Angiopoietin-2 Inhibition Rescues Arteriovenous Malformation in a Smad4 Hereditary Hemorrhagic Telangiectasia Mouse Model

Abstract

Background:

Hereditary Hemorrhagic Telangiectasia (HHT) is an autosomal dominant vascular disorder caused by heterozygous, loss-of-function mutations in four TGFβpathway members, including the central transcriptional mediator of the TGFβ pathway, Smad4. Loss of Smad4 causes the formation of inappropriate, fragile connections between arteries and veins called arteriovenous malformations (AVM), which can hemorrhage leading to stroke, aneurysm or death. Unfortunately, the molecular mechanisms underlying AVM pathogenesis remain poorly understood and the TGFβ downstream effectors responsible for HHT-associated AVM formation are currently unknown.

Methods:

To identify potential biological targets of the TGFβ pathway involved in AVM formation, we performed RNA- and ChIP-sequencing experiments on BMP9 stimulated endothelial cells (ECs) and isolated ECs from a Smad4 inducible, EC specific knockout (Smad4-iECKO) mouse model that develops retinal AVMs. These sequencing studies identified the Angiopoietin-Tek signaling pathway as a downstream target of SMAD4. We utilized monoclonal blocking antibodies to target a specific component in this pathway and assess its effects on AVM development.

Results:

Sequencing studies uncovered 212 potential biological targets involved in AVM formation, including the EC surface receptor, TEK (TEK receptor tyrosine kinase) and its antagonistic ligand, ANGPT2 (angiopoietin-2). In Smad4-iECKO mice, Angpt2 expression is robustly increased, while Tek levels are decreased resulting in an overall reduction in Angiopoietin-Tek signaling. We provide evidence that SMAD4 directly represses Angpt2 transcription in ECs. Inhibition of ANGPT2 function in Smad4 deficient mice, either before or after AVMs form, prevents and alleviates AVM formation and normalizes vessel diameters. These rescue effects are attributed to a reversion in EC morphological changes, such as cell size and shape that are altered in the absence of Smad4.

Conclusions:

Our studies provide a novel mechanism whereby loss of Smad4 causes increased Angpt2 transcription in ECs leading to AVM formation, increased blood vessel calibers and changes in EC morphology in the retina. Blockade of ANGPT2 function in an in vivo Smad4 model of HHT alleviated these vascular phenotypes further implicating ANGPT2 as an important TGFβ downstream mediator of AVM formation. Therefore, alternative approaches that target ANGPT2 function may have therapeutic value for the alleviation of HHT symptoms, such as AVMs.

Introduction

Hereditary Hemorrhagic Telangiectasia (HHT) is an autosomal dominant vascular disorder that affects 1 in 5,000 people worldwide, regardless of sex or race ¹. HHT symptoms include uncontrolled nosebleeds ², mucosal telangiectasias (dilated small blood vessels) and arteriovenous malformations (AVMs), which are enlarged, inappropriate connections between arteries and veins. In HHT, AVMs commonly form in major organs such as the liver, lung and brain, where they are prone to hemorrhaging and can lead to stroke or death ¹. HHT is genetically linked to germline, heterozygous loss-of-function mutations in TGFβ family pathway members ACVRL1, ENG, SMAD4 and BMP9 ^3–8. Mutations in ACVRL1, a Type I receptor, and ENG, an accessory receptor, comprise approximately 85% of all HHT cases. A smaller percent of patients (~4%) have mutations in the SMAD4 transcription factor and present with a combined Juvenile Polyposis syndrome (JP/HHT), while less than 1% of patients have mutations in the BMP9 ligand ^6–8.

In general, the TGFβ signaling cascade is activated when a ligand binds to a heteromeric complex of two type I and two type II receptors. These receptor complexes then recruit and phosphorylate receptor-regulated smads (R-SMADs). Typically, R-SMADs 1/5/8 are activated by BMP and anti-Muellerian receptors, while R-SMADs 2/3 are activated when TGFβ, Activin and Nodal receptors are involved ⁹. These R-SMADs form a trimeric complex with the common smad, SMAD4, to exert transcriptional activities distinct to the particular receptor-ligand interaction.

To date, the overwhelming focus in HHT research has been restricted to TGFβ cell surface components in the BMP9-ACVRL1-ENG signaling axis ^10–19. Despite SMAD4 being a central mediator of the TGFβ pathway and a direct cause of HHT itself, very little is known about downstream SMAD4 transcriptional targets in HHT, or in endothelial cells (ECs) in general. Additionally, approximately 10–15% of HHT patients have no mutations in ACVRL1, ENG, SMAD4 or BMP9 indicating that unidentified, downstream TGFβ target genes could also be genetically linked to HHT ^{20, 21}.

We previously developed a Smad4 inducible, EC specific knockout (Smad4-iECKO) mouse model that exhibited consistent and robust retinal AVMs ²², similar to that of Acvrl1 and Eng EC specific knockout mice ^{10–12, 23}. Additional vascular phenotypes in Smad4-iECKO mice closely mimicked those observed in HHT patients: increased blood vessel diameters, which is a hallmark trait of AVMs and telangiectasias ^{24, 25}, alterations in smooth muscle cell vessel coverage ^{24, 26, 27} and reduced pericyte coverage, which increases the fragility and chance of rupture in vessels ²⁸ and is hypothesized to be tied to HHT pathogenesis ²⁹. Based upon these phenotypic characteristics, Smad4-iECKO mice represented a suitable model to elucidate the molecular mechanisms underlying HHT.

