PKM2 regulates endothelial cell junction dynamics and angiogenesis via ATP production

Angiogenesis, the formation of new blood vessels from pre-existing ones, occurs in pathophysiological contexts such as wound healing, cancer, and chronic inflammatory disease. During sprouting angiogenesis, endothelial tip and stalk cells coordinately remodel their cell-cell junctions to allow collective migration and extension of the sprout while maintaining barrier integrity. All these processes require energy, and the predominant ATP generation route in endothelial cells is glycolysis. However, it remains unclear how ATP reaches the plasma membrane and intercellular junctions. In this study, we demonstrate that the glycolytic enzyme pyruvate kinase 2 (PKM2) is required for sprouting angiogenesis in vitro and in vivo through the regulation of endothelial cell-junction dynamics and collective migration. We show that PKM2-silencing decreases ATP required for proper VE-cadherin internalization/traffic at endothelial cell-cell junctions. Our study provides fresh insight into the role of ATP subcellular compartmentalization in endothelial cells during angiogenesis. Since manipulation of EC glycolysis constitutes a potential therapeutic intervention route, particularly in tumors and chronic inflammatory disease, these findings may help to refine the targeting of endothelial glycolytic activity in disease.

Angiogenesis, the formation of new blood vessels from pre-existing ones, occurs in pathophysiological contexts such as wound healing, cancer and chronic inflammatory disease. In sprouting angiogenesis, endothelial cells (ECs) receive a stimulus (usually VEGFA) which triggers the angiogenic response by the selection and specification of the endothelial tip cell 1 . This cell migrates toward the stimulus followed by the stalk cells, defined by a balanced Dll4/VEGFA signaling 2 . A fine-tuned coordination of diverse cellular responses is necessary at this initial step, including reorganization of endothelial cell junctions, tip-cell migration to generate characteristic cytoskeletal protrusions called filopodia, and proliferation and lumenization by stalk cells. Impairment in any of these processes prevents productive angiogenesis and the formation of a functional vascular plexus [3][4][5] . A shared feature of all these endothelial cell responses is the need for energy in the form of ATP 6,7 .
The predominant ATP generation route in ECs is glycolysis 8 , providing both the ATP required for cell function and the intermediate metabolites that feed into other metabolic pathways for macromolecule biosynthesis 9,10 . During sprouting angiogenesis, glycolysis is boosted in the endothelial tip cell, and angiogenesis is blunted by the genetic deletion or pharmacological inhibition of glycolytic enzymes like PFKFB3 [11][12][13] . Recent dissection of the impact of distinct metabolic routes in endothelial cells has revealed that glycolysis impacts endothelial

PKM2 is required for sprouting angiogenesis in vitro and in vivo.
To characterize the role of PKM2 in ECs during sprouting angiogenesis, we specifically silenced PKM2 expression in human umbilical vein ECs (HUVEC) with an isoform-specific siRNA that decreased PKM2 expression without affecting PKM1 levels (Fig. 1A). In the spheroid culture system used to study sprouting angiogenesis, suppression of PKM2 expression was maintained up to the final time point (day 7) (Supplementary Figure S1A,B). Although sprouts developed a proper lumen in both groups (Supplementary Figure S1C), PKM2-silenced EC spheroids produced fewer and shorter sprouts than spheroids formed by control siRNA-transfected ECs (Fig. 1B-D). Silencing PKM2 also resulted in reduced number and shorter endothelial tubes in the 'hanging drop' assay, which runs for 24 hours (Supplementary Figure S1D,E). Moreover, sprouts formed by control and PKM2-silenced EC spheroids contained similar proportions of Ki67+ proliferating cells, and tip cells in these spheroids produced fewer filopodia ( Fig. 1E-H). Silencing PKM2 with an independent siRNA in mouse lung ECs also impaired sprout formation in the spheroids confirming the need of PKM2 for in vitro sprouting angiogenesis (Supplementary Figure S1F,G).
To determine whether PKM2 was also required for sprouting angiogenesis in vivo, P6 mice were given an intravitreal injection of PKM2 siRNA, and we examined the retinal vasculature after 72 h. Western blot of retina extracts showed reduced PKM2 protein expression, and immunofluorescence analysis confirmed decreased PKM2 in the vasculature ( Fig. 2A,B). PKM2-silencing in retina diminished radial vascular growth with no impact on vascular density (Fig. 2C-E). This prompted us to analyze EC number and proliferation. PKM2-silenced and control-siRNA treated retinas had similar densities of ERG+ cells and the same proportion of Ki67+ ECs ( Fig. 2F-I). Matching the in vitro analysis, filopodia number was lower in PKM2-silenced retinas (Fig. 2J,K). Together, these observations show that PKM2 is required for sprouting angiogenesis in vitro and in vivo by mechanisms that do not seem to involve EC proliferation.
