Arterial-Lymphatic-Like Endothelial Cells Appear in Hereditary Hemorrhagic Telangiectasia 2 and Contribute to Vascular Leakage and Arteriovenous Malformations

Abstract

BACKGROUND:

Arteriovenous malformations (AVMs) are characteristic of hereditary hemorrhagic telangiectasia. Loss-of-function mutations in the activin receptor‐like kinase 1 (Alk1) are linked to hemorrhagic telangiectasia type 2.

METHODS:

Endothelial-specific deletion of Alk1, endothelial lineage tracing, transcriptomics of single-cell analysis, and electron microscopy were performed to examine the vascular phenotype and characteristics of ALK1-deficient endothelial cells (ECs) after EC-specific Alk1 deletion. Ischemia assays were used to examine the cell capacity for vascular malformation. Connectivity Map with transcriptomic analysis was applied to identify chemical compounds. Specific methods for arteriovenous malformations, such as micro–computed tomography, with other molecular and cell biological tools were also performed.

RESULTS:

We performed endothelial-specific deletion of Alk1 in mice and found severe arteriovenous malformations and vascular leakage. The transcriptomics of single-cell analysis revealed a new distinctive cell cluster formed after Alk1 deletion where the cells coexpressed arterial and lymphatic endothelial markers. The analysis projected that these cells potentially originated from arterial ECs after Alk1 deletion. This new population was referred to as arterial-lymphatic-like ECs according to its cellular markers, and its appearance was validated in the pulmonary small arteries after Alk1 deletion. Transplantation of these cells caused vascular malformations. Endothelial lineage tracing confirmed that these new arterial-lymphatic-like ECs were derived from ALK1 depleted ECs, potentially arterial ECs. We discovered that SOX17 (SRY-box transcription factor 17) induction was responsible for the derivation of these arterial-lymphatic-like ECs. We showed that direct binding of MDM2 (mouse double minute 2) was required for Sox17 to execute this activity. Inhibition of MDM2 reduced the arteriovenous malformations in the mouse model.

CONCLUSIONS:

Together, our studies revealed the mechanistic underpinnings of ALK1 signaling in regulating the endothelial phenotype and provided possibilities for new therapeutic strategies in hemorrhagic telangiectasia type 2.

Clinical Perspective.

What Is New?

• Arterial-lymphatic-like endothelial cells appear in hemorrhagic telangiectasia type 2 and contribute to arteriovenous malformations.

• Arterial-lymphatic-like endothelial cells may be derived from ALK1 (activin receptor‐like kinase 1)–depleted arterial endothelial cells.

• The derivation of arterial-lymphatic-like endothelial cells was mediated by FLT4, SOX17 (SRY-box transcription factor 17), MDM2 (mouse double minute 2), and PROX1.

What Are the Clinical Implications?

• Preventing ALK1-depleted arterial endothelial cells from acquiring lymphatic endothelial characteristics would provide a new therapeutic strategy in hemorrhagic telangiectasia type 2.

Hereditary hemorrhagic telangiectasia (HHT) is a vascular disorder characterized by malformations of blood vessels.^1–3 The vascular malformations appear as small telangiectasia in the skin and mucous membranes or as large arteriovenous malformations (AVMs) in multiple organs, such as lungs, brain, and liver.^1–3 AVMs are abnormal vessels that transport blood directly from the arteries to the veins through arteriovenous shunts, which can cause severe complications in patients with HHT.^1–3 Arteriovenous shunts in the lungs may result in hypoxemia, dyspnea, high risk of paradoxical embolic strokes, and cerebral abscesses.⁴ In the brain, because of elevated blood pressure in the arteriovenous shunts, the vessels may ultimately rupture and cause hemorrhagic strokes, often associated with high mortality.⁵ Arteriovenous shunts in the liver may lead to high-output cardiac failure, liver failure, or portal hypertension.⁶ AVMs are also a major reason for gastrointestinal bleeding causing severe iron deficiency anemia.⁷