Using Smad4-iECKO isolated retinal ECs and BMP9 stimulated ECs, we employed next generation sequencing based strategies, including RNA- and ChIP-sequencing, to uncover TGFβ effectors with roles in HHT-related AVM formation. We identified the Angiopoietin-Tek signaling pathway, which is critical to vascular development and linked to human diseases including venous malformations ^30–33, as a key contributor to retinal AVM pathogenesis. We find that in Smad4-iECKO mice the expression levels of the cell surface receptor, Tek receptor tyrosine kinase (Tek), are decreased, while its antagonistic ligand, Angiopoietin-2 (Angpt2), are strongly increased leading to a net reduction in TEK receptor signaling. ANGPT2 monoclonal antibodies administered to Smad4-iECKO mice resulted in the prevention and resolution of retinal AVMs, as well as the normalization of vessel diameters. Together, these findings point to ANGPT2 as a novel mediator of AVM formation in HHT.

Materials and Methods

The data, methods, and study materials will be made available to other researchers for purposes of reproducing the results by requesting information or materials from the corresponding author.

Mouse husbandry and injections

All animal experiments were performed in accordance with Tulane University’s Institutional Animal Care and Use Committee policy. The Smad4-iECKO mouse model was developed as previously described ²². Briefly, an endothelial-specific, tamoxifen-inducible Cre-driver line (Cdh5-Cre^ERT2) ³⁴ was crossed with a conditional Smad4 mouse (Smad4^f/f) ³⁵. Mice were genotyped using Quanta AccuStartII master mix with the following primers: Smad4 Fwd 5’-TAAGAGCCACAGGGTCAAGC-3’, Smad4 Rev 5’-TTCCAGGAAAAACAGGGCTA-3’; Cdh5 Fwd 5’-TCCTGATGGTGCCTATCCTC-3’, Cdh5 Rev 5’-CCTGTTTTGCACGTTCACCG-3’. Genotyping PCRs were run on a BioRad C1000 Touch or T100 Touch thermal cycler. The Cre-LoxP system was activated through intragastric tamoxifen injections at postnatal day 1 (P1) in Smad4^f/f;Cdh5-Cre^ERT2 (otherwise referred to as Smad4-iECKO) and Smad4^f/f littermate controls. We used 0.075 mg tamoxifen diluted in 10% ethanol and sunflower seed oil (Sigma T5648) per gram body weight of pup. ANGPT2 inhibitor (LC-10, Roche) was injected using 30 μg concentration into either P2 or P6 neonatal mice.

Endothelial Cell Isolation

Endothelial cell isolation was performed as previously described ²². For RNA sequencing experiments cells were isolated from retinas and RNA purification was done immediately. For all experiments using isolated lung endothelial cells, the cells were purified using sheep anti-rat IgG dynabeads (Invitrogen 11035) coated with PECAM antibody (BD 553370) and cultured for 7 days in EGM-2 (Lonza CC-4176) media.

RNA Sequencing

RNA was isolated using Thermo Fisher GeneJET RNA Purification Kit (Thermo K0732) and quantified using a Nanodrop (Thermo). RNA quality was assessed using Agilent RNA 6000 Nano kit (Agilent 5067–1511). Messenger RNA libraries were made with a starting concentration of 1 μg RNA using an Illumina TruSeq RNA Library Preparation Kit v2 (Illumina RS-122–2001). Library quality and quantity was assessed using Agilent DNA 1000 chip (Agilent 5067–1504) and Qubit DNA HS kit respectively. Libraries were sequenced using NextSeq 500/550 High Output v2 kit (Illumina FC-404–2004). Data was analyzed using Illumina Basespace RNA-Seq Alignment App, which utilizes STAR aligner, and DESeq2 for differential expression analysis. Criteria for significance was set at a false discovery rate of 0.05. RNA-seq data was deposited in NCBI’s Gene Expression Omnibus (GSE116230): https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE116230

ChIP Sequencing

ChIP samples were obtained using SimpleChIP (Cell Signaling #56383) and processed according to manufacturer specifications. Anti-SMAD4 antibody (Cell Signaling 46535) was used for immunoprecipitation. Cells were serum starved for 24 hours before addition of fresh media or 1 ng/mL BMP9/10 recombinant protein (R&D 5566 and 6038). Library preparations were done using TruSeq ChIP Library Preparation Kit (Illumina IP-202–1012). 10 ng starting material was used for library prep. DNA quantity was measured using Qubit DNA HS kit (Thermo 32851). Libraries were sequenced using NextSeq 500/550 High Output v2 kit. Library quality and quantity was assessed using Agilent DNA 1000 chip (Agilent 5067–1504) and Qubit DNA HS kit respectively. ChIP-Seq analysis was done using the ChIPSeq App from BaseSpace Labs (Illumina), which utilizes MACS2 for region enrichment and HOMER for motif analysis. IGV Viewer was used to generate figures. Criteria for significance was set at a false discovery rate of 0.05. Evolutionarily conserved regions were identified using ECRBrowser. ChIP-seq data was deposited in NCBI’s Gene Expression Omnibus (GSE115921): https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE115921

Statistical Analysis

GraphPad Prism software was used for all statistical analyses and to generate graphs. Sample size (n) indicates the number of independent biological samples. A minimum of three technical replicates was included per sample. Results were confirmed via blinding experiments whereby confocal images were assigned arbitrary numbers and quantified by an additional individual with no knowledge of genotype or treatment. For statistical analysis, we ran either one-way ANOVA with Tukey’s multiple comparison test applied or Mann Whitney U tests. A p value of < 0.05 was considered significant.