PKM2 is located at VE-cadherin-expressing endothelial cell junctions. Although PKM2 was mostly found in the cytoplasm of ECs, high-resolution confocal microscopy images, image deconvolution and profile intensity analysis revealed however that a pool of PKM2 localized close to VE-cadherin and cortical F-actin at the junctions (Fig. 3A,B). We confirmed PKM2 vicinity to VE-cadherin at the endothelial junctions by proximity ligation assays (Fig. 3C). Moreover, in spite of similar PKM2 protein levels, PKM2 was barely observed at the junctions of VE-cadherin-null endothelial cells compared to VE-cadherin-positive ones (Fig. 3D,E). PKM2 was also found nearby the junctions of ECs stimulated with sphingosine-1-phosphate (S1P), an inducer of EC junction remodeling and reinforcement 33 (Supplementary Figure S2A), and of 3D EC sprouts in vitro (Fig. 3F), without significant differences in PKM2 presence at the junctions of tip and stalk cells (PKM2 MFI tip = 144.4 ± 22.82 and MFI stalk = 156.3 ± 20.3, n = 19 sprouts). PKM2 was also detected at F-actin-rich filopodia and lamellipodia of migrating ECs in 3D and 2D respectively (Supplementary Figure S2B,C), in line with the reported PKM2/F-actin association 34 .
In accordance with the immunofluorescence data, PKM2 was found by western blot in both cytosolic and membrane EC fractions and the specificity of this compartmentation was shown by reduced PKM2 abundance in the membrane fraction of PKM2-silenced ECs (Fig. 3G). Cell membrane subfractionation further revealed that PKM2 was present in low-density fractions containing plasma membrane and in endosomes (Fig. 3H). control; means ± SEM, n = 3, ns non-significant, **p < 0.01 by unpaired Student t-test. (B) Bright-field microscopy images of spheroids coated with HUVECs transfected with control or PKM2 siRNA and embedded in fibrin gels for 7 days. Scale bar, 10 µm. (C) Sprout length in 3D spheroids; means ± SEM, n = 103 and 38 spheroids formed by control and PKM2 siRNA-silenced cells from one representative experiment of five performed, ****p < 0.0001 by unpaired Student t-test. (D) Sprout numbers in 3D spheroids; means ± SEM, n = 27 and 14 spheroids formed by control and PKM2 siRNA-silenced cells from one experiment representative of five performed, **p < 0.01 by unpaired Student t-test. (E) Immunofluorescence of Ki67 (red, proliferation) and Hoechst (blue, nuclei) in 3D spheroid sprouts. Scale bar, 10 µm. (F) Percentage of Ki67-positive cells per sprout in 3D spheroids; means ± SEM, n = 3 independent experiments, ns non-significant by paired Student t-test. (G) Immunofluorescence of F-actin in 3D spheroid sprouts. Scale bar, 10 µm. (H) Filopodia number in 3D spheroids; means ± SEM, n = 13 and 15 filopodia in sprouts formed by control and PKM2 siRNA-silenced cells from one representative experiment of five performed, ***p < 0.0001 by Welch´s test. MW, molecular weight. See also Figure S1.
In spite of no change in total VE-cadherin protein expression, PKM2-silenced ECs showed lower cell-junction VE-cadherin intensity than control ECs, together with a higher proportion of discontinuous junctions at baseline and after S1P treatment (Fig. 3I-M and Supplementary Figure S2D-F). Junctions formed by PKM2-silenced ECs also showed a clear decrease in plakoglobin, a marker of stable VE-cadherin cell-cell contacts ( Fig. 3N and Supplementary Figure S2G). Indeed, PKM2-silencing with an independent siRNA confirmed a similar altered junctional VE-cadherin pattern in mouse lung ECs (Supplementary Figure 2H-J). Finally, PKM2-silenced cells showed a weakened response to S1P-driven reinforcement of the EC barrier, as indicated by lower transendothelial electrical resistance (TEER) values (Fig. 3O). These data indicate defects in VE-cadherin-mediated EC barrier in the absence of PKM2 in remodeling conditions.
A recognized function of VE-cadherin at cell-cell junctions is to conduct polarity signals while maintaining barrier integrity during collective migration 36 . In line with the observed junctional defects, PKM2-silenced HUVECs were significantly less efficient than control siRNA-transfected ECs at closing the wound area in the scratch assay without showing differences in cell proliferation ( Fig. 4A-D). Accordingly, inhibition of proliferation did not change the defective wound closure and PKM2 silencing did not affect EC basal growth and proliferation (Supplementary Figure S3A-D). We next assessed the role of PKM2 in the orientation and directionality of migrating EC cells by quantifying Golgi apparatus polarization and by time-lapse microscopy 37,38 . PKM2-silenced migrating EC cultures showed more discontinuous junctions and this correlated with a higher proportion of cells polarized against the wound than control cultures ( Fig. 4E-G). Time-lapse recording and single-cell track analysis confirmed that PKM2-silenced ECs migrated faster and for longer linear (Euclidean) distances opposite to the wound but for shorter distances and less directional toward the wound compared with controls ( Fig. 4H-J and Movies S1-S2). Moreover, PKM2 silencing impaired collective migration of adjacent endothelial cells (Fig. 4K).
