Preservation of microvascular barrier function requires CD31 receptor-induced metabolic reprogramming

Endothelial barrier (EB) breaching is a frequent event during inflammation, and it is followed by the rapid recovery of microvascular integrity. The molecular mechanisms of EB recovery are poorly understood. Triggering of MHC molecules by migrating T-cells is a minimal signal capable of inducing endothelial contraction and transient microvascular leakage. Using this model, we show that EB recovery requires a CD31 receptor-induced, robust glycolytic response sustaining junction re-annealing. Mechanistically, this response involves src-homology phosphatase activation leading to Akt-mediated nuclear exclusion of FoxO1 and concomitant β-catenin translocation to the nucleus, collectively leading to cMyc transcription. CD31 signals also sustain mitochondrial respiration, however this pathway does not contribute to junction remodeling. We further show that pathologic microvascular leakage in CD31-deficient mice can be corrected by enhancing the glycolytic flux via pharmacological Akt or AMPK activation, thus providing a molecular platform for the therapeutic control of EB response.

P hysiological hyper-permeability is a typical microvasculature inflammatory response of, which facilitates diffusion of essential blood-borne immunoregulatory and pro-inflammatory mediators into extravascular tissue. Microvascular leakage results from a tightly regulated process induced by a wide range of inflammatory mediators. This event leads to phosphorylation and endocytosis of junctional vascular endothelial cadherin (VE-cadherin) complex and EC actomyosin contractility 1,2 . Endothelial contraction can open junctions via a number of molecular mediators 1,2 . Microvascular leakage can be accompanied by leukocyte extravasation 3,4 . For example, we have shown that major histocompatibility complex (MHC) moleculeligation by migrating T lymphocytes or antibodies induces transient microvascular leakage to facilitate T-cell extravasation 5 . Mechanistically, EC MHC-class-I engagement induces a rapid translocation of RhoA to the cell membrane associated with Factin stress fiber formation and cytoskeleton reorganization leading to cell contraction 6 . This effect is physiologically relevant to human disease, as capillary-leak syndrome is a key feature of vascularized allografts rejection [7][8][9] .
Following leukocyte crossing, microvascular integrity quickly recovers through re-annealing of inter-endothelial junctions, but the molecular basis of this process is poorly defined. Most mechanistic studies have been carried out in-vitro 10 . In-vivo studies have used EC-contraction mediators such as thrombin and LPS which also affect other cellular functions and cells, particularly immune cells 11,12 thus adding confounding factors which prevent the mechanistic study of EBF recovery.
Inter-endothelial CD31-homophilic interactions are required to recover vascular integrity following MHC-stimulation by migrating T-lymphocytes 5 . CD31 is a member of the immunoglobulin gene superfamily expressed at high density at the lateral borders of endothelial cells 13 . CD31 cytoplasmic tail contains two immunoreceptor-tyrosine-based-inhibitory-motifs (ITIM), which upon activation, specifically recruit src-homology 2 tyrosine phosphatases SHP-2 and SHP-1. Although CD31deficient mice do not spontaneously develop vascular leakage at steady state, CD31-deficiency has been associated with excessive vascular leakage following endothelial contraction induced by histamine 14 , thrombin 15 , and LPS-induced endotoxemia 16,17 . Importantly, loss CD31 expression by EC leads to uncontrolled T-cell extravasation in inflammatory conditions 18,19 . While this evidence points to a non-redundant role of CD31 in the recovery of endothelial continuity after barrier breach by stimuli inducing endothelial contraction, the molecular mechanisms are to-date unclear.
We have investigated the molecular mechanism of EBF recovery in response to endothelial contraction and junction release-inducing stimuli using in-vitro and in-vivo models of MHC-induced vascular leakage in the context of inter-endothelial CD31 interactions. We show that the re-establishment of microvascular integrity is dependent on a CD31-induced glycolytic response, which sustains coordinated cytoskeletal remodeling and junctional reassembly.

Results
CD31 is required for endothelial barrier recovery. To investigate the molecular mechanisms involved in the recovery of endothelial barrier function (EBF) we conducted preliminary studies to monitor the kinetics of endothelial permeability induced by MHC-class-I molecule signals (via antibody-ligation) in the presence or absence of CD31 co-engagement in confluent, WT and cd31 −/− EC monolayers. Treatment with IFN-γ was used to enhance MHC and ICAM-1 molecule expression, which is low in cultured endothelium 20 , as we have previously described 5 . As expected, antibody-triggering of MHC led to a quick and similar reduction in trans endothelial electrical resistance (TEER) by both WT and cd31 −/− endothelium (Supplementary Fig. 1a). However, while permeability of WT endothelium returned to baseline levels within 6 h, resistance of CD31-deficient endothelium remained significantly lower for up to 24 h after MHC-stimulation. TEER was not significantly affected by ICAM-1 Ab-mediated stimulation in conditions capable to induce Erk phosphorylation either in the presence or absence of CD31 expression, (Supplementary Fig. 1b). MHCtriggering did not induce EC death ( Supplementary Fig. 1c).