Immunofluorescent staining of the murine retina

Retinas were dissected and stained as previously described ^{22, 36}. Briefly, retinas were fixed in 4% paraformaldehyde for one to two hours at 4 °C and either dehydrated and stored in 75% ethanol or stained immediately. If stored, retinas were rehydrated in a step-wise fashion and washed with PBS. Retinas were then permeabilized for 30 minutes in PBS + 1% Triton (PBST) and blocked in CAS-Block (Invitrogen 008120) for 30 minutes. Retinas were stained with the following primary antibodies at a 1:100 concentration overnight: anti-Tek (RD AF762), anti-αSMA (Sigma C6198), anti-PECAM (BD 553370). Isolectin-IB4 (Invitrogen 32450, 21411, 21413) was also incubated with primary antibody overnight at a 1:200 concentration. Retinas were washed with PBS and placed into appropriate Alexa Fluor secondary antibodies diluted 1:500 in PBST for 4 hours before mounting. Retinas were mounted using Prolong Diamond antifade mountant (Invitrogen P26961) and imaged with a Nikon A1 confocal microscope. All retina analysis was performed using Nikon NIS-Elements analysis software. AVM was defined as an enlarged vessel, beyond the size of a typical capillary that directly connects an artery to a vein. For retinal vascular outgrowth, four points of reference from the optic nerve to the furthest vessel in each leaflet were measured in each retina. To measure artery and vein diameters, three vessels were measured in proximal, medial and distal positions in each retina. To measure EC areas, three fields of view were assessed in each retina and a minimum of 5 cells were measured in each image.

Western Blot

For western blots, protein was collected using RIPA buffer (Life Technologies 89900), protease inhibitor (Thermo 1860932) and phosphatase inhibitor (Sigma 04906837001) and quantified using Qubit protein assay kit (Thermo Q33211). SDS-PAGE was run using 25–50 μg of protein mixed 1:1 with lamelli buffer (BioRad 161–0737) and heated to 90°C for 10 minutes and loaded onto a Mini-PROTEAN TGX gel (BioRad 456–8094). After gel electrophoresis, samples were transferred to TransBlot Turbo Transfer Pack 0.2 μm PVDF membrane (BioRad 1704156) using TransBlot Turbo (BioRad). Membrane was blocked using 5% BSA blocking solution for 30 minutes then primary antibodies were added at a 1:100 concentration overnight at 4C. The primary antibodies used were as follows: Anti-TEK (RD AF762), Anti-ANGPT2 (RD AF623), Anti-SMAD4 (Abcam ab40759), Anti-βACTIN (Cell Signaling 3700). After washing, appropriate LICOR IRDye 800CW or 680RD secondary antibodies were used at 1:1000 and incubated at room temperature for 1 hour. Membranes were imaged using Odyessy (LICOR) and iStudio Lite was used for western quantification.

Quantitative Polymerase Chain Reaction (qPCR)

For qPCR experiments, RNA was isolated using Thermo Fisher GeneJET RNA Purification Kit (Thermo K0732) and quantified using a Nanodrop (Thermo). For each sample, 500 ng RNA was used for cDNA synthesis. cDNA was synthesized using iScript cDNA synthesis kit (Bio-Rad 1708891). Samples were run using SYBR green master mix (Thermo K0221) on a Bio-Rad CFX96 Touch Real-Time PCR Detection machine. For all experiments, the samples were run in triplicate and the ΔΔCt method was used to analyze results. Primers were designed using NCBI primer designing tool and were assessed for specificity and efficiency. The following primers were used in experiments: PECAM Fwd: 5’-CTGCAAGTCCGAAAATGGAAC-3’; PECAM Rev: 5’-CTTCATCCACCGGGGCTATC-3’; Tek Fwd: 5’-GTAAACAAGAGCGAGTGGACC-3’; Tek Rev: 5’-CCATGGCGCCTTCTACTACT-3’; Angpt2 Fwd 5’-TGACAGCCACGGTCAACAAC-3’; Angpt2 Rev 5’-ACGGATAGCAACCGAGCTCTT-3’; ODC Fwd 5’-ACCGTGCTGTGAGTGTTTCC-3’; ODC Rev 5’-TGTGGCAGTCAAACTCGTCC-3’.

Results

Genomic studies in endothelial cells reveal Smad4 targets related to HHT and the TGFβ pathway

We previously developed a mouse model of HHT whereby inducible, EC specific knockout of Smad4 (Smad4-iECKO) led to arteriovenous malformation (AVM) formations ²². Conditional deletion of Smad4 at postnatal day 1 (P1) results in retinal AVM formation, increased blood vessel diameters and reduced vascular outgrowth by P7 (Fig. 1a-b) ^{22, 37}. To identify the molecular causes of AVM formation, we performed RNA sequencing (RNA-Seq) on isolated retinal endothelial cells (iREC) collected from P7 Smad4^f/f (control) and Smad4-iECKO mice revealing 1,905 differentially expressed genes: 1,095 upregulated and 810 downregulated (Fig. 1c-d). Consistent with previous studies, expression levels of several genes, such as Angpt2, Apln, Col4a1 and Notch4, were upregulated during AVM formation (Fig. 1d) ^{22, 38}. Furthermore, Gene Ontology (GO) analyses showed enrichment in processes thought to influence AVM development including collagen-activated signaling, formation of cellular junctions and cell adhesion (Supp Fig. 1a-b) ³⁹.

Fig. 1. Genomic studies reveal endothelial SMAD4 targets related to HHT and the TGFβ pathway.

(a) Graphical representation of Cre-LoxP system components and Smad4 depletion timeline. (b) Representative Isolectin-IB4 staining of retinal blood vessels in Smad4^f/f control and Smad4-iECKO mutant animals depicting presence of AVMs (arrows). White dotted circle represents outgrowth of Smad4^f/f control mouse. A, arteries; V, veins. Scale bar represents 500 µm. (c) Heatmap depicting differential expression of genes in isolated retinal ECs (iREC) of Smad4^f/f and Smad4-iECKO animals. (d) Graphical plot of 1,095 upregulated and 810 downregulated genes in Smad4-iECKO compared to Smad4^f/f iREC. (e) Venn Diagram showing overlap between SMAD4 binding sites in unstimulated and 24 hour BMP9/10 stimulated Ms1 ECs. (f) Doughnut diagram of SMAD4 bound gene locations in unstimulated versus BMP9/10 stimulated ECs. (g) Top SMAD4 binding motifs in unstimulated (red) and BMP9/10 stimulated (green) ECs arranged by family. (h) Representative view of four SMAD4 binding motifs. Percentages below motifs are the percentage of binding sites where motifs were found in unstimulated and BMP9/10 stimulated ECs, respectively. (i,j) Highlighted SMAD4 binding peaks (red box) within the Eng and Acvrl1 mouse genes in unstimulated (grey) and BMP9/10 (blue) stimulated ECs. Evolutionary conserved regions between chimpanzee, human and mouse Eng and Acvrl1 genes show Smad4 binding peaks in Eng are located within non-coding evolutionary conserved regions (ECRs). Notice the lack of a SMAD4 binding site in the Acvrl1 gene.

