PDGFB-mediated PC recruitment regulates retinal angiogenesis and capillary size

To address the role of PDGFB-mediated PC recruitment in AVM pathogenesis, we depleted Pdgfb specifically in ECs by crossing Pdgfbfl/fl to the tamoxifen (TX) inducible Cdh5-CreERT2 mouse line to create PdgfbiΔEC (Fig. 1A). Pdgfb gene depletion was induced in newborns via intraperitoneal (i.p) TX administration at postnatal days (P0–P2 and retinal vasculature was analyzed at P7. Efficient gene depletion was confirmed by qPCR and western blot (WB) in mouse lung ECs (mLECs) isolated from TX induced P7 pups (Suppl. Fig. 1a, b). Loss of Pdgfb resulted in no significant changes in the body weight compared to control neonates (Pdgfbfl/fl) (Suppl. Fig. 1c). Labeling P7 Pdgfbfl/fl and PdgfbiΔEC retinas for myocardial neuron-glial antigen 2 (NG2), a MC marker, and Isolectin B4 (IB4) to detect the ECs, revealed severely impaired PC coverage of the developing vessels (Fig. 1B, quantified in C) with multiple vascular defects: decrease in vascular area and outgrowth, loss of side branches, and vessel dilation with the highest impact within the capillaries and a significant decline in the morphological altered tip cell number with fewer and shorter filopodia (Fig. 1D; Suppl. Fig. 1d–f), thus confirming previous findings [7, 25, 26].

Fig. 1

Loss of EC Pdgfb-mediated PC recruitment results in capillary enlargement and AV shunting. A Schematic of the experimental strategy used to delete Pdgfb in ECs (P0–P7). Arrowheads indicate i.p injection of 100 μg TX at P0-P2 in Pdgfbfl/fl and Pdgfbi∆EC. B Labeling for NG2 (white) and IB4 (green) of P7 Pdgfbfl/fl and Pdgfbi∆EC retinas. Yellow arrowheads point to loss of PC coverage in the capillaries and veins. C Quantification of the PC coverage in Pdgfbfl/fl and PdgfbiΔEC (Pdgfbfl/fl n = 5, Pdgfbi∆EC n = 8, unpaired 2-tailed t test). D Labeling with IB4 (negative images) of Pdgfbfl/fl and Pdgfbi∆EC P7 retinas. E Quantification of the number of AV shunts in Pdgfbfl/fl and PdgfbiΔEC P7 retinas (Pdgfbfl/fl n = 7, Pdgfbi∆EC n = 12, Mann–Whitney U test). F–J Images of P7 retinas (F), pial vessels (G), skin (H), GI tracts—small intestine (I) and lungs (negative images) (J) in Pdgfbfl/fl and PdgfbiΔEC mice perfused with latex red-dye. Perfused retinas in F were subsequently stained with IB4 (blue). Yellow arrowheads in D, F, G mark AV shunts. Red arrowheads in H, I and J mark latex-perfused veins in the skin, GI tracts and lungs. ***P < 0.001. a artery, v vein

Interestingly, in addition to the previously described phenotypes, loss of Pdgfb led to direct connections between arteries and veins at the vascular front, on average 1.7 connections per retina (arrowhead in Fig. 1D, quantified in E). To confirm whether these direct connections between arteries and veins are indeed AV shunts, we injected latex red-dye into the left side of the neonatal P7 hearts [20]. Due to its size, the latex cannot pass the capillary bed and remains confined to arterial branches. Yet, in PdgfbiΔEC retinas, latex was present in the enlarged capillaries and into the retinal veins, suggesting direct AV connections between arteries and veins upon Pdgfb LOF (Fig. 1F). To identify if Pdgfb loss leads to AV shunting in other organs, we analyzed other vascular beds following intracardial latex injection. Interestingly, we found AV shunting also in the pial vessels (Fig. 1G) and latex perfused veins in the skin of the scalp, gastrointestinal (GI) tract (small intestine) and lungs of mutant neonates, whereas no perfused veins were seen in control neonates (Fig. 1H–J). These findings may indicate either capillary enlargement in visceral organs, or presence of direct AV shunting at a downstream or upstream location from the visualized vascular field. Thus, PDGFB-mediated PC recruitment regulates retinal angiogenesis and capillary size.