PKM2 is required for endothelial cell junction remodeling during angiogenic sprouting in vitro and in vivo. In addition to maintenance of junction stability and to collective migration in 2D, VE-cadherin dynamics are also essential for functional EC arrangements during 3D sprouting angiogenesis, and this manifests as a balanced distribution of stable and unstable EC junctions 35 . To determine the role of PKM2 in regulating this mosaicism, we quantified the distribution of EC junction patches by classifying the morphology of EC junctions based on their VE-cadherin pattern, from discontinuous (active/unstable) to continuous (inactive/stable) according to the reported 'patching algorithm' 35 . The 3D sprouts formed in vitro by PKM2-silenced human ECs showed an altered EC-junction patch distribution, with a higher abundance of patches containing active/unstable junctions than 3D sprouts formed by control siRNA-transfected ECs; this was reflected by a change in the active/ unstable:inactive/stable patch ratio ( Fig. 5A-D). This phenotype was also confirmed in vivo in the vasculature of PKM2-silenced postnatal retinas ( Fig. 5E-G).

PKM2 activity is involved in the regulation of EC junctions and sprouting angiogenesis.
In addition to its action in glycolysis, PKM2 has been linked to non-glycolytic functions 29,30 . To determine whether the effects of PKM2 absence on EC junctions and sprouting angiogenesis are related to pyruvate kinase activity, we treated ECs with shikonin, a specific inhibitor of the PKM2 isoform 39 . Pharmacological inhibition of PKM2 activity increased the number of discontinuous junctions in basal and S1P-treated human ECs (Supplementary Figure S4A,B) and recapitulated the rest of effects of PKM2-silencing. Shikonin reduced the number and length of sprouts in the 3D spheroid assay and impaired wound closure and EC collective migration without influencing EC proliferation (Supplementary Figure S4C-G and Movies S3-S4). In contrast, the general glycolysis inhibitor 2-deoxyglucose (2-DG), in addition to impairing junction stability and EC migration, also decreased EC proliferation (Supplementary Figure S4H-K). PKM2 activity was also required in vivo since intravitreal injection of shikonin reduced vascular radial growth and filopodia number in P6 mice while having no effect on vascular density, EC number, or the percentage of proliferating ECs, thus reproducing the phenotype observed in the vasculature of PKM2-silenced retinas (Supplementary Figure S5A PKM2-silencing decreases ATP nearby EC junctions. Given the requirement of PKM2 kinase activity for its actions on EC junctions and sprouting angiogenesis, we next assessed the impact of PKM2 on EC ATP production. Analysis of the global metabolic state revealed that PKM2-silenced ECs had a lower glycolysis flux than control siRNA-transfected cells, revealed by a relative ~20% drop in the extracellular acidification ratio (ECAR) (Fig. 6A). This was reflected in significantly lower total ATP production (Fig. 6B). Mitochondrial area in P6 mouse retinas 72 hours after intravitreal siRNA-injection, means ± SEM, n = 4 mice per condition, ns non-significant by unpaired Student t-test. (H) Immunofluorescence of isolectin B4 (blue, vessels), Ki67 (green, proliferation), and ERG (red, endothelial cell nuclei) in whole-mount P6 mouse retinas 72 hours after intravitreal siRNA-injection. Scale bar, 10 µm. (I) Percentage of Ki67-positive cells per total ERG-positive cells in P6 mouse retinas 72 hours after intravitreal siRNA-injection; means ± SEM, n = 4 mice per condition, ns non-significant by unpaired Student t-test. (J) Immunofluorescence of isolectin B4 (white, vessels) in wholemount P6 mouse retinas 72 hours after intravitreal siRNA-injection. Each yellow asterisk marks one filopodia. Scale bar, 10 µm. (K) Number of filopodia per 100 μm of vascular front in P6 mouse retinas 72 hours after intravitreal siRNA-injection; means ± SEM, n = 4 mice per condition, *p < 0.05 by unpaired Student t-test. MW, molecular weight. www.nature.com/scientificreports www.nature.com/scientificreports/ activity remained intact, with similar basal and maximal oxygen consumption ratios (OCR) in PKM2 and control siRNA-silenced human ECs (Fig. 6C,D).
Following the protocol reported for other glycolytic enzymes 16,20 , we next performed sequential cell lysis with increasing proportions of saponin to allow the extraction of two fractions, one enriched for cytosol and the other less soluble for cytoskeleton and EC junctions (CSK-junction enriched) (Fig. 6E,F). Pyruvate kinase activity was ~1.8-fold higher in the CSK-junction enriched sub-fraction than in the cytosol; specificity for PKM2 was confirmed by decreased PK activity in both compartments when PKM2 expression was silenced (Fig. 6G).
