (VDS) have an increased risk of mortality after critical care initiation (16). Recent meta-analyses support the association of VDD with increased susceptibility for severe infection, sepsis, and mortality in the critically ill (17, 18). We believe that VDD is a determinant of the severity of sepsis because of effects on the host defense against infection. Previous studies of VDD have largely concentrated upon patients recruited within ICU. The prevalence and clinical significance in terms of outcomes of VDD in hospitalized mild sepsis outside the ICU is unknown. Our aim was to determine the prevalence and severity of VDD in a cohort of sepsis patients both within ICU and in a medical admissions unit (MAU) environment as soon as possible after hospital admission. We used murine sepsis models to explore the mechanistic link between preinjurious VDD and sepsis. Finally, we undertook studies in intratracheal lipopolysaccharide (IT-LPS)-treated VDD mice to demonstrate whether rescue therapy with vitamin D can attenuate inflammatory lung damage and dysregulated apoptosis associated with VDD (19). METHODS Detailed methods are available in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/C170).

Our aim was to determine the prevalence and severity of VDD in a cohort of sepsis patients both within ICU and in a medical admissions unit (MAU) environment as soon as possible after hospital admission. We used murine sepsis models to explore the mechanistic link between preinjurious VDD and sepsis. Finally, we undertook studies in intratracheal lipopolysaccharide (IT-LPS)-treated VDD mice to demonstrate whether rescue therapy with vitamin D can attenuate inflammatory lung damage and dysregulated apoptosis associated with VDD (19). METHODS Detailed methods are available in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/C170). Study Participants Patients were recruited from the acute MAUs (AMUs) and ICUs at two University Hospitals (Heart of England NHS Foundation Trust and the University Hospital Birmingham NHS Foundation Trust) between September 2012 and October 2014 as part of an observational sepsis study. Healthy elderly volunteers were recruited from a registry at the University of Birmingham and provided a cohort of healthy elderly individuals that would act as age-matched controls for the sepsis patients.

irmingham NHS Foundation Trust) between September 2012 and October 2014 as part of an observational sepsis study. Healthy elderly volunteers were recruited from a registry at the University of Birmingham and provided a cohort of healthy elderly individuals that would act as age-matched controls for the sepsis patients. Ethical Approvals All patients and healthy volunteers provided informed written consent. In circumstances where patients were unable to provide consent, a legal representative (personal or designated consultee) provided assent. Retrospective consent was sought where possible, when patients regained the ability to consent. This study received the appropriate ethical committee and local governance approvals (research ethics committee 11/YH0270). Inclusion Criteria Age greater than 18 years; documented new proven or suspected infection, and the presence of any two of the signs and symptoms of infection (WCC, > 11 or < 4 × 109/L; temperature, > 38°C or < 36°C; heart rate, > 90 per beats/min; or respiratory rate, > 20 per minute) for less than 24 hours. Patients were categorized as sepsis or severe sepsis, according to the presence of one or more organ failure at admission (20). Exclusion Criteria Recent chemotherapy, chronic steroid use, or use of other immunosuppressant drugs.

Inclusion Criteria Age greater than 18 years; documented new proven or suspected infection, and the presence of any two of the signs and symptoms of infection (WCC, > 11 or < 4 × 109/L; temperature, > 38°C or < 36°C; heart rate, > 90 per beats/min; or respiratory rate, > 20 per minute) for less than 24 hours. Patients were categorized as sepsis or severe sepsis, according to the presence of one or more organ failure at admission (20). Exclusion Criteria Recent chemotherapy, chronic steroid use, or use of other immunosuppressant drugs. Laboratory Methods 25(OH)D3 was measured by tandem mass spectrometry. The assay is calibrated using National Institute of Standards and Technology aligned material, achieving certification within Vitamin D External Quality Assessment Scheme, and described in detail previously (21). 25(OH)D3 concentrations below 50 nmol/L (20 ng/mL) were regarded as deficient. 25(OH)D3 concentrations between 50 and 75 nmol/L (30 ng/mL) were regarded as insufficient, with concentrations above 75 nmol/L (30 ng/mL) designated sufficient (22). Animal Materials and Methods Induction of VDD. Male wild-type (WT) C57Bl/6 mice once weaned were made VDD by feeding them a VDD chow (TD 89123: Harlan, Madison, WI) or maintained on normal chow for 6 weeks as a control.

Laboratory Methods 25(OH)D3 was measured by tandem mass spectrometry. The assay is calibrated using National Institute of Standards and Technology aligned material, achieving certification within Vitamin D External Quality Assessment Scheme, and described in detail previously (21). 25(OH)D3 concentrations below 50 nmol/L (20 ng/mL) were regarded as deficient. 25(OH)D3 concentrations between 50 and 75 nmol/L (30 ng/mL) were regarded as insufficient, with concentrations above 75 nmol/L (30 ng/mL) designated sufficient (22). Animal Materials and Methods Induction of VDD. Male wild-type (WT) C57Bl/6 mice once weaned were made VDD by feeding them a VDD chow (TD 89123: Harlan, Madison, WI) or maintained on normal chow for 6 weeks as a control. Murine Models of Sepsis and Lung Injury. Cecal ligation and puncture (CLP) and intratracheal instillation of LPS were performed as described previously (23) and in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/C170). Cell counts, protein permeability index (PPI), quantitative bacterial culture, cathelicidin-related antimicrobial peptide (CRAMP) levels in peritoneal lavage fluid (PLF), blood and bronchoalveolar lavage fluid (BALF), and peritoneal macrophage phagocytosis were compared between VDD and VDS mice in the CLP model. BALF cell counts, PPI, receptor for advanced glycation end-products (RAGE), and oxygen saturations were compared in the IT-LPS model.

P) levels in peritoneal lavage fluid (PLF), blood and bronchoalveolar lavage fluid (BALF), and peritoneal macrophage phagocytosis were compared between VDD and VDS mice in the CLP model. BALF cell counts, PPI, receptor for advanced glycation end-products (RAGE), and oxygen saturations were compared in the IT-LPS model. Statistics Initial power calculations for eight animals in each arm for the CLP experiment were based upon preliminary data to detect a change in lung PPI of 20% between VDD and VDS mice and six animals per arm to see a treatment effect on BALF neutrophils of 15% in the IT-LPS experiment. CLP experiments were performed in batches of four mice at different time points. Uneven numbers account for a failure in the experiment. Data were analyzed using SPSS for Windows 16.0 (SPSS, Chicago, IL). Data were tested for normality using a Shapiro-Wilks test with parametric data analyzed using unpaired t tests and a Mann-Whitney U test for nonparametric data. Data are expressed as mean (sd) unless otherwise indicated. A chi-square or Fisher exact test was used to compare proportions. A two-tailed p value of less than 0.05 was considered significant.

ng a Shapiro-Wilks test with parametric data analyzed using unpaired t tests and a Mann-Whitney U test for nonparametric data. Data are expressed as mean (sd) unless otherwise indicated. A chi-square or Fisher exact test was used to compare proportions. A two-tailed p value of less than 0.05 was considered significant. RESULTS Patient Cohorts Sepsis Patients. We enrolled 61 patients with sepsis—20 had mild sepsis and 41 had severe sepsis. Twelve patients were enrolled in the ICU and 49 were enrolled in AMUs. Two patients were transferred from AMUs to ICU postrecruitment. The etiology of sepsis was predominantly community-acquired pneumonia with a similar range of causes between sepsis and severe sepsis. As expected severity scores Acute Physiology and Chronic Health Evaluation (APACHE) II and SOFA score were significantly greater in severe sepsis than mild sepsis (Table 1) TABLE 1. Demographics and Severity of Patient Cohorts/Volunteers Healthy Donors. Blood from 20 healthy elderly donors were obtained—volunteers had no evidence of significant acute or chronic disease, normal spirometry, and were medication free. There were no significant differences in age or sex distribution between the healthy cohort, sepsis patients, and severe sepsis patients (Table 1).

Donors. Blood from 20 healthy elderly donors were obtained—volunteers had no evidence of significant acute or chronic disease, normal spirometry, and were medication free. There were no significant differences in age or sex distribution between the healthy cohort, sepsis patients, and severe sepsis patients (Table 1). 25(OH)D3 Concentrations Are Lower in Patients With Severe Sepsis Compared With Sepsis and Healthy Controls Of 61 patients, 41 met the criteria for severe sepsis with 14 requiring critical care at some point during their admission. Patients with severe sepsis had significantly lower 25(OH)D3 concentrations than either sepsis patients or controls (15.7 vs 49.5 vs 66.5 nmol/L; p = 0.0001). In contrast, patients with sepsis did not have significantly lower concentrations of 25(OH)D3 than healthy controls (Table 1) (online supplement Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). 25(OH)D3 Concentrations Are Lower in Sepsis Patients With Positive Microbial Cultures Twenty-two patients (36%) had positive cultures (blood/urine/sputum/BALF) for bacterial growth from samples taken as part of their clinical workup. Median 25(OH)D3 concentrations were significantly lower in patients who were culture positive (16.5 nmol/L) compared with culture negative patients (35.5 nmol/L; p = 0.0023) (online supplement Fig. 2, Supplemental Digital Content 1, http://links.lww.com/CCM/C170).

th from samples taken as part of their clinical workup. Median 25(OH)D3 concentrations were significantly lower in patients who were culture positive (16.5 nmol/L) compared with culture negative patients (35.5 nmol/L; p = 0.0023) (online supplement Fig. 2, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). There was an inverse relationship between 25(OH)D3 at baseline and standardized base excess and blood lactate (mmol/L) (Fig. 1, A and B). There was no relationship between 25(OH)D3 and age, sex, ethnicity, or the severity scores Sequential Organ Failure Assessment (SOFA) or APACHE II (data not shown). Figure 1. Regression plots for plasma 25-hydroxyvitamin D3 (25[OH]D3) and A, standardized base excess and B, blood lactate at admission. Regression was performed using Spearman’s rho for nonparametric data. C, Plasma 25(OH)D3 between survivors and nonsurvivors at 30 d, 90 d, and 1 yr. Box and whisker plots with median and Tukey’s distribution. 25(OH)D3 Concentrations Are Lower in Patients Who Die Within 30 Days of Admission Than Survivors Previous studies have suggested 25(OH)D3 concentrations are associated with mortality in ICU patients. In our whole cohort, there was a significant difference in median 25(OH)D3 concentrations between survivors (25.9 nmol/L) and nonsurvivors (14.5 nmol/L; p = 0.025) at 30 (11/61) days. This effect was not significant for 90-day mortality (15/61) or 1-year mortality (25/61) (Fig. 1C)

iated with mortality in ICU patients. In our whole cohort, there was a significant difference in median 25(OH)D3 concentrations between survivors (25.9 nmol/L) and nonsurvivors (14.5 nmol/L; p = 0.025) at 30 (11/61) days. This effect was not significant for 90-day mortality (15/61) or 1-year mortality (25/61) (Fig. 1C) Severe VDD Is Associated With Increased Risk of Death in Patients With Sepsis We have previously reported increased adverse postoperative inflammation and adverse events in patients undergoing esophagectomy who had severe deficiency before the operation (25[OH]D3, < 20 nmol/L) (19). In line with these data, sepsis patients with 25(OH)D3 concentrations below 20 nmol/L had a significant increased risk of 30-day mortality. Fisher exact test was significant at p value equal to 0.02 giving a relative risk of 4.71 (95% CI, 1.089–20.42). MURINE STUDIES Murine Vitamin D Status VDD was successfully established in WT C57BL/6 mice fed a deficient diet compared with a VDS diet (online supplement Table 1, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). 25(OH)D3 concentrations in the mice were equivalent to those seen in patients who died from severe sepsis. Deficiency did not result in a significant effect on serum calcium but was associated within reduced circulating bioactive 1,25(OH)2D3.

ment Table 1, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). 25(OH)D3 concentrations in the mice were equivalent to those seen in patients who died from severe sepsis. Deficiency did not result in a significant effect on serum calcium but was associated within reduced circulating bioactive 1,25(OH)2D3. VDD Is Associated With Increased Peritoneal/Systemic Bacteremia and Alveolar Bacterial Translocation After CLP VDD mice had a significantly higher bacterial load compared with VDS mice in all three compartments (peritoneal, blood, and alveolar) 16 hours after CLP (Fig. 2A). In sham experiments, there was an absence of bacteria as measured by colony forming units per milliliter in all three compartments confirming sterile procedure and surgery (data not shown). Figure 2. A, Effect of vitamin D deficient (VDD) on bacterial load in peritoneal lavage fluid (PLF), blood, and bronchoalveolar lavage fluid (BALF). Data presented as box and whisker plots with median and Tukey’s distribution, logarithmic scale to allow graphical representation. VDD n = 12; vitamin D sufficient (VDS) n = 11. Cathelicidin-related antimicrobial peptide (CRAMP) (murine cathelicidin) expression in B, PLF; C, serum; and D, BALF. Box and whisker plots with medians and Tukey’s distribution. VDD n = 12; VDS n = 11. Sham n = 4 per group. CRAMP was undetectable in sham-treated sera. CFU = colony forming units, CLP = cecal ligation and puncture, NS = not significant.

peptide (CRAMP) (murine cathelicidin) expression in B, PLF; C, serum; and D, BALF. Box and whisker plots with medians and Tukey’s distribution. VDD n = 12; VDS n = 11. Sham n = 4 per group. CRAMP was undetectable in sham-treated sera. CFU = colony forming units, CLP = cecal ligation and puncture, NS = not significant. CRAMP Is Reduced in VDD Mice CRAMP has been widely identified as a vitamin D-dependent antimicrobial peptide that binds bacteria. Cathelicidin rapidly destroys the lipoprotein membranes of microbes enveloped in phagosomes after fusion with lysosomes in macrophages (24). The CLP procedure increases CRAMP concentrations significantly in PLF, serum, and BALF in VDS mice. However, significantly lower concentrations were observed in VDD mice supporting the observation that VDD mice have reduced antimicrobial capacity (Fig. 2, B–D). CLP Does not Induce Alveolar Neutrophilia but Does Increase PPI, Which Is More Pronounced in VDD Little-to-no cellular recruitment into the alveolar compartment was observed at this time point; however, there was evidence of a mild increase in BALF PPI, suggesting early alveolar epithelial leak. This was significantly higher in VDD mice when compared with VDS mice (median, 3.30 [interquartile range (IQR), 2.69–4.64] vs 2.09 [IQR, 1.82–2.90]; p = 0.014) (online supplement Fig. 3, Supplemental Digital Content 1, http://links.lww.com/CCM/C170).

mild increase in BALF PPI, suggesting early alveolar epithelial leak. This was significantly higher in VDD mice when compared with VDS mice (median, 3.30 [interquartile range (IQR), 2.69–4.64] vs 2.09 [IQR, 1.82–2.90]; p = 0.014) (online supplement Fig. 3, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). VDD Is Associated With Increased Cellular Inflammation in the Peritoneum After CLP there was significant cellular recruitment in PLF. As major players of the acute inflammatory response, neutrophils and F4/80+ macrophages were enumerated within the peritoneal cavity. Significantly more neutrophils and F4/80+ macrophages were observed in VDD compared with VDS mice after CLP, with the neutrophil-to-macrophage ratio similar between both groups indicating a global increase in inflammatory mediators. PLF PPI was also significantly increased in VDD mice (median, 46.86 [IQR, 28.17–58.33] vs 29.81 [14.81–54.52]; p = 0.06) also indicative of more vascular damage in VDD mice after CLP (online supplement Fig. 4, Supplemental Digital Content 1, http://links.lww.com/CCM/C170).

g a global increase in inflammatory mediators. PLF PPI was also significantly increased in VDD mice (median, 46.86 [IQR, 28.17–58.33] vs 29.81 [14.81–54.52]; p = 0.06) also indicative of more vascular damage in VDD mice after CLP (online supplement Fig. 4, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). VDD Is Associated With Dysregulated Neutrophil Apoptosis and Impaired Peritoneal Macrophage Phagocytosis of Bacteria After CLP There was a significant increase in the number of apoptotic neutrophils in VDD compared with VDS PLF (median, 1.87 × 105 [IQR, 0.89–4.19 × 105] vs 0.51 × 105 [IQR, 0.20–0.54 × 105]; p = 0.007) (Fig. 3A), suggesting dysregulated neutrophil apoptosis and clearance of dying cells. To determine whether the increased bacteremia and/or accumulation of apoptotic neutrophils observed in VDD mice was due to impaired clearance by peritoneal macrophages, we assessed bacterial phagocytosis after CLP. Ex vivo phagocytosis of pHrodo-labeled Escherichia coliform bacteria was significantly reduced in F4/80+ macrophages isolated from PLF of VDD compared with VDS mice after CLP (median, 6.89% [IQR, 3.12–9.87] vs 21.12% [IQR, 17.56–24.29]; p = 0.029) (Fig. 3B). Figure 3. A, Peritoneal lavage fluid (PLF) neutrophil apoptosis. Box and whisker plot with median and Tukey’s distribution, vitamin D deficient (VDD) n = 12; vitamin D sufficient (VDS) n = 11. B, PLF macrophage phagocytosis of pHrodo-labeled E. coli bacteria. Box and whisker plot with median and Tukey’s distribution, Mann-Whitney test. VDD n = 4; VDS n = 4.

LF) neutrophil apoptosis. Box and whisker plot with median and Tukey’s distribution, vitamin D deficient (VDD) n = 12; vitamin D sufficient (VDS) n = 11. B, PLF macrophage phagocytosis of pHrodo-labeled E. coli bacteria. Box and whisker plot with median and Tukey’s distribution, Mann-Whitney test. VDD n = 4; VDS n = 4. Intraperitoneal (IP) Liquid Cholecalciferol (Vigantol) Rescue Therapy Attenuates Vitamin D-Related Inflammatory Damage Even When Given 6 Hours After IT-LPS Challenge In the United Kingdom, we can only use the CLP model of early sepsis due to home office animal license rules. To assess whether rescue therapy with vitamin D was effective postinjury, we studied our well-characterized IT-LPS challenge model, which we have previously reported results in exaggerated inflammation in VDD mice (23). Animals were administered 1,500 IU (75 µL) of cholecalciferol (Vigantol; Merck Serono GmbH, Darmstadt, Germany) or phosphate buffered saline control IP injection 6 hours postinjury and killed after 48 hours. Vigantol administration restored 25(OH)D3 concentrations in VDD mice to those similar to WT, which was sufficient to normalize the lung injury post- IT-LPS (online supplemental Fig. 5, Supplemental Digital Content 1, http://links.lww.com/CCM/C170). Vigantol treatment 6 hours post-IT-LPS reduced BALF PPI, RAGE (a marker of alveolar epithelial damage), and normalized BALF neutrophil apoptosis (Fig. 4, A–C). In addition, Vigantol attenuated the exaggerated decrease in oxygen saturations seen in this model with VDD mice (Fig. 4D).

t 1, http://links.lww.com/CCM/C170). Vigantol treatment 6 hours post-IT-LPS reduced BALF PPI, RAGE (a marker of alveolar epithelial damage), and normalized BALF neutrophil apoptosis (Fig. 4, A–C). In addition, Vigantol attenuated the exaggerated decrease in oxygen saturations seen in this model with VDD mice (Fig. 4D). Figure 4. Effect of intraperitoneal liquid cholecalciferol (Vigantol) rescue therapy upon (A) bronchoalveolar lavage fluid protein permeability index (BAL PPI); B, BAL receptor for advanced glycation end-products (BAL RAGE); C, BALF total apoptotic cell count, and D, arterial oxygen saturation in wild-type (WT) vitamin D sufficient mice given intratracheal lipopolysaccharide (IT-LPS), vitamin D deficient (VDD) mice given IT-LPS, and VDD mice given IT-LPS and 1,500 IU unit rescue therapy with cholecalciferol (postVIG) 6 hr postinjury. Mice were killed at 48 hr post-IT-LPS (n = 6 per arm). DISCUSSION We have confirmed, in a cohort of hospitalized patients with sepsis, that VDD is common, severe, and is associated with disease severity, bacterial positive culture, and 30-day mortality. To demonstrate causation of VDD as a driver of sepsis severity, our CLP mouse studies demonstrated exaggerated bacterial growth both locally and systemically, increased cellular inflammation, and dysregulated accumulation of apoptotic neutrophils in VDD mice. Using the IT-LPS challenge model, we demonstrate that the novel administration of IP cholecalciferol is an effective postinjury therapy when given 6 hours postinjury.

ated bacterial growth both locally and systemically, increased cellular inflammation, and dysregulated accumulation of apoptotic neutrophils in VDD mice. Using the IT-LPS challenge model, we demonstrate that the novel administration of IP cholecalciferol is an effective postinjury therapy when given 6 hours postinjury. We enrolled a mixed population of both mild and severe sepsis patients. VDD was both common and severe in patients with severe sepsis. 25(OH)D3 concentrations were lower in patients who died than survived as well as patients who grew culture positive bacterial specimens. Additionally, clinical markers of sepsis severity (lactate, metabolic acidosis) were associated with lower levels of 25(OH)D3 suggesting perhaps these measures could reflect a VDD population in sepsis. A criticism often levelled at observational studies, such as ours, is whether the VDD is a marker of critical illness or a mechanistic driver. Two recent meta-analyses of observational studies have confirmed a significant association between vitamin D status and susceptibility to sepsis (18), rates of infection, and 30-day mortality (17). Our findings are concordant with observational studies that have demonstrated that low vitamin D status upon admission is associated with sepsis (16), bacteremia (25), and acute respiratory distress syndrome (26, 27).

een vitamin D status and susceptibility to sepsis (18), rates of infection, and 30-day mortality (17). Our findings are concordant with observational studies that have demonstrated that low vitamin D status upon admission is associated with sepsis (16), bacteremia (25), and acute respiratory distress syndrome (26, 27). The murine studies sought to establish whether inducing VDD by diet before injury in mice leads to exaggerated sepsis and enhanced cellular inflammation/dysfunction. We successfully established severe deficiency in the mice, with concentrations of 25(OH)D3 similar to those who died from sepsis in our clinical cohort. This deficiency was reflected also in reduced circulating 1,25(OH)2D3, the bioactive form of vitamin D. In contrast, our WT mice had 25(OH)D3 and 1,25 (OH)2D3 concentrations similar to our mild sepsis patient population.

mice, with concentrations of 25(OH)D3 similar to those who died from sepsis in our clinical cohort. This deficiency was reflected also in reduced circulating 1,25(OH)2D3, the bioactive form of vitamin D. In contrast, our WT mice had 25(OH)D3 and 1,25 (OH)2D3 concentrations similar to our mild sepsis patient population. In the clinically relevant CLP model of early sepsis, VDD was associated with exaggerated bacterial growth in the peritoneal cavity, elevated systemic bacteremia as well as increased bacterial translocation to the alveolar compartment. This was associated with abnormal protein permeability of the peritoneal and alveolar capillary barrier. In the PLF, there was exaggerated cellular inflammation in VDD mice with evidence of impaired antibacterial responses in terms of CRAMP release and the ability of peritoneal macrophages to phagocytose E. coli. These cellular changes resulted in increased accumulation of apoptotic neutrophils in the PLF. Previous animal studies have shown a benefit of 1,25(OH)2D3 on sepsis-induced coagulopathy in rats (28) and our CRAMP results confirm findings by others of decreased antimicrobial peptide in VDD in sepsis and critical illness (29, 30). Our study is the first to report VDD as a predeterminant of sepsis and decreased macrophage phagocytosis in a relevant murine model. These data support our hypothesis that VDD is mechanistically important in driving sepsis and led us to the question of whether treating deficiency postinjury would be an effective therapy.

r study is the first to report VDD as a predeterminant of sepsis and decreased macrophage phagocytosis in a relevant murine model. These data support our hypothesis that VDD is mechanistically important in driving sepsis and led us to the question of whether treating deficiency postinjury would be an effective therapy. In the United Kingdom, the regulatory framework for animal experiments dictated that we could not keep VDD animals alive post-CLP for more than 16 hours because of serious adverse events so we were only able to model early sepsis using this technique. Our group has recently shown a detrimental effect of VDD with exaggerated lung injury, dysregulated cellular inflammation, and apoptosis, which manifested as reduced oxygenation in an IT-LPS direct murine lung injury model 48 hours after injury (19). We, therefore, used our IT-LPS model to test whether postinjury treatment of VDD mice attenuated the effects of VDD upon inflammatory injury.

d lung injury, dysregulated cellular inflammation, and apoptosis, which manifested as reduced oxygenation in an IT-LPS direct murine lung injury model 48 hours after injury (19). We, therefore, used our IT-LPS model to test whether postinjury treatment of VDD mice attenuated the effects of VDD upon inflammatory injury. Traditionally, vitamin D supplements have been given by mouth, intramuscular injection (cholecalciferol and ergocalciferol), or by IV infusion (calcitriol) with mixed results potentially due to poor absorption from muscle or the gut or a short IV half-life (31–33). We elected to test the effect of IP administration of 1,500 IU cholecalciferol liquid as a novel route to restore VDS—a dose that proved effective in restoring 25(OH)D3 concentrations back to those seen in WT mice. Postinjury cholecalciferol therapy was effective in reducing the exaggerated cellular inflammation, alveolar epithelial damage as measured by PPI and RAGE release, and reduced hypoxia (oxygen saturations) when given 6 hours after the insult supporting IP administration of cholecalciferol as a novel potential route of administration in patients as well as evidence that restoration of vitamin D levels may reduce inflammation with physiologic benefit.

asured by PPI and RAGE release, and reduced hypoxia (oxygen saturations) when given 6 hours after the insult supporting IP administration of cholecalciferol as a novel potential route of administration in patients as well as evidence that restoration of vitamin D levels may reduce inflammation with physiologic benefit. This study has limitations. First, patients were recruited up to 48 hours after hospital admission, so it is possible that the 25(OH)D3 concentrations seen reflected changes associated with sepsis rather than the cause. Second, this is a retrospective study of a small number of patients that could not control for patient comorbidities. The effects of sepsis and critical illness on the vitamin D metabolome are unknown and this complex interplay needs prospective large-scale studies that consider other preinsult comorbidities, chronic illness, nutritional status, and other confounders. It was for this reason we did the murine studies. In our CLP model, we studied early sepsis due to restrictions from the animal ethics committee. This meant that our animals had limited alveolar damage, which was why we undertook additional studies in the IT-LPS model. Although the VDD induced by diet design investigated whether pre-existing VDD is causal and a mechanistic driver to the severity of sepsis rather than the consequence of the sepsis insult in the murine model, it may not wholly explain the findings of the human study due to the lack of vitamin D status before sepsis and its effects on vitamin D status as discussed above. Finally, we have shown the effects of VDD in only two models of murine lung injury. Further work in other models related to sepsis particularly experimental pneumonia need to be undertaken.

dings of the human study due to the lack of vitamin D status before sepsis and its effects on vitamin D status as discussed above. Finally, we have shown the effects of VDD in only two models of murine lung injury. Further work in other models related to sepsis particularly experimental pneumonia need to be undertaken. In conclusion, we suggest that therapies aimed at restoring VDS in patients at risk of deficiency when they are admitted to hospital need to be developed to try and prevent the increasing healthcare burden of sepsis patients. Key to this will be establishing appropriate dosing regimens for vitamin D replacement in the critically ill patients both within and outside the ICU. ACKNOWLEDGMENTS We thank Sister Teresa Melody for her assistance in recruiting patients with sepsis. Supplementary Material *See also p. 376. Drs. Parekh and Patel are joint first authors. Drs. Parekh, Patel, Scott, Lax, Dancer, D’Souza, and Greenwood undertook patient recruitment sample analysis and laboratory work. Drs. Fraser, Gao, Sapey, Perkins, and Thickett designed the study and undertook analysis. Drs. Parekh, Perkins, and Thickett wrote the first draft. All authors have reviewed and approved the final version of article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

Drs. Parekh, Patel, Scott, Lax, Dancer, D’Souza, and Greenwood undertook patient recruitment sample analysis and laboratory work. Drs. Fraser, Gao, Sapey, Perkins, and Thickett designed the study and undertook analysis. Drs. Parekh, Perkins, and Thickett wrote the first draft. All authors have reviewed and approved the final version of article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Dr. Parekh disclosed other support from the British Lung Foundation and received support for article research from the Research Councils UK (RCUK), Wellcome Trust/COAF, and other. His institution received funding from the Medical Research Council UK. Dr. Patel received support for article research from the National Institute of Academic Anesthesia. Drs. Scott and Lax were supported by U.K. Medical Research Council. Dr. Dancer received support for article research from the RCUK. Her institution received funding from the U.K. Medical Research Council. Dr. D’Souza was supported by U.K. Medical Research Council. Dr. Greenwood was supported by the British Lung Foundation. Dr. Gao received support for article research from the National Institute for Health Research (NIHR) Senior Investigator Award UK. Dr. Sapey received support for article research from the RCUK. Her institution received funding from Medical Research Council and the British Lung Foundation. Dr. Perkins received support for article research from the RCUK, received funding from GlaxoSmithKline, and disclosed off-label product use (vitamin D). His institution received funding from the Medical Research Council. He is an NIHR Senior Investigator. Dr. Thickett received support for article research from the Wellcome Trust/COAF and RCUK. He was supported by the U.K. Medical Research Council. The remaining author has disclosed that he does not have any potential conflicts of interest.

Sepsis remains a leading cause of death in the world, regardless of advances in the critical care field (1). It has also been reported that a substantial number of sepsis patients surviving the initial hyperinflammatory phase of the disorder have a poor long-term outcome and low quality of life (2). Recently, a new disease concept, persistent inflammation, immunosuppression, and catabolism syndrome (PICS), was proposed to assess immunosuppression and nutritional status for patients with prolonged ICU length of stay (3). Additionally, it can contribute to mortality in the late phase of sepsis (4). Immunosuppression in sepsis is mainly caused by apoptosis of several immune cell types including macrophages, T cells, B cells, and dendritic cells. In particular, CD4+ T cells are reported to have considerable involvement in the pathophysiology of immunosuppression (5, 6). Several treatments to prevent immunosuppression have targeted apoptosis of immune cells in animal models (7–10). However, no definitive treatments for immunosuppression in clinical practice have been established so far, and effective treatments are still required to improve the late phase mortality of sepsis.

n (5, 6). Several treatments to prevent immunosuppression have targeted apoptosis of immune cells in animal models (7–10). However, no definitive treatments for immunosuppression in clinical practice have been established so far, and effective treatments are still required to improve the late phase mortality of sepsis. Apoptosis is known as “type 1 programmed cell death,” and autophagy as “type 2 programmed cell death.” Although a crosstalk between both types of cell death exists, the details of pathophysiology in sepsis have not been elucidated (11). Autophagy is a protein degradation system that is essential for cellular homeostasis. Its main functions are to recycle proteins, remove damaged organelles, eliminate microorganisms, and play a role in antigen presentation. The process of autophagy begins with the formation of an isolation membrane, which elongates to eventually form a double-membrane vesicle. Several autophagy genes, Atg3, Atg5, and Atg7, and microtubule-associated protein light chain 3 (LC3) are involved in the process (12). Recent clinical investigations and animal experiments have demonstrated that autophagy plays a protective role in several organs during sepsis (13–15). However, so far the relationship between immunosuppression and autophagy in sepsis has not been well documented. A recent study has shown that a deficiency in T cell autophagy increases mortality and immunosuppression using the cecal ligation and puncture (CLP) model (16). In this study, the T cell–specific Atg7 knockout was employed. However, the details of the mechanisms by which deficiency of autophagy induces apoptosis have not been demonstrated.

own that a deficiency in T cell autophagy increases mortality and immunosuppression using the cecal ligation and puncture (CLP) model (16). In this study, the T cell–specific Atg7 knockout was employed. However, the details of the mechanisms by which deficiency of autophagy induces apoptosis have not been demonstrated. In our study, we used T cell–specific Atg5 knockout mice (CD4-Cre recombinase/Atg5f/f mice: CD4-Cre/Atg5f/f mice) and performed a CLP procedure to prove our hypothesis that autophagy is related to apoptosis in sepsis. Consequently, we demonstrated the kinetics of autophagy in T cells during sepsis, crosstalk between autophagy and apoptosis, mitochondrial accumulation by deficiency of autophagy, and augmented interleukin (IL)-10 production by CD4+ T cells lacking autophagy. Based on the above results, we will discuss the influence on apoptosis and immunosuppression by blocking T cell autophagy in sepsis.

g sepsis, crosstalk between autophagy and apoptosis, mitochondrial accumulation by deficiency of autophagy, and augmented interleukin (IL)-10 production by CD4+ T cells lacking autophagy. Based on the above results, we will discuss the influence on apoptosis and immunosuppression by blocking T cell autophagy in sepsis. MATERIALS AND METHODS Mice Six- to 8-week-old male mice (C57BL/6) were used in the animal experiments. We used only male mice to minimize a variability of experimental results due to gender differences. Green fluorescent protein (GFP)-LC3 transgenic mice and Atg5f/f mice were kindly gifted. Transgenic Atg5f/f mice and CD4-Cre recombinase mice were crossed to generate T cell–specific Atg5 knockout mice (CD4-Cre/Atg5f/f mice), which deleted Atg5 gene in T cell by Cre-recombinase expression. All mice were acclimated to a 12-hour day/night cycle under specific pathogen-free conditions with food at least 1 week before the experiments. All experimental procedures were performed in strict accordance with the National Institute of Health guidelines and were approved by the Institutional Animal Care and Use Committees of Chiba University. CLP The above procedures were performed on mice as previously described (13). The details of the procedure are described in the supplementary data (Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Sham model mice were operated same as the CLP model, except for cecum ligation and puncture. Mice were euthanized at 6 and 24 hours to collect samples. Survival was observed every 12 hours and killed when they were moribund.

ibed in the supplementary data (Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Sham model mice were operated same as the CLP model, except for cecum ligation and puncture. Mice were euthanized at 6 and 24 hours to collect samples. Survival was observed every 12 hours and killed when they were moribund. Lymphocytes Isolation and Flow Cytometry Analysis Spleens were removed from anesthetized mice surgically and pressed with slide glasses gently. Then, they were washed with phosphate buffered saline, and RBCs were lysed with ammonium-chloride-potassium lysis buffer. Peripheral blood was obtained from inferior vena cava, and RBCs were lysed with distilled water. Centrifuged lymphocytes were resuspended in RPMI-1640 medium (Sigma-Aldrich, St. Louis, MO). Cell number and viability of lymphocytes resuspended in RPMI-1640 medium were quantified with TC20 Automated cell counter (Bio Rad, Hercules, CA). Surface marker staining, intracellular staining of p62, apoptosis, mitochondrial staining, and lysosomal staining were performed and analyzed with FACS caliber (BD Bioscience, San Jose, CA). Antibodies and reagents used in these experiments and detailed methods are described in the supplementary data (Supplemental Digital Content 1, http://links.lww.com/CCM/C114).

g of p62, apoptosis, mitochondrial staining, and lysosomal staining were performed and analyzed with FACS caliber (BD Bioscience, San Jose, CA). Antibodies and reagents used in these experiments and detailed methods are described in the supplementary data (Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Cell Sorting Prepared splenocytes were stained with anti-CD4 biotin antibody on ice for 30 minutes and then incubated with magnetic-streptavidin (Miltenyi Biotec, Bergisch Gladbach, Germany) for 15 minutes after washing. After resuspended in cell sorting buffer, CD4+ T cells were isolated with separate columns (Miltenyi Biotec) by negative selection. The purity of the CD4+ T cells was determined for more than 90%. The sorted cells were used for Western blotting, real time quantitative reverse transcription polymerase chain reaction, electron microscopic analysis, and cytokine secretion analysis. Detailed methods are described in the supplementary data (Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Statistical Analysis Data are presented as mean ± sd or absolute numbers and percentages as appropriate. We tested for differences between the two groups using an unpaired t test for continuous data and used two-way analysis of variance among different categoric independent variables. Statistical analyzes were conducted using the GraphPad Prism 6 (GraphPad Software, San Diego, CA). RESULTS Although Lymphocyte Autophagosomes are Increased by Septic Stimulation, the Process of Autophagy is Insufficient in a Murine Sepsis Model in CD4+ T Cells

Statistical Analysis Data are presented as mean ± sd or absolute numbers and percentages as appropriate. We tested for differences between the two groups using an unpaired t test for continuous data and used two-way analysis of variance among different categoric independent variables. Statistical analyzes were conducted using the GraphPad Prism 6 (GraphPad Software, San Diego, CA). RESULTS Although Lymphocyte Autophagosomes are Increased by Septic Stimulation, the Process of Autophagy is Insufficient in a Murine Sepsis Model in CD4+ T Cells To evaluate the autophagy kinetics of lymphocytes in sepsis, we performed in vitro assay to replicate the condition at first. Lymphocytes from GFP-LC3 mice were stimulated with anti-CD3/CD28 or lipopolysaccharide (LPS) for 48 hours. Mean fluorescence intensity (MFI) of GFP-LC3, which represents autophagosomes, was measured by flow cytometry. As shown in Figure 1A, MFI of GFP-LC3 in CD4+ T cells increased after activation. In B cells, the intensity transiently increased at 24 hours after the activation but declined by 48 hours. LC3B-II levels in CD4+ T cells and B cells were also increased by the stimulation sequentially in the Western blotting analysis. To determine the condition of autophagy flux, further analyses regarding selective substrates were conducted. P62, a marker protein for insufficient/inhibition of autophagy flux, was also increased in CD4+ T cells and B cells while LC3B-II levels were increased (Fig. 1B). These results suggest that autophagy in immune cells works insufficiently by mimicking septic stimulation despite increased appearance of autophagosomes.

ed. P62, a marker protein for insufficient/inhibition of autophagy flux, was also increased in CD4+ T cells and B cells while LC3B-II levels were increased (Fig. 1B). These results suggest that autophagy in immune cells works insufficiently by mimicking septic stimulation despite increased appearance of autophagosomes. Figure 1. Although lymphocyte autophagosomes are increased by septic stimulation, the process of autophagy is insufficient in a murine sepsis model in CD4+ T cells. A, Evaluation of autophagy in green fluorescent protein (GFP)-light chain 3 (LC3) transgenic mice by flow cytometry. After incubation with anti-CD3 and anti-CD28 antibodies for 48 hr for CD4+ lymphocytes and lipopolysaccharide (LPS) for B lymphocytes, mean fluorescence intensity (MFI) of harvested splenocytes was measured by flow cytometry for 48 hr. Representative data of three independent experiments are shown. B, Quantitation of autophagic protein, LC3-B and p62, was performed by western blotting. Lymphocytes were stimulated as the above procedure. The amount of each protein level was normalized by GAPDH. Representative data of independent three experiments are shown. C, FACS profiles for harvested splenocytes of GFP-LC3 transgenic mice. Mice underwent cecal ligation and puncture (CLP) and sham procedure and were killed at the time of 24 hr after the operation. Harvested splenocytes were stained with surface antigen markers and measured by flow cytometry. Representative data of five independent experiments are shown. D, Sequential MFI of lymphocytes from GFP-LC3 transgenic mice was compared between CLP- and sham-operated mice. n = 3–6 mice in each group. Results are shown as mean ± sd in a bar graph. Data are analyzed by student t test; #p < 0.05. E, MFI of Lysotracker staining lymphocytes from sham- and CLP-operated mice are shown. Each sample was stained with surface antigen markers. Data are expressed as mean and sd, and analyzed by two-way analysis of variance (ANOVA) and student t test. n = 3–4 mice in each group; #p < 0.05. F, MFI of p62 protein conjugated with fluorescent second antibody in lymphocytes from wild-type mice. Mice underwent CLP and sham procedure, and were killed at the time of 24 hr after the operation. Harvested splenocytes were stained with p62 and surface antigen markers concomitantly, and then measured by flow cytometry. Data are expressed as mean and sd, and analyzed by two-way ANOVA and student t test. n = 4–6 mice in each group; #p < 0.05.

ham procedure, and were killed at the time of 24 hr after the operation. Harvested splenocytes were stained with p62 and surface antigen markers concomitantly, and then measured by flow cytometry. Data are expressed as mean and sd, and analyzed by two-way ANOVA and student t test. n = 4–6 mice in each group; #p < 0.05. FACS = fluorescence activated cell sorting, GAPDH = glyceraldehyde-3-phosphate dehydrogenase.

ham procedure, and were killed at the time of 24 hr after the operation. Harvested splenocytes were stained with p62 and surface antigen markers concomitantly, and then measured by flow cytometry. Data are expressed as mean and sd, and analyzed by two-way ANOVA and student t test. n = 4–6 mice in each group; #p < 0.05. FACS = fluorescence activated cell sorting, GAPDH = glyceraldehyde-3-phosphate dehydrogenase. Based on the above results, we performed in vivo assay using a murine sepsis model. A CLP procedure was performed on GFP-LC3 mice and measured MFI of GFP-LC3 by flow cytometry. Autophagosomes, which were assessed by the MFI of GFP-LC3, were significantly increased in the CLP model over time, in CD4+ T cells (Fig. 1, C and D). However, no difference was observed in B cells between CLP- and sham-operated mice. Then, we investigated the lysosome kinetics with Lysotracker Red DND-99 by flow cytometry. Same as the kinetics of autophagosomes, MFI of Lysotracker in CD4+ T cells but not B cells was increased in the CLP-operated mice 24 hours after the procedure (Fig. 1E). As autophagosomes are subject to fusion with lysosomes to accomplish the autophagy process, these results indicated lysosomes accumulated in CD4+ T cells. To investigate the autophagy flux in the sepsis model, p62 protein levels of lymphocytes were measured by flow cytometry. MFI of p62 in the CLP mice was increased in CD4+ T cells but not in B cells after 24 hours compared with the sham-operated mice (Fig. 1F). We confirmed a double-membrane bound structure with membrane/organellar debris within and heterolysosomes/autolysosomes in CD4+ T cells from CLP-operated mice with transmission electron microscope (TEM) (Supplemental Fig. 1, Supplemental Digital Content 2, http://links.lww.com/CCM/C115; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Although it is suggested autophagosomes are increased by septic stimulation, autophagy process is insufficient for the CLP-operated mice.

electron microscope (TEM) (Supplemental Fig. 1, Supplemental Digital Content 2, http://links.lww.com/CCM/C115; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Although it is suggested autophagosomes are increased by septic stimulation, autophagy process is insufficient for the CLP-operated mice. Apoptosis Was Accelerated by Atg5 Deletion in T Cells During Sepsis To investigate the role of autophagy in sepsis, we generated T cell–specific autophagy knockout mice (CD4-Cre/Atg5f/f) and performed the CLP procedure. We confirmed the deleted allele in CD4+ T cells and CD8+ T cells obtained from CD4-Cre/Atg5f/f mice (Fig. 2A). The CLP procedure caused decrease of CD4+ and CD8+ T cells from spleen in the control mice. In CD4-Cre/Atg5f/f mice, percentages of CD4+ and CD8+ T cells from spleen decreased even in the sham-operated group (Fig. 2B). Since CD4-Cre/Atg5f/f mice are susceptible to the sham operation, we examined the levels of apoptosis in the autophagy-deficient mice in T cells. Annexin V positive CD4+ and CD8+ splenocytes were increased in CD4-Cre/Atg5f/f mice 24 hours after the operation (Fig. 2B). Significant differences in the proportions of apoptotic CD4+ T cells and CD8+ T cells were found between CD4-Cre/Atg5f/f mice and the control mice, and also between CLP- and sham-operated mice in CD4-Cre/Atg5f/f group (Fig. 2D). On the contrary, we did not observe any difference in B cells. Furthermore, we found significant differences in the proportions of apoptotic CD4+ T cells and CD8+ T cells from peripheral blood between CLP-operated Atg5f/f mice and CLP-operated CD4-Cre/Atg5f/f mice (Fig. 2E).

ted mice in CD4-Cre/Atg5f/f group (Fig. 2D). On the contrary, we did not observe any difference in B cells. Furthermore, we found significant differences in the proportions of apoptotic CD4+ T cells and CD8+ T cells from peripheral blood between CLP-operated Atg5f/f mice and CLP-operated CD4-Cre/Atg5f/f mice (Fig. 2E). Figure 2. Apoptosis was accelerated by Atg5 deletion in T cells during sepsis. A, Polymerase chain reaction amplification of genomic DNA from sorted lymphocytes in Atg5f/f and CD4-Cre/Atg5f/f mice. Atg5 flox allele and CD4-Cre allele are shown in the upper and middle figure, respectively. The positive controls are tail DNA from CD4-Cre/Atg5f/f mice. In the lower figure, the deleted allele is found in CD4+ T cells and CD8+ T cells extracted from CD4-Cre/Atg5f/f mice. The positive control is liver DNA from liver specific-Cre/Atg5f/f mice. B, Representative subpopulation of splenocytes in Atg5f/f and CD4-Cre/Atg5f/f mice. Mice were performed cecal ligation and puncture (CLP) or sham procedure. Splenocytes were stained with anti-CD4+/PE and anti-CD8+/APC. Representative data of five independent experiments are shown. C, FACS profiles of splenocytes stained with Annexin V and PI in Atg5f/f and CD4-Cre/Atg5f/f mice for 24 hr postoperatively. Samples were stained with anti-CD4+/PE and anti-CD8+/APC. Representative data of five independent experiments are shown. D, Subpopulation of Annexin V positive and PI negative staining splenocytes were shown for early (postoperative period, 6 hr) and late (postoperative period, 24 hr) apoptosis. Data are expressed as mean and sd; n = 8–10 mice in each group; #p < 0.05 was significance analyzed by two-way analysis of variance (ANOVA) and student t test. E, Subpopulation of Annexin V and PI staining lymphocytes from peripheral blood were shown for early (Annexin V positive and PI negative) and late (Annexin V positive and PI positive) phase of apoptosis. Data are expressed as mean and sd; n = 6–8 mice in each group; #p < 0.05 was significance analyzed by student t test between CLP-operated Atg5f/f mice and CLP-operated CD4-Cre/Atg5f/f mice. F, Relative RNA expression for apoptotic gene in sham, CLP-operated Atg5f/f mice and sham, and CLP-operated CD4-Cre/Atg5f/f mice. Total RNA in CD4+ splenic lymphocytes was extracted from experimental mice at 24 hr after the procedure, and then relative RNA expression for the several genes was analyzed.

re/Atg5f/f mice. F, Relative RNA expression for apoptotic gene in sham, CLP-operated Atg5f/f mice and sham, and CLP-operated CD4-Cre/Atg5f/f mice. Total RNA in CD4+ splenic lymphocytes was extracted from experimental mice at 24 hr after the procedure, and then relative RNA expression for the several genes was analyzed. Data are expressed as mean and sd; n = 8–10 mice in each group; #p < 0.05 was significance analyzed by two-way ANOVA and student t test. APC = allophycocyanin, Bcl-2 = B-cell leukemia/lymphoma 2, BIM = Bcl-2-like 11, FACS = fluorescence activated cell sorting, PDCD1 = programmed cell death 1, PE = phycoerythrin, PI = propidium iodide. The expression of Bcl-2-like 11 (BIM) and programmed cell death 1 (PDCD1), which have roles of apoptosis induction, was increased in CD4+ T cells from CD4-Cre/Atg5f/f mice. Otherwise, the gene expression of B-cell leukemia/lymphoma 2, which hampers apoptosis induction, was decreased in CD4+ T cells from CD4-Cre/Atg5f/f mice. Additionally, significant differences between CLP- and sham-operated CD4-Cre/Atg5f/f mice were observed in the gene expression of BIM (Fig. 2F). Based on the above results, it was suggested that deletion of T cell autophagy caused acceleration of apoptosis in the CLP model.

ed in CD4+ T cells from CD4-Cre/Atg5f/f mice. Additionally, significant differences between CLP- and sham-operated CD4-Cre/Atg5f/f mice were observed in the gene expression of BIM (Fig. 2F). Based on the above results, it was suggested that deletion of T cell autophagy caused acceleration of apoptosis in the CLP model. Mitochondrial Mass Was Increased by Autophagy Deficiency in T Cells As autophagy has a crucial role to remove damaged organelles including mitochondria, we speculated that blocking autophagy would affect mitochondrial function in T cells during sepsis. Mitochondrial mass, which was assessed by MFI of Mitotracker FM green, was significantly increased by the CLP procedure in the control mice (Fig. 3). Deficiency of T cell autophagy in CD4-Cre/Atg5f/f mice increased mitochondrial mass by both of the sham and CLP operation (Fig. 3). However, no differences were found between the CLP model and sham model in B cells from CD4-Cre/Atg5f/f (Fig. 3). As apoptosis was induced by autophagy deficiency in T cells from the above results, we evaluated mitochondrial membrane potential (MMP). In the control mice, relative MMP was increased after the CLP procedure. No increase of MMP was observed in autophagy-deficient CD4+ T cells. Furthermore, relative MMP in the CLP-operated Atg5 knockout mice is significantly lower than that of the control mice (Supplemental Fig. 2, Supplemental Digital Content 3, http://links.lww.com/CCM/C116; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114).

observed in autophagy-deficient CD4+ T cells. Furthermore, relative MMP in the CLP-operated Atg5 knockout mice is significantly lower than that of the control mice (Supplemental Fig. 2, Supplemental Digital Content 3, http://links.lww.com/CCM/C116; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Figure 3. Mitochondrial mass was increased by autophagy deficiency in T cells. FACS profiles of Mitotracker FM green staining lymphocytes in sham, cecal ligation and puncture (CLP)-operated Atg5f/f mice and sham, and CLP-operated CD4-Cre/Atg5f/f mice at 24 hr after the procedure. Representative data of five independent experiments are shown. FACS = fluorescence activated cell sorting, MFI = mean fluorescence intensity. IL-10 Secretion Was Increased by Suppression of Autophagy in T Cells We investigated the cytokine secretion in the supernatant from stimulated CD4+ T cells. IL-10 concentrations in the CLP-operated CD4-Cre/Atg5f/f mice were significantly increased (Fig. 4A). This was also confirmed by messenger RNA expression (Fig. 4B). Similarly, interferon (IFN)-γ and IL-4 concentrations were increased in CD4-Cre/Atg5f/f mice. However, no difference was observed between the CLP- and sham-operated CD4-Cre/Atg5f/f mice (Fig. 4A; and Supplemental Fig. 3, Supplemental Digital Content 4, http://links.lww.com/CCM/C117; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114).

L-4 concentrations were increased in CD4-Cre/Atg5f/f mice. However, no difference was observed between the CLP- and sham-operated CD4-Cre/Atg5f/f mice (Fig. 4A; and Supplemental Fig. 3, Supplemental Digital Content 4, http://links.lww.com/CCM/C117; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Figure 4. Interleukin (IL)-10 secretion was increased by blocking autophagy in T cells, and T cell autophagy affects the late phase mortality of sepsis. A, After lymphocytes were harvested from spleen, they were stimulated with anti-CD3 and anti-CD28 antibodies for 24 hr. Then, we measured IL-2, interferon (IFN)-γ, and IL-10 concentrations in the supernatant fluid of incubated lymphocytes by enzyme-linked immunosorbent assay. Data are expressed as mean and sd; n = 8–9 mice in each group; #p < 0.05 was significance analyzed by two-way analysis of variance (ANOVA) and student t test. B, Relative RNA expression for cytokine gene in sham, cecal ligation and puncture (CLP)-operated Atg5f/f and sham, and CLP-operated CD4-Cre/Atg5f/f mice. Total RNA in CD4+ lymphocytes was extracted from experimental mice at 24 hr after the procedure, and then relative RNA expression of IL-2, interferon-γ (IFNG) and IL-10 were analyzed. Data are expressed as mean and sd; n = 8–10 mice in each group; #p < 0.05 was significance analyzed by two-way ANOVA and student t test. C, Survival rates between CLP-operated Atg5f/f mice and CLP-operated CD4-Cre/Atg5f/f mice. Survival rates were observed every 12 hr. n = 18–20 mice in each group. p < 0.05 was significance analyzed by Kaplan-Meier method and log-rank test.

group; #p < 0.05 was significance analyzed by two-way ANOVA and student t test. C, Survival rates between CLP-operated Atg5f/f mice and CLP-operated CD4-Cre/Atg5f/f mice. Survival rates were observed every 12 hr. n = 18–20 mice in each group. p < 0.05 was significance analyzed by Kaplan-Meier method and log-rank test. Autophagy Deficiency in T Cells Decreases the Survival Rate in the Sepsis Murine Model Because apoptotic cell death in T cells was the most prominent at 24 hours after CLP surgery (Fig. 2, C–E), we needed to evaluate the survival using a model that live longer than 24 hours at least. Therefore, we performed a survival study using a less severe CLP model. The mortality is supposed to be lowered than the previous model at the same time course. To verify this condition, we used a smaller needle (27G) for puncture instead of a 23G needle. Consequently, we found that the mortality rates in CD4-Cre/Atg5f/f mice significantly increased compared with the control mice (p = 0.016) (Fig. 4C). As shown in Figure 4C, the mortality rate for controls relative to the conditional knockout mice was a more robust improvement in survival at 72 hours, that is, just after occurrence of the T cell apoptosis.

lity rates in CD4-Cre/Atg5f/f mice significantly increased compared with the control mice (p = 0.016) (Fig. 4C). As shown in Figure 4C, the mortality rate for controls relative to the conditional knockout mice was a more robust improvement in survival at 72 hours, that is, just after occurrence of the T cell apoptosis. DISCUSSION In this study, we demonstrated several findings regarding T cell autophagy in sepsis. First, an autophagy process in CD4+ T cells is insufficient during sepsis despite an increase of autophagosomes. Second, blockade of T cell autophagy accelerates apoptosis. Third, mitochondrial accumulation in T cells occurs via a blockade of autophagy during sepsis. Fourth, IL-10 production is increased in CD4+ T cells by blockade of autophagy, which may drive a further immunosuppressive state. Finally, a deficiency of autophagy in T cells decreases the survival rate in the murine sepsis model. These findings suggest that T cell autophagy plays a protective role against apoptosis and immunosuppression in the murine sepsis model.

ells by blockade of autophagy, which may drive a further immunosuppressive state. Finally, a deficiency of autophagy in T cells decreases the survival rate in the murine sepsis model. These findings suggest that T cell autophagy plays a protective role against apoptosis and immunosuppression in the murine sepsis model. As interest in the process of autophagy has developed, Crouser et al (17) have shed light on the interaction between autophagy and sepsis. In their report, they demonstrated that mitochondrial depletion was related to the removal of the damaged organelles by autophagy in a murine sepsis model. Furthermore, Watanabe et al (18) observed an increase of liver autophagosomes in sepsis patients for the first time. Although the role of autophagy in sepsis had not been clearly understood, Takahashi et al (13) demonstrated a protective role of autophagy in liver using CLP-operated mice. However, the kinetics of autophagy in T cells during sepsis have still not been elucidated. We used GFP-LC3 transgenic mice in which the autophagic activity can be monitored objectively over time using flow cytometry (19). Furthermore, we separated CD4+ T cells and confirmed the images of increased autophagosomes and autolysosomes in CLP relative to sham with our TEM data (Supplemental Fig. 1, Supplemental Digital Content 2, http://links.lww.com/CCM/C115; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Also, we found remarkable mitochondrial damages in CD4+ T cells of CLP-operated mice. To our knowledge, ours is the only study in which T cell autophagy in sepsis was monitored in such detail to date.

gital Content 2, http://links.lww.com/CCM/C115; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/C114). Also, we found remarkable mitochondrial damages in CD4+ T cells of CLP-operated mice. To our knowledge, ours is the only study in which T cell autophagy in sepsis was monitored in such detail to date. Our results demonstrated that mitochondria accumulated in the T cells of the CD4-Cre/Atg5f/f mice. It is consistent with the previous report that T cells lacking the autophagy-related genes Atg5 or Atg7 have inferior survival rates to those expressing the genes and contain expanded mitochondria (20, 21). As damaged organelles accumulate, cells are unable to maintain their physiologic functions. Although we did not demonstrate a direct relationship between apoptosis induction and mitochondrial accumulation, it has previously been shown that deletion of autophagy in T cells induces abnormal reactive oxygen species (ROS) production and apoptosis, and leads to increased mitochondrial content (21). From the above findings, it is suggested that autophagy plays a critical role to maintain mitochondrial homeostasis in sepsis.

n, it has previously been shown that deletion of autophagy in T cells induces abnormal reactive oxygen species (ROS) production and apoptosis, and leads to increased mitochondrial content (21). From the above findings, it is suggested that autophagy plays a critical role to maintain mitochondrial homeostasis in sepsis. We demonstrated the intimate interaction between autophagy and apoptosis in T cells. Although sepsis causes the increase of ROS and accumulation of damaged mitochondria, damaged mitochondria is removed by autophagic machinery to prevent apoptosis (Fig. 5A). On the other hand, blocking T cell autophagy causes accumulation of damaged mitochondria and acceleration of T cell apoptosis with augmented expression of BIM and PDCD1. Eventually, these changes result in the immunosuppressive status (Fig. 5B). It is supposed that sepsis patients, who survive the acute phase of the disorder, will suffer from PICS under the suppression of T cell autophagy often caused by nutritional stress, insulin infusion, and sustained inflammation, etc.

and PDCD1. Eventually, these changes result in the immunosuppressive status (Fig. 5B). It is supposed that sepsis patients, who survive the acute phase of the disorder, will suffer from PICS under the suppression of T cell autophagy often caused by nutritional stress, insulin infusion, and sustained inflammation, etc. Figure 5. Interaction between autophagy and apoptosis in T cell. The summary figure illustrates the crosstalk between autophagy and apoptosis in T cells during sepsis. A, Septic stresses cause the increase of reactive oxygen species (ROS) and accumulation of damaged mitochondria. Damaged organelles are cleared by autophagic machinery and these events prevent apoptosis. B, Blocking T cell autophagy accelerates T cell apoptosis and eventually results in the immunosuppressive status. Bcl-2 = B-cell leukemia/lymphoma 2, PDCD1 = programmed cell death 1.

ROS) and accumulation of damaged mitochondria. Damaged organelles are cleared by autophagic machinery and these events prevent apoptosis. B, Blocking T cell autophagy accelerates T cell apoptosis and eventually results in the immunosuppressive status. Bcl-2 = B-cell leukemia/lymphoma 2, PDCD1 = programmed cell death 1. Autophagy is required to maintain cytokine production. Production of IFN-γ and IL-2 in T cells is decreased due to blockade of Atg7 gene, which is important for macroautophagy (22). Furthermore, in the septic murine model, cytokine production of Th1, Th2, and Th17 were reduced (16). However, we demonstrated that IFN -γ and IL-10 concentrations were elevated in CD4-Cre/Atg5f/f mice. In particular, IL-10 production by T cells from CLP-operated CD4-Cre/Atg5f/f mice was markedly higher than that of the control mice, suggesting autophagy has a role in controlling cytokine production to some extent. The plausibility of this result is supported by the report that macrophages without the autophagy protein Atg16L1 show enhanced IL-1β and IL-18 production by LPS stimulation (23). Furthermore, inhibition of autophagy promotes IL-1β production in macrophages and dendritic cells, and augments secretion of IFN-γ by T cells (24). Consequently, sustained IL-10 production likely contributes to compensatory anti-inflammatory response syndrome and leads to immunosuppression together with apoptosis of immune cells. Supporting the above explanation, we observed a significant difference in mortality between the CD4-Cre/Atg5f/f mice and the control mice using the less severe CLP model. However, the survival benefit was more robust in control mice at around 72 hours after CLP surgery, which is thought to have been caused by the beneficial effects of Atg5 on T cell viability.

ved a significant difference in mortality between the CD4-Cre/Atg5f/f mice and the control mice using the less severe CLP model. However, the survival benefit was more robust in control mice at around 72 hours after CLP surgery, which is thought to have been caused by the beneficial effects of Atg5 on T cell viability. We have demonstrated that blocking autophagy leads to apoptosis and immunosuppression in concordance with an insufficient autophagy process in T cells. To summarize, acceleration of autophagy might alleviate immunosuppression. We can speculate on several treatment candidates that modulate autophagy kinetics for immunosuppression. The efficacy of these interventions for T cell autophagy should be clarified in the future investigations. CONCLUSIONS We demonstrated that blocking autophagy accelerated apoptosis and increased mortality in concordance with the insufficient autophagy process in CD4+ T cell in the septic murine model, suggesting that T cell autophagy plays a protective role against apoptosis and immunosuppression in sepsis.

We have demonstrated that blocking autophagy leads to apoptosis and immunosuppression in concordance with an insufficient autophagy process in T cells. To summarize, acceleration of autophagy might alleviate immunosuppression. We can speculate on several treatment candidates that modulate autophagy kinetics for immunosuppression. The efficacy of these interventions for T cell autophagy should be clarified in the future investigations. CONCLUSIONS We demonstrated that blocking autophagy accelerated apoptosis and increased mortality in concordance with the insufficient autophagy process in CD4+ T cell in the septic murine model, suggesting that T cell autophagy plays a protective role against apoptosis and immunosuppression in sepsis. ACKNOWLEDGMENTS We thank Dr. Noboru Mizushima (University of Tokyo, Japan) for kindly gifted green fluorescent protein (GFP)-microtubule-associated protein light chain 3 (LC3) transgenic mice and Atg5f/f mice; Dr. Masafumi Arima (institute of Dokkyou medical university, Japan) for kindly gifted CD4 T cell Cre recombinase (CD4-Cre) mice; Mrs. Aya Goda and Yoichi Teratake for their excellent technical assistance; Dr. Paul E. Swanson (Foothills Medical Centre, Canada) for validation of pathologic findings; and Ms. Yoshiko Ohashi for her excellent secretarial assistance. Supplementary Material *See also p. 145. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

ACKNOWLEDGMENTS We thank Dr. Noboru Mizushima (University of Tokyo, Japan) for kindly gifted green fluorescent protein (GFP)-microtubule-associated protein light chain 3 (LC3) transgenic mice and Atg5f/f mice; Dr. Masafumi Arima (institute of Dokkyou medical university, Japan) for kindly gifted CD4 T cell Cre recombinase (CD4-Cre) mice; Mrs. Aya Goda and Yoichi Teratake for their excellent technical assistance; Dr. Paul E. Swanson (Foothills Medical Centre, Canada) for validation of pathologic findings; and Ms. Yoshiko Ohashi for her excellent secretarial assistance. Supplementary Material *See also p. 145. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by JSPS KAKENHI grant number 15K20333. Dr. Watanabe received support for article research from JSPS KAKENHI grant number 15K20333. The remaining authors have disclosed that they do not have any potential conflicts of interest.

