Browse the corpus

Abbreviations BSAbody surface area CaO2arterial oxygen content CIcardiac index DO2oxygen delivery FIO2fraction of inspired oxygen Hbhemoglobin HRheart rate ICUintensive care unit IFDintermittent flow density MAPmean arterial pressure MFImicrovascular flow index PaO2arterial partial pressure of oxygen PPVproportion of perfused vessels PVDperfused vessel density SaO2arterial oxygen saturation SDFsidestream darkfield imaging SVIstroke volume index SVRsystemic vascular resistance SVRIsystemic vascular resistance index VDvessel density 1 INTRODUCTION Supplemental oxygen is administered to patients with arterial hypoxemia to ensure sufficient DO2 to organs. However, in clinical practice, physicians are inclined to administer oxygen profusely, even in patients who are not hypoxemic.1, 2, 3 As a result, supraphysiological oxygen tensions (hyperoxia) are frequently encountered.4 Restoring normal arterial oxygen tensions (PaO2) is obviously beneficial in hypoxemic patients, but it is uncertain whether oxygen supplementation beyond normoxia is safe and actually improves DO2.

nts who are not hypoxemic.1, 2, 3 As a result, supraphysiological oxygen tensions (hyperoxia) are frequently encountered.4 Restoring normal arterial oxygen tensions (PaO2) is obviously beneficial in hypoxemic patients, but it is uncertain whether oxygen supplementation beyond normoxia is safe and actually improves DO2. Hyperoxia may increase ICU mortality5, 6, 7 and myocardial infarct size.8 On the other hand, moderate hyperoxia may alleviate organ dysfunction after cardiac arrest.9 In mechanically ventilated ICU patients, the (retrospective) relation between the degree of hyperoxia and mortality is U‐shaped, with a nadir around 15‐20 kPa.10 Potential adverse effects of hyperoxia may occur via microvascular constriction11, 12 and a reduction in cardiac output.13, 14, 15, 16 However, findings regarding such effects are ambiguous.17, 18 The reduced perfusion and cardiac output may lead to a net loss of DO2 that had been found in some,19, 20, 21 but not all studies.22, 23, 24

effects of hyperoxia may occur via microvascular constriction11, 12 and a reduction in cardiac output.13, 14, 15, 16 However, findings regarding such effects are ambiguous.17, 18 The reduced perfusion and cardiac output may lead to a net loss of DO2 that had been found in some,19, 20, 21 but not all studies.22, 23, 24 The evidence for hyperoxia causing microvascular constriction mostly comes from animal studies. In humans, the effects of hyperoxia on the microvasculature consists of indirect measures, such as an increase in SVR15, 25, 26, 27 or a reduction in peripheral blood flow.28, 29, 30, 31 Recently, a direct effect of hyperoxia on the sublingual microcirculation was shown.32 In this study, a marked decrease in PVD (−30%) was observed, when 10 healthy volunteers breathed pure oxygen for 30 minutes. However, as with most studies on hyperoxia, only 2 inspired oxygen concentrations were studied; air (21% O2) and pure oxygen (100%). Although this comparison creates the highest contrast, its clinical relevancy is limited. An FIO2 of 1.0 is rarely used in daily practice to avoid the direct toxicity of pure oxygen to the lungs. Second, the PaO2s that arise from breathing pure oxygen by healthy volunteers is not comparable to the ones in patients with existing lung pathology. As a result, the relation between hemodynamic effects of oxygen and PaO2 at clinically relevant doses remains unknown.

id the direct toxicity of pure oxygen to the lungs. Second, the PaO2s that arise from breathing pure oxygen by healthy volunteers is not comparable to the ones in patients with existing lung pathology. As a result, the relation between hemodynamic effects of oxygen and PaO2 at clinically relevant doses remains unknown. Only a few groups investigated the dose‐response effect of oxygen on the cardiovascular system15, 26 and none directly visualized the microcirculation. It is therefore currently unknown at which PaO2 the microcirculatory effects of hyperoxia start to occur and what the nature of the dose‐response effect is. The aim of this study was to determine the dose‐response relationship between a stepwise increase in PaO2 and its associated changes in DO2 and sublingual microcirculatory perfusion. 2 MATERIALS AND METHODS 2.1 Study design and ethical approval Single‐blind, cross‐over physiological study with healthy volunteers performed at the ICU of the VU University Medical Centre (Amsterdam, the Netherlands). The study protocol was approved by the Dutch Central Committee on Research Involving Human Subjects (NL5816602916) and conformed to the standards set by the Declaration of Helsinki. 2.2 Subjects Volunteers were recruited through social media and were eligible for participation if they were 18 years or older and had no medical history of pulmonary or cardiovascular disease. A modified Allen test was performed to assess arterial competency, and subjects without a patent ulnar artery were not included. Subjects were included after written informed consent was obtained.

ligible for participation if they were 18 years or older and had no medical history of pulmonary or cardiovascular disease. A modified Allen test was performed to assess arterial competency, and subjects without a patent ulnar artery were not included. Subjects were included after written informed consent was obtained. 2.3 Protocol 2.3.1 Preparation Subjects lay in a semirecumbent position in a temperature‐controlled room at the ICU. After application of a local anesthetic (lidocaine), the radial artery was cannulated for blood sampling and blood pressure measurements. A finger cuff was placed on the index or middle finger for continuous measurement of hemodynamic parameters by the volume‐clamp method, according to the manufacturer's instructions (Nexfin®, BMEYE, Amsterdam, the Netherlands). Finally, subjects were fitted with a noninvasive ventilation mask coupled to a SERVO‐I mechanical ventilator (Maquet, Rastatt, Germany). The ventilator was set to provide zero continuous positive airway pressure or pressure support. When the subjects were accustomed to the setup (~15 minutes after radial artery cannulation), the intervention and measurements were started.

lation mask coupled to a SERVO‐I mechanical ventilator (Maquet, Rastatt, Germany). The ventilator was set to provide zero continuous positive airway pressure or pressure support. When the subjects were accustomed to the setup (~15 minutes after radial artery cannulation), the intervention and measurements were started. 2.3.2 Intervention The FIO2 was adjusted to reach target PaO2s of baseline (kPa while breathing air), 20, 40, 60 kPa, and max kPa (while breathing pure oxygen) during 5 separate phases. Five minutes into each phase, arterial blood gas analysis was performed and the FIO2 was adjusted once if PaO2 was not at the intended target. After an additional 5 minutes, a second arterial blood gas was taken. When all study measurements were performed (see below), the subject rested 5‐10 minutes before moving on to the next PaO2 target. Subjects knew they would inspire FIO2s between 21%‐100%, but were unaware of the predetermined stepwise increase. Monitors and the control of FIO2 were not visible for the participants.

ken. When all study measurements were performed (see below), the subject rested 5‐10 minutes before moving on to the next PaO2 target. Subjects knew they would inspire FIO2s between 21%‐100%, but were unaware of the predetermined stepwise increase. Monitors and the control of FIO2 were not visible for the participants. 2.4 Measurements At the end of each period, the NIV mask was removed and the sublingual microcirculation was visualized immediately (within one minute) with SDF (MicroVision Medical BV, Amsterdam, the Netherlands). In SDF imaging, green light is emitted from the device which is then absorbed by the Hb present in erythrocytes. SDF therefore relies on the presence of Hb to visualize blood vessels. Three to 5 sites were recorded and analyzed in accordance with the latest quality recommendations.33 After acquisition, the video files were stored for blinded offline semiquantitative analysis with the Automated Vascular Analysis software 3.1 (MicroVision Medical BV). In short, a grid of 5 equidistant vertical and horizontal lines is placed on top of the recording. Vessels crossing these lines are counted and classified as having either continuous, slow/sluggish, intermittent, or no flow. Vascular density (VD) is reported as the total number of vessels per mm of grid. PVD is comprised of vessels showing only continuous or slow/sluggish flow. Although not regularly reported, we also calculated the number of intermittent perfused vessels (IFD) in a similar fashion. All recordings and analyses were carried out by the same operator (BS). All data reported pertain to small vessels with a diameter of 20 μm or less.

showing only continuous or slow/sluggish flow. Although not regularly reported, we also calculated the number of intermittent perfused vessels (IFD) in a similar fashion. All recordings and analyses were carried out by the same operator (BS). All data reported pertain to small vessels with a diameter of 20 μm or less. Heart rate, CI, SVI, and SVRI were measured continuously during the entire experiment. MAP was measured via the arterial line. The average of the last 2 minutes of each exposure was used for statistics. Blood gas and metabolite parameters were measured on‐site with an ABL800 FLEX analyzer (Radiometer, Copenhagen, Denmark). 2.5 Calculations Systemic DO2 index was calculated by multiplying CI with the CaO2. For the latter, the following formula was used; CaO2 = (Hb [g/dL] × 10 × 1.36 × SaO2) + (0.0031 × [PaO2 (kPa) × 7.5]). 2.6 Statistics All values are reported as mean and standard deviation unless stated otherwise. Dose‐response relations were primarily fitted with a linear regression based on the parameters at each kPa target. An additional nonlinear regression was performed if deemed warranted based on visual inspection. Fit performance was assessed visually and by means of the Sy.x statistic (standard deviation of errors in regression). For all data, we tested whether the slope was statistically different from zero. All graphs and statistics were carried out with GraphPad Prism 7.0 (GraphPad Software, Inc., La Jolla, CA, USA).

ection. Fit performance was assessed visually and by means of the Sy.x statistic (standard deviation of errors in regression). For all data, we tested whether the slope was statistically different from zero. All graphs and statistics were carried out with GraphPad Prism 7.0 (GraphPad Software, Inc., La Jolla, CA, USA). 3 RESULTS 3.1 Volunteers and measurements Baseline characteristics of the 15 included volunteers are listed in Table 1. All participants gave written informed consent and completed the entire protocol without adverse events. On average, the duration of the study was 95 minutes (SD 8). In one subject, continuous measurement of hemodynamic parameters by the volume‐clamp method was omitted because a stable, valid waveform could not be obtained (due to peripheral vasoconstriction). Sublingual measurements and MAP values obtained through the arterial line did not differ from the other participants. Therefore, the sublingual data of this participant were included in the final analysis. Table 1 Baseline characteristics Variable Participants Number 15 Gender (male/female) 7/8 Age (y) 30 (9) Height (cm) 175 (9) Weight (kg) 71 (11) BSA (m2) 1.85 (0.21) BSA, body surface area.

3 RESULTS 3.1 Volunteers and measurements Baseline characteristics of the 15 included volunteers are listed in Table 1. All participants gave written informed consent and completed the entire protocol without adverse events. On average, the duration of the study was 95 minutes (SD 8). In one subject, continuous measurement of hemodynamic parameters by the volume‐clamp method was omitted because a stable, valid waveform could not be obtained (due to peripheral vasoconstriction). Sublingual measurements and MAP values obtained through the arterial line did not differ from the other participants. Therefore, the sublingual data of this participant were included in the final analysis. Table 1 Baseline characteristics Variable Participants Number 15 Gender (male/female) 7/8 Age (y) 30 (9) Height (cm) 175 (9) Weight (kg) 71 (11) BSA (m2) 1.85 (0.21) BSA, body surface area. John Wiley & Sons, Ltd3.2 Intervention/Respiration/Blood gas The obtained PaO2s during the 5 periods were 14 kPa (SD 1.3), 20 kPa (SD 1.8), 39 kPa (SD 3.0), 57 kPa (SD 4.9), and 73 kPa (SD 4.1). The FIO2s required to reach these oxygen tensions was 21% (SD 0), 29 (SD 2), 55 (SD 3), 80 (SD 5), and 100 (SD 0), respectively. Blood gas analysis revealed no changes in PaCO2 or pH. Glucose and lactate decreased slightly during the study period but remained within normal ranges.

57 kPa (SD 4.9), and 73 kPa (SD 4.1). The FIO2s required to reach these oxygen tensions was 21% (SD 0), 29 (SD 2), 55 (SD 3), 80 (SD 5), and 100 (SD 0), respectively. Blood gas analysis revealed no changes in PaCO2 or pH. Glucose and lactate decreased slightly during the study period but remained within normal ranges. 3.3 Systemic hemodynamic effects The absolute values for all systemic hemodynamics at each study period are reported in Table 2. Hyperoxia resulted in a linear decrease in CI (Figure 1, P slope = .009). The largest decrease occurred while breathing pure oxygen (max −10%, SD 10). This decrease was due to a similar reduction in HR (max −10%, SD 7, P slope = .009), as stroke volume remained unchanged (P slope = .75). MAP was not affected by hyperoxia (P slope = .68). SVRI increased slightly over the entire PaO2 range (Figure 2) to a maximum of +7% (SD 8, P slope = .009). On visual inspection, the SVRI data fitted a second order polynomial equation better than a linear one, although the Sy.x statistics were identical (8.1). Table 2 Measurements Period Air T1 T2 T3 Oxygen P *

3.3 Systemic hemodynamic effects The absolute values for all systemic hemodynamics at each study period are reported in Table 2. Hyperoxia resulted in a linear decrease in CI (Figure 1, P slope = .009). The largest decrease occurred while breathing pure oxygen (max −10%, SD 10). This decrease was due to a similar reduction in HR (max −10%, SD 7, P slope = .009), as stroke volume remained unchanged (P slope = .75). MAP was not affected by hyperoxia (P slope = .68). SVRI increased slightly over the entire PaO2 range (Figure 2) to a maximum of +7% (SD 8, P slope = .009). On visual inspection, the SVRI data fitted a second order polynomial equation better than a linear one, although the Sy.x statistics were identical (8.1). Table 2 Measurements Period Air T1 T2 T3 Oxygen P * Intervention (n = 15) Target (kPa) – 20 (1.5) 40 (1.5) 60 (1.5) – – PaO2 (kPa) 14 (1.3) 20 (1.8) 39 (3.0) 57 (4.9) 73 (4.1) – FIO2 (%) 21 (0) 29 (2) 55 (3) 80 (5) 100 (0) – Arterial blood gas (n = 15) SaO2 (%) 98 (1) 99 (1) 100 (0) 100 (1) 100 (1) <.001 PaCO2 (kPa) 4.7 (0.7) 4.8 (0.6) 4.6 (0.8) 4.8 (0.5) 4.8 (0.5) .98 pH 7.4 (.04) 7.4 (.03) 7.5 (.05) 7.4 (.04) 7.4 (.03) .51 Hb (mmol L−1) 8.6 (0.8) 8.5 (0.9) 8.6 (0.8) 8.6 (0.7) 8.6 (0.7) .84 Glucose (mmol L−1) 6.3 (1.3) 6.1 (1.2) 6.0 (1.0) 5.7 (0.8) 5.6 (0.6) .028 Lactate (mmol L−1) 1.0 (0.6) 0.9 (0.4) 0.9 (0.4) 0.8 (0.3) 0.7 (0.2) .015 Microcirculation (n = 15) VD (n/mm) 8.0 (0.9) 8.0 (1.0) 8.0 (1.8) 7.6 (1.2) 6.9 (1.5) .02 PVD (n/mm) 7.8 (1.0) 7.8 (1.0) 7.6 (1.6) 6.9 (1.0) 6.5 (1.5) .001 IFD (n/mm) 0.07 (.09) 0.16 (.14) 0.30 (.31) 0.46 (.32) 0.23 (.27) <.001 PPV (%) 98 (2.2) 97 (1.9) 96 (3.8) 91 (3.7) 93 (8.3) <.001 MFI 3.0 (.09) 2.9 (.15) 2.9 (.22) 2.8 (.29) 2.7 (.31) <.001 Hemodynamics (n = 14) HR (bpm) 64 (7) 62 (8) 60 (8) 58 (7) 58 (7) .009 SVI (mL min−1m−2) 56 (7) 57 (7) 58 (8) 58 (7) 57 (8) .75 CI (L min−1m−2) 3.6 (0.7) 3.5 (0.6) 3.4 (0.6) 3.3 (0.6) 3.3 (0.6) <.001 SVRI (dyn·s cm−5 m−2) 2142 (359) 2199 (368) 2236 (426) 2317 (431) 2288 (408) .009 MAP (mm Hg) 98 (16) 98 (14) 98 (15) 98 (13) 95 (13) .68 DO2 (n = 14) CaO2 (mL/L) 192 (18) 194 (20) 200 (18) 206 (16) 208 (17) <.001 DO2 (mL min−1m−2) 684 (141) 676 (132) 686 (133) 695 (141) 679 (145) .83 Data are presented as mean (SD). VD, vessel density; PVD, perfused vessel density; PPV, proportion of perfused vessels; MFI, microvascular flow index; IFD, intermittent flow density; HR, heart rate; MAP, mean arterial pressure; SVI, stroke volume index; CI, cardiac index; SVRI, systemic vascular resistance index; CaO2, arterial oxygen content; DO2, oxygen delivery. *P‐values for slope.

essel density; PPV, proportion of perfused vessels; MFI, microvascular flow index; IFD, intermittent flow density; HR, heart rate; MAP, mean arterial pressure; SVI, stroke volume index; CI, cardiac index; SVRI, systemic vascular resistance index; CaO2, arterial oxygen content; DO2, oxygen delivery. *P‐values for slope. John Wiley & Sons, LtdFigure 1 Relation between oxygen content, delivery, and CI. Oxygen content increased linearly with increasing PaO2. Inversely, CI decreased, which resulted in a stable DO 2I over the entire PaO2 range. Gray areas indicate 95% confidence intervals of the fitted curve. PaO2, arterial oxygen tension; CaO2, arterial oxygen content; DO 2I, arterial oxygen delivery index; CI, Cardiac index Figure 2 Dose‐response for PVD and SVRI. Sublingual PVD decreases in a sigmoidal fashion upon increase PaO2. SVRI shows the largest increase up to 54 kPa. Gray areas indicate 95% confidence intervals of the fitted curves. For PVD, the dotted line represents the best‐fit line based on linear regression (line is plotted without 95% CI). PaO2, arterial oxygen tension; SVRI, systemic vascular resistance index; PVD, perfused vessel density; CI, Cardiac index 3.4 Oxygen delivery The oxygen content of arterial blood increased linearly to a maximum of +8% (SD 3, P slope < .0001) when breathing pure oxygen (Table 2, Figure 1). Systemic DO2 index remained unaltered (P slope = .83).

Figure 2 Dose‐response for PVD and SVRI. Sublingual PVD decreases in a sigmoidal fashion upon increase PaO2. SVRI shows the largest increase up to 54 kPa. Gray areas indicate 95% confidence intervals of the fitted curves. For PVD, the dotted line represents the best‐fit line based on linear regression (line is plotted without 95% CI). PaO2, arterial oxygen tension; SVRI, systemic vascular resistance index; PVD, perfused vessel density; CI, Cardiac index 3.4 Oxygen delivery The oxygen content of arterial blood increased linearly to a maximum of +8% (SD 3, P slope < .0001) when breathing pure oxygen (Table 2, Figure 1). Systemic DO2 index remained unaltered (P slope = .83). 3.5 Sublingual microcirculation PaO2 reduced, in a dose‐dependent fashion, vascular density (VD) and perfused vascular density (PVD). Compared to measurements performed at baseline, VD changed with +1% (SD 13), 0% (SD 18), −4% (SD 12), and −13% (SD 17, P = .005) at a PaO2 of 20, 39, 57, and 73 kPa, respectively. Similarly, PVD changed with +0% (SD 15), −3% (SD 18), −11%(SD 13), and −15% (SD 18, P = .003). The data could be fitted with both a straight (P slope < .0001) and a sigmoidal line. The standard deviation of the values around the regression line (Sy.x) was 13.6 and 13.7, respectively. On visual inspection, the sigmoidal curve was a better fit (Figure 2). The number of vessels showing intermittent flow increased linearly up to 57 kPa, but was then relatively reduced at 73 kPa. A representative image of the microcirculation while breathing room air is shown in Figure 3A. When breathing oxygen, the number of perfused vessels was reduced and blood flow became stagnant or intermittent, visible as “dotted” vessels in Figure 3B.

sed linearly up to 57 kPa, but was then relatively reduced at 73 kPa. A representative image of the microcirculation while breathing room air is shown in Figure 3A. When breathing oxygen, the number of perfused vessels was reduced and blood flow became stagnant or intermittent, visible as “dotted” vessels in Figure 3B. Figure 3 Sublingual microcirculation. Representative images of the sublingual microcirculation acquired with the SDF device. Compared to breathing air (A), oxygen supplementation (B) decreased overall VD (vessels crossing the white grid) and caused interrupted flow (asterisk). SDF, sidestream darkfield imaging; VD, vessel density 4 DISCUSSION The main finding of this study is that in healthy volunteers, supplemental oxygen does not alter DO2, while sublingual PVD decreased in a sigmoidal fashion as PaO2 was increased stepwise from 14 up to 73 kPa. Hyperoxia decreased CI, by a reduction in HR rather than stroke volume, and increased SVRI. MAP was unchanged. In this healthy volunteer population, the increase in CaO2 caused by an increase in PaO2 was negated by a simultaneous reduction in CI. Two previous studies in healthy volunteers on hemodynamic effects of oxygen showed a slight decrease in DO2, 24, 34 while another showed no effect.35 In patients, a similar heterogeneity has been seen as DO2 was reduced in 2 studies,22, 23 but remained unaltered in 2 others.17, 19 An important conclusion from our study is that, in nonhypoxemic individuals, an intended increase in DO2 is not achieved by any level of normobaric oxygen supplementation.

wed no effect.35 In patients, a similar heterogeneity has been seen as DO2 was reduced in 2 studies,22, 23 but remained unaltered in 2 others.17, 19 An important conclusion from our study is that, in nonhypoxemic individuals, an intended increase in DO2 is not achieved by any level of normobaric oxygen supplementation. We found a significant reduction in sublingual PVD simultaneous to a stepwise increase in PaO2 in this group of healthy individuals. Changes in the sublingual microcirculation in response to an increase in the FIO2 to 1.0 also occur in patients after coronary artery bypass surgery and in a cohort of mixed ICU patients (postcardiac arrest, neurological defects, polytrauma, sepsis).18, 36 Critical illness is associated with (regional) disturbances in microcirculatory perfusion.37 It is possible that systemic oxygen‐induced changes in blood flow further impair regional perfusion and cause a mismatch between DO2 and demand.38 For example, in an animal model for severe coronary artery stenosis, hyperoxia was found to exacerbate myocardial ischemia due to coronary vasoconstriction.39 In our study, we found no evidence for cellular hypoxia (as indicated by lactate), but this may be different in the critically ill with pre‐existing perfusion defects. Either way, in terms of oxygenation, there appears to be no clear advantage to supraphysiological oxygen tensions; in a best‐case scenario, there is no effect on DO2, and in a worst‐case, an unintended reduction.

indicated by lactate), but this may be different in the critically ill with pre‐existing perfusion defects. Either way, in terms of oxygenation, there appears to be no clear advantage to supraphysiological oxygen tensions; in a best‐case scenario, there is no effect on DO2, and in a worst‐case, an unintended reduction. This is the first study in which dose‐dependent effects of oxygen on the sublingual microcirculation are described. The relation between PaO2 and PVD could be fitted with both a linear and a sigmoidal curve. Both models performed similarly based on the standard deviation of errors in regression. However, the graphical presentation of the data pleads for a sigmoidal relation between PaO2 and PVD. Also, a nonlinear relation is corroborated by observations carried out in hamsters,40, 41 rats,42 and rabbits.43 In another study with healthy volunteers, a reduction in PVD of approximately 30%32 was found, which is much larger than the effect (of ~15%) found in our study. Discrepancies may be explained by differences in study design: prolonged exposure to hyperoxia (30 vs 10 minutes) and an acute exposure to pure oxygen (vs the gradual increase in our study). The SVR, which is an indicator of vasoconstriction, shows a sharp increase after 5‐10 minutes and then further increases by a small amount over the course of an hour in healthy volunteers25, 44 and postoperative critically ill patients.17 SVR does not show a larger increase when volunteers are acutely16, 24, 26, 27, 45, 46, 47, 48 or chronically exposed to oxygen.15, 35, 49, 50, 51 The difference in effect size may therefore be partially explained by the exposure time.

e of an hour in healthy volunteers25, 44 and postoperative critically ill patients.17 SVR does not show a larger increase when volunteers are acutely16, 24, 26, 27, 45, 46, 47, 48 or chronically exposed to oxygen.15, 35, 49, 50, 51 The difference in effect size may therefore be partially explained by the exposure time. The exact mechanism behind the constrictive effect of high arterial oxygen tensions is currently unclear. Reactive oxygen species (ROS) are one possible candidate, considering that in vitro, as PO2 increases, so does the production of superoxide.52 Superoxide reacts heavily with the vasodilator nitric oxide, reducing its bioavailability and thereby causing vasoconstriction. This has been shown directly in porcine coronary arteries53 and indirectly in human studies; where the scavenging of ROS by infusion of high levels of vitamin C reduced or prevented hyperoxic constriction.11, 12, 54 However, the involvement of ROS is not found uniformly.25, 55 Intravital studies in hamsters, rat, and mice suggest that other pathways may be involved, including inactivation of calcium56, 57 and potassium channels57, 58 or the alteration of the metabolism of arachidonic acid.59, 60 For instance, high PO2 was found to decrease the activity of the enzyme cyclooxygenase, reducing the formation of dilating prostaglandins from arachidonic acid.61, 62, 63 Conversely, the production of the vasoconstrictor 20‐HETE from arachidonic acid by the CYP‐450 pathway was shown to be increased by hyperoxia.64, 65, 66 These observations are however not mutually exclusive, different mechanisms of hyperoxic vasoconstriction may be present depending on the vascular bed and/or species under investigation.

duction of the vasoconstrictor 20‐HETE from arachidonic acid by the CYP‐450 pathway was shown to be increased by hyperoxia.64, 65, 66 These observations are however not mutually exclusive, different mechanisms of hyperoxic vasoconstriction may be present depending on the vascular bed and/or species under investigation. Our study shows a small imbalance between the reduction in PVD and SVRI. At higher PaO2s (>57 kPa), the fractional decrease in PVD was larger than the increase in SVRI. Based on the data from our study, we can only speculate why this is the case. One possible explanation is heterogeneity between microvascular beds: Capillary recruitment in the sublingual area may decrease, while recruitment in other areas/organs remains unaltered or increases. As a result, the effect of increasing PaO2 on the PVD in the sublingual microcirculation may be larger than on SVR. In anaesthetized dogs, hyperoxia has been shown to redistribute blood flow to the kidney, liver, and intestines, while blood flow to the myocardium, pancreas, and skeletal muscle decreases.67 The amount of redistribution may vary at different PaO2s. Another explanation is recruitment of arteriovenous shunts, which provide relatively less resistance to flow than smaller arterioles/capillaries and therefore mitigate the increase in SVR. Shunting may also explain the decrease in VO2 that is seen in some studies after oxygen administration.17, 68 A bypass of metabolic active tissue will elevate venous PO2, reducing the arteriovenous oxygen difference used to calculate VO2. However, we did not take venous blood samples to calculate VO2, as this was beyond the scope of our study.

the decrease in VO2 that is seen in some studies after oxygen administration.17, 68 A bypass of metabolic active tissue will elevate venous PO2, reducing the arteriovenous oxygen difference used to calculate VO2. However, we did not take venous blood samples to calculate VO2, as this was beyond the scope of our study. In our study, the majority of hyperoxia‐induced changes became apparent above an oxygen tension of 20 kPa. From a hemodynamic and microcirculatory point of view, arterial oxygen tensions up to 20 kPa could therefore be suggested as “permissive hyperoxia.” Interestingly, the range of 10‐20 kPa is retrospectively associated with lower mortality in critically ill populations5, 10 and with improved organ function after cardiac arrest.9 Slight hyperoxia (up to a PaO2 of 20 kPa) may thus be beneficial, because its influence on perfusion is possibly negligible. However, even within this range, one prospective study in ICU patients showed reduced mortality in a conservative oxygen group (median PaO2 of 11.5 kPa) compared to a conventional oxygen group (median PaO2 of 13.5 kPa). Beside some methodological issues, it should be noted that the study was stopped prematurely due to slow inclusion and was therefore underpowered for a mortality endpoint.69, 70 Large randomized controlled studies in specific critically ill populations are required to determine whether slight hyperoxia is beneficial or not.

e some methodological issues, it should be noted that the study was stopped prematurely due to slow inclusion and was therefore underpowered for a mortality endpoint.69, 70 Large randomized controlled studies in specific critically ill populations are required to determine whether slight hyperoxia is beneficial or not. In healthy volunteers, supraphysiological arterial oxygen tensions, in the range of 14‐73 kPa, have no effect on systemic DO2; however, sublingual microcirculatory PVD decreased in a dose‐dependent fashion. Simultaneous with the increase in CaO2, cardiac output decreased due to a decline in HR rather than stroke volume. SVRI increased slightly, while MAP remained unaltered. All hemodynamic changes appear negligible up to a PaO2 of 20 kPa. 4.1 Limitations Our study has several limitations. For one, we used a noninvasive measurement of systemic hemodynamics which is less precise than the gold standard thermodilution method. The true effect size of hyperoxia on systemic hemodynamics may therefore be different. Second, we included a relatively low number of participants, but the study was adequately powered given the prepost study design. Third, we performed the study in a single‐blind fashion due to the incremental oxygen exposure. This approach was chosen because of the possible residual effects of oxygen inhalation on hemodynamics for up to 30 minutes.27 The risk of operator bias was reduced by blinded analysis of the microvascular recordings and the use of a set time‐period for the averaging of hemodynamic variables.

cremental oxygen exposure. This approach was chosen because of the possible residual effects of oxygen inhalation on hemodynamics for up to 30 minutes.27 The risk of operator bias was reduced by blinded analysis of the microvascular recordings and the use of a set time‐period for the averaging of hemodynamic variables. Fourth, we did not include a control group that exclusively inhaled air during the study. This means we cannot exclude the possibility of time or comfort related changes in hemodynamics (eg, a reduction in HR due to increased comfort, rather than oxygen). However, we think it is highly unlikely that an increased level of comfort toward the end of the study is a factor in our results; all volunteers had ample time to adjust to the measurements before the start of the study and they showed no signs of anxiety at any time (eg, raised blood pressure). Our results on HR, MAP and stroke volume are in line with several other studies performed in healthy volunteers.34, 35, 45, 49, 51, 71, 72 HR and sublingual microcirculatory perfusion decreased dose‐dependently throughout the entire protocol; if there was no interaction with the intervention and the effect was solely due to increased comfort, we would have expected the effect to stabilize after an initial 20 or 30 minutes. Also, we are unaware of any mechanism that may link comfort with reduced sublingual perfusion. However, we cannot completely exclude a partial effect of comfort in our results; therefore, we advise future studies to include either a time control (eg, a group inhaling air only) or a phase with return to baseline (eg, air after oxygen exposure).

ny mechanism that may link comfort with reduced sublingual perfusion. However, we cannot completely exclude a partial effect of comfort in our results; therefore, we advise future studies to include either a time control (eg, a group inhaling air only) or a phase with return to baseline (eg, air after oxygen exposure). Fifth, the sublingual microcirculation may not be representative for other parts of the human body. However, we chose to investigate the sublingual area with SDF because of its noninvasive nature and has been used extensively in studies with the critically ill; it is a clinically relevant area as it is correlated with mortality and organ failure in patients with cardiogenic shock73 and sepsis74 and with postoperative complications after abdominal surgery.75 Also, the sublingual area is perfused directly from the carotis externa, making it very closely related to the central circulation. PERSPECTIVE Despite decades of research into the cardiovascular effects of hyperoxia, the effects of clinically relevant high arterial oxygen tensions (hyperoxia) on systemic DO2 and sublingual microcirculatory perfusion are currently unknown. In this arterial oxygen tension guided study, we found that in healthy volunteers, any level of normobaric oxygen supplementation has no effect on systemic DO2. Simultaneously, sublingual PVD was substantially decreased in the hyperoxic range of 20‐73 kPa. Due to the prevalence of hyperoxia in critically ill patients, these findings warrant studies to determine whether hyperoxia exacerbates pre‐existing microcirculatory defects.