To assess whether PKM2-silencing impaired local ATP levels, particularly at CSK-junction compartment, we quantitated ATP in saponin-extracted sub-fractions. PKM2-silencing in mouse and human ECs significantly decreased ATP production in the cytosol, by ~50%, and the CSK-junction enriched sub-fraction, by ~40% (Supplementary Figure S6A and Fig. 6H), in accordance with the reduced PK activity ( Fig. 6G). To ensure quantification of ATP levels nearby EC junctions, we transduced HUVECs with lentivirus encoding the FRET-based ATP sensor GO-ATeam1 40 . PKM2 silencing resulted in a significant reduction in the GO-ATeam1 FRET signal at both cytosol and junctions (as in subcellular fraction analysis), but the ratio of ATP levels at junctions versus cytosol was significantly decreased indicating a major impact of PKM2 absence on ATP levels at the junctional region of 2D human EC monolayers (Fig. 6I,J). We observed similar results in GO-ATeam1-transduced HUVEC when PKM2 activity was inhibited by shikonin in line with the reduction of ATP levels in the CSK-junction fraction (Supplementary Figure S6B,C). These findings indicate that PKM2-silencing reduces ATP levels nearby EC junctions.

PKM2 and local ATP regulate VE-cadherin dynamics and internalization at EC junctions.
We next explored whether reduced ATP production in the CSK-junction compartment in PKM2 absence/inhibition may affect VE-cadherin turnover 41 . Gradient sub-fractionation experiments revealed that PKM2 inhibition by shikonin increased the proportion of VE-cadherin present at the plasma membrane versus endosomes compared to control ECs (Supplementary Figure S6D).
Live imaging of HUVEC transduced with lentivirus coding for VE-cadherin-EGFP or LifeAct-EGFP in basal conditions and after stimulation with VEGFA allowed us to analyze VE-cadherin, and F-actin dynamics at EC junctions 42 . Time-lapse microscopy of VE-cadherin-EGFP-expressing HUVEC showed that PKM2 inhibition by shikonin led to increased numbers of inter-cellular gaps mostly after VEGFA treatment compared to control ECs (Fig. 7A,C; Movies S5-S8); similar gaps were observed in VE-cadherin-EGFP-transduced HUVEC silenced for PKM2 (Supp Figure S6E and Movies S9-S10). These data confirmed the enhanced junction instability observed in 2D endothelial monolayers in static conditions ( Fig. 3H-M). In LifeAct-EGFP-expressing ECs, we visualized small active F-actin protrusions in the lateral border preceding VE-cadherin 'plaques' and corresponding to 'junction-associated intermittent lamellipodia' (JAIL), which increased in size and persistence after VEGFA stimulation in control cells as reported 42 (Movies S11-S14). Notably, PKM2 inhibition by shikonin significantly reduced the number of JAIL and VE-cadherin plaques in untreated and VEGFA-stimulated cells compared to controls (Fig. 7A,B,D,E).
We next investigated directly the impact of PKM2 inhibition on VE-cadherin internalization in basal conditions and after stimulation with VEGFA. VEGFA-triggered VE-cadherin internalization is a classic clathrin-dependent/ATP-consuming event 43,44 . Notably, PKM2 silencing in ECs led to a significant reduction of VEGFA-induced VE-cadherin internalization compared to controls (Fig. 7F,G), arguing for decreased local ATP level as a mediator of such effect. We obtained a similar phenotype in human ECs treated with shikonin (Supplementary Figure S6F).
Our data support a model in which PKM2 absence/inhibition reduces ATP levels nearby endothelial junctions thus hindering VE-cadherin clathrin-mediated internalization and JAlL formation and overall resulting in impaired junction remodeling, collective migration, and angiogenic sprouting.

Discussion
This study presents the first description of the impact of subcellular compartmentalization of PKM2 on ATP levels, VE-cadherin internalization and EC junction rearrangements, and sprouting angiogenesis in vitro and in vivo.