Endothelial PDGFB is required for PC maintenance to control the capillary size

To further understand if PDGFB-mediated maintenance of PCs on developing endothelium is required to control capillary size, we depleted Pdgfb at P5–P7 and analyzed retinas at P12 (Fig. 2A). Interestingly, loss of Pdgfb at this stage led to increased retinal hemorrhage (Fig. 2B), and impaired PC coverage as quantified by the NG2 + /IB4 + vascular area (Fig. 2C). By measuring vessel diameter at P12, we found a significantly altered capillary enlargement, while the diameter of veins or arteries remained unaffected (Fig. 2D). Retinal AV shunting occurred with an average of 1.7 connections per retina as shown by IB4 + endothelium and latex positive capillaries and veins (Fig. 2E, quantified in F). Surprisingly, at this stage, the AVM-like structures formed closer to the optic nerve, in higher flow regions (arrowheads in Fig. 2E), suggesting that blood flow participates in AV shunt formation upon Pdgfb depletion. Latex dye-positive veins were also observed in the GI tract (Fig. 2G). These findings argue that PDGFB is indispensable for the maintenance of PCs on developing endothelium to restrict capillary size and protect against retinal AV shunting.

Fig. 2

Endothelial PDGFB is required for PC maintenance to restrict the capillary size. A Schematic of TX administration in postnatal Pdgfbfl/fl and PdgfbiΔEC mice at P5-P7 and analysis at P12. B Representative images of hemorrhagic whole retinas isolated from Pdgfbfl/fl and PdgfbiΔEC P12 neonates. C Quantification of PC coverage in P12 retinas (Pdgfbfl/fl n = 6, Pdgfbi∆EC n = 6, unpaired 2-tailed t test with Welch’s correction). D Quantification of the retinal vessel diameter in Pdgfbfl/fl and PdgfbiΔEC retinas (Pdgfbfl/fl n = 8, Pdgfbi∆EC n = 8, Mann–Whitney U test). E Images of Pdgfbfl/fl and PdgfbiΔEC P12 retinas stained for IB4 (green) latex perfused (blue). F Quantification of the number of AV shunts in P12 retinas (Pdgfbfl/fl n = 6, Pdgfbi∆EC n = 10, Mann–Whitney U test). G Images of P12 GI tracts latex perfused. Yellow arrowheads in E mark latex perfused retinal AV shunts. Red arrowheads in G mark the latex-perfused veins in the GI tract. ns non-significant, ***P < 0.001. a artery, v vein

Endothelial PDGFB regulates capillary EC size and polarity

The Pdgfb LOF phenotype was extensively described [7, 26], yet AV shunt formation hasn’t been reported so far. Therefore, we focused on characterizing the cellular hallmarks of AVM-like structures in P7 PdgfbiΔEC retinas. Upon staining retinas for alpha smooth muscle actin (αSMA), we found a strong reduction in arterial vascular smooth muscle cells (vSMC) coverage, with scattered αSMA + immunostaining in capillaries and more abundant in veins but no ectopic αSMA within the AV shunt capillaries (Suppl Fig. 2a, quantified in b). Labelling retinas for IB4 and ICAM2, a cell adhesion molecule located at the apical EC surface further confirmed an increase in capillaries’ lumen (Suppl Fig. 2c, quantified in d). Staining for Collagen IV (ColIV) that labels the basal surface identified a higher number of empty sleeves (IB4-/ColIV +) (Suppl Fig. 2e, quantified in f), indicating excessive vascular pruning events in PdgfbiΔEC retinas.

We next reasoned that the expansion of capillary diameter is due to, either an increase in the size of capillary ECs or due to proliferating ECs clustering within capillaries. Another possibility is that perhaps disrupted EC migration against the bloodstream contributes to lumen enlargement, as capillary regression in developing vascular plexus is a cell migration–driven process from low towards higher flow regions [27].

