PI3 kinase inhibition improves vascular malformations in mouse models of hereditary haemorrhagic telangiectasia

1Haemorrhagic telangiectasia (HHT) is an autosomal-dominant inherited vascular disorder that affects ∼1 in 5,000 people. Patients develop multiple focal vascular malformations including capillary telangiectasies and arteriovenous malformations (AVMs). These lesions are fragile and prone to bleeding, and large calibre AVMs cause pulmonary and systemic shunting that can be physiologically significant. More than 95% of HHT cases are caused by mutations in transforming growth factor-β (TGF-β)/bone morphogenetic protein (BMP) signalling pathway genes, including the surface receptors Endoglin (ENG, mutated in HHT1) and ACVRL1 (hereafter referred as ALK1, mutated in HHT2), and the signalling pathway effector SMAD4-juvenile polyposis. ENG and ALK1 proteins are expressed predominantly at the surface of endothelial cells, where they bind members of the TGF-β/BMP family, including TGF-β, BMP9 and BMP10 (refs,, ). ENG acts as an auxiliary co-receptor that promotes signalling through ALK1 (ref. ). Ligand binding activates cytoplasmic regulatory SMADs 1, 5 and 8, which subsequently complex with SMAD4 and translocate into the nucleus to regulate gene expression. Thus, HHT is caused by mutations leading to deregulation in this signalling pathway, but how these mutations induce vascular malformations remains unclear. Identifying such mechanisms could provide novel approaches to prevent vascular malformations in HHT patients.

2Mouse models have provided insights into Eng and Alk1 function. Heterozygous mutations in either gene give rise to vascular lesions, but these form at low frequency and late in life, making them inconvenient to study. Homozygous global Eng and Alk1 gene inactivation in mice and zebrafish leads to embryonic death due to AVMs, again rendering the study of molecular mechanisms difficult. However, postnatal tamoxifen (Tx)-inducible, endothelial-specific homozygous deletion of either gene combined with angiogenic or pro-inflammatory stimuli induces HHT-like vascular malformations including excessive angiogenesis, enlarged veins and AVMs. Thus, loss of both copies of endothelial Alk1 or Eng genes in the context of active angiogenesis is thought to engender vascular lesions.

3Here we investigate AVM formation in the retina and gastrointestinal (GI) tract using Cdh5-CreERT2 mice crossed with Alk1fl/fl (Alk1iΔEC) mice in which endothelial-specific homozygous deletion of Alk1 can be induced postnatally with Tx and blocking antibodies (blABs) against the BMP9/10 ligands. We show that blocking BMP9-ALK1 signalling enhances pro-angiogenic signalling induced by vascular endothelial growth factor (VEGF), the major angiogenic growth factor known to date, but also enhances phosphatidyl inositol 3-kinase (PI3K)-AKT signalling independently of exogenous VEGF. We show that targeting Vegfr2 prevents angiogenesis in Alk1iΔEC mice but does not rescue normal vascular patterning and AVM formation, whereas PI3K inhibition rescues vascular malformations in BMP signalling-deficient retinas and GI tract, identifying PI3K pathway inhibition as a novel putative treatment approach for HHT patients.

4We used postnatal day 5 (P5) Cdh5-CreERT2;Alk1fl/fl mice (Alk1iΔEC) treated with a single dose of 50 μg Tx at P3 and mice treated with BMP9/10 blAB (10 mg kg−1, intraperitoneal (i.p.)) between P2 and P4 (Fig. 1a and Supplementary Fig. 1a). Tx-injected Cre-negative littermates and uninjected wild-type (no antibody) littermates were used as controls. Efficient Alk1 gene deletion was verified by quantitative PCR (qPCR) from mouse lung endothelial cells (mLECs; Supplementary Fig. 1b) and loss of endothelial staining with an anti-Alk1 antibody (Supplementary Fig. 1c,d). BMP9/10 blockade did not affect Alk1 expression (Supplementary Fig. 1c,e).