More than 600 different mutations identified within 3 genes, Endoglin, activin receptor‐like kinase (Alk1), and mothers against decapentaplegic homolog 4 (Smad4), have been linked to HHT.⁸ HHT is categorized by the mutations in these genes. Mutations in the Endoglin gene result in HHT type 1, mutations of Alk1 result in HHT type 2 (HTT2), and mutations in Smad4 result in a combination of HHT and juvenile polyposis.^9–11 ALK1 and Endoglin belong to the TGFβ (transforming growth factor β) and BMP (bone morphogenic protein) superfamily, and SMAD4 is a member of the SMAD family of transcription factors that mediates TGFβ and BMP signaling.^12,13 ALK1 is functionally linked to both Endoglin and SMAD4. Endoglin is a coreceptor of ALK1 for the activation of BMP9 or BMP10 signaling.¹⁴ SMAD4 binds to other phosphorylated SMADs to facilitate the transduction of signals from ALK1 or other TGFβ/BMP type I receptors.¹⁵ Previous studies have demonstrated that ALK1, Endoglin, and SMAD4 all play important roles in endothelial cell (EC) differentiation, and deletion of any of these genes in ECs causes AVMs,^16–18 suggesting an essential role of ECs in AVMs.

ECs are critical for maintaining normal vasculature, in which arterial, capillary, and venous ECs line the vessels and coordinate with other vascular cells to build efficient networks for supplying blood, exchanging substrates, and returning blood through the veins. ALK1 is predominantly expressed in arterial ECs and lack of ALK1 impairs the differentiation of arterial ECs.^16,19 Interestingly, previous studies have also shown that ALK1 is involved in the lymphatic vascularization and revealed that the activation of ALK1 inhibits lymphatic vascular formation.^20–23 However, it is unclear if ALK1 serves as a bridge between arterial and lymphatic endothelial lineages. It is also unknown how loss function of ALK1 affects mature arterial ECs.

In this study, we discovered a new population of arterial-lymphatic-like ECs that appear after Alk1 deletion. We also discovered a novel mechanism that potentially triggered arterial ECs to acquire unwanted lymphatic endothelial characteristics. We found that these arterial-lymphatic-like ECs cause vascular malformations and vascular leakage. Finally, we identified a compound that ameliorates AVMs by erasing the unwanted characteristics from the arterial ECs in a mouse model.

METHODS

Data Availability

The data for single-cell RNA-sequencing (scRNA-seq) were available and deposited in the Gene Expression Omnibus (GEO) database (GSE242741 and GSE269174).

Animals

VE-cadherin^cre/ERT2 mice and Alk1^flox/flox mice were obtained as gifts from Drs Ralf Adams and S. Paul Oh, respectively. Mouse double minute 2 (Mdm2^±) mice were obtained from The Jackson Laboratory (stain No. 002968). All the mice are on C57BL/6J background. Genotypes were confirmed by polymerase chain reaction (PCR),²⁴ and experiments were performed with filial generations F4 to F6. Littermates were used as controls. All mice were fed a standard chow diet (Diet 8604, Harlan Teklad Laboratory). Tamoxifen (Sigma-Aldrich, T5648, 75 mg/kg daily) was injected for 5 days. HLI373 (Sigma-Aldrich, 373226) was injected (5 µg/g daily) as in previous studies.²⁵ The studies were reviewed and approved by the institutional review board and conducted in accordance with the animal care guidelines set by the University of California, Los Angeles. The investigation conformed to the Guide for the Care and Use of Laboratory Animals of the National Research Council.²⁶

Tissue Culture

Human pulmonary arterial ECs (HPAECs) (Thermo Fisher Scientific, catalog No. C-008-5C) were cultured as previously described.²⁷ The transfection of ALK1 small interfering RNA (siRNA), SRY-box transcription factor 17 (SOX17) siRNA, or MDM2 siRNA (Thermo Fisher Scientific, ID No. 12, s34628, 122295) was performed as published.²⁷ Treatments with HLI373 (Sigma-Aldrich, catalog No. 373226), VEGF-C, VEGF-165, and VEGF-C (Cys156Ser; R&D Systems, catalog No. 9199-VC, BT-VEGF, and 752-VC) were performed as described in the Results section. Lentiviral vectors containing cytomegalovirus-SOX17 or SOX17 small hairpin RNA were purchased from GeneCopeia and applied to the cells as per the manufacturer’s protocols.