As SMAD4 is the transcriptional mediator of the TGFβ pathway and is located downstream of both ACVRL1 and ENG, we wanted to assess its direct transcriptional role in AVM pathogenesis by identifying SMAD4 binding loci in ECs. Because BMP9/10 ligands play a direct role in HHT pathology by binding directly to ACVRL1 and ENG receptors, and mutations in BMP9 lead to an HHT-like disease in humans ^{40, 41}, we conducted chromatin immunoprecipitation sequencing (ChIP-Seq) experiments on BMP9/10 stimulated and unstimulated mouse ECs. SMAD4 binding sites were found within or near 613 genes in unstimulated conditions and 1,806 genes in BMP9/10 stimulated conditions with 490 genes shared between the two datasets (Fig. 1e). As anticipated, GO analysis of both datasets revealed strong TGFβ association, whereas BMP9/10 stimulated ECs showed enrichment in R-SMAD binding indicating stimulation of the canonical TGFβ pathway (Supp Fig. 1c-f). Interestingly, BMP9/10 stimulation also caused slight increases of SMAD4 localization in promoter regions and decreases in intergenic regions compared to unstimulated ECs (Fig. 1f). Further analysis of SMAD4 bound regions revealed previously established SMAD4 DNA motifs (5’-GTCT-3’), as well as SMAD2 and 3 motifs ⁴². Unexpectedly, over 20% of target sequences contained ETS family binding motifs suggesting that SMAD4 may interact with ETS family members in ECs (Fig 1.g-h), a potentially novel connection in the endothelium.

Interestingly, we observed SMAD4 binding sites on the Eng gene but not the Acvrl1 gene, which supports our previous studies where Eng levels were downregulated but Acvrl1 expression was not affected when Smad4 was lost (Fig 1. i-j) ²². This suggests a direct transcriptional interplay with Eng and not Acvrl1 that may have implications in different HHT etiologies.

Smad4 deficiency leads to mis-regulation of Angiopoietin-Tek pathway signaling components

Given the lack of knowledge of downstream effectors in HHT, we aimed to identify genes downstream of SMAD4 that are critical in AVM formation by integrating the RNA- and ChIP-Seq datasets. This revealed 212 overlapping direct, downstream targets of SMAD4 (Fig. 2a, Supp Table 1). Interestingly, two major components of the Angiopoietin-Tek signaling pathway were identified: the cell surface receptor, TEK and its antagonistic ligand, ANGPT2. In our Smad4-iECKO mouse model, RNA-Seq experiments revealed an upregulation and downregulation in Angpt2 and Tek transcripts respectively (Fig. 1d). Quantification of transcript levels in both iRECs and isolated lung ECs (iLECs) confirmed appreciable increases of Angpt2 and downregulation of Tek in Smad4 deficient ECs (Fig. 2b-d). To assess localization changes of these transcripts, we performed in situ hybridization on P7 retinas. In P7 retinas, Angpt2 is normally expressed at very low levels in the growing vascular front but when Smad4 is lost ⁴³, Angpt2 becomes highly upregulated in this region but is absent from AVMs (Fig. 2e, Supp Fig. 2a). In contrast, Tek is normally expressed in arteries, veins and capillaries, but excluded from the vascular front ^{43, 44}. In the absence of Smad4, Tek levels throughout the vasculature are diminished, especially near the peripheral vasculature, but Tek accumulates within the AVM (Fig. 2f, Supp Fig. 2b). Similarly, we found increased ANGPT2 and decreased TEK protein levels in iLECs from Smad4-iECKO mice compared to controls (Fig. 1g). Because ANGPT2 is normally secreted from ECs and its levels were drastically elevated, we reasoned that the circulating levels of ANGPT2 in the blood might be higher in Smad4-iECKO mice as well. Indeed, we found a significant increase in soluble ANGPT2 serum levels in Smad4-iECKO compared to Smad4^f/f mice (Fig. 2h). Taken together, these results indicate that loss of Smad4 leads to significant overexpression of ANGPT2 and decreased levels of TEK, resulting in an overall net reduction in TEK receptor signaling.

Fig. 2. SMAD4 deficiency leads to changes in Angpt2 and Tek expression in the endothelium.