5At P5, the morphology of retinal and organ vasculature was analysed by injection of latex dye (Fig. 1). Control retinas and the GI tract showed staining of arterial but not venous vessels, because latex cannot perfuse capillaries (Fig. 1b,c,h,i). In contrast, both Alk1iΔEC- and BMP9/10 blAB-treated mice exhibited latex dye staining of veins in retinas (Fig. 1d–g) and the GI tract (Fig. 1j–m). Counterstaining of retinal vessels with IsolectinB4 (IB4) showed that latex perfused AVMs with blood flowing directly from an artery into a vein without an intervening capillary bed (Fig. 1d–g). Quantification revealed an average of three latex-perfused AVMs per retina in Alk1iΔEC- and two AVMs in BMP9/10 antibody-treated mice (Fig. 1n). The GI tract surface of mutants showed an average of seven in Alk1iΔEC and four in BMP9/10 blAB perfused veins per 10 mm gut length, whereas no perfused veins were seen in control mice (Fig. 1o).

6Additional analysis of IB4-stained retinas confirmed previous results that AVMs were usually located in the centre of the retina, close to the optic nerve, whereas increased vessel area and branching were observed at the vascular front in both Alk1iΔEC- and BMP9/10 blAB-treated mice (Supplementary Fig. 1f–k).

7Sorting of endothelial cells from Alk1iΔEC retinas showed that Alk1-deficient cells display enhanced cell cycle progression, with a significant reduction of cells in G1 phase and a concomitant increase of endothelial cells in S/G2/M (Supplementary Fig. 2a). Staining of BMP signalling-deficient retinas for smooth muscle actin (SMA) showed a decrease in arterial and an increase in venous SMA coverage when compared with control retinas, in particular in veins engaged in AVMs (Supplementary Fig. 2b–d). qPCR analysis on mLECs isolated from Alk1iΔEC showed unchanged levels of Vegfr2 but downregulation of Alk1, Eng and Vegfr1, as well as of Notch signalling components Notch1 and Jag1 (Supplementary Fig. 2e), in agreement with a previous report. Arterial markers Unc5b and Efnb2 (refs, ) were decreased, whereas expression of the venous marker EphB4 was upregulated in Alk1 mutant cells (Supplementary Fig. 2e). Antibody staining confirmed that Alk1 mutant cells lose expression of Vegfr1 (Supplementary Fig. 2f,g), a negative regulator of VEGF signalling, and of the Notch ligand Jagged 1 (Supplementary Fig. 2h,i), suggesting that reduced Notch and enhanced VEGF signalling could contribute to AVM formation in Alk1 mutant cells.

8To study VEGF signalling in ALK1-deficient cells, we transfected human umbilical vein endothelial cells (HUVECs) with ALK1 small interfering RNA (siRNA), which decreased ALK1 messenger RNA levels by >90% (Fig. 2a). Compared with Ctrl siRNA, ALK1-depleted cells displayed equal levels of total VEGFR2 but showed an increase in VEGFR2 phosphorylation and activation of downstream pERK and pAKT in response to 10 ng VEGFA (Fig. 2b,c). Of note, pAKT was already increased at baseline in ALK1 knockdown cells (Fig. 2b,c) in the absence of exogenous VEGF.

9We reasoned that blocking the Vegf signal-transducing receptor Vegfr2 (ref. ) might prevent vascular malformations in Alk1 mutant retinal vessels. To block Vegfr2 genetically, we used Vegfr2fl/fl mice intercrossed with Alk1iΔEC mice. One hundred micrograms of Tx was administered at P3 and P4, and Alk1;Vegfr2iΔEC mice were analysed at P5 (Fig. 2d). Tx-injected Cre-negative littermates were used as controls (Fig. 2d,e). Compared with control mice, combined deletion of both Vegfr2 and Alk1 produced varying degrees of vascular defects that correlated with the extent of Vegfr2 deletion (Fig. 2e–k). Alk1;Vegfr2iΔEC mice had a reduced number of AVMs when compared with Alk1iΔEC single mutants, but many retinas still exhibited AVMs (Fig. 2f,l), despite loss of Alk1 expression (Fig. 2i).