In this arterial oxygen tension guided study, we found that in healthy volunteers, any level of normobaric oxygen supplementation has no effect on systemic DO2. Simultaneously, sublingual PVD was substantially decreased in the hyperoxic range of 20‐73 kPa. Due to the prevalence of hyperoxia in critically ill patients, these findings warrant studies to determine whether hyperoxia exacerbates pre‐existing microcirculatory defects. CONFLICT OF INTEREST The authors declare that they have no conflicting interests. ACKNOWLEDGMENTS We would like to thank all the volunteers for their participation and E. Alberts for her assistance during the experiments.

Abbreviations A.U.arbitrary units AEadverse event BMIbody mass index CAPcold atmospheric plasma DBDdielectric barrier discharge diCAPdirect cold atmospheric plasma O2Coxygen‐to‐see rHblocal relative hemoglobin ROIregion of interest RONSreactive oxygen and nitrogen species SAEserious adverse event SDStandard deviation SO2postcapillary oxygen saturation t testStudent's t test tcPO2transcutaneous oxygen pressure TTPtissue tolerable plasma 1 INTRODUCTION In the context of medicine, the term “plasma” is typically used in conjunction with the uncoagulated cell‐free component of blood. In physics, however, “plasma” refers to the fourth state of matter—ionized gases with unique physical and chemical properties. As multicomponent systems, physical plasmas consist of neutral gas particles, charged particles (ions and electrons) as well as strongly gas‐dependent reactive gas species, and photons. The free electrons in the plasma induce excitation, ionization, and dissociation processes and are mainly responsible for the changes of physico‐chemical properties from inert gases to chemically reactive plasma.

charged particles (ions and electrons) as well as strongly gas‐dependent reactive gas species, and photons. The free electrons in the plasma induce excitation, ionization, and dissociation processes and are mainly responsible for the changes of physico‐chemical properties from inert gases to chemically reactive plasma. Plasma is usually generated using strong electric fields to induce gas breakdown. As a consequence, electric fields are of major importance for plasma engineering and can be considered as an immanent plasma component. For decades, physical plasmas have been applied in various and versatile processes, for example, to modify surface wettability or deposit functional coatings, for material etching or superficial cleaning and sterilization as well as for radiation generation.1 Most of these applications were performed at reduced pressure environments or at process temperatures that would have induced thermal damaging of organic substrates. Technical improvements over the last 15‐20 years have enabled the generation of atmospheric pressure plasmas with gas temperatures as low as room temperature.2, 3, 4, 5, 6 These CAP, TTP or, in a physically wider sense, nonthermal plasmas have extended the application field from classical inorganic surfaces to organic surfaces and even living tissue. Research and applications in the latter are often referred to as plasma medicine.7

as temperatures as low as room temperature.2, 3, 4, 5, 6 These CAP, TTP or, in a physically wider sense, nonthermal plasmas have extended the application field from classical inorganic surfaces to organic surfaces and even living tissue. Research and applications in the latter are often referred to as plasma medicine.7 Currently available plasma sources can be grouped into direct (diCAP) and indirect sources depending on their characteristic electrode configuration and treatment modalities.8, 9 For diCAP, the object to be treated is typically part of the electrical circuit and acts as a second or third electrode. As a consequence, the object surface is in direct contact with the active plasma zone and charged species as well as short‐lived may contribute to plasma interaction. For indirect CAP sources, the plasma is typically ignited by a self‐containing electrode configuration inside a cavity or restricted to a surface layer. Thereby, fluxes of charged and short‐lived species to the object surface are smaller and long‐lived species transport may be most relevant to induce biological effects. In order to maintain a necessary constant gas gap between electrode and tissue, some direct sources need to be pressed gently on the tissue surface during application thus inevitably exerting a mechanical pressure on the treated area.

r and long‐lived species transport may be most relevant to induce biological effects. In order to maintain a necessary constant gas gap between electrode and tissue, some direct sources need to be pressed gently on the tissue surface during application thus inevitably exerting a mechanical pressure on the treated area. In plasma medicine, several new fields of research have developed, for example, in oncology,10, 11 in dermatology,12, 13 and in dentistry.14, 15 Various effects induced by CAP have been observed. Consistently, a reduction in a wide range of microorganisms was found in vitro and in vivo by groups all over the globe.16, 17, 18, 19, 20, 21, 22, 23 Further effects include the induction of angiogenesis,24 stimulation or inhibition of cell proliferation,25 up‐ and downregulation of genes in skin cells,26 and virus inactivation.27 Recent studies have revealed a significant impact of short‐term diCAP application (up to 90 seconds) on the microcirculation of skin.28, 29, 30 However, in the literature, the role of mechanical pressure induced by the diCAP device was not addressed. As biological tissues are known to react to mechanical forces by, for example, changes in blood flow,31 it is so far unclear, if the device‐induced pressure to the tissue should be considered a confounding factor during diCAP application. Furthermore, it is unclear, how the tissue would react to longer treatment times with diCAP. Consequently, in this study, we set out to study the effect of diCAP application for 90 seconds, 180 seconds, and 270 seconds, in a controlled mechanical environment.

be considered a confounding factor during diCAP application. Furthermore, it is unclear, how the tissue would react to longer treatment times with diCAP. Consequently, in this study, we set out to study the effect of diCAP application for 90 seconds, 180 seconds, and 270 seconds, in a controlled mechanical environment. 2 MATERIALS AND METHODS 2.1 Ethical approval The study was approved by the local ethics commission (ethic committee University Medicine Goettingen, 7/8/16) and followed the Declaration of Helsinki. Prior to the experimental participation, the subjects signed the informed written consent. 2.2 Participants and ROI Healthy subjects (N = 10, two female and eight male) aged ≥ 18 years participated in this study. The mean age was 29.0 ± 3.4 (range: 26‐38) years. None of them presented soft tissue injuries or skin inflammation around the area of the tested skin of the dorsal forearm during the whole investigation period. They did not report comorbidities such as vasculitis, diabetes mellitus, chronic kidney or liver disease, or cardiac dysfunction. Two subjects were smokers. The mean of the BMI was 26.9 ± 3.9 kg/m² (range: 22.8‐34.3 kg/m²). The dorsal forearm of each subject was defined as the ROI.

uring the whole investigation period. They did not report comorbidities such as vasculitis, diabetes mellitus, chronic kidney or liver disease, or cardiac dysfunction. Two subjects were smokers. The mean of the BMI was 26.9 ± 3.9 kg/m² (range: 22.8‐34.3 kg/m²). The dorsal forearm of each subject was defined as the ROI. 2.3 Experimental protocol It has been observed in preliminary tests and is also known from the literature that local thermal changes, body movements, level of attention, and emotional stress can affect microcirculation parameters of the dorsal forearm.32, 33, 34, 35 To address these issues and control their impact on microcirculation parameters throughout the studies, we developed a rigid experimental protocol (Table 1) and constructed a customized mount for diCAP electrode placement and probe positioning in the ROI as depicted in Figure 1. The probe was integrated in the mount in such a way that an even contact face could be realized and the weight pressure of the mount was evenly distributed across its complete area. The probe measures in an area of approx. 1 cm² centered in the mount (position corresponded to center position of the diCAP electrode). The mount allowed to press the optical probe for microcirculation assessment as well as the diCAP electrode reproducibly onto the ROI at a defined weight pressure (optical probe: 7 mm Hg (10 g/cm²), diCAP electrode: 96 mm Hg (130 g/cm²)) during every single experimental session. The applied pressure of the diCAP electrode was 96 mm Hg, which corresponds to the typical contact pressure applied during a standard diCAP treatment with the PlasmaDerm®FLEX9060‐device. The exact same measuring position was reproducible in every subject. Heat insulation was applied to the mount at every point of contact with the skin to minimize thermal irritations of the skin. To minimize experimental uncertainties due to body movements, the subjects were placed in supine position on a padded mattress and were advised to refrain from moving any part of the body during each experimental session of approx. 100 minutes. In preliminary studies, we encountered the problem that most of the subjects fell asleep during the experiment which changed microcirculation parameters substantially. Several possibilities for prevention were investigated, for example, concentrate on breathing, listening to audio books, music, and to radio plays.

tes. In preliminary studies, we encountered the problem that most of the subjects fell asleep during the experiment which changed microcirculation parameters substantially. Several possibilities for prevention were investigated, for example, concentrate on breathing, listening to audio books, music, and to radio plays. As the most effective option to prevent falling asleep combined with achieving a level of attention as continuous as possible, we finally decided to show movies to the subjects. The plots of the movies were well known to most of the subjects thus avoiding strong emotional reactions. As a consequence of all these measures, the procedures for diCAP treatment and microcirculation measurement were very reproducible. Table 1 Consecutive actions and corresponding durations of the experimental protocol

As the most effective option to prevent falling asleep combined with achieving a level of attention as continuous as possible, we finally decided to show movies to the subjects. The plots of the movies were well known to most of the subjects thus avoiding strong emotional reactions. As a consequence of all these measures, the procedures for diCAP treatment and microcirculation measurement were very reproducible. Table 1 Consecutive actions and corresponding durations of the experimental protocol # Action Duration Data record 1 Measurement of skin temperature, pH, and moisture 5 min 2 Stabilization of perfusion 10 min 3 Measurement of microcirculation parameters in the ROI 10 min Baseline 4 Application of the diCAP electrode without plasma treatment 90, 180, or 270 s 5 Measurement of microcirculation parameters in the ROI 10 min Pressure 6 Plasma treatment 90, 180, or 270 s 7 Measurement of temperature in the ROI 30 s 8 Measurement of microcirculation parameters in the ROI 60 min Pressure + Plasma 9 Measurement of skin temperature, pH, and moisture 5 min John Wiley & Sons, LtdFigure 1 (A) Scheme of the customized mount for both reproducible plasma application and microcirculation measurements with an optical probe. (B) Photograph of the experimental setup with the optical probe (not visible) installed in the mount

Measurement of skin temperature, pH, and moisture 5 min John Wiley & Sons, LtdFigure 1 (A) Scheme of the customized mount for both reproducible plasma application and microcirculation measurements with an optical probe. (B) Photograph of the experimental setup with the optical probe (not visible) installed in the mount The experimental protocol followed a strict time lapse with the actions given in Table 1. Each of the ten subjects passed a total of three times the entire experimental protocol—each time, however, with a different treatment period of 90 seconds, 180 seconds, or 270 seconds, respectively. At first, skin temperature, pH, and moisture were measured in the ROI (#1). After resting in supine position with the forearm positioned in a controlled position within the customized mount for 10 minutes (#2), the baseline data of the microcirculation parameters in the ROI were recorded for a period of 10 minutes (#3). Directly following these measurements, the diCAP electrode was placed on the ROI for 90 seconds, 180 seconds, or 270 seconds, respectively, but no plasma treatment was conducted (#4). As soon as the diCAP electrode was removed from the ROI, the microcirculation parameters were recorded once more for a period of 10 minutes (#5). Then, the diCAP treatment was performed for 90 seconds, 180 seconds, or 270 seconds, respectively (#6), followed by measuring the skin temperature (#7) followed by microcirculation assessment over a period of 60 minutes (#8). Finally, skin temperature, pH, and moisture were measured in the ROI (#9).

of 10 minutes (#5). Then, the diCAP treatment was performed for 90 seconds, 180 seconds, or 270 seconds, respectively (#6), followed by measuring the skin temperature (#7) followed by microcirculation assessment over a period of 60 minutes (#8). Finally, skin temperature, pH, and moisture were measured in the ROI (#9). 2.4 Measurement of microcirculation parameters and data processing Microcirculation parameters were recorded by the noninvasive optical system Oxygen‐to‐see (O2C) (LEA Medizintechnik GmbH, Giessen, Germany). Conformity for the class IIa medical device with EU Guideline 93/42/EEC for medical devices is declared by the manufacturer. In brief, the O2C combines laser Doppler fluxmetry (820 nm, 30 mW) for the determination of blood flow and blood flow velocity in arbitrary units [A.U.] and spectrophotometry (450‐1000 nm, 20 W) for detection of SO2 in percentage [%] and local relative hemoglobin (rHb) in arbitrary units [A.U.]. In laser Doppler fluxmetry, the Doppler shift of erythrocytes in the tissue volume is detected and the product of moving erythrocytes N i and the corresponding velocity v i of each erythrocyte are used for the calculation of the relative blood flow via Equation (1): (1) Blood flow=∑vi·Ni

rHb) in arbitrary units [A.U.]. In laser Doppler fluxmetry, the Doppler shift of erythrocytes in the tissue volume is detected and the product of moving erythrocytes N i and the corresponding velocity v i of each erythrocyte are used for the calculation of the relative blood flow via Equation (1): (1) Blood flow=∑vi·Ni In spectrophotometry, white light is used for the determination of SO2 and rHb. The change in spectral distribution (color) of the reflected light is due to a wavelength‐dependent absorption of the applied white light and can be used for the calculation of SO2. In contrast, rHb correlates with the amount of light absorbed by the tissue (intensity). This measurement of rHb represents a hemoglobin concentration n e per tissue volume, which is independent of the erythrocyte velocity, vessel density, vessel lumen, and hemoglobin quantity in the blood. The interested reader is referred to Forst et al36 for more details on the measurement technique. For interpreting data from laser Doppler fluxmetry in the context of data from spectrophotometry, it is important to note that the procedure for determining N i only takes into account erythrocytes with velocities v i > 0, whereas the procedure for determining n e considers all erythrocytes (v i ≥ 0) and thus gives higher values.

eting data from laser Doppler fluxmetry in the context of data from spectrophotometry, it is important to note that the procedure for determining N i only takes into account erythrocytes with velocities v i > 0, whereas the procedure for determining n e considers all erythrocytes (v i ≥ 0) and thus gives higher values. For the measurements, a probe mounted on the tissue surface introduces light into the tissue and detects the backscattered signal. As stated by the manufacturer, the measuring depth of the applied probe LFX‐29 (LEA Medizintechnik GmbH) in this study is 1‐2 mm. The O2C system has been successfully applied before to measure diCAP‐induced changes in skin microcirculation.28, 29, 30 According to Table 1, the output of this study is three raw data records of time‐resolved (Δt = 25.6 ms) microcirculation parameters per subject and treatment time, which are indicated as Baseline, Pressure, and Pressure + Plasma. For time‐resolved microcirculation parameters, the raw data records with i data points were reduced to records with j data points at a temporal resolution of 30 seconds (i = 1172) according to Equation (2): (2) x¯j,time−resolved=11172∑i=1172j+11172(j+1)xi

hich are indicated as Baseline, Pressure, and Pressure + Plasma. For time‐resolved microcirculation parameters, the raw data records with i data points were reduced to records with j data points at a temporal resolution of 30 seconds (i = 1172) according to Equation (2): (2) x¯j,time−resolved=11172∑i=1172j+11172(j+1)xi For time‐averaged microcirculation parameters, raw data of the Baseline records and the Pressure records were averaged over the entire acquisition period of 10 minutes (i = 23428) per subject and treatment duration according to Equation 3a. The Pressure + Plasma records were averaged over a period of 60 minutes (i = 140625) per subject and treatment duration according to Equation 3b: (3a) x¯time−averaged,10min=123428∑i=123428xi (3b) x¯time−averaged,60min=1140625∑i=1140625xi 2.4.1 diCAP application For diCAP treatment, the class IIa medical device PlasmaDerm®FLEX9060 (CINOGY GmbH, Duderstadt, Germany) was used. Conformity with EU Guideline 93/42/EEC for medical devices is declared by the manufacturer. According to the manufacturer, the plasma area amounts to 27 cm².

For time‐averaged microcirculation parameters, raw data of the Baseline records and the Pressure records were averaged over the entire acquisition period of 10 minutes (i = 23428) per subject and treatment duration according to Equation 3a. The Pressure + Plasma records were averaged over a period of 60 minutes (i = 140625) per subject and treatment duration according to Equation 3b: (3a) x¯time−averaged,10min=123428∑i=123428xi (3b) x¯time−averaged,60min=1140625∑i=1140625xi 2.4.1 diCAP application For diCAP treatment, the class IIa medical device PlasmaDerm®FLEX9060 (CINOGY GmbH, Duderstadt, Germany) was used. Conformity with EU Guideline 93/42/EEC for medical devices is declared by the manufacturer. According to the manufacturer, the plasma area amounts to 27 cm². The device features a single high‐voltage electrode that generates sufficiently high electric fields for air breakdown only as soon as it is pressed gently on a tissue surface (or other conductive material). Electrically, the tissue acts as the counter electrode and is part of the (secondary) electrical circuit of the device. This concept to create a CAP is also referred to as DBD and belongs to the group of direct plasma sources.37 For details on typical plasma process parameters, the interested reader is referred to literature.38 In this work, the power density was determined to be 4 mW/cm² by an approved measurement technique when the single high‐voltage electrode is operated against a metal counter electrode.39

roup of direct plasma sources.37 For details on typical plasma process parameters, the interested reader is referred to literature.38 In this work, the power density was determined to be 4 mW/cm² by an approved measurement technique when the single high‐voltage electrode is operated against a metal counter electrode.39 2.5 Measurement of skin temperature, pH, and moisture Skin pH was tested with the potentiometrically operating electrode Inlab Surface (Mettler Toledo GmbH, Giessen, Germany) measuring at an accuracy of ± 0.01. Skin moisture in percentage was measured with a MoistureMeterD compact (Delfin Technologies UK Ltd., Dorking, UK) with a measurement accuracy of ± 5%. The measuring principle is based on noninvasive detection of the tissue dielectric constant. Both instruments were in contact with the skin and especially the measurement of skin moisture needed contact pressure. Therefore, these measurements could have an influence on microcirculation parameters. Consequently, we measured these two parameters only before and after the experiment and thus chose a contact‐free infrared thermometer FT90 (Beurer medical GmbH, Ulm, Germany) at an accuracy of ± 2°C. In accordance with literature, there were no decisive temperature‐driven changes in blood flow under 35°C expected.40, 41 For this reason, we expected the accuracy was sufficient. The contact‐free measurement of skin temperature enabled this measurement during the experiment without an influence on microcirculation parameters. Data are provided as mean ± SD of all subjects (N = 10) per treatment duration and time of acquisition.

ed.40, 41 For this reason, we expected the accuracy was sufficient. The contact‐free measurement of skin temperature enabled this measurement during the experiment without an influence on microcirculation parameters. Data are provided as mean ± SD of all subjects (N = 10) per treatment duration and time of acquisition. 2.6 Statistical analysis A statistical power analysis was carried out using the statistical software G*Power 3.1. Based on our previous experience, we estimated an effect size of 1.3; the probability of a type 1 error was set to 5% (alpha = 0.05) a power to 95% (beta = 0.95) this resulted in a necessary sample size of 10 with an actual power of beta = 0.954. The statistical analysis of experimental data was performed using the data management solution Origin® 2015G (OriginLab Corp., Northampton, MA, USA) and SPSS (IBM, Armonk, NY, USA). Data are provided as mean ± SD of all subjects (N = 10) per treatment duration. For pH values, statistical significance was evaluated by a paired two‐sample t test, whereas microcirculation parameters were analyzed applying univariate ANOVA testing with repeated measures followed by Bonferroni post hoc analysis. A P‐value < .05 was regarded as statistically significant. 3 RESULTS 3.1 Adverse events Throughout the experiments, three subjects reported a tingling sensation during plasma treatment which was interpreted as AE. However, these sensations did neither lead to premature termination of the plasma treatment nor the trial. No SAE were reported.

2.6 Statistical analysis A statistical power analysis was carried out using the statistical software G*Power 3.1. Based on our previous experience, we estimated an effect size of 1.3; the probability of a type 1 error was set to 5% (alpha = 0.05) a power to 95% (beta = 0.95) this resulted in a necessary sample size of 10 with an actual power of beta = 0.954. The statistical analysis of experimental data was performed using the data management solution Origin® 2015G (OriginLab Corp., Northampton, MA, USA) and SPSS (IBM, Armonk, NY, USA). Data are provided as mean ± SD of all subjects (N = 10) per treatment duration. For pH values, statistical significance was evaluated by a paired two‐sample t test, whereas microcirculation parameters were analyzed applying univariate ANOVA testing with repeated measures followed by Bonferroni post hoc analysis. A P‐value < .05 was regarded as statistically significant. 3 RESULTS 3.1 Adverse events Throughout the experiments, three subjects reported a tingling sensation during plasma treatment which was interpreted as AE. However, these sensations did neither lead to premature termination of the plasma treatment nor the trial. No SAE were reported. 3.2 Skin temperature, moisture, and pH The surface temperatures in the ROI for treatment periods of 90 seconds, 180 seconds, and 270 seconds, respectively, are depicted in Figure 2. No significant variations in skin temperature could be observed due to plasma treatment within the range of experimental uncertainty (< 2°C).

ed against a metal counter electrode), whereas in our study, the device with an electrode area of 27 cm² was operated at 4 mW/cm².39 Concluding the above findings, the applied plasma power density seems to correlate with quantitative differences in microcirculation enhancement and thus might be a key process parameter. When interpreting our results of blood flow data and blood flow velocity data (see Equation (1)), we conclude that the number of moving erythrocytes N i is disproportionally increased compared to their velocity v i thus indicating the effect of a plasma‐induced vasodilatation. To test this hypothesis, further investigations on the physiological mechanisms of diCAP on microcirculation should be conducted.

3.2 Skin temperature, moisture, and pH The surface temperatures in the ROI for treatment periods of 90 seconds, 180 seconds, and 270 seconds, respectively, are depicted in Figure 2. No significant variations in skin temperature could be observed due to plasma treatment within the range of experimental uncertainty (< 2°C). Figure 2 Skin temperature of the dorsal forearms at the beginning of the experimental protocol, immediately after plasma treatment and after completion of the experimental protocol Skin moisture in the ROI was relatively constant throughout the experiments and ranged from 35.0 to 37.5 before and after the experiments without significant changes (Figure 3). Figure 3 Skin moisture of the dorsal forearms before and after completion of the experimental protocol Figure 4 illustrates the skin pH values in the ROI. For all plasma‐treatment periods, the pH decreased by up to 0.3. For 180 seconds and 270 seconds plasma treatment, these changes were statistically significant. Figure 4 Skin pH value of the dorsal forearm before and after completion of the experimental protocol (*P < .05)

Figure 4 illustrates the skin pH values in the ROI. For all plasma‐treatment periods, the pH decreased by up to 0.3. For 180 seconds and 270 seconds plasma treatment, these changes were statistically significant. Figure 4 Skin pH value of the dorsal forearm before and after completion of the experimental protocol (*P < .05) 3.3 Microcirculation In Figure 5, the time‐resolved blood flow characteristics are given as mean values of the subjects and visually separated by colors into the three microcirculation acquisition phases—Baseline (green), Pressure (orange), Pressure + Plasma (red)—of the experimental protocol. It becomes apparent that the blood flow responses to pressure stimulation are relatively mild compared to the responses to the Pressure + Plasma stimulation. There was a short decrease in blood flow immediately after induction of pressure. However, this change was reversed in the first 10 minutes after application through adaptation, resulting in a constant plateau that was not significantly different from the baseline. Blood flow changes induced by diCAP were much more pronounced and highly individual within the ten subjects. Some subjects show a constant increase over time, whereas the reaction of others followed an asymptotic behavior. Lastly, in some individuals, the blood flow increase is very strong but then tends to decrease within the acquisition period.

y diCAP were much more pronounced and highly individual within the ten subjects. Some subjects show a constant increase over time, whereas the reaction of others followed an asymptotic behavior. Lastly, in some individuals, the blood flow increase is very strong but then tends to decrease within the acquisition period. Figure 5 Time‐resolved characteristics of the cutaneous blood flow (N = 10) after stimulation with Pressure and with Pressure + Plasma for different treatment durations (indicated by gray bars). For clarity, SD data are only provided in Figure 9 However, the mean values for all treatment durations including plasma application indicate an increase in blood flow—the longer the plasma‐treatment time, the more pronounced was the blood flow increase. It is worth noting that the predominant increase in blood flow did not occur immediately during the treatment, but minutes after the plasma was applied. 3.3.1 Local relative hemoglobin (rHb) Within the time‐averaged means over 10 minutes and 1 hour, respectively, in Figure 6, only the Pressure + Plasma treatment for 270 seconds increased the rHb statistically significant by 5.1% from 81.5 ± 8.4 to 85.6 ± 8.4 for a period of 1 hour after plasma treatment whereas Pressure stimulus alone did not compared to no intervention. Significance levels between Pressure + Plasma and Pressure alone were checked to gain deeper insights into mechanisms of rHb increase. Yet, the Pressure + Plasma data were not significantly different from the mean values on sole Pressure stimulus.

ent whereas Pressure stimulus alone did not compared to no intervention. Significance levels between Pressure + Plasma and Pressure alone were checked to gain deeper insights into mechanisms of rHb increase. Yet, the Pressure + Plasma data were not significantly different from the mean values on sole Pressure stimulus. Figure 6 Local relative hemoglobin in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) (*P < .05) 3.3.2 Blood flow velocity The time‐averaged mean of blood flow velocity in Figure 7 showed no change after 90 seconds and 180 seconds, respectively, of Pressure or Pressure + Plasma. After 270 seconds treatment duration, an increase by trend was demonstrated only for Pressure + Plasma compared to no intervention (P = .052). Figure 7 Blood flow velocity in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h)

tion (1)), we conclude that the number of moving erythrocytes N i is disproportionally increased compared to their velocity v i thus indicating the effect of a plasma‐induced vasodilatation. To test this hypothesis, further investigations on the physiological mechanisms of diCAP on microcirculation should be conducted. Due to different morbidities associated with reduced perfusion (ie, Diabetes, arteriosclerosis, polyneuropathy, overweight), a growing number of patients suffer from impaired wound healing.47 In this context, we spare an elaborated discussion on the pharmaceutical therapy, as it is usually applied preventively or in early stage of wounds.48, 49 Surgical techniques predominantly address macrovascularization to restore blood flow. Yet, blood flow on a macrovascular level alone will not reverse microcirculatory derangements, which are critical in wound healing processes.48, 50 In clinical practice, perfusion data are commonly given as tcPO2 derived from arterial oxygen concentration.51 In contrast, the noninvasive and sensitive O2C system as diagnostic device for microcirculation gives information about venous postcapillary oxygen saturation. Consequently, tcPO2 and SO2 provide different information and can thus not be compared quantitatively. In fact, SO2 data can provide substantial additional information and thus assist physicians in assessing clinical cases.

7 Blood flow velocity in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) 3.3.3 Postcapillary oxygen saturation The time‐averaged means for 10 minutes and 1 hour, respectively, for postcapillary oxygen saturation are given in Figure 8. Apparently, the sole Pressure stimulus for up to 270 seconds did not provoke alterations in the skin microcirculation compared to Baseline. In contrast, Pressure + Plasma intervention for 180 s significantly increased postcapillary oxygen saturation by 7.1% from 57.6 ± 8.6% to 64.7 ± 11.3%. With 9.4% from 59.8 ± 12.2% to 69.2 ± 10.4% after 270 seconds, this effect was even more pronounced. Once more, in order to elucidate the respective contributions of pressure or plasma stimulus to these increases, we analyzed significances between Pressure + Plasma and Pressure alone. Thereby, significant differences were found for all, 90 seconds (+6.5%), 180 seconds (+7.6%), and 270 seconds (+9.1%), respectively. Figure 8 Tissue oxygen saturation in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) (*P < .05, **P < .01)

in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) (*P < .05, **P < .01) 3.3.4 Blood flow Figure 9 depicts the response of cutaneous blood flow to stimuli for up to 270 seconds. Again, pressure‐induced effects did not lead to significant changes compared to Baseline. Yet, a significant increase by 105.5% from 32.7 ± 12.6 to 67.2 ± 20.6 over a 1‐h period occurred after diCAP treatment for 270 seconds. To analyze the particular contributions of pressure and plasma stimulus, comparison of Pressure + Plasma to sole Pressure data revealed an increase for 180 seconds treatment (+70.4%) as well as for 270 seconds treatment (+94.5%). Figure 9 Blood flow in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) (**P < .01)

lood flow in the ROI (1‐2 mm depth) at the beginning of the experimental protocol (Baseline; time average of 10 min), after pressure induction by the diCAP electrode (Pressure; time average of 10 min), and after operating the diCAP electrode at constant power density (Pressure + Plasma; time average of 1 h) (**P < .01) 4 DISCUSSION On a general note, it is worth mentioning that in accordance with the relevant literature no (S)AE was reported due to diCAP application on human tissue during our study. Positive effects on cutaneous microcirculation have already been observed.28, 29, 30 Yet, until now, it has been unclear, if the device‐induced pressure to the tissue should be considered a confounding factor in microcirculation enhancement by diCAP application. Therefore, in this study, the role of mechanical pressure induced by the diCAP device was addressed. Furthermore, it was so far unclear, how the tissue would react to longer treatment times with diCAP. Consequently, in this study, we applied diCAP for 90 seconds, 180 seconds, and 270 seconds, in a controlled mechanical environment.

this study, the role of mechanical pressure induced by the diCAP device was addressed. Furthermore, it was so far unclear, how the tissue would react to longer treatment times with diCAP. Consequently, in this study, we applied diCAP for 90 seconds, 180 seconds, and 270 seconds, in a controlled mechanical environment. It is well known, that pressure on the skin can influence skin perfusion. Whereas preclinical DBD sources can be operated without physical contact to the skin, most plasma medical products inherently exert a mechanical pressure on the treated tissue. Indeed, in the time‐resolved blood flow characteristics (Figure 5), an initial yet reversible occlusion after the pressure stimulus (96 mm Hg) becomes apparent. Autoregulative mechanisms compensate for this effect within less than 10 minutes. Within our experimental protocol, the occlusion effect is inevitably taken into account when calculating time‐averaged mean values after Pressure stimulus and leads to slightly decreased mean values for postcapillary oxygen saturation (Figure 8) and blood flow (Figure 9) compared to Baseline level. This effect could explain our counter‐intuitive observations that for 90 seconds in Figure 8 and for 180 seconds in Figure 9, we found significant differences comparing Pressure + Plasma data to sole Pressure data, whereas Pressure + Plasma data did not yield statistically significant differences when compared to Baseline results. In summary, the level of microcirculatory upregulation released by diCAP (Pressure + Plasma) was overall stronger compared to autoregulative recovery following pressure‐induced occlusion and present for a long time scale of at least 1 hour. Consequently, our first finding is that plasma‐induced effects during application of the diCAP source are dominant over mechanical pressure impact by the electrode for enhancing microcirculation.

mpared to autoregulative recovery following pressure‐induced occlusion and present for a long time scale of at least 1 hour. Consequently, our first finding is that plasma‐induced effects during application of the diCAP source are dominant over mechanical pressure impact by the electrode for enhancing microcirculation. According to literature, skin temperatures of the human forearm above 35°C show a decisive effect on blood flow.40, 41 Plasma heat transfer‐induced skin temperatures were well below this threshold, even taking the relatively low device accuracy for temperature measurements into account. Thus, our results on physiological skin parameters indicate that diCAP treatment for up to 270 seconds did not relevantly affect skin temperature. The same is true for skin moisture. Thus, we can rule out an impact of these variables on our results, which is important given that it has long been known that local temperature influences the peripheral blood flow.42 As our second finding, we state that in our experiments, temperature‐driven impact on microcirculation does not play a role.

in moisture. Thus, we can rule out an impact of these variables on our results, which is important given that it has long been known that local temperature influences the peripheral blood flow.42 As our second finding, we state that in our experiments, temperature‐driven impact on microcirculation does not play a role. The third finding is given by the observation that the skin pH significantly decreases with prolonged treatment durations. This general trend is well known for CAP generated in air and it is attributed to reactions of plasma‐generated gaseous species (reactive oxygen and nitrogen species—RONS) with omnipresent H2O, which lead to the formation of nitrous acid, nitric acid, and hydrogen peroxide.43, 44 In this context, it was hypothesized that NO penetration into the skin may be a possible mechanism for plasma‐induced changes of microcirculation.27, 45, 46 In general, for diCAP treatment durations of 90 seconds, 180 seconds, and 270 seconds, respectively, we observed a more pronounced impact on microcirculatory parameters with increasing treatment time indicated by an absolute increase in significance levels.