Staining retinas for VE-cadherin identified larger capillary single ECs and loss of VE-cadherin at cell–cell junctions (Fig. 3A, quantified in B), consistent with the increased hemorrhage observed in P12 mutant retinas (in Fig. 2B). Furthermore, labeling for the endothelial nuclear ERG1,2,3 transcription factor (ERG) [28] and IB4 identified an increase in the number of capillary EC nuclei per capillary width (Fig. 3C, quantified in D). In control retinas, capillaries contained a single EC, whereas in PdgfbiΔEC retinas, capillaries contained two to three EC nuclei (Fig. 3C, quantified in D).

Fig. 3

EC PDGFB maintains capillary EC size and EC polarity against the bloodstream. A, C, E, G–I Co-immunolabeling for VE-Cadherin (green) and IB4 (white) (A), ERG (white) and IB4 (blue) (C), EdU (red), ERG (blue) and IB4 (white) after 4 h of EdU incorporation (E), and for ERG (green), IB4 (blue) and GM130 (Golgi marker-red) in arteries, veins and capillaries (G–I) of P7 Pdgfbfl/fl and PdgfbiΔEC retinas. G–I Right side, panels illustrating EC polarization based on the position of the Golgi apparatus in relation to the nucleus in the direction of migration (green arrows). B, D, F Quantification of the EC size (Pdgfbfl/fl n = 10, PdgfbiΔEC n = 10, unpaired 2-tailed t test) (B), of the number of EC nuclei per capillary width (Pdgfbfl/fl n = 6, PdgfbiΔEC n = 6, Mann–Whitney U test) (D) and of EdU + ERG + /ERG + ECs (Pdgfbfl/fl n = 11, PdgfbiΔEC n = 14, unpaired 2-tailed t test) (F). J Quantification of EC polarization: against the direction of flow (green arrows), towards the direction of flow (red arrows) and non-oriented (blue arrows) in arteries (G), veins (H) and capillaries (I) from P7 TX induced retinas from the indicated genotypes (Pdgfbfl/fl n = 4/5, PdgfbiΔEC n = 6/8, unpaired 2-tailed t test). Yellow arrowheads in A point towards capillary ECs with loss of VE-cadherin at the cell–cell junctions, in C towards the enlarged capillaries containing more than one EC and in E towards the EdU- ECs within the AV shunt. K VE-Cadherin staining (negative images) of CTRL and PDGFB siRNA HUVECs subject to 1 DYNE/cm2. L Quantification of EC area, EC alignment and length/width ratio in CTRL and PDGFB siRNA HUVECs subject to 1 DYNE/cm2 (n = 4/5 average of 3 images (70–140 cells/image) per 3 independent experiments/group, unpaired 2-tailed t test). M Labelling of CTRL and PDGFB siRNA HUVECs subject to 12 DYNES/cm2 for GM130 (Golgi apparatus), VE-Cadherin and DAPI (nuclei) (N) quantification of EC orientation against the direction of flow, with the flow and others (non-oriented). ns non-significant, *P < 0.05, **P < 0.01, ***P < 0.001. a artery, v vein

To assess if EC proliferation leads to capillary expansion, we further quantified the proportion of ECs in S-phase (EdU + ERG + /ERG +) upon 5-ethynyl-2′-deoxyuridine (EdU) incorporation in capillary ECs at the vascular retinal front in Pdgfbfl/fl and PdgfbiΔEC neonates. We found an overall decrease in the total number of ERG + ECs with fewer EdU + ERG + cells within the AVM-like structures (Fig. 3E, quantified in F), suggesting that capillary enhancement does not involve proliferation. Furthermore, labeling retinas for P21 encoded by CDKN1A (cyclin dependent kinase inhibitor 1A), an indicator of cell cycle arrest [29], showed an increase in the number of arrested ECs (P21 + ERG + cells) within the dilated capillaries (Suppl. Fig. 2g, quantified in h).