PKM2 expression has been correlated with cell proliferation, particularly in tumor cells 26 . Our data establish that PKM2 is not required for EC proliferation, since PKM2-silenced and shikonin-treated ECs showed no differences from controls in vitro or in vivo. Despite the proposed role of PKM2 in cancer-cell proliferation, recent findings from Vander Heiden's group and others indicate that cell-cycle arrest in the absence of PKM2 is likely driven by compensatory PKM1 upregulation 27 . In this study, PKM2-null mice displayed distinct patterns of molecular weight. (I) Western blot and quantification of VE-cadherin in HUVECs transfected with control or PKM2 siRNA. Tubulin is included as a loading control. Means ± SEM, n = 3 independent experiments, ns non-significant by unpaired Student t-test. (J) Immunofluorescence of VE-cadherin (red) and PKM2 (green) in siRNA-silenced confluent HUVECs. Scale bar, 10 µm. (K and L) Mean fluorescent intensity (M.F.I.) of PKM2 (K) and VE-cadherin (L) at the junctions of siRNA-silenced confluent HUVECs; means ± SEM, n = 5 independent experiments, ****p < 0.0001 by Mann-Whitney test. (M) Percentage of discontinuous VEcadherin-junctions; means ± SEM, n = 3 independent experiments, *p < 0.05 by unpaired Student t-test. (N) Immunofluorescence of plakoglobin (red) and nuclei (blue, Hoechst) in siRNA-silenced confluent HUVECs. Scale bar, 10 µm. (O) Change in trans-endothelial electrical resistance (TEER) across monolayers of siRNAsilenced HUVECs left untreated or treated with 1 μM S1P for 3 hours; means ± SEM, n = 3 independent experiments, ns non-significant, *p < 0.05 by one-way ANOVA with Sidak post-test. See related Figure S2 www.nature.com/scientificreports www.nature.com/scientificreports/ tumor-cell proliferation depending on PKM1 expression, such that if PKM1 was expressed, proliferation capacity was reduced 27 . Moreover, PKM2-deficient primary cells derived from this mouse model also showed compensatory PKM1 upregulation, leading to impaired cell proliferation 28 . Corroborating this, proliferating embryo regions which would normally express PKM2 have been shown to initiate PKM1 expression in PKM2-null embryos, which is accompanied by a slowing of cell division and unexpected normal embryonic development 45 . In our experiments, PKM1 expression did not increase in PKM2-silenced ECs, possibly explaining the lack of any impact on EC proliferation. This is in contrast with recent reports claiming that PKM2 knockout reduced EC proliferation in which the compensatory increase of PKM1 might be a confounding factor to interpret direct PKM2 effects in EC proliferation 46,47 . Our findings further support the functional segregation of metabolic enzymes in regulating the spectrum of EC responses required for angiogenesis 14 .
PKM2 has been reported to both favor and decrease cancer cell migration [48][49][50] . Our study is in line with recent reports on the requirement of PKM2 for endothelial cell migration 46,47 . Other glycolytic enzymes have been shown to positively regulate cell migration, for example PGK1 in tumor cells 51 , PFKFB3 in ECs 11 , and HK2 in lymphatic ECs 13 . Our findings show that PKM2 activity is required for EC collective migration, evidenced by impaired EC orientation and migration in the scratch assay and by the reduced 3D sprout-length and retinal vascular radial outgrowth without changes in proliferation.
Our findings also provide insight into how differential VE-cadherin dynamics and mosaic junction activity influence productive angiogenesis. PKM2 loss reduced junctional VE-cadherin and increased the proportion of active/unstable junctions, thus compromising the balance between inactive/stable and active/unstable junctions and destabilizing the endothelial barrier. These findings thus indicate that PKM2 is primarily required for proper VE-cadherin-mediated endothelial cell-junction stability. The 'patching algorithm' developed by Bentley 35 has been especially useful in gaining insight into this process because it gives not only absolute numbers of stable and unstable junctions but also their clustered distribution. Recent studies into the role of EC junctions in angiogenesis provide evidence that they not only constitute a barrier and a signaling platform but also fine-tune cellular processes of paramount importance for efficient directionality and expansion of capillary sprouts. For example, VE-cadherin fingers from the pulling EC are essential for collective and cohesive EC migration during sprout extension 36 ; we confirm that defective VE-cadherin at junctions by PKM2 absence or inhibition results in impaired collective migration. Our findings also demonstrate that VE-cadherin internalization which is required for EC junction stability and remodeling, depends on PKM2 activity, likely through the provision of ATP for vesicle movement 20 since VEFGA-induced VE-cadherin endocytosis is clathrin-dependent and thus ATP-consuming 52 . PKM2-generated ATP at junctions could additionally favor actomyosin contractility or the phosphorylation of enzymes implicated in cell-cell junction stabilization or remodeling 6,53,54 .
PKM2 silencing also led to reduced numbers of filopodia in the endothelial tip cells. Since filopodia positively regulate cell motile velocity, they ultimately affect sprouting 55,56 . This phenotype could be related to the reported association of PKM2 with F-actin 34 since although our live imaging data favor the hypothesis of a primary impact of PKM2 inhibition on VE-cadherin internalization and JAIL formation, we cannot rule out an additional action on F-actin dynamics. Thus, PKM2 could provide ATP near to dynamic membrane protrusions such as filopodia and lamellipodia, and together with other enzymes such as PKFB3 form the actin-bound 'glycolytic hub' proposed by Carmeliet's group 19,57 . Whether a similar complex also forms at EC junctions deserves however further analysis.