Sepsis is the underlying cause in half the cases of acute kidney injury (AKI) (1). Up to 5% of total ICU admissions require renal replacement therapy for AKI and a third of them die (1, 2). Despite the clinical impact of septic AKI management is limited to supportive care. The current lack of a sensitive marker of early parenchymal kidney injury reduces the window of opportunity for early effective intervention to prevent renal dysfunction and failure. Numerous blood and urine markers of renal injury/dysfunction are being promoted but, at present, remain poorly characterized. Few studies have investigated the utility of biomarkers in predicting recovery from AKI (3). A complementary panel of markers will likely enhance interpretation of this dynamic process, although studies in septic patients are confounded by an inability to precisely time the onset of sepsis. Thus, the temporal relationship of these biomarkers to the onset, progression, and recovery of renal dysfunction and injury is unknown. Serial measurements of a panel of renal biomarkers in a well-characterized animal model with a defined onset of polymicrobial sepsis followed by recovery will provide invaluable information translatable to the patient.

these biomarkers to the onset, progression, and recovery of renal dysfunction and injury is unknown. Serial measurements of a panel of renal biomarkers in a well-characterized animal model with a defined onset of polymicrobial sepsis followed by recovery will provide invaluable information translatable to the patient. Animal models of sepsis offer the advantage of knowing precisely when “time zero” occurs and also allow control of volume status and other conditions in a relatively homogenous population. However, to be representative of the human condition, they must simulate many of the physiologic and pathologic aspects of sepsis, including a proper infectious insult, an adequate duration of study, and fluid resuscitation to avoid the consequences of untreated hypovolemia leading to organ hypoperfusion. We used a clinically relevant and well-characterized model of sepsis and recovery to define parallel changes in global hemodynamics, biochemistry, renal histology, serum cytokines, and a panel of renal biomarkers (4, 5). MATERIALS AND METHODS In Vivo Experiments Male Wistar rats (Charles River, Margate, Kent, United Kingdom) weighing 300–375 g were used. Experiments were performed under a Home Office Project License (PPL 70/7029) and local University College London Ethics Committee approval. The rats were housed in cages of six on a 12-12–hour light-dark cycle. Six time points were selected to represent early (3, 6, 12 hr), established (24 hr), and recovery (48 and 72 hr) phases of sepsis.

ormed under a Home Office Project License (PPL 70/7029) and local University College London Ethics Committee approval. The rats were housed in cages of six on a 12-12–hour light-dark cycle. Six time points were selected to represent early (3, 6, 12 hr), established (24 hr), and recovery (48 and 72 hr) phases of sepsis. All invasive and imaging techniques were performed under a brief period of isoflurane anesthesia with the animal breathing spontaneously, as described previously (4, 5). Following tunneled internal jugular line placement, rats were placed in individual cages mounted on the tether/swivel system to secure the IV catheter and allow unimpeded movement with free access to food and water. The reasons for tunneled central venous catheter (CVC) insertion are two-fold; first, to prevent the line from being pulled out by the animal and thus enable ongoing fluid resuscitation and second, to reduce the risk of CVC-related infection. At 24-hour postinstrumentation, sepsis was induced by intraperitoneal injection of fecal slurry (see Appendix for more detail). This was not performed in sham animals to prevent inadvertent bowel perforation. Antibiotics were not administered to avoid any confounding drug-induced nephrotoxicity. The optimal volume and rate of fluid administration (with 1:1 mix of 5% glucose and Hartmann’s solution) to maintain intravascular volume based on echo variables have been previously determined (4). Sham animals received a similar fluid regimen.

ere not administered to avoid any confounding drug-induced nephrotoxicity. The optimal volume and rate of fluid administration (with 1:1 mix of 5% glucose and Hartmann’s solution) to maintain intravascular volume based on echo variables have been previously determined (4). Sham animals received a similar fluid regimen. Echocardiography was performed prior to each fluid bolus and at the terminal time point as previously described (4, 5). At the terminal time point, a carotid arterial catheter was inserted under isoflurane anesthesia with the animals breathing spontaneously. Blood gas analysis was performed using 0.2 mL arterial blood taken into heparinized capillary tubes (ABL-70 analyzer, Radiometer, Copenhagen, Denmark). A laparotomy incision was made and a 22-gauge needle used to aspirate urine via bladder puncture. The left kidney was isolated and the upper pole placed into formalin and the rest snap-frozen in liquid nitrogen. Cardiac puncture was then performed to obtain blood which was placed in a heparinized tube and centrifuged at 6,500 rpm for 10 minutes. The serum was siphoned off, aliquots taken, snap-frozen in liquid nitrogen, and stored at –80°C. Serum and Tissue Sample Measurements DuoSet enzyme-linked immunosorbent assay (ELISA) kits (R&D Systems, Minneapolis, MN; BD Biosciences, Oxford, Oxon, United Kingdom) were used to assess serum cytokine levels according to the manufacturers’ instructions. Absorbance was read at 450 nm using a spectrophotometric ELISA plate reader (Anthos HTII; Anthos Labtec, Salzburg, Austria).

munosorbent assay (ELISA) kits (R&D Systems, Minneapolis, MN; BD Biosciences, Oxford, Oxon, United Kingdom) were used to assess serum cytokine levels according to the manufacturers’ instructions. Absorbance was read at 450 nm using a spectrophotometric ELISA plate reader (Anthos HTII; Anthos Labtec, Salzburg, Austria). MILLIPLEXMAP multianalyte panels (Merck Millipore, Watford, Herts, United Kingdom) were used for simultaneous detection and quantification of eight biomarkers and cytokines/chemokine in rat urine, including neutrophil gelatinase-associated lipocalin (NGAL), cystatin C, interleukin (IL)-18, monocyte chemotactic protein (MCP)-1, clusterin, calbindin, osteopontin, and kidney injury molecule (KIM)-1. The same technique was used to determine serum levels of IL-18. Assays were performed according to the manufacturer’s protocols. The plate was read on a Bio-Plex 200 multiplex system (Bio-Rad, Hemel Hempstead, Herts, United Kingdom). Urine tissue inhibitor of metalloproteinases-2 (TIMP-2) and insulin-like growth factor-binding protein 7 (IGFBP7) were analyzed by ELISA as described previously (6). Renal function (serum creatinine) was analyzed using the Jaffe assay by the Clinical Pathology laboratory at the Royal Free Hospital, London, United Kingdom. Where urine biomarkers were analyzed, matched creatinine values were used.

h factor-binding protein 7 (IGFBP7) were analyzed by ELISA as described previously (6). Renal function (serum creatinine) was analyzed using the Jaffe assay by the Clinical Pathology laboratory at the Royal Free Hospital, London, United Kingdom. Where urine biomarkers were analyzed, matched creatinine values were used. Immunohistochemistry Kidneys were fixed for 24–72 hours in formalin, transferred to 70% ethanol, and embedded in paraffin. Sections were then cut into 5 µm slices and mounted onto glass slides. For all histological analyses, sections were examined using an Olympus B×4 microscope (Olympus Optical, London, United Kingdom) at ×20 magnification. For assessment of tubular injury (tubular dilatation, brush border loss, and tubular cast formation), sections were stained with Periodic acid-Schiff. Ten random fields of view of the cortex were analyzed for each section at ×20 magnification under a light microscope. Apoptosis was identified by DNA fragments in situ using the terminal deoxyribonucleotidyl transferase (TdT)-mediated biotin-16-2-deoxyuridine 5'-triphosphate (dUTP) nick-end labeling (terminal deoxynucleotidyl transferase dUTP nick-end labeling [TUNEL] assay) using the TACS TdT In Situ Apoptosis Detection Kit (R&D Systems, Abingdon, Oxford, United Kingdom). The total number of apoptotic bodies per ×20 field was counted manually and an average taken for each group. A total of three slides for the septic groups (with at least one from the poor prognosis subgroup) and two from the naïve groups from time points 6, 12, 24, 48, and 72 hours were selected.

ford, United Kingdom). The total number of apoptotic bodies per ×20 field was counted manually and an average taken for each group. A total of three slides for the septic groups (with at least one from the poor prognosis subgroup) and two from the naïve groups from time points 6, 12, 24, 48, and 72 hours were selected. Statistics Analyses were performed using SPSS (IBM SPSS Statistics, Version 22.0; IBM Corp, Armonk, NY), and graphs drawn using Graphpad Prism Version 5.0d (GraphPad Software, La Jolla, CA). Continuous variables are presented as median (interquartile range). Differences in continuous variables between groups were compared using Mann-Whitney U test. Pearson’s correlation was performed to assess the degree of correlation between serum and urine levels of biomarkers (cystatin C, NGAL, IL-18, and MCP-1). A p value of less than 0.05 was taken as statistically significant. RESULTS Baseline measurements (weight, temperature, heart rate [HR], stroke volume [SV], cardiac output [CO]) were similar between groups. In total, 19 animals died (20%, as per the severity of the model). All deaths occurred after the 12-hour time point and before the 48-hour time point. These animals were not used for biomarker/biochemistry analyses. The sampling time (involving kill of the animals) was randomized. All animals intended for sampling at early time points survived, whereas some of the animals intended for sampling at a late time point succumbed prior to reaching this time point.

e point. These animals were not used for biomarker/biochemistry analyses. The sampling time (involving kill of the animals) was randomized. All animals intended for sampling at early time points survived, whereas some of the animals intended for sampling at a late time point succumbed prior to reaching this time point. Physiologic Markers At 3-hour postinduction of sepsis, there was a significant fall in stroke volume and cardiac output (Fig. 1). Septic animals mounted a significant tachycardia by 6 hours, which persisted at 24 hours. This paralleled the increase in core body temperature. By 48 hours, SV, CO, HR, and core temperature normalized among septic animals, and these remained stable until 72 hours. Figure 1. Seventy-two–hr characterization of sepsis and recovery phases—systemic variables. Septic animals develop a tachycardia, fall in stroke volume (SV) and fever early (3–6 hr), which resolves from 24 hr to reach baseline values at 72 hr. Points and whiskers represent median and interquartile range, respectively. *p < 0.05 sham vs sepsis; (*)p = 0.05–0.06 sham vs sepsis. CO = cardiac output, HR = heart rate.

ic animals develop a tachycardia, fall in stroke volume (SV) and fever early (3–6 hr), which resolves from 24 hr to reach baseline values at 72 hr. Points and whiskers represent median and interquartile range, respectively. *p < 0.05 sham vs sepsis; (*)p = 0.05–0.06 sham vs sepsis. CO = cardiac output, HR = heart rate. Biochemical Markers There was an early peak (3 hr) in serum urea and creatinine in septic animals (Fig. 2). Alongside the significant rise in hematocrit at 6 hours, this suggests intravascular volume depletion, despite aggressive fluid loading. Once fluid resuscitation commenced (from 2 hr), there was a progressive fall in serum urea and creatinine that normalized by 6–12 hours. By 24 hours, serum urea and creatinine were significantly elevated. The peak serum creatinine level (30 µmol/L) was 1.5-fold above that of baseline. The increases in arterial lactate levels were phasic, with a rise at 3 hours, normalization at 6 hours, a further peak at 24 hours and then a subsequent fall. The arterial base excess fell in the septic animals but normalized by 48 hours. The rise in serum cystatin C approached statistical significance by 12 hours and remained elevated at 24 hours. Compared with sham animals, serum albumin and glucose fell in the septic animals, reaching a nadir at 24 hours. This change was more pronounced in the septic rats compared with sham-operated rats. Recovery occurred by 72 hours.

ystatin C approached statistical significance by 12 hours and remained elevated at 24 hours. Compared with sham animals, serum albumin and glucose fell in the septic animals, reaching a nadir at 24 hours. This change was more pronounced in the septic rats compared with sham-operated rats. Recovery occurred by 72 hours. Figure 2. Seventy-two–hr characterization of sepsis and recovery phases—biochemistry. Biochemical changes occur from 3 hr. Apart from serum urea, changes are maximal at 24 hr followed by recovery to baseline values at 72 hr. The early rise in urea, creatinine, and lactate is corrected after fluid resuscitation, demonstrating early hemoconcentration. Points and whiskers represent median and interquartile range, respectively. *p < 0.05 sham vs sepsis; (*)p = 0.05–0.06 sham vs sepsis. Serum Cytokines Most proinflammatory cytokines were significantly elevated by 3 hours, including IL-1β, IL-6, MCP-1, and NGAL (Fig. 3). Apart from IL-6, all these cytokines remained elevated for at least 24 hours. The anti-inflammatory cytokine IL-10 was also significantly elevated at 3 hours. The earliest cytokine to peak was IL-1β at 3 hours, which remained elevated until 24 hours. Serum MCP-1 levels followed a similar pattern to IL-1β. IL-6, in contrast, peaked at 6 hours and fell sharply thereafter, to reach baseline levels by 48 hours. There was a trend toward an elevated IL-18 at 48 hours.

at 3 hours. The earliest cytokine to peak was IL-1β at 3 hours, which remained elevated until 24 hours. Serum MCP-1 levels followed a similar pattern to IL-1β. IL-6, in contrast, peaked at 6 hours and fell sharply thereafter, to reach baseline levels by 48 hours. There was a trend toward an elevated IL-18 at 48 hours. Figure 3. Characterization of sepsis and recovery phases—serum cytokines. The proinflammatory cytokine kinetic pattern is variable. Interleukin (IL)-1β and IL-6 rise early (3 hr) and remain elevated until 24 hr. The anti-inflammatory cytokine IL-10 begins to rise by 3 hr but remains elevated through to resolution. Points and whiskers represent median and interquartile range, respectively. *p < 0.05 sham vs sepsis; (*)p = 0.05–0.06 sham vs sepsis. MCP = monocyte chemotactic protein, NGAL = neutrophil gelatinase-associated lipocalin. Urine Biomarkers Ten different urine biomarkers (KIM-1, NGAL, TIMP-2, IGFBP-7, IL-18, MCP-1, calbindin, clusterin, osteopontin, cystatin C) were measured in addition to serum biomarkers of renal function (serum urea, creatinine, and cystatin C) (Fig. 4). All urine biomarkers related to tubular cell injury, apart from TIMP-2, were significantly altered to varied degrees and with different kinetics. All biomarkers returned to levels approaching those observed in sham animals with clinical recovery. Apart from urine osteopontin and IL-18, all other urine biomarkers were elevated and at an earlier time point to serum creatinine (at 24 hr).

-2, were significantly altered to varied degrees and with different kinetics. All biomarkers returned to levels approaching those observed in sham animals with clinical recovery. Apart from urine osteopontin and IL-18, all other urine biomarkers were elevated and at an earlier time point to serum creatinine (at 24 hr). Figure 4. Seventy-two–hr characterization—urine biomarkers. Urine neutrophil gelatinase-associated lipocalin (NGAL) was the most sensitive marker, rising from 3 to 24 hr. Other biomarkers were also elevated before the rise in serum creatinine, though the magnitude of their rise was much less compared with NGAL. Urine interleukin (IL)-18 was not elevated in septic animals. Points and whiskers represent median and interquartile range, respectively. *p < 0.05 sham vs sepsis; (*)p = 0.05–0.06 sham vs sepsis. IGFPB-7 = insulin-like growth factor-binding protein 7, MCP = monocyte chemotactic protein. Urine NGAL was the earliest biomarker to rise (3 hr) with a sustained peak lasting from 24–48 hours. Urine KIM-1 and calbindin peaked at 6 hours, and fell thereafter, with calbindin reaching baseline values by 24 hours and KIM-1 by 72 hours. Although urine clusterin rose early (3 hr), it lacked discriminatory values due to variability in values. Urine cystatin C, a marker of glomerular filtration rate (GFR) and intact tubular reabsorption, was raised between 24 and 48 hours in septic animals.

ndin reaching baseline values by 24 hours and KIM-1 by 72 hours. Although urine clusterin rose early (3 hr), it lacked discriminatory values due to variability in values. Urine cystatin C, a marker of glomerular filtration rate (GFR) and intact tubular reabsorption, was raised between 24 and 48 hours in septic animals. Urine IGFBP-7, a marker of cell cycle arrest, was significantly elevated at 12 hours. This followed the rise seen in tubular injury markers but preceded the rise in functional markers of filtration (i.e., serum creatinine and cystatin C). TIMP-2, another cell cycle arrest marker, was also elevated at 12 hours, but only approached statistical significance. The fall in urinary levels of KIM-1 and calbindin at 24 hours seems to predict renal recovery. Although this time point coincided with a rise in serum creatinine, there was associated improvement in hemodynamics and a fall in proinflammatory cytokines (IL-1β, IL-6, IL-18). In summary, some urine biomarkers including KIM-1 and calbindin rose early and then showed an early fall. On the other hand, NGAL also rose early, but peaked later, and remained elevated till clinical recovery. Cell cycle arrest markers rose after the injury markers and fell prior to clinical recovery. Urine cystatin C rose later, at the same time as serum creatinine (24 hours), suggestive of decreased renal functionality. Urine MCP-1, osteopontin, and IL-18 were elevated at various points in the clinical course but did not demonstrate any clear pattern of rise and fall.

y markers and fell prior to clinical recovery. Urine cystatin C rose later, at the same time as serum creatinine (24 hours), suggestive of decreased renal functionality. Urine MCP-1, osteopontin, and IL-18 were elevated at various points in the clinical course but did not demonstrate any clear pattern of rise and fall. As with current clinical use, it is unclear to what extent the urine biomarkers reflect what is filtered from the circulation into the urine, as opposed to de novo production within the kidney. There was no correlation between paired urine and serum levels of IL-18, MCP-1, or cystatin C. A modest positive correlation was seen between urine and serum NGAL values (r2 = 0.713; p < 0.001), even when including septic animals only (r2 = 0.654; p < 0.001). Renal Histology The degree of injury at 24 hours was relatively mild (Fig. 5). Predominant findings included mild tubular dilatation and brush border loss. Tubular injury was patchy, among areas of normal histology. By comparison, histology is shown of a hemorrhage-reperfusion model performed in the laboratory in similarly aged male Wistar rats (Fig. 5B). TUNEL staining revealed minimal presence of cell death, with an average of two TUNEL positive cells per 20× magnification field. Where present, TUNEL positive cells were within proximal tubular epithelial cells (PTECs) (Fig. 5).

reperfusion model performed in the laboratory in similarly aged male Wistar rats (Fig. 5B). TUNEL staining revealed minimal presence of cell death, with an average of two TUNEL positive cells per 20× magnification field. Where present, TUNEL positive cells were within proximal tubular epithelial cells (PTECs) (Fig. 5). Figure 5. Histological assessment of rat kidneys (magnification ×20). A, Renal tissue without any significant damage. B, Renal tissue obtained from a hemorrhage-reperfusion model (Dyson et al [5]), showing several characteristics of acute tubular injury including dilated tubules with loss of brush border, ischemic glomeruli, and tubular casts. C, Kidney section from a 24-hr sham-operated rat. D, Kidney section from a 24-hr septic rat. E, Terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) stain of naïve renal tissue shows two apoptotic cells per ×20 field. Apoptotic cells stain dark brown (arrow). F, TUNEL stain of 24-hr septic renal tissue shows occasional apoptotic cells. Apoptotic bodies seen mainly in proximal tubular epithelial cells (×20 magnification). Arrow = TUNEL positive cell, cast = tubular cast, CD = collecting duct, G = glomerulus, PT = proximal tubule.

ld. Apoptotic cells stain dark brown (arrow). F, TUNEL stain of 24-hr septic renal tissue shows occasional apoptotic cells. Apoptotic bodies seen mainly in proximal tubular epithelial cells (×20 magnification). Arrow = TUNEL positive cell, cast = tubular cast, CD = collecting duct, G = glomerulus, PT = proximal tubule. DISCUSSION Our study aim was to measure temporal changes in a panel of AKI biomarkers measured from the onset of sepsis through to recovery in a long-term fluid-resuscitated model of fecal peritonitis. This model demonstrates many hemodynamic, biochemical, and immunologic features consistent with clinical sepsis. As with human septic AKI, renal histology demonstrated minimal structural injury or cell death; influx of inflammatory cells was not seen. We report novel data showing an early change in markers of renal injury (urine NGAL, KIM-1, calbindin), followed by a marker of cell cycle arrest (urine IGFBP-7) and, finally, by functional markers of filtration (serum creatinine and cystatin C) (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CCM/C456). Urine NGAL was the most sensitive marker among those studied, rising from 3 to 24 hours. The rise of the functional biomarker, serum cystatin C, at 12 hours implies a fall in GFR, whereas the concurrent fall in serum creatinine is suggestive of decreased creatinine production.

Content 1, http://links.lww.com/CCM/C456). Urine NGAL was the most sensitive marker among those studied, rising from 3 to 24 hours. The rise of the functional biomarker, serum cystatin C, at 12 hours implies a fall in GFR, whereas the concurrent fall in serum creatinine is suggestive of decreased creatinine production. Serum urea and creatinine were initially elevated at 3 and 6 hours, and fell toward baseline values at 12 hours with IV fluid resuscitation. This is consistent with early hemoconcentration followed by a dilution effect. A correction factor can be applied to measured serum creatinine to correct for fluid administration (7) as acute hemodilution may mask the creatinine rise in early AKI (8). However, applying a correction factor for cumulative fluid balance over periods longer than a few hours lacks both a physiologic rationale and objective evidence of accuracy (9). After a large fluid bolus that increased circulating volume by 25%, 24-hour serum creatinine was only 2% below expected (9). The rise in serum cystatin C (consistent with a decrease in GFR) from 6 to 12 hours with concurrent falls in serum creatinine may be a consequence of decreased creatinine production (10). As with serum creatinine, sepsis decreases serum cystatin C production and increases nonrenal clearance (11). However, serum cystatin C had a faster rise and peaked more rapidly than creatinine. As such, serum cystatin C detects AKI early and better reflects inulin GFR in cecal ligation and puncture (CLP)-induced murine sepsis.

erum creatinine, sepsis decreases serum cystatin C production and increases nonrenal clearance (11). However, serum cystatin C had a faster rise and peaked more rapidly than creatinine. As such, serum cystatin C detects AKI early and better reflects inulin GFR in cecal ligation and puncture (CLP)-induced murine sepsis. The temporal changes in cytokine levels in our model were similar to a cohort study of 1,886 subjects hospitalized with community-acquired pneumonia (12). There was an early peak of both pro- and anti-inflammatory cytokines followed by a decline in the proinflammatory cytokine profiles and persistence of the anti-inflammatory cytokine IL-10 though to recovery. Food and Drug Administration–approved AKI biomarkers urine TIMP-2 and IGFBP-7 may be superior to NGAL in diagnosing AKI early in critically ill patients (13, 14). TIMP-2 and IGFBP-7 measured early in the setting of critical illness may also identify patients with AKI at increased risk of mortality or receipt of renal replacement therapy in the subsequent 9 months (15). In a CLP model of sepsis, the combination of TIMP-2 and IGFBP-7 has greater sensitivity in diagnosis of AKI compared with serum creatinine (6). However, the kinetics of these biomarkers in renal recovery has not been described.

d risk of mortality or receipt of renal replacement therapy in the subsequent 9 months (15). In a CLP model of sepsis, the combination of TIMP-2 and IGFBP-7 has greater sensitivity in diagnosis of AKI compared with serum creatinine (6). However, the kinetics of these biomarkers in renal recovery has not been described. Urine IL-18 is less promising than urine NGAL in septic AKI (13), and we report here similar findings. Two markers of distal tubular injury, calbindin and osteopontin, have not been previously evaluated in septic AKI. Urine calbindin was elevated at 6–12 hours, whereas urine osteopontin did not differ between septic and sham animals. Urine MCP-1 has also not been characterized in septic AKI; levels were significantly elevated between 12 and 18 hours. A transient and modest, albeit statistically significant rise, was seen in urine osteopontin levels at 12 hours. Within the normal kidney, osteopontin is mainly present in the loop of Henlé and distal nephron (16), but in our model, proximal tubular injury predominated. In a renal ischemia-reperfusion injury model, the distribution of osteopontin in PTECs and distal TECs (DTECs) differed (17). While DTECs showed an early and persistent increase in osteopontin, the rise seen in PTECs was delayed and mostly associated with morphological regeneration. This suggests osteopontin may promote recovery via modulation of infiltrating cells and local responses by the PTEC. Ours, we believe, is the first study to measure urine osteopontin in septic AKI.

sistent increase in osteopontin, the rise seen in PTECs was delayed and mostly associated with morphological regeneration. This suggests osteopontin may promote recovery via modulation of infiltrating cells and local responses by the PTEC. Ours, we believe, is the first study to measure urine osteopontin in septic AKI. Consistent with other studies in both patients and animal models (18–20), the renal histology in our model shows disproportionately minimal parenchymal injury and apoptosis for the degree of functional impairment measured. There was no evidence of interstitial hypercellularity at any point to suggest significant immune cell infiltration. Our preclinical model used a homogenous population of relatively young rats receiving an identical insult 24 hours after anesthesia and instrumentation. Patients with sepsis have a more variable genetic make-up, are usually older, with coexisting comorbid illnesses, and are receiving other nephrotoxic and renal physiology-modifying interventions. Other than hemodynamics and temperature, serial measurements were not made in the same animal as each time point represents the terminal experiment for collection of blood, renal tissue, and urine samples. We therefore cannot plot the trajectory of variables such as urine output or recovery from anuria over time. As the most unwell animals were anuric, urine biomarkers could not be measured in these animals.

al as each time point represents the terminal experiment for collection of blood, renal tissue, and urine samples. We therefore cannot plot the trajectory of variables such as urine output or recovery from anuria over time. As the most unwell animals were anuric, urine biomarkers could not be measured in these animals. Despite ample fluid resuscitation and becoming febrile, the rats in our peritonitis model do not develop a hyperdynamic circulation. We have previously shown in this model that significant myocardial depression yet maintained cardiac output (21). Many large animal models use an infusion of endotoxin or IV administration of live bacteria (rather than the more clinically representative fecal peritonitis insult), and these may produce a different circulatory/inflammatory profile. We avoided antibiotic use in our study to avoid potential confounding from direct nephrotoxicity and renal impairment secondary to increased inflammation (22). We thus did not wish to confound the biomarker data in our study with a potential additional impact of antibiotics, which may add nephrotoxicity. Despite the nonuse of antibiotics, 80% of rats recovered with fluid resuscitation only. Similar data have been reported by Hollenberg et al (23) in a mouse CLP model. Of note, the outcome effectiveness of antibiotics is age-dependent (24).

with a potential additional impact of antibiotics, which may add nephrotoxicity. Despite the nonuse of antibiotics, 80% of rats recovered with fluid resuscitation only. Similar data have been reported by Hollenberg et al (23) in a mouse CLP model. Of note, the outcome effectiveness of antibiotics is age-dependent (24). Serum creatinine in the septic animals rose by only 50%, less than that seen clinically. However, in this fluid-resuscitated model, animals with a greater than 50% rise in serum creatinine at 24 hours tended to not survive much longer. In pilot studies, the creatinine rise seen at 24 hours could be reached by 6 hours in the absence of fluid resuscitation (data not shown). However, in these non-resuscitated animals, mortality rates were both very high and occurred much earlier, preventing study of the recovery phase. Normalization to urinary creatinine concentration improved the prediction of developing AKI and outcome among critically ill patients, but provided no advantage in diagnosing established AKI (25). Similar findings were described in a rat model of drug-induced AKI (26). In this study, biomarker levels could not be corrected to urine creatinine concentration in many animals due to limited or absent urine output, particularly in the most severely affected.

t provided no advantage in diagnosing established AKI (25). Similar findings were described in a rat model of drug-induced AKI (26). In this study, biomarker levels could not be corrected to urine creatinine concentration in many animals due to limited or absent urine output, particularly in the most severely affected. We did not compare these biomarkers against other inflammatory, nonseptic insults, nor did we try to differentiate systemic from local production of the biomarker. Because of the need to terminally anesthetize the animals to obtain blood, urine, and tissue samples, we did not monitor biomarker levels in the same animal over time and thus could not assess the ability to prognosticate for the development of AKI. No specific intervention currently exists for septic AKI. Intuitively, initiating therapies early would be of greater benefit, but this remains speculative. Time zero, the time of onset of AKI, is rarely known in patients. If the biomarker pattern we observed in our rat model holds true for patients, this may aid selection for interventions and potentially enhance the likelihood of therapeutic benefit. The temporal relationship of biomarker change described may shed some light into pathophysiologic mechanisms.

s rarely known in patients. If the biomarker pattern we observed in our rat model holds true for patients, this may aid selection for interventions and potentially enhance the likelihood of therapeutic benefit. The temporal relationship of biomarker change described may shed some light into pathophysiologic mechanisms. Supplementary Material The in vivo and some of the ex vivo studies were undertaken at University College London, United Kingdom. The ex vivo work including multiplex analysis and histopathology were undertaken at Hammersmith Hospital, Imperial College, United Kingdom. Urine tissue inhibitor of metalloproteinases-2 and insulin-like growth factor-binding protein 7 quantifications were carried out at Department of Critical Care Medicine, University of Pittsburgh. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

Supplementary Material The in vivo and some of the ex vivo studies were undertaken at University College London, United Kingdom. The ex vivo work including multiplex analysis and histopathology were undertaken at Hammersmith Hospital, Imperial College, United Kingdom. Urine tissue inhibitor of metalloproteinases-2 and insulin-like growth factor-binding protein 7 quantifications were carried out at Department of Critical Care Medicine, University of Pittsburgh. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Dr. Arulkumaran received support for article research from Wellcome Trust/Charity Open Access Fund (COAF); he and his institution (University College London) received grant support from the Wellcome Trust Clinical Research Training Fellowship (grant in support of this project in part, including Dr. Arulkumaran’s salary); and additional funding for the biomarker analysis was provided by an U.K. Intensive Care Society New Investigator Award awarded to Dr. Arulkumaran. Dr. Kellum’s institution received funding from Astute Medical (consulting and grant support), and he has licensed unrelated technologies through the University of Pittsburgh to Astute Medical. Dr. Unwin received support for article research from Wellcome Trust/COAF. Dr. Tam’s institution received funding from AstraZeneca (research project grant); he received support for article research from Wellcome Trust/COAF and disclosed receiving a Case Fellowship joint award from Biotechnology and Biological Sciences Research Council and GlaxoSmithKline (GSK) on P2 × 7 receptor in eye disease; he disclosed that he is the chief investigator of the randomized controlled trial of Syk inhibitor in IgA nephropathy, has been on the Advisory Board of MedImmune, has a consultancy agreement with Rigel Pharmaceuticals, and has also received research grants from GSK; he disclosed he is the cosupervisor of a Wellcome Trust Clinical Research Training Fellow for Dr. Arulkumaran, who is the first author of this article. Dr. Singer’s institution received funding from Wellcome Trust (PhD studentship), U.K. Intensive Care Society Young Investigator Award, from various academic grants (e.g., Wellcome Trust, MRC, European Union), and from different companies that are paid to the institution to support research activity (e.g., with Abbott, Deltex, DSTL, Magnus Oxygen, NewB Innovation, Oxford Optronix, Probe Scientific). He has performed consultancy/advisory board work or received speaker fees from Bayer, Biotest, Deltex, Fresenius, Merck, Pfizer, SOBI, and he is Scientific Officer for Magnus Oxygen; he received support for article research from Wellcome Trust/COAF.

L, Magnus Oxygen, NewB Innovation, Oxford Optronix, Probe Scientific). He has performed consultancy/advisory board work or received speaker fees from Bayer, Biotest, Deltex, Fresenius, Merck, Pfizer, SOBI, and he is Scientific Officer for Magnus Oxygen; he received support for article research from Wellcome Trust/COAF. The remaining authors have disclosed that they do not have any potential conflicts of interest. APPENDIX Fecal Slurry The batch of slurry used for the entire set of experiments was identical. Fecal slurry was obtained from pooled stool from three healthy nonvegetarian donors. Immediately after collection, material was 1:1 diluted with a thioglycolate suspension and catalase for optimization of bacterial growth and inactivation of reactive oxygen species. Additionally, 10% glycerine was added and the suspension homogenized under anaerobic conditions. The slurry was then divided into aliquots and frozen at –80°C. Microbiology testing after thawing at different time intervals confirmed ongoing and similar bacterial growth of fecal flora. On the day of the experiment, aliquots were thawed, diluted with n-saline (1.2 mL slurry was diluted with 1.8 mL n-saline [2:3 dilution factor]), and 5 mL/kg body weight and injected intraperitoneally. The “potency” of fecal slurry does depend on the dilution factor. For this set of experiments, 1.2 mL slurry was diluted with 1.8 mL n-saline (2:3 dilution factor) and 5 mL/kg body weight was injected. A dose of 4.5–5.0 mL/kg was the LD50. The dose used in this set of experiments was 4 mL/kg, with a corresponding 20% mortality.

Hemodynamic instability associated with sepsis and septic shock has led to the long-standing dogma that sepsis-induced acute kidney injury (AKI) is primarily a consequence of renal ischemia and ensuing acute tubular necrosis (1). However, multiple analyses in both patients and animal models of AKI fail to demonstrate significant histologic injury (2–4). Furthermore, measurement of renal blood flow (RBF) in septic patients with AKI revealed preserved or even elevated RBF (5–7), whereas experimental models reported no correlation between creatinine clearance and RBF (8, 9). Thus, AKI occurs notwithstanding adequate regional oxygen delivery and tissue oxygenation, at least at the macrovascular level. Even those patients with nonsevere sepsis without significant hemodynamic compromise nor requiring intensive care admission are at increased risk of AKI (10). The paradigm of significant functional impairment despite preserved global and renal hemodynamics and histology remains poorly understood. Whether the trigger for sepsis-induced AKI is intrarenal and/or mediated via the effect of circulating factors is also uncertain.

e care admission are at increased risk of AKI (10). The paradigm of significant functional impairment despite preserved global and renal hemodynamics and histology remains poorly understood. Whether the trigger for sepsis-induced AKI is intrarenal and/or mediated via the effect of circulating factors is also uncertain. We previously proposed that organ dysfunction in sepsis is related to mitochondrial dysfunction (11). To investigate the role of mitochondrial dysfunction and the contribution of circulating factors to infection-associated renal dysfunction, we utilized a well-characterized, clinically relevant, fluid-resuscitated, nonlethal rat model of fecal peritonitis (12, 13). In addition, we performed ex vivo studies using multiphoton imaging of live kidney slices incubated in septic and control serum to assess, independent of any change in RBF or oxygen delivery, the effects of circulating factors associated with sepsis on renal tubular mitochondrial function.

f fecal peritonitis (12, 13). In addition, we performed ex vivo studies using multiphoton imaging of live kidney slices incubated in septic and control serum to assess, independent of any change in RBF or oxygen delivery, the effects of circulating factors associated with sepsis on renal tubular mitochondrial function. MATERIALS AND METHODS Animal Model of Fecal Peritonitis Male Wistar rats (Charles River, Margate, United Kingdom) weighing 300–375 g were used throughout. All experiments were performed under a Home Office Project License (PPL 70/7029) and local UCL Ethics Committee approval. All experiments were performed in accordance with relevant guidelines and regulations. A similar model has been described in detail elsewhere (12–14) and the outline illustrated in Figure 1. However, in contrast to previous experiments, this model used human fecal slurry (13) to produce nonlethal sepsis with characteristics of rat fecal slurry peritonitis (12, 14). The dose of slurry was selected to create a nonlethal model of sepsis, as the objective of this study was to investigate mechanisms of AKI in the absence of significant hemodynamic compromise. Laparotomy was performed under anesthesia at either 6- or 24-hour postinduction of sepsis, with echocardiography performed beforehand.

e of slurry was selected to create a nonlethal model of sepsis, as the objective of this study was to investigate mechanisms of AKI in the absence of significant hemodynamic compromise. Laparotomy was performed under anesthesia at either 6- or 24-hour postinduction of sepsis, with echocardiography performed beforehand. After laparotomy, a 22-gauge needle was used to puncture the renal capsule at the mid-pole. A fiberoptic optode (250 µm diameter) connected to an Oxylite monitoring system (Oxford Optronix, Didcot, Oxon, United Kingdom) allowed continuous tissue oxygen tension (tPo2) monitoring within the renal cortex (15) (Fig. 1). The left renal artery was isolated by careful blunt dissection. An ultrasonic flow probe (Transonic Systems, Ithaca, NY) of 1 mm diameter was placed around the left renal artery to measure RBF. Renal lactate clearance was calculated using the difference between renal vein and arterial lactate. Figure 1. In vivo experimental set up. A, Schema of experimental set up. Animals have a tunneled central venous catheter inserted 24 hours prior to induction of sepsis. At 2 hours, IV fluid resuscitation is commenced. Animals are anesthetized and laparotomy performed at 6 or 24 hours for measurement of cardiac output, renal blood flow, renal cortical oxygen tension and renal vein oxygen tension, and renal tissue sampling. B, Flow probe measuring renal blood flow. Oxygen sensor inserted into lower pole of renal cortex. C, Oxygen sensor inserted into the mid-pole of the renal cortex.

at 6 or 24 hours for measurement of cardiac output, renal blood flow, renal cortical oxygen tension and renal vein oxygen tension, and renal tissue sampling. B, Flow probe measuring renal blood flow. Oxygen sensor inserted into lower pole of renal cortex. C, Oxygen sensor inserted into the mid-pole of the renal cortex. Further details are provided in the supplemental data – Methods (Supplemental Digital Content 1, http://links.lww.com/CCM/D123). Histology—Light Microscopy Kidneys were fixed for 24–72 hours in formalin, transferred to 70% ethanol, and embedded in paraffin. Sections were then cut into 5 µm slices and mounted onto glass slides. Sections were examined by light microscopy (BX4, Olympus Optical, London, United Kingdom) at ×20 magnification. All sections were stained with Periodic acid-Schiff. Apoptosis was identified by DNA fragments in situ using the terminal deoxyribonucleotidyl transferase-mediated biotin-16-dUTP nick-end labeling (TUNEL) assay (TACS TdT In Situ Apoptosis Detection Kit, R&D Systems, Abingdon, Oxford, United Kingdom).

Histology—Light Microscopy Kidneys were fixed for 24–72 hours in formalin, transferred to 70% ethanol, and embedded in paraffin. Sections were then cut into 5 µm slices and mounted onto glass slides. Sections were examined by light microscopy (BX4, Olympus Optical, London, United Kingdom) at ×20 magnification. All sections were stained with Periodic acid-Schiff. Apoptosis was identified by DNA fragments in situ using the terminal deoxyribonucleotidyl transferase-mediated biotin-16-dUTP nick-end labeling (TUNEL) assay (TACS TdT In Situ Apoptosis Detection Kit, R&D Systems, Abingdon, Oxford, United Kingdom). Histology—Electron Microscopy As the instrumentation methodologies described above may potentially affect renal histology, samples for electron microscopy were taken in repeat experiments at 24 hours, where instrumentation was not performed. A kidney was obtained from septic and sham animals under anesthesia. Renal tissue was cut into sections of approximately 1 mm3 and placed immediately into glutaraldehyde fixative. The histopathology laboratory at the Royal Free Hospital, London, performed electron microscopy. Analysis was qualitative and focused on mitochondrial structure within proximal tubular epithelial cells (PTECs). Higher magnifications (×40,000) were used to evaluate changes within the mitochondria.

utaraldehyde fixative. The histopathology laboratory at the Royal Free Hospital, London, performed electron microscopy. Analysis was qualitative and focused on mitochondrial structure within proximal tubular epithelial cells (PTECs). Higher magnifications (×40,000) were used to evaluate changes within the mitochondria. Ex Vivo Assessment of Mitochondrial Function Using Confocal Microscopy Using dyes or natural fluorophores, confocal microscopy allows detailed imaging of cellular physiologic processes in intact renal tissue sections. Multiphoton imaging uses a long wavelength excitation laser that permits greater tissue penetration compared with conventional single-laser confocal fluorescence microscopy. This technique can image live kidney slices in real-time in response to various insults or drugs. We previously reported the use of multiphoton imaging of freshly prepared rat kidney slices to investigate mitochondrial function in cells along the nephron in response to toxic stimuli, including chemical anoxia (16).

py. This technique can image live kidney slices in real-time in response to various insults or drugs. We previously reported the use of multiphoton imaging of freshly prepared rat kidney slices to investigate mitochondrial function in cells along the nephron in response to toxic stimuli, including chemical anoxia (16). Further details are provided in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D123). In brief, the cationic lipophilic indicator, tetramethylrhodamine methyl ester (TMRM; ThermoFisher Scientific, Waltham, MA) was used to determine mitochondrial membrane potential, at a concentration of 50 nM. The greater the potential, the more dye accumulates and the greater the signal intensity at any given pixel. Reactive oxygen species (ROS) generation in tubules was measured using dihydroethidium (Het; ThermoFisher Scientific), at a concentration of 5 µM. As HEt fluoresces on oxidization by superoxide, the fluorescence signal increases in proportion to the rate of ROS production. At 720 nm excitation, the autofluorescence signal emitted between 435 and 485 nm (cyan) arises predominantly from mitochondrial reduced nicotinamide adenine dinucleotide (NADH), enabling monitoring of change in mitochondrial redox status. Cell viability was assessed using Calcein AM (ThermoFisher Scientific). In live cells, Calcein AM is converted to a green-fluorescent calcein by intracellular esterases and detected as a green autofluorescence emission pattern at 800 nm excitation.

de (NADH), enabling monitoring of change in mitochondrial redox status. Cell viability was assessed using Calcein AM (ThermoFisher Scientific). In live cells, Calcein AM is converted to a green-fluorescent calcein by intracellular esterases and detected as a green autofluorescence emission pattern at 800 nm excitation. Changes in cell viability, mitochondrial membrane potential, ROS production and NADH redox state of proximal tubular cells were studied in slices incubated in either 1) physiologic saline solution (PSS); 2) sham serum; 3) septic serum; or 4) septic serum coincubated with the antioxidant, 4-hydroxy-2,2,6,6-tetramethyl-piperidin-1-oxyl (4-hydroxy-TEMPO) at a final concentration of 1 nM (Sigma, Gillingham, Dorset, United Kingdom). Serum was taken at 24 hours as this timepoint corresponded to significant differences in renal function. Serum was pooled from six rats and diluted to a 1:3 ratio in PSS. The septic serum was obtained from animals that underwent a similar experimental protocol of sepsis induction and fluid resuscitation, albeit with a different batch of slurry. The confocal experiments were performed within 2 years of the in vivo work. Confocal images were taken every 10 minutes for a total of 60 minutes. A total of 7–10 sets of images were taken to assess changes in TMRM, Het, and NADH, and three sets to assess changes in calcein for cell viability.

ifferent batch of slurry. The confocal experiments were performed within 2 years of the in vivo work. Confocal images were taken every 10 minutes for a total of 60 minutes. A total of 7–10 sets of images were taken to assess changes in TMRM, Het, and NADH, and three sets to assess changes in calcein for cell viability. Mean fluorescent intensity was expressed as a percentage of mean fluorescent intensity at baseline. Images were taken at 10-minute intervals, focusing on different areas of the slice to avoid damage (bleaching) to the slice from repeated imaging of the same field.

ifferent batch of slurry. The confocal experiments were performed within 2 years of the in vivo work. Confocal images were taken every 10 minutes for a total of 60 minutes. A total of 7–10 sets of images were taken to assess changes in TMRM, Het, and NADH, and three sets to assess changes in calcein for cell viability. Mean fluorescent intensity was expressed as a percentage of mean fluorescent intensity at baseline. Images were taken at 10-minute intervals, focusing on different areas of the slice to avoid damage (bleaching) to the slice from repeated imaging of the same field. Uncoupling Protein Quantification by Western Blot Protein was extracted and estimated from renal tissue or renal slices. Samples (20 µg protein) were electrophoresed at 100 V for 1 hour through a 12% or 15% sodium dodecyl sulfate polyacrylamide gel electrophoresis gel under reducing conditions. Proteins were transferred to a polyvinylidene difluoride membrane (GE Healthcare, Amersham, Buckinghamshire, United Kingdom) at 10 V for 45 minutes and then blocked for 1 hour in 5% milk/1% tris-buffered saline. This membrane was then incubated with goat anti-mouse immunoglobulin-G polyclonal uncoupling protein (UCP-2) antibody (Santa Cruz, Dallas, TX) at 1:250 in 5% bovine serum albumin (BSA; Sigma) in phosphate-buffered saline overnight at 4oC. Following incubation with the primary antibody, horseradish peroxidase rabbit anti-goat IgG (Sigma) was added at 1:3000 in 5% milk for 1 hour. Activity was detected using ECL Plus substrate (GE Healthcare). Four animals were randomly selected from each of the sham and sepsis groups at 6 and 24 hours for analysis.

4oC. Following incubation with the primary antibody, horseradish peroxidase rabbit anti-goat IgG (Sigma) was added at 1:3000 in 5% milk for 1 hour. Activity was detected using ECL Plus substrate (GE Healthcare). Four animals were randomly selected from each of the sham and sepsis groups at 6 and 24 hours for analysis. To demonstrate evidence of renal oxidative stress in vivo, we measured levels of isoprostanes within the urine of septic animals and sham-operated animals at 24 hours. The commercially available OxiSelectTM 8- iso- Prostaglandin F2α Enzyme-Linked Immunosorbent Assay Kit (Cell Bio Labs, San Diego, CA) was used. Assays were run as per manufacturer’s protocol. Statistical Analyses Analyses were performed and graphs drawn using Graphpad Prism Version 5.0d (GraphPad Software, La Jolla, CA). In view of relatively small sample sizes, nonparametric tests were used. Continuous variables are presented as median (interquartile range). Differences in continuous variables between groups were compared using Mann Whitney U test or Kruskal-Wallis test with post hoc Dunn test for more than two groups. For in vivo data, comparisons were made between sham-operated and septic animals at 6 and 24 hours. For ex vivo data, comparisons were made between slices incubated in PSS, sham serum, septic serum, or septic serum with 4-OH-TEMPO using Kruskal-Wallis tests with post hoc Dunn test. A p value less than 0.05 was considered statistically significant.

were made between sham-operated and septic animals at 6 and 24 hours. For ex vivo data, comparisons were made between slices incubated in PSS, sham serum, septic serum, or septic serum with 4-OH-TEMPO using Kruskal-Wallis tests with post hoc Dunn test. A p value less than 0.05 was considered statistically significant. RESULTS Physiology and Biochemistry of the In Vivo Peritonitis Model For details, see Supplementary Fig. 1 (Supplemental Digital Content 2, http://links.lww.com/CCM/D124--legend, Supplemental Digital Content 3, http://links.lww.com/CCM/D125) and Table 1. TABLE 1. Physiologic and Biochemical Parameters After 6- and 24-Hour Sepsis At baseline, all measured variables were similar. All six sham-operated animals survived to study end. All septic animals (n = 6–10 per group) survived the 6- and 24-hour experiments, but two animals were peri-mortem at 24 hours and thus excluded from analysis. Compared with sham-operated animals whose body temperature was normothermic (37.5 [37.1–37.8]oC) throughout, the septic animals were febrile at both 6 hours (38.4 [37.9–38.6]oC) and 24 hours (38.3 [37.9–39.2]oC) (p = 0.006). Arterial lactate concentration was elevated only at 24 hours (2.1 [1.6–2.8] mmol/L) in septic animals compared with sham-operated animals (1.3 [1.0–1.6] mmol/L) (p = 0.026). There was no significant difference between sham-operated animals and septic animals (at either time point) in cardiac output or global oxygen delivery (Do2).

tate concentration was elevated only at 24 hours (2.1 [1.6–2.8] mmol/L) in septic animals compared with sham-operated animals (1.3 [1.0–1.6] mmol/L) (p = 0.026). There was no significant difference between sham-operated animals and septic animals (at either time point) in cardiac output or global oxygen delivery (Do2). There was no statistically significant difference between sham-operated and septic animals at either timepoint in renal DO2, oxygen consumption (VO2), or renal cortical oxygen tension (PO2). At 24 hours serum creatinine (41 [37–45] µmol/L) was elevated compared with sham-operated animals (33 [28–38] µmol/L) (p = 0.020). Renal lactate clearance reduced in septic animals at 24 hours (–13 [8 to –15]%) compared with sham-operated animals (–27 [–16 to –41]%) (p = 0.043). Microscopy For details, see Supplementary Fig. 2, Supplemental Digital Content 5, http://links.lww.com/CCM/D127--legend, Supplemental Digital Content 3, http://links.lww.com/CCM/D125). Tubular injury on light microscopy was subtle and focal. There was some loss of the brush border microvilli and mild tubular dilation at both 6 and 24 hours. TUNEL stain revealed minimal cell death. Where present, cell death was localized to PTECs. Representative images from electron microscopy showed normal tubular epithelial mitochondrial structure in both sham-operated and septic animals at 24 hours. Multiphoton Confocal Imaging For details, see Supplementary Fig. 3 (Supplemental Digital Content 6, http://links.lww.com/CCM/D128--legend, Supplemental Digital Content 3, http://links.lww.com/CCM/D125) and Table 2.

Tubular injury on light microscopy was subtle and focal. There was some loss of the brush border microvilli and mild tubular dilation at both 6 and 24 hours. TUNEL stain revealed minimal cell death. Where present, cell death was localized to PTECs. Representative images from electron microscopy showed normal tubular epithelial mitochondrial structure in both sham-operated and septic animals at 24 hours. Multiphoton Confocal Imaging For details, see Supplementary Fig. 3 (Supplemental Digital Content 6, http://links.lww.com/CCM/D128--legend, Supplemental Digital Content 3, http://links.lww.com/CCM/D125) and Table 2. TABLE 2. Percentage Change in Mean Fluorescence Intensity at 60 Minutes Compared With Baseline Following Incubation in Physiologic Saline Solution, Sham Serum, Septic Serum, or Septic Serum Coincubated With 4-Hydroxy-2,2,6,6-Tetramethyl-Piperidin-1-Oxyl

Multiphoton Confocal Imaging For details, see Supplementary Fig. 3 (Supplemental Digital Content 6, http://links.lww.com/CCM/D128--legend, Supplemental Digital Content 3, http://links.lww.com/CCM/D125) and Table 2. TABLE 2. Percentage Change in Mean Fluorescence Intensity at 60 Minutes Compared With Baseline Following Incubation in Physiologic Saline Solution, Sham Serum, Septic Serum, or Septic Serum Coincubated With 4-Hydroxy-2,2,6,6-Tetramethyl-Piperidin-1-Oxyl Using multiphoton confocal imaging of live kidney slices, cell viability, mitochondrial NADH redox state, membrane potential, and tubular ROS levels remained stable over the 60-minute time incubation period in both PSS and sham serum. However, incubation of live kidney slices in septic serum resulted in a significant progressive fall in NADH redox state in the mitochondria of PTECs compared with sham serum (74 [58–87]% vs. 102 [94–119]%, respectively; p = 0.018). A progressive fall in mitochondrial membrane potential was seen following incubation of live kidney slices in septic serum compared with sham serum (77 [70–88]% vs. 100 [91–108]%, respectively; p = 0.0007). Incubation of live kidney slices in septic serum and 4-OH-TEMPO was not associated with any changes in either PTEC NADH redox state (96 [88–99]%) or mitochondrial membrane potential (90 [85–95]%) compared with incubation of live kidney slices in sham serum (p > 0.05).

–88]% vs. 100 [91–108]%, respectively; p = 0.0007). Incubation of live kidney slices in septic serum and 4-OH-TEMPO was not associated with any changes in either PTEC NADH redox state (96 [88–99]%) or mitochondrial membrane potential (90 [85–95]%) compared with incubation of live kidney slices in sham serum (p > 0.05). ROS levels were significantly increased on incubation of live kidney slices with septic serum (163 [152–256]%) compared with sham serum (109 [87–138]%) in PTECs; p = 0.0001). Incubation of live kidney slices in septic serum and 4-OH-TEMPO was not associated with any change in PTEC ROS levels (84 [70–96%]; p > 0.05). No change in calcein uptake was seen in any groups over the 60-minute period, suggesting no change in cell viability (p = 0.733). Renal UCP-2 Quantification Western blot revealed a 1.56-fold increase in UCP-2 concentration in kidneys taken from septic rats (p = 0.05 compared with sham-operated) (Fig. 2A and B). UCP-2 levels rose modestly (1.11-fold increase) in live kidney slices exposed to septic serum over 60 minutes compared with sham serum, though this did not reach statistical significance (p = 0.10). We measured urine isoprostane levels in sham-operated and septic animals at 24 hours (Fig. 2C). Two of the samples run were spuriously high, which we excluded and analyzed the results. The levels of urine isoprostane levels are higher in 24-hour septic animals compared with sham-operated animals, though does not reach statistical significance (11,125 [6,254–31,769] pg/mL vs. 8,409 [4,285–13,014] pg/mL; p = 0.06).

Two of the samples run were spuriously high, which we excluded and analyzed the results. The levels of urine isoprostane levels are higher in 24-hour septic animals compared with sham-operated animals, though does not reach statistical significance (11,125 [6,254–31,769] pg/mL vs. 8,409 [4,285–13,014] pg/mL; p = 0.06). DISCUSSION Our in vivo model of resuscitated fecal peritonitis recapitulates many of the features of human sepsis, with an early rise in core temperature followed by a rise in arterial lactate and organ dysfunction. Consistent with published data, our model confirms that sepsis-induced AKI can still occur notwithstanding the absence of structural tubular injury, decreases in RBF and oxygen delivery, and hemodynamic instability (2–4). After 24 hours of sepsis, renal VO2, DO2, and cortical oxygenation were all stable. Renal mitochondrial ultrastructure was also normal, again suggestive of a functional rather than structural pathophysiologic mechanism.

structural tubular injury, decreases in RBF and oxygen delivery, and hemodynamic instability (2–4). After 24 hours of sepsis, renal VO2, DO2, and cortical oxygenation were all stable. Renal mitochondrial ultrastructure was also normal, again suggestive of a functional rather than structural pathophysiologic mechanism. Renal oxygenation depends on the balance between local DO2 and VO2. In health, renal VO2 depends on renal DO2, with oxygen extraction remaining stable over a wide range of local blood flow (17). Tubular reabsorption of filtered sodium is the major determinant of renal VO2 (18), while tubular transport processes are dependent on the filtered load (19, 20). Any decrease in renal function (i.e., glomerular filtration rate [GFR], as reflected in the rise in serum creatinine at 24 hr) would decrease filtered sodium and thus reduce tubular VO2. Despite this, renal VO2 remained stable at 24 hours. In conjunction with the reduction in renal function at 24 hours (rise in creatinine, reduction in lactate clearance), this suggests that renal oxygen consumption may be partially redirected away from adenosine triphosphate (ATP) production (i.e., oxidative phosphorylation). The rise in UCP-2 protein also supports the possibility of increased uncoupling of mitochondrial respiration.

(rise in creatinine, reduction in lactate clearance), this suggests that renal oxygen consumption may be partially redirected away from adenosine triphosphate (ATP) production (i.e., oxidative phosphorylation). The rise in UCP-2 protein also supports the possibility of increased uncoupling of mitochondrial respiration. The investigations on renal proximal tubular cell mitochondrial function within the live kidney slices incubated with septic serum supports these in vivo findings. Despite cell viability being maintained, the falls in tubular NADH and mitochondrial membrane potential are also consistent with increased uncoupling of mitochondrial oxygen utilization away from ATP production. Coincubation with the ROS scavenger TEMPO attenuated these changes and prevented the rise in ROS levels in the PTECs, implicating excessive oxidant stress as an important pathologic mechanism. During sepsis, high levels of ROS and reactive nitrogen species are produced, and these may overwhelm antioxidant capacity with resultant inhibition of, and damage to, the electron transport chain (21, 22). Mitochondria exposed to nonmitochondrial ROS become a source of ROS themselves (23). This positive feedback loop of ROS-induced-ROS is likely to culminate in uncoupled respiration, dysregulation, and damage to mitochondria. The reduction in membrane potential may however be a useful negative feedback mechanism to limit the amount of harmful ROS produced (24). However, excessive ATP depletion caused by the uncoupling of oxidative phosphorylation is associated with collapse of the electrochemical gradient across the mitochondrial membrane and release of ROS (25). While nonsignificant increase in renal UCP-2 protein was seen in the short time period (60 min) during which the live kidney slices were incubated in septic serum, there may be an increase in activity.

is associated with collapse of the electrochemical gradient across the mitochondrial membrane and release of ROS (25). While nonsignificant increase in renal UCP-2 protein was seen in the short time period (60 min) during which the live kidney slices were incubated in septic serum, there may be an increase in activity. To demonstrate evidence of increased renal oxidative stress in vivo, measurement of urine F2-isoprostanes in 24-hour sham-operated and septic animals was performed. Measurement of F2-isoprostanes (a product of free radical-catalyzed peroxidation of arachidonic acid) has emerged as one of the most reliable approaches to assess oxidative stress status in vivo (26). The levels of urine isoprostane levels are higher in 24-hour septic animals compared with sham-operated animals (Fig 2C), though does not reach statistical significance (p = 0.06).

alyzed peroxidation of arachidonic acid) has emerged as one of the most reliable approaches to assess oxidative stress status in vivo (26). The levels of urine isoprostane levels are higher in 24-hour septic animals compared with sham-operated animals (Fig 2C), though does not reach statistical significance (p = 0.06). Figure 2. Renal uncoupling protein by Western blot and urine isoprostane levels. Circles and squares represent individual data points in the scatter plot. A, Uncoupling protein-2 (UCP-2) in kidney homogenate from sham-operated and 24-hour septic animal. The Western blot demonstrates a clear increase in renal UCP-2 expression in septic animals (1.56-fold increase) compared with sham-operated animals (p = 0.05). B, UCP-2 in kidney slice exposed to sham and septic serum. The Western blot demonstrates a modest increase in renal UCP-2 expression in live kidney slices exposed to septic serum (1.11-fold increase) compared with sham serum (p = 0.10). C, Urine isoprostane levels from sham-operated and septic animals at 24 hours. There is an increase in urine isoprostane levels in septic animals compared with sham-operated animals at 24 hours, which approaches statistical significance (p = 0.06).

septic serum (1.11-fold increase) compared with sham serum (p = 0.10). C, Urine isoprostane levels from sham-operated and septic animals at 24 hours. There is an increase in urine isoprostane levels in septic animals compared with sham-operated animals at 24 hours, which approaches statistical significance (p = 0.06). The association between increased ROS, mitochondrial dysfunction, and renal dysfunction in sepsis in vivo has been described by others (27, 28). We have unified the in vivo and in vitro experiments to arrive at the same conclusion—therefore adding strength to the hypothesis. We have demonstrated that the changes seen both in vivo and in vitro are flow-independent. Furthermore, the changes induced in the in vitro live kidney slice experiments are from septic serum obtained from the in vivo experiments. The change in mitochondrial function seen on incubation of live kidney slices in serum from septic rats suggests a circulating mediator(s) is responsible for changes seen in sepsis. Others have also demonstrated that renal proximal tubular cells exposed to septic serum are prone to mitochondrial dysfunction (decrease in mitochondrial membrane potential, ATP depletion, and increased superoxide and peroxynitrite levels), and cytotoxicity (29). Other cell types including endothelial cells (30), cardiac myocytes (31), and fibroblasts (32) are also susceptible to mitochondrial dysfunction on exposure to septic serum. The precise mediators within septic serum that are responsible for these changes are as yet unidentified.

ynitrite levels), and cytotoxicity (29). Other cell types including endothelial cells (30), cardiac myocytes (31), and fibroblasts (32) are also susceptible to mitochondrial dysfunction on exposure to septic serum. The precise mediators within septic serum that are responsible for these changes are as yet unidentified. Previous work from our laboratory in this rat model has demonstrated a temporal change in markers of renal injury (urine NGAL, KIM-1, calbindin), followed by a rise in a marker of cell cycle arrest (urine IGFBP7) and, finally, by markers of decreased filtration (serum creatinine and cystatin C) (13). We hypothesized a model where there is direct renal tubular injury from circulating inflammatory mediators, followed by cell cycle arrest as a means of reducing tubular cell replication and, then by a reduction in function. Significant reductions in ATP result in cell death via several mechanisms, including opening of the mitochondrial transition pore (33). As such, reduction in cell cycling, a highly energy-dependent process, would facilitate cell survival in conditions where oxidative phosphorylation is reduced. Furthermore, the “functional shutdown” of GFR and tubular function may initiate various protective mechanisms including conservation of circulating volume, though at the expense of functionality (34). Mitochondrial dysfunction/uncoupling may represent an early manifestation of renal injury in vivo, which is demonstrated by the rapid changes in mitochondrial functionality seen in our healthy kidney slices incubated with septic serum.

ing conservation of circulating volume, though at the expense of functionality (34). Mitochondrial dysfunction/uncoupling may represent an early manifestation of renal injury in vivo, which is demonstrated by the rapid changes in mitochondrial functionality seen in our healthy kidney slices incubated with septic serum. Although the confocal work focuses on the proximal tubules, we acknowledge that the etiology of kidney dysfunction is likely to be multifactorial, and other renal cell types may be also involved. However, recent work strongly points to a key role of the PTEC as a site of injury. Markers of proximal tubule injury were elevated (e.g., IGFBP-7), whereas changes to distal tubular injury biomarkers were not as pronounced (13, 35). We avoided antibiotic use to avoid potential direct renal toxicity (36). Serum creatinine in the septic animals rose by 50% and did not formally measure creatinine clearance. Although the rise in creatinine is less than that seen clinically, the requirement for significant fluid resuscitation may have a dilution effect (37). Furthermore, creatinine production is reduced during early sepsis in rodents (38). Of note, in this model, animals with a greater than 50% rise in serum creatinine at 24 hours tend not to survive.

ine is less than that seen clinically, the requirement for significant fluid resuscitation may have a dilution effect (37). Furthermore, creatinine production is reduced during early sepsis in rodents (38). Of note, in this model, animals with a greater than 50% rise in serum creatinine at 24 hours tend not to survive. The time course of ROS generation in slices exposed to septic serum is significantly shorter compared with upregulation of UCP-2 protein, which was not upregulated until 24 hours in vivo. The in vitro experiments, however, are a proof of concept study to demonstrate that renal PTECs are capable of producing ROS and undergo changes consistent with uncoupling that are independent of oxygen delivery. These experiments cannot be directly compared with the in vivo experiments due to different conditions inherent to the nature of the experiments. Our in vitro mitochondrial studies were performed at levels of oxygen (room air) that are “nonphysiological.” Although clear differences were seen between kidney slices incubated with either septic or sham serum, it is yet to be determined whether the same holds true at more physiologic levels of oxygenation. Furthermore, the renal PTECs are in direct contact with serum in the in vitro experiment.

that are “nonphysiological.” Although clear differences were seen between kidney slices incubated with either septic or sham serum, it is yet to be determined whether the same holds true at more physiologic levels of oxygenation. Furthermore, the renal PTECs are in direct contact with serum in the in vitro experiment. We are not able to comment specifically on mitochondrial mass or mitophagy, as these were not measured. However, it is unlikely that this would occur so rapidly as to cause a significant decrease in mitochondrial mass in 1 hour. Although we did not assess in vivo ATP turnover, we propose that mitochondrial uncoupling would divert oxygen consumption away from ATP production. The age of rats we use are equivalent to young adults in humans. We acknowledge that the age of rats may influence immune and bioenergetic function of mitochondria. However, younger rats are used more commonly in experimental medicine due to cost implications, which is a limitation inherent to many published animal studies.

ge of rats we use are equivalent to young adults in humans. We acknowledge that the age of rats may influence immune and bioenergetic function of mitochondria. However, younger rats are used more commonly in experimental medicine due to cost implications, which is a limitation inherent to many published animal studies. Despite these limitations, our study has a number of strengths and novel findings. We demonstrate many findings consistent with the reported literature, including lack of change in renal hemodynamics and histology, notwithstanding a rise in serum creatinine and decreased renal lactate utilization. Septic rats were pyrexic, and renal UCP-2 protein was increased. As uncoupled respiration leads to increased heat production, this may be an important mechanism underlying the generation of fever in sepsis. Of note, circulating mediators within septic serum could induce changes consistent with mitochondrial uncoupling within PTECs that are mediated via excess ROS production. Using complimentary in vivo and in vitro studies, we describe a mechanism that may potentially explain the paradigm of sepsis-induced AKI and offer new therapeutic approaches. Supplementary Material *See also p. 658. The in vivo and some of the ex vivo studies were undertaken at University College London, United Kingdom. Light microscopy was undertaken at Hammersmith Hospital, Imperial College, United Kingdom, and electron microscopy was undertaken at the Royal Free Hospital, London, United Kingdom.

Supplementary Material *See also p. 658. The in vivo and some of the ex vivo studies were undertaken at University College London, United Kingdom. Light microscopy was undertaken at Hammersmith Hospital, Imperial College, United Kingdom, and electron microscopy was undertaken at the Royal Free Hospital, London, United Kingdom. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Dr. Arulkumaran received grant support from the Wellcome Trust Clinical Research Training Fellowship (grant in support of this project in part, including Dr. Arulkumaran’s salary). Mr. Pollen is supported by a New investigator award from the U.K. Intensive Care Society, the University College London MBPhD Program, and the Astor Foundation. Dr. Tam is supported by the Diamond Fund from Imperial College Healthcare Charity and Ken and Mary Minton Chair of Renal Medicine.

g Dr. Arulkumaran’s salary). Mr. Pollen is supported by a New investigator award from the U.K. Intensive Care Society, the University College London MBPhD Program, and the Astor Foundation. Dr. Tam is supported by the Diamond Fund from Imperial College Healthcare Charity and Ken and Mary Minton Chair of Renal Medicine. Drs. Arulkumaran and Tam received support for article research from Wellcome Trust/COAF. Dr. Tam’s institution received funding from Wellcome Trust Clinical Research Training Fellowship for Dr Nishkantha Arulkumaran; research project grants from AstraZeneca Limited, Baxter Biosciences, GSK, Roche Palo Alto, Rigel Pharmaceuticals, and MedImmune and has consultancy agreements with MedImmune, Novartis and Rigel Pharmaceuticals; and from Dr. Tam acting as the Chief Investigator of an international clinical trial in IgA nephropathy. Dr. Tam disclosed he is supported by the Diamond Fund from Imperial College Healthcare Charity and Ken and Mary Minton Chair of Renal Medicine, and he has consultancy agreements with Medimmune (ongoing, income to University), Novartis (ongoing, income to university), and Rigel Pharmaceuticals (past: regarding glomerulonephritis: part of income to him, part of income to university). Dr. Unwin’s institution received funding from AstraZeneca, and he received support for article research from Research Councils UK. Dr. Singer disclosed that he is a director of Magnus Oxygen, which is developing a sulfide donor for use in ischemia-reperfusion injury through inhibiting mitochondrial ROS production, and he also sits on an Advisory Board for AM Pharma, which is trialing recombinant alkaline phosphatase therapy for sepsis-induced acute kidney injury. The remaining authors have disclosed that they do not have any potential conflicts of interest.