Failure of ECs to orient and migrate against the bloodstream was proposed to trigger AVM formation upon depletion of BMP9/10 signaling receptors, Alk1 and Eng in ECs [30, 31]. This physiological process is crucial for vessel regression-mediated vascular remodeling and artery formation [27, 32,33,34]. To identify if defective flow-mediated EC migration contributes to capillary enlargement and AV shunting, we labelled Pdgfbfl/fl and PdgfbiΔEC retinas for Golph4 (to detect the Golgi apparatus), ERG and IB4. Quantifying the Golgi position in relation to the nucleus and flow direction, we found a significant altered orientation of Pdgfb mutant ECs against the bloodstream in arteries, with a more pronounced EC polarization with the flow in veins and capillary ECs engaged in AV shunts (Fig. 3G–I, quantified in J). Taken together, these results suggest that capillary enlargement does not involve proliferation but instead an increase in single EC size and disrupted EC migration against the bloodstream.

As capillary enlargement leads to changes in blood flow or fluid shear stress (FSS), we further assessed the cellular effects of PDGFB depletion (using the siRNA strategy) in HUVECs subject to low fluid shear stress (L-FSS) (1 DYNE/cm2). Remarkably, PDGFB depleted HUVECs under 1 DYNE/cm2 were larger, more elongated and aligned parallel to the direction of flow, whereas no significant changes were observed in CTRL siRNA HUVECs (Fig. 3K, quantified in L). We further addressed the effect of PDGFB loss on EC orientation in response to physiological flow (12 DYNES/cm2). Labelled HUVECs for GM130 (for Golgi apparatus), DAPI (for nucleus) and VE Cadherin showed dysregulated migration against the flow direction, with more ECs polarized with the flow upon PDGFB loss (Fig. 3M, quantified in N). Thus, PDGFB depletion amplifies morphological responses to FSS e.g. EC size, elongation and alignment and leads to disrupted EC orientation against the FSS direction.

Loss of Pdgfb leads to an altered capillary arterial-venous zonation

Previous analysis of single-cell RNAseq datasets identified defective capillary zonation with venous skewing in the brain microvasculature of Pdgfb hypomorphs (Pdgfbret/ret), pointing towards the role of PCs in maintaining capillary arterio-venous differentiation [35]. To identify if an altered arterial-venous zonation occurs in AVM-like structures in the retina, we immunolabeled control and PdgfbiΔEC retinas for several arterial and venous markers at P7. While JAG1, a ligand of NOTCH1 marks predominantly the arteries and arterioles in control retinas [36], in Pdgfb LOF retinas, JAG1 expression was retained in arteries but decreased in the arterioles (red arrows in Fig. 4A, quantified in B). In contrast, Endomucin, a transmembrane sialomucin expressed exclusively on the surface of capillary and venous ECs [37, 38], displayed an opposite pattern in PdgfbiΔEC vasculature, its expression was upregulated in all capillary ECs expanding also into arterioles (Fig. 4C, quantified in D). These results suggest gain of venous identity and loss of arterial identity. To confirm venous skewing upon Pdgfb loss, we performed qPCR on isolated mLECs and observed upregulation of the venous markers, e.g. EphB4, Nrp2, Nr2f2 and Aplnr (Fig. 4E). Surprisingly though, we identified an upregulation of the arterial markers (Sox17, Nrp1, Efnb2) as well (Fig. 4F).

Fig. 4

Endothelial PDGFB maintains capillary arterial-venous zonation. A, C, G, I Immunostaining of P7 Pdgfbfl/fl and PdgfbiΔEC retinas for Jag1 (white) and IB4 (red) (A), Endomucin (white) and IB4 (red) (C), Sox17 (green) and IB4 (white) (G), Dll4 (green) and IB4 (red) (I). B, D, H, J Quantification of the labelling intensity of Jag1 (Pdgfbfl/fl n = 6, PdgfbiΔEC n = 6, unpaired 2-tailed t test) (B), Endomucin (Pdgfbfl/fl n = 4, PdgfbiΔEC n = 4, unpaired 2-tailed t test) (D), Sox17 (Pdgfbfl/fl n = 6, PdgfbiΔEC n = 8, Mann–Whitney U test) (H) and Dll4 (Pdgfbfl/fl n = 5, PdgfbiΔEC n = 6, unpaired 2-tailed t test) (J), per vascular area in P7 Pdgfbfl/fl and PdgfbiΔEC retinas. E, F qPCR for indicated venous markers (Pdgfbfl/fl n = 4, PdgfbiΔEC n = 4, unpaired 2-tailed t test with Welch’s correction) (E), and arterial markers (Pdgfbfl/fl n = 4, PdgfbiΔEC n = 4, unpaired 2-tailed t test with Welch’s correction) (F) in mLECs from Pdgfbfl/fl and PdgfbiΔEC mice. *P < 0.05, **P < 0.01, ***P < 0.001. a artery, v vein