10Western blot analysis of Vegfr2 expression (Fig. 2j) in the lungs corresponding to the retinas in Fig. 2e–h showed that mice with inefficient Vegfr2 deletion (33%) could display close to normal retina vascular patterning (Fig. 2g,i). In contrast, mice with efficient Vegfr2 deletion (88%) had strongly reduced retinal angiogenesis, in some cases displaying barely any vasculature (Fig. 2h). Quantification of 16 retinas showed that overall vascular density and branching in double mutants was significantly lower than untreated controls (Fig. 2l), indicating that Vegfr2 inhibition was dominant over Alk1 loss of function and, if complete, prevented retinal vascular development.

11As ALK1 knockdown HUVECs exhibited increased AKT phosphorylation in the absence of VEGF stimulation (Fig. 2b,c), we investigated activation of the PI3K/AKT pathway in HUVECs (Fig. 3a,b) and mLECs (Fig. 3c,d) treated with BMP9. Two hours after treatment, AKT phosphorylation at serine 473 and phosphorylation of downstream FOXO1 were both strongly decreased compared with untreated cells (Fig. 3a–d). Upstream of PI3K-AKT, we found that phosphorylation of the Ser380/Thr382/Thr383-inactive form of the lipid phosphatase and tensin homologue (PTEN) was decreased in BMP9-stimulated HUVECs and mLECs (Fig. 3a–d), suggesting that AKT activity might be repressed by stabilizing PTEN at the cell membrane. Total PTEN levels were also decreased by BMP9 treatment (Fig. 3a,c). We next confirmed that BMP9/ALK1 signalling regulates AKT activation by using ALK1-depleted HUVECs and mLECs or cells treated with anti-BMP9/10 blABs (Fig. 3e–h). Cells were cultured in medium complemented with 2% fetal calf serum containing BMP9 proteins. ALK1-depleted cells and BMP ligand blockade both increased pAKT, pFOXO1 and pPTEN levels (Fig. 3e–h).

12In line with the western blotting results, immunostaining of Alk1iΔEC mLECs showed an increase in PI3K-Akt signalling. Compared with control cells, pFoxo1 was increased in Alk1iΔEC cells (Fig. 3i,j), whereas nuclear Foxo1 staining, which is inhibited by phosphorylation, was decreased (Fig. 3k,l). pPten immunostaining in Ctrl mLECs revealed both nuclear and membrane staining, whereas we observed an increase in nuclear and a decrease in membrane staining in Alk1iΔEC cells (Fig. 3m,n), suggesting that BMP inhibits PI3K-Akt activity via regulation of Pten localization and activity.

13The data suggested that inhibition of PI3K signalling might improve vascular malformations in Alk1iΔEC mice. To test this hypothesis, we injected 50 μg Tx into P3 Alk1iΔEC mice and administered Wortmannin (PI3Ki, 0.4 mg kg−1, i.p.) at P3, P4 and P5 (Fig. 4a). Tx-injected Cre-negative littermates were used as controls and retinas were analysed at P5. Western blotting on whole lung lysates showed that PI3Ki efficiently inhibited AKT phosphorylation, without affecting total AKT levels (Fig. 4b). PI3Ki induced a decrease of retinal vasculature density and the number of branch points when compared with untreated Ctrl mice (Fig. 4c,d). PI3Ki also efficiently normalized the vasculature of Alk1 mutants (Fig. 4f compared with Fig. 4e). Quantification showed that PI3Ki treatment prevented AV shunt formation in Alk1iΔEC retinas and normalized vessel density, as well as the number of branch points close to control levels (Fig. 4g). We also found that PI3K inhibition in Alk1iΔEC normalized the increased proliferation of the endothelial cells, which was assessed by PH3 staining (Fig. 4h).

14We next tested the efficiency of PI3Ki in rescuing AVM formation in the internal organs. In latex-perfused control versus Alk1iΔEC mice, PI3Ki efficiently reduced the number of latex-perfused retinal AVMs and veins on the surface of the GI tracts (Fig. 4i–o).