(a) Venn Diagram depicting overlap between the isolated retinal endothelial cell (iREC) RNA-Seq and BMP9/10 stimulated Ms1 cell SMAD4 ChIP-Seq (BMP9/10 Stim ChIP) datasets revealed 212 direct, downstream targets of SMAD4. (b-d) Verification of increased Angpt2 and decreased Tek transcript levels in iRECs (b-c) and isolated lung endothelial cells (iLECs; d) of Smad4-iECKO versus Smad4^f/f animals. All values are normalized to Pecam mRNA levels. Quantifications were performed in triplicate for each n. iREC: Pecam, n = 4; Tek, n = 4; Smad4, n = 3; Angpt2, n = 4. iLECs: Pecam, n = 3; Tek, n = 3; Smad4, n = 4; Angpt2, n = 3. (e,f) In situ hybridization analysis of Angpt2 and Tek transcripts in Smad4^f/f and Smad4-iECKO P7 retinas, followed by Isolectin-IB4 (IB4) immunofluorescent staining. (e) Angpt2 transcripts are barely detectable in Smad4^f/f retinas, but levels are increased dramatically at the growing vascular front of Smad4-iECKO mice. Note Angpt2 absence in AVM (asterisk). (f) In Smad4^f/f retinas, Tek is expressed in all vessels excluding ECs at the vascular edge, whereas upon loss of Smad4, Tek levels are decreased in the vascular front and are present within the AVM (asterisks). (g) Western blot analysis verifies increased ANGPT2 and decreased TEK and SMAD4 protein levels in Smad4-iECKO versus Smad4^f/f iLECs. (h) Smad4-iECKO mice exhibit increased serum levels of ANGPT2 compared to Smad4^f/f mice. (i) ChIP-Seq analysis on unstimulated and BMP9/10 stimulated mouse Ms1 ECs revealed three SMAD4 binding peaks upstream of the Angpt2 gene. All three sites are within non-coding evolutionarily conserved regions (ECRs) between chimpanzees and humans. (j) Closeup view of SMAD4 binding sites upstream of Angpt2 (Peaks 1–3; red boxes). (k) ChIP-qPCR using SMAD4 and IgG antibodies on unstimulated and BMP9/10 stimulated Ms1 ECs verified SMAD4 binding enrichment on all three peaks upstream of Angpt2 (n = 2). Control DNA primers outside of the SMAD4 binding regions show no enrichment. Primer regions and control (C) are depicted by red lines in (j). (l) Quantification of luciferase activity normalized to renilla expression in stable Ms1 cell lines expressing nonsilencing-shRNA (NS-shRNA) or Smad4-shRNA (n = 3 for all experiments; 3 readouts were analyzed per n). Sample size (n) indicates biological replicates. Error bars represent mean ± standard error. Statistics were generated using the Mann Whitney U test. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

SMAD4 directly regulates transcription of Angpt2 in the endothelium

To further understand the transcriptional relationship between the TGFβ and Angiopoietin-Tek signaling pathways, we searched for SMAD4 binding sites on the Angpt2 and Tek genes. We found one SMAD4 binding site in the first intron of the Tek gene and three SMAD4 binding sites upstream of the Angpt2 gene. Interestingly, all SMAD4 binding sites upstream of Angpt2 were located in non-coding evolutionary conserved regions (ECR) of mammalian genomes, while the SMAD4 binding site in Tek was not evolutionarily conserved (Fig. 2i-l and Supp Fig. 3a). Given the biological relevance of ECRs in conserved transcriptional mechanisms, we focused further studies on the evolutionarily conserved Smad4 bound regions within Angpt2. ChIP-qPCR confirmed SMAD4 binding to all three ChIP-seq identified regions in Angpt2 in both BMP9/10 stimulated and unstimulated conditions (Fig. 2k). Although SMAD4 has been shown to enhance, activate and repress gene transcription, we hypothesized that SMAD4 acts as a repressor of Angpt2 because loss of SMAD4 caused increased Angpt2 expression. To this end, we performed luciferase reporter assays in a constitutively active promoter (CMV) driven vector (pGL4.50) to assess transcriptional activity of each 50–100 bp Smad4 binding peak (Peaks 1–3). In stable nonsilencing-shRNA (NS-shRNA) control ECs, all three regions elicited substantial increases in luciferase activity compared to empty pGL4.50 suggesting that all three regions are transcriptionally active in ECs (Fig. 2l). However, stable knockdown of Smad4 in ECs (Smad4-shRNA) showed increased luciferase activity beyond NS-shRNA levels for all three peaks tested indicating repressive transcriptional activity of SMAD4 (Fig. 2l). Together, our data establishes SMAD4 as a direct transcriptional repressor of Angpt2 expression in ECs.

Inhibition of ANGPT2 can prevent AVM formation in Smad4-iECKO mice

Since ANGPT2 levels were greatly elevated in Smad4 deficient mice, we postulated that inhibition of ANGPT2 would result in improved HHT associated phenotypes by re-establishing Angiopoietin-Tek signaling in the vasculature. To test this idea, we used an ANGPT2 inhibitor (LC10) that has demonstrated high affinity binding for mouse ANGPT2 and inhibits its function in vitro and in vivo ^{43, 45}. To assess if LC10 could prevent AVM development, we administered LC10 or control IgG to Smad4^f/f and Smad4-iECKO mice one day after tamoxifen injections (P2) but before retinal AVM formation (Fig. 3a). In mice injected with IgG control, approximately 43% of Smad4-iECKO mice exhibited AVMs whereas no AVMs were observed in Smad4^f/f mice (Fig. 3b-c, f). Consistent with previous reports, all Smad4-iECKO mice injected with IgG control also displayed decreased vascular outgrowth, increased arterial and venous diameters and ectopic expression of α−SMOOTH MUSCLE ACTIN (αSMA) on veins (Fig. 4b-e, Supp Fig. 4b) ^{22, 37}. Similar to a previous study showing that a derivative of LC10 (LC06) resulted in decreased retinal vascular outgrowth⁴³, Smad4^f/f mice injected with LC10 exhibited reduced outgrowth compared to Smad4^f/f mice injected with IgG (Fig. 3b and d) further confirming that ANGPT2 inhibition negatively affects developmental angiogenesis. Even with these adverse effects, when LC10 was administered to Smad4-iECKO mice, the previously observed retinal phenotypes remarkably failed to form: no AVMs were observed and both artery and vein diameters remained the same size as controls (Fig. 4f-h). Furthermore, αSMA expressing smooth muscle cells, which are abundant around the retinal veins in Smad4-iECKO mice injected with IgG, were found to be largely restricted to arteries in Smad4-iECKO mice treated with LC10 (Supp Fig. 4b). Taken together, these data showed that ANGPT2 inhibition in a Smad4 depleted background prevents the main retinal vascular phenotypes associated with this model: AVM formation and increased vessel calibers.

Fig. 3. ANGPT2 inhibition prevents AVM formation in Smad4-iECKO mice by impeding changes in endothelial cell size and shape.