To dig further into possible gain of arterial identity, we labeled retinas for arterial markers: Sox17 (SRY-box transcription factor 17) which plays a crucial role in arterial identity maintenance [39] and Notch ligand Delta-like 4 (Dll4), another NOTCH1 ligand. Interestingly, upon Pdgfb loss, Sox17 expression extended exclusively throughout the AV shunt from the artery towards the vein, with some Sox17 positive ECs present within the vein (arrowheads in Fig. 4G, quantified in H). A similar pattern was observed also for Dll4 (arrowheads in Fig. 4I, quantified in J).

Thus, our results together argue that the AV shunting upon Pdgfb loss involve gain of arterial identity with more arterial ECs migrated from the artery into the veins and point towards an arterial origin for the AVMs-like structures.

AVM-like structures upon endothelial Pdgfb loss have a non-arterial origin

To test the hypothesis that AVMs-like structures upon Pdgfb LOF are of arterial origin, we intercrossed Pdgfbfl/fl mice to an artery-specific BMX non-receptor tyrosine kinase (BMX) mouse line and mTmG reporter mice to lineage-trace green fluorescent protein (GFP) in Bmx + and Pdgfb depleted ECs (PdgfbiΔBMX;mTmG). Tx was injected at P0-P2, and retinas were analyzed at P7 (Fig. 5A). GFP was detected in arteries and some arterioli of PdgfbiΔBMX;mTmG retinas confirming Cre-mediated recombination specifically in arterial ECs (Fig. 5B). We found that PdgfbiΔBMX;mTmG mutants displayed no retinal AVMs (Fig. 5B), and loss of Pdgfb specifically in arterial ECs led to a non-significant loss of PCs (Fig. 5C, quantified in E) or αSMA (Fig. 5D, quantified in F) with no obvious vascular defects (Fig. 5G, H). These results imply that PDGFB is dispensable for artery maintenance and in the same time suggest a non-arterial origin for the AVM-like upon Pdgfb loss.

Fig. 5

AVM-like structures upon Pdgfb loss have a non-arterial origin. A Schematic representation of the experimental strategy to delete Pdgfb (P0–P7) in arterial ECs using the Bmx-Cre;mTmG reporter mice. Arrowheads indicate i.p injection of 100 μg TX at P0-P2 in Pdgfbfl/fl;mTmG and Pdgfbi∆BMX;mTmG. B, C, D Representative images of P7 TX induced Pdgfbfl/fl;mTmG and Pdgfbi∆BMX;mTmG retinas labelled for GFP (green) and IB4 (red) (B), NG2 (white), GFP (green) and IB4 (red) (C), αSMA (green) and IB4 (red) (D). E–H) Quantification of  PC coverage (%) (Pdgfbfl/fl n = 4, Pdgfbi∆EC n = 4, unpaired 2-tailed t test) (E), αSMA coverage (%) (Pdgfbfl/fl n = 5, Pdgfbi∆EC n = 5, unpaired 2-tailed t test) (F), radial length (Pdgfbfl/fl n = 8, Pdgfbi∆EC n = 11, unpaired 2-tailed t test) (G) and artery diameter (Pdgfbfl/fl n = 8, Pdgfbi∆EC n = 11, unpaired 2-tailed t test) (H) in P7 Pdgfbfl/fl;mTmG and Pdgfbi∆BMX;mTmG. ns non-significant, a artery, v vein

PDGFB-mediated PC recruitment is associated with TGF-β/BMP and NOTCH signaling activation

As we identified a mixed arterial-venous zonation within the AV shunts, we next focused on signaling pathways regulating arterial-venous identity, whose disruption may contribute to AV shunting in Pdgfb LOF mice. In addition to a well-described role of NOTCH signaling in maintaining arterial identity [33, 40, 41], more recently it has been proposed that TGFβ and BMP signaling pathways regulate arterial and venous identity, respectively [42].