15To confirm the effects of PI3K pathway inhibition in rescuing AVMs in Alk1iΔEC mice we used another PI3K inhibitor, Pictilisib (PI3Ki 2). Alk1 deletion was induced at P3 with 50 μg Tx and 20 mg kg−1 of PI3Ki2 was administered i.p. at P3–P5 and retinal vasculature was analysed at P5 (Supplementary Fig. 3a). Staining for IB4 shows that inhibition of PI3K activity with PI3Ki2 efficiently prevented the development of AV shunts and rescued vascular defects in Alk1iΔEC retinas (Supplementary Fig. 3b–d).

16PI3Ki treatment also prevented vascular defects induced by BMP9/10 blockade (Fig. 5a–d). BMP9/10 blAB and PI3Ki were administered at P3 and P4, and P3–P5, respectively, and the retinal vasculature was analysed at P5 (Fig. 5a). Staining for IB4 showed that inhibition of PI3K activity efficiently prevented the development of AV shunts and rescued vascular defects induced by BMP9/10 blAB (Fig. 5b–d).

17To confirm the efficacy of PI3K inhibition in vivo, treated versus untreated controls, versus Alk1iΔEC and BMP9/10 blAB-injected mice retinas were labelled for PI3K downstream signalling components (Fig. 6a–m). Endothelial nuclear Foxo1 staining was strongly decreased in Alk1 mutant and BMP9/10 blAB-treated mice (Supplementary Fig. 4a–d). Staining with phospho-S6 (ref. ), which is positively regulated by PI3K-AKT signalling was strongly increased in BMP9/10 blAB- (Fig. 6b,c) and Alk1iΔEC-treated retinas (Fig. 6e,h) when compared with control retinas (Fig. 6a,d). PI3Ki treatment decreased the expression of phospho-S6 in controls (Fig. 6f,h) and in Alk1 mutant retinas (Fig. 6g,h). Furthermore, Eng, which is induced by Alk1 signalling and repressed by PI3K (ref. ), was decreased in Alk1 mutant retinal vessels (Fig. 6j,m) when compared with control retinas (Fig. 6i,m) and the expression was rescued by PI3Ki treatment (Fig. 6l,m).

18To address whether inhibition of PI3K was able to revert already established AVMs, we first determined the time course of AVM formation in Alk1iΔEC- and BMP9/10 blAB-treated retinas. Fifty micrograms of Tx or BMP9/10 blABs were administered at P3 in Alk1iΔEC or control pups, respectively, and retinas were analysed 24 (P4) or 40 h later (P5) (Fig. 7a). Seventy per cent of Alk1iΔEC retinas and 87% of BMP9/10 blAB-treated retinas exhibited at least one AV shunt 24 h after injection (Fig. 7b). Staining for the endothelial nuclear Erg1,2,3 transcription factor and Ve-Cadherin showed an increase in the number of endothelial cells specifically at branches of retinal veins (Fig. 7c–g). In control retinas, these branches contained a single cell (Fig. 7c), whereas Alk1iΔEC or BMP9/10 blAB-treated mice showed two to three cells (Fig. 7d–f). At 40 h, both arterial and venous branches in mutant AVMs exhibited more than three cells compared with one cell in controls (Fig. 7c–g), suggesting that AVMs arise from the enlargement of capillaries and an increase in the cell number at venous and arterial branch points.

19Having established that AVMs were present at P4, we next induced gene deletion with 50 μg Tx at P3 and administered PI3Ki 24 and 40 h later. Pups were killed at P5 (Fig. 7h). Analysis of retinal vasculature showed that PI3Ki efficiently rescued vascular anomalies in Alk1iΔEC retinas (Fig. 7i–k). Likewise, PI3Ki treatment at P4 after induction of BMP9/10 blockade at P3 and P4 also reduced vascular defects (Fig. 7l–o), suggesting that inhibition of PI3K pathway reverts already established AVMs.