(a) Timeline depicting experimental procedures including tamoxifen (Tx) injections, drug injections (LC10 or IgG), approximate time AVMs form and retinal collection. (b-e) Smad4^f/f and Smad4-iECKO P7 retinas injected with 30 μg of IgG (b,c) or LC10 (d,e), and stained for Isolectin-IB4 (magenta) and αSMA (blue). White dotted circles represent outgrowth of Smad4^f/f injected with IgG (b). (c) Smad4-iECKO P7 retinas injected with IgG revealed increased vessel diameters and AVMs (arrows). (d) Smad4^f/f P7 retinas injected with LC10 exhibited a reduction in vascular outgrowth. (e) Smad4-iECKO P7 retinas injected with LC10 did not exhibit AVMs or increased vessel diameters. A, arteries; V, veins. Scale bars represent 500 μm. (f-i) Quantification of AVM number (f), retinal vascular outgrowth (g), and vein (h) and artery (i) diameters from experimental retinas in b-e. Notice that administration of LC10 prevents vessel enlargement associated with SMAD4 loss (Smad4-iECKO + IgG versus Smad4-iECKO + LC10). (j-m) Close up images of retinal veins stained with PECAM to mark EC junctions/boundaries revealed changes in size and morphology of ECs in Smad4^f/f and Smad4-iECKO P7 retinas injected with IgG versus LC10. Several cells are uniformly enlarged below in black to highlight cell morphological changes. Scale bars represent 50 μm. (n,p) Quantification of venous and arterial EC areas. Administration of LC10 prevents endothelial cell enlargement. (o,q). Shape factor quantification of venous and arterial ECs. A cell is given a numerical value of 1 to 0 based on its shape: circle = 1; elongated line = 0. (o) Loss of Smad4 causes venous ECs to acquire an elongated, arterial shape, whereas administration with LC10 prevents this phenotype. Sample size (n) indicates biological replicates which were used for the corresponding quantifications. Mice were taken from three separate litters. Error bars represent mean ± standard error. Statistics were generated using ordinary one-way ANOVA with Tukey’s multiple comparison test applied. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

Fig. 4. ANGPT2 inhibition rescues AVM formation and EC morphology in Smad4-iECKO mice.

(a) Timeline depicting experimental procedures including tamoxifen (Tx) injections, approximate time AVMs form, drug injections (30 μg of LC10 or IgG) and retinal collection. (b-e) Smad4^f/f and Smad4-iECKO P8 retinas injected with either IgG or LC10 and stained with Isolectin-IB4 (magenta) and αSMA (blue). Scale bars represent 500 μm. White dotted circles represent outgrowth of Smad4^f/f injected with IgG. Compared to Smad4-iECKO mice treated with IgG (c), administration of LC10 to Smad4-iECKO mice results in a substantial reduction in the number of retinal AVMs and normalization of artery and vein calibers (e). (f-i) Quantification of retinal AVM number (f), vascular outgrowth (g), and vein (h) and artery (i) diameters from experimental retinas in b-e. Notice that the increases in vessel caliber in Smad4-iECKO mice treated with IgG revert to control sizes (Smad4^f/f + IgG) when LC10 is administered to Smad4-iECKO mice. (j-m) Close up images of retinal veins stained with PECAM to mark EC junctions/boundaries revealed changes in size and morphology of ECs in Smad4^f/f and Smad4-iECKO P8 retinas injected with IgG versus LC10. Several cells are uniformly enlarged below in black to highlight cell morphological changes. Scale bars represent 50 μm. (n,p) Quantification of venous and arterial EC areas. The increased cell areas observed in Smad4-iECKO mice injected with IgG revert to control sizes when LC10 is given to Smad4-iECKO mice (o,q). Shape factor quantification of venous and arterial ECs. A cell is given a numerical value of 1 to 0 based on its shape: circle = 1; elongated line = 0. (o) Loss of Smad4 causes venous ECs to acquire an elongated, arterial shape, whereas administration with LC10 significantly reverses this phenotype and cells return to a more cuboidal shape. Sample size (n) indicates biological replicates which were used for the corresponding quantifications. Mice were taken from three separate litters. Error bars represent mean ± standard error. Statistics were generated using ordinary one-way ANOVA with Tukey’s multiple comparison test applied. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

ANGPT2 inhibition prevents changes in EC size and shape caused by loss of Smad4

To assess whether cell morphological changes occur when ANGPT2 is inhibited, we stained retinas with PECAM, which marks EC junctions and provides insight into EC size and shape. We previously showed that Smad4-iECKO mice exhibit increases in vessel size due, in part, to an increase in EC cell size, which is similar to findings observed in mouse coronary arteries and human aorta lacking SMAD4 ⁴⁶, and in blood vessels of eng homozygous mutant zebrafish ^{16, 22, 47}. Consistent with our prior findings, Smad4-iECKO mice injected with IgG exhibited significant increases in both artery and vein EC areas (Fig. 4i, k, m, o; Supp. Fig. 4b). In addition to increases in cell area, both arterial and venous ECs exhibited noticeable changes in shape upon loss of SMAD4. Venous cells, which are normally cuboidal (rounded) in shape, acquired an elongated, arterial EC-like phenotype when SMAD4 is lost (Fig. 3i-j). Arterial cells, on the other hand, are normally columnar in shape; when SMAD4 is lost, arterial ECs elongate further (Fig. 3k-l, Supp Fig. 4c). Conversely, ANGPT2 inhibition largely prevented the EC size and shape changes caused by the absence of SMAD4. In venous ECs, LC10 administration resulted in cell size and morphology similar to that of Smad4^f/f mice injected with IgG (Fig. 3m-n). Similar results were observed in arterial ECs when LC10 was injected into Smad4-iECKO mice: EC areas were indistinguishable compared with Smad4^f/f mice treated with control IgG (Fig. 3o) and further elongation of cell shapes were inhibited (Fig. 3p). Interestingly, when LC10 is administered to Smad4^f/f mice, both venous and arterial EC areas significantly increase (Fig. 4j, m, o). These results suggest that impeding ANGPT2 has different effects on cell morphological properties depending on the context of EC Smad4 levels. Nevertheless, our data demonstrated that ANGPT2 inhibition counteracts Smad4 dependent changes in EC size and shape, and subsequently prevents in vivo AVM formation and dilation of blood vessels.