As endothelial contractility is associated with F-actin polymerization and stress fiber formation 21 , we further analyzed EC cytoskeletal rearrangements following MHC-triggering with or without CD31 ligation. Sub-confluent EC monolayers were used in these experiments to allow assessing the contribution of CD31 signals 'in isolation', i.e. via antibody activation, on actin polymerization. MHC-ligation led to the formation of polarized bundles of F-actin stress fibers and further separation from adjacent EC (Fig. 1a, b), a feature of EC contraction. The sparsity of the MHC-stimulated EC on the image is likely to reflect the strength of cell contraction, possibly even leading to cell detachment, as the EC were seeded in equal numbers. This was confirmed by experiments showing a similar actin configuration in MHC-stimulated CD31-deficient EC ( Supplementary Fig. 1d), while CD31 triggering on its own did not elicit any effect, coligation with MHC molecules significantly increased F-actin polymerization above the levels induced by MHC-signals, which was however associated with the cortical actin cytoskeleton and accompanied by intercellular attachment, suggesting that CD31 is required for efficient anchorage of actin fibers to the intercellular junctions during EC contraction. ICAM-1 ligation induced a slight increase in actin polymerization without endothelial contraction (Fig. 1c, d) and was not affected by co-delivery of CD31-mediated signals. Experiments designed to define the molecular mechanisms of sustained permeability of MHCstimulated cd31 −/− endothelium revealed enhanced Erk and RhoA small GTPase activation in CD31-deficient but not WT EC (Fig. 1e, f). This pathway has been previously shown to induce endothelial contraction 22 . Immunoprecipitation studies in confluent EC confirmed that CD31 becomes phosphorylated upon MHC-but not ICAM-1 molecule stimulation, and this leads to the recruitment of the Src Homology Phosphatase 2 (SHP2), a key mediator of CD31 signals (Fig. 1g).
To validate the functional consequences of these observations, we tested the effect of signals on vascular barrier function in-vivo. First, we used a selective minimal signal-i.e. MHC-antibody ligation-to avoid confounding effects related to the engagement of other endothelial receptors by T-cells. In addition, systemic administration of the antibody allowed comparing EBFin multiple organs, and estimate the specific effect of MHC-signals in the absence of other inflammatory signals. In these experiments IFNγ was not administered to avoid confounding effects. A noncomplement fixing anti-MHC-class-I antibody was mixed with a secondary antibody and injected intravenously in WT and CD31deficient mice. An anti-ICAM-1 antibody plus specific secondary antibody was also administered as a control. After 3 and 6 h, mice were injected with Evans Blue i.v. and tracer extravasation was measured. Vascular permeability was significantly increased after 6 h in the lung, liver and kidney of cd31 −/− , but not WT animals, which received anti-MHC antibody (Fig. 1h). Interestingly, no increase in the permeability of the brain vasculature was observed in either WT or cd31 −/− mice, likely due to low MHC expression typical of brain endothelium 23,24 . No effects were observed following ICAM1 stimulation ( Supplementary Fig. 1e). Systemic MHC-stimulation elicited a vascular leakage similar to that of histamine ( Supplementary Fig. 1f), a stimulus well known to induce EC contraction with amplified severity in cd31 −/− mice 17 .
We subsequently assessed the effect of MHC-triggering by Tcells both in-vitro and in-vivo.
In-vitro, allogeneic (H2-D) T-cells were generated by in-vitro stimulation with DC obtained from WT donors (H2-B) and subsequently seeded on IFN-γ-treated (48 h) WT or cd31 −/− EC monolayers grown on 0.2 μm pore transwells. TEER increased only in WT EC monolayers for the first 6 h, while it remained low in the cd31 −/− ECs (Fig. 1i). The absence of TEER decrease in this system compared to direct MHC-ligation is that adhering Tcells also contribute to higher TEER. In-vivo, we induced systemic T-cell:MHC cognate interactions by i.v. injection of staphylococcal enterotoxin A (SEA). After i.p. injection of IFN-γ 48 h earlier, SEA administration induced greater vascular leakage in CD31-deficient mice in all organs analyzed except the brain (Fig. 1j). IFN-γ alone also increased vascular permeability in some organs, but this was minimal compared to SEA administration ( Supplementary Fig. 1g).
Integrity of intercellular junctions is a major determinant of permeability of the endothelium, and de-phosphorylation of VEcadherin-catenin complexes in adherens junctions mediate both anchorage and mechanical coupling of the cytoskeleton of adjacent EC including during leukocyte extravasation 1,2,25,26 .
The causative role of CD31-mediated signals on vascular integrity was confirmed by lentiviral transduction of cd31 −/− ECs with wild-type CD31-gene constructs (KO CD31WT ) or an empty plasmid (KO ploc.1 ), or two CD31-gene constructs with mutation with loss-of-function amino acid substitutions Y663F and Y686F in the ITIMs (KO CD31Y663F , KO CD31Y686F ECs) ( Supplementary  Fig. 2a, b). Reconstitution with an intact CD31 gene rescued the ability of cd31 −/− EC to recover barrier function upon MHCligation (Fig. 2f, g). This was also achieved by KO CD31Y663F but not KO CD31Y686F EC, suggesting that phosphorylation at aminoacid residue tyrosine 686, known to be essential for SHP2 recruitment by CD31 is instrumental for its ability to maintain barrier function 27  CD31-induced metabolic reprogramming regulates EBF response. The observations above do not explain the intense membrane-associated cytoskeleton reorganization and EC contraction induced by MHC and CD31 co-engagement. Although Src-induced phosphorylation of VE-cadherin prevents the binding of β-catenin, this is not sufficient to increase endothelial permeability 29,30 . EC metabolism has recently emerged as an important regulator of endothelial function [31][32][33] .
WT EC permeability following MHC-stimulation for 3 hours (i.e. when it would normally return to baseline) was dramatically increased by inhibition of glycolysis with the glucose analog 2deoxyglucose (2-DG), indicating that recovery of junction stability upon MHC-triggering might depend on a robust glycolytic response (Fig. 3a).