20The major novel finding reported here is that inhibition of BMP9 and 10 signalling through the Alk1 receptor leads to overactivation of the PI3K-Akt pathway and that inhibition of PI3K activity rescues vascular malformations in mouse models of HHT2 (Fig. 8). The PI3K/Akt pathway stimulates endothelial cell proliferation, migration and survival downstream of various angiogenic growth factors. We show that pharmacological inhibition of PI3K signalling using two different inhibitors prevents and even rescues abnormal features in BMP signalling-deficient retinas and the GI tract. Inhibition of PI3K in wild-type retinas closely mimics the retinal angiogenesis phenotype in mice lacking the PI3K110α isoform or Akt1 gene confirming treatment specificity. However, PI3K inhibition rescued excessive retinal endothelial cell proliferation, providing one possible mechanism for its effects in vivo. In addition, a recent study showed that AVM formation in zebrafish Alk1 mutants was due to altered endothelial cell migration, which in mouse retinas is dependent on the PI3K110α isoform. Therefore, with increased endothelial cell proliferation, altered migration is likely to be implicated in the Alk1 mutant phenotype and may be targeted by PI3K inhibition. This pathway therefore represents an attractive potential therapy for Alk1-mediated HHT2 disease.

21Mechanistically, we found that enhanced PI3K-Akt pathway activation following BMP9 signalling blockade involves the upstream lipid phosphatase Pten (Fig. 8). BMP9 was previously shown to decrease Pten expression levels in mouse embryonic fibroblasts, which prompted us to investigate Pten as a possible upstream regulator of PI3K-Akt phosphorylation. We confirmed changes in total Pten levels and additionally found more dramatic effects on the C-terminal Pten phosphorylation. Pten catalyses the conversion of phosphatidylinositol 3,4,5-trisphosphate into phosphatidylinositol 4,5-bisphosphate, thereby counteracting PI3K-Akt signalling at the cell membrane. When phosphorylated, Pten assumes a closed conformation with an inactive phosphatase domain that is unable to bind the cell membrane and to dephosphorylate PI3K substrate, thereby increasing PI3K activity.

22Our in vitro experiments show that BMP9 treatment decreases carboxy-terminal PTEN phosphorylation in HUVECs and mLECs, thereby increasing PTEN activity at the cell membrane, leading to reduced phosphorylation of AKT and of downstream FOXO1 transcription factor. Conversely, loss-of-function experiments using Alk1 knockout mLECs or knockdown HUVECs, or BMP9/10 blAB-treated mLECs and HUVECs show the reverse effect: increased PTEN phosphorylation led to increased AKT and FOXO1 phosphorylation. This points to regulation of C-terminal PTEN phosphorylation by BMP9 as the critical step for AKT activation (Fig. 8). We are currently investigating the mechanistic basis for this effect. BMP9/ALK1 might signal via SMAD proteins to repress transcription of PTEN and/or induce kinases targeting the PTEN C terminus.

23PTEN mutations in humans cause vascular anomalies, suggesting PTEN as a major hub to maintain the proper balance between BMP and PI3K signalling pathways. Likewise, overactivation of the PI3K/AKT/mammalian target of rapamycin pathway has been shown to cause vascular anomalies. Activating mutations of AKT/Protein kinase B (ref. ), TIE2 receptor (ref. ) or PIK3CA (ref. ) have been associated with the development of vascular anomalies in mice and humans. Interestingly, the absence of AV shunts in these pathologies, or in Pten-deficient mice retinas suggests that an increase in PI3K signalling by itself is not sufficient to trigger AVMs. One could speculate that increased PI3K signalling leads to AVMs only in the context of reduced Smad or Notch signalling. Regardless of the precise mechanism, the data shown here demonstrate that blocking BMP9/10 signalling induces an increase in PI3K signalling and identify PI3K signalling pathway inhibition as a putative novel therapeutic approach for HHT2 patients.