The third finding is given by the observation that the skin pH significantly decreases with prolonged treatment durations. This general trend is well known for CAP generated in air and it is attributed to reactions of plasma‐generated gaseous species (reactive oxygen and nitrogen species—RONS) with omnipresent H2O, which lead to the formation of nitrous acid, nitric acid, and hydrogen peroxide.43, 44 In this context, it was hypothesized that NO penetration into the skin may be a possible mechanism for plasma‐induced changes of microcirculation.27, 45, 46 In general, for diCAP treatment durations of 90 seconds, 180 seconds, and 270 seconds, respectively, we observed a more pronounced impact on microcirculatory parameters with increasing treatment time indicated by an absolute increase in significance levels. In general, quantitative results on microcirculation data induced by diCAP are to be interpreted in the context of the individual study design. In agreement with literature data, we confirm that diCAP enhances microcirculation for a much longer period than the application time. Heuer et al presented their results with a temporal resolution of 1‐20 minutes, whereas Kisch et al measured at 1 min resolution.28, 29, 30 Even though in our study, data were recorded at a temporal resolution of 25.6 ms, our results are given as mean values over a period of up to 60 minutes. The rationale of this approach is a rather practical focus on clinical impact and applicability in patient care.

s Kisch et al measured at 1 min resolution.28, 29, 30 Even though in our study, data were recorded at a temporal resolution of 25.6 ms, our results are given as mean values over a period of up to 60 minutes. The rationale of this approach is a rather practical focus on clinical impact and applicability in patient care. Our results on local relative hemoglobin show an increase about 5.1% after 270 seconds of treatment for a period of 60 minutes. Heuer et al found an enhancement of 30% in rHb over a period of 60 minutes following a 90 seconds treatment, whereas Kisch et al observed no increase after a single application for 90 seconds. After 3 × 90 seconds treatment, they found the rHb significantly increased for 40 minutes by max. 12%.28, 29, 30 Concerning blood flow velocity, our study revealed no significant alterations due to diCAP treatment. Heuer et al observed a doubling (90 seconds treatment) after 5 minutes, which steadily decreases to a factor of 1.6 after 45 minutes.28, 29, 30 The postcapillary oxygen saturation was upregulated by 7.1% (180 seconds diCAP) and 9.4% (270 seconds diCAP) for 60 minutes. Heuer et al found an upregulation by 37% for 60 minutes after 90 seconds of treatment. Kisch et al found up to 24% increase for 8 minutes after single and up to 47% for 40 minutes after repetitive application.28, 29, 30 Statistical analysis clearly indicates that upregulation of oxygen saturation is driven by the physiological potential of the plasma and not affected by the implied electrode pressure on the skin.

al found up to 24% increase for 8 minutes after single and up to 47% for 40 minutes after repetitive application.28, 29, 30 Statistical analysis clearly indicates that upregulation of oxygen saturation is driven by the physiological potential of the plasma and not affected by the implied electrode pressure on the skin. Blood flow doubled following 270 seconds diCAP in our experiments for 1 hour. Once more, this increase is not caused by mechanical pressure impact. Heuer et al measured a roughly fourfold higher blood flow after 5 min, which decreased to threefold after 45 minues. Kisch et al found an increase up to 73% for 11 minutes after single treatment and an increase about a factor of up to 2.5 for a period of 12 minutes after repetitive treatment for 3 × 90 seconds.28, 29, 30

uer et al measured a roughly fourfold higher blood flow after 5 min, which decreased to threefold after 45 minues. Kisch et al found an increase up to 73% for 11 minutes after single treatment and an increase about a factor of up to 2.5 for a period of 12 minutes after repetitive treatment for 3 × 90 seconds.28, 29, 30 As the diagnostic device in our study is identical to the device applied in the literature, differences in quantitative results either derive from slight modifications (eg, mechanical pressure exerted by the diCAP device in comparison with the studies of Kisch et al, optical probe attachment on the skin) in the individual experimental protocols or from nonidentical diCAP process parameters. Indeed, the diCAP device applied in Heuer et al not only features a different material (ceramic instead of plastic) but also a significantly smaller geometry with at a diameter of no more than 10 mm. Furthermore, as can be judged from the provided electrical parameters, this electrode was operated by a different electronic device than in the studies of Kisch et al and our study. Unfortunately, in Heuer et al, the plasma power density, which can be considered one of the most important process parameters, is not given. Considering the relatively small plasma area in Heuer et al compared to the studies of Kisch et al and our experiments, the pulse energy as well as plasma power density can be assumed the highest in Heuer et al Therefore, the relatively high‐energy input into the skin as well as the plasma (typically leading to high RONS concentrations in the plasma volume) obviously leads to a very pronounced impact on dermal microcirculation. Kisch et al operated their diCAP device with an area of 22.5 cm² at 12 mW/cm² (when operated against a metal counter electrode), whereas in our study, the device with an electrode area of 27 cm² was operated at 4 mW/cm².39 Concluding the above findings, the applied plasma power density seems to correlate with quantitative differences in microcirculation enhancement and thus might be a key process parameter.

stic device for microcirculation gives information about venous postcapillary oxygen saturation. Consequently, tcPO2 and SO2 provide different information and can thus not be compared quantitatively. In fact, SO2 data can provide substantial additional information and thus assist physicians in assessing clinical cases. An important issue arises from the question, to what extent the effect of microcirculation enhancement by diCAP in healthy subjects is transferable to patients with impaired tissue. As a matter of fact, this question needs to be addressed in studies facilitating a cohort of patients with defined morbidities. As of now, we can only hypothesize on the basis of available literature. Beckert et al52 were the first group evaluating the O2C device. They grouped subjects with healthy skin and those with diabetic ulcers. In the latter, they defined a subgroup of healers and nonhealers. Their results showed that within the nonhealers group blood flow, rHb, and SO2 were significantly lower compared to the healer subgroup, whereas tcPO2 values (27 ± 6.0 mm Hg vs 19 ± 7.3 mm Hg) did not reveal significant differences between healers and nonhealers. The authors concluded that tcPO2 is not an optimal parameter for evaluation of oxygen supply of the wound. Furthermore, they found that venous postcapillary oxygen saturation was significantly lower in the nonhealer group compared with healthy control subjects, whereas there was no difference between control subjects (healthy skin) and healers. At the wound site, nonhealers had significantly lower mean values in postcapillary oxygen saturation (50% vs 73%), local relative hemoglobin (54 A.U. vs 77 A.U.), and especially blood flow (19 A.U. vs 150 A.U.) compared to subjects with healing wounds. Surprisingly, comparing the quantitative results of microcirculation parameters derived from intact skin of healthy volunteers with data from the nonhealer subgroup in the wound group, it was found that they match quite well. This indicates no enhanced microcirculation at the wound site of nonhealers compared to healers.52 From these observations, we hypothesize that the diCAP‐induced enhancement of microcirculation observed on intact skin of healthy volunteers in this study might be transferable to future diCAP application at the wound site of nonhealers.

o enhanced microcirculation at the wound site of nonhealers compared to healers.52 From these observations, we hypothesize that the diCAP‐induced enhancement of microcirculation observed on intact skin of healthy volunteers in this study might be transferable to future diCAP application at the wound site of nonhealers. This appears especially desirable in view of the strong blood flow increase induced by diCAP as blood flow was found to be extremely low in the wound site of nonhealers.52 Yet, upcoming studies must be performed to test this hypothesis. With regard to clinical relevance, blood flow can be considered the driving parameter for microvascular oxygen supply, as postcapillary oxygen saturation and local relative hemoglobin are primarily a result of alterations in blood flow. Applying diCAP in a topical, noninvasive treatment doubles the blood flow. Saucy et al53 increased skin blood flow of fore‐ and hindfood after bypass and endarterectomy by 50% and 40%, respectively, measured in perfusion units as derived by a laser Doppler imaging system. Taking into account that diCAP is a noninvasive intervention and so far no serious side effects have been reported, it can be considered an innovative and competitive therapy option—particularly for poor candidates for surgical or endovascular procedures, whose comorbidities and poor outflow vessels limit revascularization as a viable option.48

iCAP is a noninvasive intervention and so far no serious side effects have been reported, it can be considered an innovative and competitive therapy option—particularly for poor candidates for surgical or endovascular procedures, whose comorbidities and poor outflow vessels limit revascularization as a viable option.48 We did not observe a declining tendency of elevated blood flow after 1 hour of diCAP treatment referring to the time‐resolved blood flow data. Yet, it is a nonpermanent effect in healthy subjects. In view of promising results on the induction of neo‐angiogenetic and epithelial growths effects by CAP treatment, we hypothesize that for the repetitive treatment of wounds the increase in blood flow might become a sustainable positive effect on vascular autoregulation in impaired tissue.24, 54, 55 PERSPECTIVE The results of the present investigation provide a new therapeutic approach to sustainably enhance microcirculation in cutaneous tissue at no side effects. Consequently, diCAP emerges as a new clinically promising option to restore impaired tissue and to assist wound healing. Furthermore, diCAP might assist physicians in the prevention of impaired wound healing in potential risk patients.

oach to sustainably enhance microcirculation in cutaneous tissue at no side effects. Consequently, diCAP emerges as a new clinically promising option to restore impaired tissue and to assist wound healing. Furthermore, diCAP might assist physicians in the prevention of impaired wound healing in potential risk patients. ACKNOWLEDGMENTS This work was carried out within the research project “Plasma for Life” (funding reference no. 13FH6I04IA) with financial support from the German Federal Ministry of Education and Research (BMBF) as well as within the research project “KonChaWu” (ZW 3‐85006987) with financial support from European Regional Development Fund (ERDF) and the State of Lower Saxony. The authors would also like to express their gratitude to all subjects included in the study for their patience and cooperation. Furthermore, the technical and practical support by CINOGY GmbH and LEA Medizintechnik GmbH is gratefully acknowledged.

Abbreviations ACEIangiotensin‐converting enzyme inhibitors AMDage‐related macular degeneration ARBangiotensin receptor blockers BCVAbest‐corrected visual acuity BMIbody mass index BPblood pressure CKDchronic kidney disease DMdiabetes mellitus DVPdeep vascular plexus eGFRestimated glomerular filtration rate FAZfoveal avascular zone FD‐300foveal vessel density in 300‐μm‐wide region around FAZ LogMARlogarithm of the minimum angle of resolution OCTAoptical coherence tomography angiography RASrenin‐angiotensin system SDstandard deviation SVPsuperficial vascular plexus

DMdiabetes mellitus DVPdeep vascular plexus eGFRestimated glomerular filtration rate FAZfoveal avascular zone FD‐300foveal vessel density in 300‐μm‐wide region around FAZ LogMARlogarithm of the minimum angle of resolution OCTAoptical coherence tomography angiography RASrenin‐angiotensin system SDstandard deviation SVPsuperficial vascular plexus 1 INTRODUCTION Chronic kidney disease is a common comorbidity in ophthalmologic patients, especially among old aged, hypertensive, and diabetic patients. The prevalence is estimated to be 8%‐16%1 and increases with age. It may be as high as 30.8% at age 70 or more.2 It can also be found in 15% of non‐diabetic hypertensive patients3 and in 43%‐53% of patients with diabetes.4 The prevalence may double by 2035 with the anticipated increase in diabetic and older population.5 Chronic kidney disease has been associated with accelerated atherosclerosis, cognitive impairment, cerebrovascular disease, cardiovascular disease, and mortality.6, 7, 8, 9 In the eye, patients with CKD have higher risks of cataract, glaucoma, AMD, retinopathies, and visual impairment.10, 11, 12 The mechanism behind increased ocular diseases in patient with CKD is still being debated. It may be due to CKD and ocular diseases sharing many common systemic risk factors such as aging, DM, hypertension, smoking, and obesity.11 Alternatively, it could also be due to mechanisms related to CKD, such as increased oxidative stress by decreased filtration of free radical‐generating nitrogenous waste products or increased inflammation by activation of the RAS.11

y common systemic risk factors such as aging, DM, hypertension, smoking, and obesity.11 Alternatively, it could also be due to mechanisms related to CKD, such as increased oxidative stress by decreased filtration of free radical‐generating nitrogenous waste products or increased inflammation by activation of the RAS.11 Earlier studies have revealed that decreased retinal vessel caliber, smaller fractal dimensions, focal arteriolar narrowing, arteriovenous nicking, and opacification of the arteriolar wall can be found in patients with CKD.13, 14, 15 These retinal microvascular changes may be useful biomarkers for predicting cardiovascular diseases,16 cognitive impairment,17 and aggravation of renal function in patients with CKD.18, 19 However, there is limited information about microvascular alterations at the capillary level.20 Although increased intercapillary distance in CKD has been shown through use of scanning laser Doppler flowmetry,20 the changes in different layers of the retinal vascular plexus are unknown. It is also unclear what role other systemic comorbidities play in these microvascular changes.

rations at the capillary level.20 Although increased intercapillary distance in CKD has been shown through use of scanning laser Doppler flowmetry,20 the changes in different layers of the retinal vascular plexus are unknown. It is also unclear what role other systemic comorbidities play in these microvascular changes. Optical coherence tomography angiography is a newly developed non‐invasive diagnostic tool that provides a depth‐resolved three‐dimensional image to visualize the different layers of the retinal vascular plexuses. The purpose of this study was to evaluate early retinal microvascular changes in the superficial and DVPs in patients with CKD through use of OCTA. We also evaluated systemic factors associated with these changes. Elucidating these retinal microvascular changes may (a) shed some light on the pathogenesis of ocular diseases in CKD; (b) help interpret retinal OCTA images in patients with CKD; and (c) provide information for future pharmacological intervention to improve visual outcomes.

temic factors associated with these changes. Elucidating these retinal microvascular changes may (a) shed some light on the pathogenesis of ocular diseases in CKD; (b) help interpret retinal OCTA images in patients with CKD; and (c) provide information for future pharmacological intervention to improve visual outcomes. 2 MATERIALS AND METHODS This single‐center, cross‐sectional study was conducted between August 2017 and July 2018 by the Department of Nephrology and the Department of Ophthalmology at Keelung Chang Gung Memorial Hospital, Keelung, Taiwan. The study was approved by the Chang Gung Memorial Hospital Institutional Review Board, and it followed the tenets of the Declaration of Helsinki. The inclusion criteria for the CKD group were (a) age ≧21 years; (b) CKD stages 3‐5 (including end‐stage renal disease); and (c) no visual symptoms. The inclusion criteria for the control group were (a) age ≧21 years; (b) no major systemic disease; (c) no visual symptoms; and (d) no retinal and macular diseases. The exclusion criteria were (a) the presence of significant ocular media opacity (such as dense cataract); (b) inability to obtain adequate quality OCTA image (scan quality score <6/10 or presence of significant artifact); or (c) pregnancy.

temic disease; (c) no visual symptoms; and (d) no retinal and macular diseases. The exclusion criteria were (a) the presence of significant ocular media opacity (such as dense cataract); (b) inability to obtain adequate quality OCTA image (scan quality score <6/10 or presence of significant artifact); or (c) pregnancy. Chronic kidney disease was defined, using the criteria recommended by Kidney Disease: Improving Global Outcomes (KDIGO) 2012 Clinical Practice Guidelines, as (a) abnormalities of kidney structure or function, present for >3 months, with implications for health; and (b) decreased glomerular filtration rate to <60 mL/min/1.73 m2 for >3 months.21 Estimated glomerular filtration rate was calculated from serum creatinine concentration using the CKD Epidemiology Collaboration equation.22 Severity of CKD was defined by eGFR categories: 30‐59 mL/min/1.73 m2 (stage 3), 15‐29 mL/min/1.73 m2 (stage 4), and <15 mL/min/1.73 m2 (stage 5).23 Patients with CKD meeting the above criteria were enrolled from the Department of Nephrology. Age‐matched (same age‐group) healthy subjects without retinal disease were enrolled into the control group in 4:1 ratio. An informed consent was obtained from each subject.

.73 m2 (stage 4), and <15 mL/min/1.73 m2 (stage 5).23 Patients with CKD meeting the above criteria were enrolled from the Department of Nephrology. Age‐matched (same age‐group) healthy subjects without retinal disease were enrolled into the control group in 4:1 ratio. An informed consent was obtained from each subject. Medical histories and laboratory data for the most recent 3 months were gathered. Medical history was collected through a standardized questionnaire and electronic medical records. Taiwan's National Health Insurance provides a nationwide electronic platform that allows for the sharing of patients’ medical information, prescription records, laboratory data, and other examination reports with the patients’ informed consent. Subjects suspected to have major systemic diseases were excluded from the control group.

onal Health Insurance provides a nationwide electronic platform that allows for the sharing of patients’ medical information, prescription records, laboratory data, and other examination reports with the patients’ informed consent. Subjects suspected to have major systemic diseases were excluded from the control group. The BMI was calculated from measured weight and height. Complete ocular examinations, including BCVA, intraocular pressure, slit‐lamp biomicroscopy examination, indirect fundus ophthalmoscopy, color fundus photographs, axial length, optical coherence tomography, and OCTA, were performed. Best‐corrected visual acuity was measured on a Snellen chart and converted to the logMAR for calculation. The presence of any retinopathy was documented. Early AMD (AREDS category 2) is characterized by multiple small drusen (<63 μm in diameter), few intermediate drusen (63‐124 μm in diameter), or mild retinal pigment epithelial abnormalities.24 Diabetic retinopathy was classified via the International Clinical Diabetic Retinopathy and Diabetic Macular Edema Disease Severity Scales.25 In patients with both eyes eligible, the eye with better OCTA quality was used for statistical analysis.

‐124 μm in diameter), or mild retinal pigment epithelial abnormalities.24 Diabetic retinopathy was classified via the International Clinical Diabetic Retinopathy and Diabetic Macular Edema Disease Severity Scales.25 In patients with both eyes eligible, the eye with better OCTA quality was used for statistical analysis. 2.1 Optical coherence tomography angiography parameters AngioVue (Optovue RTVue XR Avanti; Optovue Inc.) was used for acquiring OCTA images for this study. The machine uses an 840‐nm diode laser source and has an A‐scan rate of 70 kHz. A 3 × 3‐mm scan, centered on the fovea, was performed in all eyes. An orthogonal registration algorithm was used to produce a 3‐dimensional OCTA image. Then using the machine's AngioVue software (version: A2017,1,0,151), the vascular area was automatically segmented into four layers, that is superficial, deep, outer retina, and choroidal. The default segmentation for the SVP includes vasculature between the internal limiting membrane and 10 μm above the inner plexiform layer. For the DVP, this includes the vasculature between 10 μm above the inner plexiform layer and 10 μm below the outer plexiform layer.

superficial, deep, outer retina, and choroidal. The default segmentation for the SVP includes vasculature between the internal limiting membrane and 10 μm above the inner plexiform layer. For the DVP, this includes the vasculature between 10 μm above the inner plexiform layer and 10 μm below the outer plexiform layer. The vessel density is defined as the percentage area occupied by all vessels (including terminal arterioles, venules, and capillaries) in a particular region. The data are provided in an ETDRS grid vessel density map (Figure 1). The foveal region is a 1‐mm‐diameter circle, and the parafoveal region is a 1‐mm‐wide circular annulus. The parafoveal region was further divided into the temporal, superior, nasal, and inferior quadrants. The AngioVue software automatically calculates the vessel density of the SVP and the DVP, respectively. We also evaluated other foveal parameters provided by the machine software including the FAZ size; FAZ perimeter; FAZ a‐circularity index; and FD‐300. The foveal parameters were determined from an OCTA image of the inner retina microvasculature, which contained both SVP and DVP (Figure 1).

VP and the DVP, respectively. We also evaluated other foveal parameters provided by the machine software including the FAZ size; FAZ perimeter; FAZ a‐circularity index; and FD‐300. The foveal parameters were determined from an OCTA image of the inner retina microvasculature, which contained both SVP and DVP (Figure 1). Figure 1 A 49‐y‐old healthy woman in the control group. (A) Normal color fundus photograph. (B) An OCTA image of the inner retina, which contains of both the SVP and DVP. Inner yellow line demarcates the boundary of FAZ. The inner and outer yellow lines demarcate the 300‐μm‐wide region around the FAZ. (C) An OCTA image of the SVP. The blue colored grid is an ETDRS grid that contains the foveal region in a 1‐mm‐diameter circle and the parafoveal region within a 1‐mm‐wide circular annulus. (D) An OCTA image of DVP. (E) An vessel density map of SVP. (F) An vessel density map of DVP. (G) A B‐scan image shows the segmentation site at the inner retina (between the two red lines), located at the green line in (B). (H) A B‐scan image shows the segmentation site at SVP (between the red and green lines), located at the green line in (C). (I) A B‐scan image shows the segmentation site at DVP (between the green and red lines), located at the green line in (D)

inner retina (between the two red lines), located at the green line in (B). (H) A B‐scan image shows the segmentation site at SVP (between the red and green lines), located at the green line in (C). (I) A B‐scan image shows the segmentation site at DVP (between the green and red lines), located at the green line in (D) 2.2 Statistical analysis To compare the demographic data and clinical characteristics of the CKD group with the control group, Pearson's chi‐square test was used for categorical variables and the independent sample t test was used for continuous variables. The independent sample t test was also used to analyze differences in vessel densities and foveal parameters between the two groups. Multiple linear regression models with backward stepwise method were used to determine the potential systemic factors associated with the vessel densities in SVP and DVP of all subjects (Model 1) and of patients with CKD (Model 2). Age, sex, smoking status, BMI, DM, use of anti‐hypertensive drugs, systolic BP, diastolic BP, CKD, CKD stage, and eGFR were independent variables entered into the regression models whenever applicable. A two‐tailed P value <0.05 was considered as statistically significant. Data were analyzed using SPSS Program Package version 17.0 (SPSS Inc.).

BMI, DM, use of anti‐hypertensive drugs, systolic BP, diastolic BP, CKD, CKD stage, and eGFR were independent variables entered into the regression models whenever applicable. A two‐tailed P value <0.05 was considered as statistically significant. Data were analyzed using SPSS Program Package version 17.0 (SPSS Inc.). 3 RESULTS There were 200 patients enrolled in the CKD group and 50 healthy subjects enrolled in the control group. The mean age was 62.7, SD (±) 10.1, in the CKD group, and 61.9 ± 9.7 in the control group (P = 0.622). The demographic data and clinical characteristics are summarized in Table 1. There were no significant differences in age‐group, sex, diastolic BP, smoking status, cerebrovascular disease, intraocular pressure, or axial length between two groups. However, the mean BCVA in patients with CKD (logMAR: 0.130 ± 0.151, Snellen equivalent 20/27) was slightly worse than that of the control group (logMAR: 0.069 ± 0.103, Snellen equivalent 20/23) (P = 0.001). Figure 2 shows the mean logMAR BCVA in different stages of CKD. There is a trend toward worse visual acuity with more severe CKD. Table 1 Demographic data and clinical characteristics Control group (n = 50) CKD group (n = 200) P value[Link]

3 RESULTS There were 200 patients enrolled in the CKD group and 50 healthy subjects enrolled in the control group. The mean age was 62.7, SD (±) 10.1, in the CKD group, and 61.9 ± 9.7 in the control group (P = 0.622). The demographic data and clinical characteristics are summarized in Table 1. There were no significant differences in age‐group, sex, diastolic BP, smoking status, cerebrovascular disease, intraocular pressure, or axial length between two groups. However, the mean BCVA in patients with CKD (logMAR: 0.130 ± 0.151, Snellen equivalent 20/27) was slightly worse than that of the control group (logMAR: 0.069 ± 0.103, Snellen equivalent 20/23) (P = 0.001). Figure 2 shows the mean logMAR BCVA in different stages of CKD. There is a trend toward worse visual acuity with more severe CKD. Table 1 Demographic data and clinical characteristics Control group (n = 50) CKD group (n = 200) P value[Link] Age (mean ± SD) 61.9 ± 9.7 62.7 ± 10.1 0.622 Age‐group, n (%) 50 or below 7 (14.0) 29 (14.5) 0.996 51‐60 11 (22.0) 41 (20.5) 61‐70 22 (44.0) 89 (44.5) 71 or above 10 (20.0) 41 (20.5) Sex, n (%) Female 27 (54) 81 (40.5) 0.085 Male 23 (46) 119 (59.5) BMI, mean ± SD 23.8 ± 3.1 25.7 ± 5.0 0.001 Systolic BP (mm Hg), mean ± SD 131 ± 15 139 ± 20 0.009 Diastolic BP (mm Hg), mean ± SD 75 ± 9 77 ± 13 0.215 Smoking, n (%) 5 (10) 27 (13.5) 0.508 DM, n (%) 0 (0) 91 (45.5) <0.001 Use of anti‐hypertensive drug(s), n (%) 0 (0) 176 (88) <0.001 Cardiovascular disease, n (%) 0 (0) 40 (20) 0.001 Cerebrovascular disease, n (%) 0 (0) 2 (1.1) 1.000 LogMAR BCVA (mean ± SD) 0.069 ± 0.103 0.130 ± 0.151 0.001 Intraocular pressure, mm Hg (mean ± SD) 14.9 ± 2.3 15.0 ± 2.7 0.762 Axial length, mm (mean ± SD) 24.19 ± 1.21 23.88 ± 1.35 0.131 Comparing the CKD group and control group, the P values were calculated via independent sample t test for continuous variables and chi‐square test for categorical variables.

130 ± 0.151 0.001 Intraocular pressure, mm Hg (mean ± SD) 14.9 ± 2.3 15.0 ± 2.7 0.762 Axial length, mm (mean ± SD) 24.19 ± 1.21 23.88 ± 1.35 0.131 Comparing the CKD group and control group, the P values were calculated via independent sample t test for continuous variables and chi‐square test for categorical variables. John Wiley & Sons, LtdFigure 2 The distribution of mean logMAR BCVA in the control and CKD groups Chronic kidney disease group also had significantly higher value in BMI, systolic BP, prevalence of DM, number of patients using anti‐hypertensive drug, and prevalence of cardiovascular disease. The systemic conditions and classes of anti‐hypertensive drugs in CKD group are summarized in Table 2. There were 116 (66%) patients who used more than one class of drugs. Table 2 Systemic conditions and classes of ant‐hypertensive drugs using in patients with CKD patients Systemic conditions in 200 CKD patients Etiology of CKD, n (%) DM 75 (37.5) Hypertension 49 (24.5) Gout 11 (5.5) Other systemic diseases 7 (3.5) Chronic glomerulonephritis 25 (12.5) Polycystic kidney disease 9 (4.5) Other renal or urinary tract diseases 11 (5.5) Unknown etiology 13 (6.5) CKD stage, n (%) Stage 3 81 (40.5) Stage 4 43 (21.5) Stage 5 76 (38.0) Treatments, n (%)

Etiology of CKD, n (%) DM 75 (37.5) Hypertension 49 (24.5) Gout 11 (5.5) Other systemic diseases 7 (3.5) Chronic glomerulonephritis 25 (12.5) Polycystic kidney disease 9 (4.5) Other renal or urinary tract diseases 11 (5.5) Unknown etiology 13 (6.5) CKD stage, n (%) Stage 3 81 (40.5) Stage 4 43 (21.5) Stage 5 76 (38.0) Treatments, n (%) Hemodialysis 27 (13.5) Peritoneal dialysis 33 (16.5) Kidney transplantation 3 (1.5) Creatinine (mg/dL), mean ± SD 4.68 ± 4.37 eGFR (mL/min/1.73 m2), mean ± SD 26.9 ± 19.8 Classes of drug in 176 CKD patients using anti‐hypertensive drug(s), n (%) ACEI/ARB 128 (73) Calcium channel blockers 87 (49) Beta‐blockers 75 (43) Diuretics 51 (29) Alpha‐1 blockers 17 (10) Vasodilators 9 (5) Alpha‐2 agonists 2 (1) John Wiley & Sons, LtdThe fundus pathologies in the 200 eyes in the CKD group are summarized in Table 3. The most common finding was early AMD (20.5%). Diabetic retinopathy was present in 8% of the eyes. Table 4 compares the parafoveal vessel densities and foveal parameters between the control and CKD groups. Parafoveal vessel density was significantly decreased in the CKD group, in both SVP and DVP. This finding was consistent in all four parafoveal quadrants. Localized (Figure 3) or diffuse (Figure 4) rarefaction of retinal capillaries were observed in some patients with CKD. Other possible pathological changes of retinal capillary included blunt‐ended vessels, increased vascular tortuosity, and localized non‐perfusion area (Figure 3). Table 3 Fundus pathologies in 200 eyes in CKD group

Hemodialysis 27 (13.5) Peritoneal dialysis 33 (16.5) Kidney transplantation 3 (1.5) Creatinine (mg/dL), mean ± SD 4.68 ± 4.37 eGFR (mL/min/1.73 m2), mean ± SD 26.9 ± 19.8 Classes of drug in 176 CKD patients using anti‐hypertensive drug(s), n (%) ACEI/ARB 128 (73) Calcium channel blockers 87 (49) Beta‐blockers 75 (43) Diuretics 51 (29) Alpha‐1 blockers 17 (10) Vasodilators 9 (5) Alpha‐2 agonists 2 (1) John Wiley & Sons, LtdThe fundus pathologies in the 200 eyes in the CKD group are summarized in Table 3. The most common finding was early AMD (20.5%). Diabetic retinopathy was present in 8% of the eyes. Table 4 compares the parafoveal vessel densities and foveal parameters between the control and CKD groups. Parafoveal vessel density was significantly decreased in the CKD group, in both SVP and DVP. This finding was consistent in all four parafoveal quadrants. Localized (Figure 3) or diffuse (Figure 4) rarefaction of retinal capillaries were observed in some patients with CKD. Other possible pathological changes of retinal capillary included blunt‐ended vessels, increased vascular tortuosity, and localized non‐perfusion area (Figure 3). Table 3 Fundus pathologies in 200 eyes in CKD group Fundus pathologies n (%) Early AMD 41 (20.5) Diabetic retinopathy 16 (8) Mild NPDR 5 (2.5) Moderate NPDR 5 (2.5) Severe NPDR 4 (2) PDR 2 (1) Hypertensive retinopathy 11 (5.5) Epiretinal membrane 9 (4.5) Asymptomatic retinal vein occlusion 2 (1) Suspected hydroxychloroquine retinopathy 1 (0.5) Abbreviations: NPDR, non‐proliferative diabetic retinopathy; PDR, proliferative diabetic retinopathy.

ld NPDR 5 (2.5) Moderate NPDR 5 (2.5) Severe NPDR 4 (2) PDR 2 (1) Hypertensive retinopathy 11 (5.5) Epiretinal membrane 9 (4.5) Asymptomatic retinal vein occlusion 2 (1) Suspected hydroxychloroquine retinopathy 1 (0.5) Abbreviations: NPDR, non‐proliferative diabetic retinopathy; PDR, proliferative diabetic retinopathy. John Wiley & Sons, LtdTable 4 Parafoveal vessel densities and foveal parameters in control group (50 eyes) and CKD group (n = 200 eyes) Control CKD P value SVP vessel density, % (mean ± SD, range) Parafoveal 49.7 ± 2.9, 43.7‐55.8 46.7 ± 4.3, 34.2‐54.4 <0.001 Temporal 48.5 ± 2.8, 43.5‐53.6 45.4 ± 4.5, 32.1‐54.9 <0.001 Superior 50.8 ± 3.3, 43.5‐57.3 47.8 ± 4.6, 32.8‐56.0 <0.001 Nasal 49.1 ± 3.1, 42.8‐55.2 46.3 ± 4.3, 33.7‐54.6 <0.001 Inferior 50.6 ± 3.5, 41.8‐57.4 47.3 ± 4.8, 29.6‐55.9 <0.001 DVP vessel density, % (mean ± SD, range) Parafoveal 52.6 ± 2.9, 46.2‐58.2 50.1 ± 4.1, 37.9‐58.9 <0.001 Temporal 53.1 ± 2.9, 46.6‐59.6 50.2 ± 4.1, 37.0‐59.2 <0.001 Superior 52.4 ± 3.4, 44.2‐59.0 49.9 ± 4.5, 36.7‐59.9 <0.001 Nasal 53.4 ± 2.8, 47.4‐59.0 50.7 ± 4.2, 38.4‐60.0 <0.001 Inferior 51.7 ± 3.5, 44.0‐57.8 49.4 ± 4.6, 36.2‐59.3 <0.001 Foveal parameters (mean ± SD, range)