We therefore investigated the effect of MHC-and CD31mediated signals on EC metabolism. To avoid activation of metabolic pathways due to EC proliferation or migration these experiments were performed using confluent EC monolayers. Expression of the ubiquitous glucose transporter Glut1 and uptake of the glucose-analog 6-[N-7-nitrobemz-2-oxa-1,3-diazol-4-amino] −2-deoxyglucose (6-NBDG), which accumulates in the cytoplasm in its fluorescent form, were significantly increased in CD31competent, but not -deficient EC following-stimulation ( Fig. 3b-d,  Supplementary Fig. 3a). Importantly, 2-DG pre-treatment of EC abrogated 6-NBDG uptake irrespective of CD31 expression, suggesting a positive feed-back loop of glycolysis itself on glucose uptake. We further observed that cd31 −/− EC failed to redistribute Glut1 upon MHC-stimulation, suggesting a defect in Glut1 recycling to the cell membrane ( Supplementary Fig. 3b). Accordingly, ATP levels in response to MHC-triggering were significantly reduced in CD31-deficient EC (Fig. 3e). When stimulated with insulin, however, ATP levels were increased in CD31-deficient EC compared to WT EC. As insulin is a powerful inducer of glycolysis 34 , not supplemented in EC culture medium, CD31deficient EC are capable to engage the glycolytic pathway in response to insulin, but not MHC-stimulation. Further, the possibility that addition of insulin in the insulin-free medium used might attenuate the effect of CD31-deficiency rather than engaging a CD31-independent pathway is ruled out by the in vivo experiments (i.e. in the presence of physiological amounts of insulin) which show that MHC stimulation induces vascular leakage in the absence of CD31 signals (see Fig. 1h).
It has been shown that T-cells that preferentially utilize FAO, maintain fused mitochondria 35,36 . We, therefore, compared WT and CD31-deficient EC ability to oxidize glucose, glutamine and fatty acids (FAO). As shown in Supplementary Fig. 2a, glucose oxidation did not significantly differ in unstimulated WT and cd31 −/− EC. However, while glucose oxidation was increased in WT EC in response to MHC-signals, this pathway was impaired in CD31-deficient EC ( Supplementary Fig. 3d). Glutamine oxidation, in contrast, was utilized more by resting CD31deficient EC, which were however unable to increase it following MHC-stimulation ( Supplementary Fig. 3e). FAO was significantly enhanced by MHC-stimulated CD31-deficient but not WT EC ( Supplementary Fig. 3f). Accordingly, transcription of the mitochondrial enzyme carnitine palmitoyltransferase 1A (cpt1a),  Relative mRNA expression essential for FAO, was enhanced by MHC-stimulation in CD31deficient endothelium ( Supplementary Fig. 3g). The data obtained with CD31 Y686F (Fig. 2f, g) implicate SHP-2 activation in CD31-mediated barrier recovery along with previous observations 37 . We, therefore, examined the effect of pharmacologic inhibition of SHP-2 on the metabolic response of WT EC to MHC-stimuli. In steady-state conditions, EC monolayer permeability was increased by pre-treatment with a selective SHP1/ 2 small-molecule-inhibitor, without induction of cell death ( Fig. 5a, b). The glycolytic response to MHC-triggering was abrogated in confluent monolayers of CD31-expressing EC following SHP inhibition (Fig. 5c, d). An even greater effect was observed when CD31-deficient EC were analyzed (Fig. 5e, f). In these experiments, the glycolytic pathway was impaired also in unstimulated EC, suggesting that additional SHP phosphatases might contribute to sustain glycolysis in quiescent endothelium. Mitochondrial respiration was not affected by SHP-inhibition in quiescent WT or CD31-deficient EC, suggesting that the defect in OXPHOS observed in cd31 −/− EC is not caused by lack of SHP activation. SHP inhibition did not modify OCR in MHCstimulated WT EC (Fig. 5g, h), while it was further increased in cd31 −/− endothelium (Fig. 5i, j). Thus, SHP activation is required for glycolysis while it does not affect mitochondrial respiration.
CD31 signals inhibit FoxO1 nuclear-translocation. Cellular ability to activate glycolysis depends on the availability of glycolytic enzymes 31,33 . We, therefore, investigated expression of enzymes of the glycolytic pathway known to be functional in the endothelium before and after MHC-stimulation of WT and CD31-deficient EC.
Reduced Aldolase expression by CD31-deficient EC reflected lack of increased transcription upon MHC-triggering, suggesting that CD31-signals promote a transcriptional program (Fig. 6c). Further, in WT, but not cd31 −/− EC, aldolase localized in areas of actin remodeling, including the lateral borders, suggesting a role for this enzyme in the maintenance of actin anchorage to intercellular junctions ( Supplementary Fig. 4a).
The forkhead box O (FoxO) transcription factors control fundamental cellular processes, including metabolism. FoxO1 is a negative regulator of vascular growth 32 Depending on its phosphorylation status, FoxO1 shuttles between nucleus and cytoplasm. When localized to the nucleus, FoxO1 modulates gene transcription by binding to response sequences located in the promoter. We have previously shown that CD31-induced AKTmediated phosphorylation inhibits FoxOs by preventing their nuclear localization 39 .
β-catenin contributes to metabolic responses to MHC-signals. In addition to stabilizing adherens junctions, β-catenin, a member of the Wnt signaling cascade, delivers signals to the nucleus, influencing metabolic transcriptional programs 42 .