VE-cadherin is the master regulator of EC junction formation and stability, but recently new players have been identified. For example, vinculin restrains junctions during remodeling and pacsin2 inhibits asymmetric VE-cadherin internalization, protecting EC-junction integrity from mechanical forces; moreover, the YAP/TAZ pathway increases VE-cadherin turnover, promoting cell migration while maintaining barrier function [58][59][60] . EC junctions, and particularly VE-cadherin, are thus dynamically and exquisitely remodeled during angiogenesis. Nevertheless, in spite of the recognized importance of metabolic pathways in EC responses, very little is known about how metabolism impacts EC junction stability and dynamics. Only the glycolytic enzyme PFKBF3 has been shown to modulate EC junctions, by an as-yet unknown mechanism 61 . Although PKM2 loss has been claimed to alter EC junction stability by non-enzymatic actions 46,47 , to our knowledge the present report is the first to demonstrate the presence of a pool of a glycolytic enzyme nearby the junctions of confluent ECs by immunostaining and biochemical approaches. Moreover, we show that PKM2 is required for ATP-dependent processes as VEFGA-induced clathrin-mediated VE-cadherin endocytosis 44 and thus for junction stability 41 . The fact that PKM2 at the CSK-junction enriched compartment has higher pyruvate kinase activity than cytosolic PKM2 suggests that the activity of the CSK-junction-associated PKM2 is positively regulated by mechanisms that may involve tetramer stabilization upon F-actin binding, similar to what occurs with PFKFB3 at lamellipodia 57 . The activity of PKM2 tetramers at cell-cell junctions may also be regulated by local interaction with allosteric activators or posttranslational modifications 62 . This also raises the question of how PKM2 traffics toward and locates at the plasma membrane. PKM2 interacts with huntingtin 63 , which associates with GAPDH in vesicles traveling along microtubules in neurons 20 , and it has recently been found enriched at these motile vesicles 64 . It is therefore possible that PKM2 also traffics on microtubules alone or together with GAPDH in ECs. Furthermore, we demonstrate that VE-cadherin is required for PKM2 localization at the endothelial junctions. Whether PKM2/VE-cadherin association is direct or mediated by an intermediate protein as β-catenin, a renowned VE-cadherin partner also shown to interact with PKM2 in other contexts 65 , deserves further investigation. independent experiments. Black and red tracks indicate cells migrating toward and against the wound. (I and J) Velocity, linear (Euclidean) distance and directionality of single cells moving against (I) and toward the wound (J); means ± SEM, n = 50-60 cells from 3 independent experiments, ns non-significant and *p < 0.05 by Mann-Whitney test. (K) Representative single-cell tracks of three adjacent siRNA-silenced ECs migrating during the scratch assay. See related Figure S3 and Movies S1 and S2. www.nature.com/scientificreports www.nature.com/scientificreports/ The defective vasculature of PKM2-silenced and shikonin-treated retinas in postnatal mice supports the recent evidences of a requirement for PKM2 in angiogenesis in vivo 46,47 . In our context, the vascular impairment affected radial growth and filopodia number without influencing EC proliferation, and the phenotype correlates www.nature.com/scientificreports www.nature.com/scientificreports/ with the altered ratio of stable and unstable EC junctions. The possible clinical application of these findings to stimulate or inhibit PKM2 activity in disease needs to be carefully considered. In pathological angiogenesis, the proportion of active junctions increases (with a discontinuous pattern) 35 , and sprouting defects are related to www.nature.com/scientificreports www.nature.com/scientificreports/ impaired endothelial junctions 66,67 . In contrast, there is no evidence that an increased proportion of inactive junctions (with a continuous pattern) has any effect on angiogenesis 68 . PFKFB3, a key glycolysis regulator, has been proposed as a target for inducing normalization of EC junctions 61,69 . In a model of high-dose-VEGFA-induced pathological retinal angiogenesis 35,70 , the PFKFB3 inhibitor 3PO decreased glycolysis by 15% and normalized the vasculature 69 . This effect was also observed in 3PO-treated tumor ECs 61 . However, treatment of ECs with higher 3PO concentrations, inhibiting glycolysis by 40%, induced an increase in discontinuous junctions rather than vessel normalization 71 . This finding underlines the importance of the level of glycolysis inhibition in determining the final angiogenic outcome. In line with this, we observed increased instability of VE-cadherin EC junctions upon PKM2-silencing, which also decreases glycolysis by 40%, and with the global glycolysis inhibitor 2-DG, which almost abolishes glycolysis. Moreover, high concentrations of exogenous ATP also increase discontinuous  Figure S6 and Movies S5-S8 and S11-S14. (2019) 9:15022 | https://doi.org/10.1038/s41598-019-50866-x www.nature.com/scientificreports www.nature.com/scientificreports/ junctions and decrease TEER in brain endothelial cells 72 . Together, these observations suggest a dual effect of ATP, with a precisely adjusted level required for proper dynamic EC junction rearrangement. Metabolic regulation of EC junctions thus appears to be a complex process that depends on the quantity of ATP produced, the step in the glycolytic pathway involved, and where the implicated enzyme is located. Given the likelihood that glycolysis inhibition will have a 'dose response' effect on angiogenesis, studies are warranted to address the impact of PKM2 silencing or inhibition on pathological angiogenesis.
Our findings show that PKM2 makes an essential contribution to angiogenesis in vitro and in vivo by providing ATP for the regulation of VE-cadherin internalization and EC junctions. In light of the importance of PKM2 in cancer and other diseases such as diabetes, our findings provide information relevant to the development of PKM2-based therapeutic strategies.