Parafoveal 52.6 ± 2.9, 46.2‐58.2 50.1 ± 4.1, 37.9‐58.9 <0.001 Temporal 53.1 ± 2.9, 46.6‐59.6 50.2 ± 4.1, 37.0‐59.2 <0.001 Superior 52.4 ± 3.4, 44.2‐59.0 49.9 ± 4.5, 36.7‐59.9 <0.001 Nasal 53.4 ± 2.8, 47.4‐59.0 50.7 ± 4.2, 38.4‐60.0 <0.001 Inferior 51.7 ± 3.5, 44.0‐57.8 49.4 ± 4.6, 36.2‐59.3 <0.001 Foveal parameters (mean ± SD, range) FAZ size (mm2) 0.295 ± 0.101, 0.072‐0.522 0.327 ± 0.133, 0.017‐0.824 0.118 FAZ perimeter (mm) 2.155 ± 0.419, 1.039‐3.257 2.296 ± 0.533, 1.169‐3.724 0.085 FAZ a‐circularity index 1.14 ± 0.04, 1.08‐1.27 1.16 ± 0.07, 1.07‐1.52 0.002 FD‐300 (%) 49.9 ± 4.1, 38.0‐56.7 47.6 ± 4.6, 30.4‐56.6 0.001 John Wiley & Sons, LtdFigure 3 A 41‐y‐old male patient with CKD stage 3. (A) Color fundus photograph reveals mild attenuation of retinal arteries. (B) A B‐scan image shows the segmentation site at the SVP and (C) at the DVP. (D) An OCTA image of SVP and (E) of DVP. The green arrows indicate a blunt‐ended retinal vessel. The purple arrow indicates the area with increased vessel tortuosity. A localized non‐perfusion area can be found at the nasal side of the FAZ. (F) A vessel density map of SVP. Multiple areas of capillary rarefaction are shown in deep blue. (G) A vessel density map of DVP. A few areas of capillary rarefaction can be found at the nasal and temporal‐superior side of the FAZ

sel tortuosity. A localized non‐perfusion area can be found at the nasal side of the FAZ. (F) A vessel density map of SVP. Multiple areas of capillary rarefaction are shown in deep blue. (G) A vessel density map of DVP. A few areas of capillary rarefaction can be found at the nasal and temporal‐superior side of the FAZ Figure 4 A 48‐y‐old male patient with CKDstage 5. (A) Color fundus photograph reveals mild attenuation of retinal arteries. (B) A B‐scan image shows the segmentation site at the SVP and (C) at the DVP. (D) An OCTA image of SVP and (E) of DVP. The green arrows indicate a disruption of the parafoveal capillary at SVP and DVP. (F) A vessel density map of SVP. Multiple areas of capillary rarefaction are shown in deep blue. (G) A vessel density map of DVP. DVP vessel density was well preserved in this patient

at the DVP. (D) An OCTA image of SVP and (E) of DVP. The green arrows indicate a disruption of the parafoveal capillary at SVP and DVP. (F) A vessel density map of SVP. Multiple areas of capillary rarefaction are shown in deep blue. (G) A vessel density map of DVP. DVP vessel density was well preserved in this patient Table 5 shows the multiple linear regression models for SVP and DVP vessel densities in all subjects (Model 1) and in patients with CKD (Model 2). In Model 1, age, DM, and CKD were negatively associated with both SVP and DVP vessel densities. Use of anti‐hypertensive drugs was positively associated with SVP vessel density, while smoking was negatively associated with DVP vessel density. In Model 2, age and DM were negatively associated with both SVP and DVP vessel densities. eGFR and use of anti‐hypertensive drugs were positively associated with SVP vessel density; smoking was negatively associated with DVP vessel density. Figure 5 demonstrates the distribution of mean parafoveal vessel density in the control, CKD without DM, and CKD with DM groups among different age‐groups. A trend toward vessel density reduction with aging and the presence of DM was observed in both SVP and DVP. Table 5 Multiple linear regression models for vessel densities Parafoveal SVP vessel density Parafoveal DVP vessel density Coefficient 95% CI P value Coefficient 95% CI P value Model 1 (multiple linear regression model for vessel densities in all 250 subjects)

Table 5 shows the multiple linear regression models for SVP and DVP vessel densities in all subjects (Model 1) and in patients with CKD (Model 2). In Model 1, age, DM, and CKD were negatively associated with both SVP and DVP vessel densities. Use of anti‐hypertensive drugs was positively associated with SVP vessel density, while smoking was negatively associated with DVP vessel density. In Model 2, age and DM were negatively associated with both SVP and DVP vessel densities. eGFR and use of anti‐hypertensive drugs were positively associated with SVP vessel density; smoking was negatively associated with DVP vessel density. Figure 5 demonstrates the distribution of mean parafoveal vessel density in the control, CKD without DM, and CKD with DM groups among different age‐groups. A trend toward vessel density reduction with aging and the presence of DM was observed in both SVP and DVP. Table 5 Multiple linear regression models for vessel densities Parafoveal SVP vessel density Parafoveal DVP vessel density Coefficient 95% CI P value Coefficient 95% CI P value Model 1 (multiple linear regression model for vessel densities in all 250 subjects) Age −0.116 −0.164 to −0.069 <0.001 −0.076 −0.124 to −0.028 0.002 DM −2.068 −3.133 to −1.003 <0.001 −1.512 −2.571 to −0.454 0.005 CKD −3.544 −5.431 to −1.656 <0.001 −1.953 −3.247 to −0.660 0.003 Use of anti‐hypertensive drug(s) 2.081 0.500 to 3.663 0.010 Smoking       −0.755 −1.454 to −0.056 0.034 Model 2 (multiple linear regression model for vessel densities in 200 CKD patients)

Age −0.116 −0.164 to −0.069 <0.001 −0.076 −0.124 to −0.028 0.002 DM −2.068 −3.133 to −1.003 <0.001 −1.512 −2.571 to −0.454 0.005 CKD −3.544 −5.431 to −1.656 <0.001 −1.953 −3.247 to −0.660 0.003 Use of anti‐hypertensive drug(s) 2.081 0.500 to 3.663 0.010 Smoking       −0.755 −1.454 to −0.056 0.034 Model 2 (multiple linear regression model for vessel densities in 200 CKD patients) Age −0.121 −0.176 to −0.066 <0.001 −0.067 −0.122 to −0.012 0.017 DM −2.227 −3.352 to −1.101 <0.001 −1.541 −2.658 to −0.424 0.007 eGFR 0.030 0.002 to 0.058 0.036 Use of anti‐hypertensive drug(s) 2.084 0.436 to 3.733 0.013 Smoking       −0.806 −1.607 to −0.005 0.049 Note Model 1: Multiple linear regression model with backward stepwise method in all 250 subjects: age, sex, smoking, BMI, DM, use of anti‐hypertensive drug(s), systolic BP, diastolic BP, and CKD were independent variables entered into the models. Model 2: Multiple linear regression model with backward stepwise method in 200 patients with CKD: age, sex, smoking, BMI, DM, use of anti‐hypertensive drug(s), systolic BP, diastolic BP, CKD stage, and eGFR were independent variables entered into the models. Abbreviation: CI, confidence interval. John Wiley & Sons, LtdFigure 5 The distribution of mean parafoveal vessel density in (A) the SVP and (B) the DVP. Error bars depict the 95% confidence interval

Model 2: Multiple linear regression model with backward stepwise method in 200 patients with CKD: age, sex, smoking, BMI, DM, use of anti‐hypertensive drug(s), systolic BP, diastolic BP, CKD stage, and eGFR were independent variables entered into the models. Abbreviation: CI, confidence interval. John Wiley & Sons, LtdFigure 5 The distribution of mean parafoveal vessel density in (A) the SVP and (B) the DVP. Error bars depict the 95% confidence interval 4 DISCUSSION There were two major findings in this study. First, retinal microvascular alterations may occur early in patients with CKD, before the onset of visual symptoms. Secondly, the microvasculature in different retinal layers may respond differently to systemic comorbidities. Chronic kidney disease has been associated with increased visual impairment and ocular diseases in prior epidemiology studies.10, 11, 12 In current study, the BCVA in the CKD group was worse than that of the control group. There was also a trend toward visual acuity decreasing with increased CKD severity (Figure 2). We found a high prevalence of early AMD (20.5%) among patients with CKD. Other retinopathies were not very common in this study because we enrolled visually asymptomatic patients. Our results represent the early retinal microvascular changes in patients with CKD.

cuity decreasing with increased CKD severity (Figure 2). We found a high prevalence of early AMD (20.5%) among patients with CKD. Other retinopathies were not very common in this study because we enrolled visually asymptomatic patients. Our results represent the early retinal microvascular changes in patients with CKD. We found significant retinal microvascular abnormalities in patients with CKD. The quantitative changes included capillary rarefaction in both SVP and DVP, decreased FD‐300 vessel density, and increased a‐circularity index of FAZ. Multiple regression models (Model 1) showed that CKD is an independent factor associated with decreased vessel densities in both SVP and DVP. Decreased vessel density may result from the localized or diffuse rarefaction of capillaries (Figure 3). Increased a‐circularity index of FAZ may be caused by the disruption of parafoveal capillary networks (Figure 4). Optical coherence tomography angiography also enabled us to visualize the morphological changes at the capillary level in these patients, such as blunt‐ended vessels, increased vascular tortuosity, and localized non‐perfusion area (Figure 3). The severity of retinal microvascular alteration may vary among patients with CKD. The multiple regression models (Model 2) illustrated the associated systemic factors in patients with CKD.

level in these patients, such as blunt‐ended vessels, increased vascular tortuosity, and localized non‐perfusion area (Figure 3). The severity of retinal microvascular alteration may vary among patients with CKD. The multiple regression models (Model 2) illustrated the associated systemic factors in patients with CKD. Age and DM are important factors negatively associated with vessel density in both SVP and DVP. A prior OCTA study has demonstrated that aging is associated with decreased vessel density in both the superficial and deep capillary plexus in the normal population.26 Chronic kidney disease may also contribute to premature aging of microcirculation.27 Diabetes mellitus was present in 91 (45.5%) patients in the current study. It had been well known that the reduction of vessel density is correlated to the severity of diabetic retinopathy.28, 29 Although most of our patients did not have any diabetic retinopathy, prior OCTA studies had demonstrated that microvascular changes may occur before clinically detectable diabetic retinopathy.30, 31, 32, 33, 34

well known that the reduction of vessel density is correlated to the severity of diabetic retinopathy.28, 29 Although most of our patients did not have any diabetic retinopathy, prior OCTA studies had demonstrated that microvascular changes may occur before clinically detectable diabetic retinopathy.30, 31, 32, 33, 34 Hypertension is very common among patients with CKD. About 85% of patients with CKD may have coexistent hypertension.35 In our study, the most commonly used class of anti‐hypertensive drug was ACEI/ARB (73.1%), followed by calcium channel blocker (49.7%). Prior studies showed that both ACEI/ARB and calcium channel blockers may improve the retinal arteriolar narrowing and capillary rarefaction in hypertensive patients.36, 37 The benefit of anti‐hypertensive drugs may have resulted either from better‐controlled BP38 or from other pharmacological effects independent of the BP lowering effect. For an example, a localized RAS had been found within the eye, such as in retinal microvasculature, Müller cells, and ganglion cells.39 The activation of RAS may promote retinal neovascularization, inflammation, oxidative stress, and neuronal and glial dysfunction.39 So, one of the possible mechanisms may be a protective effect in the retina via suppression of RAS by ACEI/ARB.39

, such as in retinal microvasculature, Müller cells, and ganglion cells.39 The activation of RAS may promote retinal neovascularization, inflammation, oxidative stress, and neuronal and glial dysfunction.39 So, one of the possible mechanisms may be a protective effect in the retina via suppression of RAS by ACEI/ARB.39 Cigarette smoking is a common risk factor for CKD and various ocular diseases such as AMD and cataract.11 It has been well known that cigarette smoking can result in morphological changes (ie, vessel wall injury, capillary loss) and functional changes in microcirculation.40 Interestingly, smoking was negatively associated with vessel density in DVP, but not in SVP, in the current study. A similar finding has also been reported previously in diabetic patients without diabetic retinopathy.41 Current smoker status was correlated with lower vessel density in the deep capillary plexus, but not associated with vessel density in the superficial capillary plexus.41 Further study is necessary to confirm this observation and to determine why deep retinal capillaries are more susceptible to injury from cigarette smoking.

t smoker status was correlated with lower vessel density in the deep capillary plexus, but not associated with vessel density in the superficial capillary plexus.41 Further study is necessary to confirm this observation and to determine why deep retinal capillaries are more susceptible to injury from cigarette smoking. In our study, vessel densities in SVP and DVP were associated with different systemic factors. There are three capillary plexuses over the parafoveal area, namely the superficial, intermediate, and deep capillary plexus.42 In vivo human study showed that each of these three capillary plexuses may have its own feeding arteriolar supply and draining venules.43 Each capillary plexus has different anatomical structures and may have its own autoregulation.43 The plexuses may respond differently to systemic condition alteration, such as changes in BP and oxygenation, or to retinal functional hyperemia evoked by a flickering light stimulus.44, 45, 46 In our study, anti‐hypertensive drugs and eGFR were associated with vessel density in SVP, but not in DVP. On the contrary, smoking was associated with vessel density in DVP, but not in SVP. Therefore, our results support the hypothesis that microvasculature in different retinal layers may respond differently to varying systemic factors.

hypertensive drugs and eGFR were associated with vessel density in SVP, but not in DVP. On the contrary, smoking was associated with vessel density in DVP, but not in SVP. Therefore, our results support the hypothesis that microvasculature in different retinal layers may respond differently to varying systemic factors. There are several limitations in this study. The study is limited by its small sample size and cross‐sectional study design. Longitudinal follow‐up data were not available. Furthermore, we enrolled patients without visual symptoms into the CKD group. So, our results may reflect early retinal microvascular alterations rather than late‐stage retinopathies. In summary, our study demonstrated that patients with CKD had significant rarefaction of retinal microvasculature in both SVP and DVP. Morphological changes in the retinal capillaries were observed via OCTA. The microvasculature in the different retinal layers may respond differently to varying systemic factors. Ophthalmologists should take these microvascular changes into consideration when interpreting OCTA images in patients with CKD. PERSPECTIVE Optical coherence tomography angiography showed that patients with CKD may have rarefaction and morphological changes of retinal microvasculature in the superficial and DVPs. The microvasculature in different retinal layers may respond differently to various systemic factors. CONFLICT OF INTEREST No authors have any financial/conflicts of interest to disclose.

INTRODUCTION A number of challenges to tissue homeostasis result in the expansion of the microvasculature within the affected tissue. Either as an adaptation to long-term physiological changes or in response to pathological insult, vascular beds in affected tissues often expand, leading to increases in vessel density and changes in network topology. For example, in chronic exercise, skeletal muscle vasculatures expand to increase blood perfusion volumes and capillary surface area for exchange to accommodate new metabolic demands associated with elevated muscle workloads [12,13]. Following ischemic injury, expansion of the vascular beds adjacent to the ischemic zone provides additional blood flow to the insulted area to relieve the ischemia and to facilitate tissue repair [2,34,44]. However, uncoupling of new vasculature formation from tissue needs and function leads to nonhomeostatic changes, thereby exacerbating dysfunction and creating disease [5,17,46].

nt to the ischemic zone provides additional blood flow to the insulted area to relieve the ischemia and to facilitate tissue repair [2,34,44]. However, uncoupling of new vasculature formation from tissue needs and function leads to nonhomeostatic changes, thereby exacerbating dysfunction and creating disease [5,17,46]. Effective expansion of a vasculature involves a complex interplay of vessel growth (i.e., angiogenesis) and vascular remodeling/adaptation that, in homeostatic processes, results in a vascular tree with a larger vascular volume and surface area that still retains effective perfusion potential [30]. Intimately coupled with remodeling processes such as arteriogenesis [51] and vessel diameter adaptation [45], angiogenesis, either by splitting or sprouting processes, is the primary means by which new vessel segments are added to the vascular tree. Decades of research have uncovered a vast array of cellular and molecular mechanisms underlying endothelial and perivascular cell activities related to angiogenesis, including recent discoveries of molecules orchestrating neovessel sprouting and elongation (see [1,6,9,12,16]). Despite these significant advances in our understanding of angiogenesis (sprouting angiogenesis in particular), much is still to be learned. One such area relates to the interplay between the intact, growing neovessel and the surrounding, 3-D tissue environment and how bulk tissue dynamics influences larger scale angiogenic behavior.

ignificant advances in our understanding of angiogenesis (sprouting angiogenesis in particular), much is still to be learned. One such area relates to the interplay between the intact, growing neovessel and the surrounding, 3-D tissue environment and how bulk tissue dynamics influences larger scale angiogenic behavior. During angiogenesis, growing neovessels must move through the tissue stroma, navigating around and between parenchymal cells. Furthermore, to effectively integrate into the existing vascular network, the growing neovessel must be able to locate and extend toward other vessels (likely other growing neovessels rather than existing, mature vessels). How these neovessels navigate through this three-dimensional stromal space is not clear, especially in the context of the adult tissue where there is a paucity of “global” patterning cues such as those in embryo (when body and organ plans are being established). Regardless of the mechanism, the stroma is playing a central role in neovessel navigation, acting directly or indirectly to influence neovessel behavior.

lly in the context of the adult tissue where there is a paucity of “global” patterning cues such as those in embryo (when body and organ plans are being established). Regardless of the mechanism, the stroma is playing a central role in neovessel navigation, acting directly or indirectly to influence neovessel behavior. STROMA While each tissue bed exhibits a specific and unique composition, most tissues generally contain some combination of parenchymal cells embedded within a stromal environment. This stroma is comprised of an extracellular matrix within which reside tissue support cells and the microvasculature. In general, the matrix environment is an interstitial gel consisting of collagens, glycosaminoglycans, and hyalurons all of which establish a fluid-rich, 3-D environment [18,31,35]. Matrix molecules also assemble into fibrils/fibers, which are integrated within and around the matrix gel creating a complex biochemical and mechanical milieu [29,31]. A key feature of most (if not all) stromal matrices is the viscoelastic behavior exhibited when mechanically loaded [28,39,42]. Acting simultaneously with the stretch and relaxation of the elastic elements of the matrix (typically the fibrous components) is a viscous drag created by the water trapped within the gel matrix. Thus, the stroma generally will undergo time-dependent, nonlinear changes in strain (e.g., change shape) when a stress is applied (e.g., as in stretching). This time-dependent dynamic is important for a number of tissue responses and functions [32,36,48]. Also, as will be discussed later, the viscoelasticity of the stromal matrix impacts angiogenesis and neovessel guidance.

nonlinear changes in strain (e.g., change shape) when a stress is applied (e.g., as in stretching). This time-dependent dynamic is important for a number of tissue responses and functions [32,36,48]. Also, as will be discussed later, the viscoelasticity of the stromal matrix impacts angiogenesis and neovessel guidance. As mentioned, embedded within the stromal matrix are a variety of tissue cell types. Generally thought to play critical roles in supporting parenchyma and tissue homeostasis, these cells synthesize and remodel matrix, and interact with parenchyma and the microvasculature (via paracrine and juxtacrine processes) to integrate tissue function [3,7,22,35,37]. While the stromal cells within a given tissue may have phenotypes unique to that tissue, the collection of stromal cell types across the many tissues in the body include fibroblasts, perivascular cells, mesenchymal pluripotent cells, and immune cells. No one cell appears to be more important in tissue health and function than another, although all are necessary as deficits in any one cell type may lead to dysfunction and disease. While virtually all cells synthesize and deposit matrix, the fibroblast has long been considered the primary driver cell type responsible for much of the stromal matrix deposition and remodeling, particularly in fibrotic diseases [23,49]. Only recently have the roles of other mesenchymal and immune cells in matrix deposition been uncovered [10,21,38,40,41]. Similarly, the stromal cells influence angiogenesis and vascular remodeling. Virtually all cells secrete factors that directly or indirectly regulate vascular cell activity and angiogenesis. Recently, there has been a renewed emphasis in the role the tissue macrophage plays in regulating angiogenesis either through the production of angiogenic factors and chemokines [19,33,43] or through direct interaction with the angiogenic vascular elements [14].

irectly regulate vascular cell activity and angiogenesis. Recently, there has been a renewed emphasis in the role the tissue macrophage plays in regulating angiogenesis either through the production of angiogenic factors and chemokines [19,33,43] or through direct interaction with the angiogenic vascular elements [14]. As an integrated complex of viscoelastic matrix and various cell types, the stroma creates a dynamic environment entailing a vast array of biochemical/biomolecular stimuli and biomechanical influences. These input signals to both the parenchymal and stromal cells are the means by which tissue function and health are performed and maintained. It is in this complex environment that angiogenesis occurs, during which growing neovessels must integrate with each other to effectively expand the tissue microcirculation.

ces. These input signals to both the parenchymal and stromal cells are the means by which tissue function and health are performed and maintained. It is in this complex environment that angiogenesis occurs, during which growing neovessels must integrate with each other to effectively expand the tissue microcirculation. NEOVESSEL NAVIGATION DURING SPROUTING ANGIOGENESIS Our understanding of the processes underlying the formation of a new angiogenic sprout from a parent microvessel and its growth have been well-described. Briefly, following an angiogenic stimulus, the microvessel wall is locally remodeled concomitant with the sprouting of an endothelial cell away from the microvessel. This endothelial cell establishes the growing tip of the angiogenic sprout and serves to lead the advancement of the forming neovessel during angiogenesis. Cross talk between the tip cell and the nascent stalk cells that make up the bulk of the growing neovessel (see [9,15,16]) maintains the long aspect ratio of the neovessel, keeping growth and migration directed. Thus, the tip cell serves to navigate through the stromal environment while coordinating the organization of the lagging, proliferating stalk cells.

nascent stalk cells that make up the bulk of the growing neovessel (see [9,15,16]) maintains the long aspect ratio of the neovessel, keeping growth and migration directed. Thus, the tip cell serves to navigate through the stromal environment while coordinating the organization of the lagging, proliferating stalk cells. Recently, we have observed in detail the behavior of a sprouting neovessel as it originates from the parent microvessel, advances through the stroma, and connects with a second advancing neovessel. Over a six-day period, time-lapse, two-photon video microscopy revealed that the behavior of the tip endothelial cell is dynamic, involving the extension of numerous, short filopodia during the entire neovessel growth period (Figure1). Coordinately, the neovessel advances, retracts, and changes direction as the filopodia extend and retract. Interestingly, the stalk of the neovessel does not uniformly advance with the tip cell. Instead, the cell bodies of stalk endothelial cells move episodically along in the neovessel, first trailing behind and then quickly moving forward toward the tip. It is clear from the 3-D volume renderings of these time-lapse videos that growing neovessels are able to locate and advance toward other neovessels, some of which can be separated by hundreds of micrometers. Finally, collagen fibrils are actively remodeled, condensed, and deformed at the neovessel tip and along the neovessel stalk during the entire angiogenesis process. This latter observation highlights a question that our collaborative team and others have been addressing for many years related to the dynamics between the growing neovessel and the surrounding stromal matrix. Our focus has been on the role that biomechanical factors play in this dynamic. Specifically, we have been working to determine what happens to the matrix environment during angiogenesis from a stress–strain perspective and how this reciprocally influences neovessel behavior and overall microvascular outcomes.

matrix. Our focus has been on the role that biomechanical factors play in this dynamic. Specifically, we have been working to determine what happens to the matrix environment during angiogenesis from a stress–strain perspective and how this reciprocally influences neovessel behavior and overall microvascular outcomes. Figure 1 A sequence of still frames from time-lapse video of neovessel sprouting, growth, and inosculation within a collagen gel stroma. Microvessels (red) were imaged via confocal microscopy and collagen fibril structure (green) was visualized using SHG imaging. Over the course of ∼4.5 days, a neovessel sprout (white arrow) forms, grows, changes direction to eventually inosculate (wide arrow) with a second neovessel (open arrow) that appears from out of the field of view. Brackets indicate areas of collagen condensation occurring at neovessel walls. The asterisk marks a neovessel sprout that forms and then regresses approximately three days later.

ows, changes direction to eventually inosculate (wide arrow) with a second neovessel (open arrow) that appears from out of the field of view. Brackets indicate areas of collagen condensation occurring at neovessel walls. The asterisk marks a neovessel sprout that forms and then regresses approximately three days later. GENERAL FORCES PRESENT DURING ANGIOGENESIS As the neovessel advances through the stromal matrix, there is a combination of both pulling and pushing forces [50]. As shown by many experiments using isolated cells, the tip cell is likely extending forward into the surrounding matrix, anchoring to the matrix, and then pulling itself forward [24]. Being attached to neighboring stalk cells, this effort will necessarily apply stress along the neovessel as the stalk (still attached to the parent vessel and comprised of a chain of cells) will not readily move forward with the tip cell. This stress, in part, may explain the episodic advancement of the stalk cell cytoplasm observed in the time-lapse video. It is likely that as the tip cell attempts to pull forward the adjacent stalk segment experiences tension. This tension may then pull stalk cells forward following subsequent release of downstream anchors resulting in a “sling-shot” type effect. Coordinately, there are cells within the stalk segment that are dividing, thereby providing the cellular building blocks for the growing neovessel. Given that proliferation must be constrained along the long axis of the neovessel during angiogenesis (otherwise the vessel would grow outward in diameter as opposed to lengthwise growth), the two daughter cells formed by cell division now occupy the space previously filled by the one parent cell. Either these two daughter cells are compressed and become smaller or they push outward along the neovessel length to create new space. These outward forces would then act to push forward the portion of the neovessel distal to the cell division event (conversely the stalk segment proximal to the cell division would be compressed against the parent microvessel). It is intriguing to think that perhaps the release event contributing to the rapid, episodic forward movement of cell bodies while under stress may be the result of a proliferative event and the addition of a new cell to the stalk, which is pulled forward by the tip before establishing new substrate anchors.

rovessel). It is intriguing to think that perhaps the release event contributing to the rapid, episodic forward movement of cell bodies while under stress may be the result of a proliferative event and the addition of a new cell to the stalk, which is pulled forward by the tip before establishing new substrate anchors. ANGIOGENESIS MODIFIES THE MATERIAL PROPERTIES OF THE STROMA Regardless of the types of stresses generated by the neovessel, it is clear that the neovessel mechanically remodels the surrounding stromal matrix. Using a model of sprouting angiogenesis in which neovessels grow from isolated, intact microvessel segments within type I collagen gels, our collaborative team has shown that the overall stroma harboring the angiogenesis event becomes softer initially, and then later stiffer, as angiogenesis proceeds [26]. Initially, preceding angiogenesis, the simple stroma (microvessels + collagen) is slightly less stiff (more compliant) than collagen gels alone. The stroma becomes even less stiff early following angiogenic sprouting from the parent microvessels. However, in the presence of actively growing neovessels, the stroma stiffens to 1.5× that of empty collagen gels [26] (Figure2). These changes are occurring as neovessel density increases. The period of lessened stiffening is associated with increased metalloproteinase activity [26]. However, MMP expression remains elevated during the entire angiogenesis period (Figure3), indicating that MMPs are inhibited during the later stages of active neovessel growth (i.e., after the initial sprouting event) or some other process is occurring that counteracts matrix degradation to stiffen the matrix in later growth phases. While neovessel density is increasing in this system, it is unlikely that the addition of new neovessel segments to the gel is sufficient to explain the increases in stiffness observed later in angiogenesis. Neovessels contribute less than ∼1%; to the volume of the constructs, even in fully compacted gels (Weiss JA, personal observation). Also, inclusion of nylon fibers (as inert neovessel mimics) into collagen gels without microvessels does not appreciably change the collagen stiffness (Weiss JA, unpublished data). Interestingly, concomitant with the increase in neovessel density is the contraction of the collagen stroma. However, this contraction does not occur until the later phase when stroma stiffness increases so dramatically [26], suggesting a causal relationship.

change the collagen stiffness (Weiss JA, unpublished data). Interestingly, concomitant with the increase in neovessel density is the contraction of the collagen stroma. However, this contraction does not occur until the later phase when stroma stiffness increases so dramatically [26], suggesting a causal relationship. Certainly, compaction of the collagen fibrils due to the gel contraction would produce a stiffer system as fibril density increases. Empty collagen gels do not appreciably contract or change stiffness over time [28]. Yet, the addition of cytochalasin D to the microvessel system for disruption of the cytoskeleton, and, therefore, force generation by the cells, prevented gel contraction, even at high neovessel densities [52]. Thus, growing neovessels alter matrix material properties during angiogenesis via a combination of protease activity and force generation. Figure 2 Phase microscopy images of neovessels growing from isolated parent microvessels in 3-D collagen gels over the course of 10 days. Given below is an analysis of the stiffness of the microvessel-collagen cultures over this same time course. Modified from [26]. Figure 3 Expression of select MMPs by neovessels growing through collagen over the same time course as shown in Figure2. Shown on the right are two adjacent images slices from a Z-stack of confocal images in which DQ-gelatin was used to localize gelatinase activity to a parent microvessel at a time when neovessel sprouting is just beginning. Modified from [26].

ssels growing through collagen over the same time course as shown in Figure2. Shown on the right are two adjacent images slices from a Z-stack of confocal images in which DQ-gelatin was used to localize gelatinase activity to a parent microvessel at a time when neovessel sprouting is just beginning. Modified from [26]. GROWING NEOVESSELS REORGANIZE COLLAGEN FIBRILS The contraction of the collagen stroma by the growing neovessels suggests that the endothelial cells comprising the neovessel are engaging with the collagen and exerting force. While there are many different types of collagen, fibrous collagen (types I, II, and III) makes up the bulk of the interstitial collagen of the stroma [29,47]. Type I collagen polymerization is complex and can lead to multiple types of secondary and tertiary structure and organization [47]. Collagen is synthesized and assembled by the cell into a triple helix, which is further bundled via side-by-side interactions among many helices into fibrils which contribute to the larger fibers. Because of its highly ordered structure, the fibrils within a stroma are readily visualized via different microscopic approaches [4,53]. Using SHG and two-photon microscopy to visualize the fibrils, we determined that the interstitial collagen matrix is actively reorganized by the growing neovessel during angiogenesis [20]. This reorganization presented in two different configurations: fibril rearrangement and condensation. As the sprouting neovessel advanced away from the parent vessel out into the surrounding stroma, collagen fibrils at the leading tip were organized such that they radiated out from the sprout and neovessel tip along the axis of neovessel growth (Figure4). Concomitant with this nascent fibril reorganization at the sprout and neovessel tips is an SHG-bright (i.e., fibrous) layer of collagen accumulating at regions along the neovessel wall [20]. We have interpreted this condensation as a consequence of fibril recruitment and compaction by the neovessel endothelial cells; perhaps these are the fibrils first aligned by the tip cell, which are then compacted as the neovessel moves through that fibril region. However, the condensed collagen may be due to new collagen synthesis instead of or in addition to fibril recruitment. Ongoing experiments are addressing this. It is not yet clear what role fibril alignment and condensation play during neovessel growth and navigation through the stroma.

moves through that fibril region. However, the condensed collagen may be due to new collagen synthesis instead of or in addition to fibril recruitment. Ongoing experiments are addressing this. It is not yet clear what role fibril alignment and condensation play during neovessel growth and navigation through the stroma. While not specifically addressed in this previous study, the size of the region of fibrils influenced by the neovessel depends on the density of collagen in the matrix with smaller zones of fibril aligned at the neovessel tip in more dense collagen gels (Hoying JB, personal observation). Given that the density of collagen also influences neovessel density, length and branch points [52], it is likely that fibril alignment (and the ease at which these fibrils align) may influence neovessel guidance and growth persistence. Figure 4 Two examples of collagen fibril reorientation by the tips of growing neovessels. Collagen fibrils (white) and endogenous endothelial cell fluorescence were visualized using SHG/two-photon microscopy. Yellow arrows point to new sprouts arising from the parent microvessel (green). Modified from [20].