Dephosphorylated β-catenin can translocate into the nucleus where it promotes transcription of target genes, including cMyc 42 . Phosphorylation of β-catenin leads to ubiquitination and degradation via the proteasome 43 . CD31-signals stabilize β-catenin by preventing its phosphorylation induced by MHC-engagement (see Fig. 2b, c). To investigate whether β-catenin participates to metabolic reprogramming of EC following MHC-stimulation, localization of βcatenin in MHC-stimulated EC was investigated. MHC-ligation of WT EC enhanced β-catenin nuclear translocation leading to a concomitant increase of cMyc expression in the nucleus (Fig. 7a-c). This event did not occur in CD31-deficient EC, indicating a causative role for CD31-generated signals in the nuclear shuttling of Fig. 3 CD31 regulates EBF via its effects on EC metabolism. a EC were pre-treated with the glucose analog 2-DG (5 mM, 2 h) then MHC-1-or ICAM-stimulated for 4 h prior to TEER measurements. The bar graph shows the mean of measurement collected in three separate experiments of identical design ± SD. One-way Anova with Tuckey post-hoc test. 2DG MHC vs all ***p = 0.0003. b Glut-1 expression by EC following MHC stimulation (2 h) was evaluated by qRT-PCR. Data are mean ± SD. Two-tailed Student's T test. MHC vs IsC *p = 0.0412. c, d Representative histograms of antibody-stimulated WT or CD31-deficient EC incubated with 6-NBDG for 2 h prior to analysis. 2-DG was used as a negative control. The graph shows the mean MFI measured in three independent experiments ± SD. One-way Anova with Tuckey post-hoc test. WT IsC vs WT MHC **p = 0.003, cd31 −/− IsC vs cd31 −/− MHC ns = 0.058. e ATP levels were measured in WT and CD31-deficient EC 4 h after MHC or insulin (1.8 µM) stimulation. Data are mean ± SD. Two-tailed Student's T test. WT MHC vs cd31 −/− MHC ***p = 0.0002, WT insulin vs cd31 −/− insulin **p = 0.0062. Extracellular acidification rate (ECAR) of unstimulated and antibody-stimulated WT and CD31-deficient EC is shown in (f-h). The basal and maximal glycolysis and the glycolytic reserve are shown in (i-k). Wells were first injected with anti MHC antibodies at the indicated time point. Isotype-matched and secondary antibodies were used as controls. β-catenin upon MHC-triggering. In CD31-deficient EC, nuclear localization of β-catenin and increased cMyc expression was restored following pharmacological Akt activation.
We then analyzed the transcription of both cMyc and its target aldolase upon MHC-stimulation in both WT and CD31-deficient ECs. cMyc transcription was increased as early as 30 min after stimulation, peaked at 2 h and returned to baseline by 4 h (Fig. 7d). Accordingly, aldolase transcription was also increased by 30 min, but continued to remain elevated up to 4 h after stimulation (Fig. 7e). Stimulation of cd31 −/− EC did not induce transcription of any of the genes analyzed.
Inducing CD31-independent glycolysis restores EC barrier. We next investigated whether this pathway is essential to the recovery of EC barrier function following MHC-stimulation by directly activating this metabolic response independently of CD31-signals.
First, as Akt activation is instrumental to couple CD31 signals with metabolic reprogramming of EC towards glycolysis, we assessed the effect of pharmacological Akt activation on the glycolytic flux in MHC-activated CD31-deficient endothelial cells. As shown in Fig. 8a and Supplementary Fig. 6a-c, direct Akt activation could bypass the lack of CD31 signals and enhance cd31 −/− EC glycolysis in response to MHC-ligation.
To confirm the ability of AMPK-induced glycolysis in recovering EBF independently of CD31-signals and exclude potential off-target effects of Metformin and CD31-deficiency in leukocytes 13 , we generated bone marrow (BM) chimeras in which WT BM was administered to sublethally irradiated cd31 −/− recipients (CD31 selectively-expressed by EC). Some mice were treated with the high affinity AMPK-selective small molecule allosteric activator 991 51 . Treatment with AMPK activator restored cd31 −/− EC glycolytic response to MHC-stimulation (Fig. 8g, h), TEER recovery in MHC-stimulated cd31 −/− EC invitro (Fig. 8i) as well as barrier response following systemic MHC-stimulation in cd31 −/− mice (Fig. 8j).
Finally, we investigated the physiopathological relevance of our findings both in mice and humans.
To confirm that a pathway initiated by CD31-signals and inducing glycolysis is physiologically relevant to the recovery of microvascular integrity following MHC-triggering by migrating T lymphocytes, we assessed the EBF and histopathological features of rejection of male-derived CD31 −/− skin grafts by WT female recipients. Re-establishment of endothelial continuity between recipient and graft vasculature in HY-mismatched skin graft combinations is established by 14 days after grafting, while T-cell dependent rejection becomes clinically evident at day 28-35 and it is accelerated in CD31-deficient skin 39,52 .