As a readout for TGFβ/BMP pathway activation, we labelled Pdgfbfl/fl and PdgfbiΔEC retinas for phosphorylated canonical SMADs. Surprisingly, we identified abundant expression in ECs of both activated canonical SMADs, the pSMAD3 (Fig. 6A, quantified in B) and pSMAD1/5 (Fig. 6C, quantified in D) at the highest intensity within the enlarged capillaries engaged in AVMs-like structures. Furthermore, activation of the canonical SMADs was accompanied by upregulation of TGFβ/BMP target genes [43,44,45], ID1 (Fig. 6E, quantified in F) and ENG (Fig. 6G, quantified in H). To validate these results, we performed WB for the activated SMADs (Fig. 6I) and qPCR analysis for the TGFβ/BMP target genes in mLECs isolated from TX induced P7 Pdgfbfl/fl and PdgfbiΔEC mice (Fig. 6J). We confirmed increased phosphorylation in SMAD3 and SMAD1/5 (Fig. 6I) and upregulation of TGFβ/BMP downstream target genes: Eng, Thbs1 (Thrombospondin1) and Fn1 (Fibronectin), thus validating activation of TGFβ/BMP signaling pathways upon loss of Pdgfb in ECs. Interestingly, opposite as expected, we identified upregulation of Apln (Apelin) (Fig. 6J), a known suppressed BMP target gene [46].

Fig. 6

Loss of endothelial Pdgfb is associated with activation of TGFβ/BMP and NOTCH signaling pathways. A, C, E, G Co-labelling of P7 Pdgfbfl/fl and PdgfbiΔEC retinas with IB4 (blue), Erg (white) and pSMAD3 (green) (A), IB4 (blue), ERG (white) and pSMAD1/5 (green) (C), IB4 (blue), ERG (white) and ID1 (green) (E) ENG (white) and IB4 (red) (G). B, D, F, H Quantification of the labelling intensity per vascular area (%) of pSMAD3 (B), pSMAD1/5 (D), ID1 (F) and of ENG (H) in P7 retinas Pdgfbfl/fl and PdgfbiΔEC retinas (Pdgfbfl/fl n = 6, Pdgfbi∆EC n = 6, unpaired 2-tailed t test with Welch’s correction). Yellow arrowheads point to upregulated proteins in dilated capillaries upon Pdgfb LOF. I WB of Pdgfbfl/fl and PdgfbiΔEC mLECs for activated pSMAD1/5 and pSMAD3 and total SMAD1, SMAD2/3 and VE-Cadherin as a loading control. J, K qPCR for TGFβ/BMP (J) and NOTCH target genes (Pdgfbfl/fl n = 4, Pdgfbi∆EC n = 4, unpaired 2-tailed t test with Welch’s correction) (K). *P < 0.05, **P < 0.01, ***P < 0.001. a artery, v vein

As gain of NOTCH signaling in ECs also leads to AV shunting, and we detected increased Dll4 within the enlarged capillary ECs, we performed qPCR for NOTCH transducers and transcriptional effectors and identified an increase in the expression of Notch1 and Notch4 receptors and upregulated transcription factors Hey1, Hey2 and Hes1 (Fig. 6K), indicating, additionally, EC NOTCH activation upon Pdgfb loss.

These results imply that loss of arterial-venous zonation upon Pdgfb depletion is associated with increased TGFβ/BMP and Notch activation in ECs.

Capillary enlargement results in pathological flow-mediated KLF4 upregulation

Physiological blood flow-induced FSS maintains vascular stability by mediating MC recruitment [3, 4] and promoting EC elongation, alignment [47] and arterial cell fate [48, 49]. In turn, MCs regulate the blood flow and vascular tone [5], but also the blood flow directionality [10].