While not specifically addressed in this previous study, the size of the region of fibrils influenced by the neovessel depends on the density of collagen in the matrix with smaller zones of fibril aligned at the neovessel tip in more dense collagen gels (Hoying JB, personal observation). Given that the density of collagen also influences neovessel density, length and branch points [52], it is likely that fibril alignment (and the ease at which these fibrils align) may influence neovessel guidance and growth persistence. Figure 4 Two examples of collagen fibril reorientation by the tips of growing neovessels. Collagen fibrils (white) and endogenous endothelial cell fluorescence were visualized using SHG/two-photon microscopy. Yellow arrows point to new sprouts arising from the parent microvessel (green). Modified from [20]. GROWING NEOVESSELS RESPOND TO DEFORMATION FIELDS AND NOT STRESS FIELDS It is clear from these studies that the growing neovessel is actively remodeling the stromal matrix environment during angiogenesis: metalloproteinases mediate changes in collagen dynamic stiffness and neovessel endothelial cells pull and rearrange collagen fibrils. Together, these activities can lead to relatively large deformations of the stroma environment, unless the boundaries of the stroma are anchored [11,52]. Interestingly, the extent and direction of the stroma deformation can have profound effects on the morphology and organization of neovessels [27]. Growing neovessels within round collagen gels of which all edges are anchored or free to deform uniformly randomly orient in the gel [27]. However, in a rectangular gel for which the two short sides of the gel are anchored while the two long sides are free to deform, the neovessels now orient parallel to the long axis of constraint (Figure5). Conversely, anchoring the long sides but not the short sides has little impact on neovessel orientation (Figure5). Importantly, unlike in the long-axis-constrained gels, little deformation occurs in these short-axis-constrained gels. Similar outcomes were observed if the gels were stretched (periodically or statically) along one axis allowing for deformation of the other axis [26]. Interestingly, in gels in which deformation did not occur, there was less angiogenesis and fewer neovessel branching [52].

n occurs in these short-axis-constrained gels. Similar outcomes were observed if the gels were stretched (periodically or statically) along one axis allowing for deformation of the other axis [26]. Interestingly, in gels in which deformation did not occur, there was less angiogenesis and fewer neovessel branching [52]. Figure 5 Neovessel alignment in 3-D angiogenesis cultures in constrained conditions. (A) Schematic of long-axis and short-axis-constrained gels (pink) by anchors (black bars). Gels are free to contract perpendicular to the constrained axis. (B) Projections of confocal image Z-stacks of neovessels (green) within these constrained cultures. (C) Top views of constrained cultures showing the extent and direction of gel deformation. (D) Finite element simulations of constrained gels showing relative strain fields in the long (Exx), short (Eyy), and vertical (Ezz) axes. Scale bar indicates the magnitude of the strains, which is a ratio of (L − Lo)/Lo. Negative values indicate compressive strain (contraction). Modified from [11,52].

gel deformation. (D) Finite element simulations of constrained gels showing relative strain fields in the long (Exx), short (Eyy), and vertical (Ezz) axes. Scale bar indicates the magnitude of the strains, which is a ratio of (L − Lo)/Lo. Negative values indicate compressive strain (contraction). Modified from [11,52]. Modeling of collagen and neovessel dynamics indicate that, due to the viscoelastic nature of fibrillar collagen gels [28], cellular traction forces applied by angiogenic neovessels to the collagen fibril lattice do not accumulate and are rapidly dissipated in this simple stroma system [11,52]. Experiments in which one end of the anchored long-axis-constrained culture was released in the presence of cytochalasin indicated that there is effectively no stress on these gels that is maintained longer than seconds [52]. In addition, as mentioned, neovessels do not form an aligned network in collagen gels that are fully constrained around the edges (as in the fully constrained round and short-axis-constrained configurations described above). Finally, the density of collagen fibrils and, therefore, the compliance (or stiffness) of the collagen matrix influences the degree of matrix contraction and subsequently neovessel alignment and behavior [11]. Thus, given the absence of a persistent tension in the matrix and the coordinate changes in neovessel alignment with changes in the matrix strain field, it appears that during angiogenesis, cellular traction forces produced by growing neovessels result in stromal deformation (i.e., compressive strain), the extent of which is determined by the effective compliance and boundary conditions of the matrix. When the boundary constraints allow directionality in the resulting strain field, neovessels align perpendicular to the primary direction of compressive strain and not along directions of tensile stress [52]. Interestingly, the collagen fibrils within the gel also align along the axis of constraint (i.e., perpendicular to the compressive strain axis) regardless of whether there are vessels present or not [52], suggesting that perhaps neovessels are following collagen fibril paths via contact guidance [25]. It is not yet clear whether fibril alignment precedes neovessel alignment or vice versa. But, if neovessel alignment lags behind fibril alignment, it would suggest that fibril orientation within a stromal matrix can influence neovessel guidance and orientation.

following collagen fibril paths via contact guidance [25]. It is not yet clear whether fibril alignment precedes neovessel alignment or vice versa. But, if neovessel alignment lags behind fibril alignment, it would suggest that fibril orientation within a stromal matrix can influence neovessel guidance and orientation. If true, perhaps then the active local reorientation of collagen fibrils by the neovessels described earlier may in fact be a means by which neovessel growth is mechanically directed. If neovessels preferentially grow along zones of parallel fibrils (as might be indicated in the constrained gels) and the neovessel tip is actively aligning fibrils in one direction (i.e., parallel), then it seems reasonable that this dynamic might maintain a persistent direction of neovessel growth, the path of which would be influenced by the deformability of the fibril lattice. How spatial gradients of growth factors likely present within the stroma influence directional behavior in the context of these guiding mechanical stimuli has yet to be determined.

ight maintain a persistent direction of neovessel growth, the path of which would be influenced by the deformability of the fibril lattice. How spatial gradients of growth factors likely present within the stroma influence directional behavior in the context of these guiding mechanical stimuli has yet to be determined. MICROVASCULAR NETWORK TOPOLOGY IS ALSO INFLUENCED BY STROMAL DEFORMATION As mentioned, angiogenesis in the adult results in the addition of new vessel segments to an existing microcirculation. Thus, the neovessels generated via angiogenesis must inosculate with other microvessels to form a provisional network consisting of new and existing microvessel segments eventually leading to a mature network of defined topology [30]. Our previous research has shown that stroma deformation influences angiogenesis outcomes with respect to neovessel orientation and character of neovessel growth. It appears that these same biomechanical stimuli can act to determine the final topology of a new microcirculation as well [8]. While constraining the long axis of a collagen gel containing angiogenic neovessels results in anisotropic neovessel alignment, removal of this constraint during progression to a mature network disrupts this organization, resulting in randomly oriented segments within the final network (Figure6) [8]. However, maintaining the long-axis constraint during post-angiogenesis network remodeling and maturation resulted in arrays of parallel, mature microvessels within the functional microcirculation (Figure6). Interestingly, as seen during angiogenesis, this biomechanical stimulus correlated with changes in the segment makeup of the network as the mechanically constrained networks had a higher proportion of capillaries (Figure6). Despite this change in vessel composition, the constrained, ordered networks were as equally perfused as the unconstrained, disordered networks [8]. As observed in the other studies, there was substantial uniaxial stromal deformation associated with the parallel orientation of the vessel segments in the constrained microvascular networks and uniform deformation in the unconstrained networks.

ks were as equally perfused as the unconstrained, disordered networks [8]. As observed in the other studies, there was substantial uniaxial stromal deformation associated with the parallel orientation of the vessel segments in the constrained microvascular networks and uniform deformation in the unconstrained networks. Figure 6 Architecture of and vessel type distribution in microvascular networks formed in unconstrained (A) or long-axis-constrained (B) stromal environments. The axis of constraint in (B) is from left to right. (C) Distribution of microvessel types and vessel density within the unconstrained (UF) and constrained (F) networks. Modified from [8].

d vessel type distribution in microvascular networks formed in unconstrained (A) or long-axis-constrained (B) stromal environments. The axis of constraint in (B) is from left to right. (C) Distribution of microvessel types and vessel density within the unconstrained (UF) and constrained (F) networks. Modified from [8]. Discussion During angiogenesis, the growing neovessel responds to a broad spectrum of inputs from the stroma and parenchyma [6]. Within the stroma, nonvascular cells produce a variety of soluble and insoluble paracrine signals modulating and directing neovessel growth. Molecular signals from the extracellular matrix similarly mediate neovessel stability, growth, and morphology. Growing evidence now demonstrates that the physical character of the stromal matrix also strongly influences neovessel dynamics. Furthermore, there is a reciprocal interplay between the neovessel and the matrix (Figure7) such that the neovessel actively remodels the matrix, which in turn leads to matrix deformation, which in turn influences neovessel growth and morphology. Thus, the matrix, as a biomechanical environment, is not simply a passive structure that the neovessel must move through but instead is an indirect determinant of neovessel growth and navigation. We envision the following working model of how neovessel-matrix interactions contributes to neovessel navigation through the stromal environment during angiogenesis. At the single neovessel level, as the sprouting neovessel extends forward, the preceding matrix fibrils are pulled by the neovessel tip thereby aligning the fibrils in advance of the neovessel. Because a neovessel appears to preferentially grow in the direction of aligned fibrils, it will extend into this freshly aligned zone simultaneously aligning the fibrils in the next forward zone and so on. Coordinately, matrix condensation along the growing neovessel significantly reduces adjacent fibril deformability thereby retarding changes in direction without a more active remodeling event (i.e., metalloproteinase degradation). As many neovessels (tens of thousands in our experiments) are simultaneously pulling on the matrix fibrils, there is a global contraction of the stromal environment. In the absence of any boundary constraints (or complete boundary constraint), only local fibril remodeling will occur resulting in local control of neovessel navigation.

els (tens of thousands in our experiments) are simultaneously pulling on the matrix fibrils, there is a global contraction of the stromal environment. In the absence of any boundary constraints (or complete boundary constraint), only local fibril remodeling will occur resulting in local control of neovessel navigation. However, if the stroma is not fully constrained, it will deform in bulk resulting in global changes to fibril organization. Depending on the nature of the deformation (uniaxial in our experiments), this larger fibril organization, the extent of which is determined by the matrix deformability, will possibly establish a larger scale neovessel organization as they grow within this prestructured fibril environment. Differences in stromal compliance and physical constraints may explain, in part, how the different vascular topologies arise specific to different tissues (compare the topology of the mesentery to that of skeletal muscle). Figure 7 Schematic highlighting the interplay between the growing neovessel and the surrounding matrix structure.

However, if the stroma is not fully constrained, it will deform in bulk resulting in global changes to fibril organization. Depending on the nature of the deformation (uniaxial in our experiments), this larger fibril organization, the extent of which is determined by the matrix deformability, will possibly establish a larger scale neovessel organization as they grow within this prestructured fibril environment. Differences in stromal compliance and physical constraints may explain, in part, how the different vascular topologies arise specific to different tissues (compare the topology of the mesentery to that of skeletal muscle). Figure 7 Schematic highlighting the interplay between the growing neovessel and the surrounding matrix structure. While there are a number of implications in this model, deformability of the fibrils, which can be modulated via cross-linking, fibril density, other matrix elements, and stromal cells, is a central feature. The extent at which a neovessel can deform the fibrillar matrix, considering both the forces needed to do so, and the size of the affected fibril zone extending from the neovessel tip, determines the direction of the growing neovessel tip and, perhaps, even the generation of new branch points. In stiffer, less deformable matrices, the neovessel can align a smaller preceding zone of fibrils effectively retarding neovessel activity. Conversely, a neovessel can create large zones of aligned fibrils in a highly deformable matrix, which might enable directional changes and/or branching. Changes to the matrix deformability via cross-linking or additional matrix elements that entangle and interconnect fibril networks would alter the deformability of matrix elements and fibril networks, which would in turn regulate neovessel growth direction and character. This may explain why neovessel activity is limited in collagen gels prepared with higher concentrations of collagen and therefore higher densities of fibrils [52]. Similarly, some tissues have specifically organized vasculatures, which may reflect, in part, the composition of the stromal matrix and, therefore, the matrix deformability.

why neovessel activity is limited in collagen gels prepared with higher concentrations of collagen and therefore higher densities of fibrils [52]. Similarly, some tissues have specifically organized vasculatures, which may reflect, in part, the composition of the stromal matrix and, therefore, the matrix deformability. It is difficult to resist speculating on the role matrix (and specifically fibril) deformation might play in neovessel navigation. It is generally considered that neovessels grow toward higher sources of angiogenic factors or away from repulsive factors [1]. However, growth factor gradients alone do not explain how one active neovessel is able to locate and grow toward another active neovessel resulting in the inosculation between the two neovessels. Besides the fact that two neovessels might incidentally grow toward each other as they grow to the top of a growth factor gradient, similar gradients do not adequately explain how the two neovessels can locate each other so effectively (as we see in the time-lapse video of angiogenesis). If, for example, the tip cell was to generate a paracrine gradient to attract a nearby neovessel, there would need to be many different gradient molecules such that each neovessel is responding to a separate signaling gradient and not its own (Figure8). While there may indeed be growth factor signals that direct neovessel growth over a large scale (most likely from nonvascular point sources), deformation of nascent fibrils extending from the growing neovessel tip may act to track neovessels to each other in the final stages of interconnection. In this scenario, the fibril “fan” of one neovessel might overlap with that of another, thereby creating a shared fibril track that both neovessels can follow necessarily “meeting in the middle” (Figure8). An implication is that larger zones of fibril alignment in front of neovessels, perhaps due to greater matrix deformability, would enable more zones to overlap thereby promoting more neovessel connections. Indeed, in our angiogenesis experiments, less stiff, more deformable collagen gels do indeed contain more interconnected (measured as branch points) neovessels.

alignment in front of neovessels, perhaps due to greater matrix deformability, would enable more zones to overlap thereby promoting more neovessel connections. Indeed, in our angiogenesis experiments, less stiff, more deformable collagen gels do indeed contain more interconnected (measured as branch points) neovessels. Figure 8 Diagrams of different means by which growing neovessels might locate each other. (A) An extravascular source of diffusible signal (graded pink) causes two neovessels (blue) to grow up the signal gradient incidentally approaching each other. (B) The tips of each growing neovessel produce a diffusible signal that forms a gradient the other neovessel recognizes and grows into. However, for this to occur, the signal from each neovessel would need to be a different molecule. (C) The “fan” of aligned matrix fibrils (red lines) that forms in front of each growing neovessel would act to “track” the neovessels toward each other when overlapping. (D) Consequences to neovessel location due to differences in the size of the fibril–alignment zone. A small fibril–alignment zone (due to stiffer matrix) would not readily overlap, while a larger zone (due to a less stiff matrix) would make overlapping of these tracks between neighboring neovessels more likely.

apping. (D) Consequences to neovessel location due to differences in the size of the fibril–alignment zone. A small fibril–alignment zone (due to stiffer matrix) would not readily overlap, while a larger zone (due to a less stiff matrix) would make overlapping of these tracks between neighboring neovessels more likely. In the adult, the growing neovessel in angiogenesis navigates through the complex tissue space moving through the stroma between parenchymal compartments. Clearly, there is a large spectrum of diffusible and matrix-bound molecular signals critical to neovessel behavior. In addition, much is known concerning the intra- and extracellular force dynamics underlying endothelial cell interactions with the matrix related to angiogenic neovessel activity. Adding to this broad understanding, our recent body of evidence indicates that growing neovessels sense strain fields and not stress fields within the 3-D stromal space, the responses of which also influence microvascular network topology. Furthermore, the interplay between the growing neovessel and the physical behavior of the surrounding may contribute to neovessel guidance and growth direction. How these different molecular and physical stimuli are integrated to produce an effective microcirculation remains to be understood.

vascular network topology. Furthermore, the interplay between the growing neovessel and the physical behavior of the surrounding may contribute to neovessel guidance and growth direction. How these different molecular and physical stimuli are integrated to produce an effective microcirculation remains to be understood. PERSPECTIVE There is a dynamic interplay between growing neovessels and the surrounding stroma during angiogenesis in which growing neovessels actively remodel the matrix which in turn passively influences neovascular topology. The orientation and extent of neovessel growth is sensitive to the compressive strain or deformation within the stroma, which is influenced by the neovessels as they pull and remodel the matrix fibril structure. Differences in matrix compliance (i.e., stiffness) influences the extent of stromal deformation and therefore angiogenic neovessel organization. Thus, this back and forth dynamic between neovessels and stroma matrix may explain the different microcirculatory architectures present within the many different tissues of the body, and may be manipulated in the regeneration of vascularized tissues. In addition to the numerous molecular and cellular elements of the stroma influencing neovessel activity, it is becoming clear that the integrated physical aspects of the stromal matrix also strongly influence the growing neovessels and may act to guide neovessel growth. Supported by NIH grants EB007556 and GM103507 (JBH), HL077683 (JBH, UU, JAW) and GM083925 (JAW). MMPmatrix metalloproteinase SHGsecond-harmonic generation 3Dthree-dimensional

Abbreviations Abantibody BBBblood‐brain barrier BM‐EPCbone marrow‐derived endothelial precursor cells BM‐MSCbone marrow‐derived mesenchymal stem cells CFUcolony‐forming units CNScentral nervous system ELISAenzyme‐linked immunosorbent assay NK‐1Rneurokinin‐1 receptor PDGFplatelet‐derived growth factor RPreceptor antagonist SPsubstance P VEGFvascular endothelial growth factor α‐SMAα‐smooth muscle actin 1 INTRODUCTION Brain injury, such as in the case of ischemic stroke, is accompanied by BBB damage, leading to increased vascular permeability, edema, and inflammatory cells infiltration, leading to a cascade of inflammatory secondary tissue damage.1 Many stem cell therapies using either MSCs or EPCs have been attempted in animal stroke models as well as in clinical trials to restore a functional BBB and interrupt damage in the early stages of the stroke, but no successful outcome has been reported.2 Considering that the BBB is mainly composed of inner tubular endothelial cells, outward‐encircling pericytes, and astrocyte end feet, more than one cell type may be required as reparative stem or precursor cells for the repair of the BBB following a stroke.

s of the stroke, but no successful outcome has been reported.2 Considering that the BBB is mainly composed of inner tubular endothelial cells, outward‐encircling pericytes, and astrocyte end feet, more than one cell type may be required as reparative stem or precursor cells for the repair of the BBB following a stroke. In the BBB of the CNS, the inner endothelial tube is tightly encircled by pericytes, which share a basement membrane with endothelial cells and form direct synaptic‐like peg‐socket focal contacts between two cell types through N‐cadherin and connexins.3, 4 Thereby, pericytes maintain endothelial tight junctions and regulate the capillary diameter and cerebrovascular flow under physiological conditions. Notably, close cooperation between brain endothelial cells and pericytes in the BBB has been reported in platelet‐derived growth factor (PDGF)‐B or PDGFR beta null mice, which results in a complete loss of pericytes, causing the rupture of CNS microvessels, micro‐aneurysms, and embryonic lethality.5, 6 Furthermore, after brain injury, pericytes seem to be more vulnerable than any other cell type in the BBB.6, 7, 8, 9 A reparative stem cell therapy targeting the replacement and recruitment of both pericytes and endothelial cells is in high demand.

ture of CNS microvessels, micro‐aneurysms, and embryonic lethality.5, 6 Furthermore, after brain injury, pericytes seem to be more vulnerable than any other cell type in the BBB.6, 7, 8, 9 A reparative stem cell therapy targeting the replacement and recruitment of both pericytes and endothelial cells is in high demand. Bone marrow is a reservoir of autologous stem cells such as MSCs, EPCs, and hematopoietic stem cells, which can be harvested from bone marrow aspirates and peripheral blood by the specific stem cell mobilizer.10 EPCs have been utilized for neovascularization after transplantation post‐CNS insult and myocardial infarction.11, 12 BM‐MSCs have been used in the treatment of a variety of tissue injuries and diseases because of their capacity to regulate severe immune and inflammatory responses by secreting protective trophic factors.13, 14 It is also expected that BM‐MSCs may play a similar role as pericytes in the blood vessel, based on their similar marker expression patterns and phenotype.15, 16 However, dual cell therapy with both BM‐EPCs and BM‐MSCs aimed toward tight blood vessel formation and investigation into their cooperated roles has not been widely studied.

cted that BM‐MSCs may play a similar role as pericytes in the blood vessel, based on their similar marker expression patterns and phenotype.15, 16 However, dual cell therapy with both BM‐EPCs and BM‐MSCs aimed toward tight blood vessel formation and investigation into their cooperated roles has not been widely studied. In this study, we first evaluated whether EPCs and MSCs derived from the bone marrow can concertedly work in the reconstruction of a pericyte‐covered vascular network, using a Matrigel tubular network assay. Because we are targeting the BBB, which is regarded as the most pericyte‐dense area in the body with an endothelial‐pericyte ratio between 1:1 and 3:1, in this study, BM‐EPCs and BM‐MSCs were applied at a ratio of 2:1 by considering of the physiological condition as well as the capability of BM‐MSCs on immune modulation.3 We also evaluated whether the SP/NK‐1R signaling pathway, known to be expressed and to activate cell proliferation and migration in both cell types,17, 18 is involved in the pericyte coverage of the endothelia. Finally, the important regulators of the endothelial tubular network as well as pericyte‐like coverage of the endothelia were investigated. Using this in vitro coculture system, the feasibility of a dual stem cell therapy in BBB reconstruction after transplantation following ischemic stroke can be pretested.

ndothelia. Finally, the important regulators of the endothelial tubular network as well as pericyte‐like coverage of the endothelia were investigated. Using this in vitro coculture system, the feasibility of a dual stem cell therapy in BBB reconstruction after transplantation following ischemic stroke can be pretested. 2 MATERIALS AND METHODS 2.1 Primary culture and identification of rat BM‐EPCs and BM‐MSCs BM‐EPCs and BM‐MSCs were obtained from the femurs of adult LEW/Crlj rats (Orientbio, Sungnam, Korea), as described previously, with modifications.19 Briefly, the single cells were washed several times and cultured in MSC growth medium (Lonza Inc., Walkersville, MD, USA) for BM‐MSC culture, or in endothelial growth medium‐2MV (Lonza Inc.) on a dish precoated with 1 μg/mL rat fibronectin (Sigma‐Aldrich, St. Louis, MO, USA) for BM‐EPC culture. After 24 hours of culture at 37°C in a humidified atmosphere of 5% CO₂, the unattached cells were removed. The medium was changed every 3 days, and the cells were subcultured until they reached 80% confluency, by seeding at a density of 3 × 105 cells per 100 mm2 on culture dishes. For MSC and EPC identification, marker expression and CFUs were assessed as described previously.10 Briefly, MSCs were identified by CD29+, CD106+, NK‐1R+, α‐SMA+, CD45−, and CD34− markers using FACS analysis and immunofluorescence staining, and EPCs were identified by CD34+, CD31+, UEA lectin binding+, and α‐SMA− markers. Approximately 90% of the cells satisfied these specific marker expression profiles. Cells from passage 2 to passage 4 were used in this study (Figures S1 and S2).

5−, and CD34− markers using FACS analysis and immunofluorescence staining, and EPCs were identified by CD34+, CD31+, UEA lectin binding+, and α‐SMA− markers. Approximately 90% of the cells satisfied these specific marker expression profiles. Cells from passage 2 to passage 4 were used in this study (Figures S1 and S2). 2.2 Matrigel tube formation A Matrigel tube formation assay was performed for the in vitro angiogenesis assessment.20 A 48‐well plate was treated with 150 μL of Matrigel (BD Biosciences, San Jose, CA, USA) for 30 minutes at 37°C. For optimal simulation of the cell proportions in the BBB in vivo, BM‐EPCs and BM‐MSCs were plated at a ratio of 2:1 with a density of 2 × 10⁴ and 1 × 10⁴ cells per well, respectively. To track the behavior of the inoculated cells, BM‐EPCs and BM‐MSCs were labeled with PKH26 red and PKH26 green (Sigma‐Aldrich), respectively. To analyze the effect of SP and PDGF‐BB, cells were pretreated with SP (Merck Millipore, Darmstadt, Germany) or PDGF‐BB functional blocking Ab (Abcam, Cambridge, MA, USA) at concentrations of 100 nmol/L and 20 μg/mL for 3 hours before Matrigel assay or concurrently treated at the time of coculture. The cells and Matrigel were fixed with 3.7% formaldehyde, and the tube length, bifurcation points, and numbers of the tube‐incorporated BM‐MSCs were measured under a stereomicroscope (Olympus, Tokyo, Japan) or fluorescence microscope (Leica, Wetzlar, Germany). All the experiments were performed as triplicate independent experiments, and the statistics were analyzed using NIH image J software.21

, bifurcation points, and numbers of the tube‐incorporated BM‐MSCs were measured under a stereomicroscope (Olympus, Tokyo, Japan) or fluorescence microscope (Leica, Wetzlar, Germany). All the experiments were performed as triplicate independent experiments, and the statistics were analyzed using NIH image J software.21 2.3 Enzyme‐linked immunosorbent assay (ELISA) The culture media were collected from equal cell quantities of BM‐MSCs, BM‐EPCs, or BM‐EPC/BM‐MSC cocultures at a ratio of 2:1. VEGF and PDGF‐BB levels were measured using VEGF and PDGF‐BB ELISA kits (R&D system, Inc., Minneapolis, MN, USA). Data from three independent experiments were statistically analyzed. 2.4 RT‐PCR analysis Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's protocol. The template cDNAs were synthesized using a Prime‐script RT‐PCR kit (Takara Bio. Inc., Shiga, Japan) with 1 μg of total RNA and the following primer sequences for RT‐PCR: rat VEGF (sense: TGT ACA AGT GCC AGC TAA GGA A; antisense: CAC ACG TAG TTT GCT GGA CAA G), rat VEGFR (sense: CCT TAG ATA GCC CGG AAC GC; antisense: GCC ACA CTC AGT CAC CAA CA, USA), and rat β‐actin (sense: CTT CTG CAT CCT GTC AGC GAT GC; antisense: AGA AGA GCT ATGA GCT GCC TGA CG).

ollowing primer sequences for RT‐PCR: rat VEGF (sense: TGT ACA AGT GCC AGC TAA GGA A; antisense: CAC ACG TAG TTT GCT GGA CAA G), rat VEGFR (sense: CCT TAG ATA GCC CGG AAC GC; antisense: GCC ACA CTC AGT CAC CAA CA, USA), and rat β‐actin (sense: CTT CTG CAT CCT GTC AGC GAT GC; antisense: AGA AGA GCT ATGA GCT GCC TGA CG). 2.5 Western blotting The cell lysates were prepared using a Cell Lysis Kit (Cell Signaling, Danvers, Massachusetts, USA) according to the manufacturer's instructions, subjected to separation by 10% SDS‐PAGE, and transferred to nitrocellulose membranes. The blots were blocked with 5% skim milk in PBS. Western blots were then probed with antibodies recognizing N‐cadherin (1:300; Abcam). Signals were detected using a Chemilluminator (Vilber Lourmat, Marne, France). 3 RESULTS 3.1 BM‐EPCs exhibit inherent endothelial tube‐forming capacity on the Matrigel but BM‐MSCs do not BM‐EPCs and BM‐MSCs are expected to be candidate stem and precursor cells for dual cell therapeutics for the treatment of various types of vascular damage. To test this in vitro, BM‐EPCs or BM‐MSCs were plated on Matrigel, and their tubular formations were examined (Figure 1). BM‐MSCs did not form a tubular network after 3‐hour incubation (Figure 1A) and rather formed pellet‐like structures after longer incubation times (data not shown). However, BM‐EPCs showed a rather elongated morphology at 1 hour on the Matrigel plate, tubule‐like connections at 1.5 hours, and a mature tubular network at 3 hours (Figure 1A).

a tubular network after 3‐hour incubation (Figure 1A) and rather formed pellet‐like structures after longer incubation times (data not shown). However, BM‐EPCs showed a rather elongated morphology at 1 hour on the Matrigel plate, tubule‐like connections at 1.5 hours, and a mature tubular network at 3 hours (Figure 1A). Figure 1 Tubular network‐forming capacity of BM‐MSCs and BM‐EPCs with or without SP treatment. (A) BM‐MSCs or BM‐EPCs were seeded at 2 × 104 cells/well on Matrigel‐coated wells in a 48‐well plate with or without 100 nmol/L SP. Phase‐contrast images were taken at indicated time points under a stereomicroscope. Scale bar = 50 μm. (B) The quantification of tube length in the vascular tube structure was measured using Image J. Data are presented as mean ± SEM (**P < .005, ***P < .0005, student's t test) Both BM‐EPCs and BM‐MSCs express SP receptor NK‐1R,22 and SP is known to mobilize BM‐MSCs from the bone marrow into the blood and aid in their recruitment to the injured tissue.19 We explored whether SP treatment can affect the tube‐forming capacity of the MSCs and/or EPCs. The SP treatment did not affect MSC behavior but resulted in an approximately 27% increase in EPC tube‐forming capacity (Figure 1A,B).

from the bone marrow into the blood and aid in their recruitment to the injured tissue.19 We explored whether SP treatment can affect the tube‐forming capacity of the MSCs and/or EPCs. The SP treatment did not affect MSC behavior but resulted in an approximately 27% increase in EPC tube‐forming capacity (Figure 1A,B). 3.2 BM‐MSCs in coculture with BM‐EPCs stimulate recruitment to the tubular network formed by EPCs, which is further enhanced by SP treatment Only BM‐EPCs showed an inherent capacity to reform a tubular network on Matrigel. However, the coculture of BM‐MSCs and BM‐EPCs may stimulate the incorporation of BM‐MSCs onto the outer surface of endothelial tubules as pericyte‐like cells. As this occurs in vitro, a similar outcome may be expected after cell transplantation to the vascular injury. When BM‐MSCs and BM‐EPCs were cocultured on Matrigel after the specific fluorescence labeling, PKH‐red‐labeled BM‐MSCs were no longer randomly distributed but were recruited to the PKH‐green‐labeled endothelial tubular network (Figure 2A,B). Upon SP cotreatment, branching points increased by 46% (Figure 2D), the tubule length of the endothelial tubular network increased by 30.6% (Figure 2E), and tubular recruitment of BM‐MSCs increased by 39% (Figure 2C‐F), which also resulted in a more elongated morphology of cells lining the surface of the tubules (Figure 2B). However, this tubular coverage with BM‐MSCs under the coculture decomposed following longer incubation up to 18 hours (Figure 2G). In contrast, in the SP‐treated coculture, the tubular coverage with BM‐MSCs was well maintained and more tightly incorporated for a longer period. Thus, SP treatment may stabilize and reinforce BM‐MSC incorporation into the endothelial tubular network.

decomposed following longer incubation up to 18 hours (Figure 2G). In contrast, in the SP‐treated coculture, the tubular coverage with BM‐MSCs was well maintained and more tightly incorporated for a longer period. Thus, SP treatment may stabilize and reinforce BM‐MSC incorporation into the endothelial tubular network. Figure 2 Effect of BM‐MSC and BM‐EPC coculture with or without SP treatment on the pericyte‐like incorporation and tubular network. (A) BM‐EPCs and BM‐MSCs at a density of 3 × 104 cells/well at a ratio of 2:1 were plated for the coculture on Matrigel. At 3 h, a tubular structure was clearly observed by stereomicroscope. The coculture resulted in most of the BM‐MSCs being incorporated into the tubular network. SP treatment resulted in a longer and better organized tube. (B‐G) The cells were labeled with PKH‐red for BM‐MSCs and PKH‐green for BM‐EPCs to discriminate between the two cell types. The tubular structure was imaged by a fluorescence microscope at 3 h (B‐F) and 18 h (G) postseeding. (C) The solid round circle and dash indicate branch points and tube length, respectively, for quantification analysis. The branch points (D), tube length (E), and tubular‐associated PKH‐red BM‐MSCs (F) are presented as mean ± SEM (**P < .005, student's t test). Scale bar = 50 μm

e at 3 h (B‐F) and 18 h (G) postseeding. (C) The solid round circle and dash indicate branch points and tube length, respectively, for quantification analysis. The branch points (D), tube length (E), and tubular‐associated PKH‐red BM‐MSCs (F) are presented as mean ± SEM (**P < .005, student's t test). Scale bar = 50 μm 3.3 BM‐MSCs and BM‐EPCs in coculture synergistically stimulate expression of VEGF, VEGF receptor, PDGF‐BB, and N‐cadherin BM‐MSC and BM‐EPC cocultures with SP exhibited enhanced tubular network formation as well as pericyte‐like coverage of the tubules in Matrigel. To investigate the underlying mechanism of this synergistic effect, the expression of VEGF, VEGF receptor, PDGF‐BB, and N‐cadherin, all of which are important for angiogenesis and pericyte coverage on the endothelia,5 was examined. Importantly, the coculture itself exhibited increased VEGF secretion in the medium, which was further enhanced by SP treatment (Figure 3A). This synergistic VEGF upregulation was sustained up to 3 days in the coculture, based on RT‐PCR analysis (Figure 3B‐D). Furthermore, VEGF receptor expression was also increased in the coculture (Figure 3B,D).

tself exhibited increased VEGF secretion in the medium, which was further enhanced by SP treatment (Figure 3A). This synergistic VEGF upregulation was sustained up to 3 days in the coculture, based on RT‐PCR analysis (Figure 3B‐D). Furthermore, VEGF receptor expression was also increased in the coculture (Figure 3B,D). Figure 3 Analysis of VEGF, VEGF receptor, PDGF‐BB, and N‐cadherin. (A) ELISA of VEGF. The cells were plated at a density of 3 × 105 cells/well for individual cell culture but were plated at a ratio of 2:1 in the coculture. Cells were treated with 100 nmol/L SP at the time of cell plating. VEGF was measured from the culture at day 1. Data represent the mean ± SEM (*P < .05, **P < .005, student's t test) (B) RT‐PCR analysis of VEGF and VEGFR at day 3 culture. The quantitative analysis of VEGF (C) and VEGFR (D) expressions are presented as the mean ± SEM (*P < .05, **P < .005, student's t test). (E) ELISA of PDGF‐BB at day 1 culture. PDGF‐BB expression is presented as the mean ± SEM (**P < .005, student's t test) (F,G) Western blot analysis of N‐cadherin expression at day 3 coculture with or without SP treatment. The intensity of α‐tubulin was used as an internal control. Densitometry of N‐cadherin band is presented as the mean ± SEM (*P < .05, student's t test)

sion is presented as the mean ± SEM (**P < .005, student's t test) (F,G) Western blot analysis of N‐cadherin expression at day 3 coculture with or without SP treatment. The intensity of α‐tubulin was used as an internal control. Densitometry of N‐cadherin band is presented as the mean ± SEM (*P < .05, student's t test) PDGF‐BB is one of the important factors involved in pericyte recruitment and attachment on the endothelia, possibly through stimulation of N‐cadherin‐mediated homophilic interactions. PDGF‐BB secretion was also increased in the coculture and was further stimulated by SP treatment (Figure 3E). Similarly, N‐cadherin expression in the coculture was also increased by 81.5% following SP treatment (Figure 3F,G). Thus, enhanced expression of VEGF and/or PDGF‐BB in the coculture and their further increase by SP treatment may be responsible for the synergistic effect of the BM‐EPCs and BM‐MSCs on the vascular tubular network.