24PI3K signalling is activated by various growth factors, including VEGF. Our data show that VEGF signalling is increased in ALK1 mutant cells. Several possible causes for enhanced VEGF signalling exist: first, hypersprouting and AVM formation in BMP9/10-Alk1 signalling-deficient retinas probably leads to hypoxia, which is known to increase Vegf levels, and VEGF levels are also increased in human HHT patients. Second, we and others have previously shown that BMP9 treatment leads to an increase in the expression of Notch signalling components Hey1 and Hey2, as well as in Vegfr1, and Notch and Vegfr1 counteract Vegf signalling through Vegfr2 in retina vessels. Chromatin immunoprecipitation sequencing analysis had shown that Smad1,5 binding to the promoter of Hey1 and Hey2 was increased after BMP9 treatment of HUVECs, suggesting that BMP9-Smad directly activates Notch signalling components. We and others also observe that Alk1 mutant endothelial cells show a corresponding decrease in Vegfr1 and Notch signalling components. Although we did not see effects on Vegfr2 levels, Hey2 (HESR1) was previously shown to decrease VEGFR2 promoter-luciferase reporter activity. Decreased VEGFR1 levels may be sufficient to account for the enhanced VEGF signalling shown in Fig. 2b; alternatively, VEGFR2 levels may change transiently at different time points.

25We tested whether blocking excessive VEGF could rescue vascular defects in Alk1 mutants. Administration of Vegf blAB in temporally inducible Alk1 mouse mutants has been shown to block progression of cutaneous AV shunts. Furthermore, VEGF-blAB is currently tested in clinical trials in HHT patients. We show that deletion of Vegfr2, the main signal transducing receptor for Vegf, prevents angiogenesis, but does not fully rescue normal vascular patterning and AVM formation. In fact, the severity of the vascular phenotype in Alk1;Vegfr2iΔEC mice was Vegfr2 dose dependent. We found that Cdh5-CreERT2-mediated Vegfr2 excision was highly variable between mice, as described previously. Proper interpretation of the results thus required correlating Vegfr2 protein levels in mouse lungs to the retinal vascular phenotype in each mouse. Using this approach, we found that strong inhibition of Vegfr2 expression severely impaired the development of retinal vasculature in Alk1 mutants, as previously shown on a wild-type background. Vegfr2 deletion completely abolished the hypervascular phenotype resulting from Alk1 deletion, indicating that Vegf signalling inhibition is dominant and overrides lack of Alk1 signalling. However, the retinas of the Alk1;Vegfr2iΔEC pups still displayed AV shunts, suggesting that additional signalling components may be involved in AVM formation. This idea is also supported by in vitro studies with HUVECs and mLECs, where PI3K signalling is increased in the absence of exogenous VEGF treatment. Thus, the data support the existence of a specific disease-causing enhanced PI3K activity downstream of Alk1 deficiency that is at least partially VEGF independent. Nevertheless, our data show that down-modulation of excessive Vegf signalling normalizes vasculature, supporting potential clinical benefits of anti-VEGF treatment in HHT2 patients.

26Finally, our data confirm that BMP9/10 are physiological Alk1 ligands in mice. In previous studies, blocking of BMP10 in BMP9 mutants or ligand sequestration by Alk1-Fc treatment induced retina hypervascularization, but not AVM formation, leaving open the possibility that additional Alk1 signalling mechanisms might protect from AVM formation. In these studies, treatments were administered to pups at P1, a stage when retina vascularization begins, and arteries and veins have not yet formed. When receptor deletion and ligand blockade are done at P3, a stage when retina arteries and veins have formed and carry flow, AVMs develop, suggesting that flow is required for AVM formation in mice. Furthermore, subjecting endothelial cells to laminar shear stress potentiates signalling induced by low concentrations of BMP9 in vitro. In zebrafish embryos, arresting flow prevents AVM formation in Alk1 mutants, demonstrating that flow is an evolutionary conserved trigger of AVM formation. AVMs always form between the larger arteries and veins close to the optic nerve that carry most blood flow, whereas less perfused vasculature towards the angiogenic front develops hypervascularization. Thus, existing data suggest that BMP9/10 signalling through Alk1 protects vasculature from hypervascularization in low flow conditions and at the same time prevents AVM development induced by higher flow. AVMs form via enhanced cell cycle progression and enlargement of capillary connections between retinal arteries and veins. Simply put, it appears that BMP9/10 signalling through Alk1 protects capillary size at vessel branch points, thereby allowing proper branching morphogenesis of the vasculature.

27All animal experiments were performed under a protocol approved by Institutional Animal Care Use Committee of Yale University.