Micro–Computed Tomography

Micro–computed tomography (micro-CT) was performed by Scanco USA, Inc (Southeastern, PA). The perfusion with the MICROFIL compound and the preparation of the specimens were performed as described in detail previously.^28,29 All the samples were scanned on a high-resolution, volumetric micro-CT scanner (μCT 40, Scanco Medical). The image data were acquired with the following parameters: 10-μm isotropic voxel resolution, 200-ms exposure time, 2000 views, and 5 frames per view. The micro-CT–generated Digital Imaging and Communication in Medicine files were used to analyze the samples and to create volume renderings of the regions of interest. The raw data files were viewed using the MicroView 3-D volume viewer and analysis tool (GE Healthcare).

Blue Latex Dye and Tissue Clearing

After the mice were euthanized, latex dye (blue latex, Connecticut Valley Biological Supply Co, catalog No. BR80B) was injected into the right ventricle with a 25-gauge 1-mL syringe. The stained lungs were dissected and washed in PBS, and then fixed in 10% formalin for 24 hours. The lungs were dehydrated in graded methanol (20%, 40%, 60%, 80%, and 100%, 2 hours each), and then incubated with organic solvent (benzyl alcohol/benzyl benzoate, 1:1; Sigma-Aldrich) for clearing. The imaging was performed using the Z-stack function of a Leica M205 FCA fluorescence stereo microscope (Leica Microsystems).

RNA Analysis

Real-time PCR analysis was performed as previously described.³⁰ Glyceraldehyde 3-phosphate dehydrogenase (Gapdh) was used as a control gene.³⁰ Primers and probes for mouse Alk1, Sox17, Prox1, Lyve1, and Mdm2 were obtained from Applied Biosystems as part of Taqman Gene Expression Assays.

Single-Cell RNA-Sequencing

Freshly harvested lungs of mice were enzymatically dissociated with 5 mg/mL of dispase in PBS with 2% FBS at room temperature for 1 hour. Then tissues were minced and incubated on a shaker (100 rpm) in 50 µg/mL of Deoxyribonuclease I for 30 minutes. The cell suspension was washed and filtered with 70-µm strainers. Cells were incubated with anti–vascular endothelial (VE)-cadherin (BD Pharmingen, catalog No. 562243) and anti-CD45 (BD Pharmingen, catalog No. 553080) antibodies. VE-cadherin⁺CD45^- or tdTomato⁺CD45^- cells were sorted by fluorescence-activated cell sorting (FACS) before scRNA-seq analysis.

10× Library Preparation, Sequencing, and Alignment

The scRNA-seq library was generated with the Chromium Single Cell 3′ v3 assay (10x Genomics). The libraries were sequenced using Illumina NovaSeq 6000 platform with the depth of 275 to 590 million reads. Raw reads were aligned to the mouse genome (mm10). The cellranger (v3.0.1) mkfastq function was used to generate FASTQ files, and gene counts were called using the cellranger count function.

Cell Clustering and Cell Type Annotation

Scanpy v1.9.3 was used to process scRNA-seq data and data visualization. Doublets were identified on pretrained single cell variational inference models using a semisupervised deep learning model Solo on each dataset. After doublets were removed, cells were filtered to have >200 detected genes and <15% of mitochondrial genes. Then the data were integrated and trained using a deep learning neural network model scvi.model.SCVI from single cell variational inference (scVI) tools. Dimension reduction was performed using Uniform Manifold Approximation and Projection (UMAP) and the cells were clustered using the Leiden algorithm. The cell identities were defined by known marker genes.