N‐cadherin expression in the coculture was also increased by 81.5% following SP treatment (Figure 3F,G). Thus, enhanced expression of VEGF and/or PDGF‐BB in the coculture and their further increase by SP treatment may be responsible for the synergistic effect of the BM‐EPCs and BM‐MSCs on the vascular tubular network. 3.4 PDGF‐BB blocking reduces BM‐MSC coverage of the endothelial tubules To further confirm the role of PDGF‐BB in the pericyte‐like cell recruitment of BM‐MSCs on the endothelia, PDGF‐BB expression was examined in the coculture on Matrigel (Figure 4A). The coculture strongly induced PDGF‐BB secretion at 3 hours after cell plating, which was further stimulated by the SP treatment. However, NK‐1 receptor blocker reduced but did not eliminate the expression of PDGF‐BB under coculture conditions, suggesting that SP‐NK‐1R may not be the only signaling pathway involved. To test the importance of PDGF‐BB in the BM‐MSC recruitment and tubular network, functional blocking antibodies against PDGF‐BB were applied in the coculture on Matrigel (Figure 4B,C). Blocking PDGF‐BB significantly reduced the vascular tubular network as well as the endothelia‐recruited BM‐MSCs by 59.03 ± 1.718% and 54.7 ± 4.39%, respectively, in the presence and absence of SP (Figure 4C).

etwork, functional blocking antibodies against PDGF‐BB were applied in the coculture on Matrigel (Figure 4B,C). Blocking PDGF‐BB significantly reduced the vascular tubular network as well as the endothelia‐recruited BM‐MSCs by 59.03 ± 1.718% and 54.7 ± 4.39%, respectively, in the presence and absence of SP (Figure 4C). Figure 4 Inhibition of SP‐mediated PDGF‐BB induction by its receptor blockade and inhibition of tubular incorporation of BM‐MSCs by PDGF‐BB functional blocking antibodies. (A) ELISA analysis of PDGF‐BB in the coculture on Matrigel at 3 h. SP‐stimulated PDGF‐BB secretion was blocked by NK‐1 RP. Data are presented as the mean ± SEM. (B,C) The cocultures of PKH‐red‐labeled BM‐MSCs and PKH‐green‐labeled BM‐EPCs were treated with SP and/or PDGF‐BB functional blocking Ab. At 3 h, the tubular network and tubular incorporation of BM‐MSCs were strongly diminished. The quantification analyses are presented as the mean ± SEM (*P < .05, ***P < .0005, student's t test). (D,E) Preferential tubular incorporation of SP‐pretreated BM‐MSCs. PKH‐green‐labeled BM‐EPCs were seeded on Matrigel at a density of 2 × 104 at 3 h in advance to reconstruct an endothelial tubular network. To test the effect of SP on BM‐MSC tubular incorporation capacity, PKH‐red‐labeled BM‐MSCs at a density of 2 × 103 cells/well with or without SP pretreatment for 3 h were plated. The quantitative analysis of tubular‐incorporated BM‐MSCs is presented as the mean ± SEM (***P < .0005, student's t test). Scale bar = 50 μm

To test the effect of SP on BM‐MSC tubular incorporation capacity, PKH‐red‐labeled BM‐MSCs at a density of 2 × 103 cells/well with or without SP pretreatment for 3 h were plated. The quantitative analysis of tubular‐incorporated BM‐MSCs is presented as the mean ± SEM (***P < .0005, student's t test). Scale bar = 50 μm Under ischemic conditions, such as a stroke, pericytes have been shown to be more vulnerable than endothelial cells. Early loss of pericytes leads to hyperpermeability, which results in tissue edema and inflammatory cell infiltration.23 To test whether BM‐MSCs can be incorporated into the preformed vessels lacking in pericytes, PKH‐green‐labeled EPCs alone were seeded on Matrigel to form a tubular network 3 hours in advance, and PKH‐red‐labeled MSCs with or without SP pretreatment were inoculated to examine their endothelial recruitment (Figure 4D,E). Recruitment of BM‐MSCs on the preformed endothelia tubule was significantly enhanced by SP pretreatment of the BM‐MSCs (by approximately 60%), and the cell morphology was much more elongated along the tubular network.

ith or without SP pretreatment were inoculated to examine their endothelial recruitment (Figure 4D,E). Recruitment of BM‐MSCs on the preformed endothelia tubule was significantly enhanced by SP pretreatment of the BM‐MSCs (by approximately 60%), and the cell morphology was much more elongated along the tubular network. 4 DISCUSSION Stroke is a common type of vascular occlusion or hemorrhage rupture in the brain. Currently, early resolution of the blood clots within 6 hours after a stroke is the only clinically approved treatment, and a very small portion of patients qualify for such treatment.24 Despite decades of research on numerous animals and clinical trials, the outcomes of stem cell therapies have not reached clinical expectations.19 Severe tissue‐destructive inflammatory microenvironments and a very low rate of homing and survival of the transplanted cells have been suggested as reasons contributing to their therapeutic inefficiency.19 However, using a proper combination of multiple reparative stem cells instead of just one specific cell type may be a promising method for BBB repair. In the present study, we have demonstrated in vitro that a combination of bone marrow‐derived EPCs and MSCs can be used to reconstruct the vascular network de novo synergistically by acting as endothelial cells and pericyte‐like cells, respectively. This effect can be further enhanced by SP cotreatment. Furthermore, SP‐treated BM‐MSCs were preferentially incorporated into preexisting BM‐EPC tubular structures, which mimic a capillary with pericyte loss in a micro‐aneurysm and ischemic brain injury.

acting as endothelial cells and pericyte‐like cells, respectively. This effect can be further enhanced by SP cotreatment. Furthermore, SP‐treated BM‐MSCs were preferentially incorporated into preexisting BM‐EPC tubular structures, which mimic a capillary with pericyte loss in a micro‐aneurysm and ischemic brain injury. The characteristics defining EPCs based on origin and markers expression have been debated for years.19 It was previously claimed that EPCs originate in the peripheral blood rather than the bone marrow.25 This study has revealed that BM‐EPCs are able to form an endothelial tubular network within 3 hours autonomously, which is the function of reparative EPCs in vitro. On the other hand, BM‐MSCs themselves remain in a pellet‐like morphology even after a long incubation. However, in cooperation with BM‐EPCs, BM‐MSCs can be incorporated into the EPC‐made endothelia, and this synergy can be further enhanced by SP cotreatment. A possible mechanism underlying the concerted cooperation between BM‐EPCs and BM‐MSCs in the vascular network formation could be the synergistic induction of VEGF and its receptor, PDGF‐BB, and N‐cadherin under coculture conditions.

to the EPC‐made endothelia, and this synergy can be further enhanced by SP cotreatment. A possible mechanism underlying the concerted cooperation between BM‐EPCs and BM‐MSCs in the vascular network formation could be the synergistic induction of VEGF and its receptor, PDGF‐BB, and N‐cadherin under coculture conditions. PDGF‐BB has been reported to play a critical role in the pericyte recruitment process, which leads to vessel maturation and stabilization in the developmental phase as well as tissue repair in adults.5 Our data in this study clearly support that both BM‐EPCs and BM‐MSCs synergistically act on the PDGF‐BB induction as early as 3 hours in coculture and may then induce N‐cadherin expression, which is later engaged in homotypic intercellular association between two cells. This initial event may also reinforce the tight junction formation between endothelial cells as well as further coverage of pericytes along the basal membrane of the endothelial cells. This is supported by the finding that the functional blockade of PDGF‐BB led to nearly complete inhibition of recruitment of BM‐MSCs to the endothelial tubules and partial disruption of the endothelial tube structure of BM‐EPCs. Accordingly, PDGF‐BB seems to be critical for the maintenance of both the endothelial tube itself and pericyte‐like encircling of BM‐MSCs on the endothelia.

PDGF‐BB led to nearly complete inhibition of recruitment of BM‐MSCs to the endothelial tubules and partial disruption of the endothelial tube structure of BM‐EPCs. Accordingly, PDGF‐BB seems to be critical for the maintenance of both the endothelial tube itself and pericyte‐like encircling of BM‐MSCs on the endothelia. It was reported by our group that neuropeptide SP works as a stem cell mobilizer for BM‐MSCs and then promotes tissue repair by recruiting mobilized cells to the injured tissue.10 In addition, SP has also been reported to modulate the immune environment by increasing the anti‐inflammatory M2 type macrophage after spinal cord injury, which leads to less apoptosis of neighboring cells and better functional recovery.26, 27 Taken together with the role of SP in the suppression of the injury‐mediated inflammatory response and as a BM‐MSC mobilizer, our finding that SP stimulated PDGF‐BB induction and more stable interaction of BM‐MSCs on the endothelial network provides a rationale to develop SP as a supplementary adjuvant to improve the therapeutic outcomes of dual stem cell therapies by enhancing angiogenesis and pericyte recruitment to make mature, tight BBB vessels in ischemic vascular damage repair.

uction and more stable interaction of BM‐MSCs on the endothelial network provides a rationale to develop SP as a supplementary adjuvant to improve the therapeutic outcomes of dual stem cell therapies by enhancing angiogenesis and pericyte recruitment to make mature, tight BBB vessels in ischemic vascular damage repair. 5 PERSPECTIVE This study proposes the rationale and working mechanism of dual cell therapy with BM‐EPCs and BM‐MSCs in conjunction with SP supplementation for the repair of the BBB following ischemic stroke in vivo, possibly by stimulating secretion of the angiogenic factors VEGF and PDGF‐BB and N‐cadherin expression, culminating in pericyte‐like coverage of BM‐MSCs on BM‐EPC‐regenerated endothelial tubules. Supporting information Click here for additional data file. Click here for additional data file. ACKNOWLEDGMENTS This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea Ministry of Science, ICT & Future Planning (2012M3A9C6050499, 2016M3A9B4917320).

Abbreviations ACEangiotensin‐converting enzyme Baselinebaseline capillary density BMIbody mass index cMBcerebral microbleeds CRcapillary recruitment cSVDcerebral small vessel disease DCCTDiabetes Control and Complications Trial EDICEpidemiology of Diabetes Interventions and Complications FLAIRfluid‐attenuated inversion recovery HbA1cglycated hemoglobin IQintelligence quotient MRImagnetic resonance imaging Peakpeak hyperemia capillary density RCIreliable change index SDstandard deviation T1DMtype 1 diabetes VOvenous occlusion capillary density WMLwhite matter lesions 1 INTRODUCTION Type 1 diabetes is a complex endocrine disease, which, in case of poor glycemic control, can lead to long‐term microvascular complications, affecting kidneys, peripheral and central nervous system, and eyes.1 The brain has also been recognized as a key organ that can be affected by long‐standing type 1 diabetes.2 Cross‐sectional and longitudinal studies have shown that type 1 diabetes is associated with mild‐to‐moderate impairments in multiple neurocognitive domains compared with healthy controls.3, 4

ous system, and eyes.1 The brain has also been recognized as a key organ that can be affected by long‐standing type 1 diabetes.2 Cross‐sectional and longitudinal studies have shown that type 1 diabetes is associated with mild‐to‐moderate impairments in multiple neurocognitive domains compared with healthy controls.3, 4 The pathophysiology underlying these cognitive decrements in type 1 diabetes has not yet been fully elucidated. Risk factors for cognitive impairment include longer duration and early onset of diabetes as well as poor glycemic control, which suggests that hyperglycemia‐induced damage is a pathophysiological determinant.5, 6 Additionally, the presence of diabetic microangiopathy—that is, retinopathy, nephropathy, and neuropathy—was cross‐sectionally associated with lower performance on several neuropsychological tests 3, 7, 8 and longitudinally with mild, but clinically relevant cognitive decline independent of lifetime HbA1c in the DCCT1 and its epidemiological follow‐up, the EDIC trial.6

thy—that is, retinopathy, nephropathy, and neuropathy—was cross‐sectionally associated with lower performance on several neuropsychological tests 3, 7, 8 and longitudinally with mild, but clinically relevant cognitive decline independent of lifetime HbA1c in the DCCT1 and its epidemiological follow‐up, the EDIC trial.6 In type 1 diabetes, diabetic microangiopathy has been proposed to reflect a state of generalized microvascular dysfunction, including in the brain, which may lead to cSVD. On brain MRI, cSVD presents with ischemic WML, lacunar infarcts, and non‐lobar cMB.9, 10 The link between cSVD and peripheral microvascular complications is supported by data from our own group, which show higher prevalence of cMB in type 1 diabetes patients with proliferative retinopathy compared with those without.11 In addition, diabetic retinopathy and neuropathy have been associated with the presence, volume, and severity of WML.8, 12 We therefore hypothesize that generalized microvascular dysfunction, manifest by impaired microvascular function of the skin, and cSVD may correlate with decline in cognitive function over time. In this prospective cohort study, we aimed to identify cerebral microangiopathy and skin capillary perfusion—as a surrogate marker of generalized microvascular function—as independent predictors of cognitive performance, controlling for possible confounding factors including HbA1c, diabetes duration, and the number of severe hypoglycemic events.

study, we aimed to identify cerebral microangiopathy and skin capillary perfusion—as a surrogate marker of generalized microvascular function—as independent predictors of cognitive performance, controlling for possible confounding factors including HbA1c, diabetes duration, and the number of severe hypoglycemic events. 2 MATERIALS AND METHODS This prospective cohort study was part of a larger cross‐sectional study conducted at the VU University Medical Centre,13 in which we assessed the effects of type 1 diabetes and concomitant proliferative retinopathy on cognition, brain volume, and functional and structural connectivity. A randomly selected subgroup of 25 type 1 diabetes patients with proliferative retinopathy and 25 healthy controls—matched for age, gender, BMI, and IQ—were invited to return for follow‐up measurements between May 2007 and September 2009. Inclusion criteria at baseline were as follows: age 18‐56 years, right‐handedness, and, for the type 1 diabetes patients, a diabetes duration of at least 10 years. Exclusion criteria at both time points were psychiatric comorbidity, insufficient visual acuity (below 0.3) to perform neuropsychological tests, brain trauma, previous coma unrelated to hypoglycemia, alcohol consumption (men: >21 units a week; women: >14 units a week) and drug use, use of centrally acting medication, MRI contraindications, and, for the healthy controls, hypertension at baseline. At baseline and follow‐up, all participants were subjected to evaluation of cognitive function and brain MRI. Skin microvascular function was assessed at baseline by means of nail fold capillary microscopy. To be able to perform the neuropsychological tests, all participants were required to have normal or corrected to normal vision. Additionally, none of the patients had central proliferative retinopathy, thus not affecting the central focus area of vision. This study was conducted in accordance with the Declaration of Helsinki 201314 and approved by the medical ethics committee of the VU University Medical Centre. All participants signed written informed consent at baseline and during follow‐up.

liferative retinopathy, thus not affecting the central focus area of vision. This study was conducted in accordance with the Declaration of Helsinki 201314 and approved by the medical ethics committee of the VU University Medical Centre. All participants signed written informed consent at baseline and during follow‐up. 2.1 Justification of sample size and group selection This prospective cohort study is a randomly selected subsample of a larger cross‐sectional study, in which type 1 diabetes patients with and without retinopathy and healthy controls were included. Due to funding limitations, we were able to include 50 of the 153 participants that were included at baseline. We chose to include two groups of 25 participants, instead of three groups of approximately 16 participants to increase the power of the study. The groups were selected based on the largest expected differences, taking into consideration that these results cannot be translated to type 1 diabetes patients in general.

e. We chose to include two groups of 25 participants, instead of three groups of approximately 16 participants to increase the power of the study. The groups were selected based on the largest expected differences, taking into consideration that these results cannot be translated to type 1 diabetes patients in general. 2.2 Cognitive functioning All participants underwent a detailed neuropsychological assessment covering the domains of memory, information‐processing speed, executive functions, attention, motor, and psychomotor speed. The tests were described previously.15 General cognitive ability was constructed by averaging results with respect to the above‐mentioned six domains. Raw scores were transformed into z‐scores based on the mean and SD values from healthy controls of the larger cross‐sectional study and inversed if necessary so that higher z‐scores indicated better performance. To quantify change in cognitive performance over time, the RCI was computed from baseline and follow‐up data by the following formula: ((Xfollow‐up–Xbaseline)–(mean‐controls follow‐up–mean‐controls baseline))/standard deviation delta‐score controls. Thus, for each test the mean delta‐score of the controls was subtracted from the raw subject's delta‐score and divided by the standard deviation of the delta‐score of the controls.16 To rule out confounding as a result of hypo‐ or hyperglycemia, glucose levels had to be between 4 and 15 mmol/L before and during neuropsychological testing. When outside of this range, patients were instructed to eat or inject insulin and testing was postponed for 30 min until blood glucose levels were within the required range.17

onfounding as a result of hypo‐ or hyperglycemia, glucose levels had to be between 4 and 15 mmol/L before and during neuropsychological testing. When outside of this range, patients were instructed to eat or inject insulin and testing was postponed for 30 min until blood glucose levels were within the required range.17 2.3 Brain MRI MRI scanning was performed on a 1.5‐T magnetic resonance system (Siemens‐Sonata, Erlangen, Germany). Further details on the magnetic resonance acquisition protocol have been described previously.13 Vascular WML were assessed using T2 FLAIR and cerebral microbleeds using T2* susceptibility weighted imaging. All images were rated by an experienced neuroradiologist, who was blinded to any clinical information. Both WML and cMB were scored as present or not present. The Fazekas scale is often used to indicate severity of WML18; however, in this study all but one subject had a Fazekas score of 1.

ceptibility weighted imaging. All images were rated by an experienced neuroradiologist, who was blinded to any clinical information. Both WML and cMB were scored as present or not present. The Fazekas scale is often used to indicate severity of WML18; however, in this study all but one subject had a Fazekas score of 1. 2.4 Skin microvascular function Peripheral microvascular function was assessed by skin capillary microscopy as described previously.19 Briefly, nail fold capillaries in the dorsal skin of the third finger were visualized by a capillary microscope. Baseline capillary density (baseline) was defined as the number of continuously erythrocyte‐perfused capillaries per square millimeter. Capillary density during peak reactive hyperemia (peak) was counted after 4 min of arterial occlusion. Maximal capillary density was assessed during VO. CR was calculated as the absolute and relative increase in capillary density from baseline to capillary density during peak reactive hyperemia.

per square millimeter. Capillary density during peak reactive hyperemia (peak) was counted after 4 min of arterial occlusion. Maximal capillary density was assessed during VO. CR was calculated as the absolute and relative increase in capillary density from baseline to capillary density during peak reactive hyperemia. 2.5 Statistical analyses Data were analyzed with SPSS version 22.0 (IBM‐SPSS, Chicago, IL). Data are presented as mean ± SD, median[range] (not normally distributed), or raw numbers with percentages (categorical data). We assessed differences between patients and healthy controls using the unpaired t test or Mann–Whitney U test for continuous variables, depending on the data distribution, and the X² test for categorical data. Since the sample size of this study limits the number of variables allowed in our multiple regression analysis, we first performed several simple regression analysis with change in general cognition as dependent variable and the most evident confounding factors based on current literature (ie, BMI, smoking, hypertension, HbA1c, diabetes duration, and severe hypoglycemic events in the past) as independent variables.20, 21 Subsequently, multiple linear regression models were used to assess the significance of covariate‐adjusted associations between the change in general cognitive ability (dependent variable) and variables of cerebral microangiopathy and skin microvascular function (predictors/independent variables). Age and sex (model 2) and variables demonstrating associations in the simple regression analysis with a P‐value <0.1 (model 3) qualified as independent variables for inclusion into the multiple regression analyses. Next, we added capillary perfusion variables as independent variable to the regression models with WML as predictor and vice versa. Missing data were excluded pairwise. A two‐sided P‐value <0.05 was considered statistically significant. We performed a Cook's distance analysis to examine the influence of two apparent outliers based on the scatter plot. Finally, we carried out mediation analyses to examine whether the presence of WML was a mediator between skin perfusion and the change in general cognitive ability. This analysis was performed in STATA version 13SE (StatCorp LLC, College Station, TX) using the bootstrapping method according to Preacher and Hayes.22

ot. Finally, we carried out mediation analyses to examine whether the presence of WML was a mediator between skin perfusion and the change in general cognitive ability. This analysis was performed in STATA version 13SE (StatCorp LLC, College Station, TX) using the bootstrapping method according to Preacher and Hayes.22 3 RESULTS 3.1 Participants Table 1 shows the characteristics of type 1 diabetes patients and healthy controls at baseline. There was no difference from baseline to follow‐up between type 1 diabetes patients and healthy controls (3.56 ± 0.65 and 3.94 ± 0.91 years, respectively; P = 0.098). Owing to the selection method, there also were no significant differences between the two groups in age (type 1 diabetes: 46.1 ± 6.3 years; controls: 44.3 ± 8.5 years; P = 0.410), sex (type 1 diabetes: 40% male; controls: 52% male; P = 0.395), IQ (type 1 diabetes: 112.3 ± 12.7; controls: 109.4 ± 13.1; P = 0.433), and BMI (type 1 diabetes: 26.2 ± 4.9 kg/m2; controls: 25.1 ± 2.9 kg/m2; P = 0.307). In addition, we also did not detect significant differences in diastolic blood pressure (type 1 diabetes: 75.4 ± 7.8 mmHg; controls: 78.9 ± 6.2 mmHg; P = 0.088), systolic blood pressure (type 1 diabetes: 133.0 [107.0‐151.5] mmHg; controls: 128.0 [101.0‐139.0] mmHg; P = 0.137), smoking (yes) (type 1 diabetes: 3 (12%); controls: 2 (8%); P = 0.637), and cholesterol levels (type 1 diabetes: 4.4 ± 0.8 mmol/L; controls: 4.8 ± 0.9 mmol/L; P = 0.120). By definition, HbA1c was higher in type 1 diabetes patients (7.9 ± 1.0%) than in healthy controls (5.3 ± 0.3%; P < 0.001).

.0‐139.0] mmHg; P = 0.137), smoking (yes) (type 1 diabetes: 3 (12%); controls: 2 (8%); P = 0.637), and cholesterol levels (type 1 diabetes: 4.4 ± 0.8 mmol/L; controls: 4.8 ± 0.9 mmol/L; P = 0.120). By definition, HbA1c was higher in type 1 diabetes patients (7.9 ± 1.0%) than in healthy controls (5.3 ± 0.3%; P < 0.001). Table 1 Baseline characteristics Type 1 diabetes (n = 25) Controls (n = 25) P‐value Age (years) 46.1 ± 6.3 44.3 ± 8.5 0.409 Sex (m/f(%male)) 10/15 (40%) 13/12 (52%) 0.395 Estimated IQ 112.3 ± 12.7 109.4 ± 13.1 0.433 Education levela 6 [2‐8] 6 [4‐8] 0.204 BMI (kg/m²) 26.2 ± 4.9 25.1 ± 2.9 0.307 Diastolic blood pressure (mmHg) 75.4 ± 7.8 78.9 ± 6.2 0.088 Systolic blood pressure (mmHg) 133.0 [107.0‐151.5] 128.0 [101.0‐139.0] 0.137 Hypertensionb (n(%)) 11 (44%) ‐ ‐ Smoking (n(%)) 3 (12%) 2 (8%) 0.637 Total cholesterol (mmol/L) 4.4 ± 0.8 4.8 ± 0.9 0.120 HbA1c (mmol/mol) 63 ± 10.9 34 ± 3.3 <0.001 HbA1c (%) 7.9 ± 1.0 5.3 ± 0.3 <0.001 Diabetes duration (years) 34.7 ± 8.1 ‐ ‐ Diabetes early onsetc (n(%)) 8 (32%) ‐ ‐ Severe hypoglycemic eventsd 1 [0‐25] ‐ ‐ Albuminuria (n(%))e 5 (20%) ‐ ‐ Peripheral neuropathy (n(%))f 10 (40%) ‐ ‐ Data are presented as mean ± SD, median[range] or absolute number(%). BMI, body mass index; HbA1c , glycated hemoglobin; IQ, intelligence quotient; m/f, male/female. Bold values are statistically significant results (P < 0.05). a Education level was based on a Dutch scoring system ranging from 1 to 8, One indicates unfinished primary school, and 8 indicates a completed university study at master's level.