Alterations in EC size and shape occur in the absence of flow and are rescued when ANGPT2 is inhibited in vitro

Previous in vitro experiments showed that loss of Smad4 could cause significant increases in EC sizes independent of flow ²². Therefore, we aimed to understand how ANGPT2 inhibition, via LC10, worked irrespective of flow using the NS-shRNA and Smad4-shRNA Ms1 cell lines. First, we confirmed that Smad4-shRNA cells recapitulated gene expression changes seen in Smad4-iECKO mice: mRNA levels of Smad4 and Tek, were decreased, while Angpt2 was significantly increased compared to NS-shRNA cells (Supp. Fig. 5a). LC10 or IgG was then added to the cells for 24 hours and cell size and shape were analyzed (Supp. Fig. 5b-d). In Smad4-shRNA cells treated with IgG, increases in cell area were consistently detected, similar to Smad4-iECKO mice injected with IgG (Supp. Fig. 5b,c). Likewise, Smad4-shRNA cells subjected to LC10 exhibited partial but significant reductions in EC sizes compared to IgG (Supp. Fig. 5b,c). In regards to cell morphology, inhibition of ANGPT2 also had rescuing affects on EC shape. Using a program that measures cell shapes, LC10 treatment showed partial rescue of EC morphology in Smad4-shRNA cells (Supp. Fig. 5b,d), while measurements of cell length over width pointed to a more complete rescue with LC10 (Supp. Fig. 5b,e). Interestingly, although inhibition of ANGPT2 largely rescued cell shape changes, exogenous ANGPT2 alone did not appear to influence cell shape and had modest effects on increasing cell size (Supp. Fig. 5e-h). Together, these experiments revealed that ANGPT2 inhibition in a Smad4 deficient background can reduce cell areas and improve cell morphologies in the absence of flow.

Inhibition of ANGPT2 rescues AVMs in Smad4-iECKO mice

AVMs are thought to arise en utero or during early childhood development, thus making preventative strategies problematic for HHT patients. Therefore, approaches that alleviate AVMs after development would be more clinically feasible. To assess whether inhibition of ANGPT2 could improve HHT phenotypes, such as reducing size or severity of AVMs, we administered LC10 at P6, approximately 2 days after AVMs have formed and when the vascular network is more mature. Retinas were collected at P8 to allow LC10 48 hours to exert its affects (Fig. 4a; Supp Fig. 6a). When administering IgG to Smad4^f/f and Smad4-iECKO mice, we observed the same phenotypes as previously described (Fig. 3). Namely, Smad4-iECKO mice injected with IgG developed robust AVMs and exhibited reduced vascular outgrowth, increased artery and vein diameters and ectopic accumulation of αSMA expressing cells on mutant veins (Fig. 4b-c, Supp Fig. 6b). Conversely, when LC10 was administered, the reduction in vascular outgrowth seen in Smad4^f/f mice was effectively mitigated suggesting that ANGPT2 inhibition has reduced effects on this process at later stages (Fig. 4d, f, Fig. 4d, f). More impressively, AVMs were nearly absent in Smad4-iECKO mice injected with LC10 and the overall vasculature appeared relatively normal. In fact, we only identified two AVMs, which were smaller in diameter than most AVMs observed in Smad4 mutants treated with IgG (Supp Fig. 6d). Additionally, vascular outgrowth extended to nearly control lengths, while artery and vein diameters, and αSMA expression patterns returned to normal (Fig. 4b-h, Supp Fig. 6b-c). These data indicated that inhibition of ANGPT2 after AVM development can alleviate retinal vascular abnormalities caused by a lack of SMAD4.

ANGPT2 inhibition rescues adverse cell morphological effects caused by loss of Smad4

Since ANGPT2 inhibition alleviated AVM formation and other adverse vascular affects at later stages, we hypothesized that EC morphological properties would likewise be rescued. As anticipated, ANGPT2 inhibition after AVM formation yielded a similar phenotype as ANGPT2 inhibition before AVM formation: specifically, cell area and shape in both veins and arteries reverted to control dimensions (Fig. 4i-p, Supp Fig. 6d). Moreover, when LC10 was administered to Smad4^f/f control mice, no appreciable negative effects were observed suggesting that inhibition of ANGPT2 has increased, negative effects when administered during angiogenic development. Based upon our observations, we conclude that inhibition of ANGPT2 normalizes EC sizes and shapes, and subsequently alleviates AVM formation and reverts vessel diameters in a Smad4 mouse model of HHT.

Discussion

While the BMP9/10-ACVRL1-ENG signaling axis has garnered the majority of attention in the HHT field, TGFβ downstream effectors and the molecular mechanisms underlying HHT-related AVM formation remain largely unresolved. Therefore, we aimed to identify direct, downstream effectors of the TGFβ pathway that are relevant to HHT-mediated AVM formation. Our data linked the TGFβ and Angiopoietin-Tek signaling pathways via SMAD4 transcriptional inhibition of Angpt2. We demonstrated that AVMs could be prevented and rescued in our Smad4-iECKO mouse model by inhibiting ANGPT2 function.