Pseudotime Trajectory Construction

R package Monocle3 (v1.3.1) was used for pseudotime trajectory construction. The EC lineages from the processed data were further analyzed to construct single-cell trajectories. The cells were ordered along a learned trajectory and pseudotime. The expression dynamics of genes of interest were also plotted along the pseudotime.

RNA Velocity

RNA velocity analysis was performed by integrating the measurements of both spliced and unspliced transcripts with temporal dynamics in single-cell gene expression using python package scVelo (v0.3.2). This analysis (velocity stream plot) inferred the differentiation direction of the cell clusters. Partition-based graph abstraction velocity graph (with connectivity [dashed] and transitions [solid/arrows]) further mapped the transition of the cell clusters.

Connectivity Map and Compound Identification

From each cell type in the scRNA-seq analysis, we selected the top 100 marker genes. Connectivity Map (CMap) from the Broad Institute (Massachusetts Institute of Technology and Harvard University) is a database with a collection of gene expression profiles that were obtained from 9 human cell lines treated with various small compounds. To identify a small molecule from 2837 compounds in the LINCS database (The Library of Integrated Network-based Cellular Signatures), we used the next-generation CMap (CLUE) platform (Broad Institute) to directly explore the connectivity of gene signatures among perturbagens using the Touchstone tool (Broad Institute). The drug similarity was ranked according to the CMap connectivity score (from −100 to 100). Connectivity scores >95 or below −95 were considered to be strong scores for predicting whether small candidate molecules corresponded to the query of selected gene expression signatures between different cell types or clusters.

Fluorescence-Activated Cell Sorting

FACS analysis was performed as previously described.³⁰ The cells were stained with fluorescein isothiocyanate–, phycoerythrin-, or Alexa Fluor 488–conjugated antibodies against LYVE1 (Thermo Fisher Scientific, catalog No. 12-0443-82) and Ephrin B2 (R&D Systems, catalog No. AF496). Nonspecific fluorochrome- and isotype-matched IgGs (BD Pharmingen) served as controls.

Immunoblotting and Immunofluorescence

Equal amounts of tissue or cell lysates were used for immunoblotting, which was performed as previously described.³⁰ We used β-actin as a loading control for immunoblotting to examine several proteins in the experiments. Because the size of some proteins was close to β-actin or the same secondary antibodies were used to examine both the targeted protein and β-actin, the signal of both the targeted protein and β-actin may be disrupted if performed on the same membrane. In addition, stripping of the blots may decrease their quality. To avoid these issues, we prepared several individual membranes loaded with equal amounts of protein for each set of samples. We then performed immunoblotting using individual membranes probed with individual antibodies to ensure as much as possible loading of the same amount of protein. The blots were incubated with specific antibodies to ALK1, SOX17, LYVE1, FLT4, PROX1, and MDM2 (Abcam, catalog Nos. ab51870, ab191699, ab183501, ab154079, ab124910, and ab226939). β-Actin (Sigma-Aldrich, catalog No. A2228) was used as a loading control. Immunofluorescence was performed as previously described in detail.³¹ We used specific antibodies to PROX1, SOX17, LYVE1, and CD31 (Thermo Fisher Scientific, catalog Nos. PA5-85552, PA5-72815, 12-0443-82, 14-0311-82), Ephrin B2 (R&D Systems, AF496), and VE-cadherin (BD Biosciences, 562243). The nuclei were stained with 4’,6-diamidino-2-phenylindole (Sigma-Aldrich, catalog No. D9564).