Diabetes duration (years) 34.7 ± 8.1 ‐ ‐ Diabetes early onsetc (n(%)) 8 (32%) ‐ ‐ Severe hypoglycemic eventsd 1 [0‐25] ‐ ‐ Albuminuria (n(%))e 5 (20%) ‐ ‐ Peripheral neuropathy (n(%))f 10 (40%) ‐ ‐ Data are presented as mean ± SD, median[range] or absolute number(%). BMI, body mass index; HbA1c , glycated hemoglobin; IQ, intelligence quotient; m/f, male/female. Bold values are statistically significant results (P < 0.05). a Education level was based on a Dutch scoring system ranging from 1 to 8, One indicates unfinished primary school, and 8 indicates a completed university study at master's level. b Hypertension was defined as a systolic blood pressure of 140 mmHg or above, a diastolic blood pressure of 90 mmHg or above or the use of antihypertensive drugs. c Early onset of type 1 diabetes was defined as diabetes onset before the age of seven. d Severe hypoglycemic events were self‐reported. e Albuminuria was defined as an albumin: creatinine ratio >2.5 mg/mmol for men and >3.5 mg/mmol for women and assessed with 24‐hour urine sampling. f Peripheral neuropathy was based on medical records or, in case they were not available, based on self‐report.

c Early onset of type 1 diabetes was defined as diabetes onset before the age of seven. d Severe hypoglycemic events were self‐reported. e Albuminuria was defined as an albumin: creatinine ratio >2.5 mg/mmol for men and >3.5 mg/mmol for women and assessed with 24‐hour urine sampling. f Peripheral neuropathy was based on medical records or, in case they were not available, based on self‐report. John Wiley & Sons, Ltd3.2 Neurocognitive functioning At baseline, type 1 diabetes patients had significantly lower general cognitive ability (z‐score) relative to healthy controls (type 1 diabetes: −0.3830 ± 0.45; controls: −0.0030 ± 0.36; P = 0.002), which was driven by significantly lower information‐processing speed (type 1 diabetes: −0.7622 ± 0.84; controls: −0.0002 ± 0.56; P < 0.001) and motor speed (type 1 diabetes: −0.4928 ± 0.91; controls: −0.0012 ± 0.81; P = 0.048). At follow‐up, general cognitive ability did not change significantly in both the patient and control group. The reliable change index of executive function was significantly lower in the type 1 diabetes patients (−0.337 ± 0.53) compared with the healthy controls (−0.003 ± 0.34; P = 0.010). These data are shown in Table 2 and were published previously 23). We have measured plasma glucose levels before neurocognitive testing. There were no significant differences between these measurements (baseline: 7.65 ± 3.75 versus follow‐up: 8.29 ± 3.44; P = 0.559). Table 2 Baseline and reliable change index of neurocognitive function

John Wiley & Sons, Ltd3.2 Neurocognitive functioning At baseline, type 1 diabetes patients had significantly lower general cognitive ability (z‐score) relative to healthy controls (type 1 diabetes: −0.3830 ± 0.45; controls: −0.0030 ± 0.36; P = 0.002), which was driven by significantly lower information‐processing speed (type 1 diabetes: −0.7622 ± 0.84; controls: −0.0002 ± 0.56; P < 0.001) and motor speed (type 1 diabetes: −0.4928 ± 0.91; controls: −0.0012 ± 0.81; P = 0.048). At follow‐up, general cognitive ability did not change significantly in both the patient and control group. The reliable change index of executive function was significantly lower in the type 1 diabetes patients (−0.337 ± 0.53) compared with the healthy controls (−0.003 ± 0.34; P = 0.010). These data are shown in Table 2 and were published previously 23). We have measured plasma glucose levels before neurocognitive testing. There were no significant differences between these measurements (baseline: 7.65 ± 3.75 versus follow‐up: 8.29 ± 3.44; P = 0.559). Table 2 Baseline and reliable change index of neurocognitive function Baseline (z‐scores) Reliable change index Type 1 diabetes (n = 25) Controls (n = 25) P‐value Type 1 diabetes (n = 25) Controls (n = 25) P‐value General cognitive ability −0.383 ± 0.45 −0.003 ± 0.36 0.002 0.045 ± 0.32 0.001 ± 0.19 0.550 Memory −0.353 ± 0.64 −0.000 ± 0.62 0.052 0.160 ± 0.44 0.000 ± 0.50 0.233 Information‐processing speed −0.762 ± 0.84 −0.000 ± 0.56 <0.001 0.110 ± 0.66 0.000 ± 0.52 0.516 Executive function 0.056 ± 0.57 −0.004 ± 0.45 0.684 −0.337 ± 0.53 −0.003 ± 0.34 0.010

−0.383 ± 0.45 −0.003 ± 0.36 0.002 0.045 ± 0.32 0.001 ± 0.19 0.550 Memory −0.353 ± 0.64 −0.000 ± 0.62 0.052 0.160 ± 0.44 0.000 ± 0.50 0.233 Information‐processing speed −0.762 ± 0.84 −0.000 ± 0.56 <0.001 0.110 ± 0.66 0.000 ± 0.52 0.516 Executive function 0.056 ± 0.57 −0.004 ± 0.45 0.684 −0.337 ± 0.53 −0.003 ± 0.34 0.010 Attention −0.272 ± 0.93 0.000 ± 0.75 0.267 0.024 ± 0.74 0.000 ± 0.79 0.915 Motor speed −0.493 ± 0.91 0.001 ± 0.81 0.048 0.144 ± 0.69 0.000 ± 0.61 0.439 Psychomotor speed −0.473 ± 0.80 0.000 ± 0.10 0.071 0.169 ± 0.77 0.000 ± 1.00 0.506 Data are presented as mean ± SD. Reliable change index = to quantify change in cognitive performance over time the reliable change index was computed from baseline and follow‐up data by the following formula: ((Xfollow‐up−Xbaseline)−(mean‐controls follow‐up−mean‐controls baseline))/standard deviation delta‐score controls. Bold values are statistically significant results (P < 0.05). John Wiley & Sons, Ltd3.3 White matter lesions and cerebral microbleeds At baseline, the presence of WML (type 1 diabetes: 6 (24%); controls: 6 (24%); P = 0.999) and cMB (type 1 diabetes: 5 (20%); controls: 4 (17.4%); P = 0.941) was not significantly different between type 1 diabetes patients and healthy controls in (Table 3). Table 3 Baseline parameters of cerebral microangiopathy and capillary perfusion

John Wiley & Sons, Ltd3.3 White matter lesions and cerebral microbleeds At baseline, the presence of WML (type 1 diabetes: 6 (24%); controls: 6 (24%); P = 0.999) and cMB (type 1 diabetes: 5 (20%); controls: 4 (17.4%); P = 0.941) was not significantly different between type 1 diabetes patients and healthy controls in (Table 3). Table 3 Baseline parameters of cerebral microangiopathy and capillary perfusion Cerebral microangiopathy Type 1 diabetes (n = 25) Controls (n = 25) P‐value CMB (n(%))—present yes/no 5 (20.0%) 4 (17.4%) 0.941 WML (n(%))—present yes/no 6 (24.0%) 6 (24.0%) 0.999 Skin capillary densitya Type 1 diabetes (n = 17) Controls (n = 20) Baseline perfusion (n/mm²) 45.0 ± 7.7 50.6 ± 11.1 0.087 Peak perfusion (n/mm²) 63.7 ± 13.6 70.4 ± 19.2 0.238 Venous occlusion (n/mm²) 65.2 ± 13.3 70.6 ± 18.9 0.334 Capillary recruitment (absolute values) 18.6 ± 7.9 19.7 ± 11.3 0.745 Capillary recruitment (%) 40.7 ± 16.1 38.4 ± 17.6 0.685 Data are presented as mean ± SD or number (percentage). CMB, cerebral microbleeds; CR, capillary recruitment (ie, increase of capillaries in n/mm² from baseline to peak perfusion); WML, white matter lesions. a 13 out of 50 perfusion measurements were excluded based on improper quality of the video microscopy image.

Cerebral microangiopathy Type 1 diabetes (n = 25) Controls (n = 25) P‐value CMB (n(%))—present yes/no 5 (20.0%) 4 (17.4%) 0.941 WML (n(%))—present yes/no 6 (24.0%) 6 (24.0%) 0.999 Skin capillary densitya Type 1 diabetes (n = 17) Controls (n = 20) Baseline perfusion (n/mm²) 45.0 ± 7.7 50.6 ± 11.1 0.087 Peak perfusion (n/mm²) 63.7 ± 13.6 70.4 ± 19.2 0.238 Venous occlusion (n/mm²) 65.2 ± 13.3 70.6 ± 18.9 0.334 Capillary recruitment (absolute values) 18.6 ± 7.9 19.7 ± 11.3 0.745 Capillary recruitment (%) 40.7 ± 16.1 38.4 ± 17.6 0.685 Data are presented as mean ± SD or number (percentage). CMB, cerebral microbleeds; CR, capillary recruitment (ie, increase of capillaries in n/mm² from baseline to peak perfusion); WML, white matter lesions. a 13 out of 50 perfusion measurements were excluded based on improper quality of the video microscopy image. John Wiley & Sons, Ltd3.4 Skin microvascular function At baseline, we did not detect significant differences between type 1 diabetes patients and healthy controls in skin capillary perfusion: that is, baseline (type 1 diabetes 45.0 ± 7.7; controls: 50.6 ± 11.1; P = 0.087), peak (type 1 diabetes 63.7 ± 13.6; controls: 70.4 ± 19.2; P = 0.238), and VO (type 1 diabetes 65.2 ± 13.3; controls: 70.6 ± 18.9; P = 0.334) capillary density, as well as absolute CR (type 1 diabetes 18.6 ± 7.9; controls: 19.7 ± 11.3; P = 0.745) and CR percentage (type 1 diabetes 40.7 ± 16.1; controls: 38.4 ± 17.6; P = 0.685) (Table 3).

tes 63.7 ± 13.6; controls: 70.4 ± 19.2; P = 0.238), and VO (type 1 diabetes 65.2 ± 13.3; controls: 70.6 ± 18.9; P = 0.334) capillary density, as well as absolute CR (type 1 diabetes 18.6 ± 7.9; controls: 19.7 ± 11.3; P = 0.745) and CR percentage (type 1 diabetes 40.7 ± 16.1; controls: 38.4 ± 17.6; P = 0.685) (Table 3). 3.5 Simple regression analysis A simple regression analysis with general cognitive ability (the mean of all tests) as dependent variable and BMI, smoking, hypertension, HbA1c, diabetes duration, early onset of diabetes, and severe hypoglycemic events in the past (self‐reported) as independent variables was performed to detect possible confounding factors. With a significance level of P < 0.100, HbA1c and the number of severe hypoglycemic events in the past were identified as possible confounding factors and were subsequently added to the multiple regression model (model 3).

st (self‐reported) as independent variables was performed to detect possible confounding factors. With a significance level of P < 0.100, HbA1c and the number of severe hypoglycemic events in the past were identified as possible confounding factors and were subsequently added to the multiple regression model (model 3). 3.6 Multiple regression analysis In type 1 diabetes patients, a significant association was detected between poorer general cognitive ability over time and presence of WML (ß = −0.419; P = 0.037), as well as with lower baseline capillary density (ß = 0.753; P < 0.001), peak hyperemia capillary density (ß = 0.743; P = 0.001), venous occlusion capillary density (ß = 0.675; P = 0.003), and absolute capillary recruitment (ß = 0.549; P = 0.022) at baseline (Figure 1). These associations were independent of age, sex (model 2), and HbA1c and severe hypoglycemic events (model 3) and were not present in the control population (Table S4). There was no relationship between the presence of cMB and general cognitive ability in both type 1 diabetic patients and healthy controls. The presence of WML was not associated with one of the parameters of skin capillary perfusion (baseline: ß = −0.319; P = 0.213, peak: ß = −0.168; P = 0.519, VO: ß = −0.294; P = 0.252, CR absolute: ß = 0.020; P = 0.938, and CR percentage: ß = 0.117; P = 0.656). However, the relationship between WML and general cognitive ability decreased significantly (β change > 10%) and was no longer significant when combined with baseline (β = −0.184; P = 0.320), peak (β = −0.289; P = 0.100), and VO (β = −0.227; P = 0.269) capillary density. Adding WML to the regression analysis only slightly (<10%) changed the associations between capillary perfusion and general cognitive ability (baseline: β = 0.694; P = 0.002; peak: β = 0.695; P = 0.001; VO: β = 0.608; P = 0.008, and CR (abs) β = 0.558; P = 0.012). Cook's distance analysis for detecting influential cases showed no observations above 1. Further analysis of the six independent cognitive domains demonstrated in type 1 diabetes patients and controls that the presence of WML was significantly associated with decline in motor speed (type 1 diabetes: ß = −0.399; P = 0.048; controls: ß = −0.477; P = 0.025) (Table S5) (Figure 1), independent of age, sex, HbA1c, and (for type 1 diabetes patients only) severe hypoglycemic events.

in type 1 diabetes patients and controls that the presence of WML was significantly associated with decline in motor speed (type 1 diabetes: ß = −0.399; P = 0.048; controls: ß = −0.477; P = 0.025) (Table S5) (Figure 1), independent of age, sex, HbA1c, and (for type 1 diabetes patients only) severe hypoglycemic events. Furthermore, in type 1 diabetes patients, lower baseline capillary density (ß = 0.743; P = 0.001) and lower peak capillary density (ß = 0.703; P = 0.002) were associated with decline in attention (Table S6) (Figure 1), independent of sex, age, HbA1c, and severe hypoglycemia. The significant associations between changes in attention with VO (ß = 0.670; P = 0.003) and absolute CR (ß = 0.489; P = 0.046) disappeared after correcting for age and sex (VO: ß = 0.667; P = 0.052 and absolute CR: ß = 0.316; P = 0.279) (Table S6). Next, we performed mediation analyses, with the change in general cognitive ability as outcome variable, skin perfusion parameters as independent variables, and WML as mediator, to assess whether WML (partly) mediated the effect of skin perfusion on the change in general cognitive ability. In type 1 diabetic patients, the total effect of skin perfusion (baseline: B = 0.030; P < 0.001, peak: B = 0.017; P = 0.001, VO: B = 0.015; P = 0.003) was not mediated by WML, indicated by the non‐significant indirect coefficients (baseline: B = 0.011 P = 0.564; peak: B = 0.005 P = 0.806; VO: B = 0.008 P = 0.489).

ity. In type 1 diabetic patients, the total effect of skin perfusion (baseline: B = 0.030; P < 0.001, peak: B = 0.017; P = 0.001, VO: B = 0.015; P = 0.003) was not mediated by WML, indicated by the non‐significant indirect coefficients (baseline: B = 0.011 P = 0.564; peak: B = 0.005 P = 0.806; VO: B = 0.008 P = 0.489). Figure 1 Correlations between RCI of general cognitive ability with (A) WML and nail fold capillary density during (B) baseline capillary density (baseline), (C) peak hyperemia (peak), (D) VO, and (E) absolute capillary recruitment (CR) in type 1 diabetic patients. And, correlations between RCI of motor speed with WML (F) and the RCI of attention with nail fold capillary density during baseline (G) and peak (H) in type 1 diabetic patients. †Cook's distance analysis below 1

(C) peak hyperemia (peak), (D) VO, and (E) absolute capillary recruitment (CR) in type 1 diabetic patients. And, correlations between RCI of motor speed with WML (F) and the RCI of attention with nail fold capillary density during baseline (G) and peak (H) in type 1 diabetic patients. †Cook's distance analysis below 1 4 DISCUSSION In the present longitudinal study, we aimed to identify cerebral microangiopathy and skin capillary perfusion—as a surrogate marker for generalized microvascular function—as predictors of cognitive performance. We demonstrated that, in type 1 diabetes patients with proliferative retinopathy, the presence of WML and lower skin capillary perfusion at baseline was associated with lower performance in general cognitive ability over time, independent of age, sex, HbA1c, and severe hypoglycemic events. The association between WML and lower performance in general cognitive ability is driven by lower performance in motor speed, whereas lower capillary perfusion is related to lower performance in the attention domain. In contrast, in the healthy controls these associations were not found. These data suggest a possible influence of WML and impaired microvascular function on cognitive performance in type 1 diabetes patients. In addition, we showed that the relationship between WML and lower cognitive performance was significantly reduced when capillary perfusion variables were added to the regression model, suggesting that generalized microvascular dysfunction may underlie WML‐associated lower cognitive performance.

pe 1 diabetes patients. In addition, we showed that the relationship between WML and lower cognitive performance was significantly reduced when capillary perfusion variables were added to the regression model, suggesting that generalized microvascular dysfunction may underlie WML‐associated lower cognitive performance. The etiology of cerebral WML is not yet fully understood. A multifactorial etiology is presumed, including but not limited to impairment of the blood‐brain barrier, ischemia, hypoxia, immune activation, and altered cell metabolic pathways.24 Previous cross‐sectional studies on the influence of WML on cognitive performance have shown correlations between WML severity (ie, Fazekas score) and location with (subjective) cognitive failures in non‐diabetic populations.25, 26 Furthermore, in type 1 diabetes patients, higher WML volume was associated with lower information‐processing speed.12 It has been shown previously that patients with childhood‐onset type 1 diabetes have more severe WML compared to healthy controls.12 In our study, the presence of WML was similar in type 1 diabetes patients and healthy controls at baseline. This was unexpected, however consistent with a previous study with a larger sample size.27 Study differences may be explained by variations in baseline characteristics such as age, diabetes duration, and the presence of early‐onset diabetes. Despite a similar prevalence in WML, we could only detect a relationship between the presence of WML at baseline and changes in general cognitive ability in the type 1 diabetes patients and not in the healthy controls. This suggests that WML do not have an effect on cognition in a healthy population with no other abnormalities and enough cognitive reserve capacity, whereas in a population with an underlying disease such as T1DM, WML may contribute to lower cognitive performance. Then again, when analyzing separate neurocognitive domains, the association between changes in motor speed and WML was found in both type 1 diabetes patients and healthy controls. The idea of a cognitive reserve capacity, which determines whether people experience cognitive decline or not has been postulated before.28 In this study, we included middle‐aged type 1 diabetes patients and control subjects, which we followed over a relatively short follow‐up period. In this period, we did not detect a mean difference in general cognitive ability.

determines whether people experience cognitive decline or not has been postulated before.28 In this study, we included middle‐aged type 1 diabetes patients and control subjects, which we followed over a relatively short follow‐up period. In this period, we did not detect a mean difference in general cognitive ability. We speculate that in older subjects we would most likely find more WML, more cMB as well as more pronounced microvascular dysfunction in both control subjects and type 1 diabetic patients, since aging itself is known to be an important contributor to both (skin) microvascular dysfunction29 and cSVD.28 This may lead to more pronounced effects of general cognitive ability, and perhaps more pronounced differences between healthy subjects and type 1 diabetic patients.

l subjects and type 1 diabetic patients, since aging itself is known to be an important contributor to both (skin) microvascular dysfunction29 and cSVD.28 This may lead to more pronounced effects of general cognitive ability, and perhaps more pronounced differences between healthy subjects and type 1 diabetic patients. Diabetes is known to accelerate microvascular aging.30 Mechanisms leading to the impairment of microcirculation in diabetes are extensive, including increased polyol pathway flux, enhanced formation of glycation end products (AGEs), abnormal activation of signaling cascades such as protein kinase C and increased hexosamine pathway flux.31, 32 These processes result from hyperglycemia‐induced overproduction of reactive oxygen species by the mitochondrial electron transport chain.32, 33, 34 These alterations in the vasculature lead to vascular leakage, a pro‐inflammatory, pro‐thrombotic and more vasoconstrictive state, and are involved in the development of both macro‐ and microvascular complications of diabetes. Skin microvascular function measurements are easy, non‐invasive, and fast and are considered a representative model for generalized microvascular (dys)function,19 including microvascular (dys)function in the brain. Several studies have demonstrated abnormalities in peripheral35, 36, 37 and cerebral38, 39, 40 microvascular function in type 1 diabetes patients. In the present study, we were unable to detect significant differences in baseline skin microvascular function, yet baseline and peak hyperemia capillary density are lower in the type 1 diabetes patients, indicating a possible power problem. Indeed, a previous analysis performed by us in a larger sample showed a trend across groups toward lower baseline capillary function in patients with type 1 diabetes compared to controls (type 1 diabetes with retinopathy: 45 ± 7 capillaries/mm2; type 1 diabetes patients without retinopathy: 46 ± 9 capillaries/mm2; healthy controls: 48 ± 10 n/mm2; P = 0.05). Similar results were shown for capillary density after arterial occlusion (peak reactive hyperemia).11 We therefore assume that the differences we detected in this study are real, but not reaching statistical significance due to our small sample size.

46 ± 9 capillaries/mm2; healthy controls: 48 ± 10 n/mm2; P = 0.05). Similar results were shown for capillary density after arterial occlusion (peak reactive hyperemia).11 We therefore assume that the differences we detected in this study are real, but not reaching statistical significance due to our small sample size. The treatment of type 1 diabetes is based on the balance between lowering HbA1c levels, without increasing (the risk of) hypoglycemic events. In our study, both HbA1c and severe hypoglycemic events were associated with cognitive performance over time, but neither influenced the relationship between WMLs and skin capillary perfusion with cognitive performance over time in the multiple regression model. The relationship between hypoglycemic episodes and cognitive dysfunction in middle‐aged type 1 diabetes is less evident than the relationship between hyperglycemia‐related damage on cognitive dysfunction. Retrospective studies in adult patients with type 1 diabetes have suggested an association between a history of recurrent severe hypoglycemia and a modest‐to‐severe degree of cognitive impairment.41 However, large prospective studies have failed to confirm this association.5, 41, 42 These contradictory results concerning the relationship between hypoglycemia and cognitive decline in middle‐aged patients with type 1 diabetes may be partially explained by the positive relationship between the frequency of hypoglycemic episodes and glycemic control (lower HbA1c), of which the latter improves cognitive function.

adictory results concerning the relationship between hypoglycemia and cognitive decline in middle‐aged patients with type 1 diabetes may be partially explained by the positive relationship between the frequency of hypoglycemic episodes and glycemic control (lower HbA1c), of which the latter improves cognitive function. In this study, lower skin microvascular function at baseline was associated with lower cognitive performance over time in type 1 diabetes patients. Furthermore, the association between the presence of WML and lower performance in general cognitive ability over time was significantly reduced to non‐significant levels when adjusting for baseline capillary perfusion. Interestingly, we did not detect a correlation between skin microvascular function and cognition when analyzing our baseline data, even though the sample size of the cross‐sectional study was larger. Using longitudinal data has the great advantage of being less hampered by inter‐individual differences, which may have revealed these correlations. Nonetheless, we should be aware that, in this subsample, two data points have a large influence on the detected associations, despite the fact that they are not indicated as influential outliers by Cook's distance analyses. Theoretically, there is ground for a causal mechanism that links microvascular dysfunction to lower cognitive performance. Maintenance of adequate cerebral perfusion is vital for the preservation of normal brain function, since the brain has no buffer for nutrients and oxygen and relies exclusively on perfusion to meet neuronal metabolic demand.43 Cerebral autoregulation, including myogenic responses to changes in blood pressure, is an important mechanism for maintaining stable cerebral blood flow and to prevent hypoxia, hypo‐ and hypercapnia.43, 44, 45 In addition to myogenic responses, cerebral perfusion also depends on microvascular endothelial function. Endothelium‐dependent NO production is a key contributor of moment‐to‐moment adjustment of regional cerebral perfusion to changes in neuronal activity.43 Lower peak hyperemia and lower capillary recruitment of the skin could translate to both loss of this myogenic cerebral autoregulation and endothelium‐dependent microvascular function. Loss of these functions can potentially reduce oxygen delivery and alter neuronal activation and may therefore have detrimental effects on the brain.

eremia and lower capillary recruitment of the skin could translate to both loss of this myogenic cerebral autoregulation and endothelium‐dependent microvascular function. Loss of these functions can potentially reduce oxygen delivery and alter neuronal activation and may therefore have detrimental effects on the brain. Indeed, in type 1 diabetes patient total gray matter cerebral blood flow is reduced 39, 46 and regional cerebral hemodynamic response to incremental exercise is blunted compared with control subjects.47 Furthermore, in animal models of diabetes, improvement of cerebral blood flow by chronic treatment with an angiotensin‐converting enzyme inhibitor (ACE‐inhibitor) was found to be associated with improvement of cognitive function.48

emodynamic response to incremental exercise is blunted compared with control subjects.47 Furthermore, in animal models of diabetes, improvement of cerebral blood flow by chronic treatment with an angiotensin‐converting enzyme inhibitor (ACE‐inhibitor) was found to be associated with improvement of cognitive function.48 Limitations of our study design have to be considered, including the small sample size and observational character of the study. The small sample size made it impossible to correct for all possible confounding factors. In studies using MRI, this is often the case, since MRI techniques are expensive and time‐consuming. We therefore chose several confounding factors based on current literature 20, 21 and performed a simple regression analysis to identify variables for the multiple regression model. Furthermore, this was an observational study, which makes it impossible to draw conclusions on causality. Nevertheless, the prospective correlations we found fit the hypothesis that generalized impairment of microvascular function is involved in cognitive performance over time in subjects with type 1 diabetes. Other limitations include the follow‐up time, which was fairly brief (<4 years), and may explain why there were no significant changes in mean cognition in both groups. The use of a 1.5‐T magnetic resonance system was standard during the initiation of the baseline study in 2006, yet stronger 3T systems more sensitively detect cMB with a lower inter‐observer variability. To circumvent this problem as much as possible, we used susceptibility weighted imaging, which is highly sensitive to cMB. Nevertheless, our MRI field strength may have led to an underestimation of the amount of cMB and consequently a decreased chance of finding significant differences in cMB between the two study groups or correlations between cMBs and cognition over time. Third, in this study we selected T1DM patient with proliferative retinopathy, and therefore, the data cannot be translated to T1DM patients in general. T1DM patients with proliferative retinopathy have more deteriorated skin capillary perfusion compared with those without,11 which may have a more pronounced effect on cognitive decline, assuming it reflects a further deterioration of generalized microvascular dysfunction.

nnot be translated to T1DM patients in general. T1DM patients with proliferative retinopathy have more deteriorated skin capillary perfusion compared with those without,11 which may have a more pronounced effect on cognitive decline, assuming it reflects a further deterioration of generalized microvascular dysfunction. Finally, the associations between general cognitive ability and baseline capillary density (Figure 1, panel B), VO capillary density (Figure 1, panel D), and capillary recruitment (Figure 1, panel E) are not significant when deleting the two most extreme data points from the analysis, although a positive non‐significant correlation is found (B = 0.090, P = 0.670; B = 0.391, P = 0.134; B = 0.325, P = 0.219, respectively). It is likely that correlations will diminish when the two most influential data points are deleted from a relatively small dataset. Since these data points are not statistical outliers, and the data points in these correlations are not derived from two specific individuals, we consider it correct to keep these data points in our analysis. In addition, the association between general cognitive ability and peak capillary density (Figure 1, panel C) remains statistically significant when deleting these two points (B = 0.533; P = 0.034).

elations are not derived from two specific individuals, we consider it correct to keep these data points in our analysis. In addition, the association between general cognitive ability and peak capillary density (Figure 1, panel C) remains statistically significant when deleting these two points (B = 0.533; P = 0.034). In conclusion, this study demonstrates that in type 1 diabetes patients with proliferative retinopathy, the presence of WML and lower skin capillary perfusion at baseline is associated with lower performance in general cognitive ability over time. In addition, the relationship between WML and cognitive decline was significantly reduced when correcting for capillary perfusion measurements. We previously showed that type 1 diabetes patients with proliferative retinopathy have more cMB compared to those without retinopathy and that in patients with cMB capillary perfusion is impaired.11 Together, these data fit our hypothesis that cSVD is a manifestation of generalized microvascular dysfunction, leading to cognitive dysfunction. Future research with a larger sample size and longer follow‐up should confirm these observations. In addition, including type 1 diabetes patients without retinopathy helps to discriminate between hyperglycemia and microvascular damage as an underlying cause of lower cognitive performance.

ding to cognitive dysfunction. Future research with a larger sample size and longer follow‐up should confirm these observations. In addition, including type 1 diabetes patients without retinopathy helps to discriminate between hyperglycemia and microvascular damage as an underlying cause of lower cognitive performance. PERSPECTIVE These are the first prospective data that show a relationship between cognitive decline, cerebral small vessel disease, and microvascular dysfunction. These data fit our hypothesis that cerebral microangiopathy is a manifestation of generalized microvascular dysfunction, leading to cognitive dysfunction. Future research with a larger sample size and longer follow‐up should confirm these observations. CONFLICT OF INTEREST There is no conflict of interests to declare. AUTHORS' CONTRIBUTIONS A.L.E. performed the statistical analysis and wrote the manuscript. E.v.D. participated in the design of the study, collected all magnetic resonance imaging data and part of skin capillary data, and wrote the manuscript. M.P.W. rated the magnetic resonance imaging scans. M.K., F.B. and F.J.S., MD and RGI participated in the design of the study. E.C.E. and E.H.S. supervised the statistical analysis of the study. In addition, all authors were involved in drafting the manuscript and making revisions to the manuscript. Supporting information Click here for additional data file. ACKNOWLEDGMENTS This research is supported by grant 2005.00.006 of the Dutch Diabetes Research Foundation and the European Foundation for the Study of Diabetes.

ed of allograft rejection with normal serum cyclosporine A and tacrolimus concentrations received renal allograft biopsy. We evaluated the application of CEUS in the assessment of different pathologic renal allograft dysfunction (AR and CR) and further established a novel and simple noninvasive model for predicting CR. 2 MATERIALS AND METHODS 2.1 Patients A total of 66 renal transplant recipients in the derivation group were enrolled in this prospective study from January 2011 to December 2016. For the validation group, 38 recipients were enrolled from January 2017 to September 2017. The validation group underwent the same studies as the derivation group. All patients received living‐related or deceased donor kidneys. Due to increased SCr either rapidly or slightly, all patients received renal allograft biopsy for pathological diagnosis. Before biopsy, all patients were admitted to our abdominal ultrasound unit for ultrasound examination. Patients with renal allograft artery stenosis, artery or venous thrombus, renal allograft urinary obstruction, perirenal hematoma, ATN, BK virus‐associated nephropathy, tubulointerstitial nephritis, CNI toxicity, and thrombotic microangiopathy were excluded. This study protocol was approved by the Ethics Committee. Procedures in this study were in accordance with the Helsinki Declaration of 2000, with informed consent from the participants.

atoma, ATN, BK virus‐associated nephropathy, tubulointerstitial nephritis, CNI toxicity, and thrombotic microangiopathy were excluded. This study protocol was approved by the Ethics Committee. Procedures in this study were in accordance with the Helsinki Declaration of 2000, with informed consent from the participants. 2.2 Ultrasound examination and principle of contrast‐enhanced ultrasonography All patients in our study underwent ultrasound‐based measurement of RI and assessment of renal allograft perfusion by CEUS quantification by two experienced ultrasound physicians with at least 10 years of experience in clinical ultrasound examinations. The doctors were blinded to the clinical and laboratory data of patients. To calculate the volume of the transplanted kidney, we used the ellipsoid formula as follows: volume (cm3) = length (cm) × width (cm) × thickness (cm) × π/6. The three dimensions of the transplanted kidney were measured by B‐mode ultrasound examination.

Abbreviations ABMRantibody‐mediated rejection ARacute rejection ATNacute tubular necrosis AUROCarea under receiver operating characteristic curve BMIbody mass index CEUScontrast‐enhanced ultrasonography CNIcalcineurin inhibitor CRchronic rejection eGFRestimated glomerular filtration rate ISPimmunosuppressive protocol +LRpositive likelihood ratio ‐LRnegative likelihood ratio MDRDModification of Diet in Renal Disease MMFmycophenolate mofetil NPVnegative predictive value PPVpositive predictive value PRApanel reactive antibody Predprednisone QOFquality of fit RIresistive index ROCreceiver operating characteristic RTrising time RTcrising time of cortex RTirising time of interlobar artery RTmrising time of medulla RTsrising time of segmental artery SCrserum creatinine Sensensitivity Spespecificity SRLsirolimus TCMRT cell‐mediated rejection TICstime‐intensity curves TTPtime to peak TTPctime to peak of cortex TTPitime to peak of interlobar artery TTPmtime to peak of medulla TTPstime to peak of segmental artery ΔRTm‐cchange in RT between medulla and cortex ΔTTPm‐cchange in TTP between medulla and cortex

SCrserum creatinine Sensensitivity Spespecificity SRLsirolimus TCMRT cell‐mediated rejection TICstime‐intensity curves TTPtime to peak TTPctime to peak of cortex TTPitime to peak of interlobar artery TTPmtime to peak of medulla TTPstime to peak of segmental artery ΔRTm‐cchange in RT between medulla and cortex ΔTTPm‐cchange in TTP between medulla and cortex 1 INTRODUCTION Conventional ultrasound in combination with color Doppler imaging is a common and protocol examination for patients after renal transplantation. However, conventional ultrasound in combination with Doppler imaging can diagnose renal artery stenosis and vein thrombosis, but it is not possible to display subtle microvascular tissue perfusion. Apart from immunological factors, one of the most important factors to ensure a stable allograft function is renal blood supply, which is mainly influenced by parenchymal blood perfusion. Over 90% of renal blood flow in the renal cortex is provided by small renal arterioles and capillaries.1, 2 Therefore, conventional ultrasound is not able to provide accurate and more information for the evaluation of acute and chronic allograft dysfunctions.