28Eight weeks old Alk1iΔEC and Alk1;Vegfr2iΔEC mice of mixed genetic background were intercrossed for pups. Offsprings of both genders were used. Gene deletion was induced by intra-gastric injections with 50 μg Tx (Sigma, T5648; 1 mg ml−1) into Alk1iΔEC and 100 μg Tx into Alk1;Vegfr2iΔEC pups at P3 and P4. Tx-injected Cre-negative littermates were used as controls. BMP9/10 blABs (10 mg kg−1 per day) and the PI3K inhibitors were injected i.p.: BMP9/10 blABs between P2–P4, Wortmannin (0.4 mg kg−1 per day) or Pictilisib (20 mg kg−1 per day) once at P3, twice at P4 and once at P5.

29P5 pups were anaesthetized on ice, and abdominal and thoracic cavities were opened. The right atrium was cut, blood was washed out with 2 ml PBS and 1 ml of latex dye (Blue latex, BR80B; Connecticut Valley Biological Supply Co.) was slowly and steadily injected into the left ventricle with an insulin syringe. Retinas and GI tracts were washed in PBS and fixed with 4% paraformaldehyde (PFA) overnight. Retinas were stained with Isolectin B4 and GI tracts were cleared in 10% glycerol, 4 M urea and 0.1% Triton for 1–2 weeks before imaging.

30For immunodetection: anti-Alk1 (AF 770, 1 μg ml−1, R&D systems), anti-Eng (AF1320, 1 μg ml−1, R&D systems), anti-Jagged1 (AF599, 1 μg ml−1, R&D systems), anti-Vegfr1 (AF471, 2 μg ml−1, R&D systems), anti-Erg1/2/3 (SC353, 2 μg ml−1, Santa Cruz), anti-α-SMA CY3 (CY3-SMA, C6198, 1:200 working solution, Sigma), anti-phosphoS6 (5364, 1:200 working solution, Cell Signalling), anti-CD144 (555289, 2 μg ml−1, BD Pharmingen), IB4 (121412, 10 μg ml−1, Life Technologies), anti-pFOXO1 (9464, 1:100 working solution, Cell Signalling), anti-FOXO1 (2880, 1:100 working solution, Cell Signalling), anti-pPTEN (9549, 1:100 working solution, Cell Signalling), anti-PTEN (9188, 1:100 working solution, Cell Signalling) and Dapi (D1306, 1 μg ml−1, Life Technologies).

31For western blotting: anti-Vegfr2 (9698, 1:2,000 working solution, Cell Signalling), anti-pVegfr2 (2478, 1:2,000 working solution, Cell Signalling), anti-pAKT (4060, 1:1,000 working solution, Cell Signalling), anti-AKT (4060, 1:1,000 working solution, Cell Signalling), anti-p-p44/42 mitogen-activated protein kinase (phospho-ERK, 9106, 1:2,000 working solution, Cell Signalling), anti-p44/42 mitogen-activated protein kinase (total ERK, 9102, 1:2,000 working solution, Cell Signalling), anti-pFOXO1 (9464, 1:1,000 working solution, Cell Signalling), anti-FOXO1 (2880, 1:1,000 working solution, Cell Signalling), anti-pPTEN (9549, 1:1,000 working solution, Cell Signalling) and anti-PTEN (9188, 1:1,000 working solution, Cell Signaling). Appropriate secondary antibodies were fluorescently labelled (Alexa Fluor donkey anti-rabbit, Alexa Fluor donkey anti-goat, Alexa Fluor donkey anti-rat, 1:400 working solution, Invitrogen) or conjugated to horseradish peroxidase (Anti-Rabbit IgG (H+L), 1:8,000 working solution, Vector Laboratories).

32BMP9/10 blABs were from Genentech and PI3K inhibitors Wortmannin S2758 and Pictilisib S1065 were purchased from Selleckchem.

33Retinas were harvested at P5, dissected in cold PBS and digested in 1 mg ml−1 Collagenase type II (Sigma) in DMEM (Sigma). Cells were incubated with CD31-APC (BD 551262, 2 ng ml−1) and CD45-FITC (11-0452-85, 5 ng ml−1, eBioscience), Hoechst (B2261, 25 μg ml−1, Sigma Aldrich) and Pyronin Y (P9172, 200 ng ml−1, Sigma Aldrich). CD31+/CD45− endothelial cells were isolated by FACS and further analysed for their cell cycle distribution by two-dimensional analysis of Hoechst and Pyronin Y fluorescence signal.