As shown in Fig. 8k, l, CD31-deficient, male-derived skin was grafted onto female WT littermates and, as a control, WT females received CD31-deficent or WT, female-derived skin. Some mice were treated with Metformin (+Met), or vehicle alone. Two weeks later, mice were injected with Evans Blue i.v. and tracer extravasation in the grafts was measured. As shown in Fig. 8k, CD31-deficient male-derived skin grafts displayed significantly higher vascular leakage compared to WT-derived grafts. Importantly, female-derived skin graft vasculature appeared functionally intact irrespective of CD31 expression, confirming that the leakage was caused by MHC:HY complexes ligation by allospecific T-cells. Metformin-treatment significantly reduced tracer extravasation in CD31-deficient male-derived skin grafts. Comparison of histological features revealed substantial, proteinrich papillary dermal edema-as demonstrated by the large quantity of eosinophil material deposed between cells, indicative of severe microvascular leakage-in CD31-deficient compared to WT male-derived skin grafted onto female WT recipients (Fig. 8l). Edema in CD31-deficient grafts was reduced by treatment with Metformin. WT and CD31-deficient, female-derived skin graft did not display pathological features.
In humans, we correlated the effect of loss of CD31 expression and vascular leakage by comparing soluble CD31 (sCD31) molecules in the serum of patients with sepsis and septic shock (Supplementary Tables 1 and 2, respectively) 53 . CD31-deficient mice have severely reduced survival to LPS-induced endotoxic shock, associated with enhanced vascular permeability 16,17 . In humans, CD31 is shed from endothelium activated by strong proinflammatory stimuli 54,55 , and sCD31 levels have been correlated with sepsis severity 56 . As shown in Supplementary Fig. 6i, serum levels of sCD31 were significantly elevated in patients with septic shock compared with those of sepsis patients, consistent with the hypothesis that loss of CD31 expression by the endothelium undermines its ability to preserve vascular integrity.

Discussion
By modeling endothelial cell contraction and vascular leakage induced by MHC-signals, we show that recovery of endothelial integrity involves a complex integration of Akt-dependent molecular events, including nuclear exclusion of FoxO1, βcatenin stabilization and nuclear translocation, and induction of cMyc expression, leading to engagement of the glycolytic flux (summarized in Fig. 9). These events likely reflect the requirement for ATP-sustained actin remodeling to re-establish junctional anchorage to cytoskeletal components. The dramatic metabolic impairment due to loss of CD31 expression selectively applies endothelial adaptation in response to stimuli that induce EC contraction and barrier breach. Previous studies have demonstrated that endothelial migration and proliferationwhich also involve intense cytoskeleton reorganization-are dependent on a similar metabolic response 31,32 . Other receptors on CD31-deficient EC can induce metabolic reprogramming-as vascular functions such as angiogenesis and endothelial proliferation, which are highly dependent on glycolysis, appear to be normal in these mice.
Our observations suggest that MHC-signals per se can induce stress fiber formation independently of CD31-induced signals, Fig. 4 CD31-deficient EC increase mitochondrial respiration upon MHC stimulation. Oxygen consumption rate (OCR), basal, maximal respiration, and spare capacity of unstimulated WT and CD31-deficient EC is shown in (a, b), respectively (N = 3 independent experiments). The error bars represent SD. One-way Anova with Tuckey post-hoc test. Basal respiration WT vs cd31 −/− ****p < 0.0001, max respiration WT vs cd31 −/− ****p < 0.0001, spare capacity WT vs cd31 −/− ****p < 0.0001. Oxygen consumption rate (OCR) of unstimulated and MHC-stimulated WT and cd31 −/− EC is shown in (c, d), respectively. The maximal respiration and spare capacity are shown in the right-had-side of each panel. Wells were first injected with anti MHC antibodies at the indicated time point. Isotype matched and secondary antibodies were used as controls. Further injections followed at the time point indicated (arrows) introducing the indicated compounds into the wells. N = 3 independent experiments. The error bars represent SD. One-way Anova with Tuckey post-hoc test. d Rotenone injection IsC vs Rotenone injection MHC *p = 0.04, Max respiration IsC vs MHC **p = 0.002, spare capacity IsC vs MHC *p = 0.0327. e EM analysis of EC mitochondria and mitochondrial cristae of WT and cd31 −/− EC MHC-or IsC-stimulated over 4 h. Scale bar = 500 nm, represents 2 experiments. The mean transverse diameter and number of mitochondria measured in 15 images from each experimental group is shown in (f, g), respectively. In h maximum cristae width was measured from 15 images from each experimental group using ImageJ software and it is expressed in arbitrary units, Bar graphs depicts mean ± SEM (N = 3 independent experiments). One-way Anova with Tuckey post-hoc test. We further show that SHP-2 activation by CD31 signals is instrumental to the recovery of junctional stability, through the initiation of Akt-dependent metabolic reprogramming required for optimal cytoskeletal reorganization and junction re-assembly.
The role of CD31 in the maintenance of junctional stability has been attributed to mechanical bridging of EC and junction stabilization by de-phosphorylation of β-catenin and VE-cadherin 57 . Our observations rather support a prominent role for this receptor in the initiation of a signaling pathway culminating in the production of energy to sustain actin remodeling and junction annealing.  Akt activation is central to this function of the CD31 receptor. The two major mediators, which define the contribution of Akt signaling in metabolic reprogramming leading to junction stabilization, are β-catenin and FoxO1. FoxO1 is prevented from translocating to the nucleus after phosphorylation by Akt 58 . Concomitantly, Akt inactivates GSK-3β, which sequester βcatenin in the cytoplasm 59 leading to β-catenin stabilization and nuclear translocation 60 . Ultimately, CD31 maintains the efficiency of the glycolytic machinery in EC by promoting βcatenin-mediated c-Myctranscription, which in turn promotes expression of the glucose transporter Glut-1 and Aldolase A 40 .