, which is mainly influenced by parenchymal blood perfusion. Over 90% of renal blood flow in the renal cortex is provided by small renal arterioles and capillaries.1, 2 Therefore, conventional ultrasound is not able to provide accurate and more information for the evaluation of acute and chronic allograft dysfunctions. Contrast‐enhanced ultrasonography, which employs microbubble contrast agents and complementary harmonic pulse sequences to demonstrate parenchymal perfusion, can be a helpful problem‐solving tool in several clinical scenarios, including kidney diseases.3 The main ultrasound contrast agent approved in Europe is SonoVue (BR1, Bracco, Milan, Italy). Contrast‐enhanced ultrasonography is safe and well‐tolerated in patients without renal toxicity and cross‐allergy anaphylactic reaction.3, 4 The application of CEUS in renal transplants highlights its versatility in immediate problem solving without recourse to other potentially nephrotoxic agents. In addition, CEUS can uniquely provide additional information not available from other modalities about the microcirculation.5

allergy anaphylactic reaction.3, 4 The application of CEUS in renal transplants highlights its versatility in immediate problem solving without recourse to other potentially nephrotoxic agents. In addition, CEUS can uniquely provide additional information not available from other modalities about the microcirculation.5 In our previous prospective study, we discriminated AR from ATN in renal allografts using CEUS. After establishing a mathematic model, AR can also be distinguished from non‐AR recipients at any time period after transplantation.6 However, whether CEUS can diagnose renal allograft CR has not been reported. Up to 50% of all rejection episodes are subclinical and occur without changes in standard parameters, such as creatinine or blood urea nitrogen.7, 8 CR, which is mainly mediated by antibodies against donor antigens, has been the major cause of long‐term graft loss. Vasculopathy and disturbances in allograft perfusion occur in CR.9, 10 Most of these vascular insults affect small parenchymal arteries and arterioles, which cannot be assessed by conventional Doppler ultrasound. Thus, it might be helpful to provide CR diagnostic evidence by evaluating renal allograft microperfusion. In this study, patients who were suspected of allograft rejection with normal serum cyclosporine A and tacrolimus concentrations received renal allograft biopsy. We evaluated the application of CEUS in the assessment of different pathologic renal allograft dysfunction (AR and CR) and further established a novel and simple noninvasive model for predicting CR.

tors were blinded to the clinical and laboratory data of patients. To calculate the volume of the transplanted kidney, we used the ellipsoid formula as follows: volume (cm3) = length (cm) × width (cm) × thickness (cm) × π/6. The three dimensions of the transplanted kidney were measured by B‐mode ultrasound examination. All patients were examined at 8‐10 am on an empty stomach without fluid infusion or caffeine intake. Patients were examined in the supine position. CEUS examination was performed using a Philips iU‐22 ultrasonic apparatus with a C5‐1 probe (Philips, Amsterdam, the Netherlands) with an intravenous bolus injection of 0.6 mL SonoVue. After routine B‐mode ultrasound and color Doppler examination, we chose the longitudinal section of the transplant kidney as the fixed section for CEUS, which showed the renal hilus and the maximum area of the transplant kidney. Image acquisition began at the start of the SonoVue injection, and a capture of 120 s was recorded continuously onto the local hard drive as a DICOM file. The main gain, focus position, TGC, and other presets remained constant when CEUS was being performed. The mechanical index (MI) was set at 0.07.

transplant kidney. Image acquisition began at the start of the SonoVue injection, and a capture of 120 s was recorded continuously onto the local hard drive as a DICOM file. The main gain, focus position, TGC, and other presets remained constant when CEUS was being performed. The mechanical index (MI) was set at 0.07. The proprietary software SonoLiver (TomTec Imaging Systems Gmbh, Munich, Germany) was employed for quantitative data analysis. A 50 mm2 ROI was placed over the area, including the cortex, medulla, interlobar artery, and segmental artery in the mid pole of the transplanted kidneys; consequently, 4 TICs were generated for each patient (Figure 1). In this study, the segmental artery was chosen as the reference region, and the QOF ≥ 75% between the reference region and the analysis region was regarded as the permission of enrolling. We used two quantitative parameters, RT and TTP, for further analysis. The repeatability was good, as previously described.6 Figure 1 CEUS quantification measurement and TIC. A, The 4 regions of interest were demonstrated. Yellow circle: segmental artery; green circle: interlobar artery; purple dotted circle: medulla; purple solid circle: cortex. B, In AR and CR kidneys, the TIC was coarse, especially in the AR kidney, with apparent ups and downs. In addition, the peak of the TIC for CR was sharper than AR

of interest were demonstrated. Yellow circle: segmental artery; green circle: interlobar artery; purple dotted circle: medulla; purple solid circle: cortex. B, In AR and CR kidneys, the TIC was coarse, especially in the AR kidney, with apparent ups and downs. In addition, the peak of the TIC for CR was sharper than AR 2.3 Blood and pathological examinations Blood samples were taken on the same day. All clinical blood tests were performed in the Department of Clinical Laboratory, including routine blood and renal function tests. The eGFR was calculated with the simplified MDRD Study equation. Renal allograft biopsy and histologic examination based on the Banff 2013 classification were performed.

n on the same day. All clinical blood tests were performed in the Department of Clinical Laboratory, including routine blood and renal function tests. The eGFR was calculated with the simplified MDRD Study equation. Renal allograft biopsy and histologic examination based on the Banff 2013 classification were performed. 2.4 Data management and statistical analysis The results were expressed as the mean values ± SD or median (interquartile range). In the derivation group, one‐way ANOVA tests or independent sample t tests were used to compare the markers among the groups. Significant variables from the univariate analysis (P < 0.05) were then subjected to multivariate analysis by forwarding logistic regression to identify independent factors associated with either end point. CR was considered as a positive result and AR as a negative result. A predictive index was constructed by modeling the values of the independent variables. The diagnostic values of parameters were assessed by calculating the AUROCs. The best cutoff points were selected from the ROC curve to identify the presence or absence of CR. For that purpose, we selected cutoff points with a 90% certainty for the presence and absence of CR, thus regarding 10% false‐negative or false‐positive results as clinically acceptable; we also selected an optimal cutoff point according to the best Youden index. The diagnostic accuracy was calculated using Sen, specificity, NPV and PPV, and likelihood ratios. Then, the index derived in the derivation group was tested in the validation group. All data were analyzed using SPSS 19.0 (SPSS Inc., Chicago, USA).

so selected an optimal cutoff point according to the best Youden index. The diagnostic accuracy was calculated using Sen, specificity, NPV and PPV, and likelihood ratios. Then, the index derived in the derivation group was tested in the validation group. All data were analyzed using SPSS 19.0 (SPSS Inc., Chicago, USA). 3 RESULTS 3.1 Demographics and baseline characteristics In the derivation group, the mean age of all 66 enrolled patients (53 males, 13 females) was 39.1 ± 12.1 years. All patients were administered triple immunosuppressants, including MMF, cyclosporine A or tacrolimus, SRL, and Pred. The mean time from renal transplantation was 51 months (duration of 1 week to 180 months). In the validation group, the mean age of all patients (28 males, 10 females) was 41.4 ± 9.9 years. Patients in the validation group also received a standard triple ISP. The mean time from renal transplantation was 61 months (duration of 3 weeks to 228 months). In the derivation group, 41 patients presented AR, and 25 patients presented CR with a biopsy‐proven histologic examination. In the validation group, there were 22 patients with AR and 16 patients with CR. There were no significant differences between the AR and CR groups in terms of sex, weight, BMI, or ISP in either the derivation or validation groups. The demographics and baseline characteristics of each group are shown in Table 1. For the derivation group, the baseline SCr and eGFR in the AR and CR groups were 115.4 ± 30.2 vs. 105.6 ± 25.4 μmol/L and 75.2 ± 37.6 vs. 80.7 ± 44.4 ml/min/1.73 m2, respectively. The baseline SCr and eGFR were not significantly different between the AR and CR groups. The renal function data from the biopsy and ultrasound examination are presented in Table 2.

r and eGFR in the AR and CR groups were 115.4 ± 30.2 vs. 105.6 ± 25.4 μmol/L and 75.2 ± 37.6 vs. 80.7 ± 44.4 ml/min/1.73 m2, respectively. The baseline SCr and eGFR were not significantly different between the AR and CR groups. The renal function data from the biopsy and ultrasound examination are presented in Table 2. Table 1 Demographic characteristics and ultrasound indexes of patients Derivation group (n = 66) Validation group (n = 38) AR (n = 41) CR (n = 25) P value AR (n = 22) CR (n = 16) P value Sex (Male/Female) 33/8 20/5 >0.05 18/4 10/6 >0.05 Age (Years) 36.29 ± 12.21 43.76 ± 10.59 0.014 40.14 ± 8.54 43.13 ± 11.34 >0.05 Weight (kg) 63.87 ± 4.87 62.76 ± 5.34 >0.05 65.01 ± 5.23 64.91 ± 6.08 >0.05 BMI 23.82 ± 1.87 23.39 ± 2.12 >0.05 23.75 ± 2.01 23.12 ± 2.32 >0.05 Post‐transplant time at US examination (month) 27.61 ± 32.04 90.76 ± 53.64 <0.001 35.98 ± 57.12 96.94 ± 53.96 <0.001 Pretransplant PRA (%) Class I 3.59 ± 12.92 4.92 ± 12.35 0.087 0.59 ± 2.06 4.00 ± 8.21 >0.05 Class II 1.32 ± 5.96 12.20 ± 25.24 0.008 3.00 ± 10.19 2.00 ± 5.57 >0.05 ISP     NA     NA CNI + MMF + Pred 41 25   20 14 SRL + MMF + Pred 0 0   2 2 RI 0.60 ± 0.07 0.64 ± 0.10 >0.05 0.65 ± 0.12 0.62 ± 0.09 >0.05 Kidney volume change (%) 11.56 ± 15.92 ‐10.90 ± 17.54 <0.001 20.44 ± 26.38 ‐15.30 ± 17.58 <0.001 RTm 12.97 ± 6.13 15.85 ± 3.82 0.040 10.21 ± 3.05 13.82 ± 3.87 0.003 TTPm 14.25 ± 6.08 17.39 ± 4.61 0.030 12.45 ± 4.33 16.25 ± 5.29 0.024 Data are presented as the mean values with the standard error of the mean. John Wiley & Sons, LtdTable 2 Renal function

Derivation group (n = 66) Validation group (n = 38) AR (n = 41) CR (n = 25) P value AR (n = 22) CR (n = 16) P value Sex (Male/Female) 33/8 20/5 >0.05 18/4 10/6 >0.05 Age (Years) 36.29 ± 12.21 43.76 ± 10.59 0.014 40.14 ± 8.54 43.13 ± 11.34 >0.05 Weight (kg) 63.87 ± 4.87 62.76 ± 5.34 >0.05 65.01 ± 5.23 64.91 ± 6.08 >0.05 BMI 23.82 ± 1.87 23.39 ± 2.12 >0.05 23.75 ± 2.01 23.12 ± 2.32 >0.05 Post‐transplant time at US examination (month) 27.61 ± 32.04 90.76 ± 53.64 <0.001 35.98 ± 57.12 96.94 ± 53.96 <0.001 Pretransplant PRA (%) Class I 3.59 ± 12.92 4.92 ± 12.35 0.087 0.59 ± 2.06 4.00 ± 8.21 >0.05 Class II 1.32 ± 5.96 12.20 ± 25.24 0.008 3.00 ± 10.19 2.00 ± 5.57 >0.05 ISP     NA     NA CNI + MMF + Pred 41 25   20 14 SRL + MMF + Pred 0 0   2 2 RI 0.60 ± 0.07 0.64 ± 0.10 >0.05 0.65 ± 0.12 0.62 ± 0.09 >0.05 Kidney volume change (%) 11.56 ± 15.92 ‐10.90 ± 17.54 <0.001 20.44 ± 26.38 ‐15.30 ± 17.58 <0.001 RTm 12.97 ± 6.13 15.85 ± 3.82 0.040 10.21 ± 3.05 13.82 ± 3.87 0.003 TTPm 14.25 ± 6.08 17.39 ± 4.61 0.030 12.45 ± 4.33 16.25 ± 5.29 0.024 Data are presented as the mean values with the standard error of the mean. John Wiley & Sons, LtdTable 2 Renal function Derivation group (n = 66) Validation group (n = 38) AR (n = 41) CR (n = 25) P value AR (n = 22) CR (n = 16) P value SCr (μmol/L) 230.32 ± 134.44 340.52 ± 205.09 0.012 236.50 ± 171.70 226.00 ± 164.49 >0.05 eGFR (mL/min/1.73 m2) 39.45 ± 17.83 26.29 ± 14.62 0.003 39.32 ± 18.87 39.06 ± 19.86 >0.05 Data are presented as the mean values with standard error of the mean.

= 38) AR (n = 41) CR (n = 25) P value AR (n = 22) CR (n = 16) P value SCr (μmol/L) 230.32 ± 134.44 340.52 ± 205.09 0.012 236.50 ± 171.70 226.00 ± 164.49 >0.05 eGFR (mL/min/1.73 m2) 39.45 ± 17.83 26.29 ± 14.62 0.003 39.32 ± 18.87 39.06 ± 19.86 >0.05 Data are presented as the mean values with standard error of the mean. John Wiley & Sons, Ltd3.2 Comparison of renal function and CEUS parameters The patients with CR demonstrated significantly higher SCr and lower eGFR compared to the AR group (Figure 2A,B). No difference was observed in RI between these two groups (Figure 2C). The kidney volume change was calculated as (volume 2‐volume 1)/volume 1. The volume 2 was tested by B‐mode ultrasound examination before CEUS was performed. The volume 1 was tested within 24 hours post‐transplantation. In the CR group, the kidney volume change was significantly decreased compared to that in the AR group (Figure 2D). The RT and TTP of the medulla (RTm and TTPm) were significantly longer in the CR group than those in the AR group (Figure 2E,F). Figure 2 Comparison of renal function and ultrasound features between AR and CR patients in the derivation group. A, SCr; (B) eGFR; (C) RI; (D) kidney volume change; (E) RTm; (F) TTPm; (G) New index between the AR and CR groups; (H) New index between the TCMR and ABMR groups in CR patients. Data are expressed as the mean ± SD

2 Comparison of renal function and ultrasound features between AR and CR patients in the derivation group. A, SCr; (B) eGFR; (C) RI; (D) kidney volume change; (E) RTm; (F) TTPm; (G) New index between the AR and CR groups; (H) New index between the TCMR and ABMR groups in CR patients. Data are expressed as the mean ± SD 3.3 Predictors of CR and establishment of the diagnostic model All patients enrolled were divided into the CR group and AR group. In the univariate analysis, variables including age, SCr, eGFR, kidney volume change, RTm, and TTPm were identified as predictors of CR. Among them, three variables were identified as independent predictors by multivariate analysis as follows: age (OR = 1.077, 95% CI 1.012‐1.146, P = 0.019), kidney volume change (OR = 5.45E‐5, 95% CI 2.28E‐7‐0.013, P < 0.001), and TTPm (OR = 1.122, 95% CI 1.001‐1.258, P = 0.048). ROC curves were estimated for individual markers in all patients. The AUROCs for age, kidney volume change, and TTPm in diagnosing renal allograft CR were 0.696, 0.831, and 0.687, respectively (Table 3 and Figure 3A). Table 3 ROC analysis in the derivation group to evaluate discrimination ability

3.3 Predictors of CR and establishment of the diagnostic model All patients enrolled were divided into the CR group and AR group. In the univariate analysis, variables including age, SCr, eGFR, kidney volume change, RTm, and TTPm were identified as predictors of CR. Among them, three variables were identified as independent predictors by multivariate analysis as follows: age (OR = 1.077, 95% CI 1.012‐1.146, P = 0.019), kidney volume change (OR = 5.45E‐5, 95% CI 2.28E‐7‐0.013, P < 0.001), and TTPm (OR = 1.122, 95% CI 1.001‐1.258, P = 0.048). ROC curves were estimated for individual markers in all patients. The AUROCs for age, kidney volume change, and TTPm in diagnosing renal allograft CR were 0.696, 0.831, and 0.687, respectively (Table 3 and Figure 3A). Table 3 ROC analysis in the derivation group to evaluate discrimination ability Parameters AUROC 95% CI P value RTs 0.650 0.519‐0.781 0.042 RTi 0.681 0.551‐0.812 0.014 RTc 0.672 0.542‐0.802 0.020 RTm 0.701 0.577‐0.826 0.006 ΔRTm‐c 0.663 0.528‐0.799 0.027 TTPs 0.579 0.439‐0.718 0.287 TTPi 0.668 0.540‐0.796 0.023 TTPc 0.682 0.554‐0.810 0.014 TTPm 0.687 0.561‐0.814 0.011 ΔTTPm‐c 0.645 0.507‐0.782 0.050 eGFR 0.709 0.582‐0.836 0.005 Age 0.696 0.566‐0.828 0.008 Kidney volume change 0.831 0.729‐0.933 <0.001 RI 0.640 0.502‐0.778 0.058 New index 0.886 0.807‐0.965 <0.001 John Wiley & Sons, LtdFigure 3 Area under ROC curves. AUROCs estimated the diagnostic performance of the new index, age, RTm, TTPm, eGFR and kidney volume change in the derivation group (A) and validation group (B)

dney volume change 0.831 0.729‐0.933 <0.001 RI 0.640 0.502‐0.778 0.058 New index 0.886 0.807‐0.965 <0.001 John Wiley & Sons, LtdFigure 3 Area under ROC curves. AUROCs estimated the diagnostic performance of the new index, age, RTm, TTPm, eGFR and kidney volume change in the derivation group (A) and validation group (B) Based on the multivariate analysis results and AUROCs of individual markers, we constructed a new index expressed by the following formula: P=−5.424+0.074×age−9.818×kidney volume change+0.115×TTPm New Index=eP/(1+eP) We found that this new index was significantly elevated in CR patients compared with AR patients (median (interquartile range), 0.64 (0.48, 0.87) vs 0.09 (0.03, 0.39), P = 1.71E‐7) (Figure 2G). No significant differences were observed between the TCMR and ABMR subgroups in CR patients (Figure 2H). Moreover, the highest AUROC was observed for this index (AUROC = 0.89, 95% CI 0.81‐0.97). Two cutoff values were chosen to identify the absence (less than 0.36) and presence (greater than 0.70) of renal allograft CR (Table 4). By applying the lower cutoff value (a score below 0.36), 30 of the 31 patients with AR were correctly identified, and the Sen and specificity were 96% and 73%, respectively. At the cutoff value of 0.70, the Sen and specificity were 48% and 93%, respectively, and 12 of 15 patients with CR were correctly identified. Using these two cutoff values, approximately 70% of patients could be correctly diagnosed, with over 90% accuracy.

ed, and the Sen and specificity were 96% and 73%, respectively. At the cutoff value of 0.70, the Sen and specificity were 48% and 93%, respectively, and 12 of 15 patients with CR were correctly identified. Using these two cutoff values, approximately 70% of patients could be correctly diagnosed, with over 90% accuracy. Table 4 Sensitivity, specificity, predictive values, and likelihood ratios of new index according to different cutoffs for the diagnosis of CR Cutoff CR predicted by New index All patients n (%) Renal allograft biopsy Sen Spe NPV PPV ‐LR +LR AR n (%) CR n (%) Derivation group (n = 66) 0.36 AR 31 (47%) 30 (73%) 1 (4%) 96% 73% 97% 69% 0.05 3.6 CR 35 (53%) 11 (27%) 24 (96%) 0.70 AR 51 (77%) 38 (93%) 13 (52%) 48% 93% 75% 80% 0.56 6.6 CR 15 (23%) 3 (7%) 12 (48%) Validation group (n = 38) 0.36 AR 21 (55%) 18 (82%) 3 (19%) 81% 82% 86% 76% 0.23 4.5 CR 17 (45%) 4 (18%) 13 (81%) 0.70 AR 28 (74%) 21 (95%) 7 (44%) 56% 95% 75% 90% 0.46 12 CR 10 (26%) 1 (5%) 9 (56%) John Wiley & Sons, Ltd3.4 Comparison of the new index with individual markers for predicting CR Area under ROC curves were used to evaluate the overall diagnostic performance of this new index and individual markers (Figure 3A). We found that this new index had a better AUROC than that of kidney volume change, eGFR, and individual CEUS markers. Additionally, this new index was superior to individual markers when optimal cutoff values were applied (Table 5). Table 5 Sensitivity, specificity, predictive values, and likelihood ratios of models according to optimal cutoff for the diagnosis of CR in the derivation group

Derivation group (n = 66) 0.36 AR 31 (47%) 30 (73%) 1 (4%) 96% 73% 97% 69% 0.05 3.6 CR 35 (53%) 11 (27%) 24 (96%) 0.70 AR 51 (77%) 38 (93%) 13 (52%) 48% 93% 75% 80% 0.56 6.6 CR 15 (23%) 3 (7%) 12 (48%) Validation group (n = 38) 0.36 AR 21 (55%) 18 (82%) 3 (19%) 81% 82% 86% 76% 0.23 4.5 CR 17 (45%) 4 (18%) 13 (81%) 0.70 AR 28 (74%) 21 (95%) 7 (44%) 56% 95% 75% 90% 0.46 12 CR 10 (26%) 1 (5%) 9 (56%) John Wiley & Sons, Ltd3.4 Comparison of the new index with individual markers for predicting CR Area under ROC curves were used to evaluate the overall diagnostic performance of this new index and individual markers (Figure 3A). We found that this new index had a better AUROC than that of kidney volume change, eGFR, and individual CEUS markers. Additionally, this new index was superior to individual markers when optimal cutoff values were applied (Table 5). Table 5 Sensitivity, specificity, predictive values, and likelihood ratios of models according to optimal cutoff for the diagnosis of CR in the derivation group Optimal cutoff point Sen Spe PPV NPV +LR −LR New index 96% 76% 71% 97% 3.9 0.05 Kidney volume change 72% 80% 69% 83% 3.7 0.35 RTm 80% 61% 56% 83% 2.1 0.33 eGFR 64% 71% 57% 76% 2.2 0.51 The optimal cutoff points of the new index, kidney volume change (rate), RTm and eGFR are 0.37, 4.0%, 12.7 and 30.0, respectively, based on the best Youden index in our study.

−LR New index 96% 76% 71% 97% 3.9 0.05 Kidney volume change 72% 80% 69% 83% 3.7 0.35 RTm 80% 61% 56% 83% 2.1 0.33 eGFR 64% 71% 57% 76% 2.2 0.51 The optimal cutoff points of the new index, kidney volume change (rate), RTm and eGFR are 0.37, 4.0%, 12.7 and 30.0, respectively, based on the best Youden index in our study. John Wiley & Sons, Ltd3.5 Validation of the new index The characteristics of patients in the validation group are summarized in Table 1. No differences were observed in SCr, eGFR, and RI values between the AR and CR groups (Figure 4A‐C). Similar to the derivation group, the kidney volume change was significantly lower in the CR group than that in the AR group (Figure 4D). The RTm and TTPm were increased in the CR group compared to the AR group (Figure 4D,E). The new index was significantly elevated in CR patients compared with AR patients (median (interquartile range), 0.72 (0.51, 0.94) vs 0.05 (0.13, 0.32), P = 7.68E‐7) (Figure 4G). The AUROC is shown in Figure 3B (AUROC = 0.90, 95% CI 0.81‐0.99, P < 0.001). Upon applying a high cutoff value (New Index = 0.70), CR was predicted in 26% of patients with PPV 90%, and upon applying a low cutoff value (New Index = 0.36), CR was excluded with 86% certainty in 55% of patients (Table 4). Finally, the new index showed no significant differences between the TCMR and ABMR subgroups in CR patients (Figure 4H).

f value (New Index = 0.70), CR was predicted in 26% of patients with PPV 90%, and upon applying a low cutoff value (New Index = 0.36), CR was excluded with 86% certainty in 55% of patients (Table 4). Finally, the new index showed no significant differences between the TCMR and ABMR subgroups in CR patients (Figure 4H). Figure 4 Comparison of renal function and ultrasound features between AR and CR patients in the validation group. A, SCr; (B) eGFR; (C) RI; (D) Kidney volume change; (E) RTm; (F) TTPm; (G) New index between the AR and CR groups; (H) New index between the TCMR and ABMR groups in CR patients. Data are expressed as the mean ± SD

4 Comparison of renal function and ultrasound features between AR and CR patients in the validation group. A, SCr; (B) eGFR; (C) RI; (D) Kidney volume change; (E) RTm; (F) TTPm; (G) New index between the AR and CR groups; (H) New index between the TCMR and ABMR groups in CR patients. Data are expressed as the mean ± SD 4 DISCUSSION Although the CEUS demands special analysis software and a more expensive contrast agent, the general advantages of CEUS versus conventional ultrasound in combination with color Doppler imaging in kidney transplantation are notable. Contrast‐enhanced ultrasonography examination is not contraindication for impaired kidney function. Contrast‐enhanced ultrasonography displays microvascular tissue perfusion and allows renal blood flow to be quantified, and this assessment demands no special experience of the investigator.11 There are two substantial arguments for the application of CEUS in the evaluation of renal allograft function. Renal allograft underlies progressive vascular remodeling after transplantation. Most of these vascular insults affect small parenchymal arteries and arterioles, which cannot be assessed by Doppler ultrasonography. Therefore, the major advantage of CEUS is evaluating microperfusion to augment diagnostic evidence and to permit early administration of the appropriate therapy. Second, the superficial position of the transplanted kidney and minimum organ movements due to respiration facilitates examination with a contrast agent largely, while the kidney should be kept in a stable position for the assessment of renal blood flow.12

and to permit early administration of the appropriate therapy. Second, the superficial position of the transplanted kidney and minimum organ movements due to respiration facilitates examination with a contrast agent largely, while the kidney should be kept in a stable position for the assessment of renal blood flow.12 Based on the results, we found that the essential difference between AR and CR is the microperfusion speed. CR demonstrated a lower microperfusion speed, which means a higher resistance of renal artery and arteriole. Allograft rejection is a complex process that involves the interplay of different cellular and molecular pathways that cause a broad range of allograft injuries. Allograft rejection can be hyperacute (occurring within minutes after the vascular anastomosis), acute (occurring days to weeks after transplantation), late acute (occurring 3 months after transplantation), or chronic (occurring months to years after transplantation). The major pathological changes of AR were lymphocytic infiltration, interstitial edema, intimal arteritis, tubulitis and arteritis.13, 14 However, CR is much more complex than AR. Currently, the main pathology of CR is ABMR. Despite the major advances in molecular biology and gene rearrangement, the diagnosis of ABMR is still dependent on histologic findings.15 The typical pathology of chronic ABMR is vascular endothelium injury.16 Therefore, CEUS, a quantitative method for evaluating microperfusion, is able to discriminate CR from AR. In addition, RT and TTP parameters in CR are significantly increased compared to those in the AR group.

ll dependent on histologic findings.15 The typical pathology of chronic ABMR is vascular endothelium injury.16 Therefore, CEUS, a quantitative method for evaluating microperfusion, is able to discriminate CR from AR. In addition, RT and TTP parameters in CR are significantly increased compared to those in the AR group. Allograft transplantation, not only kidney transplantation but also other solid organ transplantation, needs consecutive monitoring of allograft status. Some doctors suggest protocol biopsies at fixed time points after transplantation to diagnose any subclinical allograft injury or rejection to improve transplant outcomes. However, core needle biopsies are invasive, and frequent biopsies may be associated with severe complications. Moreover, sampling errors and variability in biopsy analysis confound conclusions about graft health. CEUS is an emerging noninvasive method but has not been reported for CR diagnosis thus far. Interestingly, Fischer et al recently tested the efficacy of CEUS in detecting allograft AR and CR using a murine heart transplantation model.17 Compared to the syngeneic groups, a progressive decline in microperfusion was demonstrated in the allografts undergoing acute transplant rejection (40%, 64%, and 92% on days 4, 6, and 8 post‐transplantation, respectively) and CR (33%, 33%, and 92% on days 5, 14, and 30 post‐transplantation, respectively). The data suggest that early endothelial cell injury and platelet aggregation contributed to the early microperfusion decline. Although a 33% decrease in microperfusion in the CR group seems less than 40% or 64% in the AR group, the detection time points are not the same. Animal experiments using CEUS in renal transplants are still lacking.

early endothelial cell injury and platelet aggregation contributed to the early microperfusion decline. Although a 33% decrease in microperfusion in the CR group seems less than 40% or 64% in the AR group, the detection time points are not the same. Animal experiments using CEUS in renal transplants are still lacking. In addition to the CEUS parameter TTPm, kidney volume change is also an important parameter for the new index formula. In the AR and CR groups, the kidney volume presented absolutely different variation tendencies. Mechanically, the swelling of the renal pyramids that occurs in AR is caused by the intense reaction initiated by the immune response. Although CR is also mediated by immune rejection similar to AR, the immune response is weaker and slower. Han et al analyzed 351 living‐donor kidney transplantation patients. They also found that the low‐graft‐volume group conferred a greater risk of rejection, chronic change, and graft loss than that in the high‐graft‐volume group.18

lso mediated by immune rejection similar to AR, the immune response is weaker and slower. Han et al analyzed 351 living‐donor kidney transplantation patients. They also found that the low‐graft‐volume group conferred a greater risk of rejection, chronic change, and graft loss than that in the high‐graft‐volume group.18 In addition to CR, interstitial fibrosis and tubular atrophy (IF/TA) is another confounding factor in the late stage of renal transplantation. Therefore, IF/TA is indeed a very important factor that is worth considering. This factor is also a limitation of the current study and will be investigated in the future. Although the precise CEUS parameters that may best predict disease still warrant systematic evaluations, animal models, and limited clinical trials in humans, our study raises hopes that CEUS could outcompete other modalities as a first‐line tool for assessing renal perfusion noninvasively.19 However, due to the limited number of kidney transplantation cases, the validation group was rather small, and a power analysis was not performed. Our results should be validated in further studies with large sample sizes, and the cutoff values may need to be adjusted or modified based on the results from further larger studies.

ue to the limited number of kidney transplantation cases, the validation group was rather small, and a power analysis was not performed. Our results should be validated in further studies with large sample sizes, and the cutoff values may need to be adjusted or modified based on the results from further larger studies. In our study, the new index cannot discriminate sub‐phenotypes of CR, such as TCMR and ABMR. Although the major cause of CR is ABMR, the therapies for TCMR and ABMR are different. ABMR is mediated by donor‐specific antibodies generated by plasma cells. The common therapeutic strategy for ABMR is IVIG, plasmapheresis, antithymocyte globulin, or rituximab. However, steroid pulse therapy and increased immunosuppressants are usually applied for TCMR. Mechanically, TCMR is mediated by T cells, rather than B cells. Therefore, if CEUS could differentiate TCMR and ABMR, it will be helpful for doctors to makes decisions for treatments. It is inspiring that there are some preliminary animal studies using targeted CEUS. For instance, in a rat kidney transplant study, microbubble contrast agents were coupled with anti‐CD3, anti‐CD4, and anti‐CD8 antibodies. Strikingly, CD3‐mediated ultrasound, which suggests T‐cell infiltration, allows the detection of AR as early as postoperative day 2.20 Similarly, Sun et al detected C4d deposition in vivo in rat kidney and heart transplant models of ABMR using C4d‐targeted microbubbles with a streptavidin‐biotin conjugation.21, 22 These studies further enhanced the specificity of CEUS in detecting allograft rejection and provide the potential for discriminating between ABMR and TCMR in the future.

C4d deposition in vivo in rat kidney and heart transplant models of ABMR using C4d‐targeted microbubbles with a streptavidin‐biotin conjugation.21, 22 These studies further enhanced the specificity of CEUS in detecting allograft rejection and provide the potential for discriminating between ABMR and TCMR in the future. In conclusion, CEUS is a reliable and useful tool for the diagnosis and follow‐up after kidney transplantation, allows for the visualization of kidney allograft microperfusion in different circumstances and is an accurate, specific, and sensitive method for the assessment of CR. Moreover, the new index established in our study provides assistance in the diagnosis of CR with a high degree of accuracy and convenience. PERSPECTIVE Contrast‐enhanced ultrasonography (CEUS) is a noninvasive imaging tool for collecting quantitative measurements of regional renal perfusion and microvascular function. Here we established a model to diagnose renal allograft chronic rejection using CEUS. We believe that antibody‐coupled CEUS which named as molecular ultrasound image will be more specific and attractive in the future. CONFLICT OF INTEREST The authors declare no conflicts of interest.

PERSPECTIVE Contrast‐enhanced ultrasonography (CEUS) is a noninvasive imaging tool for collecting quantitative measurements of regional renal perfusion and microvascular function. Here we established a model to diagnose renal allograft chronic rejection using CEUS. We believe that antibody‐coupled CEUS which named as molecular ultrasound image will be more specific and attractive in the future. CONFLICT OF INTEREST The authors declare no conflicts of interest. ACKNOWLEDGMENTS This study was supported by the National Key R&D Program of China (2018YFA0107502 to CY), the Medical and Health Talents Training Plan for the Excellent Youth of Shanghai Municipal (2018YQ50 to CY), Shanghai Rising‐Star Program (19QA1406300 to Cheng Yang), the National Natural Science Foundation of China (81270833, 81570674, 81400752, 81770746, and 81500457), and the Science and Technology Commission of Shanghai Municipality (15411964900).