Damage to the CNS in the form of traumatic brain injury (TBI) can lead to multiple organ dysfunction syndrome (1). The lung is the most frequent organ to be affected, manifested by the development of acute respiratory distress syndrome (ARDS) (1, 2). ARDS occurs in 20–25% of patients with isolated brain injury (3) and is a critical independent factor affecting mortality. The mechanisms for ARDS development in this population remain to be defined. Lung-protective strategies can improve the outcome of TBI patients (4); however, life-threatening pulmonary dysfunction is extremely difficult to treat (5). Furthermore, this has major implications for lung transplantation, with less than one third of nonsurviving severe TBI patients considered suitable for lung donation, of which more than 80% fail to donate (6). A “double hit” model has been hypothesized to explain the increased pulmonary susceptibility to injury after severe TBI (7) involving a sympathetic catecholamine storm that induces type II pneumocyte damage of the pulmonary endothelium (8). This, coupled with a systemic inflammatory response induced by increased intracranial proinflammatory mediator production and its release into the circulation (9), alongside activation of various receptors within the lung (Toll-like receptor 4 [TLR-4], receptor for advanced glycation end products [RAGEs]) by damage-associated molecular pattern molecules (DAMPs) (10, 11), may drive the development of pulmonary injury through increased mediator production and infiltration of activated neutrophils (7, 12, 13).

ous receptors within the lung (Toll-like receptor 4 [TLR-4], receptor for advanced glycation end products [RAGEs]) by damage-associated molecular pattern molecules (DAMPs) (10, 11), may drive the development of pulmonary injury through increased mediator production and infiltration of activated neutrophils (7, 12, 13). In this study, we examined the effects of mild TBI on pulmonary neutrophil priming. To our knowledge, this has never been investigated because most researchers have studied the pulmonary response to moderate/severe TBI models. We show for the first time that mild TBI induces massive pulmonary sequestration of interstitial neutrophils that then transmigrate into the alveolar compartment following a subsequent small inflammatory insult such as acid aspiration to induce injury. This has potential implications for considering earlier prophylactic measures for targeting neutrophil migration from the interstitium to the alveolar space in patients with TBI. METHODS Animals Eight-week-old male CD1 mice were purchased from Harlan (Harlan, United Kingdom) and given 1 week to acclimatize before experimentation. Mice were maintained in 12-hour light/12-hour dark cycles with free access to food and water. All experimental animal procedures were approved by the University of Edinburgh and were performed in accordance with Home Office guidelines (Animal [Scientific Procedures] Act 1986).

eek to acclimatize before experimentation. Mice were maintained in 12-hour light/12-hour dark cycles with free access to food and water. All experimental animal procedures were approved by the University of Edinburgh and were performed in accordance with Home Office guidelines (Animal [Scientific Procedures] Act 1986). Fluid Percussion Injury Adult (25–35 g) male CD1 mice were anesthetized with isoflurane (Merial, Woking, United Kingdom) and prepared for a 1.5 atmosphere fluid percussion injury (FPI) according to published methodology (14). For full details and flow charts outlining experimental protocols, see online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740) and Supplementary Figure E1 (Supplemental Digital Content 2, http://links.lww.com/CCM/D741; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Animals (n = 3–6 per group) were retrieved at 0, 6, 24, and 48 hours post-FPI.

ta supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740) and Supplementary Figure E1 (Supplemental Digital Content 2, http://links.lww.com/CCM/D741; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Animals (n = 3–6 per group) were retrieved at 0, 6, 24, and 48 hours post-FPI. Brain Histology and Immunohistochemistry Preparation For histologic and immunohistochemical assessment of neuronal damage and cellular infiltration, brains were frozen in −38°C isopentane (277258-1L; Sigma-Aldrich Company, St. Louis, MO) before being placed in storage at −80°C. Tenmicrometer coronal cryostat sections were cut using the Leica CM1900 (Leica, Wetzlar, Germany) before being mounted on Leica Surgipath X-tra Adhesive (3800050, Leica) precleaned micro slides and stored at −80°C. Light microscopy images were obtained using a Zeiss Axioskop (Carl Zeiss, Welwyn Garden City, United Kingdom) light microscope connected to a Qimaging Micropublisher 3.3 camera (Qimaging, Surrey, BC, Canada).

e being mounted on Leica Surgipath X-tra Adhesive (3800050, Leica) precleaned micro slides and stored at −80°C. Light microscopy images were obtained using a Zeiss Axioskop (Carl Zeiss, Welwyn Garden City, United Kingdom) light microscope connected to a Qimaging Micropublisher 3.3 camera (Qimaging, Surrey, BC, Canada). Acid Fuchsin To assess neuronal damage, brain sections were stained with acid fuchsin. Frozen 10-µm coronal cryostat sections were dried at 40°C overnight before being fixed in cold 4% paraformaldehyde in PBS for 1 hour. Slides were washed in phosphate-buffered saline (3 × 5 min) before being stained for 30 seconds in 1% Acid Fuchsin (A3908; Sigma-Aldrich Company, St. Louis, MO) with three drops per 100 mL of glacial acetic acid. Slides were washed with water, dehydrated through 70%, 90%, and 100% ethanol for 2 minutes each, and cleared in xylene for 2 minutes before being mounted with Pertex (3808707E; Leica, Wetzlar, Germany). To quantify neuronal damage, three sections located throughout the hippocampus of each animal were imaged, with acid fuchsin positive cells counted using Image J (National Institutes of Health, Bethesda, MD). Myeloperoxidase Immunohistochemistry To assess neutrophil infiltration, myeloperoxidase immunohistochemistry was performed. For full details, see online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Flow Cytometric Analysis of Brain and Lung Tissues Tissue digests and flow cytometry methods are detailed in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740).

Myeloperoxidase Immunohistochemistry To assess neutrophil infiltration, myeloperoxidase immunohistochemistry was performed. For full details, see online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Flow Cytometric Analysis of Brain and Lung Tissues Tissue digests and flow cytometry methods are detailed in the online data supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Bronchoalveolar Lavage Bronchoalveolar lavage fluid (BALF) was collected according to a published protocol (15). For full details, see online supplement (Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Enzyme-Linked Immunosorbent Assay Enzyme-linked immunosorbent assay (ELISA) kit for the measurement of interleukin (IL)-1β in BALFs (DuoSet; R&D Systems, Minneapolis, MN) was used according to the manufacturers’ instructions. Total Protein Total protein within BALF was performed using a Pierce BCA Total Protein Assay Kit (23227; Thermo Scientific, Waltham, MA) as per the manufacturers’ instructions.

Enzyme-Linked Immunosorbent Assay Enzyme-linked immunosorbent assay (ELISA) kit for the measurement of interleukin (IL)-1β in BALFs (DuoSet; R&D Systems, Minneapolis, MN) was used according to the manufacturers’ instructions. Total Protein Total protein within BALF was performed using a Pierce BCA Total Protein Assay Kit (23227; Thermo Scientific, Waltham, MA) as per the manufacturers’ instructions. Hydrochloric Acid Microaspiration Model To model acid microaspiration following TBI, a subclinical hydrochloric acid (HCl) aspiration model was developed (16, 17). HCl (318965; Sigma-Aldrich Company, St. Louis, MO) was diluted to pH 1.75 in saline (0.9% sodium chloride, UKF7124; Baxter, Deerfield, IL), and 50 µL was administered via an “intratracheal” route. 11.25 µl of HCl was instilled in a volume of 50 µL. An equivalent scaling factor in humans for lung weight and size is approximately 700-fold; therefore, the equivalent volume in a human would be approximately 7 mL. After receiving HCl, a preemptive intraperitoneal injection of saline (200 µL) was given. Mice were then placed in a warm humidified oxygen chamber (2 L/min flow rate) for 30 minutes before reversal of anesthesia. Mice were kept on a heat mat before retrieval 6 hours later. To model acid microaspiration following TBI, the HCl model was applied immediately following FPI after recording the righting time.

en. Mice were then placed in a warm humidified oxygen chamber (2 L/min flow rate) for 30 minutes before reversal of anesthesia. Mice were kept on a heat mat before retrieval 6 hours later. To model acid microaspiration following TBI, the HCl model was applied immediately following FPI after recording the righting time. Neutrophil Depletion—Anti-Lymphocyte Antigen 6 Complex Locus G6D Antibody Neutrophil depletion was achieved with an anti-lymphocyte antigen 6 complex locus G6D (LY-6G) monoclonal antibody (clone IA8; BioXCell, West Lebanon, NH) as described (15). For details of flow cytometric analysis of blood, see online data supplement, (Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Cytokine Bead Array Cytokine levels within BALF were measured using the BD Cytometric Bead Array Mouse Inflammation Kit (552364; BD Biosciences, San Jose, CA) as per the manufacturers’ instructions. Statistics Data are represented as mean ± sd or sem. Quantification of histology/immunohistochemistry was performed blinded by the investigator. Statistical comparisons were made using two-tailed Student t test or one-way/two-way analysis of variance with Bonferroni posttest for multiple comparisons. A p value of less than 0.05 was considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001). All graphs and statistics were performed using the statistical package Graphpad Prism 5 for Windows (Graphpad Software, La Jolla, CA).

o-way analysis of variance with Bonferroni posttest for multiple comparisons. A p value of less than 0.05 was considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001). All graphs and statistics were performed using the statistical package Graphpad Prism 5 for Windows (Graphpad Software, La Jolla, CA). RESULTS A 1.5 ATM FPI in Mice Induces a Unilateral Cortical Insult With Cerebral Inflammatory Cell Influx and No Measureable Clinical Sequelae Mice received a lateral FPI using an established methodology (14). Sham and FPI were significantly different in righting times (time to turn prone from supine), indicating that a significant cortical impact had occurred (Supplementary Fig. E2, Supplemental Digital Content 3, http://links.lww.com/CCM/D742; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). However, once the mice had recovered after the impact (within 15 min), no behavioral or physical differences could be distinguished between the two groups. To assess neuronal injury, brain sections were stained with acid fuchsin. Homogenous staining and cells with regular morphology were seen following sham treatment, indicating no neuronal damage (for all histology and flow cytometry plots, see Supplementary Fig. E3, Supplemental Digital Content 4, http://links.lww.com/CCM/D743; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). However, significant neuronal damage, evident by the appearance of triangularly shaped, intensely stained neurons with condensed cellular morphology and pyknotic nuclei, could be seen 6 and 24 hours after FPI (Fig. 1A). Damage was located to the injury site and the parietal cortex within the ipsilateral hemisphere. No damage was evident within the contralateral hemisphere or within the corpus callosum or hippocampus following FPI or sham procedures.

ular morphology and pyknotic nuclei, could be seen 6 and 24 hours after FPI (Fig. 1A). Damage was located to the injury site and the parietal cortex within the ipsilateral hemisphere. No damage was evident within the contralateral hemisphere or within the corpus callosum or hippocampus following FPI or sham procedures. Figure 1. Characterization of mild fluid percussion-induced brain injury. Individual hemispheres were digested 6, 24, and 48 hr after receiving fluid percussion injury (FPI)/sham procedures for flow cytometry or processed for histology. A, Neuronal damage via acid fuchsin. Significant neuronal damage was detected in the ipsilateral hemisphere 6 and 24 hr after receiving FPI. B, Myeloperoxidase (MPO) immunohistochemistry on brain sections 24 hr after sham/FPI procedure. Significant neutrophil infiltration was seen at 24 hr after FPI. C, Ipsilateral neutrophil quantification via flow cytometry. A small number of neutrophils were detected in the ipsilateral hemisphere 6, 24, and 48 hr after FPI. Data are represented as mean ± sem or sd (Fig. 1C) and were analyzed using two-way analysis of variance (n = 3–5 per group, *p < 0.05, **p < 0.01, ***p < 0.001).

psilateral neutrophil quantification via flow cytometry. A small number of neutrophils were detected in the ipsilateral hemisphere 6, 24, and 48 hr after FPI. Data are represented as mean ± sem or sd (Fig. 1C) and were analyzed using two-way analysis of variance (n = 3–5 per group, *p < 0.05, **p < 0.01, ***p < 0.001). Neuronal injury following FPI was associated with significant neutrophil infiltration in the ipsilateral hemisphere adjacent to the injury/craniotomy site compared with sham treatment (Fig. 1B). This was confirmed using flow cytometry, which identified significant yet mild infiltration into the ipsilateral (injured) hemisphere following FPI at all time points (Fig. 1C). The temporal quantification of infiltration differed slightly to that seen with immunohistochemistry, most likely due to the increased sensitivity of flow cytometry and its ability to count neutrophils that were extravasating from blood vessels. When healthy, untouched brains were digested, no neutrophils were detected.

he temporal quantification of infiltration differed slightly to that seen with immunohistochemistry, most likely due to the increased sensitivity of flow cytometry and its ability to count neutrophils that were extravasating from blood vessels. When healthy, untouched brains were digested, no neutrophils were detected. Mild Fluid Percussion Cortical Injury Induces Massive Pulmonary Interstitial Neutrophil Migration Without Pulmonary Vascular Leak Lungs from FPI and sham animals were assessed for evidence of inflammation. Both the alveolar and interstitial compartments were evaluated at 0, 6, 24, and 48 hours after FPI. In BALF, only resident alveolar macrophages were present from sham- or FPI-treated animals (Supplementary Fig. E4, Supplemental Digital Content 5, http://links.lww.com/CCM/D744; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). No differences in BALF cell counts, alveolar IL-1β levels, or vascular permeability were observed between FPI- and sham-treated animals (Supplementary Fig. E5, Supplemental Digital Content 6, http://links.lww.com/CCM/D745; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740).

/links.lww.com/CCM/D740). No differences in BALF cell counts, alveolar IL-1β levels, or vascular permeability were observed between FPI- and sham-treated animals (Supplementary Fig. E5, Supplemental Digital Content 6, http://links.lww.com/CCM/D745; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). To quantify pulmonary interstitial myeloid cell accumulation, whole lungs were digested after perfusion and lavage for multiparametric flow cytometric analysis as previously described (17). Significant neutrophil accumulation occurred within the interstitium 6 and 24 hours after FPI compared with sham treatment (Fig. 2A; and Supplementary Fig. E6, Supplemental Digital Content 7, http://links.lww.com/CCM/D74; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). This was confirmed using immunohistochemistry for myeloperoxidase (Fig. 2B). To investigate one potential mechanism driving high interstitial neutrophil accumulation, the intercellular adhesion molecule (ICAM)-1 expression of circulating and pulmonary neutrophils was assessed. No detectable changes in ICAM-1 expression were identified (Supplementary Fig. E7, Supplemental Digital Content 8, http://links.lww.com/CCM/D747; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). BALF was also analyzed on a proteome profiler array to characterize cytokine expression. Pooled samples from three mice per group were assessed. FPI was shown to increase the expression of 105 proteins, including neutrophil chemokines chemokine (C-C motif) ligand 2 (CCL2)/monocyte chemotactic protein (MCP)-1, chemokine (C-X-C motif) ligand 2 (CXCL2)/macrophage inflammatory protein (MIP)-2, IL-6, and tumor necrosis factor (TNF), all of which were absent following sham treatment, as well as an increase in RAGE expression (Supplementary Fig. E8, Supplemental Digital Content 9, http://links.lww.com/CCM/D748; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740; and Supplementary Table E1, Supplemental Digital Content 1, http://links.lww.com/CCM/D740).

nt following sham treatment, as well as an increase in RAGE expression (Supplementary Fig. E8, Supplemental Digital Content 9, http://links.lww.com/CCM/D748; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740; and Supplementary Table E1, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Figure 2. Pulmonary interstitial infiltration of neutrophils following concussion fluid percussion injury (FPI). Interstitial lung tissue was collected 6, 24, and 48 hr after FPI/sham procedure to assess signs of pulmonary injury/inflammation. A, Interstitial neutrophils. Significant neutrophil accumulation was present 6 and 24 hr after FPI procedure. B, Myeloperoxidase (MPO) immunohistochemistry on lung sections following sham/FPI procedure. Significantly more neutrophils were seen 6 hr after FPI, which confirmed the findings seen with flow cytometry. Data are represented as mean ± sd or sem (B) and were analyzed using two-way analysis of variance (n = 3–6 per group, *p < 0.05, ***p < 0.001). Intratracheal Acid Aspiration After FPI Results in Alveolar Neutrophilia and Pulmonary Hemorrhage The observation that mild concussive FPI results in massive interstitial neutrophil accumulation without alveolar neutrophil ingress suggested that the intra-alveolar compartment may be primed for subsequent intrapulmonary insult. To test this, we established a “double hit” model in which mice received a small volume of HCl, which in naive mice caused no lung injury, directly into the lungs immediately after FPI or sham injury.

ar neutrophil ingress suggested that the intra-alveolar compartment may be primed for subsequent intrapulmonary insult. To test this, we established a “double hit” model in which mice received a small volume of HCl, which in naive mice caused no lung injury, directly into the lungs immediately after FPI or sham injury. In naive mice, intratracheal HCl (pH, 1.75) induced no increase in BALF cell counts when compared with PBS (Fig. 3A). However, after mild concussive FPI, HCl administration significantly increased alveolar cell counts. This was associated with significant pulmonary hemorrhage (Fig. 3, B and C). Quantification of neutrophils confirmed significant alveolar infiltration in FPI-HCl–treated mice compared with sham-HCl (Fig. 3D). Nonsignificant neutrophil levels were detected within the pulmonary interstitium following HCl administration alone or in sham-HCl mice; however, levels were significantly increased following FPI-HCl (Fig. 3E).

rophils confirmed significant alveolar infiltration in FPI-HCl–treated mice compared with sham-HCl (Fig. 3D). Nonsignificant neutrophil levels were detected within the pulmonary interstitium following HCl administration alone or in sham-HCl mice; however, levels were significantly increased following FPI-HCl (Fig. 3E). Figure 3. Bronchoalveolar lavage fluid (BALF) and interstitial analysis following fluid percussion injury (FPI) and intratracheal hydrochloric acid (HCl). Fifty-microliter pH 1.75 HCl was instilled via an intratracheal route immediately after FPI/sham procedure and lungs retrieved 6 hr later. A, FPI-treated mice had significantly higher BALF cell counts 6 hr after receiving HCl compared with controls. B, BALF cell differential. HCl administration after both FPI and sham treatments resulted in neutrophil influx; however, FPI-treated mice had significant pulmonary hemorrhage. C, Representative sham-HCl and FPI-HCl cytocentrifuge preparations. Neutrophils and resident alveolar macrophage are the predominant cells present within sham-HCl BALF. RBCs, however, were also present following FPI-HCl. D, BALF neutrophil count. FPI-HCl–treated mice had significantly more neutrophils within the BALF than in controls. E, Interstitial neutrophil accumulation. Significant neutrophil accumulation was seen following FPI-HCl. Data are represented as mean ± sd and were analyzed using one-way analysis of variance (n = 4 per group, *p < 0.05). Scale bar represents 100 μm.PBS = phosphate-buffered saline.

ils within the BALF than in controls. E, Interstitial neutrophil accumulation. Significant neutrophil accumulation was seen following FPI-HCl. Data are represented as mean ± sd and were analyzed using one-way analysis of variance (n = 4 per group, *p < 0.05). Scale bar represents 100 μm.PBS = phosphate-buffered saline. Neutrophil Depletion With Anti-LY-6G Depleting Antibody Attenuates FPI-HCl Injury To determine if neutrophils had a direct association with alveolar pulmonary hemorrhage, neutrophil depletion studies were performed using an anti-LY-6G depleting antibody as described (15). Time course experiments confirmed that 24 hours after injection, anti-LY-6G–depleting antibody successfully depleted neutrophils within the pulmonary interstitium (Fig. 4A; and Supplementary Fig. E9, Supplemental Digital Content 10, http://links.lww.com/CCM/D749; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Total depletion of circulating neutrophils was confirmed throughout the duration of the double hit experiment (Fig. 4B).

the pulmonary interstitium (Fig. 4A; and Supplementary Fig. E9, Supplemental Digital Content 10, http://links.lww.com/CCM/D749; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Total depletion of circulating neutrophils was confirmed throughout the duration of the double hit experiment (Fig. 4B). Figure 4. Confirmation of neutrophil depletion with anti-lymphocyte antigen 6 complex locus G6D (LY-6G). Mice were preinjected with anti-LY-6G 24 hr before receiving fluid percussion injury. A, Pulmonary interstitial neutrophils. Neutrophils could still be identified following isotype control; however, anti-LY-6G successfully depleted neutrophils within the lung. B, Intravascular neutrophil mobilization. Anti-LY-6G injection resulted in total depletion of circulating neutrophils throughout the experiment. Data are represented as mean ± sd and were analyzed using Student t test and one-way analysis of variance (n = 3–6 per group, *p < 0.05, **p < 0.01, ***p < 0.001). A significant reduction in BALF cell counts was seen in the FPI-HCl double hit following neutrophil depletion (Fig. 5A). BALF macrophage levels remained unchanged (Supplementary Fig. E10, Supplemental Digital Content 11, http://links.lww.com/CCM/D750). This reduction in neutrophil number was associated with a decrease in pulmonary hemorrhage (Fig. 5B). Addition of anti-LY-6G–depleting antibody resulted in significant depletion of interstitial neutrophils (Fig. 5C).

ned unchanged (Supplementary Fig. E10, Supplemental Digital Content 11, http://links.lww.com/CCM/D750). This reduction in neutrophil number was associated with a decrease in pulmonary hemorrhage (Fig. 5B). Addition of anti-LY-6G–depleting antibody resulted in significant depletion of interstitial neutrophils (Fig. 5C). Figure 5. Hydrochloric acid (HCl) double hit at righting time with anti-lymphocyte antigen 6 complex locus G6D (LY-6G). Mice were preinjected with anti-LY-6G 24 hr before receiving fluid percussion injury (FPI)-HCl. Bronchoalveolar lavage fluid (BALF) was retrieved 6 hr after FPI. A, BALF analysis. Administration of anti-LY-6G resulted in a significant drop in BALF cell count due to a reduction in BALF neutrophils. B, BALF cell differential and representative cytocentrifuge preparations of FPI-HCl and FPI-HCl + anti-LY-6G. Anti-LY-6G resulted in a significant drop in BALF neutrophils and a nonsignificant reduction in pulmonary hemorrhage. C, Interstitial neutrophil accumulation. Administration of anti-LY-6G resulted in a significant reduction in interstitial neutrophil accumulation. D, BALF total protein. An increase in vascular permeability was observed in FPI-HCl–treated mice compared with sham-HCl. Neutrophil depletion resulted in a reduction in vascular permeability. Data are represented as mean ± sd and were analyzed using Student t test and one-way analysis of variance (n = 3–6 per group, *p < 0.05, **p < 0.01, ***p < 0.001). Scale bar represents 100 μm.

observed in FPI-HCl–treated mice compared with sham-HCl. Neutrophil depletion resulted in a reduction in vascular permeability. Data are represented as mean ± sd and were analyzed using Student t test and one-way analysis of variance (n = 3–6 per group, *p < 0.05, **p < 0.01, ***p < 0.001). Scale bar represents 100 μm. The increase in vascular permeability observed in FPI-HCl–treated mice was prevented following neutrophil depletion (Fig. 5D). When inflammatory cytokine levels were assessed, a trend of increased TNF, MCP-1, and IL-6 was seen in FPI-HCl–treated mice compared with sham-HCl (Supplementary Fig. E10, Supplemental Digital Content 11, http://links.lww.com/CCM/D750; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Neutrophil depletion resulted in a reduction in TNF and IL-6 levels, however, had no effect on MCP-1 levels.

een in FPI-HCl–treated mice compared with sham-HCl (Supplementary Fig. E10, Supplemental Digital Content 11, http://links.lww.com/CCM/D750; legend, Supplemental Digital Content 1, http://links.lww.com/CCM/D740). Neutrophil depletion resulted in a reduction in TNF and IL-6 levels, however, had no effect on MCP-1 levels. DISCUSSION Lung injury following TBI is a common and serious consequence. Understanding the pathophysiologic mechanisms that underlie the development of lung injury after TBI may help identify novel therapeutic avenues because current treatment interventions are limited. The brain-lung axis has been postulated to involve complex reciprocal cross talk (18), and pulmonary dysfunction is a predictor of poor neurologic outcome and mortality (3). Hence, it is now well recognized that ventilatory strategies after brain injury should include lung-protective approaches (19), which also have the potential to increase the number of eligible lungs for donation (20). We show for the first time that mild TBI increases pulmonary susceptibility to a secondary innocuous insult such as miniscule acid microaspiration. Even more importantly, we identify pulmonary interstitial neutrophils as the key cells responsible for driving augmented injury.

number of eligible lungs for donation (20). We show for the first time that mild TBI increases pulmonary susceptibility to a secondary innocuous insult such as miniscule acid microaspiration. Even more importantly, we identify pulmonary interstitial neutrophils as the key cells responsible for driving augmented injury. To explore the brain-lung axis, we hypothesized that subclinical pulmonary priming may contribute to subsequent lung injury. To test this, we developed an experimental model of TBI and subsequently delivered acid to the lung. An injury force of 1.5 ATM was chosen because this was expected to produce a moderate yet fully clinically reversible phenotype mimicking a concussion (21). To ensure sufficient cortical injury, the righting times for mice to recover following FPI/sham procedure were recorded as previous studies in rats have shown correlation between length of unconsciousness and degree of pathologic consequences (22). An exclusion time of 270 seconds was chosen (23); hence, any animal receiving FPI with a righting time less than 270 seconds was removed from the study. At an injury level of 1.5 ATM, righting times were well above 270 seconds, no mortality was seen and no immediate or long-term clinical sequelae were observed once the mice began to mobilize.

seconds was chosen (23); hence, any animal receiving FPI with a righting time less than 270 seconds was removed from the study. At an injury level of 1.5 ATM, righting times were well above 270 seconds, no mortality was seen and no immediate or long-term clinical sequelae were observed once the mice began to mobilize. To characterize the cerebral injury, brains were sectioned and stained with acid fuchsin, an anionic dye that stains acidophilic irreversibly damaged neurons a dark red (24, 25). Damage was observed directly below the injury site, resulting from the fluid pulse. By 24 hours, neuronal damage had increased in size, suggesting secondary injury. This was associated with significant neutrophil infiltration 6, 24, and 48 hours after FPI in the ipsilateral hemisphere. Neutrophils, capable of secreting neurotoxic substances (26), exacerbate excitotoxic insults and increase neuronal death in vitro (27). DAMPs (11) and proinflammatory cytokines (IL-1β, IL-6, and TNF) (28) released from the injured brain attract inflammatory cells to contused cerebral tissue. The extent of neuronal damage (confined only to cortex and small cell number), small degree of neutrophil infiltration (< 500 cells), and lack of any overt clinical signs between FPI and sham groups demonstrated that this injury was more representative of mild “concussive” TBI, even though 1.5 ATM has been previously considered within the moderate severity range in mice (21).

small cell number), small degree of neutrophil infiltration (< 500 cells), and lack of any overt clinical signs between FPI and sham groups demonstrated that this injury was more representative of mild “concussive” TBI, even though 1.5 ATM has been previously considered within the moderate severity range in mice (21). BALF differential cell counts showed no differences in alveolar cell count between FPI- and sham-treated animals at any time point. Alveolar neutrophil infiltration was absent, and no differences were observed in vascular permeability.

small cell number), small degree of neutrophil infiltration (< 500 cells), and lack of any overt clinical signs between FPI and sham groups demonstrated that this injury was more representative of mild “concussive” TBI, even though 1.5 ATM has been previously considered within the moderate severity range in mice (21). BALF differential cell counts showed no differences in alveolar cell count between FPI- and sham-treated animals at any time point. Alveolar neutrophil infiltration was absent, and no differences were observed in vascular permeability. Surprisingly, mild TBI induced extensive interstitial neutrophil accumulation, which remained elevated 24 hours after FPI. There were no observable differences in pulmonary vascular permeability, suggesting that a sympathetic catecholamine storm was not responsible for driving high interstitial neutrophil accumulation in this model as endothelial dysfunction is a recognized major component of sympathetic catecholamine storms. TBI has previously been shown to increase levels of pulmonary IL-1β, IL-6 (13), prostaglandin-synthesizing enzyme cyclooxygenase-2 (5), and leukotriene B4; the latter being particularly chemotactic for neutrophils (13). Despite no differences in IL-1β being seen in this model at 6 hours post-FPI, increased expression of other direct or indirect neutrophil chemoattractants were identified using the proteome profiler array (CCL2/MCP-1, CXCL1/KC, CXCL2/MIP-2, IL-6, TNF) (13, 29–31). RAGE was also up-regulated following FPI. This receptor has been implicated with the development of pulmonary inflammation following TBI through its interaction with high-mobility group box-1, a DAMP released from the injured brain that leads to sustained activation of nuclear factor κ-light-chain-enhancer of activated B cells (leading to proinflammatory cytokine production) and increased RAGE expression to ensure amplification of the inflammatory signal (11, 32). TLR-4 activation by DAMPs may also account for the increase in proinflammatory cytokines (10).

to sustained activation of nuclear factor κ-light-chain-enhancer of activated B cells (leading to proinflammatory cytokine production) and increased RAGE expression to ensure amplification of the inflammatory signal (11, 32). TLR-4 activation by DAMPs may also account for the increase in proinflammatory cytokines (10). Seeing as no differences in ICAM-1 expression were detected in circulating and pulmonary neutrophils, reverse transmigration of neutrophils from the injured brain to the lung was ruled out (33). To investigate whether mild TBI increased susceptibility to pulmonary injury, an acid microaspiration “double hit” model was established. Acid microaspiration is a common sequelae in patients who are concussed or lose airway protection as damage to the CNS has been associated with an increased risk of stomach content microaspiration (4). To model acid microaspiration following TBI, HCl (pH 1.75 in sodium chloride resembling the acidity and osmolality similar to that of patient gastric sections) (34) was administered via an intratracheal route immediately following FPI (after recording the righting time). Pulmonary consequences were evaluated 6 hours later.

odel acid microaspiration following TBI, HCl (pH 1.75 in sodium chloride resembling the acidity and osmolality similar to that of patient gastric sections) (34) was administered via an intratracheal route immediately following FPI (after recording the righting time). Pulmonary consequences were evaluated 6 hours later. As expected, HCl administration in naive mice was innocuous due to the large size of mice (25–35 g) and high pH when compared with previous work (16). HCl following FPI, however, resulted in a significant increase in alveolar neutrophil infiltration which was not seen in sham-HCl mice. This was accompanied with increases in the proinflammatory cytokines TNF, MCP-1, and IL-6 and significant interstitial neutrophil accumulation. Alveolar and interstitial neutrophil accumulation were associated with pulmonary hemorrhage and increased vascular permeability. Neutrophil-specific depletion with an anti-LY-6G antibody confirmed neutrophil dependence of vascular leak.

ytokines TNF, MCP-1, and IL-6 and significant interstitial neutrophil accumulation. Alveolar and interstitial neutrophil accumulation were associated with pulmonary hemorrhage and increased vascular permeability. Neutrophil-specific depletion with an anti-LY-6G antibody confirmed neutrophil dependence of vascular leak. Neutrophil-mediated lung injury is central to the pathogenesis of many lung conditions including ARDS (17, 35, 36), and the data presented here suggest that even in mild experimental TBI, neutrophilic interstitial infiltration is an important prelude to subsequent innocuous intrapulmonary microaspiration. Although the mechanisms remain to be delineated, inhibition or attenuation of neutrophil recruitment/activity may be a potential therapeutic intervention following TBI. These data suggest that TBI patients with reduced consciousness may be susceptible to the immediate effects of microaspiration or indeed to other injurious iatrogenic insults such as ventilation-induced stretch that could direct the intra-alveolar recruitment of neutrophils from the interstitial pool.

on following TBI. These data suggest that TBI patients with reduced consciousness may be susceptible to the immediate effects of microaspiration or indeed to other injurious iatrogenic insults such as ventilation-induced stretch that could direct the intra-alveolar recruitment of neutrophils from the interstitial pool. The findings in this article support the underlying hypothesis that the brain-lung axis may contribute to priming for secondary neutrophil-dependent lung injury. To investigate mechanisms of neutrophil recruitment, further experiments include blocking key neutrophil pathways that may contribute to recruitment and activation such as phosphatidylinositide 3-kinase and inhibitors of early alarm cytokines such as TNF. Additionally, augmenting neutrophil clearance through promoting apoptosis is also a viable experimental intervention (37). Early administration of glucocorticoids may also help reduce inflammation; however, care must be given to the dosage and timing because glucocorticoids also promote neutrophil survival, thus potentially prolonging inflammation.

ophil clearance through promoting apoptosis is also a viable experimental intervention (37). Early administration of glucocorticoids may also help reduce inflammation; however, care must be given to the dosage and timing because glucocorticoids also promote neutrophil survival, thus potentially prolonging inflammation. Furthermore, in the clinical setting, it will be necessary to prove whether these observations also prevail after minor TBI. Human experimental studies are now warranted to quantify the accumulation of neutrophils in the lung post-TBI through human molecular imaging approaches including both whole-body approaches such as positron emission tomography (38) and evolving endomicroscopic approaches that have the potential to visualize neutrophil accumulation at the bedside in critical care (39). These studies would enable targeted therapeutic intervention. ACKNOWLEDGMENTS We thank the Queen’s Medical Research Institute Flow Cytometry and Histology Facilities at the University of Edinburgh for assistance. Supplementary Material Drs. Andrews, Rhodes, and Dhaliwal conceived the project. Drs. Humphries, Rhodes, and Dhaliwal designed the experiments. Dr. Humphries, Dr. O’Neill, and Ms. Scholefield performed experiments. Drs. Dorward and Rhodes provided methods. Drs. Mackinnon, Rossi, Haslett, Andrews, and Rhodes provided guidance and edited the article. Drs. Humphries and Dhaliwal wrote the article.

ct. Drs. Humphries, Rhodes, and Dhaliwal designed the experiments. Dr. Humphries, Dr. O’Neill, and Ms. Scholefield performed experiments. Drs. Dorward and Rhodes provided methods. Drs. Mackinnon, Rossi, Haslett, Andrews, and Rhodes provided guidance and edited the article. Drs. Humphries and Dhaliwal wrote the article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported mainly by a U.K. Medical Research Council (MRC) PhD Studentship to Dr. Humphries. Also supported by the MRC (MR/K013386/1 [Drs. Rossi and Haslett]) and the Wellcome Trust WT096497 (Dr. Dorward). Dr. Dorward received support for article research from Wellcome Trust/Charity Open Access Fund. Drs. Humphries and Haslett received support for article research from Research Councils UK (RCUK). Dr. Dorward’s institution received funding from Wellcome Trust. Dr. Haslett’s institution received funding from Medical Research Council studentship. Drs. Haslett and Dhaliwal received funding from Edinburgh Molecular Imaging. The remaining authors have disclosed that they do not have any potential conflicts of interest.

Background The process of fibrinolysis is inevitable to regain microvessel patency and restore vital organ perfusion after intravascular clotting. Endothelial tissue-type plasminogen activator (t-PA) ensures clot dissolution by converting plasminogen into plasmin at the site of primary vascular damage averting permanent circulatory compromise and subsequent thrombotic organ failure. In contrast, primary hyperfibrinolysis occurs without preceding intravascular clotting and is associated with poor outcome in several critical conditions including trauma and sepsis (1, 2). In particular in traumatic coagulopathy, fibrinolysis has been investigated intensively and identified as an early and independent predictor of mortality (1, 3). Although underlying mechanisms likely differ from that in trauma-associated fibrinolysis, which involves tissue injury and significant crystalloid hemodilution (4), previous studies likewise reported an association between the presence of fibrinolysis and unfavorable prognosis in cardiac arrest (5, 6), with the highest fibrinolytic activity found in patients with early death.

trauma-associated fibrinolysis, which involves tissue injury and significant crystalloid hemodilution (4), previous studies likewise reported an association between the presence of fibrinolysis and unfavorable prognosis in cardiac arrest (5, 6), with the highest fibrinolytic activity found in patients with early death. Hypoperfusion resulting in release of endothelial t-PA (6) is considered one possible mechanism of fibrinolysis occurring during cardiopulmonary resuscitation (CPR) (7). The importance of hypoxia as a causal factor is best seen in the severe hyperfibrinolysis of young and previously healthy drowning patients where preceding activation of coagulation can be excluded (6). Hence, the extent of fibrinolysis at admission reflects prolonged or insufficient resuscitation efforts and may therefore provide reliable prediction of poor outcome (6). In this context, previous studies have already identified fibrin degradation products, such as d-dimer, as an early indicator of unfavorable outcome in cardiac arrest (8, 9). D-dimer is a well-known marker of fibrinolysis and can be routinely assessed, but however lacks specificity, which makes the definition of a reliable predictive threshold unlikely. Yet, thrombelastometry has proven useful for point-of-care detection of increased fibrinolysis in cardiac arrest (6, 7, 10). The current study aimed to assess whether there is an optimal fibrinolysis cutoff value assessed by thrombelastometry to predict poor outcome in a preselected cohort of successfully resuscitated adult patients with out-of-hospital cardiac arrest.

Yet, thrombelastometry has proven useful for point-of-care detection of increased fibrinolysis in cardiac arrest (6, 7, 10). The current study aimed to assess whether there is an optimal fibrinolysis cutoff value assessed by thrombelastometry to predict poor outcome in a preselected cohort of successfully resuscitated adult patients with out-of-hospital cardiac arrest. METHODS Study Population Eligible for inclusion were adults (≥ 18 yr) with out-of-hospital cardiac arrest of presumed cardiac origin, subjected to targeted temperature management, who had achieved return of spontaneous circulation (ROSC) at admission to the ICU section of the emergency department at the Medical University of Vienna. Exclusion criteria comprised thrombolytic therapy and application of intravascular cooling or extracorporeal bypass device, as both may affect fibrinolytic activity. All patients were treated with therapeutic hypothermia at a target temperature range of 33°C ± 1°C. Patients were cooled using cooling pads (EMCOOLS Flex.Pad, Emcools AG, Pfaffstaetten, Austria) (11) or water-circulating gel-coated pads (Arctic Sun 5000 Temperature Management System; Medivance, Louisville, CO) with or without cold fluids. Target temperature between 32°C and 34°C was maintained for 24 hours after first achievement of less than 34°C. Rewarming was performed at a rate of 0.25–0.5°C per hour. Body temperature was recorded with an esophageal and bladder probe.

erature Management System; Medivance, Louisville, CO) with or without cold fluids. Target temperature between 32°C and 34°C was maintained for 24 hours after first achievement of less than 34°C. Rewarming was performed at a rate of 0.25–0.5°C per hour. Body temperature was recorded with an esophageal and bladder probe. Resuscitation-related variables were analyzed and reported according to Utstein recommendations as described previously (12). The primary endpoint was a composite of poor neurologic function or death, defined as a Cerebral Performance Category (CPC) of 3–5 (severe cerebral disability; coma or vegetative state; or brain death, respectively) at day 30 post resuscitation (13). Neurologic function at 30 days was assessed by study fellows through structured face-to-face interview with the patient or by means of structured telephone interview with the patient, the relatives, treating physicians, or nursing home members. Study fellows assessing outcome were all blinded to results obtained by rotational thrombelastometry (ROTEM).

assessed by study fellows through structured face-to-face interview with the patient or by means of structured telephone interview with the patient, the relatives, treating physicians, or nursing home members. Study fellows assessing outcome were all blinded to results obtained by rotational thrombelastometry (ROTEM). Sustained ROSC was defined as recovery of spontaneous circulation for more than 20 minutes. No- and low-flow intervals were defined as the time from collapse to initiation of CPR and the time from CPR initiation to sustained ROSC, respectively. No- and low-flow intervals were determined through immediate structured telephone interviews with the dispatch center, the emergency physicians, and paramedics at the scene and the bystander who performed the emergency call. The International Society on Thrombosis and Haemostasis disseminated intravascular coagulation (DIC) eight-point score was applied for DIC calculation (14). A score greater than or equal to 5 is compatible with overt DIC (Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/CCM/D848). For sample size calculation, we used the methods described by Buderer (15) based on specificity. We anticipated a specificity level of 90% with 95% CI (width of 10%) and a given prevalence of hyperfibrinolysis of 40% based on previously published results (7, 10). The high level of anticipated specificity was estimated based on our own results from drowning patients, where a very high predictive value for death was demonstrated, if signs of hyperfibrinolysis were present (6).

f 10%) and a given prevalence of hyperfibrinolysis of 40% based on previously published results (7, 10). The high level of anticipated specificity was estimated based on our own results from drowning patients, where a very high predictive value for death was demonstrated, if signs of hyperfibrinolysis were present (6). The study was approved by the Ethics Committee of the Medical University of Vienna and carried out in accordance with the Declaration of Helsinki. A waiver was obtained for informed consent at admission, and patients were informed of their study participation on regaining consciousness. Laboratory Methods Blood sampling for thrombelastometry and laboratory studies were performed as soon as vascular access was available. Analysis of whole blood viscoelastic properties was done in 3.8% sodium-citrated whole blood samples using ROTEM (TEM International GmBH, Munich, Germany) as described previously (16). The following ROTEM tests were applied: EXTEM, which tests the extrinsic pathway of coagulation, and APTEM, which corresponds to EXTEM but additionally comprises the antifibrinolytic aprotinin to definitively reveal the presence of increased fibrinolysis. Fibrinolysis is given as maximum lysis (ML) (%), which represents the percentile difference between the highest and lowest clot amplitude (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CCM/D848; reference range < 15%, as specified by the manufacturer).

the presence of increased fibrinolysis. Fibrinolysis is given as maximum lysis (ML) (%), which represents the percentile difference between the highest and lowest clot amplitude (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CCM/D848; reference range < 15%, as specified by the manufacturer). For enzyme-linked immunoassay (ELISA) analysis, blood was collected into tubes containing EDTA or 3.8% citrated plasma. Obtained samples were centrifuged for 10 minutes at 2,000g. Plasma was stored at –80°C until being tested. Tissue plasminogen activator antigen (TECHNOZYM t-PA Combi Actibind; Technoclone, Vienna, Austria), plasminogen activator inhibitor (PAI)–1 (TECHNOZYM PAI-1 Actibind; Technoclone) (16), and prothrombin fragments F1 + 2 (EnzygnostF 1 + 2; Siemens, Marburg, Germany) (17) assays were performed using commercially available ELISA kits. The lower limits of detection are 0.01 ng/mL (t-PA antigen), 20 pmol/L (F1 + 2), and 0.49 IU/mL (PAI-1), respectively. All assays were performed according to manufacturer’s instructions.

in fragments F1 + 2 (EnzygnostF 1 + 2; Siemens, Marburg, Germany) (17) assays were performed using commercially available ELISA kits. The lower limits of detection are 0.01 ng/mL (t-PA antigen), 20 pmol/L (F1 + 2), and 0.49 IU/mL (PAI-1), respectively. All assays were performed according to manufacturer’s instructions. Statistical Analysis Variables are presented as absolute values (n), relative frequencies (%), and median (interquartile ranges [IQRs]). Prevalence of increased fibrinolysis and overt DIC is given as a proportion with a 95% CI. Potentially missing data for demographic variables were not imputated. Between-group comparisons were performed using the Mann-Whitney U test for continuous variables or the chi-square test/Fisher exact test for nominal variables. We performed exact bivariable logistic regression to estimate the effect on poor neurologic function or death at day 30 of ML and remaining candidate predictors, which were judged to be clinically plausible, including age, no-flow time, low-flow time, pH, lactate, d-dimer levels, sex, CPC prior to cardiac arrest, arrest site, witness status, presence or absence of bystander resuscitation, initial rhythm, epinephrine dose administered during resuscitation, and the rate of sustained ROSC at admission. Results are given as odds ratios/median unbiased estimates with 95% CI and are available in Supplemental Table 4 (Supplemental Digital Content 1, http://links.lww.com/CCM/D848).

e of bystander resuscitation, initial rhythm, epinephrine dose administered during resuscitation, and the rate of sustained ROSC at admission. Results are given as odds ratios/median unbiased estimates with 95% CI and are available in Supplemental Table 4 (Supplemental Digital Content 1, http://links.lww.com/CCM/D848). The optimal cutoff for ML to predict poor 30-day outcome was assessed by computing a receiver operating characteristic curve. Specificity and sensitivity were calculated with 95% CI. The Kaplan-Meier method was used to describe survival according to the optimal cutoff for ML. Generally, a two-sided p value of less than 0.05 was considered statistically significant. We used IBM SPSS Statistical Software, Version 22.0 (IBM Corp., Armonk, NY) and Stata Statistical Software: Release 15 (StataCorp., College Station, TX) for statistical analysis and GraphPad Prism Version 7.00 for Windows (GraphPad Software, La Jolla, CA) to draw figures.

0.05 was considered statistically significant. We used IBM SPSS Statistical Software, Version 22.0 (IBM Corp., Armonk, NY) and Stata Statistical Software: Release 15 (StataCorp., College Station, TX) for statistical analysis and GraphPad Prism Version 7.00 for Windows (GraphPad Software, La Jolla, CA) to draw figures. RESULTS Seventy-eight patients (median age 59 yr; 47–69; 78% male) were included in the study. Results from thrombelastometry at admission and all data for the primary outcome were available. Median core temperature at admission was 35.3°C (34.8–35.8°C). Fibrinolysis exceeding the normal reference value of 15% was present in 36% of patients (28/78; 95% CI, 25–48%). The rate of overt DIC was 4% overall (3/78; 95% CI, 1–11%). In total, 54% of patients (42/78) had a poor 30-day outcome including 23 nonsurvivors (30%). Kaplan-Meier estimates of survival to day 30 according to ML cutoff of greater than or equal to 20% are available with the supplement (Supplemental Fig. 2, Supplemental Digital Content 1, http://links.lww.com/CCM/D848). While nine patients died due to multiple organ failure, in 14 patients a decision to withdraw life-sustaining therapy was made by treating physicians after determination of unfavorable neurologic prognosis. Characteristics of the study patients according to 30-day outcome (CPC) are shown in Table 1. TABLE 1. Characteristics of the Study Patients According to 30-Day Outcome

While nine patients died due to multiple organ failure, in 14 patients a decision to withdraw life-sustaining therapy was made by treating physicians after determination of unfavorable neurologic prognosis. Characteristics of the study patients according to 30-day outcome (CPC) are shown in Table 1. TABLE 1. Characteristics of the Study Patients According to 30-Day Outcome The admission ML cutoff predicting poor outcome with 100% specificity (95% CI, 90–100) was greater than or equal to 20% (sensitivity 41%; 95% CI, 26–58) (Fig. 1). The corresponding positive and negative predictive values were 100% and 59%, respectively. The receiver operating characteristic curve of ML for prediction of poor 30-day outcome as well as specificity, sensitivity, and cumulative frequency distributions of covariables are given with the supplement (Supplemental Figs. 3 and 4, Supplemental Digital Content 1, http://links.lww.com/CCM/D848).

59%, respectively. The receiver operating characteristic curve of ML for prediction of poor 30-day outcome as well as specificity, sensitivity, and cumulative frequency distributions of covariables are given with the supplement (Supplemental Figs. 3 and 4, Supplemental Digital Content 1, http://links.lww.com/CCM/D848). Figure 1. Sensitivity and specificity (%) of increasing fibrinolysis values (maximum lysis [ML], %) to predict poor neurologic function or death at day 30 after admission (n = 78). Specificity (gray squares), sensitivity (black dots, both left y-axis), cumulative frequency distribution (%, light-colored and rich-colored red area, right y-axis), 100% specificity value (vertical dashed red line), percentage of patients predicted with 100% specificity to have poor neurologic function or death (rich-colored red area). Error bars indicate 95% CI. ML equal to or greater than 20% had a 100% specificity for poor neurologic function or death. ML greater than or equal to 20% enables 100% poor outcome prediction in every fifth successfully resuscitated cardiac arrest patient.

have poor neurologic function or death (rich-colored red area). Error bars indicate 95% CI. ML equal to or greater than 20% had a 100% specificity for poor neurologic function or death. ML greater than or equal to 20% enables 100% poor outcome prediction in every fifth successfully resuscitated cardiac arrest patient. Patients presenting with a ML greater than or equal to 20% (17/78, 22%; 59% male) were older (median 64 yr [IQR, 56–71 yr] vs 56 yr [46–67 yr]; p = 0.004) and tended to have a higher frequency of nonshockable rhythm (35% vs 15%; p = 0.09), had on average a 83% longer low-flow time (44 min [35–58 min] vs 24 min [15–36 min]; p < 0.001), slightly higher lactate (8.3 mmol/L [6–14 mmol/L] vs 6.9 mmol/L [4–10 mmol/L]; p < 0.001), 2.7-fold higher d-dimer (20 μg/mL [10–37 μg/mL] vs 7.4 μg/mL [2.7–14 μg/mL]; p < 0.001), and 79% higher t-PA levels (52 ng/mL [26–79 ng/mL] vs 29 ng/mL [17–49 ng/mL]; p = 0.036) and tended to receive higher doses of epinephrine (3 mg [2–6 mg] vs 2 mg [1–4 mg]; p = 0.06). Median core temperature at admission (35.3°C [34.0–35.4°C] vs 35.3°C [34.8–35.8°C]; p = 0.34), platelet counts (215 × 109/L [165–293 × 109/L] vs 205 × 109/L [176–244 × 109/L]; p = 0.55), prothrombin times (80% [68–93%] vs 77% [64–90%]; p = 0.20), fibrinogen levels (238 mg/dL [167–335 mg/dL] vs 295 mg/dL [246–322 mg/dL]; p = 0.08), PAI-1 levels (1.15 IU/mL [0.14–2.87 IU/mL] vs 0.70 IU/mL [0–3.32 IU/mL]; p = 0.60), and prothrombin fragments levels (1,201 pmol/L [518–1,201 pmol/L] vs 1,134 pmol/L [513–1,201 pmol/L]; p = 0.82) were similar compared with patients without increased fibrinolysis. Patients presenting with a ML greater than or equal to 20% had a median time to death of 6.5 days. The time course of fibrinolysis from admission to rewarming is available in Supplemental Tables 2 and 3 (Supplemental Digital Content 1, http://links.lww.com/CCM/D848). ML greater than or equal to 20% was an independent predictor of poor neurologic function or death (Supplemental Table 4, Supplemental Digital Content 1, http://links.lww.com/CCM/D848).

rom admission to rewarming is available in Supplemental Tables 2 and 3 (Supplemental Digital Content 1, http://links.lww.com/CCM/D848). ML greater than or equal to 20% was an independent predictor of poor neurologic function or death (Supplemental Table 4, Supplemental Digital Content 1, http://links.lww.com/CCM/D848). DISCUSSION The current study prospectively investigated the value of fibrinolysis to predict 30-day outcome early after cardiac arrest. The study was built on previous studies identifying hypoperfusion and hypoxia as triggers of primary fibrinolysis (6, 7). Although prolonged or poor resuscitation efforts may cause hyperfibrinolysis, the prognostic relevance of fibrinolysis occurring during CPR, however, remained unclear. This study specifically tested the hypothesis that increased fibrinolysis in thrombelastometry may predict poor outcome. The prevalence of increased fibrinolysis (ML > 15%) in our study is in good agreement with previous findings (10). However, we were interested in the optimal cutoff for fibrinolysis to specifically predict poor outcome. The greater than or equal to 20% cutoff value that predicted poor outcome with 100% specificity found in the current study corresponds well to the cutoff for increased fibrinolysis recently observed in healthy volunteers (18).

were interested in the optimal cutoff for fibrinolysis to specifically predict poor outcome. The greater than or equal to 20% cutoff value that predicted poor outcome with 100% specificity found in the current study corresponds well to the cutoff for increased fibrinolysis recently observed in healthy volunteers (18). From previous research, we interpret fibrinolysis as a marker of tissue hypoperfusion due to prolonged resuscitation efforts or poor-quality CPR performance and consecutive accumulation of t-PA (6). This hypothesis is mainly based on data obtained from patients with drowning-related out-of-hospital cardiac arrest (OHCA), who are characterized by severely prolonged no- and low-flow intervals along with massive bleeding at hospital admission (6). Their bleeding phenotype was accompanied by increased t-PA levels and absent clotting signature in thrombelastometry, which were reversed by adding aprotinin in vitro. A subsequent forearm-ischemia model conducted in healthy volunteer confirmed the increase of t-PA levels following interruption of arterial blood flow. Further previous studies likewise reported high plasma fibrinolytic activity in cardiac arrest patients with early death (5) and a good correlation between t-PA levels and markers of hypoperfusion (7). In accordance, patients with ML greater than or equal to 20% had significantly higher t-PA levels at admission. Levels of the t-PA inhibiting protein PAI-1 were comparatively low in patients with and without increased fibrinolysis, which is in agreement with current literature (19). This relates to the lack of a storage compartment for PAI-1—in contrast to t-PA stored in endothelial cells—which forbids a readily available release upon hypoxic conditions. Upon acute t-PA release, PAI-1 is consumed (20, 21), and therefore it seems plausible that its activity is low in the very early phase after arrest. PAI-1 levels have been found to likewise increase only within 6 hours after acute hypoxia, contributing to the fibrinolytic shutdown in the subacute phase of the postcardiac arrest syndrome (5, 22).

I-1 is consumed (20, 21), and therefore it seems plausible that its activity is low in the very early phase after arrest. PAI-1 levels have been found to likewise increase only within 6 hours after acute hypoxia, contributing to the fibrinolytic shutdown in the subacute phase of the postcardiac arrest syndrome (5, 22). Low-flow times and lactate levels in the current study were significantly higher in patients with a ML greater than or equal to 20%, which supports the above findings. Likewise, the rate of nonshockable rhythm tended to be higher, which may suggest longer preceding no-flow intervals before initiation of CPR in these patients. No-flow times did not differ significantly between groups, but are, however, usually only estimates with low reliability. We suggest that fibrinolysis is mainly triggered by hypoperfusion which could also explain the low negative predictive value of ML found in the current study. It is conceivable that patients with prolonged no-flow times but high-quality CPR performance—and thus, hypothetically, rapid clearance of t-PA—suffer poor neurologic recovery despite lack of fibrinolysis at admission. The quality of CPR performance is, however, difficult to assess routinely.

ound in the current study. It is conceivable that patients with prolonged no-flow times but high-quality CPR performance—and thus, hypothetically, rapid clearance of t-PA—suffer poor neurologic recovery despite lack of fibrinolysis at admission. The quality of CPR performance is, however, difficult to assess routinely. An alternative or possibly contributing mechanism of increased fibrinolysis in cardiac arrest may be thrombin-related plasmin activation following blood exposure to interstitial tissue factor through hypoxic endothelial damage. Along these lines, Adrie et al (5) found a significant association between thrombin levels and organ dysfunction after successful resuscitation. Thrombin may also contribute to delayed t-PA (23) and/or urokinase-type plasminogen activator (24) release from endothelial cells. In our study, patients’ prothrombin fragments were markedly elevated in both, patients with and without increased fibrinolysis, suggesting early thrombin generation in cardiac arrest. Yet, although thrombin-mediated coagulopathy likely becomes important hours or days after successful resuscitation, possibly as part of the postcardiac arrest syndrome, its contribution to increased fibrinolysis detected by thrombelastometry very early after successful resuscitation is questionable. Furthermore, in tissue factor/thrombin-driven coagulopathy, we would expect substantial consumption of clotting factors and platelets resulting in what is referred as DIC with fibrinolytic phenotype (25).

to increased fibrinolysis detected by thrombelastometry very early after successful resuscitation is questionable. Furthermore, in tissue factor/thrombin-driven coagulopathy, we would expect substantial consumption of clotting factors and platelets resulting in what is referred as DIC with fibrinolytic phenotype (25). Consumptive coagulopathy, however, was very rare in our study patients, and the rate of overt DIC at admission did not differ between patients with and those without increased fibrinolysis. The low prevalence of overt DIC in our study patients, however, is in striking contrast to data from a retrospective Asian study, which reported overt DIC rates of 33% in resuscitated cardiac arrest with greater than 90% hospital mortality (26). As we analyzed a highly preselected subset of patients, the results of the two studies are difficult to compare. Yet, it would be of interest whether these differences result from different resuscitation policies (including termination-of-care rules) and whether overt DIC simply represents a premortem sign. Another possible contributor to the impairment of coagulation may be hypothermia, but, however, a substantial impact on increased fibrinolysis in our patients seems unlikely. Median body temperature at admission was above 35°C, which is in agreement with recent literature (27), and did not differ significantly between patients with and without ML greater than or equal to 20%. Furthermore, previous studies did not report patterns of increased fibrinolysis under intentional hypothermia (28–31).

dian body temperature at admission was above 35°C, which is in agreement with recent literature (27), and did not differ significantly between patients with and without ML greater than or equal to 20%. Furthermore, previous studies did not report patterns of increased fibrinolysis under intentional hypothermia (28–31). We were further interested in whether exogenous epinephrine may contribute to the extent of increased fibrinolysis in cardiac arrest. Catecholamines are known to promote release of t-PA from the endothelium (32), and the amount of epinephrine given during resuscitation has previously been linked to poor outcome regardless of the length of resuscitation (33). In the current study, there was an insignificant trend of higher cumulative epinephrine doses in patients who presented with ML greater than or equal to 20%. In regression analysis, however, the dosage of epinephrine administered had no effect on mortality after adjustment for ML greater than or equal to 20, which may suggest no causative relationship. Whether exogenous epinephrine directly contributes to fibrinolysis or simply reflects longer resuscitation efforts, and thus prolonged preceding hypoperfusion remains speculative and might be answered in a further study. Furthermore, yet, no data are available on the interaction between endogenous catecholamine levels and fibrinolysis in cardiac arrest. Our own recent data derived from a subset of 1,188 cardiac arrest patients underline the outcome-predictive role of sympathoadrenergic activation following resuscitation indicated by both, increased short- and long-term mortality in cardiac arrest along with the amount of immature peripheral neutrophils at admission (34). It is unclear whether this simply reflects the gradual extent of stress to the body following resuscitation or whether endogenous catecholamines may have a role in subsequent alterations of the coagulation and fibrinolytic system, as it has been described in trauma and septic patients (35, 36). In addition to increased fibrinolysis caused by stress-mediated sympathoadrenergic catecholamine release, neutrophil elastase has been linked as trigger for fibrinolysis in trauma (25). However, to date, no comparable data are available on the contribution of neutrophil elastase to promote fibrinolysis in cardiac arrest.

id chromatography-mass spectrometry quadrupole time-of-flight instrument in profiling mode, according to protocols described by Sarafian et al (13). The identity of the bile acids was confirmed by comparison of their retention times and mass spectra with those of reference standards also included in the analytical run. Fecal DNA Extraction and Bacterial 16S Ribosomal RNA Gene Sequencing Whole microbial genome DNA was extracted from fecal samples using PowerFecal DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA). Aliquots of extracted genomic DNA were quantified using Qubit dsDNA HS Assay Kit (Life Technologies, Waltham, MA). The DNA was amplified with Illumina adapter and indexed polymerase chain reaction primers. Bacterial 16S ribosomal RNA was sequenced using the Illumina MiSeq sequencing platform (Illumina, Inc., San Diego, CA) as previously described (14). Further details are given in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693). Bioinformatic and Statistical Analysis for Bacterial 16S Sequence Data Multivariate diversity analysis between patient and control samples was performed using PERmutational Multivariate ANalysis Of VAriance (PERMANOVA) using the Adonis function from the R package VEGAN (15).

addition to increased fibrinolysis caused by stress-mediated sympathoadrenergic catecholamine release, neutrophil elastase has been linked as trigger for fibrinolysis in trauma (25). However, to date, no comparable data are available on the contribution of neutrophil elastase to promote fibrinolysis in cardiac arrest. Some limitations need to be considered while interpreting the results. The current prospective observational study analyzed a strictly preselected cohort of OHCA patients including only those with presumed cardiac cause of cardiac arrest, who had achieved ROSC at admission, and were subjected to targeted temperature management. The possibly associated selection bias has to be considered, and study results need to be interpreted with appropriate caution. Furthermore, it has to be mentioned that there is currently no widely accepted “gold standard” assay available for detection of systemic fibrinolysis. A possible risk of bias relating to the lack of a reference standard must be taken into account. Future studies determining thrombelastometry thresholds for increased fibrinolysis in OHCA need to validate their findings against standardized methods for detection of fibrinolysis, which may become available in the future (37). Further investigations also need to confirm the predictive performance of a greater than or equal to 20% ML cutoff found in this study.

thresholds for increased fibrinolysis in OHCA need to validate their findings against standardized methods for detection of fibrinolysis, which may become available in the future (37). Further investigations also need to confirm the predictive performance of a greater than or equal to 20% ML cutoff found in this study. CONCLUSIONS Increased fibrinolysis at admission may be interpreted as cumulative surrogate marker for hypoperfusion and hypoxia, that is, the duration of no-flow time and resuscitation quality. The current study provides a predictive cutoff value for a readily available bedside marker with 100% positive predictive value for poor outcome, which could be of interest for both treating physicians and relatives. ACKNOWLEDGMENT We thank Gerhard Ruzicka, Karin Petroczi, and Christa Drucker for their valuable support with thrombelastometry and laboratory analysis. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported by the Austrian Science Fund FWF grant SFB 54/APF05404FW (Special Research Program-Cellular Mediators Linking Inflammation and Thrombosis, Medical University of Vienna). The authors have disclosed that they do not have any potential conflicts of interest.

Sepsis is a major health concern with increasing frequency and an estimated global mortality rate of 5.3 million deaths per year (1). Time is crucial in the management of septic patients and early treatment, including antibiotic administration and source control are the decisive first steps that influence patients’ outcome dramatically (2–4). Current guidelines recommend the initiation of antimicrobial therapy as early as possible and preferably within 1 hour (5). However, most of early treatments are empirical, and 46% of empirical antibiotic treatments were shown to be inappropriate and associated with 35% mortality (6). About 50% was either unnecessary or too broad spectrum, increasing the risk for resistance and toxicity. Early recognition of the infecting microorganism is therefore crucial for a targeted antimicrobial therapy, reducing side effects for the patient and improving patients’ outcome. The current standard of care blood culture (BC) however is limited by long time to positivity, low sensitivity, and low specificity. We recently published a proof of concept study, in which the general applicability of microbial circulating cell-free DNA (cfDNA) to diagnose causative pathogens in sepsis using a significance scoring system and to identify single-gene resistances via next-generation sequencing (NGS) was demonstrated (7).