Interestingly, in the absence of CD31 signals, PFKFB3 protein expression by EC is increased, suggesting that the regulation of this enzyme transcription is cMyc independent. PFKFB3 gene expression has been shown to be inhibited by shear stress through Kruppel-like factor-2 (KLF2) in EC 61 , and was enhanced by HIF Serum Response Factor in various tumor cell lines 61 , but has never been associated with cMyc-induced transcription.
Unlike what was observed in human EC transduced with a gain-of-function FoxO1 allele rendering FoxO1 constitutively nuclear 32 , our data indicate that enhanced FoxO1 nuclear translocation leads to increased oxidative metabolism, probably fueled by fatty acid oxidation (FAO). The reason for this discrepancy is at present unclear, but could be related to phosphatase targets other than the FoxO1 pathway in our system. It is worth noting that the identity of phosphatase enzyme(s), which dephosphorylate FoxO1 is at present unknown. In addition, the pattern of FoxO phosphorylation can differentially affect its activities 62 and it is possible that different phosphorylation patterns might reflect a stimulus-specific modulation which leads to stimulus specific-effects.
An unexpected feature of CD31-deficient EC is their severely compromised mitochondrial respiration associated with a large number of enlarged mitochondria. It has been suggested that in response to stressors, mitochondria become enlarged decreasing the rate of oxygen consumption, leading to decreases in the rate of ROS production and ATP 63 . Relevant to the focus of this study, metformin-induced AMPK activation restores the EBF and glycolysis without affecting mitochondrial respiration, suggesting that oxidative phosphorylation is not required for restoring endothelial continuity after contraction and barrier breach. In addition, SHP inhibition does not affect OCR by WT ECs, thus ruling out a role for this CD31-signaling mediator. Understanding the molecular mechanism and the functional consequences of the defective oxphos in CD31-deficient EC will require further investigations. Of note, oxidative phosphorylation is plays a role in sustaining endothelial cell division 64 and attenuation of vascular development has been observed in CD31deficient mice 65 .
We further show that, unlike MHC-stimulation, exposure to insulin results in an increased ATP levels in CD31-deficient EC, suggesting that CD31-deficient EC are capable to engage the glycolytic pathway in response to insulin receptor, but not MHCstimulation. This also implies that the metabolic alterations of CD31-deficient EC might become apparent only as a result of selected signals, such as MHC-triggering or histamine, and that insulin receptor signals are not affected by CD31 activity. The enhanced ATP production in response to insulin by CD31deficient EC compared to WT also indicates that compensatory mechanisms are in place to compensate for the severe deficiency in mitochondrial respiration in these cells.
The contribution of CD31 signals in the maintenance of barrier integrity has long been contentious 57 . In inflammatory conditions, the impact of CD31 activity on junction stability during leukocyte migration appears to be stimulus-specific 66 . Similarly, endothelial contraction is not required for IL-1β-induced leukocyte migration 67 . Genetic deletion of the actin nucleationpromoting factor cortactin in mice is associated with reduced neutrophil recruitment but increased vascular permeability invivo 68 , thus demonstrating that transendothelial migration and vascular leakage can be associated, but not necessarily coupled. In agreement with these reports, our data show that ICAM-1mediated signals do not result in endothelial contraction and temporary vascular leakage, while suggesting that metabolic adaptation by EC is paramount to the barrier response to severe 'breaching' signals, such as those induced by histamine, thrombin and MHC-ligation. Physiologically, this would reflect a prominent role of CD31 signaling during inflammatory conditions characterized by high endothelial contractility (histamine, MHCtriggering) leading to potential vascular damage. CD31-induced metabolic responses appear to be largely dependent on a specific transcriptional program. While our fluxometry studies show that an increased glycolytic rate immediately follows MHC-stimulation of WT EC and MHCstimulation rapidly induces translocation to Glut1 to the endothelial surface, full re-establishment of endothelial integrity invivo (see Fig. 1h, i) requires~2-3 h. Such a prolonged energy requirement is likely to require, as we show, a rapid and sustained transcriptional reprogramming leading to increased expression of glycolysis-promoting genes such as cMyc and aldolase.
In further support of the requirement of endothelial metabolic adaptation in inflammatory conditions characterized by severe barrier breach, treatment with Metformin has been known to promote favorable outcome in sepsis, a condition characterized Fig. 5 The metabolic response to endothelial barrier breach is src-phosphatase-dependent. Confluent EC monolayers were MHC-stimulated and treated with the selective SHP1/2 inhibitor (50 μM for 3 h) or vehicle as a control. a The bar graph shows the mean TEER values measured in three independent experiments. b Percentage of EC viability following treatment with SHP1/2 inhibitor determined by the trypan blue exclusion assay in three independent experiments. c-f ECAR measured in WT (c, d) or cd31 −/− EC (e, f) following MHC antibody-stimulation with or without pre-treatment with the SHP1/2 inhibitor (50μM for 45 min). Basal and maximal glycolysis and glycolytic reserve are shown in (d, f). g-j OCR measured in WT g-h or CD31-deficient EC I, j following MHC antibody-stimulation with or without pre-treatment with the SHP1/2 inhibitor. Basal and maximal respiration and spare capacity are shown in (i, j). (N = 3 independent experiments). Data are shown as mean ± SEM; significant differences were determined using one-way Anova with Tuckey post-hoc test. a IsC vehicle vs MHC vehicle ****p < 0.0001, MHC vehicle vs MHC + SHP inhibitor ****p < 0.0001; c Glu/Oligo injection WT MHC vs all ****p < 0.   by systemic endothelial leakage [69][70][71] . Similarly, inhibition of Abl family kinases, which phosphorylate β-catenin, attenuates vascular leakage induced by thrombin, histamine, vascular endothelial growth factor (VEGF), lipopolysaccharide (LPS), and oxidative stress [72][73][74][75][76] . The present study provides a molecular basis for these clinical observations. Thus, in-depth knowledge of endothelial metabolic adaptation in health and disease, and of the endothelial receptors regulating metabolic reprogramming might provide novel targets for the prevention and therapy of diseases characterized by endothelial dysfunction.