Mouse lung endothelial cells (MLECs) isolated from C57BL/6 mice and VE-cadherin-null mouse endothelial cells provided by E. Dejana (FIRC Institute of Molecular Oncology, Italy) were cultured as described previously 6  Animals were intravitreally injected with 1 µg of siRNA in Lipofectamine RNAimax at P3 and sacrificed at P6. For pharmacological inhibition of PKM2 activity, 100 µM of shikonin (or DMSO control) were injected at P5, and the mice were sacrificed at P6. Eyes were fixed overnight in 2% paraformaldehyde (PFA) and then enucleated, and retina was isolated as reported 74 . siRNA interference and plasmid and lentivirus transfection. Silencer ™ Select siRNA oligonucleotides against isoform M2 of pyruvate kinase (Thermo Fisher) were transfected using Lipofectamine RNAimax (Thermo Fischer) in Optimem medium (Lonza) at a final siRNA concentration of 10 nM. Oligonucleotide sequences were 5′CCAUAAUCGUCCUCACCAA dTdT 3′ for human PKM2 and 5′ GUGCGAGCCUCCAGUCACU dTdT 3′ Ambion ™ In Vivo for mouse PKM2. Control siRNA was from Ambion ™ In Vivo. Scratch assays. Confluent HUVECs were serum-starved for 3 hours and a wound was made with a plastic pipette tip. After washing the cells with PBS, migration was stimulated by adding medium containing 5% FBS. In the experiments in which proliferation was inhibited, cells were treated before and throughout the experiment with 2 μg ml −1 mitomycin C, 1 µM Shikonin or corresponding vehicle control. Migration was quantified as the proportion of the initial wound area remaining uncovered at the end of the experiment. (2019) 9:15022 | https://doi.org/10.1038/s41598-019-50866-x www.nature.com/scientificreports www.nature.com/scientificreports/ For collective migration, time-series pictures were taken every 20 minutes during 16 hours; only those experiments in which the endothelial monolayer remained confluent >80% after siRNA silencing or shikonin treatment were included in the analysis.
Cell count assay. Cells were seeded at equal density in P96 wells. On each day of the experiment, cells in individual wells were detached and counted.
Microbead spheroid assay. Cytodex3 microbeads were purchased from Amersham, and the protocol was performed as previously described 75 . Transendothelial electrical resistance (TEER). TEER assays with an electric cell-substrate impedance sensing system (ECIS 1600R; Applied Biophysics 76 ;) were performed as described 77,78 . The S1P-induced TEER increase was expressed as the percentage difference in resistance between unstimulated ECs and ECs stimulated with S1P for 3 hours.
Immunofluorescence. Cells were fixed for 15 min in 4% PFA, blocked and permeabilized in 0.01% Triton X-100 2% BSA in PBS. Primary antibodies were incubated overnight in a cold room and secondary antibodies for 1 hour at 37 °C. Cell orientation was quantitated according to the relative position of Golgi versus nucleus toward or against the wound (±180°).
Fibrin gels were fixed in 1% PFA PBS for 30 minutes and washed. Gels were then permeabilized and blocked for 2 hours in blocking and permeabilization (BP) buffer (2% BSA, 0.3% Triton X-100, 0.3 M glycine, 0.2% azide, and 1% goat serum in PBS). Gels were incubated with primary antibodies overnight in the same buffer. The next day, gels were post-fixed in 4% PFA for 15 minutes. After a wash and 30 minutes in BP buffer, gels were incubated with secondary antibodies.
Retinas were blocked and fixed overnight in 2% BSA, 1% goat serum, 0.2% deoxycholate, 0.5% Triton X-100, and 0.2% azide in PBS. The next day, primary antibodies were added and incubations were maintained overnight. Retinas were then incubated overnight with Isolectin B4 in PBLEC buffer (1 mM MgCl 2 , 1 mM CaCl 2 , and 0.1 mM MnCl in PBS pH 6.8). Finally, secondary antibodies were added and incubations were maintained overnight. All preparations were mounted in Fluoromount-G.
Proximity ligation assay (PLA). HUVECs were grown to confluence, fixed and permeabilized. PLA was performed next day following the manufacturer's instructions (Duolink PLA, Sigma Aldrich). Briefly, samples were blocked with the blocking solution for 30 min at 37 °C and incubated overnight with the primary antibodies, rabbit anti-PKM2 (#4053, Cell Signalling) and goat anti-VE-cadherin (sc-6458, Santa Cruz biotechnology), at 4 °C. The following day, the samples were washed and incubated for 1 h at 37 °C with PLA probes for rabbit and goat antibodies. After washing, samples were incubated with the ligase for 30 min at 37 °C, washed again and incubated with the polymerase for 100 min at 37 °C. Finally, the samples were washed, dried completely and mounted in Duolink In Situ Mounting Medium with Hoechst. Rabbit and goat IgGs were used as negative control.