Sepsis is a major health concern with increasing frequency and an estimated global mortality rate of 5.3 million deaths per year (1). Time is crucial in the management of septic patients and early treatment, including antibiotic administration and source control are the decisive first steps that influence patients’ outcome dramatically (2–4). Current guidelines recommend the initiation of antimicrobial therapy as early as possible and preferably within 1 hour (5). However, most of early treatments are empirical, and 46% of empirical antibiotic treatments were shown to be inappropriate and associated with 35% mortality (6). About 50% was either unnecessary or too broad spectrum, increasing the risk for resistance and toxicity. Early recognition of the infecting microorganism is therefore crucial for a targeted antimicrobial therapy, reducing side effects for the patient and improving patients’ outcome. The current standard of care blood culture (BC) however is limited by long time to positivity, low sensitivity, and low specificity. We recently published a proof of concept study, in which the general applicability of microbial circulating cell-free DNA (cfDNA) to diagnose causative pathogens in sepsis using a significance scoring system and to identify single-gene resistances via next-generation sequencing (NGS) was demonstrated (7). The aim of the presented study was to evaluate the performance of the NGS-based diagnostic approach with a larger cohort of patients and benchmark it by direct comparison to the current standard of care BC. Due to the limitations of BC, our findings were reviewed by an independent expert jury for plausibility, and the antimicrobial treatment regimen was reassessed for putative changes.

e NGS-based diagnostic approach with a larger cohort of patients and benchmark it by direct comparison to the current standard of care BC. Due to the limitations of BC, our findings were reviewed by an independent expert jury for plausibility, and the antimicrobial treatment regimen was reassessed for putative changes. METHODS Study Design Data result from a secondary analysis of septic patients (n = 50) participating in a previously published, prospective observational clinical study of our workgroup, which was conducted in the surgical ICU of Heidelberg University Hospital, Germany between November 2013 and January 2015 (German Clinical Trials Register: DRKS00005463) (8). The focus of this primary study was the immune response to fungal infections in patients suffering from septic shock, and three patients (S16, S25, and S35) were already described with special focus on fungal infections in detail, including NGS results as heat maps within this study. All study patients or their legal designees gave written informed consent. In total 50 patients suffering from septic shock according to the criteria of the Surviving Sepsis Campaign: International Guidelines for Management of Severe Sepsis and Septic Shock 2012 were enrolled in this study (9). Treatment of patients with septic shock included early-goal directed therapy (10), elimination of the septic focus, and broad-spectrum antibiotic therapy (10–12). Blood samples were collected at sepsis onset (T0) and 1 day (T1), 2 days (T2), 7 days (T3), 14 days (T4), 21 days (T5), and 28 days (T6) thereafter. Relevant baseline data (demographic data and primary site of infection), clinical data (disease severity scores such as Simplified Acute Physiology Score II, Sequential Organ Failure Assessment [SOFA] score, Acute Physiology and Chronic Health Evaluation II score, surgical procedures, antifungal therapy, and outcome variables) as well as routine infection variables (e.g., leukocytes, C-reactive protein [CRP], procalcitonin, and body temperature) were collected. In addition, 20 postoperative patients following major abdominal surgery without any evidence of infection were included as controls. Routine infection variables (e.g., leukocytes, CRP, procalcitonin, and body temperature), BCs, and other clinical microbiological specimens were without pathologic findings in this group. Plasma samples from the post-surgery group were collected prior to surgery (T0), immediately after the surgical procedure (T1), and 24 hours later (T2).

riables (e.g., leukocytes, CRP, procalcitonin, and body temperature), BCs, and other clinical microbiological specimens were without pathologic findings in this group. Plasma samples from the post-surgery group were collected prior to surgery (T0), immediately after the surgical procedure (T1), and 24 hours later (T2). Two septic patients as well as three patients of the post-surgery control group were retrospectively excluded from further analyses due to technical reasons, resulting in 48 septic patients and 17 post-surgery control patients. A workflow diagram to illustrate the study design and the NGS diagnostics workflow in context with clinical data is provided in Figure S1 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Additional details about patients, study time points, and samples are provided in Data S2 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Whole blood for BC and for plasma preparation for NGS testing was drawn on the same day (different EDTA tubes). Study and control patients or their legal designees signed written informed consent. All study procedures were approved by the local ethics committee (Ethics Committee of the Medical Faculty of Heidelberg, Trial Code No. S-097/2013).

r plasma preparation for NGS testing was drawn on the same day (different EDTA tubes). Study and control patients or their legal designees signed written informed consent. All study procedures were approved by the local ethics committee (Ethics Committee of the Medical Faculty of Heidelberg, Trial Code No. S-097/2013). Clinical Microbiology At Heidelberg University Hospital BC testing is routinely performed as described previously (13). Quantification of herpes simplex virus 1 DNA and cytomegalovirus DNA from plasma or tracheal secretion was carried out via quantitative real-time polymerase chain reaction as described previously (14). Cultivation of wound swabs, catheter, and stool samples was carried out as previously described (15, 16). Plasma Preparation and Nucleic Acid Isolation Plasma was prepared, and nucleic acid isolation was performed as described previously (8). If plasma volumes were below 1,000 μL after centrifugation, the respective samples were excluded (Data S2, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Contamination controls were prepared following the same procedure, and quality control steps were carried out as already described (8).

eviously (8). If plasma volumes were below 1,000 μL after centrifugation, the respective samples were excluded (Data S2, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Contamination controls were prepared following the same procedure, and quality control steps were carried out as already described (8). Preparation of NGS Libraries and Sequencing Library preparation and sequencing were carried out as previously described (7) from 1 ng cfDNA using the Nextera XT library preparation kit (Illumina, San Diego, CA) with a Biomek FXP liquid handling robot (Beckman Coulter, Brea, CA). The utilized raw data for NGS-based diagnostics are deposited in the European Nucleotide Archive under the following accession numbers: PRJEB21872 and PRJEB30958.

) from 1 ng cfDNA using the Nextera XT library preparation kit (Illumina, San Diego, CA) with a Biomek FXP liquid handling robot (Beckman Coulter, Brea, CA). The utilized raw data for NGS-based diagnostics are deposited in the European Nucleotide Archive under the following accession numbers: PRJEB21872 and PRJEB30958. Bioinformatics Bioinformatic processing and sepsis indicating quantifier (SIQ) score calculation were carried out as already described (7). Briefly, after bioinformatic removal of human sequences and taxonomic classification of nonhuman sequences, the SIQ score provides a quantitative and probabilistic assessment of every detected microbe in the respective sample based on a noninfected control group and permits a comparison between different samples, when identically processed. After normalization of read counts to the library size, a likelihood estimate of the probability to observe a certain species in the control group is generated. Under the assumption that all read counts for a certain species are Poisson distributed based on data generated from the control group, a p value that assesses the likelihood for read abundances outside this Poisson distribution is calculated. This p value along with a species-specific factor and the normalized read abundance gives rise to the individual SIQ score of a certain species in a patient sample. Control samples from elective surgery patients, which passed the quality control restrictions were added to our database to serve as the noninfected control group. The same criteria as previously described (8) were applied to exclude species hits (Fig. S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364) with the following exception: thresholds for bacterial, viral, and fungal hits were 10 normalized reads.

to our database to serve as the noninfected control group. The same criteria as previously described (8) were applied to exclude species hits (Fig. S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364) with the following exception: thresholds for bacterial, viral, and fungal hits were 10 normalized reads. Evaluation Clinical Expert Panel To assess the plausibility of the calculated SIQ scores, a panel of eight clinicians specialized in intensive care, clinical microbiology, and infectiology were asked to answer a questionnaire (Fig. S4, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). The participants were from Heidelberg University Hospital but had no role in the study. As most positive BCs were obtained at sepsis onset, only the first three time points (T0, T1, T2) were subjected to evaluation. Every case was introduced with the patient’s anamnesis, antibiotic therapy and all results from BC and NGS at T0, T1, and T2 as well as microbiological results from other specimens, if available. The concept of the SIQ score was explained to the clinical experts as well as the visualization of the results in form of heat maps, with highest scores in dark red and low scores without coloring. Color scaling was however only comparable within the results from one patient. The experts were asked to make an overall assessment of the plausibility of reported species for the individual time point, not for each species individually and note if they deemed the overall result plausible with respect to the patient history of underlying primary diseases, surgeries, complications, and microbiological results from other specimens. Majority rules were obtained as further described in the figure legend of Figure S4 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). How NGS results and clinical data were integrated is furthermore illustrated in Figure S1 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

other specimens. Majority rules were obtained as further described in the figure legend of Figure S4 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). How NGS results and clinical data were integrated is furthermore illustrated in Figure S1 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). RESULTS Participants and Study Design In total, 48 patients with septic shock were included in the investigation, whose characteristics are described in detail in Table 1. The primary septic focus was the abdomen (n = 43; 90%), followed by the lung (n = 4; 8%) as well as the genitourinary tract (n = 1; 2%). In 31 patients (65%), sepsis was due to postoperative peritonitis following surgery of the gastrointestinal tract (n = 36; 75%) or hepatobiliary surgery (n = 10; 21%). The overall 28-day as well as 90-day mortality was 19% (n = 9) and 32% (n = 15), respectively. The median length of ICU as well as hospital stay was 20 days and 47 days, respectively. The median SOFA score among sepsis patients was 11 at sepsis onset. Seventeen patients with elective major abdominal surgery served as uninfected controls (Table 1). TABLE 1. Patient and Control Group Characteristics

RESULTS Participants and Study Design In total, 48 patients with septic shock were included in the investigation, whose characteristics are described in detail in Table 1. The primary septic focus was the abdomen (n = 43; 90%), followed by the lung (n = 4; 8%) as well as the genitourinary tract (n = 1; 2%). In 31 patients (65%), sepsis was due to postoperative peritonitis following surgery of the gastrointestinal tract (n = 36; 75%) or hepatobiliary surgery (n = 10; 21%). The overall 28-day as well as 90-day mortality was 19% (n = 9) and 32% (n = 15), respectively. The median length of ICU as well as hospital stay was 20 days and 47 days, respectively. The median SOFA score among sepsis patients was 11 at sepsis onset. Seventeen patients with elective major abdominal surgery served as uninfected controls (Table 1). TABLE 1. Patient and Control Group Characteristics For the direct comparison of BC results with NGS, same day blood samples were acquired for each procedure for up to seven time points. In total, 256 blood samples were obtained from septic patients for NGS-based analysis and BC, respectively. Several samples had to be excluded as they failed to meet defined quality standards, resulting in 239 plasma samples of septic patients and 34 plasma samples of control patients for NGS. A detailed overview of patients, samples, and results is given in Supplementary Data S2 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Results of cfDNA quantifications are shown in Figure S5 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

34 plasma samples of control patients for NGS. A detailed overview of patients, samples, and results is given in Supplementary Data S2 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Results of cfDNA quantifications are shown in Figure S5 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364). NGS Diagnosis Yields More Positive Results Than BC At sepsis onset (T0), out of 48 patients with septic shock 33% (n = 16) had positive BCs. NGS of 44 plasma samples at sepsis onset revealed 72% (n = 32) positive samples with relevant pathogens identified (Fig. 1A). For the whole study period of 28 days, out of 256 BCs from 48 patients with septic shock, 11% (n = 29) were positive (Fig. 1A). In contrast, NGS of 239 septic patients’ cfDNA samples followed by SIQ score calculation yielded 169 positive results (71%) over the whole study period, which were similar for all time points (69–74%) except for T5 (52%) (Fig. 1A). As previously described (7, 17), approximately 98% of reads were of human origin and less than 1% of reads could be classified to microbes (Fig. S6, Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

over the whole study period, which were similar for all time points (69–74%) except for T5 (52%) (Fig. 1A). As previously described (7, 17), approximately 98% of reads were of human origin and less than 1% of reads could be classified to microbes (Fig. S6, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Figure 1. Positive results and species comparison from blood culture (BC) and sepsis indicating quantifier (SIQ) calculation. A, Percentage of positive results from BC (blue) and SIQ (orange) over the study time points. B, Percentage of species detected in the top 10 positive BC (blue) or SIQ (orange) results are displayed in decreasing order, sorted by BC results. The top two species are distributed similarly, with additional overlap in other species. A proportion of species is only observed in either one of the methods. The frequently observed potential skin contaminants Coagulase-negative Staphylococci (part of which are Staphylococcus epidermidis and S. hominis) as well as Propionibacteria are grouped at the bottom of the graph.

ditional overlap in other species. A proportion of species is only observed in either one of the methods. The frequently observed potential skin contaminants Coagulase-negative Staphylococci (part of which are Staphylococcus epidermidis and S. hominis) as well as Propionibacteria are grouped at the bottom of the graph. Out of 27 positive BCs with any NGS result, 48% (n = 13) identified typical skin commensals such as Coagulase-negative (CoN) Staphylococci or Propionibacteria. Three patients were found to suffer from fungemia, caused by Candida albicans or Candida glabrata (15%; n = 4). Predominant species in positive BCs were Escherichia coli (18%; n = 5) and Enterococcus faecium (15%; n = 4). In total, 13 unique species were identified by BC. Of these 13 species, two were fungi (C. albicans and C. glabrata) and 11 bacteria (73% gram-positive, 27% gram-negative). By application of bioinformatics filtering and stringent decision processes (Fig. S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364), we found 169 samples with positive SIQ scores. In these 169 samples, 438 organisms and 74 unique species were identified (Tables S1 and S2, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Most of them (n = 67) were bacteria and viruses (n = 4), but also archae (n = 2) and the fungus C. albicans (n = 1). Of these bacteria, 52% were gram-negative and 48% gram-positive.

es, 438 organisms and 74 unique species were identified (Tables S1 and S2, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Most of them (n = 67) were bacteria and viruses (n = 4), but also archae (n = 2) and the fungus C. albicans (n = 1). Of these bacteria, 52% were gram-negative and 48% gram-positive. Remarkably, the two species with the highest relative abundance in BC and NGS results were similar in prevalence (Fig. 1B), including E. coli (NGS: 12.6%; n = 55 and BC: 14.7%; n = 5) and E. faecium (NGS: 10.0%; n = 44 and BC: 11.8%; n = 4). Normalized read abundances of the most frequently identified species by BC and NGS show a tendency for higher abundances in patient samples in comparison to the control cohort (Fig. S7, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Of 17 specimens, which were positive in BC and NGS, 7 BCs revealed CoN Staphylococci, whereas different species were identified via NGS (Fig. S8A, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Of the remaining 10 positive specimens, the same species were identified by both methods in nine cases, with additional species identified either by NGS or BC in six instances, indicating a high concordance of positively identified species.

NGS (Fig. S8A, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Of the remaining 10 positive specimens, the same species were identified by both methods in nine cases, with additional species identified either by NGS or BC in six instances, indicating a high concordance of positively identified species. Ten specimens were solely positive by BC, of which 60% were CoN Staphylococci or Propionibacteria (Fig. S8A, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). In four cases from positive BCs with other species than potential skin contaminants, SIQ analysis was negative. In contrast, of 169 positive SIQ results, 152 were negative by BC (Fig. S8B, Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

8A, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). In four cases from positive BCs with other species than potential skin contaminants, SIQ analysis was negative. In contrast, of 169 positive SIQ results, 152 were negative by BC (Fig. S8B, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Confirmation of NGS-Based Diagnostic Results by an Independent Expert Evaluation The contingency table summarizes the number of positive and negative findings for BC and NGS (Fig. S8B, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Seventeen samples (7%) were excluded from direct comparison because these samples did not pass one of several quality filters for NGS. When taking the current gold standard BC as a reference, 17 samples were positive by BC and NGS, 10 were positive by BC only, which led to a sensitivity of 17/(17 + 10) = 62.96%. Exclusion of potential false-positive results by BC (Staphylococcus epidermidis, S. hominis, or Propionibacteria), results in 10 samples that were positive by BC and NGS and four that were positive by BC only, leading to a sensitivity of 10/(10 + 4) = 71.43%. Correspondingly, the calculated specificity was 28.30% (29.33%), positive predictive value 10.06% (5.92%), and negative predictive value 85.71% (94.29%).

or Propionibacteria), results in 10 samples that were positive by BC and NGS and four that were positive by BC only, leading to a sensitivity of 10/(10 + 4) = 71.43%. Correspondingly, the calculated specificity was 28.30% (29.33%), positive predictive value 10.06% (5.92%), and negative predictive value 85.71% (94.29%). Due to the well-known limitations of BC with respect to sensitivity and specificity, BC may not represent a reliable reference to evaluate the plausibility of the calculated SIQ scores, despite being the current gold standard. We therefore held an expert adjudication assessing NGS results for the first three time points (T0, T1, T2) with respect to their clinical plausibility in the context of the patient’s medical history (Fig. S1, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). For time points T0–T2 out of 121 specimens, 73% (n = 88) were SIQ positive and 27% (n = 33) were negative (Fig. 2A). Of 131 corresponding BCs, 13% were positive (n = 17) and 87% were negative (n = 114) (Fig. 2A). Following a majority rule, SIQ results were evaluated as plausible for 96% of SIQ+-samples. Plausibility for SIQ–-samples was 61% (Fig. 2B). All results of the expert evaluation are shown in Figure S9 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364), with the corresponding inter-rater agreement statistics, which were adequate for plausibility of the SIQ results (Fig. S10, Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

g. 2B). All results of the expert evaluation are shown in Figure S9 (Supplemental Digital Content 1, http://links.lww.com/CCM/E364), with the corresponding inter-rater agreement statistics, which were adequate for plausibility of the SIQ results (Fig. S10, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Figure 2. Expert evaluation of next-generation sequencing diagnosis regarding plausibility. A, Percentage of positive and negative results obtained for sepsis indicating quantifier (SIQ) and blood culture (BC) distributed over the first three time points (T0, T1, T2, varying shades of gray) or the average over T0–T2 (blue). B, Assessment of the plausibility by majority vote of the positive or negative SIQ score result over the first three time points (T0, T1, T2, varying shades of gray) or the average over T0–T2 (orange). This graph refers to question 1 in the questionnaire.

1, T2, varying shades of gray) or the average over T0–T2 (blue). B, Assessment of the plausibility by majority vote of the positive or negative SIQ score result over the first three time points (T0, T1, T2, varying shades of gray) or the average over T0–T2 (orange). This graph refers to question 1 in the questionnaire. Clinical Impact of NGS-Based Diagnostics The clinical implications of NGS-based diagnosis are exemplified for patient S3, whose detailed case description is given in the figure legend (Fig. S11, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). NGS was able to detect the major infecting pathogen E. faecium for patient S3 already at sepsis onset and all later time points, in contrast to BC. The clinical experts regarded the NGS-based results as plausible and voted for a retrospective change in the antimicrobial treatment, here a de-escalation of antibiotics. Based on NGS findings, the adjustment to monotherapy would have been possible already at sepsis onset. All heat maps for all patients with integrated NGS and clinical microbiology results are shown in the Supplemental Data File 2 (Supplemental Digital Content 2, http://links.lww.com/CCM/E365).

scalation of antibiotics. Based on NGS findings, the adjustment to monotherapy would have been possible already at sepsis onset. All heat maps for all patients with integrated NGS and clinical microbiology results are shown in the Supplemental Data File 2 (Supplemental Digital Content 2, http://links.lww.com/CCM/E365). Based on SIQ score analyses the experts would have changed the antibiotic regimen in 53% of all patients analyzed in this study (Fig. 3A), recommending a de-escalation of antimicrobial therapy in 80% and escalation (to include other antibiotics) in 40% (Fig. 3B; and Fig. S9, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Inter-rater agreement regarding therapy was only slight to fair (Fig. S10, Supplemental Digital Content 1, http://links.lww.com/CCM/E364), which is influenced by the large size of the expert jury and potentially multifaceted views on the nature of the therapy change. Another reason is that only for a subset of patients a therapy change was advised, but all patients were included in the calculation of the inter-rater agreement. According to the expert evaluation, patients were assigned into two groups: group 1, with a retrospectively recommended antimicrobial change based on NGS results (n = 24) and group 2, where the majority vote recommended a continuation of the antimicrobial therapy (n = 17). Remarkably, 28-day as well as 90-day mortality in group 1 were higher by 13% and 14%, respectively (Table S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364; and Fig. 3C). Furthermore, we observed a significantly increased consumption of antimicrobials during the 28-day observation period in group 1 compared with group 2 (30 vs 19 cumulated daily therapies of prescribed antimicrobials; Table S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Furthermore, the need for renal replacement therapies was also increased in group 1 (45.8% vs 17.6%; Table S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364).

oup 2 (30 vs 19 cumulated daily therapies of prescribed antimicrobials; Table S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Furthermore, the need for renal replacement therapies was also increased in group 1 (45.8% vs 17.6%; Table S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Figure 3. Expert evaluation of the impact of next-generation sequencing (NGS) results on the antimicrobial therapy (AT). A, Results from the majority vote regarding the hypothetical therapy intervention based on NGS results (question two of the questionnaire). B, Percentage of answers evaluated by majority vote regarding the nature of therapy intervention (question three of the questionnaire). C, Kaplan-Meier curve for patient’s survival concerning group 1 (inappropriate AT) versus group 2 (appropriate AT). The orange line indicates group 1, the black line indicates group 2. Patients’ survival is given as cumulative survival over a period of 90 d.

f therapy intervention (question three of the questionnaire). C, Kaplan-Meier curve for patient’s survival concerning group 1 (inappropriate AT) versus group 2 (appropriate AT). The orange line indicates group 1, the black line indicates group 2. Patients’ survival is given as cumulative survival over a period of 90 d. DISCUSSION The overall positivity rate of BC was relatively low in this study (11%), which is in line with literature (ranging from 10% [18] up to 30% [19]). The most prevalent microorganisms detected by BC were CoN Staphylococci and E. coli. CoN Staphylococci represent the most common bona fide BC contaminants (20) with a false positivity rate of around 80% (21, 22). Another frequent finding in BC were Propionibacteria, also rarely representing true pathogens in sepsis (23). The low sensitivity of BC coupled with the high rate of contaminants expectedly led to poor values for specificity and positive predictive value for NGS diagnostics, whereas sensitivity and negative predictive value were reasonable.

ng in BC were Propionibacteria, also rarely representing true pathogens in sepsis (23). The low sensitivity of BC coupled with the high rate of contaminants expectedly led to poor values for specificity and positive predictive value for NGS diagnostics, whereas sensitivity and negative predictive value were reasonable. The proportion of positive BCs was highest at sepsis onset (T0) and strongly decreased at later time points. In contrast, the rate of positive NGS results was constant over the different time points, which is in line with other DNA-based diagnostic tests (24). Our results suggest a higher sensitivity over BC and independence of antimicrobial treatment. In addition, NGS-based analyses frequently covered double-stranded DNA viruses (herpesviruses) and fungi. The accuracy, with which cfDNA reflects the dynamics of infection (7) is affirmed by the high concordance of identified species in specimens which were positive by BC and NGS and the relative distribution of the most abundant species (excluding potential contaminants) in separately positive BC and NGS samples. However, there were 10 instances, in which BC was positive while NGS was negative. Of these positive BCs, 60% revealed skin bacteria. Four of these were included by the expert evaluation, and two were assessed as plausibly negative, representing bona fide contaminants, whereas the other half was assessed as implausibly negative, together with the other four specimens positive by BC (Table S4, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). The BC positive samples contained species such as Enterobacter cloacae, E. coli, C. albicans, E. faecium, and S. aureus which are usually confident indicators of bacteremia or fungemia when isolated from blood (23). In case of E. cloacae, E. faecium, and E. coli (S04 T0, S22 T0, S53 T0), only few reads were counted for the respective species, and no SIQ score was calculated due to the stringent filter of less than 10 normalized read counts. Although the respective species were identified, however, the threshold settings might have been too stringent, as already described for fungi (8). Another confounding factor might be the acquisition of samples, which were acquired on the same day but not always at the same time. The lack of NGS-identified reads for S. aureus for patient S12 T0 and T1 despite two positive BCs remains unclear.

tings might have been too stringent, as already described for fungi (8). Another confounding factor might be the acquisition of samples, which were acquired on the same day but not always at the same time. The lack of NGS-identified reads for S. aureus for patient S12 T0 and T1 despite two positive BCs remains unclear. For this patient, a cecum perforation occurred as complication during surgery, which is in line with the spectrum of gut bacteria found by NGS. The expert panel’s vote regarding the NGS data for T0 was therefore plausible. The expert evaluation covered only the plausibility of NGS results, which does not imply implausibility of the BC results. Here, we can only speculate that positive S. aureus BCs for patient S12 T0 and T1 might be due to an S. aureus colonized central venous catheter (T0). Although only 2% of positive NGS results were regarded as implausible, for negative NGS results the proportion was 33%, indicating the unsatisfactory lack of a result in context of a clinically expected infection, with any diagnostic method.

For this patient, a cecum perforation occurred as complication during surgery, which is in line with the spectrum of gut bacteria found by NGS. The expert panel’s vote regarding the NGS data for T0 was therefore plausible. The expert evaluation covered only the plausibility of NGS results, which does not imply implausibility of the BC results. Here, we can only speculate that positive S. aureus BCs for patient S12 T0 and T1 might be due to an S. aureus colonized central venous catheter (T0). Although only 2% of positive NGS results were regarded as implausible, for negative NGS results the proportion was 33%, indicating the unsatisfactory lack of a result in context of a clinically expected infection, with any diagnostic method. Finally, the majority of samples did not yield a positive BC result but were positive by NGS. As indicated by the comparison of top 10 species between BC and NGS separately as well as the expert panel, most of these species seem to be pathogenic organisms which are not found in BC due to its limited sensitivity. However, a number of identified species are previously described contaminants (25), which becomes especially obvious in batch effects of different kits used in the sample preparation (Fig. S7, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). Although the most frequent contaminants were nonpathogenic environmental species, a certain overlap also of pathogenic species could be observed, for instance for E. cloacae or Salmonella enterica in the control cohort (Fig. S7, Supplemental Digital Content 1, http://links.lww.com/CCM/E364). As indicated in our decision tree (Fig. S3, Supplemental Digital Content 1, http://links.lww.com/CCM/E364), we aimed at not ruling out critical pathogens based on literature research, despite a certain overlap between described contaminants and pathogens. The occurrence of background signals in the control group stresses again the importance of a statistical qualifier such as the SIQ score to bioinformatically filter out contaminating sequences. Further data will lead to a continuous refinement of our databases and algorithms and render this platform even more precise.

The occurrence of background signals in the control group stresses again the importance of a statistical qualifier such as the SIQ score to bioinformatically filter out contaminating sequences. Further data will lead to a continuous refinement of our databases and algorithms and render this platform even more precise. Patient S3 illustrates the benefits of NGS-based diagnosis: 1) timely results, which were available since the earliest time point; 2) consistent results between NGS findings but also between NGS and other cultivation results; 3) stringent filtering and improved decision processes in combination with calibration to a control group led in this case to one significant species, while potential contaminants were mostly excluded; 4) the quantitative score, calculated individually for each pathogen allows for a direct comparison between different samples and is therefore excellently suited to monitor the time course of a patient; and 5) NGS results would have indicated an earlier change to a targeted monotherapy with vancomycin. As the patient recovered well, a reduction in the length of stay would have been a potential benefit of NGS diagnostics.

een different samples and is therefore excellently suited to monitor the time course of a patient; and 5) NGS results would have indicated an earlier change to a targeted monotherapy with vancomycin. As the patient recovered well, a reduction in the length of stay would have been a potential benefit of NGS diagnostics. There were some limitations of this study. The most important limitation for a direct comparison is sample acquisition. Although this was closely matched, a split sample for NGS and BC would have been ideal which was unfortunately not possible in the clinical setup of this study. The importance of a robust and diverse control group cannot be overestimated to effectively deal with laboratory, reagent, workflow, or database contaminants. The clinical case mixture of the control cohort might however influence the test characteristics intrinsic to the usage of the SIQ score. Therefore, it is highly advisable to only compare results generated with the same control-database. Since the nature of the SIQ score is based on a Poisson distribution parameterized by the control cohort, it becomes apparent that more controls included, and a more heterogenic composition of this control cohort, will be of great benefit to the diagnostic accuracy generated by the SIQ score method since the median/mean microbial cfDNA burden in healthy people will be more accurately captured/modeled (25). In this context, our control cohort was limited regarding the number and the population type (monocentric study, only abdominal surgery). Using our workflow, results can be achieved within 24 hours upon arrival of the patient’s sample in the laboratory. However, the 24-hour turnaround time was calculated by adding the individual processing steps within a research environment and has yet to withstand real-life conditions in clinical laboratories. Further developments in real-time sequencing might reduce time-to-diagnosis to only a few hours (26). Although this time frame is currently out of range for a first-line decision on antimicrobial treatment, it is certainly compatible for a reevaluation and possible de-escalation of broad-spectrum antibiotics.

ratories. Further developments in real-time sequencing might reduce time-to-diagnosis to only a few hours (26). Although this time frame is currently out of range for a first-line decision on antimicrobial treatment, it is certainly compatible for a reevaluation and possible de-escalation of broad-spectrum antibiotics. CONCLUSIONS Our main findings from this study are an over six-fold higher positivity rate of NGS over BC for serial samplings taken over the 28-day study period. The results of NGS diagnostics were assessed as plausible in 96% of positive SIQ results and would have led to a change in antimicrobial therapy in 53%. Despite the limited cohort size, we could observe remarkable trends in the two groups, which were retrospectively assessed in the expert evaluation as treated adequately or inadequately based on NGS results. In the adequately treated group, 28- and 90-day survival was higher and the overall use of antimicrobials was reduced, indicating the potential benefit of an adequate treatment for the patient, even without resistance identification. Therefore, we think that this method provides valuable data to support intensive care specialists in their treatment of patients with complex manifestations of sepsis.

ll use of antimicrobials was reduced, indicating the potential benefit of an adequate treatment for the patient, even without resistance identification. Therefore, we think that this method provides valuable data to support intensive care specialists in their treatment of patients with complex manifestations of sepsis. ACKNOWLEDGMENTS We thank all patients, referring physicians, and study nurses who submitted samples. We would like to acknowledge Armin Kalenka, MD, Marcel Hochreiter, MD, Alexandra Heininger, MD, Cornelius Busch, MD, Götz Hofmann, MD, Christoph Eisner, MD, Sascha Klemm, MD, and Stefan Zimmermann, MD, for their valuable participation in the expert evaluation. We also acknowledge Ute Krauser for excellent technical support. Furthermore, we acknowledge Annette Sigl and Kevin Tourelle for their aid to collect and assign patient samples. Supplementary Material Drs. Brenner and Sohn share senior authorship.

ACKNOWLEDGMENTS We thank all patients, referring physicians, and study nurses who submitted samples. We would like to acknowledge Armin Kalenka, MD, Marcel Hochreiter, MD, Alexandra Heininger, MD, Cornelius Busch, MD, Götz Hofmann, MD, Christoph Eisner, MD, Sascha Klemm, MD, and Stefan Zimmermann, MD, for their valuable participation in the expert evaluation. We also acknowledge Ute Krauser for excellent technical support. Furthermore, we acknowledge Annette Sigl and Kevin Tourelle for their aid to collect and assign patient samples. Supplementary Material Drs. Brenner and Sohn share senior authorship. Drs. S. Grumaz, Decker, Hofer, Brenner, and Sohn conceived of, designed, and supervised the study. Drs. Decker, Hofer, and Brenner collected clinical samples and clinical data. Drs. S. Grumaz, C. Grumaz, and Glanz performed all sample processing and next-generation sequencing experiments. Drs. Vainshtein and Stevens performed bioinformatic data processing and statistical analyses. Drs. S. Grumaz, C. Grumaz, Stevens, and Sohn analyzed the data. Drs. Hofer, Weigand, Brenner, and Sohn provided materials. Drs. S. Grumaz and C. Grumaz prepared the tables and figures. Drs. S. Grumaz and Sohn wrote the article with contributions from all other authors. All authors read, critically revised, and approved the final article. A data visualization tool associated with this article can be viewed here: https://lippincott.shinyapps.io/Sohn/.

Drs. S. Grumaz, Decker, Hofer, Brenner, and Sohn conceived of, designed, and supervised the study. Drs. Decker, Hofer, and Brenner collected clinical samples and clinical data. Drs. S. Grumaz, C. Grumaz, and Glanz performed all sample processing and next-generation sequencing experiments. Drs. Vainshtein and Stevens performed bioinformatic data processing and statistical analyses. Drs. S. Grumaz, C. Grumaz, Stevens, and Sohn analyzed the data. Drs. Hofer, Weigand, Brenner, and Sohn provided materials. Drs. S. Grumaz and C. Grumaz prepared the tables and figures. Drs. S. Grumaz and Sohn wrote the article with contributions from all other authors. All authors read, critically revised, and approved the final article. A data visualization tool associated with this article can be viewed here: https://lippincott.shinyapps.io/Sohn/. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by the Fraunhofer IGB (Stuttgart, Germany), Fraunhofer Future Foundation, Department of Anesthesiology (University of Heidelberg, Germany), Heidelberg Foundation of Surgery (University of Heidelberg, Germany). Our study sponsors were not involved in study design, collection, analysis, or interpretation of data.

n part, by the Fraunhofer IGB (Stuttgart, Germany), Fraunhofer Future Foundation, Department of Anesthesiology (University of Heidelberg, Germany), Heidelberg Foundation of Surgery (University of Heidelberg, Germany). Our study sponsors were not involved in study design, collection, analysis, or interpretation of data. Drs. Decker and Brenner received project funding from Heidelberg Foundation of Surgery. Drs. S. Grumaz, Stevens, and Sohn disclosed that they are co-founders of the company Noscendo active in molecular diagnostics for infectious diseases. The remaining authors have disclosed that they do not have any potential conflicts of interest.

There has been a surge of interest in the contribution of the gut microbiome to health and disease. Bacteria residing in the intestinal tract are, in health, compartmented from the host but have a symbiotic relationship, contributing to metabolic, endocrine, and immune functions. The composition and diversity of the intestinal microbiome in critical illness is likely to be impacted by poor intestinal perfusion, hypoxia, lack of enteral feeds, and antimicrobial therapy (1–4). This creates opportunities for the proliferation of potentially pathogenic species associated with adverse outcomes, including secondary infection and mortality (5–9). Profiling of the intestinal microbiome has generally been undertaken through sequencing of microbial DNA. However, more detailed functional information about the intestinal microbiome is possible through capture of metabolic outputs. 1H-nuclear magnetic resonance (1H-NMR) spectroscopy and mass spectrometry (MS) allow simultaneous detection of both human and microbial metabolites within biological samples. This may offer an insight into host-microbe interactions in the complex human system. Furthermore, such methods could provide a more cost-effective approach to understanding the clinical impact of intestinal dysbiosis in critical illness.

eous detection of both human and microbial metabolites within biological samples. This may offer an insight into host-microbe interactions in the complex human system. Furthermore, such methods could provide a more cost-effective approach to understanding the clinical impact of intestinal dysbiosis in critical illness. We previously identified significant plasma metabolic signatures of critical illness and organ failure in critically ill children (10). In the current study, we set out to characterize the functional capacity of the intestinal microbiome in critical illness, through multi-compartmental metabolic profiling. We examined changes in bacterial and host metabolites in urine and feces in order to characterize the gut microbiota-host relationship in severe illness. MATERIALS AND METHODS Study Population Critically ill children between 1 and 16 years old were consecutively enrolled at the time of admission to the PICU if they were mechanically ventilated and an admission urine sample was available within 48 hours of PICU admission. Children who were on chronic steroid or immune suppressant treatment were excluded. Data were obtained from the clinical health records. Healthy children, recruited from the local community, were eligible if they were well, had a normal healthy diet, and had not received antibiotics in the prior 3 months.

CU admission. Children who were on chronic steroid or immune suppressant treatment were excluded. Data were obtained from the clinical health records. Healthy children, recruited from the local community, were eligible if they were well, had a normal healthy diet, and had not received antibiotics in the prior 3 months. Approval for the study was granted by the East Midlands-Nottingham 2 Research Ethics Committee for recruitment from the PICUs at Cambridge University Hospitals NHS Foundation Trust, Great Ormond Street Hospital NHS Foundation Trust, and Imperial College Healthcare NHS Foundation Trust and from the City Road and Hampstead Research Ethics committee for recruitment of healthy children in the Cambridge vicinity. Parental informed consent was obtained prior to participation in the study. Samples Urine samples were collected via indwelling catheters as soon as possible after admission to PICU (timepoint 1). Further samples were obtained at day 3–5 and days 6–8 (timepoint 2 and timepoint 3, respectively) after admission. Urine samples from healthy children were collected directly into 20 mL universal containers and stored at –70°C until use. Fecal samples were collected from early (within first 2 d of PICU admission) and late (days 5–8 of PICU admission) in critically ill children, and a single sample obtained from healthy children. Samples were collected from nappies, placed in sterile plastic containers and stored at –70°C until use.

Samples Urine samples were collected via indwelling catheters as soon as possible after admission to PICU (timepoint 1). Further samples were obtained at day 3–5 and days 6–8 (timepoint 2 and timepoint 3, respectively) after admission. Urine samples from healthy children were collected directly into 20 mL universal containers and stored at –70°C until use. Fecal samples were collected from early (within first 2 d of PICU admission) and late (days 5–8 of PICU admission) in critically ill children, and a single sample obtained from healthy children. Samples were collected from nappies, placed in sterile plastic containers and stored at –70°C until use. Clinical Data Acquisition Disease severity was defined through collation of bedside physiologic data. In addition, routinely collected clinical data were recorded for multivariable analysis to identify associations with cardiovascular failure (the maximal inotrope score [11]), respiratory failure (days free of mechanical ventilation at 30 d), and critical illness duration (days free of PICU at 30 d). 1H-NMR Spectroscopy Fecal water and urine samples were prepared for 1H-NMR Spectroscopy according to published protocols (12). Acquired spectroscopic data were processed using the TopSpin 3.1 software package (Bruker Biospin, Rheinstetten, Germany). Data processing was undertaken using Matlab (Version 8.3.0.532 R2014a; Mathworks, Natick, MA). Further details are given in supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693).

ctroscopic data were processed using the TopSpin 3.1 software package (Bruker Biospin, Rheinstetten, Germany). Data processing was undertaken using Matlab (Version 8.3.0.532 R2014a; Mathworks, Natick, MA). Further details are given in supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693). Chemometric Analysis of Spectroscopic Data Processed spectroscopic data were imported to the SIMCA 13.0 software package (Umetrics AB, Umeå, Sweden) to conduct unsupervised multivariate statistical analysis. Principal components analysis was used to evaluate similarities/differences in urinary and fecal metabolite composition between groups. The R2 and Q2 variables provided an indication of goodness of fit (R2) as well as goodness of prediction (Q2) of the models. Supervised Orthogonal Projections to Latent Structures Discriminant Analysis models were calculated using one predictive and two orthogonal components. The models were assessed based on variance explained (R2Y) and predictive ability (Q2Y) metrics. Further details are given in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693). Bile Acid Quantification Bile acid data were acquired using an liquid chromatography-mass spectrometry quadrupole time-of-flight instrument in profiling mode, according to protocols described by Sarafian et al (13). The identity of the bile acids was confirmed by comparison of their retention times and mass spectra with those of reference standards also included in the analytical run.

Fecal DNA Extraction and Bacterial 16S Ribosomal RNA Gene Sequencing Whole microbial genome DNA was extracted from fecal samples using PowerFecal DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA). Aliquots of extracted genomic DNA were quantified using Qubit dsDNA HS Assay Kit (Life Technologies, Waltham, MA). The DNA was amplified with Illumina adapter and indexed polymerase chain reaction primers. Bacterial 16S ribosomal RNA was sequenced using the Illumina MiSeq sequencing platform (Illumina, Inc., San Diego, CA) as previously described (14). Further details are given in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693). Bioinformatic and Statistical Analysis for Bacterial 16S Sequence Data Multivariate diversity analysis between patient and control samples was performed using PERmutational Multivariate ANalysis Of VAriance (PERMANOVA) using the Adonis function from the R package VEGAN (15). Rank-based indirect gradient analysis “nonmetric multidimensional scaling (NMDS)” was used for the visualization of taxonomic differences between the different groups, using metMDS in R (15, 16). NMDS attempts to represent, as closely as possible, the pairwise dissimilarity between objects in a low-dimensional space. Further details are given in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693).

axonomic differences between the different groups, using metMDS in R (15, 16). NMDS attempts to represent, as closely as possible, the pairwise dissimilarity between objects in a low-dimensional space. Further details are given in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/E693). Statistical Analysis Multiple linear and logistic regression analyses were applied using the Statistical Package for the Social Sciences Version 21 (SPSS Statistics 22, IBM Corp., Armonk, NY). Categorical variables were analyzed using chi-square and Fisher exact tests as appropriate. Continuous variables were compared using two-sided t tests (for parametric variables) and the Mann-Whitney U test (nonparametric variables). A false discovery rate (FDR) was used to adjust for multiple metabolite testing. Metabolites with FDR less than 0.05 were considered significant. Multivariate beta diversity analysis between groups, including age, gender, and diagnostic category, was performed using R studio and the Statistical Package for the Social Sciences (SPSS Statistics 22, IBM Corp.). All presented p values were corrected for multiple comparisons using the Benjamini-Hochberg FDR method.

Statistical Analysis Multiple linear and logistic regression analyses were applied using the Statistical Package for the Social Sciences Version 21 (SPSS Statistics 22, IBM Corp., Armonk, NY). Categorical variables were analyzed using chi-square and Fisher exact tests as appropriate. Continuous variables were compared using two-sided t tests (for parametric variables) and the Mann-Whitney U test (nonparametric variables). A false discovery rate (FDR) was used to adjust for multiple metabolite testing. Metabolites with FDR less than 0.05 were considered significant. Multivariate beta diversity analysis between groups, including age, gender, and diagnostic category, was performed using R studio and the Statistical Package for the Social Sciences (SPSS Statistics 22, IBM Corp.). All presented p values were corrected for multiple comparisons using the Benjamini-Hochberg FDR method. RESULTS Participant Demographics The mean (sd) ages and weights were 70.28 months (52.8 mo) and 21.5 kg (13.5 kg) for patients; 73.5 months (43.7 mo) and 20.9 kg (10.1–41 kg) for controls. Four children died in the patient cohort (5.9%). All the critically ill children had received one or more broad-spectrum antibiotics at the time of sampling. Further clinical details and antibiotic exposure are given in Table 1.

1.5 kg (13.5 kg) for patients; 73.5 months (43.7 mo) and 20.9 kg (10.1–41 kg) for controls. Four children died in the patient cohort (5.9%). All the critically ill children had received one or more broad-spectrum antibiotics at the time of sampling. Further clinical details and antibiotic exposure are given in Table 1. Urinary Metabolic Profiling Demonstrates Loss of Intestinal Bacterial Metabolic Activity in Critical Illness We identified robust differences in the global 1H-NMR metabolic profiles of admission (timepoint 1) urine samples from critically ill compared with healthy children (Fig. 1A), indicating differences in both endogenous and intestinal microbiome-derived metabolites. Using supervised multivariate statistical analysis, we observed R2Y and Q2Y values of the generated model at 0.89 and 0.8, respectively (Fig. 1B).

point 1) urine samples from critically ill compared with healthy children (Fig. 1A), indicating differences in both endogenous and intestinal microbiome-derived metabolites. Using supervised multivariate statistical analysis, we observed R2Y and Q2Y values of the generated model at 0.89 and 0.8, respectively (Fig. 1B). Figure 1. Urinary and fecal 1H-nuclear magnetic resonance (1H-NMR) global metabolic profiles of critically ill and healthy children. A, Unsupervised principal components analysis (PCA) scores plot of admission urine samples from age-matched critically ill (red) and healthy (blue) children. R2 = 0.16, Q2 = 0.09. B, Supervised Orthogonal Projections to Latent Structures Discriminant Analysis (O-PLS-DA) loadings line plot. Urinary metabolites higher in critically ill children (up) compared with age-matched healthy children (down). The color bar indicates the correlation coefficient (R2) (i.e., the redder the peak, the higher the correlation). R2Y = 0.89, Q2Y = 0.80. C, Unsupervised PCA scores plot of fecal samples from age-matched critically ill (red, n = 27) and healthy control (blue, n = 41) samples. R2 = 0.23, Q2 = 0.08. D, Supervised O-PLS-DA loadings line plot. Fecal metabolites higher in critically ill children (up) compared with age-matched healthy children (down). R2Y = 0.96, Q2Y = 0.85. R2Y = variance explained, Q2Y= predictive ability. a.u = arbitrary units, N-AG = n-acetylglucosamine, NMNA = n-nethylnicotinamide.

23, Q2 = 0.08. D, Supervised O-PLS-DA loadings line plot. Fecal metabolites higher in critically ill children (up) compared with age-matched healthy children (down). R2Y = 0.96, Q2Y = 0.85. R2Y = variance explained, Q2Y= predictive ability. a.u = arbitrary units, N-AG = n-acetylglucosamine, NMNA = n-nethylnicotinamide. Urinary excretion of several mammalian-microbial co-metabolites were reduced in critically ill patient samples (Supplementary Table 1, Supplemental Digital Content 2, http://links.lww.com/CCM/E694). These included hippurate, 4-cresol sulphate, and formate. In addition, two bile acid peaks were associated with samples from the patient cohort, and a strong reduction in urinary citrate excretion observed in samples from critically ill compared with healthy children. We did not observe a strong separation in the global urine metabolic profile based on sample timing during the course of PICU stay (Supplementary Fig. 1, a and b, Supplemental Digital Content 3, http://links.lww.com/CCM/E695; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). There was no difference in global metabolic profiles based on gender, comorbidity status, duration or class of antibiotic exposure, and survival.

CU stay (Supplementary Fig. 1, a and b, Supplemental Digital Content 3, http://links.lww.com/CCM/E695; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). There was no difference in global metabolic profiles based on gender, comorbidity status, duration or class of antibiotic exposure, and survival. Untargeted Fecal Metabolic Profiling Indicates Loss of Bacterial Fermentation Activity in Critical Illness As with urinary metabolic profiles, the global 1H-NMR metabolic profiles of fecal water in critically ill compared with healthy children were strongly separated (Fig. 1C). Using supervised multivariate statistical analysis, we observed R2Y and Q2Y values of the generated model at 0.7 and 0.17, respectively (Fig. 1D). Metabolites showing discrimination between critically ill and healthy samples included short-chain fatty acid (SCFA) concentrations (among them butyrate, acetate, and propionate) and the nucleobase thymine which associated with healthy child sample profiles. Lactate and N-acetylglycoprotein associated with samples from critically ill children (Supplementary Table 2, Supplemental Digital Content 4, http://links.lww.com/CCM/E696). As with the urine, we did not identify differences based on gender, comorbidities, duration or class of antibiotic exposure, and survival status.

Untargeted Fecal Metabolic Profiling Indicates Loss of Bacterial Fermentation Activity in Critical Illness As with urinary metabolic profiles, the global 1H-NMR metabolic profiles of fecal water in critically ill compared with healthy children were strongly separated (Fig. 1C). Using supervised multivariate statistical analysis, we observed R2Y and Q2Y values of the generated model at 0.7 and 0.17, respectively (Fig. 1D). Metabolites showing discrimination between critically ill and healthy samples included short-chain fatty acid (SCFA) concentrations (among them butyrate, acetate, and propionate) and the nucleobase thymine which associated with healthy child sample profiles. Lactate and N-acetylglycoprotein associated with samples from critically ill children (Supplementary Table 2, Supplemental Digital Content 4, http://links.lww.com/CCM/E696). As with the urine, we did not identify differences based on gender, comorbidities, duration or class of antibiotic exposure, and survival status. Intestinal Bacterial Metabolism of Bile Acids Is Compromised During Critical Illness Based on the presence of increased urinary bile acid concentrations in patient urine samples, we went on to examine intestinal bile acid metabolism using targeted MS. We observed clear separation in overall fecal bile acid profile between critically ill and healthy children (Fig. 2A), with observed R2Y and Q2Y values of the generated model at 0.47 and 0.27, respectively (Supplementary Fig. 2, Supplemental Digital Content 5, http://links.lww.com/CCM/E697; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701).

al bile acid profile between critically ill and healthy children (Fig. 2A), with observed R2Y and Q2Y values of the generated model at 0.47 and 0.27, respectively (Supplementary Fig. 2, Supplemental Digital Content 5, http://links.lww.com/CCM/E697; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). Figure 2. Fecal liquid chromatography-mass spectrometry bile acid (BA) profiles of critically ill and healthy children. A, Unsupervised principal components analysis scores plot of critically ill (red) and healthy control (blue) samples. R2 = 0.47, Q2 = 0.16. B, Changes in the metabolism of BAs in critically ill (in red) compared with healthy children (in blue) illustrate accumulation of primary BAs and a reduction in lithocholic acid in critically ill children. 5β-CA-3β, 12a-diol = 5β-cholanic acid-3β, 12a-diol, FDR 2.952E-6, 23 nor 5β- CA-3α, 12a-diol = 23-nor-5b-cholanic acid-3a, 12a-diol, FDR 1.664E-7, 3KCA = 3 ketocholanic acid, FDR 3.505E-7, 3a-H-12 KLCA = 3a-hydroxy-12 ketolithocholic acid, FDR 4.174E-6, CA = cholic acid, FDR 2.142E-7, DCA = deoxycholic acid, FDR 0.004, FDR = false discovery rate, ILCA = isolithocholic acid, FDR 1.596E-10, LCA = lithocholic acid, FDR 6.157E-10, PC = principal component, TCA = taurocholic acid, FDR 0.001.

id, FDR 3.505E-7, 3a-H-12 KLCA = 3a-hydroxy-12 ketolithocholic acid, FDR 4.174E-6, CA = cholic acid, FDR 2.142E-7, DCA = deoxycholic acid, FDR 0.004, FDR = false discovery rate, ILCA = isolithocholic acid, FDR 1.596E-10, LCA = lithocholic acid, FDR 6.157E-10, PC = principal component, TCA = taurocholic acid, FDR 0.001. In samples from critically ill children, a statistical increase in primary bile acid concentrations (cholic and choledeoxycholic acids) was observed. A reduction in concentration of secondary bile acids dependent on commensal bacterial metabolism was observed, including deoxycholic and lithocholic acids along with other bile acid metabolites arising from bacterial dehydroxylation, epimerization, or oxidation (Fig. 2B). Loss of Intestinal Metabolic Capacity Is Associated With Intestinal Dysbiosis in the Critically Ill Child Using the Shannon alpha diversity index as an indicator of fecal microbial diversity, we observed mean (sd) alpha diversity at genus level of 2.82 (0.34) compared with 2.12 (0.87) in samples from healthy versus critically ill children, respectively (p < 0.0001).

ted With Intestinal Dysbiosis in the Critically Ill Child Using the Shannon alpha diversity index as an indicator of fecal microbial diversity, we observed mean (sd) alpha diversity at genus level of 2.82 (0.34) compared with 2.12 (0.87) in samples from healthy versus critically ill children, respectively (p < 0.0001). NMDS demonstrated greater inter-individual variability in patient 16S profiles compared with those of healthy children (Fig. 3). Hierarchical analysis demonstrated that healthy children had higher prevalence of Bacteroides, Faecalibacterium, and Ruminococcus genera. Samples from critically ill children showed increased presence of noncommensals (e.g., Enterococcus and Streptococcus), and of normally low-prevalence microbial genera (Supplementary Fig. 3, Supplemental Digital Content 6, http://links.lww.com/CCM/E698; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). Figure 3. Bacterial composition in age-matched critically ill and healthy children. Nonmetric multidimensional scaling (NMDS) plot of fecal microbial composition at genus level. High variability is seen in samples from critically ill children (red) compared with healthy controls (orange). PERmutational Multivariate ANalysis Of Variance (PERMANOVA) p = 0.0001, F = 3.314. Samples from healthy children were tightly clustered together, suggesting greater similarity in composition, while those from patients were scattered widely across the plot and statistically separated from the healthy profiles (PERMANOVA test: F = 9.78; p < 0.001).

Nalysis Of Variance (PERMANOVA) p = 0.0001, F = 3.314. Samples from healthy children were tightly clustered together, suggesting greater similarity in composition, while those from patients were scattered widely across the plot and statistically separated from the healthy profiles (PERMANOVA test: F = 9.78; p < 0.001). We did not observe any statistical difference in the overall fecal microbial profile based on specific antibiotic class exposure or the days of antibiotic exposure prior to sampling. However, that the proportional abundance of the Enterococcus genus in patient samples directly correlated with the number of antibiotic classes administered prior to sample collection (r = 0.43; p = 3 × 10–6). Patterns of Change in Composition and Function of the Intestinal Microbiome in Critical Illness We undertook integrated analysis of fecal microbial 16S, and metabolic profiles using regression analysis and unsupervised clustering. Metabolites associated with samples from healthy children included acetate, propionate, and butyrate. An inverse correlation was noted of fecal SCFA levels with proportional abundance of Enterococcus, Bifidobacteria, Escherichia-Shigella, and Staphylococcus genera (Supplementary Table 1, Supplemental Digital Content 2, http://links.lww.com/CCM/E694). Linear regression demonstrated that abundance of Faecalibacterium, Fusicatenibacter, and Phascolarctobacterium were most predictive of fecal butyrate levels (R2 = 0.32; F = 9.44; p < 0.0001).

chia-Shigella, and Staphylococcus genera (Supplementary Table 1, Supplemental Digital Content 2, http://links.lww.com/CCM/E694). Linear regression demonstrated that abundance of Faecalibacterium, Fusicatenibacter, and Phascolarctobacterium were most predictive of fecal butyrate levels (R2 = 0.32; F = 9.44; p < 0.0001). We observed that intermediate metabolites of the citric acid cycle including fecal succinate, lactic acid, oxaloacetic acid, pyruvic, and acetoacetic acid were associated with the 16S profiles of critically ill compared with healthy children (Supplementary Fig. 4a, Supplemental Digital Content 7, http://links.lww.com/CCM/E699; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). We examined the association between fecal microbial composition and bile acid secretion in the patient cohort. We noted an inverse correlation between alpha diversity with primary bile acids including taurocholic acid (r = –0.44; p = 0.023) and taurohyocholic acid (r = –0.55; p = 0.003), and a direct correlation with secondary bile acids lithocholic acid (r = –0.48; p = 0.011) and isolithocholic acid (r = –0.6; p = 0.0001). Bacterial genera including Bacteroides, Ruminococcus, Eubacterium, Lachnospiraceae, and Faecalibacterium were directly correlated with levels of secondary bile acids including deoxycholic, lithocholic, and isolithocholic acid. Conversely, these bile acids were negatively correlated with abundance of Enterococcus and Staphylococcus (Supplementary Table 3, Supplemental Digital Content 8, http://links.lww.com/CCM/E700).

rium were directly correlated with levels of secondary bile acids including deoxycholic, lithocholic, and isolithocholic acid. Conversely, these bile acids were negatively correlated with abundance of Enterococcus and Staphylococcus (Supplementary Table 3, Supplemental Digital Content 8, http://links.lww.com/CCM/E700). In a linear regression model, we observed that proportional abundance of Faecalibacterium, Coprococcus, Ruminococcus, Catabacter, Enterococcus, Oscillibacter, and Pseudobutyrivibrio genera were predictive of fecal lithocholic acid concentration (R2 = 0.27; F = 2.34; p = 0.032). Metabolic Markers of Disease Severity in Critical Illness The peak integral values for fecal butyrate in the early admission samples of critically ill children correlated directly with days free of intensive care at 30 days (r = 0.38; p = 0.03). Peak integral values for urinary formate were inversely correlated with vasopressor requirement as measured by the inotrope score (17) (r = –0.2; p = 0.037). Urinary citrate directly correlated with inotrope score in patient samples (r = –0.27; p = 0.004). TABLE 1. Clinical Details of Study Participants DISCUSSION Although the adverse effects of critical illness on the composition of the intestinal microbiome in adults and children have been reported (5, 6, 9), the functional consequences of this on the developing pediatric microbiome have not been well described. We identified several changes in intestinal microbial activity during critical illness, including the fermentation of terminal carbohydrate metabolites into SCFAs and bile acid metabolism.

e been reported (5, 6, 9), the functional consequences of this on the developing pediatric microbiome have not been well described. We identified several changes in intestinal microbial activity during critical illness, including the fermentation of terminal carbohydrate metabolites into SCFAs and bile acid metabolism. Undigested carbohydrate and protein are key substrates for fermentation by colonic microbiota, and the resulting metabolites include SCFAs, branched-chain fatty acids, ammonia, amines, phenolic compounds, and gases including hydrogen, methane, and hydrogen sulphide. A healthy trophic network of SCFAs is maintained by a diverse number of commensal species in the gut (18, 19). The three main SCFAs, acetate, propionate, and butyrate are used in a number of host metabolic processes including signaling effector roles to modulate distal organ function (20), and in maintaining the health and barrier function of the colonic mucosa (21). Sustained reduction in fecal levels of SCFAs compared with healthy controls has been previously documented in critically ill adults (3). Commensal microbes regulate bile acid metabolism in the lower intestine (22) and are involved in deconjugation and dehydroxylation of bile acids released by the gall bladder into the intestinal lumen (23). Ours is the first study to demonstrate dysregulation of fecal bile acid metabolism in pediatric critical illness. Our data fits with previous observations of reduced luminal SCFA and secondary bile acid concentrations in murine microbiome depletion models (24, 25).

ed by the gall bladder into the intestinal lumen (23). Ours is the first study to demonstrate dysregulation of fecal bile acid metabolism in pediatric critical illness. Our data fits with previous observations of reduced luminal SCFA and secondary bile acid concentrations in murine microbiome depletion models (24, 25). The patient microbiome was characterized by over-representation of opportunistic pathogenic species and with species typically associated with the small intestine or with the use of antibiotics in critical illness including (Enterococcus) (6, 26) or with diarrhea (Streptococcus) (27). We were surprised not to see any microbial or metabolic patterns of change over the course of PICU admission, given the dynamic physiology of critical illness. However, it is likely that the ongoing exposure to systemic antibiotics is likely to influence the rate of recovery of intestinal microbial activity in this population. In critically ill adults, the fecal microbiome appears to continue to show significant dysbiosis at the time of ICU discharge (29). Our patient samples were collected at or within the first few days of PICU admission, so there was little variance in duration of antibiotic exposure. Broad-spectrum antibiotics are frequently administered to treat suspected infection in children and adults with critical illness (6, 28). Often it is done for life-saving reasons. The fact that all patients in our study received broad-spectrum antibiotics is a limitation, and future studies should include samples from nonantibiotic exposed children.

frequently administered to treat suspected infection in children and adults with critical illness (6, 28). Often it is done for life-saving reasons. The fact that all patients in our study received broad-spectrum antibiotics is a limitation, and future studies should include samples from nonantibiotic exposed children. Our finding that the number of antibiotic classes administered was associated with proportional abundance of the Enterococcus genus is in keeping with studies in critically ill adults (30). The most comparable study in critically ill children (6) did not explore the issue of antibiotic class in detail, although as in our study, administration of antibiotics was widespread (89% of a population where 94% were mechanically ventilated). We were surprised not to find a differentiation in 16S or metabolic profiles between patients with and without preexisting comorbidity. It suggests that the combination of adverse exposures in critical illness is greater than the impact of preexisting illness. Although there have been studies of the microbiome in patients with a protracted course of critical illness (26), we did not identify any studies that compared ICU admission microbiome profiles in previously well individuals compared with those living with life-long conditions.

r than the impact of preexisting illness. Although there have been studies of the microbiome in patients with a protracted course of critical illness (26), we did not identify any studies that compared ICU admission microbiome profiles in previously well individuals compared with those living with life-long conditions. We saw no significant impact of gender on the composition of the microbiome in our patient cohort, which was not unexpected given that the majority of children in the study were pre-pubertal. Studies in healthy neonates have not demonstrated gender-based differences in the neonatal urinary metabolome (31), and in older healthy children, population-specific factors (age, sex, body mass index, ethnicity, dietary, and country of origin) appear to be better captured in serum than in urine metabolic profiles (32). We did not find any published data of gender-based differences in critically ill children or adults. In a cohort study of 43 newborns over 2 years, the most significant influences on fecal microbiome maturation were birth mode, antibiotic exposure, and diet (33). Our integrated analysis reflects the accumulation of intermediate fermentation metabolites in critical illness as a result of intestinal dysbiosis (Supplementary Fig. 4b, Supplemental Digital Content 7, http://links.lww.com/CCM/E699; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). The gross decline of Bacteroides, Faecalibacterium, Roseburia, and Prevotella and their associations with the production of acetate, butyrate, and propionate substantiate this.

Fig. 4b, Supplemental Digital Content 7, http://links.lww.com/CCM/E699; legend, Supplemental Digital Content 9, http://links.lww.com/CCM/E701). The gross decline of Bacteroides, Faecalibacterium, Roseburia, and Prevotella and their associations with the production of acetate, butyrate, and propionate substantiate this. We did not observe any microbial or metabolic signatures of nonsurvival, although our study was not powered for this outcome. In adults, pathogen colonization of the fecal microbiome has been shown to be associated with death (9). Fecal SCFA measurement has been shown to be an effective tool to monitor recovery of gut microbiome in neonatal and murine probiotic supplementation trials (34, 35). The ability to assess the intestinal microbiome using urine metabolic profiling is advantageous in this patient population, since fecal samples may not be available acutely. The methodology to undertake the metabolite assays is scaleable and could be used to monitor gut health during and after recovery from critical illness. We have demonstrated that profiling of bacterial metabolites offers an insight into the functional capacity of the intestinal microbiome. A reduced abundance of these metabolites is linked to clinical disease severity.

Fecal SCFA measurement has been shown to be an effective tool to monitor recovery of gut microbiome in neonatal and murine probiotic supplementation trials (34, 35). The ability to assess the intestinal microbiome using urine metabolic profiling is advantageous in this patient population, since fecal samples may not be available acutely. The methodology to undertake the metabolite assays is scaleable and could be used to monitor gut health during and after recovery from critical illness. We have demonstrated that profiling of bacterial metabolites offers an insight into the functional capacity of the intestinal microbiome. A reduced abundance of these metabolites is linked to clinical disease severity. Dietary and microbiome-based therapies are being explored for the potential to support recovery of healthy gut commensal populations during and after critical illness. The panel of bacterial metabolites we have identified could be used to stratify and monitor such interventions. Beyond critical illness, the methodology is applicable to other disorders such as inflammatory bowel disease or severe malnutrition. The technology is scalable and adapting it to a clinically relevant format is feasible.

bacterial metabolites we have identified could be used to stratify and monitor such interventions. Beyond critical illness, the methodology is applicable to other disorders such as inflammatory bowel disease or severe malnutrition. The technology is scalable and adapting it to a clinically relevant format is feasible. ACKNOWLEDGMENTS We acknowledge the support of the Imperial College Clinical Phenotyping Centre, a core facility of the National Institute for Health Research Imperial Biomedical Research Centre’s Institute of Translational Medicine and Therapeutics. We would like to thank the children and families participating in the study, along with the clinicians treating them. We also thank the core informatics, sequencing, and pathogen informatics teams at the Wellcome Trust Sanger Institute. Supplementary Material Drs. Wijeyesekera and Wagner contributed equally.

ACKNOWLEDGMENTS We acknowledge the support of the Imperial College Clinical Phenotyping Centre, a core facility of the National Institute for Health Research Imperial Biomedical Research Centre’s Institute of Translational Medicine and Therapeutics. We would like to thank the children and families participating in the study, along with the clinicians treating them. We also thank the core informatics, sequencing, and pathogen informatics teams at the Wellcome Trust Sanger Institute. Supplementary Material Drs. Wijeyesekera and Wagner contributed equally. Dr. Wijeyesekera developed and supervised the metabolic profiling strategy, undertook data analysis, and wrote the article. Dr. Wagner developed and supervised the microbial profiling strategy, undertook data analysis, and wrote the article. Dr. De Goffau analyzed the microbial data and co-wrote the article. Ms. Thurston undertook sample processing and data analysis. Drs. Rodrigues Sabino and Zaher, Ms. White, Ms. Ridout, and Dr. Valla undertook sample processing, data collection, and analysis. Dr. Meyer undertook data analysis. Drs. Peters, Branco, Torok, Meyer, and Klein contributed to protocol development, supervised data analysis, and co-wrote the article. Dr. Parkhill developed the microbial profiling protocol, supervised all aspects of the microbial data analysis, and co-wrote the article. Drs. Frost and Holmes developed the metabolic profiling protocol, supervised all aspects of the metabolic data analysis, and co-wrote the article. Dr. Pathan conceived and supervised the study and wrote the article.

ial profiling protocol, supervised all aspects of the microbial data analysis, and co-wrote the article. Drs. Frost and Holmes developed the metabolic profiling protocol, supervised all aspects of the metabolic data analysis, and co-wrote the article. Dr. Pathan conceived and supervised the study and wrote the article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Aspects of the work were funded by an Imperial College Biomedical Research Centre award (to Drs. Holmes and Pathan), the Evelyn Trust (to Drs. Parkhill and Pathan), a Wellcome Trust Core Informatics Award (to Dr. Parkhill), Great Ormond Street Hospital Children’s Charity (to Drs. Peters and Ramnarayan), and a Levi-Montalcini award from the European Society of Intensive Care Medicine (to Dr. Pathan). The research was supported by the National Institute for Health Research Biomedical Research Centres based at Cambridge University Hospitals National Health Service (NHS) Foundation Trust, Great Ormond Street Hospital NHS Foundation Trust, Imperial College Healthcare NHS Trust, and Imperial College London.