Immunoprecipitation

Antibodies were cross-linked to Dynabeads Protein G (Thermo Fisher Scientific). A total of 40 μL Dynabeads was incubated with antibodies for 1 hour before washing. Dynabeads-antibody complexes were then added to cell extracts isolated from the lungs of mice with gentle rotation at 4 °C for 4 hours. Beads were subsequently washed 3 times with 10 mM Tris-HCl pH 7.5, 50 mM KCl. Elution was performed by heating the beads for 10 minutes at 70 °C in 20 μL NuPage loading buffer (Invitrogen) containing 100 mM Dithiothreitol, or SDS elution buffer (2% SDS, 100 mM Tris-HCl pH 7.5, 10% glycerol, 0.5 mM EDTA, 100 mM DTT, and bromophenol blue as tracer). After removal of the supernatant, the procedure was repeated. The supernatants were combined. The eluates were analyzed by immunoblotting using equal amounts of cell lysates. The blots for the inputs were the controls that ensured that equal amounts of protein were used in the immunoprecipitation. The specific antibodies to ALK1, FLT4 (ab154079), SOX17, MDM2 (Abcam, ab51870, ab154079, ab191699, and ab226939), VEGF-C, and FLT4 (AFL4; Invitrogen, PA5-29772 and 14-5988-82) were used.

Mouse Surgery

The murine model of hindlimb ischemia was performed as previously described.³² A 10-mm-long incision of the skin was made toward the medial thigh. The femoral artery was exposed and separated from the femoral vein and nerve. Silk sutures were used to tie the proximal and distal ends of the femoral artery with double knots. Cells to be tested (5×10⁵) were transplanted into the surgical area, and the incision was closed. Laser Doppler perfusion imaging was used to monitor the blood flow at different time points. Histology and immunostaining were used to examine the vascularization after transplantation.

Specimens From Human AVMs

Deidentified specimens were obtained from the Department of Pathology, David Geffen School of Medicine at University of California, Los Angeles. The specimens were not obtained specifically for this research, and none of the investigators involved in the research were able to ascertain the identity of the subjects. Sections of the specimens were subjected to immunofluorescence staining as described in the immunofluorescence section using the same antibodies. Three cases of deidentified human AVMs were used in this study.

Transmission Electron Microscopy

Transmission electron microscopy (TEM) was performed as previously described.³³ Dissected tissues were immersed in a solution containing 2% glutaraldehyde and 2% paraformaldehyde in 0.1 M PBS, pH 7.4, for 2 hours at room temperature, then incubated at 4 °C overnight. The next day, 0.5% of tannic acid was added to the tissues and incubated for 1 hour at room temperature. The tissue blocks were subsequently washed 5 times in 0.1 M PBS and postfixed in a solution of 1% OsO₄ in PBS, pH 7.2 to 7.4. Combination of tannic acid/glutaraldehyde/paraformaldehyde followed by osmification increased the staining of the membranes. The samples were washed 4r times in 0.1 M sodium acetate buffer, pH 5.5, and block-stained in 0.5% uranyl acetate in 0.1 M sodium acetate buffer, pH 5.5, for 12 hours at 4 °C. The samples were dehydrated in graded ethanol (50%, 75%, 95%, 100%, 100%, and 100%, 10 minutes each), passed through propylene oxide, and infiltrated in mixtures of Epon 812 and propylene oxide (1:1 and 2:1, 2 hours each). The tissues were subsequently infiltrated in pure Epon 812 overnight. Embedding was performed in pure Epon 812, and curing was done in an oven at 60 °C for 48 hours. Sections of 60 nm thickness (gray interference color) were cut on an ultramicrotome (RMC MTX) using a diamond knife. The sections were deposited on single-hole grids coated with Formvar and carbon and double-stained in aqueous solutions of 8% uranyl acetate for 25 minutes at 60 °C, and 26.6 mM lead citrate for 3 minutes at room temperature. The sections were ultimately examined with a 100CX JEOL electron microscope.