34MLECs were isolated from collagenase-digested lung tissue using rat-anti-mouse CD31 monoclonal antibody-coated Dynabeads (11035, Invitrogen). Primary mLECs were cultured in DMEM supplemented with 10% fetal bovine serum (FBS), 100 U ml−1 penicillin, 100 μg ml−1 streptomycin, 100 μg ml−1 endothelial cell mitogen (Biomedical Technologies, Inc.) and 10 U ml−1 heparin for 10 days.

35The eyes of P5/P6 pups were prefixed in 4% PFA for 20 min at room temperature (rt). Retinas were dissected, blocked for 30 min at rt in blocking buffer (1% fetal bovine serum, 3% BSA, 0.5% Triton X-100, 0.01% Na deoxycholate, 0.02% Na azide in PBS at pH 7.4) and then incubated with specific antibodies (mentioned above) in blocking buffer overnight. The following day retinas were washed and then incubated with IB4 and corresponding secondary antibody for 1 h at rt and mounted in fluorescent mounting medium (DAKO, Carpinteria, CA, USA).

36Endothelial cells in culture were fixed in 4% PFA for 15 min, permeabilized in 0.1% Triton X-100 in PBS for 10 min, blocked in 3% BSA (AB01088-00100, AmericanBIO) and incubated with the indicated primary antibodies overnight at 4 °C (mentioned above) followed by 1 h incubation with appropriate secondary antibodies at rt.

37High-resolution pictures were acquired using a Leica SP5 confocal microscope with a Leica spectral detection system (Leica 15 SP detector) and the Leica application suite advanced fluorescence software. Quantification of retinal vasculature was done using ImageJ.

38HUVEC cells were obtained from the Yale University Vascular Biology and Therapeutics Core Facility and cultured in EGM2-Bullet kit medium (CC-3156 & CC-4176, Lonza). Depletion of ALK1 was achieved by transfecting 25 pmol siRNA (Qiagen, mixture of S102659972 and S102758392) with Lipofectamine RNAiMax (Invitrogen), following the manufacturer's instructions. Transfection efficiency was assessed by western blotting and qPCR. Experiments were performed 48 h post transfection and results were compared with siRNA Ctrl (ON-TARGETplus Non-targeting Pool D-001810-10-05)

39Cells were lysed with Laemmli buffer including phosphatase and protease inhibitors (Thermo Scientific, 78420, 1862209). Twenty micrograms of proteins were separated on 4–15% Criterion precast gel (567-1084, Bio-Rad) and transferred on nitrocellulose membrane (Biorad). Western blottings were developed with chemiluminescence horseradish peroxidase substrate (Millipore, WBKLS0500) on a Luminescent image Analyser, ImageQuant LAS 4000 mini (GE Healthcare). Bands were quantified using ImageJ. See Supplementary Fig. 5 for the uncropped immunoblots.

40RNAs from HUVEC or from mLECs were purified using RNeasy-kit (Qiagen). One RNA was reverse transcribed using SuperScript III (Invitrogen) and quantitative PCR were assayed using the corresponding primers (Qiagen): MmAcvrl1 (QT00161434), MmEng (QT00148981), MmUnc5b (QT00167846), MmFlt1 (QT00096292), MmNotch1 (QT00156982), MmJag1 (QT00115703), MmEfnb2 (QT00139202) and MmEphB4 (QT00120295). The expression levels were normalized to MmActin (QT00095242) and MmCdh5 (QT00110467).

41The authors declare that the data supporting the findings of this study are included within the article and its Supplementary Information files, or are available from the authors upon request.

42How to cite this article: Ola, R. et al. PI3 kinase inhibition improves vascular malformations in mouse models of hereditary haemorrhagic telangiectasia. Nat. Commun. 7, 13650 doi: 10.1038/ncomms13650 (2016).

43Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.