Dr. Pathan). The research was supported by the National Institute for Health Research Biomedical Research Centres based at Cambridge University Hospitals National Health Service (NHS) Foundation Trust, Great Ormond Street Hospital NHS Foundation Trust, Imperial College Healthcare NHS Trust, and Imperial College London. Dr. Rodrigues Sabino’s institution received funding from National Institute for Health Research Imperial Biomedical Research Centre Institute of Translational Medicine and Therapeutics Call for Experimental Medicine Proposals. Dr. Valla received funding from Baxter and Nutricia. Dr. Meyer received funding from academic lectures for Danone, Nestle and Mead Johnson, and from the Mead Johnson Allergy Advisory Board. Dr. Frost’s institution received funding from Nestle and Heptares, and he received support for article research from Research Councils UK and Bill & Melinda Gates Foundation. Drs. Frost, Parkhill, and Pathan received support for article research from Wellcome Trust/Charity Open Access Fund. Dr. Parkhill’s institution received funding from Wellcome Trust, and he received funding from Next Gen Diagnostics Llc. Dr. Pathan’s institution received funding from European Society of Intensive Care Medicine, Evelyn Trust, and Wellcome Trust. The remaining authors have disclosed that they do not have any potential conflicts of interest.

Prolonged admissions to an ICU have an associated high resource utilization and personal cost to the patient. It is estimated that 23–45% of ICU bed days are occupied by long-stay patients (1). In a study of admissions to Canadian ICUs, those lasting more than 14 days constituted only 7.3% of admissions but accounted for 43.5% of ICU bed days (2). There is no consensus on what constitutes a prolonged intensive care admission and a range of definitions have been used. Previous descriptions of prolonged length of stay have ranged from 2 (3) to greater than 21 days (4). Previous studies have reported an association between total ICU length of stay and ICU and hospital mortality (5, 6). As a result, prolonged care on an ICU is sometimes viewed as an indicator of a slow response to treatment and hence poor prognosis. However, on any day of an ICU admission, there is uncertainty about the influence of the preceding length of stay on a patient’s ultimate outcome. Conditional survival is defined as the probability of future survival, given that the patient has already survived for a certain period of time (7, 8). This has previously been used in studies of patients with cancer but not in those treated on an ICU. Applying these methods to patients treated in an ICU may provide useful information about how prognosis evolves over time, which can be factored into decision making about continuing therapy.

tain period of time (7, 8). This has previously been used in studies of patients with cancer but not in those treated on an ICU. Applying these methods to patients treated in an ICU may provide useful information about how prognosis evolves over time, which can be factored into decision making about continuing therapy. As progressively older and more comorbid patients are admitted to ICUs (9), it is important to understand how their response to a prolonged ICU admission differs from a younger cohort. For example, older and frail patients may lose muscle mass and power with increasing length of stay, while remaining exposed to the other risks of treatment on an ICU, reducing the probability of future survival. Conversely, they may take longer to respond to conventional therapy and longer to recover after their initial insult. Despite accounting for a large proportion of bed days, the actual proportion of patients admitted to an ICU for an extended period is small (2). Large clinical databases therefore provide a cost-effective method of assessing conditional survival with sufficient power to analyze this cohort of patients. Using three unique intensive care databases, we compute conditional survival with increasing duration of ICU admission first in unselected patients and then in a subgroup analysis of the population dichotomized by age.

ve method of assessing conditional survival with sufficient power to analyze this cohort of patients. Using three unique intensive care databases, we compute conditional survival with increasing duration of ICU admission first in unselected patients and then in a subgroup analysis of the population dichotomized by age. We hypothesized that the probability of future survival may change as duration of ICU admission increases and that this will not necessarily be a linear decrease in survival. Our primary objective is to assess probability of future survival with increasing duration of ICU admission.

ve method of assessing conditional survival with sufficient power to analyze this cohort of patients. Using three unique intensive care databases, we compute conditional survival with increasing duration of ICU admission first in unselected patients and then in a subgroup analysis of the population dichotomized by age. We hypothesized that the probability of future survival may change as duration of ICU admission increases and that this will not necessarily be a linear decrease in survival. Our primary objective is to assess probability of future survival with increasing duration of ICU admission. MATERIALS AND METHODS Data Source We performed an observational study using three databases. The Post Intensive Care Risk-Adjusted Alerting and Monitoring (PICRAM) database contain details of all admissions between 2008 and 2016 to general ICUs in two English hospitals, a tertiary referral center and district general hospital. The Medical Information Mart for Intensive Care III (MIMIC-III) is an open-source clinical database, developed and maintained by Massachusetts Institute of Technology (MIT), Philips Healthcare, and Beth Israel Deaconess Medical Center (BIDMC) in the United States. Patients included in this database were admitted to one of the ICUs in BIDMIC between 2001 and 2012. The electronic ICU (eICU) collaborative database is composed of patients admitted to an ICU in the Philips telehealth program in the years 2014 and 2015. Two-hundred eight hospitals across continental United States contribute data to the database. The use of three independent databases allows us to limit bias that may exist within one dataset and further to demonstrate whether findings are consistent between the United Kingdom and North America, where different healthcare systems are in place and ICUs admit a different case-mix. Further, using MIMIC in addition to PICRAM extends the period of the study to 2001–2016.

it bias that may exist within one dataset and further to demonstrate whether findings are consistent between the United Kingdom and North America, where different healthcare systems are in place and ICUs admit a different case-mix. Further, using MIMIC in addition to PICRAM extends the period of the study to 2001–2016. Patient Population We included patients 18 years old or older at ICU admission with a record of vital status at hospital discharge. We only included data recorded during their first admission to ICU. We extracted age, sex, length of stay in an ICU, and survival to hospital discharge. Cohorts were dichotomized on age less than 75 or greater than or equal to 75 years old for an a priori planned subgroup analysis as 75 years has previously been used as cut off for “elderly” patients (10). The Oxford Acute Severity of Illness Score was used to describe severity of illness. This score is composed of 10 variables and has been demonstrated to have comparable predictive accuracy with other severity of illness score. It has previously been reported in Oxford data, MIMIC-III, and the eICU database (11). The primary outcome of interest was survival to hospital discharge. Patients with missing data were excluded from the analysis. A flow chart of inclusion/exclusion criteria is included in Supplementary Figure 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/F27).

Patient Population We included patients 18 years old or older at ICU admission with a record of vital status at hospital discharge. We only included data recorded during their first admission to ICU. We extracted age, sex, length of stay in an ICU, and survival to hospital discharge. Cohorts were dichotomized on age less than 75 or greater than or equal to 75 years old for an a priori planned subgroup analysis as 75 years has previously been used as cut off for “elderly” patients (10). The Oxford Acute Severity of Illness Score was used to describe severity of illness. This score is composed of 10 variables and has been demonstrated to have comparable predictive accuracy with other severity of illness score. It has previously been reported in Oxford data, MIMIC-III, and the eICU database (11). The primary outcome of interest was survival to hospital discharge. Patients with missing data were excluded from the analysis. A flow chart of inclusion/exclusion criteria is included in Supplementary Figure 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/F27). Statistical Analysis Simple descriptive statistics were calculated for each database and reported as number and percentage for binary variables and median and interquartile range (IQR) for continuous variables. To assess the main objective we calculated conditional survival, which we define as the proportion of patients surviving to hospital discharge given they have already survived a certain period of time on the ICU. We calculated this for each day of admission from zero until the day where fewer than 50 patients reached that length of stay (due to increasingly large CI widths after this day). The result at day zero is equivalent to the average hospital survival probability for all ICU admissions. We were able to use simple binomial methods rather than survival methods since discharge status was known for all patients, and therefore, no censoring occurred. We plotted survival to discharge against increasing length of stay and visually inspected for changes in the probability of future survival. We repeated this analysis with age subgroups (< 75 and ≥ 75 yr old).

than survival methods since discharge status was known for all patients, and therefore, no censoring occurred. We plotted survival to discharge against increasing length of stay and visually inspected for changes in the probability of future survival. We repeated this analysis with age subgroups (< 75 and ≥ 75 yr old). We generated binomial CIs for conditional survival using the method by Agresti and Coull (12). Graphs were smoothed using local estimated scatterplot smooth (LOESS) with a span of 0.5. LOESS is a nonparametric smoothing procedure using a locally weighted least squares method to correct for influence of outliers and obtain a robust estimate of the trends. All statistical analysis was performed in R Core v3.4.4 (R: A Language and Environment for Statistical Computing, Vienna, Austria). Additional packages used were dplyr (version 0.8.1; https://cran.r-project.org/package=dplyr) and binom (version 1.1-1; https://cran.r-project.org/package=binom) This study was conducted following STROBE guidelines and a checklist of recommendations and evidence is included in the supplementary table (Supplemental Digital Content 2, http://links.lww.com/CCM/F28) (16).

We generated binomial CIs for conditional survival using the method by Agresti and Coull (12). Graphs were smoothed using local estimated scatterplot smooth (LOESS) with a span of 0.5. LOESS is a nonparametric smoothing procedure using a locally weighted least squares method to correct for influence of outliers and obtain a robust estimate of the trends. All statistical analysis was performed in R Core v3.4.4 (R: A Language and Environment for Statistical Computing, Vienna, Austria). Additional packages used were dplyr (version 0.8.1; https://cran.r-project.org/package=dplyr) and binom (version 1.1-1; https://cran.r-project.org/package=binom) This study was conducted following STROBE guidelines and a checklist of recommendations and evidence is included in the supplementary table (Supplemental Digital Content 2, http://links.lww.com/CCM/F28) (16). Ethical Approval PICRAM 1 was approved by the NRES Committee Oxford C on December 28, 2011 (ref: 11/SC/0440) and by the National Information Governance Board on February 2, 2012 (ref: ECC 7-05(f)/2011). The MIMIC-III and eICU databases have received ethical approval from the Institutional Review Boards (IRBs) at BIDMC and MIT and because the database does not contain protected health information, a waiver for the requirement for informed consent was included in the IRB approval.

bruary 2, 2012 (ref: ECC 7-05(f)/2011). The MIMIC-III and eICU databases have received ethical approval from the Institutional Review Boards (IRBs) at BIDMC and MIT and because the database does not contain protected health information, a waiver for the requirement for informed consent was included in the IRB approval. RESULTS PICRAM Database Eleven-thousand six-hundred forty-eight index admissions were included in the study after 524 were excluded for missing data (4.3%). Median age was 64 years (IQR, 50–74 yr), 58.7% were male, median length of stay was 1.9 days (IQR, 1.0–4.3 d). Hospital mortality was 18.5% (81.5% survival to hospital discharge) (Table 1). Ninety patients were admitted for longer than 40 days. Conditional survival declined over the initial 5–10 days but then plateaued prior to day 10 (Fig. 1). Conditional survival with a length of stay greater than 9 days declined to 74.2% (CI, 71.7–77.3%), at which point changes in conditional survival plateau with increasing length of stay. TABLE 1. Descriptive Statistics for Three ICU Populations, Unselected and Then Dichotomized by Age Figure 1. Conditional survival—survival to hospital discharge on each day of ICU admission for Post Intensive Care Risk-Adjusted Alerting and Monitoring (PICRAM) (left), Medical Information Mart for Intensive Care III (MIMIC-III) (middle), and electronic ICU (eICU) (right) populations. Dashed lines represent 95% CIs.

TABLE 1. Descriptive Statistics for Three ICU Populations, Unselected and Then Dichotomized by Age Figure 1. Conditional survival—survival to hospital discharge on each day of ICU admission for Post Intensive Care Risk-Adjusted Alerting and Monitoring (PICRAM) (left), Medical Information Mart for Intensive Care III (MIMIC-III) (middle), and electronic ICU (eICU) (right) populations. Dashed lines represent 95% CIs. When patients were dichotomized into less than 75 years old and greater than or equal to 75, there was an overall reduced survival from initial admission in the older age group, which was followed by a steeper decrease in conditional survival with increasing length of stay (Fig. 2). Fewer than 50 patients over the age of 75 years had durations of stay longer than 21 days in ICU. The probability of future survival continued to decrease up to ~15 days in the older cohort with 52.3% (CI, 43.0–61.4%) conditional survival with duration of stay greater than 4 days. This was contrasted with 79.2% (CI, 75.8–82.2%) at the same time point in the younger cohort. Figure 2. Conditional survival—survival to hospital discharge on each day of ICU admission stay for Post Intensive Care Risk-Adjusted Alerting and Monitoring (PICRAM) (left), Medical Information Mart for Intensive Care III (MIMIC-III) (middle), and electronic ICU (eICU) (right) populations dichotomized by age. Dot-dash line represents patients less than 75 yr old (dotted lines represent 95% CIs); solid line represents patients greater than or equal to 75 yr old (dashed lines represent 95% CIs).

cal Information Mart for Intensive Care III (MIMIC-III) (middle), and electronic ICU (eICU) (right) populations dichotomized by age. Dot-dash line represents patients less than 75 yr old (dotted lines represent 95% CIs); solid line represents patients greater than or equal to 75 yr old (dashed lines represent 95% CIs). MIMIC-III Thirty-eight–thousand five-hundred thirty-two index admissions were included in the initial cohort, no admissions were excluded for missing data, median age was 66 years (IQR, 52–78 yr), 56.6% were male, median length of stay was 2 days (IQR, 1–4 d), and 88.9% survived to hospital discharge (Table 1). This initial cohort reduced to 166 patients who were admitted for longer than 40 days. In all patients, conditional survival declined over the initial 10 days but then plateaued with minimal changes in conditional survival as length of stay increased (Fig. 1). Initial survival of 88.9% (CI, 88.5–89.2%) for all patients admitted to the ICU declined to a conditional survival of 78.8% (CI, 77.4–80.0%) in patients admitted for greater than 9 days and remained at this level (supplementary table, Supplemental Digital Content 1, http://links.lww.com/CCM/F27).

ay increased (Fig. 1). Initial survival of 88.9% (CI, 88.5–89.2%) for all patients admitted to the ICU declined to a conditional survival of 78.8% (CI, 77.4–80.0%) in patients admitted for greater than 9 days and remained at this level (supplementary table, Supplemental Digital Content 1, http://links.lww.com/CCM/F27). When patients were dichotomized by age, there was a steeper decrease in conditional survival in the older age group with increased length of stay (Fig. 2). The probability of hospital survival was lower at admission in older patients, 82.7% (CI, 82–83.3%) compared with 91.7% (CI, 91.3–92.0%). Conditional survival continued to decrease in the older population to 68.0% (CI, 65.2–70.1%) in patients admitted for greater than 9 days before plateauing (supplementary table, Supplemental Digital Content 1, http://links.lww.com/CCM/F27). eICU One-hundred sixty-five–thousand one-hundred twenty-five index admissions were included in the initial cohort, 1,696 admissions were excluded due to missing data (1%), median age was 65 years (IQR, 52–76 yr), 53.8% were male, median length of stay was 1 day (IQR, 0–2 d) and 91.0% survived to hospital discharge (Table 1). This initial cohort reduced to 181 patients who were admitted for longer than 40 days. Conditional survival declined over the initial 10 days but then plateaued with minimal changes in as length of stay increased (Fig. 1). Conditional survival at admission decreased from 90.3% to 84.2% in patients admitted to ICU for greater than 9 days.

reduced to 181 patients who were admitted for longer than 40 days. Conditional survival declined over the initial 10 days but then plateaued with minimal changes in as length of stay increased (Fig. 1). Conditional survival at admission decreased from 90.3% to 84.2% in patients admitted to ICU for greater than 9 days. When patients were dichotomized by age, survival at admission was 85.9% in the older cohort compared with 92.8% in the younger cohort and was followed by a steeper decrease in conditional survival in the older age group (supplementary table, Supplemental Digital Content 1, http://links.lww.com/CCM/F27). Although the initial decline was greater in the older population, the conditional survival plateaued at a similar time point of around 10 days of ICU (Fig. 2). Comparison of Databases The median ages of the patients in each database were similar, median length of stay was shorter in the eICU database (Table 1). Hospital mortality rates were higher in the PICRAM database, 18.5% compared with 11.1% and 9.0% in MIMIC-III and eICU, respectively. Median OASIS scores were also higher in PICRAM, 33 compared with 30 in MIMIC and eICU. Patients in the eICU databases had lower rates of mechanical ventilation compared with PICRAM and OASIS.

ital mortality rates were higher in the PICRAM database, 18.5% compared with 11.1% and 9.0% in MIMIC-III and eICU, respectively. Median OASIS scores were also higher in PICRAM, 33 compared with 30 in MIMIC and eICU. Patients in the eICU databases had lower rates of mechanical ventilation compared with PICRAM and OASIS. DISCUSSION In an observational retrospective study of three intensive care databases, we have demonstrated that in unselected patients admitted to an ICU, after an initial period of 5–10 days conditional survival to hospital discharge does not decrease with length of ICU stay. Assessment of conditional survival generates prognostic information in how the probability of surviving to hospital discharge changes as length of ICU stay increases. When the populations were dichotomized by age, less than 75 years, and greater than or equal to 75 years, this finding persisted in the younger cohorts, but in the older cohorts, conditional survival decreased with increasing length of stay for a longer period and also declined more sharply.

ngth of ICU stay increases. When the populations were dichotomized by age, less than 75 years, and greater than or equal to 75 years, this finding persisted in the younger cohorts, but in the older cohorts, conditional survival decreased with increasing length of stay for a longer period and also declined more sharply. To our knowledge, this is the first report to assess ICU mortality as conditional survival with the aim of describing changing mortality over the duration of an ICU admission. This method has previously been used in oncology to describe changing hazard over time and describes the probability of ongoing survival if a patient has survived to a point in time (7). Previous studies have used regression methods to assess how a patient’s total length of stay is linked to their probability of survival. However, here we describe the probability of future survival given they have already survived a certain period of time.

oing survival if a patient has survived to a point in time (7). Previous studies have used regression methods to assess how a patient’s total length of stay is linked to their probability of survival. However, here we describe the probability of future survival given they have already survived a certain period of time. The mortality rate is higher in the PICRAM database of English ICUs which is possibly due to these ICUs admitting patients with higher illness severity scores as has previously been reported (14). However, the OASIS score does not differ as significantly as might be expected considering the difference in mortality which is approximately double in PICRAM compared with the eICU database. The OASIS score only uses 10 variables, and it is likely that it does not account fully for the differences in severity of illness between databases. Further, if eICU contains a significantly different case-mix from the conventional ICUs used to develop and validate the score, it may perform less well. The eICU population has a shorter length of stay than MIMIC-III and PICRAM, median length of stay 1 day contrasted with 2 days in the other populations. This may represent more complex patients being transferred to tertiary referral centers. However, it should be noted that patients in all three databases have a relatively short median length of stay. PICRAM also contains a high proportion of elective surgical patients which is associated with lower mortality. With bed pressures in the United Kingdom, it is likely that only the most high-risk surgical procedures involve an elective ICU admission. Finally, it is noted that PICRAM has the highest proportion of intubated patients at admission. Intubation is associated with high mortality and is included with the OASIS score.

y. With bed pressures in the United Kingdom, it is likely that only the most high-risk surgical procedures involve an elective ICU admission. Finally, it is noted that PICRAM has the highest proportion of intubated patients at admission. Intubation is associated with high mortality and is included with the OASIS score. There is no consensus on what constitutes a “prolonged stay” in an ICU. However, attempts have been made to characterize a subpopulation of patients with “persistent critical illness” (15). This occurs when the admitting diagnosis is no longer the reason for their continued treatment on an ICU (16). Previous studies in Scotland and Australia have characterized patients with very long ICU admissions of greater than 30 or greater than 60 days and have shown that many survive (17, 18). However, with a median duration of ICU stay of 2 days (IQR, 1–5 d) (19), it may be that a much shorter duration of ICU admission could be considered prolonged. In all three cohorts assessed, there appears to be an initial phase of worsening probability of future survival with increasing length of stay. However, this plateaued within 10 days with little change in conditional survival after this time. Our results suggest that the change in phenotype to that of long stayer may occur as early as 10 days. In the PICRAM population, there was a favorable trend after 15 days not seen in the other populations, it is unclear whether this is a true effect or variance from low patient numbers.

tional survival after this time. Our results suggest that the change in phenotype to that of long stayer may occur as early as 10 days. In the PICRAM population, there was a favorable trend after 15 days not seen in the other populations, it is unclear whether this is a true effect or variance from low patient numbers. Our findings of an initial period of reducing survival with increased stay is supported by the findings of Williams et al (20) who report in an Australian ICU population a linear decrease in long term mortality for an initial 5-day period followed by a relatively constant risk. Other reports have also demonstrated favorable outcomes in patients with prolonged admissions to ICU (17, 21). This information is useful for clinicians and patients because it highlights that a prolonged or increasing length of stay is not a reason for pessimism. It may be that a belief that increasing duration of stay confers a poor prognosis contributing to the sharper decline in the older population where life-sustaining measures are being withdrawn.

for clinicians and patients because it highlights that a prolonged or increasing length of stay is not a reason for pessimism. It may be that a belief that increasing duration of stay confers a poor prognosis contributing to the sharper decline in the older population where life-sustaining measures are being withdrawn. In patients greater than or equal to 75 years old their survival at admission is lower than a younger cohort and their conditional survival declines more rapidly and plateaus after a longer duration of ICU admission. In the PICRAM population, only conditional survival appeared to improve in the older population around day 15, however, the number of patients at this time point is small and the significance of this trend is unclear. The finding that older patients have worse outcomes has been highlighted in previous reports (10) and age is usually a component of illness severity scoring systems such as Acute Physiology and Chronic Health Evaluation II. However, the difference in response to increasing duration of stay is novel. The impact of age on probability of future survival may, in part, be due to a reduction in physiologic reserve and faster rate of deconditioning that occurs with aging. Older patients have been demonstrated to decondition within the first 24 hours of hospital admission (22). A prospective analysis of patients greater than 80 years old admitted to Canadian ICUs demonstrated a poor functional recovery and high mortality rate (23). However, our retrospective analysis, cannot correct for any selection bias that may be resulting from the medical team withdrawing care earlier in elderly patients.

prospective analysis of patients greater than 80 years old admitted to Canadian ICUs demonstrated a poor functional recovery and high mortality rate (23). However, our retrospective analysis, cannot correct for any selection bias that may be resulting from the medical team withdrawing care earlier in elderly patients. This study has a number of important limitations. As duration of ICU admission increases, there are fewer patients remaining within the cohort to assess conditional survival. To minimize any bias, this may introduce we a priori set an arbitrary cutoff of a minimum of 50 patients for continued analysis. However, as length of stay increases, the precision in our findings diminishes as represented by widening error bars on Figs. 1 and 2. The limited cohort size for these more extended stays indicates the findings may not be generalizable to all ICU patients with prolonged admissions. As this is an observational study, we cannot account for any impact of length of stay on decisions to continue or withdraw care which may have an impact on future survival. Further, a proportion of patients will have had their care stepped down from level 3 to level 2 care or alternatively may have delayed discharge to the ward. We have been unable to account for this in our analysis. We have included within our analysis a minority of patients who have been transferred out either to an ICU or a ward at an alternative hospital. A proportion of these patients may die prior to discharge for the same episode of illness, these events would not be captured and contribute to our primary outcome.

analysis. We have included within our analysis a minority of patients who have been transferred out either to an ICU or a ward at an alternative hospital. A proportion of these patients may die prior to discharge for the same episode of illness, these events would not be captured and contribute to our primary outcome. The strengths of this study include a large population of unselected ICU patients, which provides good generalizability. Further, we have replicated the findings in two North American ICU populations and demonstrated consistent findings. That these findings remain consistent despite the differences between the ICU populations lends considerable support to the validity of our findings. Large clinical datasets are limited by their reliability of data collection and assumptions that are required to construct the database. Therefore, we have limited our study to reporting objective measures, specifically age, sex, length of stay, ICU, and hospital mortality. In PICRAM and eICU, outcome data were not available for 4% and 1% of the cohorts, respectively. Exclusion of these subjects may have biased our findings toward higher or lower hospital survival, but the effect would be limited given the small exclusion rates. Further work may focus on case-mix adjustment to assess the impact of initial and evolving severity of illness scores on conditional survival. Additionally, analysis of subpopulations of patients with different diagnoses may yield contrasting patterns of conditional survival and further inform clinicians when assessing the impact of length of stay on patient outcome.

assess the impact of initial and evolving severity of illness scores on conditional survival. Additionally, analysis of subpopulations of patients with different diagnoses may yield contrasting patterns of conditional survival and further inform clinicians when assessing the impact of length of stay on patient outcome. CONCLUSIONS After an initial period of 5–10 days, probability of future survival does appear to decrease with increasing length of stay in unselected patients admitted to ICUs in United Kingdom and United States. Length of stay in itself should therefore not be factored in to decisions around withdrawal of life-sustaining measures. In a subpopulation of patients, 75 years or older probability of future survival continued to decrease with increasing length of stay. Supplementary Material Drs. Marshall, Young, and Watkinson designed the study. Drs. Marshall and Hatch and Mr. Gerry conducted the analysis. All authors were responsible for interpreting the data and drafting of the article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

Supplementary Material Drs. Marshall, Young, and Watkinson designed the study. Drs. Marshall and Hatch and Mr. Gerry conducted the analysis. All authors were responsible for interpreting the data and drafting of the article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). This publication has been made possible through access to a research database that was created with support from the Health Innovation Challenge Fund (HICF-0510-006; WT-094951), a parallel funding partnership between the Department of Health and Wellcome Trust. Access was granted by the owners of the research database, the University of Oxford Critical Care Research Group.

to a research database that was created with support from the Health Innovation Challenge Fund (HICF-0510-006; WT-094951), a parallel funding partnership between the Department of Health and Wellcome Trust. Access was granted by the owners of the research database, the University of Oxford Critical Care Research Group. Dr. Young’s institution received funding from Health Innovation Challenge Fund (joint venture of Wellcome Trust and U.K. Department of Health. Dr. Watkinson’s institution received funding from Drayson Health and National Institute for Health Research (NIHR) Biomedical Research Center, Oxford, and he received funding from Drayson Health. Dr. Hatch is funded by an NIHR Academic Clinical Fellowship. Gerry is funded by an NIHR Doctoral Fellowship (DRF-2016-09-073). Dr. Watkinson has developed an electronic observations application for which Drayson Health (now Sensyne Health) has purchased a sole license. The company has a research agreement with the University of Oxford and pay personal fees. No other authors have financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years, nor other relationships or activities that could appear to have influenced the submitted work. Dr. Marshall has disclosed that he does not have any potential conflicts of interest.

Sepsis is a life-threatening condition that occurs when the body’s response to infection causes tissue damage, organ failure, or death (1–3). In the United States, nearly 1.7 million people develop sepsis and 270,000 people die of sepsis each year; over one third of people who die in U.S. hospitals have sepsis (4). Globally, an estimated 30 million people develop sepsis and 6 million people die of sepsis each year (5). Costs for managing sepsis in U.S. hospitals exceed those for any other health condition at $24 billion annually (13% of U.S. healthcare expenses); a majority of these costs are for patients who develop sepsis during their hospital stay (6). The developing world faces additional expenses from sepsis management and higher risks of adverse outcomes. Altogether, sepsis is a major public health issue responsible for significant morbidity, mortality, and healthcare expenses (7–10).

these costs are for patients who develop sepsis during their hospital stay (6). The developing world faces additional expenses from sepsis management and higher risks of adverse outcomes. Altogether, sepsis is a major public health issue responsible for significant morbidity, mortality, and healthcare expenses (7–10). The reliable and early identification of sepsis is often complicated by its syndromic nature, which can contribute to delays in treatment. The importance of early identification and treatment of sepsis is highlighted in two recent studies that suggest an increase in the adjusted mortality of septic patients who experienced delays in antibiotic therapy (11, 12). This effect is even more profound in patients suffering from septic shock, where hourly delays were associated with an 3.6–9.9% increase in mortality per hour (13). Professional critical care societies have proposed clinical criteria for recognizing and treating sepsis (1–3); however, the fundamental need for early and reliable identification of sepsis remains unmet (14). The PhysioNet/Computing in Cardiology Challenge is an international competition focused on open-source solutions for complex physiologic signal processing and medical classification problems (15). In 2019, the Challenge’s 20th year, we asked participants to develop automated techniques for the early detection of sepsis from clinical data.

ing in Cardiology Challenge is an international competition focused on open-source solutions for complex physiologic signal processing and medical classification problems (15). In 2019, the Challenge’s 20th year, we asked participants to develop automated techniques for the early detection of sepsis from clinical data. Computational approaches promise to improve the early detection of sepsis. Such approaches typically apply machine learning techniques to clinical data (e.g., see Refs. [16–18]), with the goal of making real-time predictions up to a day before clinical recognition of sepsis. However, the relative strengths and weaknesses of algorithmic approaches are unclear for a variety of reasons. The PhysioNet/Computing in Cardiology Challenge 2019 provided an opportunity to explore the limits of computational approaches for detecting sepsis.

edictions up to a day before clinical recognition of sepsis. However, the relative strengths and weaknesses of algorithmic approaches are unclear for a variety of reasons. The PhysioNet/Computing in Cardiology Challenge 2019 provided an opportunity to explore the limits of computational approaches for detecting sepsis. First, algorithms for the early detection of sepsis frequently address subtly different problems, and they tend to have been developed and tested in different patient cohorts with different clinical variables and labels arising from different clinical criteria for sepsis. For the Challenge, we provided a common problem statement using the same clinical variables and sepsis criteria. We shared data from two separate hospital systems and sequestered data from a third hospital system. Algorithms that overfit on the shared databases typically underperformed on the hidden database, particularly if they encoded data collection behaviors specific to a given hospital system. Furthermore, we ran algorithms only once on the full hidden dataset to prevent sequential training on the hidden data, and we compared algorithms to identify teams that attempted to circumvent the rules and have more “bites of cherry” than other teams.

oded data collection behaviors specific to a given hospital system. Furthermore, we ran algorithms only once on the full hidden dataset to prevent sequential training on the hidden data, and we compared algorithms to identify teams that attempted to circumvent the rules and have more “bites of cherry” than other teams. Second, different studies often employ different evaluation metrics, and such metrics do not necessarily reflect the clinical utility of sepsis detection and treatment. Traditional scoring metrics, such as area under the curve (AUC) metrics, do not explicitly reward early detection or penalize false alarms or overtreatment. For the Challenge, we devised a novel evaluation metric that addresses these issues and could be generally applicable to predicting infrequent events in time series data. Third, the complexity of such algorithms is nearly impossible to adequately describe in a research article. For the Challenge, we encouraged and facilitated the open sourcing of algorithms to ensure that subtle implementation details are provided and reproducible. METHODS Challenge Objective The goal of this Challenge was the development of algorithms for the early prediction of sepsis using routinely available clinical data. Early predictions of sepsis are potentially lifesaving, whereas late or missed predictions are potentially life threatening, and false alarms consume hospital resources and erode trust in the algorithms themselves (19).

evelopment of algorithms for the early prediction of sepsis using routinely available clinical data. Early predictions of sepsis are potentially lifesaving, whereas late or missed predictions are potentially life threatening, and false alarms consume hospital resources and erode trust in the algorithms themselves (19). For this Challenge, we asked participants to design and implement working, open-source algorithms that can, based only on the provided clinical data, automatically identify a patient’s risk of sepsis and make a positive or negative prediction of sepsis for every hourly time window in the patient’s clinical record. In particular, we asked participants to predict sepsis at least 6 hours (but not more than 12 hr) before the onset time of sepsis according to Sepsis-3 clinical criteria (1–3). To evaluate each algorithm, we designed a new clinical utility-based scoring metric that rewards algorithms for early sepsis predictions and penalizes them for late and missed sepsis predictions as well as for false alarms. The winners of this Challenge were the team whose algorithm gave predictions with the highest clinical utility score for patients in a hidden test set across three hospital systems.

hat rewards algorithms for early sepsis predictions and penalizes them for late and missed sepsis predictions as well as for false alarms. The winners of this Challenge were the team whose algorithm gave predictions with the highest clinical utility score for patients in a hidden test set across three hospital systems. We awarded prizes to teams with winning algorithms. Although we allowed both noncommercial and commercial entities to enter, only open-source entries were eligible for prizes. All code was required to be submitted to ensure that methods were replicable and because no teams had access to the hidden data. This allowed for the comparison of winning teams with commercial entities and increased the competitive landscape. Challenge Data We obtained the data for the Challenge from three geographically distinct U.S. hospital systems with three different Electronic Medical Record (EMR) systems: Beth Israel Deaconess Medical Center (hospital system A), Emory University Hospital (hospital system B), and a third, unidentified hospital system (hospital system C). These data were collected over the past decade with approval from the appropriate institutional review boards. We deidentified and labeled the data using Sepsis-3 clinical criteria (1–3). Data and labels for 40,336 patients from hospital systems A and B were posted publicly for download, and data and labels for 24,819 patients from hospital systems A, B, and C were sequestered as hidden test sets.

e institutional review boards. We deidentified and labeled the data using Sepsis-3 clinical criteria (1–3). Data and labels for 40,336 patients from hospital systems A and B were posted publicly for download, and data and labels for 24,819 patients from hospital systems A, B, and C were sequestered as hidden test sets. The Challenge data consisted of a combination of hourly vital sign summaries, laboratory values, and static patient descriptions. In particular, the data contained 40 clinical variables: eight vital sign variables, 26 laboratory variables, and six demographic variables; Table 1 describes these variables. Altogether, these data included over 2.5 million hourly time windows and 15 million data points. TABLE 1. Feature Summary Data extracted from the EMR underwent a series of preprocessing steps prior to formal analysis and model development. All patient features were condensed into hourly bins simplifying model development and testing; for example, multiple heart rate measurements in an hourly time window were summarized as the median heart rate measurement. Multiple Logical Observation Identifiers Names and Codes codes describing the same clinical parameter were condensed into a single variable; for example, serum hemoglobin and arterial hemoglobin became hemoglobin. We labeled patient data in accordance with Sepsis-3 clinical criteria (1–3). For each septic patient, we specified the following three time points to define the onset time tsepsis of sepsis:

Data extracted from the EMR underwent a series of preprocessing steps prior to formal analysis and model development. All patient features were condensed into hourly bins simplifying model development and testing; for example, multiple heart rate measurements in an hourly time window were summarized as the median heart rate measurement. Multiple Logical Observation Identifiers Names and Codes codes describing the same clinical parameter were condensed into a single variable; for example, serum hemoglobin and arterial hemoglobin became hemoglobin. We labeled patient data in accordance with Sepsis-3 clinical criteria (1–3). For each septic patient, we specified the following three time points to define the onset time tsepsis of sepsis: tsuspicion: Clinical suspicion of infection identified as the earlier timestamp of IV antibiotics and blood cultures within a given time interval. If IV antibiotics were given first, then the cultures must have been obtained within 24 hours. If cultures were obtained first, then IV antibiotic must have been ordered within 72 hours. In either case, IV antibiotics must have been administered for at least 72 consecutive hours. tSOFA: Occurrence of organ failure as identified by a two-point increase in the Sequential Organ Failure Assessment (SOFA) score within a 24-hour period. tsepsis: Onset of sepsis identified as the earlier of tsuspicion and tSOFA as long as tSOFA occurred no more than 24 hours before or 12 hours after tsuspicion.

tsuspicion: Clinical suspicion of infection identified as the earlier timestamp of IV antibiotics and blood cultures within a given time interval. If IV antibiotics were given first, then the cultures must have been obtained within 24 hours. If cultures were obtained first, then IV antibiotic must have been ordered within 72 hours. In either case, IV antibiotics must have been administered for at least 72 consecutive hours. tSOFA: Occurrence of organ failure as identified by a two-point increase in the Sequential Organ Failure Assessment (SOFA) score within a 24-hour period. tsepsis: Onset of sepsis identified as the earlier of tsuspicion and tSOFA as long as tSOFA occurred no more than 24 hours before or 12 hours after tsuspicion. Missing and erroneous data were intentionally preserved as part of the Challenge. However, patients with less than 8 hourly time windows of data in the ICU were not included, and patients with tsepsis less than 4 hours after ICU admission were not included. Patient records were truncated after ICU discharge, and patients with more than 2 weeks of hourly time windows were truncated to 2 weeks.

lenge. However, patients with less than 8 hourly time windows of data in the ICU were not included, and patients with tsepsis less than 4 hours after ICU admission were not included. Patient records were truncated after ICU discharge, and patients with more than 2 weeks of hourly time windows were truncated to 2 weeks. Supplemental Table 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/F206) summarizes the datasets for the two shared hospital databases. Figure 1 shows the densities of entries (i.e., the fraction of non-empty hourly measurements) for each vital sign and laboratory value in each patient record; most vital signs were updated on an hourly basis in most patient records, and most laboratory values were updated on a daily basis. Supplemental Figure 1 (Supplemental Digital Content 2, http://links.lww.com/CCM/F207) shows the distributions of these entries across patient records. Supplemental Figure 2 (Supplemental Digital Content 3, http://links.lww.com/CCM/F208) quantifies the difference between the vital sign and laboratory value distributions between hospital systems using Jensen-Shannon divergence. Note that most clinical variables have similar distributions across hospital systems. Figure 1. Densities of vital sign (rows 1-8) and laboratory value (rows 9-34) entries (fraction of non-empty entries) in the shared and hidden datasets for hospital systems A, B, and C.

Supplemental Table 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/F206) summarizes the datasets for the two shared hospital databases. Figure 1 shows the densities of entries (i.e., the fraction of non-empty hourly measurements) for each vital sign and laboratory value in each patient record; most vital signs were updated on an hourly basis in most patient records, and most laboratory values were updated on a daily basis. Supplemental Figure 1 (Supplemental Digital Content 2, http://links.lww.com/CCM/F207) shows the distributions of these entries across patient records. Supplemental Figure 2 (Supplemental Digital Content 3, http://links.lww.com/CCM/F208) quantifies the difference between the vital sign and laboratory value distributions between hospital systems using Jensen-Shannon divergence. Note that most clinical variables have similar distributions across hospital systems. Figure 1. Densities of vital sign (rows 1-8) and laboratory value (rows 9-34) entries (fraction of non-empty entries) in the shared and hidden datasets for hospital systems A, B, and C. Challenge Scoring We scored each algorithm’s predictions using a novel evaluation metric that we created for the Challenge. To better capture the clinical utility of sepsis detection and treatment, this metric rewarded algorithms for early sepsis predictions in septic patients, and it penalized algorithms for late or missed sepsis predictions in septic patients and for sepsis predictions in nonseptic patients.

we created for the Challenge. To better capture the clinical utility of sepsis detection and treatment, this metric rewarded algorithms for early sepsis predictions in septic patients, and it penalized algorithms for late or missed sepsis predictions in septic patients and for sepsis predictions in nonseptic patients. Each algorithm made a binary sepsis prediction for each hourly time window in each patient record. To evaluate each algorithm, we first defined a score for each prediction and then aggregated these scores over all hourly time windows and all patient records. Given an algorithm’s prediction for an hourly time window t in a patient record s, we define a score [1] where UTP(s, t), UFP(s, t), UFN(s, t), and UTN(s, t) are illustrated in Figure 2A for an example septic patient and in Figure 2B for an example nonseptic patient. These scores were chosen to reflect the broad clinical realities of sepsis detection and treatment, and the actual utility values and time points in [1] and Figure 2 can be chosen to capture the specific preferences or trade-offs of any particular hospital system. Figure 2. Diagrams of utility of positive and negative predictions for sepsis and non-septic patients; the time tsepsis = 48 of sepsis onset is given as an example.

[1] where UTP(s, t), UFP(s, t), UFN(s, t), and UTN(s, t) are illustrated in Figure 2A for an example septic patient and in Figure 2B for an example nonseptic patient. These scores were chosen to reflect the broad clinical realities of sepsis detection and treatment, and the actual utility values and time points in [1] and Figure 2 can be chosen to capture the specific preferences or trade-offs of any particular hospital system. Figure 2. Diagrams of utility of positive and negative predictions for sepsis and non-septic patients; the time tsepsis = 48 of sepsis onset is given as an example. For patients who become septic during their ICU stay, early sepsis detection tends to be beneficial. Therefore, sepsis predictions in septic patients who were at least 12 hours before and at most 3 hours after the onset time tsepsis of sepsis were rewarded with a maximum reward at 6 hours before tsepsis, and sepsis predictions that are more than 12 hours before tsepsis were slightly penalized. Similarly, for patients who become septic during their ICU stay, very early predictions may be implausible or unhelpful, and late or missed septic predictions are generally harmful. Therefore, sepsis predictions in septic patients who were more than 12 hours before tsepsis were slightly penalized, and nonsepsis predictions that were less than 6 hours before tsepsis were increasingly penalized.

predictions may be implausible or unhelpful, and late or missed septic predictions are generally harmful. Therefore, sepsis predictions in septic patients who were more than 12 hours before tsepsis were slightly penalized, and nonsepsis predictions that were less than 6 hours before tsepsis were increasingly penalized. For patients who do not become septic during their ICU stay, sepsis predictions contribute to alarm fatigue and lower confidence in algorithms, antibiotic overuse, and overall poor allocation of hospital attention and resources. Therefore, sepsis predictions in nonseptic patients were slightly penalized. Similarly, nonsepsis predictions in nonseptic patients were neither rewarded nor penalized. Given an algorithm’s predictions for all hourly time windows T (s) in each patient record s, we define the total score for an algorithm as the sum [2] over all predictions. For easier interpretability, we normalize [2] so that the optimal algorithm with the highest possible score receives a normalized score of 1 and a completely inactive algorithm that only makes nonsepsis predictions receives a normalized score of 0, that is, [3] Each algorithm received a score from [3], and the algorithm with the highest value of [3] on the full sequestered dataset from hospital systems A, B, and C won the Challenge. Challenge Submissions Challenge participants submitted their algorithms for evaluation on the sequestered data. This strategy encouraged reproducibility and gave participants the ability to validate their algorithms on real-world datasets.

[3] Each algorithm received a score from [3], and the algorithm with the highest value of [3] on the full sequestered dataset from hospital systems A, B, and C won the Challenge. Challenge Submissions Challenge participants submitted their algorithms for evaluation on the sequestered data. This strategy encouraged reproducibility and gave participants the ability to validate their algorithms on real-world datasets. Each team was allowed a total of five scored entries during an unofficial phase of the Challenge from February 8, 2019, to April 14, 2019. This phase allowed for beta testing and socialization of the submission system, rules, and scoring mechanism, and teams were required to submit at least one entry during the unofficial phase for Challenge eligibility. Subsequently, each eligible team was allowed a total of 10 scored entries during the official phase of the Challenge from April 25, 2019, to August 25, 2019. This phase allowed teams to submit their models for evaluation on test data from hospital system A; scoring on the full hidden test data occurred only after the official phase at the end of the Challenge. This limit also improved the tractability of the Challenge. Because we did not heavily restrict the languages and libraries that teams could use, many teams required technical support for their submissions.

ystem A; scoring on the full hidden test data occurred only after the official phase at the end of the Challenge. This limit also improved the tractability of the Challenge. Because we did not heavily restrict the languages and libraries that teams could use, many teams required technical support for their submissions. The submission system relied on containers that were orchestrated, as pipelines, on the Google Cloud Platform; Supplemental Figure 3 (Supplemental Digital Content 4, http://links.lww.com/CCM/F209) illustrates this system. A container is a standard unit of software that packages code and its dependencies so that the application runs readily and reliably in different computing environments. For the Challenge, we used the Docker containerization environment. Participants packaged their entries and uploaded them to a GitHub repository (Microsoft, San Francisco, CA), which was shared privately with the Challenge organizers. For each submission, the submission system cloned the repository, created a pipeline that consisted of the entry and our scoring function, and launched this pipeline on Google Cloud. This system allowed us to score multiple entries in parallel. During the unofficial and official phases of the Challenge, we processed over a thousand submissions in Julia (https://julialang.org), MATLAB (MathWorks, Natick, MA), Python (https://www.python.org), and R (R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org) from over a hundred participants.

ng the unofficial and official phases of the Challenge, we processed over a thousand submissions in Julia (https://julialang.org), MATLAB (MathWorks, Natick, MA), Python (https://www.python.org), and R (R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org) from over a hundred participants. Each entry was run in a virtual machine with two central processing units and 12 GB of random access memory, and each entry was allowed 24 hours of run time on each hidden test set. The submission system orchestrator, Cromwell (The Eli and Edythe L. Broad Institute of MIT and Harvard, Cambridge, MA), typically requested a n2-highmem-2 machine type on Google Cloud. Implementations of Evaluation Metric and Baseline Model To provide a baseline model, we trained a Weibull-Cox regression model and provided open-source implementations of this model in Julia, MATLAB, Python, and R. These implementations also served as examples of how to devise a working prediction algorithm in each language that we accepted for the Challenge. We also provided open-source implementations of our clinical utility-based scoring function. The code is available online at https://github.com/physionetchallenges.

, and R. These implementations also served as examples of how to devise a working prediction algorithm in each language that we accepted for the Challenge. We also provided open-source implementations of our clinical utility-based scoring function. The code is available online at https://github.com/physionetchallenges. Analysis of Entry Independence, Collusion, and Plagiarism After the conclusion of each Challenge, we frequently build a meta-algorithm from the final entries that are weighted by their independence; agreement between highly similar algorithms can suggest a false consensus of predictions. To increase the independence of algorithms, we therefore prohibited teams from collaboration at any point of the Challenge. Specifically, we note the following: Multiple teams from a single entity (such as a company, university, or department) were permitted as long as the teams were truly independent and did not share team members, code, or ideas at any point. Multiple teams from the same research group or unit within a company were not allowed because we did not believe that true independence between teams could be maintained when team members may frequently interact. New team members could join as long as they had not previously been involved with another team or had communicated with a team member from another team concerning this year’s Challenge.

Multiple teams from a single entity (such as a company, university, or department) were permitted as long as the teams were truly independent and did not share team members, code, or ideas at any point. Multiple teams from the same research group or unit within a company were not allowed because we did not believe that true independence between teams could be maintained when team members may frequently interact. New team members could join as long as they had not previously been involved with another team or had communicated with a team member from another team concerning this year’s Challenge. Teams could use public code if it had been posted before the competition. Members of teams were not allowed to publicly post code during the competition or use another competitor’s code that was posted during the competition whether or not it was intentionally made public. Members of teams were not allowed to publicly post information describing their methods or give a talk outside of their own research group at any point during the competition that revealed the methods they have employed or planned to employ in the Challenge. Members of teams were allowed to present or publish on methods on other data as long as they did not indicate that they planned to apply it to Challenge data until after the competition. Members of teams were required to use the same team name and email address throughout the course of the competition, including for abstract submissions to the public forum at which they defended their work, that is, at Computing in Cardiology.

Members of teams were not allowed to publicly post information describing their methods or give a talk outside of their own research group at any point during the competition that revealed the methods they have employed or planned to employ in the Challenge. Members of teams were allowed to present or publish on methods on other data as long as they did not indicate that they planned to apply it to Challenge data until after the competition. Members of teams were required to use the same team name and email address throughout the course of the competition, including for abstract submissions to the public forum at which they defended their work, that is, at Computing in Cardiology. Although the rules of the Challenge strictly prohibited teams from more than 10 scored entries during the official phase of the Challenge, several entries from apparently different teams achieved exactly the same score. An investigation of their submissions showed strong similarities between these teams, which, when questioned, either did not reply or claimed not to have colluded. By examining associations among email addresses, team names, and GitHub repositories, we were able to identify several prohibited collaborations. Supplemental Figure 4 (Supplemental Digital Content 5, http://links.lww.com/CCM/F210) illustrates associations among email addresses, team names, and GitHub users from Challenge submissions, where each team was expected to have only one email address, team name, and GitHub user. Some associations with multiple email addresses, team names, and/or GitHub users indicated prohibited collaborations and resulted in disqualifications.

among email addresses, team names, and GitHub users from Challenge submissions, where each team was expected to have only one email address, team name, and GitHub user. Some associations with multiple email addresses, team names, and/or GitHub users indicated prohibited collaborations and resulted in disqualifications. Results A total of 104 teams from academia and industry submitted a total of 853 entries during the official phase of the Challenge; of these, 88 distinct teams with a total of 430 entries were able to be scored. Recall that each team received training data and labels for hospital systems A and B but not for hospital system C. Each successful entry received scores on the test data for hospital system A during the unofficial and official phases of the Challenge, and each team nominated its favorite successful entry for evaluation on the full test data containing patient records from hospital systems A, B, and C. Table 2 summarizes the teams with the highest-scoring entries. TABLE 2. Top Clinical Utility Scores By curating clinical data from multiple hospital systems and sharing different amounts of data and information from these systems, we demonstrated that algorithms generally performed much better in two hospital systems for which we provided training data than a third hospital system for which we provided no training data.

By curating clinical data from multiple hospital systems and sharing different amounts of data and information from these systems, we demonstrated that algorithms generally performed much better in two hospital systems for which we provided training data than a third hospital system for which we provided no training data. Although algorithms that performed well by one evaluation metric might be expected to perform well by another metric, we saw that this was generally not the case for traditional evaluation metrics and the clinical utility score that we devised for the Challenge. Figure 3A compares each algorithm’s area under the receiver operating characteristic curve (AUROC) with its utility score on the test sets from each of the hospital systems. AUROC and utility scores are positively correlated on test sets A and B (Spearman rank correlation coefficients ρ = 0.791 and ρ = 0.839, respectively). These scores are poorly correlated on test set C (Spearman rank correlation coefficient ρ = 0.054), which corresponds to the hospital system for which participants did not receive training data. Furthermore, even on test sets A and B, algorithms with high utility scores did not necessarily have high AUROC scores, demonstrating that traditional evaluation metrics do not necessarily capture the clinical utility of predictions.

rresponds to the hospital system for which participants did not receive training data. Furthermore, even on test sets A and B, algorithms with high utility scores did not necessarily have high AUROC scores, demonstrating that traditional evaluation metrics do not necessarily capture the clinical utility of predictions. Figure 3. Comparison of each algorithm’s AUROC and utility scores on test data from hospital systems A, B, and C, where we shared training data for hospital systems A and B but not for hospital system C. A, Comparison of each algorithm’s area under the receiver operating characteristic curve (AUROC) and utility scores on test sets A, B, and C. B, Comparison of each algorithm’s AUROC and utility scores on test sets A and B. C, Comparison of each algorithm’s AUROC and utility scores on test sets A and C. D, Comparison of each algorithm’s AUROC and utility scores on test sets B and C. E, Ranked performance of the final algorithms on test sets A, B, and C. Red indicates a high overall ranking across all three databases, and blue indicates a low overall ranking. Lines from top to bottom indicate how the individual algorithm ranking changed when considering the performance on each database. Algorithms that performed well on test sets A and B generally performed relatively poorly on test set C.

overall ranking across all three databases, and blue indicates a low overall ranking. Lines from top to bottom indicate how the individual algorithm ranking changed when considering the performance on each database. Algorithms that performed well on test sets A and B generally performed relatively poorly on test set C. Furthermore, the choice of evaluation metric influenced how transferable algorithms appeared to be across hospital systems. Figure 3, B–D compares each algorithm’s AUROC or utility score on test sets from different hospital systems. Although AUROC scores are strongly correlated for each pair of hospital systems (Spearman rank correlation coefficients ρ = 0.973 for hospital systems A and B, ρ = 0.938 for hospital systems A and C, and ρ = 0.947 for hospital systems B and C), this is not true for utility scores. Utility scores are strongly correlated between the two hospital systems for which we provided training data (Spearman rank correlation coefficient ρ = 0.949 for test sets A and B), but they are poorly correlated with the third hospital system for which we did not provide training data (Spearman rank correlation coefficients ρ = –0.033 and ρ = 0.013 for hospital systems A and B, respectively, with hospital system C). Figure 3E further shows that the methods with the highest scores on data from hospital systems with shared training databases were not necessarily the methods with the highest scores on the hidden database from a separate hospital system.

nd ρ = 0.013 for hospital systems A and B, respectively, with hospital system C). Figure 3E further shows that the methods with the highest scores on data from hospital systems with shared training databases were not necessarily the methods with the highest scores on the hidden database from a separate hospital system. Our use of clinical data from multiple hospital systems and our application of a clinical utility-based evaluation metric provided a more nuanced view of predictive generalizability than results on one system with traditional evaluation metrics would present. DISCUSSION The PhysioNet/Computing in Cardiology Challenge 2019 asked participants to develop automated, open-source algorithms for the early detection of sepsis from clinical data. We assembled over 60,000 patient records from three hospital systems, with two shared publicly and one remaining hidden. By posting two databases publicly, we provided participants the opportunity to create training methodologies that do not overfit to one medical center. The third hidden database provided a strong indication of how well participants had accomplished this critical task. We also proposed and used a novel evaluation metric that captures the clinical utility of early sepsis detection, weighted by the relative “earliness” or “lateness” of each prediction.

DISCUSSION The PhysioNet/Computing in Cardiology Challenge 2019 asked participants to develop automated, open-source algorithms for the early detection of sepsis from clinical data. We assembled over 60,000 patient records from three hospital systems, with two shared publicly and one remaining hidden. By posting two databases publicly, we provided participants the opportunity to create training methodologies that do not overfit to one medical center. The third hidden database provided a strong indication of how well participants had accomplished this critical task. We also proposed and used a novel evaluation metric that captures the clinical utility of early sepsis detection, weighted by the relative “earliness” or “lateness” of each prediction. We suggest that this metric should be considered for wider adoption in clinical care because it does not suffer from many of the problems of F-measures (and related metrics such as accuracy, sensitivity, and positive predictive value) or standard AUC metrics (such as AUROC and area under the precision recall curve), which either assume a one-shot decision or no decision threshold, respectively. In particular, this novel evaluation metric shows that algorithms that perform well in one hospital system may perform poorly in another.

e predictive value) or standard AUC metrics (such as AUROC and area under the precision recall curve), which either assume a one-shot decision or no decision threshold, respectively. In particular, this novel evaluation metric shows that algorithms that perform well in one hospital system may perform poorly in another. A third novelty in this Challenge is the development of graphical and analytical approaches to measure the similarity between entries between supposedly independent Challenge teams. We identified and disqualified teams that appeared to be highly related to each other and did not provide satisfactory explanations of these relationships. We received 853 entries from 104 participants in academia and industry, providing a diverse view of algorithmic approaches to early sepsis detection. Combined, these efforts provide a more complete picture of how algorithms can provide early sepsis predictions. A subsequent analysis of the best performing and most interesting algorithms submitted to the Challenge will combine the strengths of different approaches to push the boundaries of automated approaches to early sepsis prediction. Supplementary Material Drs. Nemati, Clifford, and Sharma contributed equally to this work. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

Supplementary Material Drs. Nemati, Clifford, and Sharma contributed equally to this work. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by the Gordon and Betty Moore Foundation. This work was also supported by the National Center for Advancing Translational Sciences of the National Institutes of Health (NIH) under Award Number UL1TR002378 and the National Institutes of Health-sponsored Research Resource for Complex Physiologic Signals (www.physionet.org) (R01GM104987). The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

es of Health (NIH) under Award Number UL1TR002378 and the National Institutes of Health-sponsored Research Resource for Complex Physiologic Signals (www.physionet.org) (R01GM104987). The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Drs. Reyna, Jeter, Nemati, and Clifford are partially funded by the National Science Foundation under award number 1822378 (Leveraging Heterogeneous Data Across International Borders in a Privacy Preserving Manner for Clinical Deep Learning). Mr. Shashikumar and Dr. Nemati are also funded by the National Institutes of Health (NIH) award number K01ES025445. Drs. Reyna, Josef, Jeter, Westover, Clifford, and Sharma received support for article research from the NIH. Dr. Josef’s institution received funding from NIH (T-32 Grant Trainee: T32GM095442-09) and Henry M. Jackson Foundation for role as a post-doctoral researcher for the Surgical Critical Care Institute, www.sc2i.org, funded through the Department of Defense’s Defense Health Program—Joint Program Committee 6/Combat Casualty Care (Uniformed Services University of the Health Sciences HT9404-13-1-0032 and HU0001-15-2-0001) and was supported by a grant from the NIH, United States (NIH grant: 5T32GM095442-09). Dr. Clifford’s institution received funding from the Gordon and Betty Moore Foundation and NIH and he received cloud credits from Google Cloud. Dr. Sharma’s institution received funding from the Gordon and Betty Moore Foundation, and he received funding from Google (travel reimbursement for a talk at a seminar). Dr. Sharma and the development of the cloud-based scoring system were partially supported by the National Cancer Institute (U24CA215109). The remaining authors have disclosed that they do not have any potential conflicts of interest.

Sudden cardiac arrest (CA) both in and out of hospital is common and has high morbidity and mortality (1). Early, high-quality cardiopulmonary resuscitation (CPR) has been demonstrated to increase survival, but its effectiveness deteriorates within minutes if its initiation is delayed (2, 3). Delayed CPR is common and often associated with cardiogenic shock resulting in hemodynamic instability and poor neurologic outcomes (4). The severity of post-CPR shock may also contribute to the extent of neurologic outcomes in surviving patients (5). Post-CPR cardiogenic shock occurs even in the absence of acute coronary artery occlusion and is a component of the “post-CA syndrome” (5, 6). The pathophysiology of post-CPR cardiogenic shock is unknown, and effective therapies are lacking.

also contribute to the extent of neurologic outcomes in surviving patients (5). Post-CPR cardiogenic shock occurs even in the absence of acute coronary artery occlusion and is a component of the “post-CA syndrome” (5, 6). The pathophysiology of post-CPR cardiogenic shock is unknown, and effective therapies are lacking. Myocardial ischemia of short duration followed by adequate coronary flow restoration results in reversible myocardial dysfunction without necrosis. This cardiac pathology is termed “myocardial stunning” (7, 8). It was originally used to describe regional non–infarcted ventricular wall movement abnormalities, following brief coronary artery occlusion/reperfusion injuries, but has since been used to describe patients experiencing cardiogenic shock after percutaneous coronary artery intervention and cardiopulmonary bypass surgery (9, 10). Myocardial stunning is not commonly recognized as mediating post-CPR shock, and although it has been described in the setting of ventricular fibrillation–induced CA (8, 11), it has not been studied in other forms of CA, such as asystolic CA. Furthermore, the molecular mechanisms mediating myocardial stunning are unknown because it has been described as “lacking clinical relevance” (9).

shock, and although it has been described in the setting of ventricular fibrillation–induced CA (8, 11), it has not been studied in other forms of CA, such as asystolic CA. Furthermore, the molecular mechanisms mediating myocardial stunning are unknown because it has been described as “lacking clinical relevance” (9). Myocardial mitochondria occupy one third of the heart’s volume and are central regulators of calcium, reactive oxygen species (ROS), and metabolism. Mitochondria are dynamic organelles undergoing regulated fusion (joining) and fission (dividing) events (12, 13). Our group was the first to demonstrate evidence of mitochondrial fission following CA, its mediation of myocardial dysfunction through fission-induced ROS generation (14). In addition to mitochondrial fission, the accumulation of succinate during cellular ischemia results in increased electron leak and generation of superoxide and/or hydrogen peroxide (H2O2) (15, 16). Limiting electron leak and ROS generation using inhibitors of mitochondrial electron transport during ischemia/reperfusion (IR) have shown promise, but their utility is limited by their negative effects on metabolism (17–19).

increased electron leak and generation of superoxide and/or hydrogen peroxide (H2O2) (15, 16). Limiting electron leak and ROS generation using inhibitors of mitochondrial electron transport during ischemia/reperfusion (IR) have shown promise, but their utility is limited by their negative effects on metabolism (17–19). Recently, Brand et al (16) have identified compounds that protect against H2O2 production induced by electron leak at sites IQ (the ubiquinone-binding site of complex I, the active site during reverse electron transport), IIF (the flavin site of complex II), or IIIQ0 (the outer ubiquinone-binding site of complex III) in isolated skeletal muscle. One compound, the suppressor of site IQ electron leak (S1QEL), limited ROS generation at complex I without affecting normal electron transport. S1QEL also attenuated oxidative damage in several cell types while limiting infarct size in Langendorff-perfused mouse hearts. In this study, we hypothesized that the severity of cardiogenic shock following CPR is dependent on the length of CA and by virtue of lacking myocardial necrosis, is reversible, consistent with myocardial stunning. Furthermore, we hypothesized that myocardial stunning is mediated by increased mitochondrial complex I ROS generation and that its therapeutic targeting by S1QEL could improve postresuscitation outcomes. Findings in this study support the hypothesis and suggest that S1QEL has potential as a therapeutic agent to improve outcomes in CA.

hypothesized that myocardial stunning is mediated by increased mitochondrial complex I ROS generation and that its therapeutic targeting by S1QEL could improve postresuscitation outcomes. Findings in this study support the hypothesis and suggest that S1QEL has potential as a therapeutic agent to improve outcomes in CA. MATERIAL AND METHODS CA Mice Model Asystolic CA was induced in adult (age, 6–8 mo; 20–30 g) retired breeder female C57BL/6 mice as previously described (14). Briefly, mice were anesthetized (3% vaporized isoflurane) and instrumented. Asystolic CA was induced by 0.08 mg/g potassium chloride injection via the jugular vein. Following 12 minutes of CA, the ventilator was reconnected and manual chest compression was performed at a rate of 350~400 beats/min. After 90 seconds of CPR, 1.5 μg of epinephrine was injected. The CA protocol used in this study is illustrated in Supplemental Figure 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/F61). In accordance with National Institutes of Health guidelines, the University of Chicago Institutional Animal Care and Use Committee approved all animal procedures. In total, 121 mice entered the study. Twenty two mice died due to surgical failure and 49 mice could not be resuscitated. Additional details are described in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61).

stitutional Animal Care and Use Committee approved all animal procedures. In total, 121 mice entered the study. Twenty two mice died due to surgical failure and 49 mice could not be resuscitated. Additional details are described in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Neurologic Scoring of Animals. Neurologic deficits after CA (2, 6, 24, 48, and 72 hr) in mice were determined using a 12-point mouse neurologic scoring system (20). Scores ranged from 0 (no response or worst) to 2 (normal) along six domains: paw pinch, righting reflex, breathing, spontaneous movement, motor-global, and motor-focal. The scores for each of the six domains were determined in a blinded fashion and summed to achieve the neurologic score. Mitochondria Isolation. Mitochondria were obtained from post-CA hearts as previously described (13). Briefly, hearts from Sham and post-CA mice were collected at 15 minutes after CPR, then minced and incubated with trypsin before homogenization with a glass/teflon Potter Elvehjem homogenizer (Fisher Scientific, Hanover Park, IL). Heart homogenates were centrifuged at 800g × 5 minutes at 4°C and the supernatant collected and centrifuged at 8,000g × 5 minutes at 4°C twice to obtain purified cardiac mitochondria. See the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61) for further details.

fic, Hanover Park, IL). Heart homogenates were centrifuged at 800g × 5 minutes at 4°C and the supernatant collected and centrifuged at 8,000g × 5 minutes at 4°C twice to obtain purified cardiac mitochondria. See the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61) for further details. Mitochondrial Permeability Transition Pore Opening Mitochondrial permeability transition pore (mPTP) opening induced by calcium was determined in freshly isolated cardiac mitochondria (13). The absorbance was continuously measured using a Cytation 3 (BioTek, Winooski, VT) 96-well plate reader at 540 nm. Additional details are described in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Complex I Enzyme Activity Complex I activity was measured using an enzyme activity dipstick assay (Abcam, Cambridge, MA) following the manufacturer protocol. In principle, immunocaptured Complex I oxidizes nicotinamide adenine dinucleotide, reduced form (NADH) and the resulting hydrogen ion (H+) reduces nitrotetrazolium blue (NBT) to form a blue-purple precipitate at the complex I antibody line on the dipstick immersed in complex I activity buffer containing NADH (substrate) and NBT (electron acceptor). The signal intensity of this precipitate corresponds to the level of complex I enzyme activity (blue band) in the sample. The intensity was analyzed by using Fiji 6 (National Institutes of Health, Bethesda, MD).

the dipstick immersed in complex I activity buffer containing NADH (substrate) and NBT (electron acceptor). The signal intensity of this precipitate corresponds to the level of complex I enzyme activity (blue band) in the sample. The intensity was analyzed by using Fiji 6 (National Institutes of Health, Bethesda, MD). Superoxide-H2O2 Production in Cardiac Mitochondria To induce H2O2 production from site IQ in cardiac mitochondria, 20-mM glycerol 3-phosphate was added to isolated mitochondria (1 μg/100 μL) in respiration medium with 50-μM Amplex Red and 2-mU/mL horseradish peroxidase (ThermoFisher, Waltham, MA) (16). Fluorescence was monitored using a microplate reader (SpectraMax iD3; Molecular Devices, Sunnyvale, CA) for excitation at 540 nm and emission detection at 590 nm at 37°C after 30 minutes incubation. Seahorse Measurement of Mitochondrial Oxygen Consumption Rates Isolated mitochondria (1 μg/100 μL) from the hearts of Sham and post-CPR mice were suspended in 24-well plates. Oxygen consumption rates (OCRs) were determined using the Seahorse XF24 Extracellular Flux Analyzer (Seahorse Bioscience, Billerica, MA), as previously described (21). Complex I OCR was measured using the substrates 10-mM pyruvate + 2-mM malate. Complex II OCR was measured using the substrate 10-mM succinate and an inhibitor of reverse electron flow, 2-µM rotenone. Additional details are described in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61).

lex I OCR was measured using the substrates 10-mM pyruvate + 2-mM malate. Complex II OCR was measured using the substrate 10-mM succinate and an inhibitor of reverse electron flow, 2-µM rotenone. Additional details are described in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Statistics Comparisons between groups containing normally distributed data were made using analysis of variance with Tukey test or the Student t test. Mann-Whitney U test and Kruskal-Wallis test were applied for nonparametric statistics. The survival curves were compared using a log rank (Mantel Cox) test. Analysis was performed using Prism software (Graph Pad, La Jolla, CA). Data were presented as mean ± sem. Values of p less than 0.05 were considered statistically significant. Supplemental Methods Details regarding mouse echocardiography and different staining methods are provided in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61).