Capillary expansion upon Pdgfb loss results in an altered blood flow and/or an altered FSS within the capillaries. To visualize changes in flow patterns upon Pdgfb loss, we labelled retinas for KLF4, a mechanosensor previously identified as a robust readout of physiological FSS in the retina vasculature [21]. Whereas KLF4 is little expressed in low flow regions-vascular front and capillary ECs in control retinas, in PdgfbiΔEC retinas, KLF4 expression expanded throughout the capillary plexus with more KLF4 + ECs in the capillaries and veins engaged in the direct connections at the vascular front (Fig. 7A and B, quantified in C).

Fig. 7

Capillary enlargement results in pathological flow-mediated Klf4 upregulation. A, B—small insets Co-labelling of P7 Pdgfbfl/fl and PdgfbiΔEC retinas for KLF4 (white) and IB4 (red). Red arrowheads point towards KLF4 + ECs within the enlarged capillaries and in veins of PdgfbiΔEC. C, D Quantification of KLF4 + ECs and of KLF4 labeling intensity per EC in arteries, capillaries and veins from Pdgfbfl/fl and PdgfbiΔEC retinas (Pdgfbfl/fl n = 6 Pdgfbi∆EC n = 6 unpaired 2-tailed t test with Welch’s correction). E, F mRNA expression of Klf4, Bmp6 and Tgf-β1 in mLECs isolated from P7 Pdgfbfl/fl and PdgfbiΔEC mice (Pdgfbfl/fl n = 4/7 Pdgfbi∆EC n = 4/7 unpaired 2-tailed t test with Welch’s correction). G WB for TGF-β1 and BMP6 in mLECs isolated from P7 Pdgfbfl/fl n = 4 Pdgfbi∆EC (H) Acvr1 and Tgfrb1 mRNA expression in mLECs isolated from P7 Pdgfbfl/fl and PdgfbiΔEC mice (Pdgfbfl/fl n = 7, Pdgfbi∆EC n = 7, unpaired 2-tailed t test). I, J Co-imunolabelling for IB4 (red) and BMP6 (white) (I) or TGF-Β1 (white) (J) of P7 Pdgfbfl/fl and PdgfbiΔEC retinas. Red arrowheads point towards BMP6 + and TGF-Β1 + vessels. K, L Quantification of BMP6 (K) and TGF-β1 (L) immunolabelling intensity from P7 Pdgfbfl/fl and PdgfbiΔEC retinas (Pdgfbfl/fl n = 6, Pdgfbi∆EC n = 6, unpaired 2-tailed t test with Welch’s correction). ns non-significant, **P < 0.01, ***P < 0.001. a artery, v vein

Interestingly, measuring KLF4 labeling intensity per single EC as a readout for its activation upon flow, KLF4 appeared more enhanced in capillaries and veins engaged in AV shunts in contrast to control retinas where KLF4 was intensely labeling the high flow arteries and the arterial branch points (Fig. 7D). Upregulation of Klf4 expression was also validated by qPCR in PdgfbiΔEC mLECs (Fig. 7E). Together, these results emphasize that high KLF4 within the enlarged capillaries is likely a consequence of the combination of pathological flow and increased sensitivity of ECs to flow, thus supporting our in vitro data (Fig. 3K–L).

Interestingly, KLF4 was previously identified as a direct transcriptional activator of BMP6 [50], TGF-β1 [51] and Dll4 [52] ligands. As increased Dll4 and NOTCH activation was already confirmed (Fig. 4I, 6K), we performed qPCR and WB for Tgf-β1 and Bmp6 in Pdgfbfl/fl and PdgfbiΔEC mLECs. Both Tgf-β1 and Bmp6 mRNAs and proteins were upregulated upon Pdgfb loss (Fig. 7F, G). Interestingly, Acvr1 and Tgfbr1 coding for the Alk2 and Alk5 receptors were also upregulated (Fig. 7H).To validate these findings, we next labelled retinas for Bmp6 (Fig. 7I) and Tgf-β1 (Fig. 7J). While, the ligands were very little expressed in control retinas, both BMP6 and TGF-β1 showed increased expression at the vascular front and in the AV shunts (Fig.