Methods
Mice. cd31 −/−77 and wild type (WT) male and female mice were bred in house in SPF conditions and used at the age of 8-10 weeks. All in-vivo experiments were conducted with strict adherence to the Home Office guidelines (PPL P71E91C8E) following approval by the Queen Mary University of London Ethics committee. The number of animals required to obtain statistical significance was estimated based on similar studies previously performed. Animals were not randomized, and no blinding was done. Antibody-mediated EC activation. Monolayers of WT EC were stimulated with 1 μg/ml mouse-anti-mouse H-2Ld/H-2Db (BD Biosciences 14-5999-85) or isotype control and 0.5 μg/ml goat-anti-mouse IgG (BioLegend 405301) with and without coligation of CD31 molecules (rabbit-anti-mouse CD31 5 μg/ml plus goat-antirabbit Ig 2.5 μg/ml). In addition, WT EC underwent MHC class 1 molecule stimulation in the presence or absence of ICAM-1 ligation (rat-anti-mouse CD54 (eBioscience 14-0549-82) 5 μg/ml plus goat anti-rat Ig 2.5 μg/ml (Biolegend 400202).

Patients
Measurement of trans-endothelial electrical resistance (TEER). EC were grown on Transwell polycarbonate filters (pore size, 0.4 µm; Sigma-Aldrich, CLS3396) first coated with calfskin collagen type I (Sigma-Aldrich, C3867) and bovine plasma fibronectin (Sigma-Aldrich, F1141). Transendothelial resistance across the monolayer was determined by using an Endohmeter (World Precision Instruments, Sarasota, FL) stabilized at 148 ± 12 Ω. Resistance from coated cell-free inserts was always subtracted from the resistance obtained in the presence of endothelial cells.
ATP assay. ATP production by EC was measured using the ATP Assay Kit (Abcam, ab83355). 10 6 WT and CD31 deficiency EC with or without treatments were harvested and homogenized, centrifuged to remove insoluble material, and the supernatant collected for use in subsequent steps according to the manufacturer's instructions.
Widefield deconvolution fluorescence microscopy. EC were cultured in DMEM medium and fixed with 3.7% formaldehyde. After fixing, they were stained with the antibodies indicated in the figures. Coverslips were extensively washed, air dried, and mounted in Vectorshield mounting medium for fluorescence with DAPI (Vector Laboratories, H-1800) on glass slides. Cells were visualized using a Zeiss Z1 fluorescence microscope (Carl Zeiss) equipped with an AxioCam MRm cooled monochrome digital camera and an ApoTome.2 imaging unit. Images were acquired using a Plan Apochromat 10×/0.8 NA objective. Standard epiilluminating fluorescein and rhodamine fluorescence filter cubes were used, and 12-bit image datasets were generated using Axiovision software version 4.8.
The thermal cycling profile for amplification was 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 54°C for 1 min. Amplification was 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. To ensure the amplification specificity, the melting curve program was set as follows: 95°C for 15 s, 60°C for 1 min, and 95°C for 15 s, right after the PCR cycles.
The qPCR data were analyzed using the delta CT method by taking the CT values of the genes of interest from the house keeping gene following by normalization to the wildtype control sample.
Cellular Fractionation and Isolation of Nuclei. Cells (5-10 × 10 6 ) were cultured in 60-mm diameter culture dishes until~80% confluency. For isolation of nuclear extracts, the cells were then collected into microtubes, centrifuged for 20 s in a microcentrifuge, and resuspended in 200 μl of 10.0 mM Hepes (Gibco, 15630080), pH 7.9, containing 10.0 mM KCl, 1.5 mM MgCl 2 , and 0.5 mM dithiothreitol. After incubation at 4°C for 15 min, the cells were lysed by passing 10 times through a 22gauge needle. Next, the cells were centrifuged for 20 s in a microcentrifuge, and the supernatant, cytoplasmic fraction was collected and frozen in small aliquots. The pellet, which contained the nuclei, was resuspended in 150 μl of 20 mM Hepes, pH 7.9, containing 20% v/v glycerol, 0.1 mM KCl, 0.2 mM EDTA (Invitrogen, AM9912), 0.5 mM dithiothreitol (Thermo Scientific, A39255), and 0.5 mM phenylmethanesulfonyl fluoride (Sigma Aldrich,10837091001) and then stirred at 4°C for 30 min. The nuclear extracts were then centrifuged for 20 min at 4°C in a microcentrifuge. The supernatant was collected, aliquoted into small volumes, and stored at −80°C.
Western blotting. Nuclear or whole-cell proteins derived from each cell sample were fractionated by SDS-PAGE, blotted onto a nitrocellulose membrane (Whatman, WHA10402506). Membranes were blocked for 2 hours at room temperature in 5% milk/TBS-Tween 20 and were incubated overnight at 4°C with the primary antibodies listed in Reagents and Antibodies in the main text and subsequently with HRP-conjugated secondary antibody (1:5,000; Amersham Bioscience). Films then were scanned, and the intensity of the bands was quantified using ImageJ Software v.1.37c (NIH).