Statistics Comparisons between groups containing normally distributed data were made using analysis of variance with Tukey test or the Student t test. Mann-Whitney U test and Kruskal-Wallis test were applied for nonparametric statistics. The survival curves were compared using a log rank (Mantel Cox) test. Analysis was performed using Prism software (Graph Pad, La Jolla, CA). Data were presented as mean ± sem. Values of p less than 0.05 were considered statistically significant. Supplemental Methods Details regarding mouse echocardiography and different staining methods are provided in the supplemental methods (Supplemental Digital Content 1, http://links.lww.com/CCM/F61). RESULTS CA Duration Determines Post-CPR Myocardial Dysfunction and Resuscitation Outcomes Using our previously established model of induced asystolic CA, we investigated the effects of cardiac duration on resuscitation outcomes (14). Baseline characteristics of the mice and CPR quality were recorded (Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Increasing the durations of CA reduced rates of return to spontaneous circulation (ROSC) and increased the CPR time needed to achieve ROSC (ROSC rate: 100%, 93%, 71%, and 44% in 4, 8, 12, and 16-min group, respectively; time to ROSC: 88 ± 2, 145 ± 17, 189 ± 23, and 227 ± 29 min, respectively; Fig. 1, A and B). Post-CPR myocardial dysfunction was proportional to the duration of CA (fractional shortening at 15 min post-ROSC: 35% ± 2%, 33% ± 1%, 24% ± 3%, 16% ± 1%, and 9% in sham, 4, 8, 12, and 16-min group, respectively; Fig. 1C) and predicted survival 2 hours after ROSC (Fig. 1D). CA duration also correlated strongly with the severity of neurologic injury over 72 hours after ROSC (Fig. 1E). Despite being severely depressed during the first 6 hours following ROSC, myocardial function gradually improved to near-baseline measurements over the following 72 hours (fractional shortening at 72 hr post-ROSC: 41% ± 1% in sham, 39% ± 2% in 12-min group; Fig. 1F). These data are consistent with clinical observations, which show that post-CPR outcomes worsen as a function of CA duration (2, 6) and that post-CPR myocardial dysfunction recovers over time (22).

ver the following 72 hours (fractional shortening at 72 hr post-ROSC: 41% ± 1% in sham, 39% ± 2% in 12-min group; Fig. 1F). These data are consistent with clinical observations, which show that post-CPR outcomes worsen as a function of CA duration (2, 6) and that post-CPR myocardial dysfunction recovers over time (22). Figure 1. Duration of cardiac arrest determines postcardiopulmonary resuscitation (CPR) outcomes. A, Return of spontaneous circulation (ROSC) rates following 4, 8, 12, and 16 min of cardiac arrest (CA). B, Time of CPR to achieve ROSC. n = 12, 15, 14, and 9, respectively. *p < 0.05; **p < 0.01; ***p < 0.001 versus 4-min group. C, Percent left ventricular fractional shortening 15 min after achieving ROSC for different durations of CA. n = 17, 12, 12, 12, and 1, respectively. D, Kaplan-Meyer Curve demonstrating survival following different durations of CA. n = 22, 28, 34, and 9, respectively. E, Neurologic scores following CA of increasing duration. n = 12, respectively. F, Percent left ventricular fractional shortening recovery over time following CA. n = 7, 7, and 9, respectively. *p < 0.05; **p < 0.01; ***p < 0.001 versus sham. #p < 0.05.

ng different durations of CA. n = 22, 28, 34, and 9, respectively. E, Neurologic scores following CA of increasing duration. n = 12, respectively. F, Percent left ventricular fractional shortening recovery over time following CA. n = 7, 7, and 9, respectively. *p < 0.05; **p < 0.01; ***p < 0.001 versus sham. #p < 0.05. Post-CPR Myocardial Dysfunction Is Consistent With Myocardial Stunning Next we sought to determine whether post-CPR myocardial dysfunction was the result of cardiomyocyte cell death. Tetrazolium staining and histologic examination revealed no evidence of myocardial necrosis (Fig. 2, A and B), whereas terminal deoxynucleotidyl transferase dUTP nick-end labeling staining and cluster of differentiation 31 staining showed no evidence of cardiomyocyte apoptosis or endothelial cell loss (Fig. 2, C and D; and Supplemental Fig. 2, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Increased sensitivity to mPTP opening which is associated with myocardial infarction was not observed in mitochondria isolated 15 minutes following ROSC compared with shams (Fig. 3A). However, time-dependent increases in ROS were measured in post-CA tissue (Fig. 3B) and mitochondria (Fig. 3C) compared with shams (Supplemental Fig. 3, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). In addition, complex I activity reduced in post-CA mitochondria compared with shams (Fig. 3D). Together, these observations indicate that post-CA myocardial dysfunction is associated with post-CPR mitochondrial ROS and mitochondrial dysfunction.

tal Fig. 3, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). In addition, complex I activity reduced in post-CA mitochondria compared with shams (Fig. 3D). Together, these observations indicate that post-CA myocardial dysfunction is associated with post-CPR mitochondrial ROS and mitochondrial dysfunction. Figure 2. Postcardiopulmonary resuscitation myocardial dysfunction occurs in the absence of myocardium necrosis. A, Tetrazolium staining of hearts 2 hr following a 12-min cardiac arrest (CA). Hematoxylin and eosin staining (B), terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) staining (C), and cluster of differentiation 31 (CD31) staining (D) of left ventricle sections 2 hr following CA compared with sham. AU = arbitrary units, DAPI = 4′,6-diamidino-2-phenylindole, ROSC = return to spontaneous circulation.

in staining (B), terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) staining (C), and cluster of differentiation 31 (CD31) staining (D) of left ventricle sections 2 hr following CA compared with sham. AU = arbitrary units, DAPI = 4′,6-diamidino-2-phenylindole, ROSC = return to spontaneous circulation. Figure 3. Increased reactive oxygen species (ROS) production and decreased complex I activity postcardiopulmonary resuscitation (CPR) resuscitation. A, Calcium-induced mitochondrial swelling from sham and post-CPR heart. n = 2, 3, 3, and 3, respectively. B, MitoSox staining (ThermoFisher, Waltham, MA) from cardiac arrest (CA) and sham mice heart. Fluorescence quantification is demonstrated in left graphic. n = 4, respectively. C, Fluorescence quantification of MitoSox staining on mitochondria isolated from CA and sham mice with the present of 10-mM pyruvate + 2-mM malate. n = 4, respectively. D, Complex I activity measurement directly from cardiac mitochondria. n = 4, respectively. 12 min CA (CA12) + reperfusion 15 min (R15) = 12-min CA + 15-min resuscitation; *p < 0.05; ***p < 0.001 versus sham. AU = arbitrary units, mPTP = mitochondrial permeability transition pore.

ate + 2-mM malate. n = 4, respectively. D, Complex I activity measurement directly from cardiac mitochondria. n = 4, respectively. 12 min CA (CA12) + reperfusion 15 min (R15) = 12-min CA + 15-min resuscitation; *p < 0.05; ***p < 0.001 versus sham. AU = arbitrary units, mPTP = mitochondrial permeability transition pore. Mitochondrial Injury and Complex I and II Function Following Successful CPR We next measured mitochondrial oxygen consumption in isolated mitochondria from post-CA and sham mice to further characterize post-CA mitochondrial dysfunction. Following the administration of adenosine diphosphate (ADP) to induce mitochondrial respiration, OCRs (Fig. 4, A and B) increased as expected in both post-CPR mitochondria and sham mitochondria (919 ± 55 vs 729 ± 57 pM/min). Paradoxically, these increases were greater in the damaged post-CPR mitochondria than in the sham mitochondria but occurred in the context of increased mitochondrial proton leak (241 ± 16 vs 154 ± 9 pM/min; Fig. 4E), suggesting that ADP-stimulated increases in OCR were reflective of increased ROS production rather than that of ATP production. Further evidence of post-CPR mitochondria damage was the depressed OCR observed upon maximal OCR respiration stimulated by the uncoupler carbonilcyanide p-triflouromethoxyphenylhydrazone (FCCP) (1,547 ± 97 vs 2,127 ± 86 pM/min in sham; Fig. 4D) and decreases in mitochondrial efficiency of oxygen consumption based on state 3/state 4 ratios following CA compared with sham (3.7 ± 0.4 vs 6.3 ± 0.4; Fig. 4C). These results are consistent with the decreased complex I activity measured directly from cardiac mitochondria after CA and further demonstrate the association of mitochondrial injury at complex I following post-CA resuscitation (Fig. 3D). Similar to complex I, experiments designed to measure OCR at complex II found decreases OCR in post-CPR mitochondria stimulated by FCCP compared with shams (1,554 ± 83 vs 2,082 ± 115 pM/min; Supplemental Fig. 4A, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). However, unlike complex I, complex II ADP-dependent OCR decreased compared with sham mitochondria (340 ± 52 vs 985 ± 74 pM/min; Supplemental Fig. 4B, Supplemental Digital Content 1, http://links.lww.com/CCM/F61) with no significant differences in proton leak (998 ± 93 vs 838 ± 76 pM/min; Fig. 4E).

nks.lww.com/CCM/F61). However, unlike complex I, complex II ADP-dependent OCR decreased compared with sham mitochondria (340 ± 52 vs 985 ± 74 pM/min; Supplemental Fig. 4B, Supplemental Digital Content 1, http://links.lww.com/CCM/F61) with no significant differences in proton leak (998 ± 93 vs 838 ± 76 pM/min; Fig. 4E). These experiments demonstrate that mitochondrial injury occurs following CA resuscitation and that complex I injury differs fundamentally from complex II injury suggesting increased ROS production from this site. Figure 4. Postcardiopulmonary resuscitation mitochondrial complex I injury. Oxygen consumption rate (OCR) measurements of cardiac mitochondria from cardiac arrest (CA) and sham. A, The sequential injection of mitochondrial inhibitors is indicated by arrows. B, Adenosine diphosphate (ADP)-stimulated OCR. C, State 3/state 4 respiration. D, Carbonilcyanide p-triflouromethoxyphenylhydrazone-stimulated OCR. E, Calculated proton leak. n = 7, respectively. 12 min CA (CA12) + reperfusion 15 min (R15) = 12-min CA + 15-min resuscitation; *p < 0.05; **p < 0.01; ***p < 0.001 versus sham.

denosine diphosphate (ADP)-stimulated OCR. C, State 3/state 4 respiration. D, Carbonilcyanide p-triflouromethoxyphenylhydrazone-stimulated OCR. E, Calculated proton leak. n = 7, respectively. 12 min CA (CA12) + reperfusion 15 min (R15) = 12-min CA + 15-min resuscitation; *p < 0.05; **p < 0.01; ***p < 0.001 versus sham. Inhibition of Complex I–Specific Superoxide Generation Reduces Myocardial Stunning and Improves Post-CPR Survival Because our results were indicative of increased ROS production from complex I following CA resuscitation, we next investigated whether a site-specific complex I superoxide inhibitor S1QEL would improve post-CPR outcomes. In dose-response trials, we found that a 10 μM of S1QEL was sufficient to inhibit H2O2 production in isolated mitochondria induced by 5-mM succinate at site IQ (Fig. 5A), while having no observable effects on cardiac function, neurological scores, or survival in normal mice (Supplemental Fig. 5A-C, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). We next tested the effects of blinded, randomized administration of S1QEL or phosphate-buffered saline at the initiation of CPR (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Baseline animal characteristics and CPR quality were similar in both groups (Supplemental Table 2, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). S1QEL (10 μM) reduced ROS production 15 minutes post-CA (Fig. 5B) and increased the ROSC rate without altering the CPR time to ROSC (Fig. 5C). S1QEL was associated with improved post-CPR myocardial contractility, neurologic function, and overall survival (fraction shortening at 2-hr post-CPR: 26% ± 2% vs 18% ± 1%; neurologic score at 72-hr post-CPR: 9.5 ± 1.0 vs 4.9 ± 1.4; survival rate at 72-hr post-CPR: 74% vs 30%; Fig. 5D–F). The beneficial effects of S1QEL occurred in a dose-dependent manner (Supplemental Fig. 6, Supplemental Digital Content 1, http://links.lww.com/CCM/F61) although S1QEL did not improve the outcomes following prolonged CA (16-min CA) (Supplemental Fig. 7, Supplemental Digital Content 1, http://links.lww.com/CCM/F61).

–F). The beneficial effects of S1QEL occurred in a dose-dependent manner (Supplemental Fig. 6, Supplemental Digital Content 1, http://links.lww.com/CCM/F61) although S1QEL did not improve the outcomes following prolonged CA (16-min CA) (Supplemental Fig. 7, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). Figure 5. Suppressor of site IQ (the ubiquinone-binding site of complex I, the active site during reverse electron transport) electron leak (S1QEL) reduces postcardiopulmonary resuscitation (CPR) myocardial stunning and improves post-CPR resuscitation outcomes. A, Effects of S1QEL (0.1, 1, and 10 μM) on succinate-induced H2O2 production at site IQ of complex I post cardiac arrest (CA). n = 16, 16, 22, and 16, respectively. ***p < 0.001 versus CA group. B, Images and bar graph show that MitoSox staining in the heart tissue following CPR with and without S1QEL. n = 9, 10, and 8, respectively. *p < 0.05; ***p < 0.001 versus sham. ##p < 0.01 versus CA group. C, Return to spontaneous circulation (ROSC) following 12 min of CA and CPR time to ROSC with S1QEL and controls. n = 53, 39, respectively. D, Left ventricular fractional shortening following 12-min CA with S1QEL and controls. n = 10, 8, respectively. E, Neurologic scores in mice following CA with S1QEL and controls. n = 14, 17, respectively. F, Survival curve following CA with S1QEL and controls. S, S1QEL; n = 53, 39, respectively. *p < 0.05; **p < 0.01; ***p < 0.001 versus CA group. ROS = reactive oxygen species.

A with S1QEL and controls. n = 10, 8, respectively. E, Neurologic scores in mice following CA with S1QEL and controls. n = 14, 17, respectively. F, Survival curve following CA with S1QEL and controls. S, S1QEL; n = 53, 39, respectively. *p < 0.05; **p < 0.01; ***p < 0.001 versus CA group. ROS = reactive oxygen species. DISCUSSION In this study, we have made “three key findings.” First, post-CPR myocardial dysfunction following asystolic CA is due to myocardial stunning rather than myocardial necrosis (Figs. 1 and 2). Although myocardial stunning is typically associated with ventricular wall movement abnormalities following brief coronary occlusion/reperfusion, our study demonstrates that stunning can occur in the context of global cardiac IR injury, which is experienced by patients resuscitated from CA. Myocardial dysfunction following induced asystolic CA has been described previously by our laboratory and others (14, 18), but in this study, we demonstrate for the first time in an asystolic CA model that post-CPR myocardial dysfunction is dependent on the length of arrest, not associated with myocardial necrosis/apoptosis, and is reversible, consistent with myocardial stunning. This stunning is similar to that previously reported in the setting of ventricular fibrillation in other animal models (8, 11). Findings of myocardial stunning described in our study and others are also consistent with reports of early recovery of myocardial function in survivors following CA in several clinical studies (23, 24). Importantly, our study demonstrates that the severity of myocardial stunning is determined by the length of CA which is related to ROSC rates and survival. Stunning is a key determinant of early post-CPR mortality and is clinically relevant. Understanding the pathophysiology of stunning is of great translational relevance in the setting of post-CA resuscitation.

severity of myocardial stunning is determined by the length of CA which is related to ROSC rates and survival. Stunning is a key determinant of early post-CPR mortality and is clinically relevant. Understanding the pathophysiology of stunning is of great translational relevance in the setting of post-CA resuscitation. Second, we discovered that post-CPR myocardial stunning occurs in the context of mitochondrial injury at complex I and II, resulting in a paradoxical increase in oxygen consumption at electron transport chain (ETC) complex I (Fig 4; and Supplemental Fig. 4, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). As expected, the reduced maximal OCR at complex I and II and the decreased complex I activity were observed, supporting the finding of complex I injury following CA (Fig. 3D). Mitochondrial injury following post-CA resuscitation has been reported previously (25), but our unexpected observations of increased OCR with ADP administration and increased proton leak at complex I suggest that complex I could be the site of increased ROS in post-CPR ventricular tissue and mitochondria. These observations are consistent with prior reports of complex I injury associated with increased oxygen consumption and ROS generation after prolonged cardiac ischemia-reperfusion (18, 26, 27). The ROS generated following CPR in our study was not sufficient to generate opening of the mPTP but could be responsible for the observed post-CPR myocardial dysfunction given that superoxide has been demonstrated to reduce myocardial filament contractile activity in vitro in a dose-responsive manner (28).

18, 26, 27). The ROS generated following CPR in our study was not sufficient to generate opening of the mPTP but could be responsible for the observed post-CPR myocardial dysfunction given that superoxide has been demonstrated to reduce myocardial filament contractile activity in vitro in a dose-responsive manner (28). Third, we determined that S1QEL, a site IQ-specific H2O2 production suppressor, limited ROS generation and neurologic injury while improving ROSC rate, myocardial function, and survival following CA (Fig. 5; and Supplemental Fig. 6, Supplemental Digital Content 1, http://links.lww.com/CCM/F61). It is well known that myocardial IR injury increases ROS generation and that targeting complex I–mediated ROS generation during reperfusion has therapeutic utility (9, 17, 18). However, a major limitation of these approaches is that they not only reduce ROS production but also limit electron flow through ETC thus disrupting normal mitochondrial function to a significant degree. Brand et al (16) have shown S1QEL overcomes these limitations and has protective effects against oxidative damage, endoplasmic reticulum stress and IR injury in the isolated perfused heart in a Langendorff preparation. To our knowledge, S1QEL has not been studied previously in vivo in mammals. Here, we show that S1QEL improves post-CPR mitochondria function resulting in reduced ROS generation and improved cardiac, neurologic, and survival outcomes in a mouse CA model. Our work has translational significance because S1QEL was administered at the time of CPR initiation and limited the effects of reperfusion injury following CA. Future research into agents that can be administered to patients by paramedics in the field to limit post-CPR reperfusion injury could represent a major advance in the caring of post-CA patients.

nce because S1QEL was administered at the time of CPR initiation and limited the effects of reperfusion injury following CA. Future research into agents that can be administered to patients by paramedics in the field to limit post-CPR reperfusion injury could represent a major advance in the caring of post-CA patients. Our study has several limitations. First, our study was performed in a murine model of asystolic CA. Although this model has several advantages, including the ability to perform survival outcome studies and cost, our findings on the efficacy of S1QEL on myocardial function could benefit from study in other models of CA. Second, our study was not designed to determine the mechanism of S1QEL’s neuroprotective effects. It is possible that S1QEL could have had direct effects on brain ischemia-reperfusion injury although it is unknown if it is able to cross blood-brain barrier. Additional experiments will be needed to address the effects of S1QEL specifically on post-CPR neurologic injury.

mechanism of S1QEL’s neuroprotective effects. It is possible that S1QEL could have had direct effects on brain ischemia-reperfusion injury although it is unknown if it is able to cross blood-brain barrier. Additional experiments will be needed to address the effects of S1QEL specifically on post-CPR neurologic injury. SUMMARY In conclusion, post-CPR cardiogenic shock reflects ischemia/reperfusion-induced myocardial stunning, the severity of which depends upon the length of cardiac standstill prior to CPR. This stunning can occur following asystolic CA or following arrhythmogenic-induced CA (8, 11). Myocardial stunning is associated with a pattern of mitochondrial injury indicative of increased mitochondrial ROS generation at complex I. Targeting mitochondrial complex I ROS in the setting of post-CPR with specific inhibitors of electron leak (S1QEL) represents a novel, practical strategy to improve post-CPR resuscitation outcomes. ACKNOWLEDGMENTS We thank Dr. Yun Fang for kindly helping us on measuring mitochondrial permeability transition pore opening. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by National Institutes of Health RO1HL133675 (to Dr. Sharp).

Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by National Institutes of Health RO1HL133675 (to Dr. Sharp). Drs. Mutlu, Archer, and Sharp received support for article research from the National Institutes of Health (NIH). Dr. Dezfulian’s institution received funding from Mallinckrodt Pharmaceuticals. Dr. Archer is supported by the Canadian Institutes of Health Research, The Henderson Foundation, and the Queen’s Cardiopulmonary Unit. Dr. Sharp’s institution received funding from NIH RO1HL133675. The remaining authors have disclosed that they do not have any potential conflicts of interest.

ing to standard formulae. RRT-derived parameters were recorded from the RRT monitor including hemofilter pressure drop and transmembrane pressure. Plasma neutrophil gelatinase-associated lipocalin was measured at inclusion using a fluorescence immunoassay (Alere Triage® Neutrophil Gelatinase-Associated Lipocalin Test). Statistical Analyses The primary endpoint was a 20% reduction in Paco2 at 20 minutes after ECCo2R initiation (34). With a mean Paco2 before ECCo2R of 50 ± 10 Torr, a sample of 10 patients would be required to detect a bilateral difference of 5% with a power of 80%. We extended the sample to 11 patients to increase the statistical power of our comparison between positions of the membrane oxygenator within the circuitry. Distribution of the data was tested using a Kolmogorov-Smirnoff test. Categorical variables are expressed as %, and continuous data are expressed as mean ± sd or median and interquartile 25–75% range, as appropriate. Change in Paco2 after initiation of ECCo2R is expressed as % of relative reduction with mean and 95% CI values. Comparisons between data over time were performed using one-way repeated measures analysis of variance (ANOVA) for parametric data or Friedman repeated measures ANOVA on ranks. If overall significance was achieved, the Holm-Sidak test for all pairwise multiple comparisons was applied for parametric data or the Tukey test for nonparametric data. To compare the effect of time and position of the membrane oxygenator, we used two-way repeated measures ANOVA (General Linear Model) with the Holm-Sidak post hoc procedure. Missing data were integrated in the calculation. Correlations were tested for using the Pearson product-moment test. The significance level was fixed at p < 0.05. Data were analyzed using Sigma Stat v3.5 software (Systat, San Jose, CA).

eated measures ANOVA (General Linear Model) with the Holm-Sidak post hoc procedure. Missing data were integrated in the calculation. Correlations were tested for using the Pearson product-moment test. The significance level was fixed at p < 0.05. Data were analyzed using Sigma Stat v3.5 software (Systat, San Jose, CA). RESULTS Twelve combined therapies were conducted among the 11 patients included. In one patient, the hemofilter clotted between 12 and 24 hours, and a new circuit was implemented according to the protocol. The membrane oxygenator was placed upstream of the hemofilter in seven therapies and downstream in five. Of the 12 therapies, seven (58%) were led to the 72nd hour. In one patient, the therapy ended just before the last study measurement due to an alarm of the RRT device; thus, we may consider that two thirds of patients received 72 hours of combined therapy. Three patients died prematurely: two from refractory multiorgan failure and one from ventricular arrhythmia, despite initial improvement. The characteristics of the patients at inclusion are presented in Table 1. The main clinical and biological variables at inclusion are presented in Table 2. TABLE 1. Characteristics of the Study Population (n = 11 Patients) TABLE 2. Pulmonary, Hemodynamic, and Renal Parameters of the Study Population at Inclusion (n = 11 Patients)

Inflammasomes have emerged as important cellular complexes involved in the detection of both pathogen-associated molecular patterns (PAMPs) and endogenous danger-associated molecular patterns (DAMPs) by immune cells (1). Once activated, inflammasomes assemble as a cytoplasmic multiprotein complex to activate caspase-1, which proteolytically converts the precursors of proinflammatory cytokines, pro-interleukin (IL)-1β and pro-IL-18 into active molecules (2). In concert with peripheral immune cell signaling, neural mechanisms orchestrate the inflammatory response via autonomic-immune effector pathways (3). However, the mechanisms underlying detection of PAMPs and DAMPs by the nervous system remain unclear.

es, pro-interleukin (IL)-1β and pro-IL-18 into active molecules (2). In concert with peripheral immune cell signaling, neural mechanisms orchestrate the inflammatory response via autonomic-immune effector pathways (3). However, the mechanisms underlying detection of PAMPs and DAMPs by the nervous system remain unclear. The onset of critical illness is frequently accompanied by tachycardia and tachypnea (4). Although these physiological features are commonplace in the critically ill, they often occur in the absence of systemic hypoxia, hypercapnia and/or acidosis (5, 6), major stimuli for increased peripheral chemoreceptor activity and chemoreceptor-driven cardiovascular and respiratory responses (7), including rapid alterations in heart rate variability (8, 9). Circulating DAMPs, may therefore be an important additional stimulus for increasing peripheral chemoreceptor activity, beyond (and in addition to) previously identified cytokine mediators (10). We therefore hypothesized that chemosensitive glomus cells of the carotid body—a peripheral afferent organ of neural crest origin—are capable of detecting PAMPs/DAMPs by inflammasome activation, to relay immune/inflammatory signals to the central nervous system. We used the yeast cell wall product zymosan, which activates the NLRP3 inflammasome via toll-like receptor-2 (11), to produce a robust model of sterile systemic inflammation in vivo (12, 13) and in cultured cells (12).

DAMPs by inflammasome activation, to relay immune/inflammatory signals to the central nervous system. We used the yeast cell wall product zymosan, which activates the NLRP3 inflammasome via toll-like receptor-2 (11), to produce a robust model of sterile systemic inflammation in vivo (12, 13) and in cultured cells (12). MATERIALS AND METHODS All experiments were performed in accordance with the United Kingdom Animals (Scientific Procedures) Act (1986). All drugs were acquired from Sigma (Poole, UK) unless stated otherwise.

DAMPs by inflammasome activation, to relay immune/inflammatory signals to the central nervous system. We used the yeast cell wall product zymosan, which activates the NLRP3 inflammasome via toll-like receptor-2 (11), to produce a robust model of sterile systemic inflammation in vivo (12, 13) and in cultured cells (12). MATERIALS AND METHODS All experiments were performed in accordance with the United Kingdom Animals (Scientific Procedures) Act (1986). All drugs were acquired from Sigma (Poole, UK) unless stated otherwise. Carotid body Cell Cultures Sprague-Dawley rat pups (7–11 days old; local breeding colony at University College London) were terminally anesthetized with isoflurane prior to bilateral excision of the carotid bodies (14). Chemosensitive glomus cells were enzymatically isolated by exposing the carotid body tissue to a 0.1% trypsin-0.1% collagenase solution for 1hr at 37°C followed by mechanical dissociation. The dispersed cell suspension was then collected and triturated in growth medium (F-12 nutrient medium supplemented with 10% (v/v) fetal bovine serum, 80U/L insulin, 0–6% (w/v) glucose, 2mM glutamine and 1% penicillin-streptomycin). Dispersed cells were then placed onto poly-L-lysine coated cover slips. Cultures were grown at 37°C in a humidified atmosphere of 95% air/5% CO2 for 72 hours. Each cell culture contained multiple cell clusters containing approximately 5–30 chemosensitive glomus cells that were easily identified under bright field microscopy. To activate inflammasome-dependent signaling pathways, cultures were incubated in a growth medium containing 0, 0.5, or 5 μg/mL of zymosan for 24 hours.

Each cell culture contained multiple cell clusters containing approximately 5–30 chemosensitive glomus cells that were easily identified under bright field microscopy. To activate inflammasome-dependent signaling pathways, cultures were incubated in a growth medium containing 0, 0.5, or 5 μg/mL of zymosan for 24 hours. Primary Human Neutrophils Primary human neutrophils (>97% purity) were isolated as described previously (12) and incubated with lipopolysaccharide (10ng/mL; Escherichia coli B0111:B4) or phosphate-buffered saline for 1.5 hours prior to cell fixation. Lipopolysaccharide was used since previous studies have shown that in primary human neutrophils the NLRP3 inflammasome is not activated by zymosan (15).

described previously (12) and incubated with lipopolysaccharide (10ng/mL; Escherichia coli B0111:B4) or phosphate-buffered saline for 1.5 hours prior to cell fixation. Lipopolysaccharide was used since previous studies have shown that in primary human neutrophils the NLRP3 inflammasome is not activated by zymosan (15). Immunofluorescence Studies Cell cultures were fixed with 4% paraformaldehyde solution and permeabilized with 0.025% Triton X-100. Blocking solution contained 1% bovine serum albumin and 10% of either goat or donkey normal serum. Nuclear structures were stained with 4′,6-diamidino-2-phenylindole (DAPI). Tyrosine hydroxylase (TH) immunostaining was performed routinely to confirm the glomus cell phenotype. Cultures were simultaneously incubated with sheep anti-TH antibody (1:250 Abcam, Cambridge, UK) with the addition of one of the following antibodies: goat anti-TLR4 (1:500; Santa Cruz, Heidelberg, Germany), rabbit anti-TLR2 (1:250; Santa Cruz), goat anti-IL-1β (epitope mapping at the C-terminus (16, 17); 1:500; Santa Cruz); goat anti-IL-1ra/IL-1F3 (1:250; R & D Systems, Abingdon, UK); rabbit anti-NLRP1 (1:500; Abcam) and goat anti-NLRP3 (1:500, Abcam). Negative controls were performed by the omission of primary or secondary antibodies.

R2 (1:250; Santa Cruz), goat anti-IL-1β (epitope mapping at the C-terminus (16, 17); 1:500; Santa Cruz); goat anti-IL-1ra/IL-1F3 (1:250; R & D Systems, Abingdon, UK); rabbit anti-NLRP1 (1:500; Abcam) and goat anti-NLRP3 (1:500, Abcam). Negative controls were performed by the omission of primary or secondary antibodies. Confocal Microscopy Imaging Images were captured using a Confocal Laser Scanning Microscope, (Zeiss 510, Welwyn Garden City, UK) with 40X or 60X oil-immersion objectives. UV, argon (458, 477, 488, 504nm), or helium-neon (543nm, 633nm) lasers were used as appropriate. All images were captured using the same settings to enable comparisons of fluorescence intensities between different cell cultures. Intensity of fluorescence was estimated using Zeiss LSM 510 software and expressed in arbitrary units (AU). Intracellular Ca2+ Imaging Glomus cells were loaded for 40min at room temperature with Fluo-4 AM (4 μM) or Fura-2 AM (4μM) to determine changes in cytosolic [Ca2+] in response to hypoxia or IL-1β application. Model of Systemic Inflammation C57BL/6 inbred mice (~20g; Charles River, UK) received intraperitoneal injections of zymosan (500mg kg-1; n = 12) (12) to induce systemic inflammation or sterile saline (0.2mL; controls; n = 12). Severity scores were recorded as described previously (18).

Intracellular Ca2+ Imaging Glomus cells were loaded for 40min at room temperature with Fluo-4 AM (4 μM) or Fura-2 AM (4μM) to determine changes in cytosolic [Ca2+] in response to hypoxia or IL-1β application. Model of Systemic Inflammation C57BL/6 inbred mice (~20g; Charles River, UK) received intraperitoneal injections of zymosan (500mg kg-1; n = 12) (12) to induce systemic inflammation or sterile saline (0.2mL; controls; n = 12). Severity scores were recorded as described previously (18). Whole-body Plethysmography Ventilation was measured in conscious, freely moving C57BL/6 mice (Charles River, UK) injected with either zymosan or sterile saline (n = 5 per group) immediately before being placed in a whole-body plethysmograph, as described previously (13, 18, 19). The recording chamber (250mL) was flushed continuously with a humidified mixture of 79% nitrogen and 21% oxygen, at a rate of 500mL/min with temperature maintained at 22°C. Levels of oxygen and carbon dioxide were monitored using a fast-response gas analyzer (Morgan Medical, Hertford, UK). Tidal volume was determined following calibration of the plethysmograph through repeated injections and withdrawal of different volumes of air from within the recording chamber. The mice were left to acclimatize to the chamber environment for at least 30minutes before measurements of baseline ventilation were taken. Minute ventilation was calculated from the product of respiratory frequency and tidal volume measurements.

withdrawal of different volumes of air from within the recording chamber. The mice were left to acclimatize to the chamber environment for at least 30minutes before measurements of baseline ventilation were taken. Minute ventilation was calculated from the product of respiratory frequency and tidal volume measurements. Carotid Sinus Nerve Recordings Three hours after intraperitoneal injections of zymosan or sterile saline, mice were euthanized under 5% isoflurane anesthesia. The carotid bifurcation region was rapidly removed, placed into a recording chamber (~3mL), carotid sinus nerve (CSN) was isolated, de-sheathed, and recordings were made using a suction electrode, as described previously (14, 19, 20). Nerve activity signal was amplified (×20,000) and filtered (200–3,000 Hz), with data acquired and stored using Spike2 software (Cambridge Electronic Design Ltd, Cambridge, UK). Single-unit analysis was performed offline using the spike-sorting function of Spike2 software (18). CSN chemoafferent activity was recorded during baseline normoxia and hypoxia, produced by exposure of the preparations to Krebs’ solution saturated with 95% N2/5% CO2 for 3 minutes. Peak CSN activity was estimated at the end of each experiment by application of sodium cyanide (0.03% w/v; 50 μL bolus).

e2 software (18). CSN chemoafferent activity was recorded during baseline normoxia and hypoxia, produced by exposure of the preparations to Krebs’ solution saturated with 95% N2/5% CO2 for 3 minutes. Peak CSN activity was estimated at the end of each experiment by application of sodium cyanide (0.03% w/v; 50 μL bolus). Cytokine Measurements Cytokine levels in plasma were determined using an enzyme-linked immunosorbent assay kit (ProteoPlex Murine CytokineArray, Novagen, Germany), according to manufacturer’s instructions. Arterial blood samples were drawn from the carotid artery 3 hours after injection of zymosan. Plasma was separated by centrifugation and stored at −80°C until assayed. Statistical Analysis The data were analyzed using GraphPad Prism 5 software. Statistical significance was tested by Student t test for paired or unpaired data (two groups) or by repeated measures ANOVA (multiple groups) followed by Tukey-Kramer testing for multiple comparisons between groups. Mean values (±se) are presented, unless stated otherwise. Significance was accepted at p values of less than 0.05.

significance was tested by Student t test for paired or unpaired data (two groups) or by repeated measures ANOVA (multiple groups) followed by Tukey-Kramer testing for multiple comparisons between groups. Mean values (±se) are presented, unless stated otherwise. Significance was accepted at p values of less than 0.05. RESULTS Expression of Inflammasome Complexes in Cultured Chemosensitive Carotid Body Glomus Cells We investigated the presence and changes in the expression level of inflammasomes in cultured chemosensitive glomus cells following incubation with zymosan. Carotid body glomus cells (Fig. 1A and B) were found to display TLR-2 (Fig. 1C), TLR-4 (Fig. 1D), NLRP1 (Fig. 1E) and NLRP3 (Fig. 1G) inflammasome immunoreactivities. NLRP1 staining was found to be dispersed throughout the cell body (Fig. 1E). As reported previously (15), distribution of NLRP3 protein could be different, and in glomus cells being concentrated around the edge of the cell (Fig. 1G). NLRP3 expression by glomus cells was up-regulated (Fig. 1H, L; p = 0.023) following incubation with zymosan, whereas expression of NLRP1 was unaffected (Fig. 1F and L). Incubation of primary human neutrophils with lipopolysaccharide induced similar in magnitude increases in the expression of both NLRP1 and NLRP3 inflammasomes (Fig. 1I–K).

glomus cells was up-regulated (Fig. 1H, L; p = 0.023) following incubation with zymosan, whereas expression of NLRP1 was unaffected (Fig. 1F and L). Incubation of primary human neutrophils with lipopolysaccharide induced similar in magnitude increases in the expression of both NLRP1 and NLRP3 inflammasomes (Fig. 1I–K). Figure 1. Expression of inflammasome complexes in cultured carotid body glomus cells. A and B, Representative confocal images showing clusters of the carotid body glomus cells stained with 4′,6-diamidino-2-phenylindole and identified on the basis of tyrosine hydroxylase immunoreactivity. C and D, Toll-like receptor (TLR2 and TLR4) immunoreactivities in glomus cells. E and F, NLRP1 immunoreactivity in glomus cells following incubation in control medium (C) or in the presence of 500 ng/mL of zymosan (Z). G and H, NLRP3 immunoreactivity in glomus cells following incubation in control medium (C) or in the presence of 500 ng/mL of zymosan (Z). Note increased immunofluorescence levels in cultures incubated with zymosan. I and J, NLRP1 and NLRP3 immunoreactivities in primary human neutrophils following incubation in control medium (C) or in the presence of 10 ng/mL of lipopolysaccharide (L). K, Summary data showing significant increases in NLRP1 and NLRP3 immunofluorescence intensities in primary human neutrophils in response to lipopolysaccharide (L). L, Summary data showing significant increase in NLRP3 immunofluorescence intensity in glomus cells in response to zymosan (Z). Data are presented as means ± SE. *Significant difference (p < 0.05).

eases in NLRP1 and NLRP3 immunofluorescence intensities in primary human neutrophils in response to lipopolysaccharide (L). L, Summary data showing significant increase in NLRP3 immunofluorescence intensity in glomus cells in response to zymosan (Z). Data are presented as means ± SE. *Significant difference (p < 0.05). Effect of Zymosan on IL-1β Expression in Chemosensitive Carotid Body Glomus Cells Incubation of glomus cells in the presence of zymosan (0.5 or 5 µg/mL) resulted in a profound up-regulation of IL-1β expression (Fig. 2A and B; p < 0.0001). We also confirmed the presence of IL-1 receptor immunoreactivity in glomus cells (Fig. 2C) suggesting that IL-1β produced by these cells may act in an autocrine manner. Figure 2. Effects of zymosan on interleukin (IL)-1β expression in glomus cells of the carotid body. A, IL-1β immunoreactivity in glomus cells following incubation in control medium (C) or in the presence of 0.5 (Z 0.5) and 5 µg/mL (Z 5) of zymosan. B, Summary data showing significant increases in IL-1β immunofluorescence intensity in glomus cells in response to zymosan. C, IL-1 receptor immunoreactivity in glomus cells. Data are presented as means ± SE. * Significant difference (p < 0.05).

medium (C) or in the presence of 0.5 (Z 0.5) and 5 µg/mL (Z 5) of zymosan. B, Summary data showing significant increases in IL-1β immunofluorescence intensity in glomus cells in response to zymosan. C, IL-1 receptor immunoreactivity in glomus cells. Data are presented as means ± SE. * Significant difference (p < 0.05). Effect of IL-1β on [Ca2+]i in Chemosensitive Carotid Body Glomus Cells Next, we confirmed that IL-1β is capable of eliciting a physiological response in the carotid body glomus cells. As expected, hypoxia activated glomus cells, as characterized by immediate high amplitude increases in [Ca2+]i with a rapid return to basal levels shortly after reoxygenation (Fig. 3A and B). In conditions of normoxia, glomus cells were activated in the presence of IL-1β (n = 23 cells; three independent experiments; Fig. 3C and D). Application of IL-1β induced a delayed [Ca2+]i response in glomus cells, which was observed ~10 minutes after cytokine application and was characterized by sustained high-amplitude oscillations in [Ca2+]i. IL-1β-induced Ca2+ responses in glomus cells were abolished in the presence of IL-1 receptor antagonist in the incubation media (Fig. 3E).

uced a delayed [Ca2+]i response in glomus cells, which was observed ~10 minutes after cytokine application and was characterized by sustained high-amplitude oscillations in [Ca2+]i. IL-1β-induced Ca2+ responses in glomus cells were abolished in the presence of IL-1 receptor antagonist in the incubation media (Fig. 3E). Figure 3. Effects of hypoxia and IL-1β on [Ca2+]i in glomus cells of the carotid body. A, Clusters of glomus cells loaded with the fluorescent Ca2+ indicator Fluo-4. Fluorescent images were obtained before (left panel) and at the peak of [Ca2+]i response to hypoxia (right panel). B, Representative trace showing changes in [Ca2+]i elicited in an individual glomus cell in response to hypoxia. C, Representative trace showing changes in [Ca2+]i elicited in an individual glomus cell in response to interleukin (IL)-1β. D. IL-1β-induced [Ca2+]i responses in glomus cells are abolished in the presence of IL-1-receptor antagonist.

wing changes in [Ca2+]i elicited in an individual glomus cell in response to hypoxia. C, Representative trace showing changes in [Ca2+]i elicited in an individual glomus cell in response to interleukin (IL)-1β. D. IL-1β-induced [Ca2+]i responses in glomus cells are abolished in the presence of IL-1-receptor antagonist. Carotid Body Chemoafferent Activity During Systemic Inflammation Since IL-1β actions on glomus cells of the carotid body mimicked the effect of hypoxia, we sought in vivo evidence that systemic inflammation augments carotid body chemosensitivity and increases ventilation. Zymosan-treated mice showed classical clinical signs of systemic inflammation, corroborated by high cytokine levels in plasma (Table 1). Arterial blood oxygen (zymosan-treated: 9.1±1.6 kPa; control: 10.0±1.7 kPa) and carbon dioxide tensions (zymosan: PaCO2 3.5 kPa; control: 3.6 kPa) were similar (n = 5 each group). Developing of systemic inflammation following administration of zymosan (n = 5) was accompanied by significant increases in respiratory frequency and hence minute ventilation 2 hours after the injections (p < 0.05; Fig. 4). Core temperature (34.9±1.1°C) was lower in mice injected with zymosan (13), compared with mice that received sterile saline (36.2±0.9°C; p < 0.05). TABLE 1. Plasma Cytokine Levels in Mice Injected With Zymosan (500 mg/kg Interperitoneal)

Carotid Body Chemoafferent Activity During Systemic Inflammation Since IL-1β actions on glomus cells of the carotid body mimicked the effect of hypoxia, we sought in vivo evidence that systemic inflammation augments carotid body chemosensitivity and increases ventilation. Zymosan-treated mice showed classical clinical signs of systemic inflammation, corroborated by high cytokine levels in plasma (Table 1). Arterial blood oxygen (zymosan-treated: 9.1±1.6 kPa; control: 10.0±1.7 kPa) and carbon dioxide tensions (zymosan: PaCO2 3.5 kPa; control: 3.6 kPa) were similar (n = 5 each group). Developing of systemic inflammation following administration of zymosan (n = 5) was accompanied by significant increases in respiratory frequency and hence minute ventilation 2 hours after the injections (p < 0.05; Fig. 4). Core temperature (34.9±1.1°C) was lower in mice injected with zymosan (13), compared with mice that received sterile saline (36.2±0.9°C; p < 0.05). TABLE 1. Plasma Cytokine Levels in Mice Injected With Zymosan (500 mg/kg Interperitoneal) Figure 4. Zymosan-induced systemic inflammation is accompanied by increased ventilatory activity in mice. Group data illustrating changes in the respiratory rate (A), tidal volume (B), and minute ventilation (C) in conscious mice up to 2 hr after intraperitoneal injections of saline or zymosan (500 mg/kg; n = 5 per group). Data are presented as means ± SE. * Significant difference (p < 0.05).

ventilatory activity in mice. Group data illustrating changes in the respiratory rate (A), tidal volume (B), and minute ventilation (C) in conscious mice up to 2 hr after intraperitoneal injections of saline or zymosan (500 mg/kg; n = 5 per group). Data are presented as means ± SE. * Significant difference (p < 0.05). To determine whether zymosan-induced systemic inflammation alters peripheral chemoreceptor activity, we recorded CSN activity in a superfused in vitro carotid body/CSN preparation free of possible confounding factors present in vivo such as metabolic acidosis and circulating inflammatory mediators. Representative examples of chemoafferent activity recorded in the carotid body/CSN preparations taken from the animals injected with saline (control) and zymosan are shown in Fig. 5A. Resting CSN discharge frequency was increased five-fold in preparations taken from zymosan-treated mice (p = 0.01; Fig. 5A, B). Chemoafferent responses to hypoxia (Fig. 5A and C) or application of cyanide (Fig. 4D) were also significantly enhanced in the carotid body/CSN preparations of mice injected with zymosan.

5A. Resting CSN discharge frequency was increased five-fold in preparations taken from zymosan-treated mice (p = 0.01; Fig. 5A, B). Chemoafferent responses to hypoxia (Fig. 5A and C) or application of cyanide (Fig. 4D) were also significantly enhanced in the carotid body/CSN preparations of mice injected with zymosan. Figure 5. Zymosan-induced systemic inflammation is accompanied by increased carotid body chemoafferent activity. A, Raw data illustrating changes in the carotid sinus nerve chemoafferent activity recorded in the in vitro carotid body/carotid sinus nerve preparations taken 3 hr after intraperitoneal injections of saline (control) or zymosan (500mg/kg). B–D, Population data [n = 7 mice per group] demonstrating that the development of systemic inflammation in response to zymosan leads to higher baseline (A), hypoxia (B) and cyanide induced (C) chemoafferent discharge. C and D, Data are shown as absolute increases from the baseline level of activity. Data are presented as means ± se. *Significant difference (p < 0.05). CSN chemoafferent activity is composed of individual spikes with distinguishable amplitude and shape (19). This allows analysis and comparison of single-unit chemoafferent activity between control and zymosan-treated mice. Although mean discharge frequency of individual CSN fibers increased only in control mice during hypoxia (Fig. 6A), the overall higher CSN discharge in mice treated with zymosan was attributable to the recruitment of more individual fibers both at baseline conditions and during hypoxia (p = 0.0001; Fig. 6B).

ated mice. Although mean discharge frequency of individual CSN fibers increased only in control mice during hypoxia (Fig. 6A), the overall higher CSN discharge in mice treated with zymosan was attributable to the recruitment of more individual fibers both at baseline conditions and during hypoxia (p = 0.0001; Fig. 6B). Figure 6. Single-unit analysis of carotid sinus nerve discharge. A, Discharge frequency of individual fibers only increases in controls. Data are presented as mean ± SE. * Significant difference (p < 0.01; [n = 7 mice per group]). B, More single units are recruited in preparations taken from zymosan-treated mice during normoxic (* p = 0.0001) and hypoxic (# p = 0.0001) conditions (n = 7 mice per group).

y of individual fibers only increases in controls. Data are presented as mean ± SE. * Significant difference (p < 0.01; [n = 7 mice per group]). B, More single units are recruited in preparations taken from zymosan-treated mice during normoxic (* p = 0.0001) and hypoxic (# p = 0.0001) conditions (n = 7 mice per group). DISCUSSION Chemosensitive glomus cells of the carotid body express both NLRP1 and NLRP3 inflammasomes. When stimulated with zymosan, glomus cells up-regulate the expression of NLRP3, suggesting that peripheral chemoreceptors of neural crest origin utilize the same mechanism as innate immune cells to detect PAMPs and, potentially, DAMPs. The data obtained also confirm the results of previous studies showing that, in addition to sensing arterial levels of PO2, PCO2, and pH, peripheral chemoreceptors are capable of detecting other modalities, including glucose concentration, various electrolytes, neurohormones, and inflammatory mediators (21), many of which are released and/or their levels are altered during systemic inflammation. Indeed, several groups have reported that various proinflammatory cytokines, including TNFα, impact on carotid body chemoreceptor function. Interestingly, experimental chronic intermittent hypoxia itself increases expression of proinflammatory cytokines within the carotid body, perhaps through oxidative stress (22). The generation of reactive oxygen species is crucial in regulating inflammasome gene expression (23). Here we extend these concepts by demonstrating that inflammasome complexes that detect an array of DAMPs regulate the expression of IL-1β by the chemosensitive carotid body glomus cells (24).

ugh oxidative stress (22). The generation of reactive oxygen species is crucial in regulating inflammasome gene expression (23). Here we extend these concepts by demonstrating that inflammasome complexes that detect an array of DAMPs regulate the expression of IL-1β by the chemosensitive carotid body glomus cells (24). We have also confirmed recently reported data suggesting that IL-1β is a cytokine capable of altering carotid body function (25), but have extended this observation by demonstrating one of the possible sources of IL-1β as the glomus cells themselves following activation of TLR-2 and inflammasome complexes. Critically, the effects of TLR-2 receptor activation by zymosan were observed in an immune cell-free, aseptic carotid body cell culture free of many confounding factors such as hypoxia or inflammatory mediators. In contrast to the reported depressant effects of TNFα on the carotid body chemosensory function (26), IL-1β (given in pathophysiologically relevant concentrations) was found to trigger [Ca2+]i oscillations in glomus cells, similar in magnitude to the responses elicited by hypoxia. IL-1β is known to induce expression of the hypoxia-inducible factor(s), which are essential for the maintenance of normal carotid body activity during hypoxia (27). Therefore, IL-1β appears to mimic the responses of the carotid body to hypoxia and may, therefore, act in an autocrine manner to enhance the peripheral chemoreceptor drive during systemic inflammation.

oxia-inducible factor(s), which are essential for the maintenance of normal carotid body activity during hypoxia (27). Therefore, IL-1β appears to mimic the responses of the carotid body to hypoxia and may, therefore, act in an autocrine manner to enhance the peripheral chemoreceptor drive during systemic inflammation. What are the possible implications of inflammatory mediator–dependent alterations in the peripheral chemosensory function on autonomic and neurohormonal control? In this study we show that peripheral chemoafferent activity in normoxia and in conditions of hypoxia is markedly enhanced during early systemic exposure to inflammatory stimuli. Consistent with these data, chronic (proinflammatory) bleomycin-induced lung injury is accompanied by higher baseline respiratory rate and augmented hypoxic ventilatory response (28). The marked increase in the carotid body chemoafferent activity during early systemic inflammation implies that peripheral chemoreceptors may play a key role in the pathophysiology of early sepsis, perhaps by affecting several homeokinetic mechanisms that underlie biological variability. Carotid sinus denervation is associated with shorter survival time in rats administered lethal doses of endotoxin (29), although the impact of concomitant baro-denervation and consequent disruption of baroreflex mechanisms considerably complicate interpretation of any possible protective role of peripheral chemoreceptors.

otid sinus denervation is associated with shorter survival time in rats administered lethal doses of endotoxin (29), although the impact of concomitant baro-denervation and consequent disruption of baroreflex mechanisms considerably complicate interpretation of any possible protective role of peripheral chemoreceptors. Clinical data have revealed an association between peripheral chemoreflex dysfunction and outcome from critical illness (30, 31). It is striking, however, that an array of the commonest respiratory, cardiovascular, endocrine, and renal responses observed in critically ill patients can also be produced by discrete activation of the carotid chemoreflex (7). In addition to tachypnea and an increased depth of breathing, peripheral chemoreceptor stimulation results in increased airway resistance and secretions (32). Release of neurohormones, including cortisol and vasopressin, is also increased in response to hypoxemia and carotid sinus activation (5). Clinically, such alterations in autonomic neural control are detected by profound changes in heart rate variability (4, 8).

sults in increased airway resistance and secretions (32). Release of neurohormones, including cortisol and vasopressin, is also increased in response to hypoxemia and carotid sinus activation (5). Clinically, such alterations in autonomic neural control are detected by profound changes in heart rate variability (4, 8). CONCLUSION Perturbations of autonomic control mechanisms are likely to play a key role at the onset of critical illness (8). The early detection of blood-borne DAMPs and PAMPs by the carotid chemoreceptors and the consequent inflammasome-dependent cytokine-induced activation of the peripheral chemoreflex are likely to play a pivotal role in triggering autonomic dysregulation. Understanding how afferent function is altered during systemic inflammation and sepsis may provide novel therapeutic opportunities and/or monitoring modalities (33) to explore. This work was undertaken at laboratories in University College London. Supported, in part, by Academy Medical Sciences/Health Foundation Clinician, Scientist Award (Dr. Ackland), UK Intensive Care Society Young Investigator award (Dr. Ackland), and The Wellcome Trust (Dr. Gourine is a Wellcome Trust Senior Research Fellow). Part of this work was undertaken at UCLH/UCL that received a proportion of funding from the UK Department of Health’s NIHR Biomedical Research Centre’s funding scheme. Drs. Ackland and Gourine are consultants for Baxter Healthcare.

Supported, in part, by Academy Medical Sciences/Health Foundation Clinician, Scientist Award (Dr. Ackland), UK Intensive Care Society Young Investigator award (Dr. Ackland), and The Wellcome Trust (Dr. Gourine is a Wellcome Trust Senior Research Fellow). Part of this work was undertaken at UCLH/UCL that received a proportion of funding from the UK Department of Health’s NIHR Biomedical Research Centre’s funding scheme. Drs. Ackland and Gourine are consultants for Baxter Healthcare. Drs. Ackland, Kaymov, Marina, Singerm, and Gourine received funding from the Academy of Medical Sciences/Health Foundation Clinician Scientist aware and the National Institute for Health Research UK.

Acute respiratory distress syndrome (ARDS) is a common condition in ICUs and results from various etiologies that cause either direct or indirect lung injury (1, 2). It is characterized by a proinflammatory response leading to pulmonary edema, surfactant dysfunction, and alteration of the alveolocapillary barrier (3). There is a wide heterogeneity in the distribution of alveolar injury. Collapsed or fluid-filled lung areas predominate in dependent regions and normally aerated areas in nondependent, the net result being a loss of lung air volume (4, 5).

dema, surfactant dysfunction, and alteration of the alveolocapillary barrier (3). There is a wide heterogeneity in the distribution of alveolar injury. Collapsed or fluid-filled lung areas predominate in dependent regions and normally aerated areas in nondependent, the net result being a loss of lung air volume (4, 5). Positive pressure mechanical ventilation facilitates gas exchange and unloads respiratory muscles. During insufflation, tidal volume (TV) distributes primarily to normally aerated alveoli with a risk of overdistension, whereas previously collapsed alveoli are submitted to cyclic opening and closing phenomenon (6–8). Both may lead to excessive strain and stress resulting in stretch-induced mechanical transcription of proinflammatory signals, mechanical disruption of the alveolar barrier, and capillary stress failure (9, 10). Reducing the TV from 12 to 6 mL/kg predicted body weight (PBW) in human ARDS was associated with a 22% relative reduction in mortality (11). However, there is a growing body of evidences that a further reduction in the mechanical stress produced by the ventilator may be further “lung protective” (12–17). Animal studies have demonstrated that a TV of 3–4 mL/kg reduces lung edema and preserves, at least in part, alveolar epithelial and endothelial integrity (18–20). Few studies have been conducted in patients and all conclude toward a reduction in the proinflammatory response, both at the pulmonary and at the plasma level (21, 22).

ies have demonstrated that a TV of 3–4 mL/kg reduces lung edema and preserves, at least in part, alveolar epithelial and endothelial integrity (18–20). Few studies have been conducted in patients and all conclude toward a reduction in the proinflammatory response, both at the pulmonary and at the plasma level (21, 22). Decreasing TV reduces alveolar ventilation, providing that respiratory rate (RR) is constant. Thus, Paco2 increases and promotes respiratory acidosis, which may constitute a limitation to implement widely a lung protective ventilation strategy (23). Extracorporeal Co2 removal (ECCo2R), a technique described more than 40 years ago, appears as a convenient solution to prevent hypercarbia and opens up the field to a wide range of clinical studies (24–27). Briefly, blood is driven, either passively from the arterial side or actively with a pump from the venous side, toward a membrane oxygenator where fresh gas, free of Co2, sweeps hollow fibers and removes Co2 by diffusion. Several devices have been designed and some have been tested in clinical settings. Among them, the Decap system (Hemodec, Salerno, Italy) uses a roller pump to drive blood first through a membrane oxygenator and then through a hemofilter. The effluent is reinfused upstream of the oxygenator, which not only increases the overall Co2 removal capacity but also prevents from any blood purification. To date, no device combining blood purification and Co2 removal capacities is available.

blood first through a membrane oxygenator and then through a hemofilter. The effluent is reinfused upstream of the oxygenator, which not only increases the overall Co2 removal capacity but also prevents from any blood purification. To date, no device combining blood purification and Co2 removal capacities is available. Acute kidney injury (AKI) may develop in 25–60% of patients with ARDS, especially when sepsis is the underlying disease (28, 29). The Kidney Disease Improving Global Outcome has recently proposed a classification of the severity of renal injury and suggested renal replacement therapy (RRT) be initiated when stage 2 or 3 occurs (30). Because RRT is common practice in ICUs and because some patients may present together ARDS and AKI (31), we hypothesized that we could combine RRT and ECCo2R through the integration of a membrane oxygenator within a hemofiltration circuit. This study was designed in two parts: first, to test the efficacy of ECCo2R at a fixed TV (6 mL/kg); and second, to test whether a ventilation at lower TV (4 mL/kg) for a period of 72 hours is safe and feasible. Finally, by testing two configuration of circuitry, we also addressed the issue of the most efficacious position of the membrane oxygenator. MATERIALS AND METHODS The protocol of this study was approved by an independent ethics committee (Comité de Protection des Personnes Sud-Méditérranée I) and by French Health Authority (AFSSAPS, 2010-A00397-32) and was registered at clinical.trials.gov (NCT 01239966). Written informed consent was obtained from next of kin before enrollment.

The protocol of this study was approved by an independent ethics committee (Comité de Protection des Personnes Sud-Méditérranée I) and by French Health Authority (AFSSAPS, 2010-A00397-32) and was registered at clinical.trials.gov (NCT 01239966). Written informed consent was obtained from next of kin before enrollment. From December 2011 to December 2014, 11 consecutive patients presenting with both ARDS (Berlin criteria) and AKI (Kidney Disease Improving Global Outcome stage 2 or 3) who required RRT were included (30, 32). For the detailed inclusion and exclusion criteria, see the supplemental data (Supplemental Digital Content 1, http://links.lww.com/CCM/B420). Patients were sedated with remifentanil, midazolam, and ketamine and paralyzed with cisatracurium. Others medications, including antibiotics, fluids, catecholamines, and transfusions, were left to the discretion of the attending physician. All patients were equipped with a 4-Lumen subclavian central venous catheter and an indwelling radial artery catheter (20G × 5 cm). The arterial line was connected to a continuous uncalibrated cardiac output monitor (Flotrac-Vigileo; Edwards, Irvine, CA). Central venous pressure and a five-lead ECG were continuously displayed on a M540 monitor (Dräger, Lubeck, Germany).

central venous catheter and an indwelling radial artery catheter (20G × 5 cm). The arterial line was connected to a continuous uncalibrated cardiac output monitor (Flotrac-Vigileo; Edwards, Irvine, CA). Central venous pressure and a five-lead ECG were continuously displayed on a M540 monitor (Dräger, Lubeck, Germany). Volume-controlled mechanical ventilation was delivered using an Engström CareStation ventilator (General Electric, Madison, WI) including a metabolic monitor that measured end-tidal CO2 (Petco2), oxygen consumption (Vo2), and Co2 elimination by the lung during 1 minute (Vco2Lung). Inspiratory gases were heated and humidified with a MR 850 device (Fischer&Paykel, Australia). A closed suction system allowed tracheobronchial aspiration. Respiratory settings included a TV of 6 mL/kg PBW, a positive end-expiratory pressure (PEEP)/Fio2 combination according to the ARDS Net protocol, an RR adjusted to maintain Petco2 less than or equal to 45 Torr, an end-inspiratory pause of 0.4 second, and an inspiratory flow to obtain complete exhalation. RRT was delivered with a PrismaFlex v6.0 monitor (Gambro, Lund, Sweden) using a 1.5 m2 AN69 membrane (M150; Hospal, Meyzieu, France). The RRT mode was continuous venovenous hemofiltration. Pump blood flow was progressively increased until its maximum value: 450 mL/min minus the predilution rate. Effluent flow was 45 mL/kg/h with a predilution rate of 33%, a setting designed to reduce the risk of clotting (33). The replacement solution was Hemosol B0 (Gambro, Lund, Sweden).

uous venovenous hemofiltration. Pump blood flow was progressively increased until its maximum value: 450 mL/min minus the predilution rate. Effluent flow was 45 mL/kg/h with a predilution rate of 33%, a setting designed to reduce the risk of clotting (33). The replacement solution was Hemosol B0 (Gambro, Lund, Sweden). A 0.65 m2 polymethylpentene heparin-coated hollow fiber membrane oxygenator (MEDOS HILITE 2400 LT; MEDOS Medizintechnik AG, Stolberg, Germany) was inserted prior to the RRT circuit’s priming, either upstream or downstream of the hemofilter, using ¼ inches luer-lock connection tubes (priming volume 95 mL, maximum blood flow 4,800 mL/min). Inlet and outlet pressures were monitored using side ports at both ends of the membrane. Blood flow was recorded using a ¼ inches ultrasonic flow probe (SonoTT Clap-on Transducer; EMTEC, Munich, Germany) clamped on the inlet tube of the membrane oxygenator. The hemofilter circuit was primed with saline (2,000 mL containing 10,000 IU of heparin). Heparin was further administrated, within the RRT circuit, as a bolus after connection (80 IU/kg) followed by continuous infusion (18 IU/kg/h) if the platelet count was greater than 50 G/L and prothrombin was greater than 30%. An active clotting time ratio of 1.5 was targeted.

2,000 mL containing 10,000 IU of heparin). Heparin was further administrated, within the RRT circuit, as a bolus after connection (80 IU/kg) followed by continuous infusion (18 IU/kg/h) if the platelet count was greater than 50 G/L and prothrombin was greater than 30%. An active clotting time ratio of 1.5 was targeted. Study Protocol Inclusion’s measurements were obtained prior connecting the RRT + ECCo2R device. Thereafter, a 15.5F double-lumen extracorporeal circuit catheter (JFFS; Jet Medical, La Chaux-de-Fonds, Switzerland) was inserted into the right venous jugular vein, using echographic guidance. The 15-cm-long catheter was primarily chosen, but a 24 cm was also available. Baseline’s measurements were obtained once the RRT + ECCo2R device had been connected, but with zero sweep gas flow. Then, a sweep gas flow of 8 L/min with FO2 = 1 was added and not varied for the remainder of the study. Twenty minutes after initiation of ECCo2R, we performed the primary endpoint set of measurements. Thereafter, TV was reduced to 4 mL/kg PBW for the remainder of the study (72 hr). Other measurements were performed at 1–6–12–24–36–48–72 hours.

8 L/min with FO2 = 1 was added and not varied for the remainder of the study. Twenty minutes after initiation of ECCo2R, we performed the primary endpoint set of measurements. Thereafter, TV was reduced to 4 mL/kg PBW for the remainder of the study (72 hr). Other measurements were performed at 1–6–12–24–36–48–72 hours. Respiratory settings were modified according to the PEEP/Fio2 table of the ARDS Net protocol (supplemental data, Supplemental Digital Content 1, http://links.lww.com/CCM/B420). A recruitment maneuver (i.e., 2 cm H2O stepwise increases in PEEP until plateau pressure = 35 cm H2O) was performed after a tracheobronchial aspiration if SpO2 had decreased by greater than 2%. RR was adjusted to maintain Petco2 less than 45 mm Hg. Pump blood flow and effluent flow management are described in the supplemental data (Supplemental Digital Content 1, http://links.lww.com/CCM/B420). Fluid removal by RRT was not allowed during the first hour of therapy.

ion if SpO2 had decreased by greater than 2%. RR was adjusted to maintain Petco2 less than 45 mm Hg. Pump blood flow and effluent flow management are described in the supplemental data (Supplemental Digital Content 1, http://links.lww.com/CCM/B420). Fluid removal by RRT was not allowed during the first hour of therapy. Measurements Arterial, preoxygenator (inlet) and postoxygenator (outlet) blood gases were obtained at 20 minutes (M20) and at 1, 12, 24, 48, and 72 hours. Oxygen content (CtO2), Co2 content (CtCo2), Co2 removal (Vco2Oxy), and oxygen uptake rate by the membrane oxygenator were calculated according to standard formulae (supplemental data, Supplemental Digital Content 1, http://links.lww.com/CCM/B420). Oxygenator’s blood flow and pressure drop (inlet minus outlet pressure) were recorded at each time. Mean arterial pressure, central venous pressure, and heart rate were obtained from the cardioscope. Respiratory parameters were recorded from the ventilator (TV, RR, Petco2, Vco2Lung, and Vo2). Total Co2 elimination was calculated as the sum of the Vco2 by the lung and by the membrane oxygenator. Plateau pressure was obtained after a 2-second end-inspiratory pause and total PEEP after a 2-second end-expiratory pause. Quasi-static respiratory system compliance and deadspace fraction of tidal ventilation were calculated according to standard formulae. RRT-derived parameters were recorded from the RRT monitor including hemofilter pressure drop and transmembrane pressure. Plasma neutrophil gelatinase-associated lipocalin was measured at inclusion using a fluorescence immunoassay (Alere Triage® Neutrophil Gelatinase-Associated Lipocalin Test).