Cell live imaging. Live imaging was performed with a LSM700 Zeiss confocal microscope (Carl Zeiss) at a constant temperature of 37 °C and 5% CO 2 . PKM2-GFP was detected with 488 nm excitation and 510 nm emission. Go-ATeam1 was detected with 488 nm excitation and emission at 510 nm and 560 nm. VE-cadherin-GFP and LifeAct-GFP were detected with 488 nm excitation and 510 nm emission, time-series images were taken every minute for 15 minutes and after addition of 50 ng/ml of VEGFA for a further 15 min. Image analysis was performed using Fiji software.
Membrane subfractionation. The membrane subfractionation protocol was performed as previously described 79 . Briefly, HUVEC membrane fractions were obtained and separated by ultracentrifugation on a discontinuous Optiprep density gradient (AxisShield). Protein fractions were precipitated, resuspended in Laemmli buffer, resolved by SDS-PAGE, and analyzed by western blot.
Western blotting. Cells were lysed in RIPA buffer as described 6 . Retinas were isolated as reported 80 , and then RIPA buffer was added to the exposed vessels. Centrifugation was performed after 1 hour of incubation, and supernatant was recovered. Protein lysates were separated by 8% SDS-PAGE under reducing conditions. After transfer to a nitrocellulose membrane (Bio-Rad), blots were blocked with 5% BSA-PBS and incubated for 1 hour at room temperature (RT) with primary antibody diluted in 2% BSA-PBS. Membranes were then washed in 0.2% Tween 20-PBS and incubated for 1 hour at RT with horseradish peroxidase-conjugated secondary antibody. Protein bands were visualized by enhanced chemiluminescence (ECL; GE Healthcare). Band intensities were analyzed with Fiji software.
Pyruvate kinase assay. PK activity was measured in saponin-permeabilized HUVEC fractions using a commercial kit from Biovision.

VE-cadherin internalization.
Internalization assays were performed as previously described 82 . Briefly, to measure internalization of endogenous VE-cadherin, confluent HUVEC were incubated with the VE-cadherin antibody (TEA1/31) at 4 °C for 1 h in 1% BSA medium. Then, we rinsed the cells in cold culture medium to remove unbound antibody and treated them with 50 ng/ml of VEGFA for 30 min at 37 °C to allow the internalization. Cells were acid-washed (2 mM PBS-glycine, pH 2.0) or washed with cold PBS 3 times for 10 min each (used as control) and fixed and processed for immunofluorescence.
Confocal image quantification. Images were obtained with a Zeiss LSM700 confocal microscope (Carl Zeiss Microscopy) and maximum intensity projections are shown in all figures unless otherwise indicated. For analysis of mean fluorescence intensity, regions of interest (ROIs) were selected around junctions with Fiji software. For junction morphology analysis, we used reported criteria to distinguish discontinuous and continuous junctions 83 . For spheroid assays, mean length of sprouts and sprout number were analyzed in bright-field images with Fiji. Filopodia number was determined using phalloidin-stained F-actin. Data were normalized to cell length. For Go-aTeam1 analysis, the region of interest (ROI) for EC junctions was drawn ≤5 µm width around the cell-cell contacts defined by a 16 colors-pseudocolor plugin of Fiji software (validated by VE-cadherin staining). The cytosol ROI was defined as the rest of the cell.
Retinal parameters were quantified as described 74,84 . Radial growth was calculated as the average length in each lobule of one eye, from the optic nerve to the vascular front. Vascular density was calculated as the percentage of total area occupied by isolectin B4. Ki67 positive ECs were calculated as the percentage of Ki67 + ERG + cells of total ERG + cells, and normalized by Isolectin B4 area. Junction morphology was analyzed using the 'patching algorithm' implemented in MATLAB software 35 .
For VE-cadherin internalization experiments, images were taken using Nikon A1R confocal microscopy with a 40x objective and image analysis were performed using Fiji software.
For time-lapse scratch assay, images were taken with a Nikon time-lapse microscope with a 10x objective. Analysis of the tracking was done using Fiji with manual tracking plugin and Chemotaxis tool from ibidi.

Statistical analysis.
Absolute values were represented for most of the quantitated parameters. In cases of high variability in such values among experiments, fold induction versus control was calculated for each experiment as indicated in the corresponding legend. All values were analysed with Prism GraphPad. First all values were subjected to Grubbs's outlier test (p < 0.05), and then D' Agostino-Pearson test for analysis of normal distribution was performed. The proper tests to compare samples were used as indicated in each figure legend. In general, for comparison of two samples, unpaired Student t-test was used in normal (Gaussian) data with equal variance, Welch's test was used in normal (Gaussian) data with no equal variance and Mann-Whitney test for non-normal data; for comparison of three or more conditions, One-way ANOVA with Sidak-post test was used. Differences were considered significant at p < 0.05.

Data availability
All relevant data are available from the authors upon request.