In-vitro 6NBDG uptake assay. Freshly isolated EC were washed in PBS and resuspended in glucose-free cell medium (Gibco, Cat11879-020) containing various mentioned 30 6-NBDG (Life Technologies, CatN23106) in glucose free cell medium was then added to the cells and the cells were further incubated for an additional 10-15 min. Finally, the cells were washed twice with warm PBS and resuspended in flow cytometry buffer and placed on ice. Immediate analysis was performed using flow cytometry to observe fluorescence uptake by the EC.
Measurement of ECAR and OCR. Real time bioenergetics analysis of extracellular acidification rates (ECAR) and oxygen consumption rates (OCR) of T-cells subjected to antibody stimulation was performed using the XF analyzer (Seahorse Biosciences). T-cells were cultured in serum free, unbuffered XF assay medium (Seahorse biosciences, Cat 102365-100) for 1 h. The cells were then seeded (6 × 10 5 / well) into the seahorse XF24 cell plates for analysis. Perturbation profiling of the use of metabolic pathways by T-cells was achieved by the addition of oligomycin (1 μM), FCCP (1 μM), Antimycin A (1 μM), rotenone (1 μM), D-glucose (10 mM), 2-deoxy-D-glucose (2DG, 50 mM; all from Seahorse biosciences, Cat# 103020-100 and 103015-100). Experiments with the Seahorse system were done with the following assay conditions: 2 min mixture; 2 min wait; and 4-5 min measurement. Metabolic parameters were calculated by Wave v2.4.1 Software. Experiments were done in at least triplicate wells.
Mitochondrial oxidation of glucose, glutamine, and fatty acid oxidation. Oxidation rates of glucose, glutamine, and FAO were measured using an XFp Extracellular Flux Seahorse Analyzer (Seahorse Bioscience). WT and CD31 deficiency cells (15,000/well) were treated with MHC triggering for 45 min prior to the start of the XFp Mito Fuel Flex Test kit, which was performed in accordance with manufacturer′s instructions. Each plotted value is the mean of at least 6 replicates and is normalized to Hoechst signal in each well.
Flow cytometry. Cells were suspended in FACS buffer (PBS, 1% BSA, 0.01% sodium azide), stained with the appropriate concentration of fluorescenceconjugated antibodies, or isotype control antibodies, according to the manufacturer's instructions, fixed in fix buffer (PBS, 4% paraformaldehyde, 1% FCS), and analyzed by a FACSAria cell sorter (Becton Dickinson). Acquired samples were analyzed using FlowJo 7.6 software (Tree Star, Inc.).  (1). Erk phosphorylation is modulated by CD31 signals, possibly via SHP-2 (2). MHC signals induce CD31 ITIM phosphorylation and SHP-2 recruitment. SHP-2 prevents the phosphorylation of b-catenin (5) and VE-cadherin (6), thus stabilizing the junctional complex. In addition, dephosphorylated bcatenin can transfer to the nucleus where it induces cMyc transcription. In parallel, SHP-2 induces AKT activation which in turn inhibits FoxO1 nuclear translocation, thus preventing inhibition of cMyc transcription. This leads to enhanced transcription of glycolysis enzymes and enhanced glycolysis required for actin remodeling and maintenance of junctional anchorage.
Measurement of endothelial permeability in-vivo. Vascular permeability was quantitatively evaluated by extravasation of Evans blue as a marker of albumin extravasation. Briefly, mice received anti-MHC and secondary cross-linking antibody (0.67 μg and 0.33 μg/kg body weight, respectively) or anti-ICAM1 and secondary cross-linking antibody (3.35 μg and 1.7 μg/kg body weight, respectively) before Evans blue dye (2 mg/kg) i.v. injection. 45 min later. As indicated, some MHC-stimulated mice received either an Akt activator (administered i.p. at a dose of 7 mg/kg) or Metformin (i.p. 125 mg/kg) before administration of the antibodies. WT and cd31-/-mice from treatment and control groups were deeply anesthetized with chloral hydrate, and blood was obtained by cardiac puncture. After the mice were sacrificed, organs were collected, weighted and incubated in formamide (1 ml/ 100 mg) for 48 hours at 56°C to extract Evans blue. Absorbance at 600 nm was normalized by tissue weight, and by the relative concentration of Evans blue present in the blood of the corresponding mice.
Skin grafting. Skin grafting was conducted using a method previously described by Billingham and Medawar 79 using tail skin grafted onto the lateral thorax. Briefly, donor tail skin was removed and cut into 1 cm 2 sections. A piece of skin was removed from the right flank of anesthetized recipient mice to create a graft bed and a 1 cm 2 piece of tail skin was placed in the graft bed. The graft was covered with muslin and a plaster cast was then wrapped around the midriff and graft. Plasters were removed 7-10 days after grafting and grafts were inspected every other day.
Electron microscopy of resin-embedded cells. For high-pressure freezing suspension, cultured endothelial cells were harvested by filtering and immediately frozen in a high-pressure freezing apparatus (HPF010; Bal-Tec, Balzers, Liechtenstein). For subsequent freeze substitution, the material was kept at −85°C for 60 h before slowly being warmed to 0°C for a period of 18 h. Substitution was performed in an AFS freeze substitution unit (Leica, Bensheim, Germany). The sections were poststained with aqueous uranyl acetate/lead citrate, and images were captured with a Hitachi H7650 transmission electron microscope (Hitachi High-Technologies) operating at 80 kV.