Three patients died prematurely: two from refractory multiorgan failure and one from ventricular arrhythmia, despite initial improvement. The characteristics of the patients at inclusion are presented in Table 1. The main clinical and biological variables at inclusion are presented in Table 2. TABLE 1. Characteristics of the Study Population (n = 11 Patients) TABLE 2. Pulmonary, Hemodynamic, and Renal Parameters of the Study Population at Inclusion (n = 11 Patients) Arterial Blood Gas, Gas Exchange, and the Primary Endpoint Adding ECCo2R at a TV of 6 mL/kg decreased Paco2 by 21 % (95% CI, 17–25%), from 47 ± 11 to 37 ± 4 Torr (p < 0.001; Fig. 1). The relative reduction in Paco2 did not differ according to the position of the membrane oxygenator, but tended to be higher when the oxygenator was placed upstream of hemofilter (22%±7% vs 18%±6%). Figure 1. Individual changes in Paco2 between baseline and 20 min after ECCo2R initiation (p < 0.001) at constant tidal volume (6 mL/kg predicted body weight). Decreasing TV to 4 mL/kg PBW increased Paco2 by 27% (95% CI, 24–30%), from 37 ± 8 to 48 ± 10 Torr (p < 0.001). The time course of Paco2 throughout the study period is presented in Figure 2A. Change in pH followed inversely those of Paco2, with significant increases after ECCo2R initiation and a drop after reduction of TV (Fig. 2B). There were no significant variations during the study period for Pao2/Fio2. Other variables are reported in Table 3. TABLE 3. Arterial Blood Gases and Gas Exchange During the Study Period in the 12 Therapies

Decreasing TV to 4 mL/kg PBW increased Paco2 by 27% (95% CI, 24–30%), from 37 ± 8 to 48 ± 10 Torr (p < 0.001). The time course of Paco2 throughout the study period is presented in Figure 2A. Change in pH followed inversely those of Paco2, with significant increases after ECCo2R initiation and a drop after reduction of TV (Fig. 2B). There were no significant variations during the study period for Pao2/Fio2. Other variables are reported in Table 3. TABLE 3. Arterial Blood Gases and Gas Exchange During the Study Period in the 12 Therapies Figure 2. Time course of Paco2 (A) and pH (B) during the study period; *p < 0.05 vs baseline; †p < 0.05 vs 20 min (M20); £p < 0.05 vs 1 hr (H1). Respiratory Parameters Lowering TV from 6 to 4 mL/kg PBW (i.e., from 383 ± 63 to 258 ± 46 mL) decreased minute ventilation by 33%±2%, from 7.8 to 5.2 mL/min and reduced plateau pressure by 18%±2%, from 25 ± 4 to 21 ± 3 cm H2O (all p < 0.001; Fig. 3). The PEEP level and the quasi-static respiratory system compliance remain unaltered. The deadspace fraction of ventilation increased after reduction of TV, from 26%±16% to 32%±15% (p < 0.001). The rate of Co2 elimination by the lung (Vco2Lung) decreased significantly after ECCo2R initiation and was further reduced after reduction of TV (Fig. 4). Other variables are presented in Table 4. TABLE 4. Respiratory Parameters During the Study Period in the 12 Therapies

Respiratory Parameters Lowering TV from 6 to 4 mL/kg PBW (i.e., from 383 ± 63 to 258 ± 46 mL) decreased minute ventilation by 33%±2%, from 7.8 to 5.2 mL/min and reduced plateau pressure by 18%±2%, from 25 ± 4 to 21 ± 3 cm H2O (all p < 0.001; Fig. 3). The PEEP level and the quasi-static respiratory system compliance remain unaltered. The deadspace fraction of ventilation increased after reduction of TV, from 26%±16% to 32%±15% (p < 0.001). The rate of Co2 elimination by the lung (Vco2Lung) decreased significantly after ECCo2R initiation and was further reduced after reduction of TV (Fig. 4). Other variables are presented in Table 4. TABLE 4. Respiratory Parameters During the Study Period in the 12 Therapies Figure 3. Time course of tidal volume (VT; A), minute ventilation (B), plateau pressure (Pplat; C), and positive end-expiratory pressure (PEEP; D) during the study period; *p < 0.05 vs baseline; †p < 0.05 vs 20 min (M20). .VE = minute ventilation. Figure 4. Time course of the total rate of Co2 elimination (total Vco2) with respective contribution of the Vco2 by the natural lung and the Vco2 by the membrane oxygenator; p = NS for total Vco2 (Tables 4 and 5 for Vco2Lung and Vco2Oxy variations, respectively). Note that the total Vco2 increased at 20 min (M20) due to additional Co2 removal provided by the membrane oxygenator. Also note that the total Vco2 decreased at 1 hr (H1) due to the reduction of tidal volume from 6 to 4 mL/kg predicted body weight, with values similar to baseline.

and Vco2Oxy variations, respectively). Note that the total Vco2 increased at 20 min (M20) due to additional Co2 removal provided by the membrane oxygenator. Also note that the total Vco2 decreased at 1 hr (H1) due to the reduction of tidal volume from 6 to 4 mL/kg predicted body weight, with values similar to baseline. Membrane Oxygenator–Related Parameters and the Effect of Position On an average of both positions, the oxygenator’s blood flow was 410 ± 30 mL/min. There was a moderate decrease after 24 hours. Blood flow was significantly higher when the membrane oxygenator was placed upstream of the hemofilter (432 ± 25 mL/min) than when the membrane oxygenator placed downstream (382 ± 29 mL/min; p < 0.001 at all time, Fig. 5A). Figure 5. Time course of the oxygenator blood flow (A) and Co2 removal rate by the membrane oxygenator (B) during the study period according to the position of the membrane oxygenator within the renal replacement therapy circuit (upstream of hemofilter [HF] or downstream of HF); *p < 0.05 between positions; †p < 0.05 vs 20 min (M20); £p < 0.05 vs 1 hr (H1). On an average of both positions, the oxygenator’s Co2 removal rate (Vco2 Oxy) was 83 ± 20 mL/min without significant variation over time, and it accounted for 42%±9% of the total Co2 elimination. Vco2Oxy did not significantly differ according to the position of the membrane, but values were higher when the position was upstream of the hemofilter (91 ± 49 mL/min) than when the position was downstream (72 ± 59 mL/min; p = 0.083; Fig. 5B).

iation over time, and it accounted for 42%±9% of the total Co2 elimination. Vco2Oxy did not significantly differ according to the position of the membrane, but values were higher when the position was upstream of the hemofilter (91 ± 49 mL/min) than when the position was downstream (72 ± 59 mL/min; p = 0.083; Fig. 5B). Inlet and outlet oxygenator pressures significantly decreased over time, whereas the oxygenator’s pressure drop remained constant and remarkably low. Inlet PCO2 and inlet CtCo2 significantly increased over time. A significant correlation (r = 0.92) was observed between the inlet PCO2 and the difference between inlet and outlet (delta) PCO2 (Fig. S1, Supplemental Digital Content 2, http://links.lww.com/CCM/B421, which illustrates the correlation between the inlet partial pressure of Co2 [PCO2 inlet] and the difference between the inlet and the outlet PCO2 [Delta PCO2]). Other variables are presented in Table 5. TABLE 5. Membrane Oxygenator–Related Parameters During the Study Period in the 12 Therapies RRT-Related Parameters Pump blood flow was maintained at maximum rate (420 mL/min) without any increase in the inlet arterial pressure over time. Effluent flow was maintained close to 45 mL/kg/h over time, with 33% predilution. Predilution and postdilution rates decreased over time, whereas fluid removal progressively increased. As a result, the filtration fraction remained constant around 15%. There was a slight but nonsignificant increase in the hemofilter’s pressure drop. In contrast, the transmembrane pressure of the hemofilter markedly increased over time (Table 6).

rates decreased over time, whereas fluid removal progressively increased. As a result, the filtration fraction remained constant around 15%. There was a slight but nonsignificant increase in the hemofilter’s pressure drop. In contrast, the transmembrane pressure of the hemofilter markedly increased over time (Table 6). TABLE 6. Hemofiltration Parameters During the Study Period in the 12 Therapies Hemodynamic Parameters Heart rate decreased significantly over time. Mean arterial pressure increased significantly by 20 minutes after initiation of the combined therapy when compared with baseline values, from 74 ± 14 to 90 ± 15 mm Hg (p < 0.001). A modest increase in cardiac output was also observed at 20 minutes. Norepinephrine doses decreased, but not significantly, over time; meanwhile two patients evolved toward multiorgan failure requiring high amount of vasopressor (Fig. S2, Supplemental Digital Content 3, http://links.lww.com/CCM/B422, which illustrates the time course of individual doses of norepinephrine throughout the study period among the 10 patients who received norepinephrine. Note that the two patients with high and increasing values evolved toward multiorgan failure and died.). Other variables are presented in Table 7. TABLE 7. Hemodynamic Parameters During the Study Period in the 12 Therapies Coagulation and Heparin Therapy All patients received continuous infusion of heparin at a mean rate of 521 ± 18 IU/h. The mean active clotting time ratio was 2 ± 0.9 during study period when compared with 1.5 ± 0.3 at inclusion.

Hemodynamic Parameters Heart rate decreased significantly over time. Mean arterial pressure increased significantly by 20 minutes after initiation of the combined therapy when compared with baseline values, from 74 ± 14 to 90 ± 15 mm Hg (p < 0.001). A modest increase in cardiac output was also observed at 20 minutes. Norepinephrine doses decreased, but not significantly, over time; meanwhile two patients evolved toward multiorgan failure requiring high amount of vasopressor (Fig. S2, Supplemental Digital Content 3, http://links.lww.com/CCM/B422, which illustrates the time course of individual doses of norepinephrine throughout the study period among the 10 patients who received norepinephrine. Note that the two patients with high and increasing values evolved toward multiorgan failure and died.). Other variables are presented in Table 7. TABLE 7. Hemodynamic Parameters During the Study Period in the 12 Therapies Coagulation and Heparin Therapy All patients received continuous infusion of heparin at a mean rate of 521 ± 18 IU/h. The mean active clotting time ratio was 2 ± 0.9 during study period when compared with 1.5 ± 0.3 at inclusion. Safety We did not observe the presence of air within the circuit. There was one episode of hemofilter clotting but none for the membrane oxygenator. One extracorporeal catheter needs to be replaced by a longer one (from 15- to 24-cm length) because of excessive negative arterial pressure. There was no pump dysfunction or any bleeding complications. In one patient, an alarm “Gain Limit Reached” prematurely ended the therapy (just before the 72-hr time). The insertion of a membrane oxygenator did not modify the normal operation of the RRT device.

cm length) because of excessive negative arterial pressure. There was no pump dysfunction or any bleeding complications. In one patient, an alarm “Gain Limit Reached” prematurely ended the therapy (just before the 72-hr time). The insertion of a membrane oxygenator did not modify the normal operation of the RRT device. DISCUSSION In the present study, we demonstrate that adding a membrane oxygenator within a CRRT circuit, in patients presenting with both ARDS and AKI, is safe and provides efficient extracorporeal Co2 removal with a reduction of Paco2 by 21%. Such a combined therapy also allows mechanical ventilation at reduced TV together with blood purification for a period of 72 hours.

ding a membrane oxygenator within a CRRT circuit, in patients presenting with both ARDS and AKI, is safe and provides efficient extracorporeal Co2 removal with a reduction of Paco2 by 21%. Such a combined therapy also allows mechanical ventilation at reduced TV together with blood purification for a period of 72 hours. The idea of integrating an oxygenator membrane within an RRT circuit was first reported in 2013 by Forster et al (35). They used continuous venovenous hemodialysis with a 13.5F double-lumen catheter and a 0.67 m2 membrane oxygenator that was inserted downstream of a high-flux polysulfone 1.4 m2 hemofilter. However, they did not standardize their therapy in terms of blood flow, sweep gas flow, and ventilator settings and thus reported scattered data, mostly in patients with ARDS. They observed a marked decrease in Paco2 (from 69 to 49 Torr) and increase in pH (from 7.18 to 7.3) at 4 hours after starting ECCo2R, but data on the Co2 removal rate were not available. The correction of respiratory acidosis was associated with rapid hemodynamic improvement, which led to a marked decreased in the need for norepinephrine. They observed two episodes of clotting: one from the hemofilter and the other from the membrane oxygenator.

Co2R, but data on the Co2 removal rate were not available. The correction of respiratory acidosis was associated with rapid hemodynamic improvement, which led to a marked decreased in the need for norepinephrine. They observed two episodes of clotting: one from the hemofilter and the other from the membrane oxygenator. In the present study, we used continuous venovenous hemofiltration with a larger catheter (15.5F), and we were able to maintain high blood flow throughout the study period. The mean Co2 removal rate was similar to those reported with the RAS Hemolung system (ALung, Pittsburg, USA) and with the Abylcap system (Belco, Mirandola, Italy) at a similar blood flow (400 mL/min) (36, 37). Of note, the effective blood flow delivered by the PrismaFlex device corresponds to the pump blood flow minus the predilution rate, the result being limited to 450 mL/min. We did not observe any clotting of the membrane oxygenator, whereas it occurred once in the hemofilter. The time course of the pressure drop profile for both the hemofilter and the membrane oxygenator clearly showed that the risk of clotting was higher within the hemofilter.

ate, the result being limited to 450 mL/min. We did not observe any clotting of the membrane oxygenator, whereas it occurred once in the hemofilter. The time course of the pressure drop profile for both the hemofilter and the membrane oxygenator clearly showed that the risk of clotting was higher within the hemofilter. This is the first study to address a relevant issue, namely the position of the membrane oxygenator within the RRT circuit. We observed a higher level of blood flow and a higher level of Co2 removal when the membrane oxygenator was placed upstream of the hemofilter, albeit the latter did not reach statistical significance. During hemofiltration, when the oxygenator is placed downstream of the hemofilter, the flow that exits the hemofilter and crosses the oxygenator corresponds to the subtraction of the inlet hemofilter blood flow minus the effluent flow. Therefore, the difference in flow between each position, at constant pump blood flow, depends on the amount of effluent. In our study, we observed a difference in flow of 50 mL/min between each position, which resulted in a difference of 19 mL/min in the rate of Co2 removal. Hence, we recommend clinicians to place the membrane oxygenator upstream of the hemofilter.

position, at constant pump blood flow, depends on the amount of effluent. In our study, we observed a difference in flow of 50 mL/min between each position, which resulted in a difference of 19 mL/min in the rate of Co2 removal. Hence, we recommend clinicians to place the membrane oxygenator upstream of the hemofilter. Although one of the main limitations with the Decap system is the inability to provide blood purification, we designed this study to combine both an appropriate intensity of renal support together with enhanced lung protective ventilation. We were able to provide sustained purification throughout the study period; however, the transmembrane pressure of the hemofilter increased systematically over time. Using a relatively high effluent dose (45 mL/kg/h); we observed, as others, a rapid hemodynamic improvement that led to a substantial decrease in the need for norepinephrine in most patients (38, 39). Finally, we observed an increase in the plasma concentration of bicarbonate ion (HCO3–) and in the Co2 content over the study period. This may be related to the high concentration of HCO3– (32 mmol/L) within our replacement solution (Hemosol B0), which provides a faster acidosis correction but increases the blood Co2 content. Whether other HCO3– concentrations in the replacement solution would have changed results need to be explored.

e study period. This may be related to the high concentration of HCO3– (32 mmol/L) within our replacement solution (Hemosol B0), which provides a faster acidosis correction but increases the blood Co2 content. Whether other HCO3– concentrations in the replacement solution would have changed results need to be explored. Using a standardized protocol of ventilation based on the ARDS Net protocol, we were able to demonstrate that our combined RRT + ECCo2R system decreased arterial PCo2 by 21%. This result is similar to that reported with the Decap system, performed at a lower blood flow (300 mL/min) (34). Although the magnitude of the Paco2 reduction is less impressive than in the study by Forster et al (35), the range of Paco2 observed in our study is markedly lower. Because the removal of Co2 from the blood through the membrane oxygenator is a passive phenomenon, gas transfer depends on the gradient of partial pressure between both sides of the membrane (blood/gas). Therefore, the higher the inlet PCo2, the higher the diffusion will be (Fig. S1, Supplemental Digital Content 2, http://links.lww.com/CCM/B421). As others have reported, we also did not observe any improvement in systemic oxygenation as the range of extracorporeal blood flow achieved was inappropriate for this purpose.

Therefore, the higher the inlet PCo2, the higher the diffusion will be (Fig. S1, Supplemental Digital Content 2, http://links.lww.com/CCM/B421). As others have reported, we also did not observe any improvement in systemic oxygenation as the range of extracorporeal blood flow achieved was inappropriate for this purpose. Lowering TV to 4 mL/kg led to a 33% decrease in minute ventilation and to a mean decrease of 4 cm H2O in plateau pressure, whereas Paco2 returned close to its initial value (i.e., 6 mL/kg without ECCo2R). We did not need to increase the PEEP level throughout the study period, and the quasi-static compliance of the respiratory system remained unaltered. Although some studies have suggested that lowering TV may be associated with derecrutement, this seems not to be the case in our study (40). We hypothesized that our PEEP levels were sufficient to prevent massive alveolar collapse, as recently observed in a CT scan study on a patient with ARDS in which the size of the nonaerated area of the lung did not increase after reduction of TV from 6 to 4 mL/kg with constant PEEP level (41). Although reducing the RR and airflow has recently been shown to decrease lung and plasma cytokines in an animal model, we were not able to decrease the RR with regards to the range of Co2 removal achieved (37). A strategy that combined both reduction in TV and RR may require higher level of Co2 removal and thus higher extracorporeal blood flow.

nd airflow has recently been shown to decrease lung and plasma cytokines in an animal model, we were not able to decrease the RR with regards to the range of Co2 removal achieved (37). A strategy that combined both reduction in TV and RR may require higher level of Co2 removal and thus higher extracorporeal blood flow. This study has some limitations. First, the aim was to test whether the combination of RRT + ECCo2R was feasible over a 72-hour period, but we did not demonstrate that our ventilation strategy had reduced lung injury. This point has already been addressed previously and was not in the scope of the present study. Second, our population was small and limited to patients who presented with ARDS during the early phase of the disease and AKI who required RRT initiation. Therefore, extrapolating results from this study to other conditions requires caution. Third, such a population seems quite rare, at least if both conditions are expected early after ICU admission. Whether the implementation of such a combined therapy during the stay will extend the field of the technique needs to be investigated. Four, the use of such a combined therapy should be limited to a period of 72 hours, beyond there is a theoretical risk of rupture of the circuit. Finally, we were not able to demonstrate a significant difference in the rate of Co2 removal between positions of the membrane oxygenator due to our small sample, but our result is very close to the threshold and we believe that the difference would have reach the significance level with the appropriate statistical power.

ally, we were not able to demonstrate a significant difference in the rate of Co2 removal between positions of the membrane oxygenator due to our small sample, but our result is very close to the threshold and we believe that the difference would have reach the significance level with the appropriate statistical power. CONCLUSION We have demonstrated that a strategy that combined CRRT with ECCo2R through the insertion of a membrane oxygenator within an RRT circuit is safe and allows sustained blood purification together with enhanced lung protective ventilation during the early phase of ARDS and AKI. A higher efficacy was observed when the membrane oxygenator was placed upstream of the hemofilter. The present study may constitute the rationale for the design of a randomized controlled study to address the effect of such a combined organ-support strategy on mortality. ACKNOWLEDGMENTS We thank Dr. Sylvain Thuaudet for his contribution to the circuit arrangement. We are also grateful to Dr. Sylvie Jordana for its assistance in neutrophil gelatinase-associated lipocalin measurement. We thank the nurses and personal from the ICUs in the Hospital Ambroise Paré and the Hospital Paul Desbief, which merge in a new hospital, the Hospital Européen, all in Marseille, France. Following are the study sites: Service de Réanimation, Hôpital Ambroise Paré, Marseille, France; Service de Réanimation, Hôpital Paul Desbief, Marseille, France; and Service de Réanimation, Hôpital Européen Marseille, Marseille, France. Supplementary Material *See also p. 2683.

ACKNOWLEDGMENTS We thank Dr. Sylvain Thuaudet for his contribution to the circuit arrangement. We are also grateful to Dr. Sylvie Jordana for its assistance in neutrophil gelatinase-associated lipocalin measurement. We thank the nurses and personal from the ICUs in the Hospital Ambroise Paré and the Hospital Paul Desbief, which merge in a new hospital, the Hospital Européen, all in Marseille, France. Following are the study sites: Service de Réanimation, Hôpital Ambroise Paré, Marseille, France; Service de Réanimation, Hôpital Paul Desbief, Marseille, France; and Service de Réanimation, Hôpital Européen Marseille, Marseille, France. Supplementary Material *See also p. 2683. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by a research grant from ARARD, Aubagne, France. Dr. Allardet-Servent’s institution received grant support from ARARD (contribution to the paiement of membrane oxygenator). The remaining authors have disclosed that they do not have any potential conflicts of interest.

Sepsis remains a major cause of morbidity and mortality. Current treatment options are limited to prompt source control, appropriate antibiotics, cardiorespiratory resuscitation, and general organ support (1), with attempts to develop new immunological therapies proving unsuccessful (2–6). A better understanding of how sepsis affects leukocyte function may help identify new therapeutic strategies. Altered leukocyte function occurs in response to inflammation. Monocytes in critically ill patients can display a phase of activation and one of refractoriness. This refractory state has been termed “deactivation” (7) or “reprogramming” (8) and is thought to represent immune-suppression (9–12). It is characterized by reduced expression of human leukocyte antigen-D-related peptide (HLA-DR) and reduced in vitro release of tumor necrosis factor (TNF) (7), but its clinical significance is not fully understood.

termed “deactivation” (7) or “reprogramming” (8) and is thought to represent immune-suppression (9–12). It is characterized by reduced expression of human leukocyte antigen-D-related peptide (HLA-DR) and reduced in vitro release of tumor necrosis factor (TNF) (7), but its clinical significance is not fully understood. TNF mediates a considerable degree of pleiotropy (13) through activation of different receptors; TNF receptor (TNFR)-1 and TNFR-2 (14, 15). TNF-α–converting enzyme (TACE) is a trans-membrane protease enzyme that cleaves membrane-bound (mem)-TNF to soluble (sol)-TNF (16), a process termed “ectodomain shedding.” TACE is widely expressed, including on monocytes (17), the largest producers of systemic TNF. In response to inflammatory stimuli, monocyte TACE activity increases rapidly (18, 19) consistent with post-translational modification. TACE can shed proinflammatory (sol-TNF) and anti-inflammatory (TNFR-1 and TNFR-2) substrates (20); hence, modulation of TACE activity may alter cellular inflammatory balance. Studies measuring TACE expression or indirectly quantifying activity suggest that it may regulate systemic inflammation (21, 22) and potentially be a useful surrogate of inflammation and organ injury in the systemic inflammatory response syndrome (SIRS). TACE and other protease enzymes are thought to form interlinked “networks” that regulate epithelial cell barrier functions and may display altered functionality in inflammatory conditions, such as sepsis (23), whereas conditional murine TACE knockout appears protective in endotoxic shock (24).

response syndrome (SIRS). TACE and other protease enzymes are thought to form interlinked “networks” that regulate epithelial cell barrier functions and may display altered functionality in inflammatory conditions, such as sepsis (23), whereas conditional murine TACE knockout appears protective in endotoxic shock (24). We have previously described a method to measure TACE catalytic activity based on TNF cleavage (25) via fluorescence resonance energy transfer (FRET) assay technology and have used this to elucidate upstream TACE signaling in human monocytes. This occurs through reactive oxygen species phosphorylating (activating) p38-mitogen activated protein kinase (MAPK) (18), which activates TACE. We hypothesized that monocyte TACE activity would be altered by sepsis and may reflect illness severity and/or cellular inflammatory balance. We performed an observational clinical study, obtaining monocytes from patients with sepsis and determining their TACE activity. Activity was determined at several levels: membrane expression, catalytic activity (basal and stimulus induced), as well as expression, and stimulus-induced ectodomain shedding of TNFR-1 and TNFR-2. To provide a baseline for comparison, data were obtained from healthy volunteers and a subgroup of patients with noninfectious SIRS.

s determined at several levels: membrane expression, catalytic activity (basal and stimulus induced), as well as expression, and stimulus-induced ectodomain shedding of TNFR-1 and TNFR-2. To provide a baseline for comparison, data were obtained from healthy volunteers and a subgroup of patients with noninfectious SIRS. MATERIALS AND METHODS Patients and Controls An independent research ethics committee approved the study (North London REC 3 reference: 10/H0709/77). Patients were recruited from Charing Cross Hospital between May 2010 and April 2012 within 48 hours of ICU admission. Recruitment was limited to patients who fulfilled two SIRS criteria, in whom sepsis was suspected and were expected to remain intubated for a further 48 hours. Patients with immunosuppression or receiving granulocyte colony-stimulating factor were excluded. Healthy volunteers were recruited from Imperial College London staff. Exclusion criteria for volunteers were immunosuppression, chronic pathology, or intercurrent illness. We calculated that to have 80% power to detect 1 sd difference in TACE activity between two groups (at p < 0.05), 24 participants would be required (ie, n = 12 in sepsis and volunteer groups). In total, 16 patients with sepsis and 15 healthy volunteers were recruited. Eight additional patients with noninfectious SIRS were recruited to ascertain whether changes in sepsis monocyte TACE activity were present within the wider critical care population.

ould be required (ie, n = 12 in sepsis and volunteer groups). In total, 16 patients with sepsis and 15 healthy volunteers were recruited. Eight additional patients with noninfectious SIRS were recruited to ascertain whether changes in sepsis monocyte TACE activity were present within the wider critical care population. Data Collection Baseline demographic data collected included age, sex, Acute Physiology and Chronic Health Evaluation II (APACHE II) score (26) and other clinical parameters. Clinical data were collected daily including Sequential Organ Failure Assessment (SOFA) scores (27), leukocyte counts, microbiological data, and organ support requirements. ICU lengths of stay, as well as ICU and hospital mortality, were also recorded. Sample Collection and Processing Patient blood (30 mL) was obtained from arterial cannula or central venous catheter at baseline (D0), day 2 (D2), and day 4 (D4) while in ICU. A single peripheral blood sample (30 mL) was taken from healthy volunteers. All samples were collected onto ice and processed immediately.

Data Collection Baseline demographic data collected included age, sex, Acute Physiology and Chronic Health Evaluation II (APACHE II) score (26) and other clinical parameters. Clinical data were collected daily including Sequential Organ Failure Assessment (SOFA) scores (27), leukocyte counts, microbiological data, and organ support requirements. ICU lengths of stay, as well as ICU and hospital mortality, were also recorded. Sample Collection and Processing Patient blood (30 mL) was obtained from arterial cannula or central venous catheter at baseline (D0), day 2 (D2), and day 4 (D4) while in ICU. A single peripheral blood sample (30 mL) was taken from healthy volunteers. All samples were collected onto ice and processed immediately. Peripheral Blood Mononuclear Cells Peripheral blood mononuclear cells (PBMCs) were isolated from blood using Histopaque-1077 (Sigma-Aldrich, St. Louis, MO) in conjunction with Leucosep tubes (Greiner, Stonehouse, United Kingdom). The concentration of CD14-positive cells was determined using an anti-CD14 antibody (BD Biosciences, Oxfordshire, United Kingdom), and PBMCs were resuspended at 5 × 106 CD14-positive cells/mL. For stimulation, cells were incubated for 1 hour at 37°C with Escherichia coli ultrapure lipopolysaccharide (LPS; Invivogen, San Diego, CA) at 1 μg/mL. To determine p38MAPK activation, PBMCs were fixed, permeated, and stained using antibodies specific for the phosphorylated (phospho)-p38MAPK (Thr180/Tyr182) isoform (Cell Signaling, Danvers, MA). An LPS stimulation of 1 μg/mL was used for 15 or 30 minutes. To measure the response, nonactivated MAPK phosphorylation levels were subtracted from LPS-activated levels. To determine sol-TNF production, PBMCs were stimulated for 4 hours at 37°C ± LPS (1 μg/mL). Supernatants were collected and sol-TNF levels measured by enzyme-linked immunosorbent assay (R&D Systems, Oxfordshire, United Kingdom). Patient interleukin (IL)-6 and sol-TNF levels were both measured by enzyme-linked immunosorbent assay of diluted heparinized plasma samples (Ebioscience, San Diego, CA).

PS (1 μg/mL). Supernatants were collected and sol-TNF levels measured by enzyme-linked immunosorbent assay (R&D Systems, Oxfordshire, United Kingdom). Patient interleukin (IL)-6 and sol-TNF levels were both measured by enzyme-linked immunosorbent assay of diluted heparinized plasma samples (Ebioscience, San Diego, CA). Monocyte Isolation Monocytes were isolated using magnetic-activated cell selection using a CD14-positive bead selection strategy (Miltenyi Biotech, Surrey, United Kingdom). “Two-Hit” LPS Model For chronic LPS stimulus, monocytes were placed in polytetrafluoroethylene inserts (Millipore, Carrigtwohill, Ireland) within 12-well plates (Becton Dickinson, Franklin Lakes, NJ) at 3 × 106 per mL for 16 hours at 37°C ± LPS (1 μg/mL). For acute LPS stimulus, cells were harvested from tissue plates and stimulated as previously described. Flow Cytometry Surface marker expression was determined by flow cytometry using anti-HLA-DR (clone L243; Biolegend, San Diego, CA), anti-TNFR-1 (clone 16803), anti-TNFR-2 (clone 22235), and anti-TACE (clone 111633, all R&D Systems) antibodies in conjunction with the manufacturer recommended isotype control. Cells were incubated with fluorophore-conjugated anti-human antibodies for 30 minutes at 4°C. A Cyan ADP fluorescence-activated cell sorter (Beckman Coulter, High Wycombe, United Kingdom) using Summit version 4.3.02 (Beckman Coulter) was used for acquisition and Flowjo V.7.5 (Tree Star, Ashland, OR) for analysis. Expression levels are reported as the geometric mean of the fluorescence intensity (MFI).

s at 4°C. A Cyan ADP fluorescence-activated cell sorter (Beckman Coulter, High Wycombe, United Kingdom) using Summit version 4.3.02 (Beckman Coulter) was used for acquisition and Flowjo V.7.5 (Tree Star, Ashland, OR) for analysis. Expression levels are reported as the geometric mean of the fluorescence intensity (MFI). FRET Assay The FRET assay was conducted as previously described (25). In brief, monocytes were resuspended at 2.5 × 104 cells per well in 384-well plates and incubated with a peptide containing two fluorophores conjugated by a TACE-sensitive 13 amino acid mem-TNF sequence. Cleavage of the peptide results in fluorescence due to loss of internal quenching between the donor (fluorescein) and acceptor (tetramethylrhodamine) fluorophores. Measurement of fluorescent signal over time allows TACE activity to be determined and quantified in fluorescence units (FU). For all groups, TACE activity was determined without stimulation (basal activity) and in response to an LPS stimulus of 1 μg/mL for 1 hour (LPS-induced activity). Data Analysis A Shapiro-Wilkes test was used to determine normality and appropriate analysis was based on the results of this. For two variables, a t test or Mann-Whitney U test was used, whereas for three or more variables, a one-way analysis of variance or Kruskal-Wallis test was used as appropriate. For categorical data chi-square analysis was used. Where correlation analysis was performed using Pearson or Spearman r values are provided.

s. For two variables, a t test or Mann-Whitney U test was used, whereas for three or more variables, a one-way analysis of variance or Kruskal-Wallis test was used as appropriate. For categorical data chi-square analysis was used. Where correlation analysis was performed using Pearson or Spearman r values are provided. RESULTS Details of the 16 patients with sepsis and eight with noninfectious SIRS are in Table 1. Apart from a younger age in the SIRS cohort, the demographics were similar. Vasopressor use was confined to the sepsis group (31.3%), and the predominance of pneumonia as the source of sepsis likely explains the lower Pao2:Fio2 ratio within this cohort. TABLE 1. Demographic, Physiological Variables at Baseline (D0) and Outcomes in the Sepsis and Systemic Inflammatory Response Syndrome Cohorts Although the median length of stay was 9.5 days, seven patients died and five were discharged earlier than expected meaning that serial samples were obtained from 14 patients (58%). Day 0 refers to sampling baseline; 78% of patients were sampled within 48 hours of admission and 100% within 72 hours. A single sample was obtained from 15 healthy volunteers; this group was younger (mean age, 33.3 ± 3.6 yr) and 60% (n = 9) were men.

ted meaning that serial samples were obtained from 14 patients (58%). Day 0 refers to sampling baseline; 78% of patients were sampled within 48 hours of admission and 100% within 72 hours. A single sample was obtained from 15 healthy volunteers; this group was younger (mean age, 33.3 ± 3.6 yr) and 60% (n = 9) were men. Standard Markers of Immune Phenotype Produced Interindividual Variability We sought to characterize patient inflammatory phenotype and used a cellular expressed marker (HLA-DR) in combination with an inducible marker (sol-TNF release in response to LPS stimulation) to determine whether monocytes displayed evidence of “deactivation” or “reprogramming” as previously described (7).

Variability We sought to characterize patient inflammatory phenotype and used a cellular expressed marker (HLA-DR) in combination with an inducible marker (sol-TNF release in response to LPS stimulation) to determine whether monocytes displayed evidence of “deactivation” or “reprogramming” as previously described (7). Human monocytes can be divided into subsets based on their expression of CD14, CD16, and CD64 (28–30). We attempted to subdivide monocytes into subsets for analysis; however, grossly altered expression levels of these surface markers on sepsis patients’ monocytes, presumably due to inflammatory modulation of marker expression (31–35) independently of subset type, precluded this. Instead, for consistency, monocytes were analyzed as a single population (CD11b+, CD56–, and CD14+) throughout. As expected, monocytes from patients with sepsis and SIRS displayed decreased HLA-DR expression and sol-TNF release (Fig. 1, A and B). As these markers behaved similarly across the patient groups, they appear to lack specificity when distinguishing sepsis from SIRS. Two patients in the sepsis group received renal replacement therapy, which has previously been shown to increase monocyte HLA-DR expression in sepsis (36). However, HLA-DR expression data from these two patients lay within the range seen in other patients and their exclusion did not alter the results.

sepsis from SIRS. Two patients in the sepsis group received renal replacement therapy, which has previously been shown to increase monocyte HLA-DR expression in sepsis (36). However, HLA-DR expression data from these two patients lay within the range seen in other patients and their exclusion did not alter the results. Figure 1. Monocytes from patients with sepsis and systemic inflammatory response syndrome (SIRS) appeared phenotypically similar on the basis of human leukocyte antigen (HLA)-DR expression and sol- tumor necrosis factor (TNF) release. Monocytes were isolated from healthy volunteers (HV; n = 15), patients with sepsis (Se; n = 16) and patients with SIRS (Si; n = 8). Monocyte HLA-DR expression levels, as determined by flow cytometry, were attenuated in sepsis and SIRS when compared with values obtained from healthy volunteer cells (A). Cells were then incubated with lipopolysaccharide 1 μg/mL for 4 hr and in vitro sol-TNF release quantified. Sol-TNF release was attenuated across sepsis and SIRS patients when compared with healthy volunteers (B). D0 is sampling baseline, D2 is day 2 and D4 is day 4. Data shown as median ± interquartile range. *p < 0.01 versus healthy volunteers.

th lipopolysaccharide 1 μg/mL for 4 hr and in vitro sol-TNF release quantified. Sol-TNF release was attenuated across sepsis and SIRS patients when compared with healthy volunteers (B). D0 is sampling baseline, D2 is day 2 and D4 is day 4. Data shown as median ± interquartile range. *p < 0.01 versus healthy volunteers. There were no significant correlations between APACHE II score and D0 HLA-DR expression (sepsis: r = –0.37, p = 0.16; SIRS: r = –0.29, p = 0.5) or in vitro sol-TNF release (sepsis: r = –0.25, p = 0.35; SIRS: r = –0.36, p = 0.44) in either group. Because there was a significant difference in age between the patient groups, and when compared with healthy volunteers, we sought to determine whether age had any association with our measured variables. There were no correlations between age and HLA-DR levels (sepsis: r = –0.18, p = 0.34; SIRS: r = 0.21, p = 0.41) or sol-TNF release (sepsis: r = 0.29, p = 0.25; SIRS: r = –0.11, p = 0.73) in either group. Among patients with sepsis, HLA-DR expression levels and SOFA scores were not correlated (r = 0.06; p = 0.75) and neither were SOFA scores or in vitro sol-TNF release (r = 0.24; p = 0.22). HLA-DR levels (MFI, 174.5; interquartile range [IQR], 34.3–312.1 survivors vs MFI, 57.5; IQR, 31–166 nonsurvivors; p = 0.25) and in vitro sol-TNF levels (268 pg/mL; IQR, 80–523 pg/mL survivors vs 579 pg/mL; IQR, 115–829 pg/mL nonsurvivors; p = 0.14) did not differ between patients with sepsis who survived and those who died.

levels (MFI, 174.5; interquartile range [IQR], 34.3–312.1 survivors vs MFI, 57.5; IQR, 31–166 nonsurvivors; p = 0.25) and in vitro sol-TNF levels (268 pg/mL; IQR, 80–523 pg/mL survivors vs 579 pg/mL; IQR, 115–829 pg/mL nonsurvivors; p = 0.14) did not differ between patients with sepsis who survived and those who died. Plasma levels of IL-6 and sol-TNF were measured at D0 within the patient groups. In contrast to the immune-depressed monocyte phenotype suggested by the HLA-DR and in vitro sol-TNF data, D0 IL-6 levels appeared raised in the sepsis group, suggesting continued inflammation (Fig. 2A). Sepsis IL-6 levels were correlated with APACHE II score (r = 0.71; p = 0.003) (Fig. 2B), but not with SOFA scores (r = 0.38; p = 0.16) and did not significantly differ between survivors and nonsurvivors (68 pg/mL; IQR, 3–242 pg/mL survivors vs 85 pg/mL; IQR, 62–433 pg/mL nonsurvivors; p = 0.43). IL-6 levels and HLA-DR expression were not correlated (r = –0.32; p = 0.25).

re (r = 0.71; p = 0.003) (Fig. 2B), but not with SOFA scores (r = 0.38; p = 0.16) and did not significantly differ between survivors and nonsurvivors (68 pg/mL; IQR, 3–242 pg/mL survivors vs 85 pg/mL; IQR, 62–433 pg/mL nonsurvivors; p = 0.43). IL-6 levels and HLA-DR expression were not correlated (r = –0.32; p = 0.25). Figure 2. Plasma interleukin (IL)-6 levels were raised in the sepsis group, and correlated at D0 with Acute Physiology and Chronic Health Evaluation II (APACHE II) scores. Plasma IL-6 and sol-tumor necrosis factor (TNF) levels were determined at D0 for both sepsis and patients with systemic inflammatory response syndrome (SIRS). The trend (p = 0.06) toward increased IL-6 levels in the sepsis group (A) is consistent with continued inflammation and (B) correlated with APACHE II score (r = 0.75; p < 0.01). In contrast, plasma sol-TNF was elevated among only a few patients with sepsis (C) and did not differ significantly between the groups (p = 0.55). Data shown as median ± interquartile range. Plasma sol-TNF levels were elevated in a few sepsis patients only (Fig. 2C) and did not correlate with APACHE II score (r = 0.17; p = 0.57). TACE Activity as Determined by FRET Assay We characterized TACE activity and expression profiles to determine whether they were altered by sepsis. Basal (no stimulation) TACE activity was elevated in the patients with sepsis (Fig. 3A), yet TACE expression and LPS-induced TACE activity appeared unaltered (Fig. 3, B and C). No changes in SIRS patient basal and LPS-induced TACE activity or expression were present (Fig. 3, A–C).

e whether they were altered by sepsis. Basal (no stimulation) TACE activity was elevated in the patients with sepsis (Fig. 3A), yet TACE expression and LPS-induced TACE activity appeared unaltered (Fig. 3, B and C). No changes in SIRS patient basal and LPS-induced TACE activity or expression were present (Fig. 3, A–C). Figure 3. Monocyte basal tumor necrosis factor-α–converting enzyme (TACE) activity was elevated in patients with sepsis compared with healthy volunteers. Monocytes isolated from healthy volunteers (HV), patients with sepsis (Se), and SIRS (Si) had basal (unstimulated) TACE activity determined. Within the sepsis group, significantly increased basal activity was present at D0 and D2 (A). Basal TACE activity did not differ in healthy volunteers and patients with SIRS. In sepsis and patients with SIRS, TACE expression (B) and lipopolysaccharide (LPS)–induced TACE activity (C) were unaltered. Sepsis patient D0 basal TACE activity was positively correlated with Acute Physiology and Chronic Health Evaluation II (APACHE II) score (r = 0.75; p < 0.01; D). Data shown as median + interquartile range. *p < 0.01 compared with HVs.

expression (B) and lipopolysaccharide (LPS)–induced TACE activity (C) were unaltered. Sepsis patient D0 basal TACE activity was positively correlated with Acute Physiology and Chronic Health Evaluation II (APACHE II) score (r = 0.75; p < 0.01; D). Data shown as median + interquartile range. *p < 0.01 compared with HVs. Sepsis monocyte D0 basal TACE activity was strongly correlated with APACHE II score (r = 0.75; p = 0.002) (Fig. 3C). Basal TACE activity and age were not correlated within the sepsis cohort (r = 0.13; p = 0.52). There was no relationship between sepsis monocyte basal TACE activity and ICU mortality (85.9 ± 50.1 FU/min survivors vs 86 ± 38.4 FU/min nonsurvivors; p = 1.0) or hospital mortality (88 ± 52.1 FU/min survivors vs 82.8 ± 36.7 FU nonsurvivors; p = 0.78). Basal TACE activity in healthy volunteers was not correlated with age (r = 0.33; p = 0.24). As expected (18), volunteer monocytes increased their basal TACE activity when LPS stimulated, whereas, in contrast, stimulation-induced activity changes were attenuated in sepsis (Fig. 4A). This attenuated activity negatively correlated with APACHE II and SOFA scores (Fig. 4, B and C). Among patients with sepsis, those who survived had larger TACE activity increases on LPS stimulation than those who died. The median fold increase was 1.7 (IQR, 1.4–2.4) for survivors versus 1.2 (IQR, 1.1–1.6) for nonsurvivors, p = 0.005. Similarly, the absolute LPS-induced TACE activity values also differed between these groups (146.7 ± 75.3 FU/min survivors vs 98.8 ± 28.7 FU/min for nonsurvivors; p = 0.05).

mulation than those who died. The median fold increase was 1.7 (IQR, 1.4–2.4) for survivors versus 1.2 (IQR, 1.1–1.6) for nonsurvivors, p = 0.005. Similarly, the absolute LPS-induced TACE activity values also differed between these groups (146.7 ± 75.3 FU/min survivors vs 98.8 ± 28.7 FU/min for nonsurvivors; p = 0.05). Figure 4. Monocytes isolated from patients with sepsis displayed an attenuated response to lipopolysaccharide (LPS). Monocytes from healthy volunteers and patients were stimulated with LPS 1 μg/mL at 37°C for an hour and their tumor necrosis factor-α–converting enzyme (TACE) activity increase determined. Sepsis patients (Se) displayed an attenuated TACE response to LPS (A) when compared with healthy volunteers (HV) and systemic inflammatory response syndrome (Si) patients. The attenuated TACE response to LPS seen in patients with sepsis was negatively correlated with APACHE II (r = –0.34; p = 0.07; B) and Sequential Organ Failure Assessment (SOFA) scores (r = –0.48; p = 0.09; C). Median + interquartile range. *p < 0.01 compared with HVs. Similar to healthy volunteers, patients with SIRS appeared to display a preserved response to LPS (Fig. 4A). No relationship between activity increases and mortality was present (median fold increase, 2.0; IQR, 1.7–2.4 ICU survivors vs 2.4; IQR, 1.7–2.9 ICU nonsurvivors; p = 0.57). LPS-stimulated TACE activity values were not correlated with age within either group (sepsis: r = –0.1, p = 0.61; SIRS: r = –0.05, p = 0.84).

4A). No relationship between activity increases and mortality was present (median fold increase, 2.0; IQR, 1.7–2.4 ICU survivors vs 2.4; IQR, 1.7–2.9 ICU nonsurvivors; p = 0.57). LPS-stimulated TACE activity values were not correlated with age within either group (sepsis: r = –0.1, p = 0.61; SIRS: r = –0.05, p = 0.84). TNFR Shedding We examined the behavior of TNFR-1 and TNFR-2 using a substrate-shedding index and this confirmed the results obtained using the FRET assay. To determine this index, monocytes were stimulated with LPS and receptor expression measured pre and post stimulation, allowing percentage shedding of each receptor to be calculated and used as a marker of TACE activity: (Baseline expression – LPS-induced expression) × 100/baseline expression. There were no differences in baseline expression of TNFR-1 or TNFR-2 on patient monocytes (Fig. 5, A and B). LPS-induced shedding of both TNFRs appeared reduced in sepsis (Fig. 5, C and D). Shedding of TNFR-1 at D0 was negatively correlated with both APACHE II (r = –0.5; p = 0.06) and SOFA scores (r = –0.44; p = 0.02). TNFR-1 shedding and age were not correlated (r = –0.1; p = 0.61), and shedding did not differ between survivors and nonsurvivors (% shedding 52.2 ± 23.6 survivors vs 49.7 ± 22.5 nonsurvivors; p = 0.82). There were no correlations between TNFR-2 shedding (D0) and APACHE II score (r = –0.26; p = 0.35), SOFA scores (r = –0.20; p = 0.33), or age (r = –0.28; p = 0.16). TNFR-2 shedding did not differ between survivors and nonsurvivors (% shedding 65.1 ± 20.2 survivors vs 50.2 ± 29.5 nonsurvivors; p = 0.17).

vors; p = 0.82). There were no correlations between TNFR-2 shedding (D0) and APACHE II score (r = –0.26; p = 0.35), SOFA scores (r = –0.20; p = 0.33), or age (r = –0.28; p = 0.16). TNFR-2 shedding did not differ between survivors and nonsurvivors (% shedding 65.1 ± 20.2 survivors vs 50.2 ± 29.5 nonsurvivors; p = 0.17). Figure 5. Attenuation of tumor necrosis factor receptor (TNFR)-1 and TNFR-2 shedding in patients with sepsis. Monocyte expression of TNFR-1 and TNFR-2 was determined for healthy volunteers (HV), sepsis (Se), and systemic inflammatory response syndrome (SIRS [Si]) patients (A and B). No significant alterations were detected in expression across the two patient groups. Percentage shedding of each substrate (in response to lipopolysaccharide 1 μg/ml for an hour) was then determined. In patients with sepsis, shedding was attenuated at D0 for TNFR-1 and TNFR-2, but did not appear altered by noninfectious SIRS (C and D). Data shown as median ± interquartile range. *p < 0.05 compared with HVs. In contrast, in response to LPS stimulation, healthy volunteer and SIRS monocytes shed the majority of membrane TNFR-1 and TNFR-2 (Fig. 5, C and D).

Figure 5. Attenuation of tumor necrosis factor receptor (TNFR)-1 and TNFR-2 shedding in patients with sepsis. Monocyte expression of TNFR-1 and TNFR-2 was determined for healthy volunteers (HV), sepsis (Se), and systemic inflammatory response syndrome (SIRS [Si]) patients (A and B). No significant alterations were detected in expression across the two patient groups. Percentage shedding of each substrate (in response to lipopolysaccharide 1 μg/ml for an hour) was then determined. In patients with sepsis, shedding was attenuated at D0 for TNFR-1 and TNFR-2, but did not appear altered by noninfectious SIRS (C and D). Data shown as median ± interquartile range. *p < 0.05 compared with HVs. In contrast, in response to LPS stimulation, healthy volunteer and SIRS monocytes shed the majority of membrane TNFR-1 and TNFR-2 (Fig. 5, C and D). In Vitro Modeling of the Changes in TNF and TNFR Cleavage Present in Sepsis Differences in TACE behavior between the patient groups may reflect exposure to pathogen-associated molecular patterns (PAMPs). We developed a “two-hit” inflammatory model to reproduce sepsis TACE activity profiles. Volunteer monocytes were cultured with LPS for 16 hours (chronic primary sepsis stimulus), exposed to a second LPS stimulus for 1 hour (acute secondary sepsis stimulus) and their TACE response determined. After untreated 16-hour culture (no chronic sepsis stimulus), cells remained capable of acute TACE activation on LPS stimulation (Fig. 6A). After chronic primary, LPS stimulus volunteer cells displayed increased basal TACE activity (Fig. 6A) and unaltered TACE expression (Fig. 6B) but were refractory to acute secondary LPS stimulus (Fig. 6A). In addition, shedding of TNFR-1 and TNFR-2 in response to the acute secondary LPS stimulus was also reduced by the chronic primary LPS stimulus (Fig. 6, C and D). Such changes may indicate that a ceiling of maximal TACE activity has been reached in these cells, with similar behavior observed in sepsis patient monocytes. However, it is important to recognize that substrate shedding will also be affected by TACE-independent factors, such as substrate expression and availability.

. Such changes may indicate that a ceiling of maximal TACE activity has been reached in these cells, with similar behavior observed in sepsis patient monocytes. However, it is important to recognize that substrate shedding will also be affected by TACE-independent factors, such as substrate expression and availability. Figure 6. The tumor necrosis factor (TNF)-α–converting enzyme (TACE) profiles seen in the patients with sepsis could be induced in healthy volunteer monocytes through the use of a two-hit inflammatory model. A two-hit inflammatory model was used to induce changes in TACE activity profiles in healthy volunteer cells. Monocytes were cultured with lipopolysaccharide (LPS) for 16 hours (primary chronic LPS stimulus) and then further stimulated with LPS (secondary acute LPS stimulus) and their TACE activity profiles determined. Changes elicited were consistent with those seen in sepsis patients: basal activity was increased and refractory to further LPS stimulus (A). No differences in TACE expression after primary LPS stimulus were present (B). Shedding of the TNF receptors (TNFRs)-1 (C) and TNFR-2 (D) were attenuated by LPS exposure. Sol-TNF was released after primary chronic LPS stimulus with no secondary acute stimulus (E), but not after secondary acute stimulus in cells exposed to both primary and secondary LPS stimuli (F). n = 8 (A and B), n = 4 (C–F), data shown as mean + sd. *p < 0.05, **p < 0.01, ***p < 0.001. E, Control refers to monocytes cultured without primary chronic LPS stimulus.F, Control refers to monocytes cultured without primary chronic LPS stimulus subsequently exposed to secondary acute LPS stimulus.

ary LPS stimuli (F). n = 8 (A and B), n = 4 (C–F), data shown as mean + sd. *p < 0.05, **p < 0.01, ***p < 0.001. E, Control refers to monocytes cultured without primary chronic LPS stimulus.F, Control refers to monocytes cultured without primary chronic LPS stimulus subsequently exposed to secondary acute LPS stimulus. To determine whether these altered profiles mirrored the overall changes in monocyte responsiveness, sol-TNF release was determined in the model. After untreated 16-hour culture (no chronic primary sepsis stimulus), monocytes remained capable of releasing sol-TNF on secondary acute LPS stimulus (Fig. 6, E and F). In contrast, this release following acute secondary LPS stimulus was undetectable after chronic primary LPS stimulus (Fig. 6F) and was similar to the attenuated sol-TNF release in sepsis and SIRS monocytes.

sis stimulus), monocytes remained capable of releasing sol-TNF on secondary acute LPS stimulus (Fig. 6, E and F). In contrast, this release following acute secondary LPS stimulus was undetectable after chronic primary LPS stimulus (Fig. 6F) and was similar to the attenuated sol-TNF release in sepsis and SIRS monocytes. P38-MAPK Signaling in Patients With Sepsis We previously demonstrated that upstream LPS-induced TACE activation is dependent on p38-MAPK phosphorylation (18). To investigate the mechanisms responsible for the altered TACE activity seen in patients with sepsis, we measured the p38MAPK response to LPS in the last seven patients with sepsis recruited, and found it to be attenuated (Fig. 7). As expected, elevated levels of phospho-p38MAPK were present in healthy volunteer monocytes after LPS stimulation (Fig. 7). The attenuated response in sepsis suggests a direct link between regulation of this pathway and the observed reduction of TACE activation. However, unlike TACE basal activity in sepsis, unstimulated basal (t = 0) levels of phospho-p38MAPK were not significantly elevated in the sepsis group compared with healthy volunteers. This difference suggests that TACE can be maintained in an activated state without the need for prolonged up-regulation of phospho-p38MAPK levels.

ike TACE basal activity in sepsis, unstimulated basal (t = 0) levels of phospho-p38MAPK were not significantly elevated in the sepsis group compared with healthy volunteers. This difference suggests that TACE can be maintained in an activated state without the need for prolonged up-regulation of phospho-p38MAPK levels. Figure 7. Alterations in basal and lipopolysaccharide (LPS)–induced p38-mitogen activated protein kinase (MAPK) signaling were seen in sepsis. Cells from a subgroup of patients with sepsis (n = 7) were LPS stimulated (1 μg/mL) and had levels of phospho-p38MAPK determined at baseline (t = 0) min, 15 (t = 15) min, and 30 (t = 30) min. Levels of phospho-p38MAPK were not significantly elevated at 15 or 30 min after LPS stimulation. In contrast, when cells taken from healthy volunteers (HVs) were exposed to the same protocol, LPS-induced phospho-p38MAPK levels were significantly increased at t = 15 (*p < 0.01 vs HV t = 0). Data shown as mean + sd.

min. Levels of phospho-p38MAPK were not significantly elevated at 15 or 30 min after LPS stimulation. In contrast, when cells taken from healthy volunteers (HVs) were exposed to the same protocol, LPS-induced phospho-p38MAPK levels were significantly increased at t = 15 (*p < 0.01 vs HV t = 0). Data shown as mean + sd. DISCUSSION We hypothesized that TACE activity profiles would be altered by severe sepsis and may reflect illness severity and/or cellular inflammatory balance. In this observational study, comparing and contrasting patients with sepsis with healthy volunteers and a small cohort of noninfected critically ill patients, we found that basal TACE activity was increased and LPS-induced TACE activity changes appeared attenuated among patients with sepsis. Although patient numbers were limited, the increase in basal activity was correlated with illness severity and was associated with attenuated TNFR shedding within our cohort. A two-hit inflammatory in vitro model could reproduce the observed changes in TACE activity profiles and, by examining upstream signaling, we found alterations in sepsis patients’ p38-MAPK signaling that were characterized by a reduced response to LPS. It is plausible that prolonged exposure to inflammatory stimuli result in increased basal TACE activity and an attenuated TACE response to further LPS stimulation, through altered p38MAPK signaling. By demonstrating that TACE is in a higher activation state but refractory to further stimulation, these results suggest a resetting of the monocyte phenotype rather than simple deactivation as part of the response to sepsis.

and an attenuated TACE response to further LPS stimulation, through altered p38MAPK signaling. By demonstrating that TACE is in a higher activation state but refractory to further stimulation, these results suggest a resetting of the monocyte phenotype rather than simple deactivation as part of the response to sepsis. Monocytes obtained from both patient groups showed reduced HLA-DR expression and sol-TNF release as expected. Consistent with previous work (37, 38), there appeared to be no differences in the magnitude and kinetics of these changes between sepsis and SIRS. We found no association between HLA-DR expression and patient outcomes, but it should be noted that our study was not powered to detect such an association. Relating HLA-DR expression to patient outcomes has produced conflicting results, and a review of studies in this area failed to find a threshold in HLA-DR expression that successfully predicted unfavorable outcomes in sepsis (12). There is no accepted consensus around threshold levels or in its measurement and thus the clinical significance of HLA-DR expression remains uncertain.

nflicting results, and a review of studies in this area failed to find a threshold in HLA-DR expression that successfully predicted unfavorable outcomes in sepsis (12). There is no accepted consensus around threshold levels or in its measurement and thus the clinical significance of HLA-DR expression remains uncertain. Plasma IL-6 levels were elevated among the patients with sepsis at D0 in a manner that correlated with APACHE II scores. Given that the half-life of IL-6 is under 6 hours (39), elevated levels indicate continued inflammation at the same time that other markers (HLA-DR and sol-TNF) suggest immune-suppression. It should be noted that IL-6 levels represent responses from a number of different cells, hence may be different from a monocyte-only response. Therefore, the relationship between pro- and anti-inflammatory processes appears complex, with both potentially occurring concurrently, and not necessarily consecutively, as has been suggested (40, 41). Given this complex picture, it may be beneficial to assess inflammatory balance at a cellular level, comparing basal and inducible responses in the same individual. Such an approach may help resolve some of the intrinsic interindividual variability when other markers such as HLA-DR, in vitro sol-TNF, and plasma IL-6 levels are used in conjunction with each other.

ial to assess inflammatory balance at a cellular level, comparing basal and inducible responses in the same individual. Such an approach may help resolve some of the intrinsic interindividual variability when other markers such as HLA-DR, in vitro sol-TNF, and plasma IL-6 levels are used in conjunction with each other. TACE activity has not previously been directly measured in critical illness. Our data provide the first direct functional insight into monocyte inflammatory function in this context. Within the sepsis group, basal TACE activity appeared increased and correlated with APACHE II score. Combined with this increase in basal activity, LPS-induced up-regulation of enzyme activity decreased and these changes were not seen in patients with SIRS. The refractoriness to LPS seen in sepsis may represent tolerance in a specific pathway, or a general inability to respond to PAMPs (cross-pathway tolerance), and a state of apparent cellular immune-suppression. It is interesting to note that in vitro sol-TNF release was attenuated across both groups, whereas TACE activity profiles appeared altered only by sepsis. This may reflect a lack of substrate availability as, if mem-TNF expression was reduced equally in both groups, but the shedding mechanism suppressed only in the sepsis group, it would explain these results.

l-TNF release was attenuated across both groups, whereas TACE activity profiles appeared altered only by sepsis. This may reflect a lack of substrate availability as, if mem-TNF expression was reduced equally in both groups, but the shedding mechanism suppressed only in the sepsis group, it would explain these results. We examined expression levels and shedding profiles of both TNFRs to assess for any relationship with TACE behavior changes and association with clinical status. Because the FRET assay uses a TNF-specific cleavage sequence peptide, extrapolating TACE catalytic activity values from this assay directly to proteolytic cleavage of alternate substrates is not possible, and a shedding index was used as a substrate-specific surrogate marker of activity. Distinct from any implications of TACE in the inflammatory process (42–44), enhanced release of TNFRs represents an anti-inflammatory process by reducing TNF signaling in cells and neutralizing sol-TNF (20), and therefore is of importance when attempting to draw conclusions as to cellular inflammatory balance. Within the sepsis group, there was attenuated shedding of TNFR-1 at D0, but no changes in expression. Although not statistically significant, attenuated TNFR-2 shedding was also present in the sepsis group. Overall, therefore, LPS-inducible TNFR-1 and TNFR-2 shedding was attenuated in patients with sepsis, but not in patients with SIRS, indicating a similar pattern to LPS-induced TNF cleavage. The TACE activity and shedding profiles obtained from the two patients who received renal replacement therapy were comparable with data from other patients within the sepsis group.

2 shedding was attenuated in patients with sepsis, but not in patients with SIRS, indicating a similar pattern to LPS-induced TNF cleavage. The TACE activity and shedding profiles obtained from the two patients who received renal replacement therapy were comparable with data from other patients within the sepsis group. Because we found no alterations in the surface expression of either TNFR-1 or TNFR-2, it is difficult to draw firm conclusions on the implications of this reduced shedding, but it is likely that this will affect paracrine signaling. Together with the TACE activity changes, altered shedding provides further information that TACE-mediated functions are altered in sepsis. It has been shown that plasma membrane microdomains regulate TNFR-1 shedding in human endothelial cells by colocalizing TACE to membrane expressed TNFR-1 (45), and this process may be altered in sepsis. Altered colocalization of protease enzymes and substrates may be of importance in the regulation of protease networks. Alternatively, it is plausible that the reduced shedding of TNFR-1 may result from the mitochondrial dysfunction seen in sepsis (46) because some investigators have implicated mitochondria in TNFR-1 ectodomain shedding (47).

of protease enzymes and substrates may be of importance in the regulation of protease networks. Alternatively, it is plausible that the reduced shedding of TNFR-1 may result from the mitochondrial dysfunction seen in sepsis (46) because some investigators have implicated mitochondria in TNFR-1 ectodomain shedding (47). The p38MAPK pathway appears altered in sepsis and the attenuated signaling we report is temporally related to changes in TACE activity. Therefore, TACE and/or p38MAPK may be of benefit when determining any evolving inflammatory response in longitudinal patient studies. MAPK signaling has been implicated in TNFR-1 and TNFR-2 downstream signaling in humans (48) and murine TNFR-2 shedding is p38MAPK dependent (49). Hence, the reduced TNFR shedding in sepsis monocytes is likely related to the attenuated p38MAPK activation that we observed.

matory response in longitudinal patient studies. MAPK signaling has been implicated in TNFR-1 and TNFR-2 downstream signaling in humans (48) and murine TNFR-2 shedding is p38MAPK dependent (49). Hence, the reduced TNFR shedding in sepsis monocytes is likely related to the attenuated p38MAPK activation that we observed. Our data would suggest that sepsis may “reprogram” cells, altering their response state. As we could reproduce sepsis TACE behavior in vitro using an LPS model, it is possible that reprogramming may be mediated through exposure to PAMPs. The fact that these changes are not seen within the SIRS cohort may reflect an absence of PAMPs and suggest a lack of single- or cross-pathway tolerance therein. In our septic cohort, both basal TACE activity and TNFR-1 shedding correlated with APACHE II scores. Data from a larger patient population may help delineate whether TACE (and/or other protease) enzyme behavior better reflects illness severity than HLA-DR or in vitro sol-TNF release in sepsis. Although TACE knockout mice are nonviable (50, 51), mice with myeloid cell–inactivated TACE have a survival benefit in endotoxic shock (24), and knockouts of other protease enzymes also confer a survival benefit in murine models of sepsis (52). Thus, an enhanced understanding of how enzyme function is altered in sepsis may ultimately lead to new therapeutic strategies.

mice with myeloid cell–inactivated TACE have a survival benefit in endotoxic shock (24), and knockouts of other protease enzymes also confer a survival benefit in murine models of sepsis (52). Thus, an enhanced understanding of how enzyme function is altered in sepsis may ultimately lead to new therapeutic strategies. There are some limitations of this observational study that should be considered. There is considerable heterogeneity within the sepsis group, with patients presenting with infection in different tissues and with different organisms. In future, it may be more appropriate to try and study a more homogenous population by concentrating on specific pathologies/organisms. Also the groups differ in their demographic data and therefore potential confounding effects cannot be excluded. Despite performing a power calculation and targeting patient recruitment based on this, numbers are low (especially within the SIRS subgroup) and some nonsignificant correlations may reflect this limited power. As patients improved and were discharged, or deteriorated and died, they were lost from the study with consequent reduction in our sample size over time. Although not the major focus of the study, we attempted to analyze individual monocyte subsets but were unable to clearly identify distinct populations in all patients with the markers we used. The lack of follow-up to determine whether TACE basal activity changes indicate resolution or chronic low-grade inflammation should also be considered a limitation.

udy, we attempted to analyze individual monocyte subsets but were unable to clearly identify distinct populations in all patients with the markers we used. The lack of follow-up to determine whether TACE basal activity changes indicate resolution or chronic low-grade inflammation should also be considered a limitation. CONCLUSIONS Monocyte TACE activity appeared altered by sepsis in a manner that reflects illness severity, could be reproduced in vitro through exposure to LPS, and may be mediated through attenuated p38MAPK signaling. These data support the reprogrammed monocyte concept and future work should focus on determining the functional capacity (ie, migratory and phagocytic capability) of this inflammatory phenotype in a larger patient cohort. Real-time measurements of enzymatic function, or other dynamic responses involving post-transcriptional protein modification, may provide a more immediate and complete determination of monocyte immune phenotype, revealing both the underlying inflammatory status and the response to further stimulation. This approach would allow a more sophisticated determination of cell signaling, differentiating sepsis from sterile inflammation, potentially identifying therapeutic targets and stratifying patients for immune-modulating therapies. * See also p. 1532.