Previous studies on various HHT models have suggested a relationship whereby loss of TGFβ signaling leads to increased ANGPT2 levels but failed to provide a direct mechanistic link: Acvrl1 germline deletion showed increased Angpt2 expression in brain and spinal AVMs ¹⁵, BMP9/10 inhibition led to overexpression of Angpt2 in the neonate retina ³⁸ and EC specific deletion of Smad4 led to increased embryonic ANGPT2 expression ⁴⁸. Our studies confirm these findings in a SMAD4 postnatal retinal mouse model of HHT and establish a direct mechanistic link between TGFβ and Angiopoietin-Tek signaling pathways whereby SMAD4 is directly responsible for Angpt2 transcriptional inhibition (Fig. 5). Comparable to our findings, gene profiling of human sporadic brain AVM tissue revealed substantial increases in Angpt2 and decreased Tek expression suggesting a more expansive impact of this pathway on non-HHT related AVM development ^{49, 50}.

Fig 5. Model: SMAD4 directly represses Angpt2 transcription in ECs to maintain a normal vascular network and prevent HHT associated phenotypes.

SMAD4 is a direct repressor of Angpt2 transcription in ECs. Loss of SMAD4 causes increased Angpt2 expression and leads to changes in EC size and shape that results in blood vessel dilation and AVM formation. Inhibition of ANGPT2 function can prevent and alleviate AVM formation and vessel caliber increases by regulating EC size and shape.

However, we note that not all studies are in accordance with our findings. For example, ANGPT2 levels in blood plasma from Acvrl1 HHT patients were decreased, while levels in Eng HHT patients were not affected ⁵¹. Also, cultured blood outgrowth endothelial cells (BOEC) of Acvrl1 and Eng patients, and Eng^+/− mice exhibited reduced ANGPT2 expression ⁵². Various possibilities for these discrepancies include the method of BOEC culture, the time period in which blood plasma was collected (active angiogenesis vs quiescence), or simply the possibility that Smad4 mutations lead to a slightly different phenotype then Acvrl1 or Eng mutations. Given Smad4’s converging importance in the TGFβ pathway, if the latter explanation is correct, it would have substantial implications in the field. However, this prospect seems less likely due to the similar vascular phenotypes shared between murine Acvrl1, Eng, Bmp9/10 and Smad4 models ^{10–12, 14, 23, 36–39, 53–55}. Alternatively, it is possible that the circulating levels of ANGPT2 have less physiological importance on AVM formation than the actual local effects of ANGPT2 on the AVM tissues themselves. Therefore, it will be important going forward to investigate Angiopoietin-Tek signaling in AVM tissues of HHT patients.

In our Smad4-iECKO model, ANGPT2 inhibition effectively prevented and alleviated AVM formation, and led to the normalization of EC morphological defects. Eng mutant zebrafish exhibited similar cellular defects, including increased EC size and defective morphology. These studies concluded that AVMs arise in Eng mutants due to an inability of ECs to respond to hemodynamic-responsive mechanisms that normally limit vessel caliber ¹⁶. In support of this, it has been demonstrated that blood flow is required for AVM formation in Acvrl1 mutant zebrafish ¹⁷. It will be interesting to see how ANGPT2 inhibition affects other HHT phenotypes such as increased proliferation ^{10, 11, 23, 36, 37}, defective migration ¹⁸, and pericyte coverage ^{10, 11, 37, 53} that are common in Acvrl1, Eng and Bmp9/10 mammalian models of HHT. Additionally, in order to map out a more complete molecular pathway of HHT and to understand how certain ligand-receptor interactions in the TGFβ pathway influence Angpt2 transcription, further research is needed to address the relationships between SMAD4 and transcriptional cofactors, including R-SMADs and non-canonical binding partners such as Ets.

Although our Smad4-iECKO retinal AVM model has provided an excellent starting point for uncovering the molecular mechanisms involved in AVM formation, future studies will need to focus on aspects of AVM pathogenesis that are more clinically relevant. This includes the molecular examination of AVMs that form in biologically relevant organs such as the lung, liver and brain, and the generation of HHT models that mimic mature AVMs commonly found in patients. These salient features will allow a more thorough platform for preclinical testing of potential drugs targets, such as those that target ANGPT2 function.

Currently, HHT patients have no proven pharmacological options for AVM treatment. Typically, patients must undergo invasive surgical procedures to remove or reduce AVMs. Our studies provide evidence illustrating the effectiveness of ANGPT2 as a novel biological target to improve HHT-related AVMs. Phase I clinical trials of ANGPT2 inhibitor, MEDI3617, have shown that it is well tolerated among patients ⁵⁶ and nesvacumab is currently in Phase II trials for wet age-related macular degeneration (AMD) treatment ⁵⁷. In addition to ANGPT2 inhibition alone, drugs that simultaneously inhibit ANGPT2 and activate TEK signaling, such as the ABTAA antibody, have shown positive results in murine sepsis models and in normalization of the tumor vasculature ^{58, 59}. Therefore, drugs that target ANGPT2 function may hold promising therapeutic value for treating HHT symptoms such as AVM, epistaxis or telangiectasia.

Clinical Perspective.

What is new?

• Integration of comprehensive RNA- and ChIP-sequencing genomic screens revealed SMAD4 target genes in endothelial cells.

• Loss of SMAD4 causes defects in Angiopoietin-Tek signaling through decreased TEK and increased ANGPT2 levels.

• SMAD4 is a transcriptional repressor of Angpt2 in endothelial cells; loss of this repression represents a novel regulatory mechanism governing AVM formation.

• AVM formation is accompanied by increased endothelial cell size and elongated shape, leading to an overall increase in vessel diameter.

• Inhibition of ANGPT2 can rescue and prevent AVM formation in Smad4 mutant animals by rescuing endothelial cell size and shape defects.

2). What are the clinical implications?

• We demonstrated that targeted inhibition of ANGPT2 holds potential therapeutic value for treating HHT-related AVMs.

• ANGPT2 inhibition may have beneficial implications in non-HHT related, sporadic AVMs, which also exhibit decreased and increased levels of TEK and ANGPT2 respectively.

• Using next generation sequencing methods, we provide a list of potential therapeutic candidates for HHT targeted therapy.