CONCLUSIONS Monocyte TACE activity appeared altered by sepsis in a manner that reflects illness severity, could be reproduced in vitro through exposure to LPS, and may be mediated through attenuated p38MAPK signaling. These data support the reprogrammed monocyte concept and future work should focus on determining the functional capacity (ie, migratory and phagocytic capability) of this inflammatory phenotype in a larger patient cohort. Real-time measurements of enzymatic function, or other dynamic responses involving post-transcriptional protein modification, may provide a more immediate and complete determination of monocyte immune phenotype, revealing both the underlying inflammatory status and the response to further stimulation. This approach would allow a more sophisticated determination of cell signaling, differentiating sepsis from sterile inflammation, potentially identifying therapeutic targets and stratifying patients for immune-modulating therapies. * See also p. 1532. Supported, in part, by the Intensive Care Foundation through a Young Investigators Award and the National Institute for Health Research (NIHR) supported it through the Comprehensive Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London, and the infrastructure provided by the Critical Care Specialty Group of the Comprehensive Clinical Research Network.

s Award and the National Institute for Health Research (NIHR) supported it through the Comprehensive Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London, and the infrastructure provided by the Critical Care Specialty Group of the Comprehensive Clinical Research Network. This article presents independent research funded by the National Institute for Health Research (NIHR). The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.

s Award and the National Institute for Health Research (NIHR) supported it through the Comprehensive Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London, and the infrastructure provided by the Critical Care Specialty Group of the Comprehensive Clinical Research Network. This article presents independent research funded by the National Institute for Health Research (NIHR). The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. Dr. Gordon is a National Institute for Health Research (NIHR) Clinician Scientist Fellowship award holder and the study was included in the NIHR Clinical Research Network (NIHR CRN) portfolio (ID 10492). He lectured for Orion Pharmaceuticals and consulted for Ferring Pharmaceuticals and Baxter Healthcare (consultancy in area of sepsis, unrelated to this work). His institution received grant support from the UK Intensive Care Foundation (funded by a Young Investigator Award) and from the National Institute for Health Research (support from the NIHR Imperial Biomedical Research Centre); served as a board member for the Intensive Care Foundation (Director of Research); and received grant support from Tenax Therapeutics (Unrestricted grant for sepsis trial, unrelated to this work), Orion Pharmaceuticals (non-financial grant support), and NIHR (Several sepsis trial grants). Dr. O’Callaghan is employed by the Imperial College Healthcare NHS Trust and disclosed other support from the British Boxing Board of Control (Payment for work as a ringside physician). His institution received grant support from the Intensive Care Foundation (Young Investigators Award) and the National Institute for Health Research (support from the NIHR Imperial Biomedical Research Centre). Dr. O’Dea’s institution received grant support from the Intensive Care Foundation, UK; National Institute for Health Research, UK; and Wellcome Trust (Project grant to provide scientific infrastructure for this work). Dr. Alasdair is employed by Imperial College Healthcare NHS Trust and received support for article research from the NIH and from the Young Investigator Award. His institution received grant support from the Young Investigator Award and the NIHR. Dr. Takata’s institution received grant support from the Intensive Care Foundation, UK; National Institute for Health Research, UK; and Wellcome Trust (Project grant to provide scientific infrastructure for this work).

Congenital heart disease is relatively common, affecting between 4 and 14 individuals in every 1,000 live births (1). Around one third of affected children require surgery during early childhood. Interventions to correct or palliate congenital heart disease present their own challenges. Young age, abnormal circulatory physiology, altered blood flow and hypothermia during cardiopulmonary bypass, and the trauma of surgery may cause splanchnic hypoperfusion and tissue ischemia-reperfusion. The resulting dysregulation of homeostatic pathways regulating inflammation, metabolism, and the endocrine system has important clinical consequences including the severity of organ failure and secondary morbidity (2–5).

pass, and the trauma of surgery may cause splanchnic hypoperfusion and tissue ischemia-reperfusion. The resulting dysregulation of homeostatic pathways regulating inflammation, metabolism, and the endocrine system has important clinical consequences including the severity of organ failure and secondary morbidity (2–5). A better understanding of the interaction between homeostatic derangement and host organ failure in severe illness could improve diagnostic accuracy and patient stratification through the use of clinically relevant biomarkers. These are chemical entities that can be objectively measured and evaluated as an indicator of biological processes. Many biomarker studies have focused on genomic and transcriptomic methods, alongside protein identification in tissues and body fluids. New developments in metabolic profiling techniques such as nuclear magnetic resonance, spectroscopy, and mass spectrometry have opened up the possibility to build a picture of the metabolic environment in cells, tissues, or biofluids and identify key metabolic biomarkers of disease risk, severity and outcome. This approach of assessing metabolites within a biological sample has been used as a tool for biomarker identification and evaluation in a range of disease states, including surgical oncology, sepsis, obesity, and malnutrition (6–12).

iofluids and identify key metabolic biomarkers of disease risk, severity and outcome. This approach of assessing metabolites within a biological sample has been used as a tool for biomarker identification and evaluation in a range of disease states, including surgical oncology, sepsis, obesity, and malnutrition (6–12). We sought to evaluate the metabolic profiling approach and explore the effect of a tight versus conventional glycemic control regime on the metabolic response to surgery in a group of children with congenital heart disease. We used this pilot study to test the following hypotheses: 1) tight glycemic control post surgery reduces the inflammatory and metabolic dysregulation following surgery; 2) metabolic profiling has a role in augmenting stratification of and clinical management of children undergoing congenital heart surgery. METHODS Patients The study was approved by the Royal Brompton Hospital Research Ethics Committee (REC Ref 10/H0801/8). Children undergoing elective surgery for congenital heart disease were recruited preoperatively, with informed parental/patient consent. The children in this study were also enrolled into the Control of Hyperglycemia in Pediatric Critical Illness trial, with a treatment arm of using insulin to target blood glucose to a range of 72 to 126 mg/dL (4.0–7.0 mmol/L), and control arm of conventional glycemic control, with a target level below 216 mg/dL (12.0 mmol/L) (13).

in this study were also enrolled into the Control of Hyperglycemia in Pediatric Critical Illness trial, with a treatment arm of using insulin to target blood glucose to a range of 72 to 126 mg/dL (4.0–7.0 mmol/L), and control arm of conventional glycemic control, with a target level below 216 mg/dL (12.0 mmol/L) (13). Children were eligible to participate in the study if they were between 36 weeks of corrected gestational age and 16 years of age, if they had been admitted to a PICU, and if they had an arterial catheter in place and were receiving mechanical ventilation and vasoactive drugs, with an anticipated duration of treatment of at least 12 hours. Children were excluded if they had diabetes mellitus, a confirmed or suspected inborn error of metabolism, if withdrawal of treatment was being considered, or if they had been in a PICU for more than 5 days. Patient Biological Samples Peripheral blood was obtained from intra-vascular lines preoperatively (at induction of anesthesia), and at serial timepoints postoperatively following admission to the PICU. Specifically, samples were taken on admission (0 hr), and then at 6, 24, and 48 hr post surgery for as long as an indwelling vascular catheter was in situ. Blood was collected into endotoxin-free sodium heparin (168 IU) Vacutainer tubes (Beckton Dickinson, Oxford, United Kingdom). Plasma was extracted by centrifugation of blood at 1,200g for 10 minutes and the supernatant was stored at –80°C until analysis.

rgery for as long as an indwelling vascular catheter was in situ. Blood was collected into endotoxin-free sodium heparin (168 IU) Vacutainer tubes (Beckton Dickinson, Oxford, United Kingdom). Plasma was extracted by centrifugation of blood at 1,200g for 10 minutes and the supernatant was stored at –80°C until analysis. Cytokine Assays Cytokine concentrations in plasma samples were determined in duplicate using MSD MULTI-SPOT immunoassay kits for interleukins (IL) 10, 6, 8 and IL-1 receptor antagonist (IL-1ra) (Mesoscale Discovery, Gaithersburg, MD), according to the manufacturer’s protocol. Analysis was undertaken using MSD Workbench, version 3.0.18, on the Sector Imager 2400 (MSD) Statistical Analysis of Demographic Variables Values are shown as median (interquartile range) except where indicated. 1H NMR Spectroscopic Analysis The protocols for sample preparation and NMR acquisition are based on published protocols by Dona et al (14). A Bruker AVANCE III spectrometer (Bruker Biospin, Billerica, MA) with a broadband inverse 600 MHz 5 mm Z gradient probe with automatic tuning and matching was used for all the 1H NMR spectroscopy experiments. Further details are provided in the supplementary methods (Supplemental Digital Content 1, http://links.lww.com/CCM/B267).

ANCE III spectrometer (Bruker Biospin, Billerica, MA) with a broadband inverse 600 MHz 5 mm Z gradient probe with automatic tuning and matching was used for all the 1H NMR spectroscopy experiments. Further details are provided in the supplementary methods (Supplemental Digital Content 1, http://links.lww.com/CCM/B267). Data Reduction and Multivariate Analysis The acquired 1H NMR spectra of plasma samples were zero filled to 132 K points, Fourier transformed, phase and baseline corrected using Bruker Topspin 3.1 (Bruker Biospin) Using both in-house developed scripts (Drs T. Ebbels, O. Cloarec, and M. Rantalainen) for MATLAB (version 2012b; The MathWorks, Inc, Natick, MA) and python 2.7.4 (15–17). Further details on the analysis and model fitting (18–23) are provided in the supplementary methods (Supplemental Digital Content 1, http://links.lww.com/CCM/B267). Peak Fitting For deconvolution and relative metabolite quantification in the spectra the R (22) package BATMAN (Bayesian AuTomated Metabolite Analyzer for NMR spectra) was used (23). Relative concentrations for the metabolites were used to obtain Pearson correlation matrices and plot heatmaps in python (15–17) using the plotting library matplotlib (24). Fifteen metabolites were relatively quantified in this way: lactate, creatine, creatinine, glucose, citrate, formate, 3-hydroxybutyrate, acetone, acetate, acetoacetate, alanine, valine, isoleucine, leucine, and threonine. Further details are provided in the supplementary methods (Supplemental Digital Content 1, http://links.lww.com/CCM/B267).

relatively quantified in this way: lactate, creatine, creatinine, glucose, citrate, formate, 3-hydroxybutyrate, acetone, acetate, acetoacetate, alanine, valine, isoleucine, leucine, and threonine. Further details are provided in the supplementary methods (Supplemental Digital Content 1, http://links.lww.com/CCM/B267). RESULTS Patients There were 28 patients with a median age of 6.6 months (4.0–18.9 mo), and median weight of 6.2 kg (4.0–8.52). The median number of PICU-free days at 28 days was 24 (20.25–26). Children in this cohort were relatively undernourished, with a median weight age z score of –2.03 (–2.94 to –1.49). As a marker of organ failure, we used the Pediatric Logistic Organ Dysfunction (PELOD) score. During the first 24 hours, children enrolled in the study had a median PELOD score of 11 (11–20.75). Further clinical details on the patient clinical variables are shown in Table 1. The natures of the congenital heart lesions of the children enrolled in the study are shown in Table 2. TABLE 1. Clinical Characteristics of the Patients Enrolled in the Study TABLE 2. Congenital Heart Defects of Patients Enrolled to the Study Fifteen children had tight glycemic control postoperatively, and the remaining 13 had conventional blood glucose management. Median glucose (±interquartile range) for tight and conventional glycemic control groups was 6.1 mmol/L (5.8–6.5) and 6.7 mmol/L (6–8.02), respectively (p = 0.046). There was no difference in age, weight, weight age Z score, PELOD score, Inotrope Score, or duration of ventilation and PICU-free days between the two groups (Table 3).

e (±interquartile range) for tight and conventional glycemic control groups was 6.1 mmol/L (5.8–6.5) and 6.7 mmol/L (6–8.02), respectively (p = 0.046). There was no difference in age, weight, weight age Z score, PELOD score, Inotrope Score, or duration of ventilation and PICU-free days between the two groups (Table 3). TABLE 3. Comparison of Clinical Variables Between Those Children Undergoing Conventional Management and Those With Tight Glycemic Control Metabolic Profiles Showed No Difference Between Glycemic Control and Conventional Glucose Management Groups 1H NMR spectra of plasma profiles showed no difference in postoperative samples from children with glycemic control intervention compared with the control group. The prediction metrics for the PLS-DA model of the global 1H NMR profile are R2Y 0.75, Q2Y –0.44, and R2Y 0.46, Q2Y –0.78 for the model generated using a subset of 15 relatively quantified metabolites. The model validation metric Q2Y is obtained through cross-validation, and is therefore a more robust estimate against model overfitting. Because in both cases the obtained Q2Y values are negative, our data do not support the hypothesis of a difference in metabolic profiles caused by a tighter glycemic control compared with the conventional regime.

2Y is obtained through cross-validation, and is therefore a more robust estimate against model overfitting. Because in both cases the obtained Q2Y values are negative, our data do not support the hypothesis of a difference in metabolic profiles caused by a tighter glycemic control compared with the conventional regime. Metabolic Profiles Showed a Temporal Evolution That Returned Toward the Presurgical Profile With Clinical Recovery Substantial differences between pre- and postoperative samples were found. This is summarized in the principal components analysis scores plot including all pre- and postoperative timepoints showing that samples are generally well clustered according to the time at which they were taken (Fig. 1). As expected from a fairly heterogeneous population, there was a considerable amount of patient variability that became particularly visible postoperatively. The preoperative samples formed a tight cluster in the top left corner of the plot inferring relatively low variation in the biochemical composition of plasma, whereas samples obtained postoperatively were dispersed across the scores plot indicative of variation in the individual response to surgery or ultrafiltration. In general, there was a time trend post surgery with the samples immediately postoperatively (orange circles) migrating in the first principal component (PC) with the 6-hour (yellow) and 24-hour (green) samples differentiating from the preoperative samples in the second PC. By 48 hours postoperatively (purple circles), some of the patients were comapping with the preoperative samples, indicating a recovered metabolic profile.

migrating in the first principal component (PC) with the 6-hour (yellow) and 24-hour (green) samples differentiating from the preoperative samples in the second PC. By 48 hours postoperatively (purple circles), some of the patients were comapping with the preoperative samples, indicating a recovered metabolic profile. Figure 1. Principal component analysis score plots for the full nuclear magnetic resonance profile (A) and cytokine measurements (B). Coordinates are colored according to time of sampling and show systematic changes in both metabolite and cytokine profiles over time. Timepoint 1 (red): preop; timepoint 2 (orange): 0 hr postop; timepoint 3 (yellow): 6 hr postop; timepoint 4 (green): 24 hr postop; timepoint 5 (purple): 48 hr postop. We draw attention to this for two patients with different clinical severities by connecting the coordinates of their plasma profile in chronological order. The patient with mild disease (blue line) has a smaller trajectory than the patient with severe disease (red), who had a more deranged trajectory that did not recover to baseline by the end of the sampling period (48 hr post surgery).

erities by connecting the coordinates of their plasma profile in chronological order. The patient with mild disease (blue line) has a smaller trajectory than the patient with severe disease (red), who had a more deranged trajectory that did not recover to baseline by the end of the sampling period (48 hr post surgery). Changes in Metabolic Profile Were Related to Pre- and Postsurgical Factors The strength of the metabolic response to surgery was influenced by surgical and/or disease severity. This is demonstrated by trajectories for two patients that are highlighted in Figure 1. Patient 19 underwent major surgery and had a more prolonged postoperative recovery (RACHS-1 score 4, invasive ventilation for 241 hr and 7 d free of PICU at 28 d). The pattern of change in metabolic profile over time demonstrates that this patient had a correspondingly strong metabolic response. In contrast, a tighter trajectory was seen in samples from patient 26 whose surgery was less severe and recovery quicker (RACHS-1 score 2, invasive ventilation 21 hr and 25 d free of PICU at 28 d). Comparison of the preoperative samples with samples collected immediately postoperatively showed clear differentiation in the first PC (PC1), with coclustering of preoperative samples (Fig. 2A). Figure 2. Principal component analysis score plots for preoperative (cyan) and postoperative (red) detailing the variability in the full nuclear magnetic resonance profile (A), cytokine concentrations (B), and quantified metabolites (C).

Comparison of the preoperative samples with samples collected immediately postoperatively showed clear differentiation in the first PC (PC1), with coclustering of preoperative samples (Fig. 2A). Figure 2. Principal component analysis score plots for preoperative (cyan) and postoperative (red) detailing the variability in the full nuclear magnetic resonance profile (A), cytokine concentrations (B), and quantified metabolites (C). We analyzed the loading plots for the PCA model (Fig. 3, A and B) to identify metabolites that had significantly higher concentrations in the postoperative samples. Those identified included acetate, acetoacetate, acetone, alanine, citrate, formate, glucose, 3-hydroxybutyrate, isoleucine, leucine, N-acetylated glycoprotein, threonine, and valine. Figure 3. Loadings for the first principal component, projected onto the median nuclear magnetic resonance spectra for both pre- and postsurgery samples: δ 4.3–0.6 region (A) and δ 8.1–5.1 region (B). Analysis of Key Metabolites in Predictive Models of Clinical Outcome The loadings plot highlighted the influence of therapeutically administered drug and nutrition resonances (Fig. 3, B). This made it crucial to develop a sound analytical approach to identify relevant compounds that might be hidden by the peaks of some of these agents.

etabolites in Predictive Models of Clinical Outcome The loadings plot highlighted the influence of therapeutically administered drug and nutrition resonances (Fig. 3, B). This made it crucial to develop a sound analytical approach to identify relevant compounds that might be hidden by the peaks of some of these agents. We recorded all the administered drugs and feeds and undertook analysis of their metabolic profiles. We then incorporated these data into our patient sample analysis. We then examined those integrals from fitted peaks that showed a relationship with clinical factors, but which were independent of the influence of any exogenous agents. We selected the most significant metabolites identified in the multivariate analyses, then deconvolved and relatively quantified them. In addition to the metabolites identified through this method, we included known markers of organ dysfunction in critical illness. These included lactate (a well-established marker of tissue perfusion), creatine and creatinine (markers of muscle metabolism and renal function, respectively). PCA scores plots based solely on the quantified metabolites gave similar results to those models using the global spectral profile (Fig. 2, B). However, samples collected preoperatively were not so tightly clustered and one individual, denoted by an asterisk, appeared as an outlier postoperatively. The profile from this individual was distinguished from the rest mainly by high levels of lactate.

ts to those models using the global spectral profile (Fig. 2, B). However, samples collected preoperatively were not so tightly clustered and one individual, denoted by an asterisk, appeared as an outlier postoperatively. The profile from this individual was distinguished from the rest mainly by high levels of lactate. To more specifically identify the spectral features associated with the observed time course, OPLS regression was performed against time. Analysis of the regression coefficients revealed that the predominant variables largely match the highest loadings of PC1 on the pre- versus postoperative whole profile PCA plot. Next, we used orthogonal partial least squares discriminant (OPLS) analysis and OPLS-DA to screen for patterns in the full plasma profiles across individual timepoints. Using the 15 fitted metabolites, the OPLS models on samples taken immediately postoperation indicated moderate predictive capacity of the fitted metabolites for clinical outcome. These included RACHS-1 score (R2Y = 0.70, Q2Y = 0.26), PELOD (R2Y = 0.66, Q2Y = 0.17), and PICU-free days (R2Y = 0.71, Q2Y = 0.11). By 6 hours post operation, the ability of the metabolite panel to predict clinical outcome was stronger: RACHS (R2Y = 0.77, Q2Y = 0.70), PELOD (R2Y = 0.73, Q2Y = 0.70), and PICU-free days (R2Y = 0.70, Q2Y = 0.69). In addition, some predictive capacity of the metabolite panels for inotrope score (R2Y = 0.64, Q2Y = 0.59) and mechanical ventilation hours (R2Y = 0.68, Q2Y = 0.37) was evident.

ct clinical outcome was stronger: RACHS (R2Y = 0.77, Q2Y = 0.70), PELOD (R2Y = 0.73, Q2Y = 0.70), and PICU-free days (R2Y = 0.70, Q2Y = 0.69). In addition, some predictive capacity of the metabolite panels for inotrope score (R2Y = 0.64, Q2Y = 0.59) and mechanical ventilation hours (R2Y = 0.68, Q2Y = 0.37) was evident. Cytokine Profiles Had a Similar Postoperative Time Trajectory PCA scores plots of the cytokine measurements (Figs. 1, B and 2, C) displayed a similar scattering pattern to the corresponding metabolite scores plots, with preoperative samples clustering tightly and the samples from subsequent timepoints demonstrating a trajectory through time. However, in contrast to the metabolic profiles, cytokine levels regressed to preoperative values as early as 6 hours postoperatively for some patients. The key cytokines influencing the deviation in PC1 (corresponding to the immediate response to surgery) were an increase in the IL-6/IL-10 ratio and IL-1ra concentrations, which increased rapidly post surgery and then smoothly decayed back to the baseline. The other cytokines followed a similar but less defined pattern. At 6 hours post surgery, concentrations of some cytokines displayed inverse significant correlations with the subsequent number of days free of PICU at 28 days (IL-1ra – r: –0.62, p = 0.017; IL-6 – r: –0.56, p = 0.034; IL-8 – r: –0.64, p = 0.013; IL-10 – r: –0.71, p = 0.0041). For IL-6 and IL-8, concentrations measured at 24 hours continued to have association with PICU-free days at 28 days (IL-6 – r: –0.73, p = 0.026; IL-8 – r: –0.76, p = 0.017).

days free of PICU at 28 days (IL-1ra – r: –0.62, p = 0.017; IL-6 – r: –0.56, p = 0.034; IL-8 – r: –0.64, p = 0.013; IL-10 – r: –0.71, p = 0.0041). For IL-6 and IL-8, concentrations measured at 24 hours continued to have association with PICU-free days at 28 days (IL-6 – r: –0.73, p = 0.026; IL-8 – r: –0.76, p = 0.017). As with the metabolic profile data, no significant difference was noted in cytokine concentrations between the postoperative glycemic control groups. Assessment of the Integrated Response of Metabolites and Cytokines to Congenital Heart Surgery To examine relationships of the key metabolic and inflammatory markers, we prepared a correlation map, which showed associations between coherent groups of variables. We observed a significant association between IL-6 and IL-8, the ketone bodies (acetone, acetoacetate and 3-d-hydroxybutyrate), and the branched chain amino acids (valine, leucine, and isoleucine) across all time points (Fig. 4, A–C). Figure 4. Pearson correlation heat maps for the presurgery (A), immediately postsurgery (B), and combined 6–48 hr postsurgery samples (C). Cells with opaque coloring specify correlations with a value of p < 0.05 (Student t test). Red cells indicate positive correlation and blue cells negative correlation. Some relationships varied across time. For example, preoperatively, alanine was inversely correlated with ketone body levels. Lactate and creatine were strongly correlated with IL-6, IL-8, and the IL-6/IL-10 ratio (Fig. 4, A).

Figure 4. Pearson correlation heat maps for the presurgery (A), immediately postsurgery (B), and combined 6–48 hr postsurgery samples (C). Cells with opaque coloring specify correlations with a value of p < 0.05 (Student t test). Red cells indicate positive correlation and blue cells negative correlation. Some relationships varied across time. For example, preoperatively, alanine was inversely correlated with ketone body levels. Lactate and creatine were strongly correlated with IL-6, IL-8, and the IL-6/IL-10 ratio (Fig. 4, A). Postoperatively, we observed a change in association pattern. Immediately postoperatively 3-d-hydroxybutyrate, acetoacetate, and to some extent acetone were positively correlated with the branch chain amino acids, particularly leucine (Fig. 4, B). Another significant correlation immediately postoperatively was an inverse association between the ketone bodies and cytokines IL-6 and IL-8. Neither of these correlations persisted through to later timepoints (Fig. 4, C). Other correlations of interest included citrate and acetate, evident only in samples obtained immediately post operation, and between alanine and the branched chain amino acids, which showed a strong positive correlation postoperatively. We also investigated the correlation of the quantified lactate with the IL-6/IL-10 ratio and found a strong correlation on preoperative samples (r = 0.98; p = 3.15–5). This correlation becomes nonsignificant immediately post operation (r = –0.38; p = 0.169), but reverts to the preoperative status afterward (r = 0.76; p = 1.75–6).

estigated the correlation of the quantified lactate with the IL-6/IL-10 ratio and found a strong correlation on preoperative samples (r = 0.98; p = 3.15–5). This correlation becomes nonsignificant immediately post operation (r = –0.38; p = 0.169), but reverts to the preoperative status afterward (r = 0.76; p = 1.75–6). DISCUSSION Metabonomic technology has been applied to several studies in critical care and neonatal medicine (28). However, previous studies have largely ignored confounders such as drugs (in this case, particularly antibiotics, sedatives, and anesthetics) and feed type. The spectral signatures from critically ill patients are known to be heavily “contaminated” by signals from exogenous resonances from treatments such as drugs, intravenous feeds, and organ perfusion solutions (29). The raw data from plasma spectra were confounded by metabolites derived from administered feeds and drugs and displayed very complex and dynamic biological composition. We were concerned that this would obfuscate large regions of the spectra. Therefore, we developed a method of peak fitting and relative signal quantification in plasma spectra, which was shown to be robust by demonstrating association with known metabolic biomarkers of outcome (such as lactate) and some of the important clinical outcomes, despite being a small patient cohort.

of the spectra. Therefore, we developed a method of peak fitting and relative signal quantification in plasma spectra, which was shown to be robust by demonstrating association with known metabolic biomarkers of outcome (such as lactate) and some of the important clinical outcomes, despite being a small patient cohort. No Effect of Glycemic Control Was Found on Clinical Outcome or on Metabolite or Cytokine Responses to Surgery Our patient cohort was a subset of patients recruited to a larger clinical trial of glycemic control in 1,300 critically ill children (13). We examined the effect of tight glycemic control on the global metabolic profile and a cytokine panel in response to congenital heart surgery when compared with the standard glycemic control procedures. While glucose concentrations differed between groups, the magnitude was small, but statistically significant (0.6 mmol/L or 10 mg/dL). In the wider clinical trial, a similar small but significant difference in mean blood glucose concentrations between the two groups of 0.42 mmol/L or 7.6 mg/dL was reported (13). We did not observe any difference in inflammatory or metabolic parameters, or in clinical outcome. It was not possible to obtain any reliable OPLS-DA model discriminating between the two glycemic control regimes at any of the timepoints after insulin administration started, using full plasma metabolic profile, relatively quantified metabolites, or cytokine measurements.

metabolic parameters, or in clinical outcome. It was not possible to obtain any reliable OPLS-DA model discriminating between the two glycemic control regimes at any of the timepoints after insulin administration started, using full plasma metabolic profile, relatively quantified metabolites, or cytokine measurements. Our data are consistent with results from the wider clinical trial, which concluded that tight glycemic control in critically ill children had no significant effect on major clinical outcomes (13). Metabolic Panels Reflect the Effects of Surgery and Clinical Outcome Our analytical approach enabled us to uncover profound changes in plasma composition following heart surgery in critically ill infants and children. Each patient’s metabolic response took a trajectory that demonstrated distinct phases relating to the pre-/postoperative stage. Each individual demonstrated a unique response to the surgery and this could be mapped as a patient journey, whereby both the metabolic and the cytokine panels could be followed and modeled temporally. We noted, even in this small patient cohort, that severity of surgery and critical illness affected the nature of the postoperative metabolic trajectory and its return to baseline. It would be interesting to explore this in a larger cohort and uncover important preoperative and postoperative factors including cyanosis, heart failure, splanchnic ischemia, and the effects of any intra- or postoperative complications.

d the nature of the postoperative metabolic trajectory and its return to baseline. It would be interesting to explore this in a larger cohort and uncover important preoperative and postoperative factors including cyanosis, heart failure, splanchnic ischemia, and the effects of any intra- or postoperative complications. Our population was undernourished, as measured by the weight age z score. Children with congenital heart disease are known to exhibit early and progressive falls in their growth trajectory compared with healthy children, with reductions in WAZ score, head circumference, and length for age z score (30, 31). It is also known that surgery and bypass, and the burden of cardiac failure and chronic disease, result in significant metabolic and nutritional stress. Furthermore, inadequate nutritional intake in the postoperative period results in further challenges to restoring normal growth parameters (32, 33). Our patient cohort received standard nutritional care pre- and postoperatively. No additional dietetic interventions are currently in place in our institution. Our analysis has demonstrated that changes in feed have a rapid and strong effect on metabolic profile, and it would be interesting to compare the profile with that in well-nourished children. The reduction in plasma very low-density lipoprotein concentrations postoperatively and increase in ketone bodies is consistent with fasting of these children. Thus, despite the confounding variables, it was possible to observe some interesting relationships that warrant further investigation.

well-nourished children. The reduction in plasma very low-density lipoprotein concentrations postoperatively and increase in ketone bodies is consistent with fasting of these children. Thus, despite the confounding variables, it was possible to observe some interesting relationships that warrant further investigation. Lactate (34–36), inflammatory cytokines, and IL-6/IL-10 ratio (5, 37) have been shown to be markers of outcome severity in critically ill patients. The former has been shown to correlate well with classical scoring systems of critical care, such as Acute Physiology and Chronic Health Evaluation II (37, 38) Our data demonstrate an association between inflammatory cytokine balance (IL-6/IL-10 ratio) and lactate, suggesting a close relationship between inflammation and metabolic derangement in the days after surgery for congenital heart disease. The increase in N-acetylated glycoprotein fragments immediately post-surgery is consistent with inflammation, which also reflects the general spike in plasma cytokines.

io) and lactate, suggesting a close relationship between inflammation and metabolic derangement in the days after surgery for congenital heart disease. The increase in N-acetylated glycoprotein fragments immediately post-surgery is consistent with inflammation, which also reflects the general spike in plasma cytokines. Using the relatively quantified metabolites, it was possible to obtain reliable OPLS models for the PELOD and RACHS-1 clinical scores, particularly in the samples taken in the early postoperative period (on PICU admission and at 6 hr post surgery). Analysis of the OPLS model coefficients showed that both the RACHS-1 and the PELOD scores were mainly associated with lactate, citrate, and to a lesser extent creatinine and alanine. Ketone bodies, particularly acetoacetate and creatine, displayed the inverse association with these scores. In all analyses, creatine showed the inverse trend of creatinine (more creatinine and less creatine associated with higher scores and greater severity). All these metabolic parameters were also associated with PICU-free days but with their trends reversed, as expected. The main metabolites positively associated with inotropic score were alanine, citrate, and lactate, with creatine inversely correlated. In general, ketone bodies seem to correlate with better surgical outcomes, whereas citrate, lactate, alanine, and higher creatinine-to-creatine ratio with the inverse.

ed, as expected. The main metabolites positively associated with inotropic score were alanine, citrate, and lactate, with creatine inversely correlated. In general, ketone bodies seem to correlate with better surgical outcomes, whereas citrate, lactate, alanine, and higher creatinine-to-creatine ratio with the inverse. Our data are similar to previously published work regarding metabolism in adults with critical illness (39). Increased catecholamine, cortisol, and glucagon secretion in critical illness are known to promote lipolysis and ketone body generation (39, 40). Triglycerides are metabolized to fatty acids and glycerol, a gluconeogenic substrate. High glucagon and low insulin concentrations also promote oxidation of free fatty acids to acyl CoA. The latter is hepatically converted to ketone bodies (β-hydroxybutyrate, acetoacetate, and acetone), as a water-soluble fuel source.

Triglycerides are metabolized to fatty acids and glycerol, a gluconeogenic substrate. High glucagon and low insulin concentrations also promote oxidation of free fatty acids to acyl CoA. The latter is hepatically converted to ketone bodies (β-hydroxybutyrate, acetoacetate, and acetone), as a water-soluble fuel source. To our knowledge, this work is the first metabonomic study in children undergoing surgery for congenital heart disease. The study demonstrates that nuclear magnetic resonance can be applied to biological samples from children with congenital heart disease, and that changes occur in metabolic profile during the response to the surgical insult. Because the postoperative metabolic spectra are heavily confounded by resonances deriving from feeds and drugs, the use of the whole 1H NMR plasma or urine profile can be quite limited, other than to identify gross outliers. Despite these shortcomings, we show that through an improvement in the prep-processing procedures, it is possible to reliably measure the evolution of various markers of interest.

iving from feeds and drugs, the use of the whole 1H NMR plasma or urine profile can be quite limited, other than to identify gross outliers. Despite these shortcomings, we show that through an improvement in the prep-processing procedures, it is possible to reliably measure the evolution of various markers of interest. We accept that our study group was small, and for this reason we could not keep separate training and test datasets for a more stringent assessment of the multivariate models. To further validate the PLS models in the setting of a small sample size, a permutation test was performed on the most relevant clinical parameters (such as surgical and disease severity scores and outcomes). In this test, the models are refitted and cross-validated multiple times, after permuting class labels or dependent variable values. The p values for the tests to assess the metabolite panel were highly significant (< 0.01) for RACHS score, PELOD score, and PICU-free days in models using data from samples at 6 hours post operation. Nevertheless, any putative marker identified in this study has to be properly validated in a larger cohort, as it is likely that the full extent of variability in this type of critical care population has been under sampled.

e, PELOD score, and PICU-free days in models using data from samples at 6 hours post operation. Nevertheless, any putative marker identified in this study has to be properly validated in a larger cohort, as it is likely that the full extent of variability in this type of critical care population has been under sampled. Because ketone bodies and lipid metabolism were indicated as predictive of clinical outcomes and displayed alterations in response to surgical insult, additional studies should include evaluation of liquid chromatography-mass spectrometry to profile the lipidomic response to critical illness. Liquid chromatography-mass spectrometry methods generate 2D datasets (chromatographic retention time and mass-to-charge ratio) and have less form overlap between exogenous and endogenous molecules. Our data on inflammatory links to metabolic shifts and clinical outcome post surgery suggest that targeted mass spectrometry methods could be of interest to identify other biological markers of acute inflammation.

tion time and mass-to-charge ratio) and have less form overlap between exogenous and endogenous molecules. Our data on inflammatory links to metabolic shifts and clinical outcome post surgery suggest that targeted mass spectrometry methods could be of interest to identify other biological markers of acute inflammation. In summary, our study suggests that metabolic profiling could be a clinically relevant tool for stratifying patient responses to congenital heart surgery. Furthermore, we show evidence that tight glycemic control does not impact the postsurgical metabolic profiles, supporting the clinical trial from which this study was drawn that failed to show improvement in early clinical outcomes with tight glycemic control. With improved data analysis workflows and targeted metabolite quantification, detailed and informative metabolic data could inform patient stratification and critical care management. In summary, we have identified a panel of metabolites, which is predictive of clinical outcome. This metabolite panel will require validation in a larger cohort of children.

In summary, our study suggests that metabolic profiling could be a clinically relevant tool for stratifying patient responses to congenital heart surgery. Furthermore, we show evidence that tight glycemic control does not impact the postsurgical metabolic profiles, supporting the clinical trial from which this study was drawn that failed to show improvement in early clinical outcomes with tight glycemic control. With improved data analysis workflows and targeted metabolite quantification, detailed and informative metabolic data could inform patient stratification and critical care management. In summary, we have identified a panel of metabolites, which is predictive of clinical outcome. This metabolite panel will require validation in a larger cohort of children. ACKNOWLEDGMENTS We are grateful to the patients, their parents and the staff at the Royal Brompton Hospital PICU for supporting this study. Samples were prepared and spectral data acquired at the Imperial-National Institute for Health Research (NIHR) Clinical Phenotyping Centre. The centre is supported by the NIHR Imperial Biomedical Research Centre based at Imperial College Healthcare National Health Service (NHS) Trust and Imperial College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal).

ACKNOWLEDGMENTS We are grateful to the patients, their parents and the staff at the Royal Brompton Hospital PICU for supporting this study. Samples were prepared and spectral data acquired at the Imperial-National Institute for Health Research (NIHR) Clinical Phenotyping Centre. The centre is supported by the NIHR Imperial Biomedical Research Centre based at Imperial College Healthcare National Health Service (NHS) Trust and Imperial College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by the Cardiovascular Biomedical Research Institute at the National Heart and Lung Institute at the Royal Brompton Hospital.

Supplementary Material Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccmjournal). Supported, in part, by the Cardiovascular Biomedical Research Institute at the National Heart and Lung Institute at the Royal Brompton Hospital. Mr. Correia is supported by the Imperial College Stratified Medicine Graduate Training Programme in Systems Medicine and Spectroscopic Profiling (STRATiGRAD). Dr. Pathan’s institution received grant support from the British Heart Foundation (research grant) and a Higher Education Funding Council for England clinical senior lecturer award. Dr. Ng received grant support from the British Heart Foundation (Researcher salary) and received support for article research from the British Heart Foundation. Dr. Jimenez consulted for Metabometrix is employed by the Imperial College London, and received support for article research from the National Institutes of Health (NIH). Her institution received grant support from the Cardiovascular Biomedical Research Institute. Dr. Macrae is employed by Royal Brompton and Harefield NHS Foundation Trust. Dr. Holmes is employed by the Imperial College London and received support for article research. Her institution received grant support from the Imperial College London (unrelated research grants in the field of metabolic medicine). The remaining authors have disclosed that they do not have any potential conflicts of interest.

Sepsis and septic shock are the most common causes of acute kidney injury (AKI) in critically ill patients, and sepsis-induced AKI accounts for 40–60% cases of AKI in ICUs (1–3), associated with a very high mortality, prolonged hospital stay, and increased costs of care (3–5). Few proven effective preventative or therapeutic interventions on septic AKI exist (6). Given the factors that contribute to AKI and AKI-associated mortality vary with time and differ in the immediate and long term (3), the pathogenesis of sepsis-induced AKI is complex and remains elusive. Hemodynamic factors might play a role in the loss of glomerular filtration rate (7); however, they might not act through the induction of renal damage in sepsis. A growing body of evidence suggests that nonhemodynamic mechanisms are likely to be crucial, including immunologic, toxic, and inflammatory factors that may affect renal microvasculature and tubular cells (2, 5). Among these mechanisms, apoptosis may turn out to be important (8, 9). Guo et al (10) found that apoptotic renal cells might actually be a source of local inflammation, contributing to subsequent nonapoptotic renal injury in endotoxemia-induced AKI mice. When treated with broad-spectrum caspase inhibitor, mice were protected against endotoxemia-induced AKI, characterized by significantly less apoptosis and less multiple markers of inflammation. Lee et al (11) also found that tubular cell apoptosis was prominent in septic kidneys, and inhibition of apoptosis by caspase-3 inhibitor contributed to attenuation of renal dysfunction. Therefore, inhibiting apoptosis or reducing inflammation might be a potential therapeutic strategy for septic AKI.

s of inflammation. Lee et al (11) also found that tubular cell apoptosis was prominent in septic kidneys, and inhibition of apoptosis by caspase-3 inhibitor contributed to attenuation of renal dysfunction. Therefore, inhibiting apoptosis or reducing inflammation might be a potential therapeutic strategy for septic AKI. MicroRNAs (miRNAs) are endogenous, small noncoding RNAs, which function as negative gene regulators at the posttranscriptional level, and play an important role in various biological processes. miR-21, a strong antiapoptotic factor (12, 13), has been shown to promote proliferation, has been shown to inhibit apoptosis, and is involved in the pathogenesis of kidney injury and tissue repair process (14–16). In our previous studies, we found that xenon exposure significantly increased the expression of miR-21 in a time-dependent manner in mouse kidney, and miR-21 contributed to the renoprotective effect of xenon preconditioning on ischemia-reperfusion injury and nephrotoxicity (17, 18). We hypothesize that xenon preconditioning could protect against septic AKI via upregulation of miR-21. In this study, we examined the effect of xenon preconditioning on lipopolysaccharide (LPS)-induced AKI in mice, by focusing on the systemic immune response as well as kidney inflammation and apoptosis, and further studied the underlying mechanisms.

MicroRNAs (miRNAs) are endogenous, small noncoding RNAs, which function as negative gene regulators at the posttranscriptional level, and play an important role in various biological processes. miR-21, a strong antiapoptotic factor (12, 13), has been shown to promote proliferation, has been shown to inhibit apoptosis, and is involved in the pathogenesis of kidney injury and tissue repair process (14–16). In our previous studies, we found that xenon exposure significantly increased the expression of miR-21 in a time-dependent manner in mouse kidney, and miR-21 contributed to the renoprotective effect of xenon preconditioning on ischemia-reperfusion injury and nephrotoxicity (17, 18). We hypothesize that xenon preconditioning could protect against septic AKI via upregulation of miR-21. In this study, we examined the effect of xenon preconditioning on lipopolysaccharide (LPS)-induced AKI in mice, by focusing on the systemic immune response as well as kidney inflammation and apoptosis, and further studied the underlying mechanisms. MATERIALS AND METHODS Experimental Animals Male C57BL/6 mice were obtained commercially (Animal Center of Fudan University, Shanghai, China) and studied at 10 weeks of age, weighing 20–25 g, housed in temperature- and humidity-controlled cages, with free access to water and rodent food, and a 12-hour light/dark cycle. This study was approved by the Institutional Animal Care and Use Committee of Fudan University and adhered strictly to the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

emperature- and humidity-controlled cages, with free access to water and rodent food, and a 12-hour light/dark cycle. This study was approved by the Institutional Animal Care and Use Committee of Fudan University and adhered strictly to the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Mouse Model of Gas Exposure and LPS-Induced AKI According to our previous study (17), mice were exposed to either 70% xenon or 70% nitrogen balanced with 30% oxygen for 2 hours through a close-loop ventilation system containing a reservoir bag, in which oxygen and xenon or nitrogen were mixed and delivered. Twenty-four hours after gas exposure, mice received a single intraperitoneal injection of Escherichia coli LPS (Sigma, St. Louis, MO) at a dose of 20 mg/kg. In Vivo Knockdown of miR-21 Using Locked Nucleic Acid-Modified Anti-miR Locked nucleic acid (LNA)-modified antiscrambled or anti-miR-21 oligonucleotides (Exiqon, Woburn, MA) were diluted in saline (5 mg/mL) and administered intraperitoneally (10 mg/kg) within 30 minutes before xenon exposure, referred to our previous study (17).

ockdown of miR-21 Using Locked Nucleic Acid-Modified Anti-miR Locked nucleic acid (LNA)-modified antiscrambled or anti-miR-21 oligonucleotides (Exiqon, Woburn, MA) were diluted in saline (5 mg/mL) and administered intraperitoneally (10 mg/kg) within 30 minutes before xenon exposure, referred to our previous study (17). Histopathological Examinations Kidney and liver slices were fixed in 10% formalin, embedded in paraffin wax, cut into 5-μm sections, and stained with hematoxylin and eosin. The tissues were evaluated under light microscopy by a pathologist blinded to the origin of preparations. Histologic injury scores were determined using scoring system, as described in previous study (17). The percentage of morphologic changes that displayed tubular cell necrosis, loss of brush border, vacuolization, tubule dilation, cast formation, and inflammatory cells infiltration were scored as follows: no injury (0), mild: less than 25% (1), moderate: less than 50% (2), severe: less than 75% (3), and very severe: more than 75% (4). Blood Chemistry Examination and Enzyme-Linked Immunosorbent Assay of Cytokines Serum creatinine (Scr) and alanine aminotransferase (ALT) were examined by an autoanalyzer (Vet test 8008; Idexx, Westbrook, ME). Concentrations of cytokines in blood and tissue homogenate were examined by commercially available enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems, Minneapolis, MN) for interleukin (IL)-6, IL-10, and tumor necrosis factor (TNF)-α, according to the manufacturer’s protocol.

et test 8008; Idexx, Westbrook, ME). Concentrations of cytokines in blood and tissue homogenate were examined by commercially available enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems, Minneapolis, MN) for interleukin (IL)-6, IL-10, and tumor necrosis factor (TNF)-α, according to the manufacturer’s protocol. Measurement of Nuclear Factor-κB Activity The nuclear factor (NF)-κB activity was measured with ELISA-based TransAM method using a commercial kit (TransAM NF-κB p65 Assay Kit; Active Motif, Carlsbad, CA) according to the manufacturer’s protocol. Briefly, a 96-well plate coated with an oligonucleotide containing the NF-κB consensus binding site (5′-GGGACTTTCC-3′) was used. The active form of NF-κB in the renal tissue nuclear extracts binds to the consensus site and is detected by a primary antibody specific for the activated NF-κB p65 subunit. Then, a horseradish peroxidase-conjugated secondary antibody was used for colorimetric quantification by spectrophotometry at 450 nm. The results were expressed as the optical density value.

nuclear extracts binds to the consensus site and is detected by a primary antibody specific for the activated NF-κB p65 subunit. Then, a horseradish peroxidase-conjugated secondary antibody was used for colorimetric quantification by spectrophotometry at 450 nm. The results were expressed as the optical density value. Terminal Deoxynucleotidyl Transferase-Mediated dUTP Nick-End Labeling Staining Kidney sections were stained for apoptotic nuclei with the terminal deoxynucleotidyl transferase-mediated dUTP nick-end labeling (TUNEL) method by using a commercially available in situ cell death detection kit (In situ Cell Death Detection kit, Peroxidase; Roche, Mannheim, Germany), according to the manufacturer’s protocol. The number of TUNEL-positive cells from 10 areas of randomly selected renal cortex was counted under a light microscope. In brief, formalin-fixed, paraffin-embedded kidneys were cut into 4-μM sections. Slides were rehydrated into a series of graded alcohols, followed by proteinase K treatment and incubation in 3% H2O2/methanol. Specimens were incubated with terminal deoxynucleotidyl transferase/bromide- deoxynucleoside triphosphate mixture, followed by anti-BrdU treatment and incubation in streptavidin-horseradish peroxidase, then subsequent detection with diaminobenzidine. Sections were examined under light microscopy for TUNEL-positive nuclei.

cimens were incubated with terminal deoxynucleotidyl transferase/bromide- deoxynucleoside triphosphate mixture, followed by anti-BrdU treatment and incubation in streptavidin-horseradish peroxidase, then subsequent detection with diaminobenzidine. Sections were examined under light microscopy for TUNEL-positive nuclei. Immunohistochemistry Immunohistochemical staining was performed in 4-μm paraffinized sections. In brief, after being dewaxed and dehydrated, the sections were incubated with 3% H2O2, treated with normal goat serum, and then incubated with primary antibody against lymphocyte antigen 6 (Ly-6G/-6C, rat monoclonal 1:100; Abcam, Cambridge, MA). Then, the sections were incubated with horseradish peroxidase-conjugated secondary antibody (anti-rat IgG), stained with 3,3′-diaminobenzidine (Sigma), and counterstained with hematoxylin. Slides were evaluated under light microscopy.

against lymphocyte antigen 6 (Ly-6G/-6C, rat monoclonal 1:100; Abcam, Cambridge, MA). Then, the sections were incubated with horseradish peroxidase-conjugated secondary antibody (anti-rat IgG), stained with 3,3′-diaminobenzidine (Sigma), and counterstained with hematoxylin. Slides were evaluated under light microscopy. Real-Time Reverse Transcription Polymerase Chain Reaction Total RNA from dissected kidney tissue was extracted using Trizol reagent (Invitrogen, Carlsbad, CA), followed by quantification. Extracted RNA was reverse transcribed to complementary DNA (PrimeScript RT Reagent Kit; TaKaRa, Kyoto, Japan) and then for real-time polymerase chain reaction (PCR) (SYBR Premix Ex TaqTM TaKaRa). PCR primers (Sangon, Shanghai, China) were designed with sequences as follows: TNF-α forward: 5′-GCCTCTTCTCATTCCTGCTTGT-3′, reverse: 5′-TTGAGATCCATGCCGTTG-3′; IL-6 forward: 5′-GCTACCAAACTGGATATAATCAGGA-3′, reverse: 5′-CCAGGTAGCTATGGTACTCCAGAA-3′; IL-10 forward: 5′-ACTGCACCCACTTCCCAGT-3′, reverse: 5′-TGTCCAGCTGGTCCTTTGTT-3′; β-actin forward: 5′-GATTACTGCCCTGGCTCCTA-3′, reverse: 5′-TCATCGTACTCCTGCTTGCT-3′. Expression level of miR-21 was quantified using real-time reverse transcription-PCR with the Taqman chemistry (Applied Biosystems, Hayward, CA), as described previously (19). U6 small nuclear RNA and β-actin messenger (mRNA) were used as endogenous control for miRNAs and mRNAs, respectively. Relative changes in mRNA and miR-21 expression were determined using the 2–ΔΔCt method. Relative gene levels were expressed as ratios to control.

Biosystems, Hayward, CA), as described previously (19). U6 small nuclear RNA and β-actin messenger (mRNA) were used as endogenous control for miRNAs and mRNAs, respectively. Relative changes in mRNA and miR-21 expression were determined using the 2–ΔΔCt method. Relative gene levels were expressed as ratios to control. Western Blot Assay The dissected renal tissues were homogenized in ice-cold lysis buffer, which included 50 mm Tris (pH 7.4), 1% Triton X-100, 150 mm NaCl, 1% sodium deoxycholate, 0.1% sodium dodecyl sulfate, and protease inhibitors. After being centrifuged at 12,000g for 15 minutes at 4°C, the supernatant was collected. Samples (50 μg per lane) were loaded and then separated on a sodium dodecyl sulfate-polyacrylamide gel and transferred to a polyvinylidene difluoride membrane. The membrane was blocked with 5% nonfat milk and incubated with the primary antibodies against programmed cell death protein 4 (PDCD4) (rabbit polyclonal 1:1,000; Novus, Littleton, CO), phosphatase and tensin homolog deleted on chromosome 10 (PTEN) (rabbit monoclonal 1:1,000; Abcam), total protein kinase B (Akt) (rabbit monoclonal 1:1,000; Cell Signaling Technology, Danvers, MA), phospho-Akt (rabbit monoclonal 1:1,000; Cell Signaling Technology), B-cell lymphoma-2 (Bcl-2, rabbit monoclonal 1:1,000; Cell Signaling Technology), IκB-α and phospho-IκB-α (mouse monoclonal 1:1,000; Cell Signaling Technology) overnight at 4°C, then incubated with a horseradish peroxidase-conjugated secondary antibody, and developed by chemiluminescent Horseradish Peroxidase Substrate (Millipore, Billerica, MA). Results were normalized with glyceraldehyde-3-phosphate dehydrogenase or β-actin and expressed as ratios to control.

ng Technology) overnight at 4°C, then incubated with a horseradish peroxidase-conjugated secondary antibody, and developed by chemiluminescent Horseradish Peroxidase Substrate (Millipore, Billerica, MA). Results were normalized with glyceraldehyde-3-phosphate dehydrogenase or β-actin and expressed as ratios to control. Statistical Analysis Statistical analysis was performed using the statistical software SPSS Version 16.0 (SPSS, Chicago, IL). All numerical data were presented as a mean ± sem. For comparison of means between two groups, two-tailed, unpaired t tests were used. For comparison of means between three or more groups, one-way analysis of variance with Bonferroni posttest was applied. The values of score were presented as a class variable and analyzed by the Mann-Whitney or Kruskal-Wallis nonparametric test. A p value of less than 0.05 was considered significant.

tests were used. For comparison of means between three or more groups, one-way analysis of variance with Bonferroni posttest was applied. The values of score were presented as a class variable and analyzed by the Mann-Whitney or Kruskal-Wallis nonparametric test. A p value of less than 0.05 was considered significant. RESULTS Xenon Exposure Protected Against LPS-Induced AKI We demonstrated that systemic injection of LPS resulted in an increase in Scr level in a time-dependent manner in mice, which peaked at 24 hours after injection (control-24 hr, 0.37 ± 0.04 mg/dL vs LPS-24 hr, 0.78 ± 0.10 mg/dL; p = 0.000) (Fig. 1A). A 2-hour xenon preconditioning provided functional and morphologic renoprotection against LPS-induced AKI. Compared with the mice in LPS group and nitrogen + LPS group, mice in the xenon + LPS group showed a significant decrease in Scr 24 hours after LPS injection (xenon + LPS, 0.57 ± 0.12 mg/dL vs LPS, 0.71 ± 0.07 mg/dL, p = 0.000; or vs nitrogen + LPS, 0.69 ± 0.09 mg/dL, p = 0.040) (Fig. 1B). Parallel to the deterioration of renal function, morphological damage occurred in the kidneys after LPS administration, characterized by tubular cell vacuolization, loss of brush border, tubule dilation, cast formation, and interstitial inflammatory cells infiltration, but minimal tubular cell necrosis. The regions were prominently localized in the cortex and outer stripe of the outer medulla, whereas the inner stripe of the outer medulla was less affected. In contrast, mice pretreated with xenon showed mild morphological damage (Fig. 1C). Morphologic evidence was assessed using histopathologic scoring, and the score in LPS group and nitrogen pretreatment group was significantly higher than that in xenon pretreatment group (Fig. 1D).

f the outer medulla was less affected. In contrast, mice pretreated with xenon showed mild morphological damage (Fig. 1C). Morphologic evidence was assessed using histopathologic scoring, and the score in LPS group and nitrogen pretreatment group was significantly higher than that in xenon pretreatment group (Fig. 1D). Figure 1. Xenon preconditioning protected against lipopolysaccharide (LPS)-induced renal dysfunction. A and B, Serum creatinine concentration. Mice received intraperitoneal injection of LPS (20 mg/kg). A, Serum creatinine was measured at 2, 6, 12, 24, and 48 hr after LPS injection. Control (n = 6 for each time point), LPS (2, 6, 12, and 24 hr, n = 6 for each time point; LPS-48 hr, n = 5, for one died at 48 hr after LPS injection), *p < 0.05 and **p < 0.01 compared with control group at each time point. For different time points of LPS group, #p < 0.01 compared with LPS-12 hr. B, Mice were pretreated with xenon (Xe) or nitrogen (N2) for 2 hr. Serum creatinine was measured 24 hr after LPS injection. n = 8 mice/group. *p < 0.05 and **p < 0.01 compared with LPS group. C and D, Morphologic injury in kidney and quantification of histologic scoring 24 hr after LPS injection. Kidney sections were stained with hematoxylin and eosin (HE) and photographed at ×100 magnification. Arrow indicates damage in renal tubules. n = 8 mice/group. **p < 0.01 compared with LPS group. Bar = 50 μm. E, Morphologic injury in liver 24 hr after LPS injection. Liver sections were stained with HE and photographed at ×100 magnification. Arrow indicates necrosis of hepatic cells. F, Serum alanine aminotransferase (ALT) was measured 24 hr after LPS injection. n = 8 mice/group. **p < 0.01 compared with LPS group. Data are mean ± sem.

gic injury in liver 24 hr after LPS injection. Liver sections were stained with HE and photographed at ×100 magnification. Arrow indicates necrosis of hepatic cells. F, Serum alanine aminotransferase (ALT) was measured 24 hr after LPS injection. n = 8 mice/group. **p < 0.01 compared with LPS group. Data are mean ± sem. In addition, we evaluated the effects of xenon on liver injury induced by LPS. At 24 hours after LPS challenge, mild to moderate necrosis of hepatic cells was observed in the liver tissue of LPS-challenged mice and LPS-challenged mice pretreated with nitrogen. However, liver tissue injury was attenuated in the LPS-challenged mice pretreated with xenon (Fig. 1E). Parallel to the changes of morphology, the level of serum ALT was reduced significantly in xenon + LPS group (56.0 ± 12.1 IU/L) when compared with LPS group (82.6 ± 9.1 IU/L, p = 0.000) and nitrogen + LPS group (79.5 ± 13.6 IU/L, p = 0.005) (Fig. 1F).

tenuated in the LPS-challenged mice pretreated with xenon (Fig. 1E). Parallel to the changes of morphology, the level of serum ALT was reduced significantly in xenon + LPS group (56.0 ± 12.1 IU/L) when compared with LPS group (82.6 ± 9.1 IU/L, p = 0.000) and nitrogen + LPS group (79.5 ± 13.6 IU/L, p = 0.005) (Fig. 1F). Xenon Exposure Suppressed LPS-Induced Systemic Inflammation as well as Renal Inflammation, Apoptosis To explore the effect of xenon preconditioning on inflammatory response, we performed immunohistochemical study to localize lymphocyte antigen 6 (Ly-6G/-6C) expression to evaluate changes in the number of neutrophils in kidneys. Ly-6G/-6C was upregulated and prominent in the renal interstitial space 24 hours after LPS injection. Pretreatment with xenon significantly decreased the LPS-induced neutrophils infiltration (Fig. 2, A and B). Meanwhile, to assess systemic and kidney inflammation, we examined multiple cytokines in the circulation and kidneys at 24 hours after LPS injection. The concentrations of inflammatory cytokines, TNF-α, and IL-6 in serum and kidneys significantly increased, and the concentration of anti-inflammatory cytokine IL-10 moderately increased concomitantly. However, pretreatment with xenon significantly decreased the concentrations of TNF-α and IL-6 and further increased the concentration of IL-10 in serum and kidneys 24 hours after LPS administration when compared with LPS group and nitrogen + LPS group (Fig. 2, C–H). In addition, quantitative real-time reverse transcription-PCR analysis showed that TNF-α, IL-6, and IL-10 mRNA expression were also upregulated after LPS administration, parallel to the elevation of protein levels. Likewise, pretreatment with xenon significantly decreased the mRNA levels of TNF-α and IL-6 and further increased IL-10 mRNA level in kidneys 24 hours after LPS administration, when compared with the LPS group and nitrogen + LPS group (Fig. 2, I–K).

after LPS administration, parallel to the elevation of protein levels. Likewise, pretreatment with xenon significantly decreased the mRNA levels of TNF-α and IL-6 and further increased IL-10 mRNA level in kidneys 24 hours after LPS administration, when compared with the LPS group and nitrogen + LPS group (Fig. 2, I–K). Figure 2. Neutrophil infiltration and cytokines production 24 hr after lipopolysaccharide (LPS) injection with or without gas preconditioning. A, Immunohistochemical staining for neutrophil infiltration in kidneys. Representative photomicrographs, ×100 magnification, with brown color indicating positive staining. Bar = 50 μm. B, Mean value of staining-positive cells in renal sections. C–H, The concentrations of inflammatory and anti-inflammatory cytokines in serum and kidneys. C–E, Enzyme-linked immunosorbent assay (ELISA) of interleukin (IL)-6, tumor necrosis factor (TNF)-α, and IL-10 production in serum. F–H, ELISA of IL-6, TNF-α, and IL-10 production in kidneys. I–K, Quantitative real-time reverse transcription polymerase chain reaction analysis of IL-6, TNF-α, and IL-10 messenger RNAs (mRNAs) in kidneys. n = 6 mice/group. *p < 0.05 and **p < 0.01 compared with LPS group. Data are mean ± sem. N2 = nitrogen, Xe = xenon.

–H, ELISA of IL-6, TNF-α, and IL-10 production in kidneys. I–K, Quantitative real-time reverse transcription polymerase chain reaction analysis of IL-6, TNF-α, and IL-10 messenger RNAs (mRNAs) in kidneys. n = 6 mice/group. *p < 0.05 and **p < 0.01 compared with LPS group. Data are mean ± sem. N2 = nitrogen, Xe = xenon. To estimate renal injury at cellular level, TUNEL staining was used to analyze apoptosis of renal cells. Quantitatively, TUNEL-positive cells in kidneys were significantly fewer in xenon pretreatment group (8 ± 2 per high-power field) than those in LPS group (17 ± 4 per high-power field; p = 0.000) and nitrogen pretreatment group (16 ± 5 per high-power field; p = 0.000) (Fig. 3, A and B).

analyze apoptosis of renal cells. Quantitatively, TUNEL-positive cells in kidneys were significantly fewer in xenon pretreatment group (8 ± 2 per high-power field) than those in LPS group (17 ± 4 per high-power field; p = 0.000) and nitrogen pretreatment group (16 ± 5 per high-power field; p = 0.000) (Fig. 3, A and B). Figure 3. Renal cell apoptosis, miR-21, and apoptosis-related protein expression in mouse kidney 24 hr after lipopolysaccharide (LPS) administration following xenon exposure. A, Terminal deoxynucleotidyl transferase-mediated dUTP nick-end labeling (TUNEL)-positive cells in renal sections from all groups, photographed at ×100 magnification. Bar = 50 μm. B, Mean value of TUNEL-positive cells in renal sections (n = 6 mice/group). **p < 0.01 compared with LPS group. C, Mice were exposed to air (control),70% nitrogen (N2), or 70% xenon (Xe), and miR-21 expression in kidneys was detected 24 hr after gas exposure (n = 4 mice/group). *p < 0.05 compared with control group. D, Xenon preconditioning increased miR-21 expression 24 hr after LPS injection (n = 6 mice/group). **p < 0.01 compared with LPS group. E–G, Western blot analysis of phosphatase and tensin homolog deleted on chromosome 10 (PTEN), protein kinase B (Akt)/phospho-Akt, B-cell lymphoma (Bcl)-2, and cleaved caspase-3 in kidneys 24 hr after LPS injection with or without xenon preconditioning. Xenon suppressed PTEN expression (E), upregulated phospho-Akt expression (F), and upregulated Bcl-2 and downregulated cleaved caspase-3 expression (G). Data are representative of three independent experiments. n = 6 mice/group. *p < 0.05 and **p < 0.01 compared with LPS group. Data are mean ± sem.

enon preconditioning. Xenon suppressed PTEN expression (E), upregulated phospho-Akt expression (F), and upregulated Bcl-2 and downregulated cleaved caspase-3 expression (G). Data are representative of three independent experiments. n = 6 mice/group. *p < 0.05 and **p < 0.01 compared with LPS group. Data are mean ± sem. Xenon Preconditioning Caused Upregulation of miR-21 and Inhibition of Proapoptotic Signaling Pathway PTEN/Akt in Kidney In our previous study, we observed that xenon preconditioning significantly upregulated miR-21 expression in mouse kidney in time-dependant manner (17). Here, we similarly found xenon significantly increased miR-21 expression in mouse kidney 24 hours after xenon exposure or 24 hours after LPS injection following xenon preconditioning (Fig. 3, C and D). It has been shown that miR-21 is a strong antiapoptotic factor (12, 13). We further measured the target effector of miR-21, PTEN, a proapoptotic factor, and Akt pathway, which is negatively regulated by PTEN. Mice receiving xenon pretreatment showed a significant downregulation of PTEN and upregulation of p-Akt when compared with mice receiving nitrogen pretreatment or only LPS administration without gas pretreatment (Fig. 3, E and F). Furthermore, in comparison with LPS group and nitrogen + LPS group, mice receiving xenon pretreatment caused a significant increase in expression of antiapoptotic Bcl-2 and a decrease in activity of caspase-3 in kidneys (Fig. 3G).

rogen pretreatment or only LPS administration without gas pretreatment (Fig. 3, E and F). Furthermore, in comparison with LPS group and nitrogen + LPS group, mice receiving xenon pretreatment caused a significant increase in expression of antiapoptotic Bcl-2 and a decrease in activity of caspase-3 in kidneys (Fig. 3G). Xenon Preconditioning Inhibited Expression of PDCD4 and Activity of NF-κB To understand the mechanism whereby xenon preconditioning blocked LPS-induced cytokine production (TNF-α and IL-6) and increased anti-inflammatory factor (IL-10), we examined the expression of PDCD4, a proinflammatory protein and target effector of miR-21, and activity of NF-κB. Compared with the mice in LPS group and nitrogen + LPS group, mice receiving xenon pretreatment caused a significant downregulation of PDCD4 (Fig. 4A), and p-IκB-α was also markedly downregulated in mice receiving xenon pretreatment (Fig. 4B), which was pivotal to NF-κB activation. We then examined NF-κB activity in kidneys 24 hours after LPS injection with ELISA-based TransAM method (20) and found LPS administration strongly increased NF-κB activity. This increase in NF-κB activity was significantly attenuated in mice pretreated with xenon (Fig. 4C).