Statistical Analysis

The analyses were performed using GraphPad Instat, version 8.0 (GraphPad Software). Data were analyzed by either unpaired 2-tailed Student t test or 1-way ANOVA with Tukey multiple-comparisons test for statistical significance.

RESULTS

Arterial-Lymphatic-Like ECs Appeared After Endothelial-Specific Deletion of Alk1 to Cause AVMs and Vascular Leakage

To examine ECs after endothelial gene deletion of Alk1, we crossbred VE-cadherin^cre/ERT2 mice with Alk1^flox/flox mice. At 10 weeks of age, we administered tamoxifen to the VE-cadherin^cre/ERT2Alk1^flox/flox mice by intraperitoneal injection for 5 consecutive days. After injection, we examined the Alk1 expression in different mouse organs, including the lungs, liver, brain, retina, kidneys, intestine, skin, and tongue. The real-time PCR showed a robust decrease of Alk1 transcripts in these organs (Figure S1), supporting efficient deletion of Alk1 deletion in the ECs.

In the lungs of tamoxifen-injected VE-cadherin^cre/ERT2Alk1^flox/flox mice, micro-CT showed multiple abnormal nidus-like structures that connected arteries and veins (Figure 1A and 1B; Figure S2). We examined at least 6 mice in each group by micro-CT; quantification revealed that all VE-cadherin^cre/ERT2Alk1^flox/flox mice had niduses in their lungs after Alk1 deletion (Figure 1C). The visualization of the venous system in the VE-cadherin^cre/ERT2Alk1^flox/flox mice supported that these nidus-like structures were arteriovenous shunts (Figure 1A and 1B; Figure S2). No abnormalities of the pulmonary vasculature or niduses were found in the Alk1^flox/flox or VE-cadherin^cre/ERT2 mice after tamoxifen injection (Figure 1A and 1B; Figure S3).

Figure 1.

Vascular leakage and AVMs in VE-cadherin^cre/ERT2Alk1^flox/flox mice after tamoxifen administration. A and B, Micro-CT images of the pulmonary vasculature in VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=6). Stars and arrowheads represent niduses or arteriovenous (A-V) shunts. A, artery. V, vein. f/f, flox/flox. Scale bars=1 mm. C, Quantification of niduses, and analysis of vessel radii and total number of branches in the pulmonary vasculature of VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=6). D and E, The amount of fluid in the chest cavity (D) and the abdominal cavity (E) of VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=10). F, TEM of small blood vessels in VE-cadherin^cre/ERT2Alk1^flox/flox mice after 2 or 5 days of tamoxifen administration. Alk1^flox/flox mice were used as controls (n=6). Red arrowheads represent the endothelium. RBC, red blood cells. Magnification for TEM, 3.7×10³. Scale bars=800 nm. G, H&E and Evans Blue staining of the lungs from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=5). Scale bars=100 nm (top) and 1 mm (bottom). C through E were analyzed for statistical significance by the unpaired 2-tailed Student t test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values. *P<0.01; **P<0.001; ***P<0.0001.

We also examined the frequency of vessels with different radii. The results showed that the frequency of vessels with radii <50 µm dramatically increased, especially in the vessels between 25 amd 50 µm, in the VE-cadherin^cre/ERT2Alk1^flox/flox mice. On the other hand, the frequency of vessels with radii >75 µm decreased in the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 1C). We quantified the total number of branches in the vessels, which showed a decrease in the lungs of the VE-cadherin^cre/ERT2Alk1^flox/flox mice compared with the Alk1^flox/flox mice (Figure 1C). The increase in niduses together with the increase in the frequency of vessels between 25 and 50 µm suggested that the niduses were composed of small vessels in the lungs, as shown by the micro-CT (Figure 1A through 1C; Figure S2). The decrease of vessels with radii >75 µm as well as the number of vascular branches point to a shrinkage of the local networks because of short-circuiting by niduses in the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 1A through 1C; Figure S2).

In addition, we examined the retinas of tamoxifen-injected VE-cadherin^cre/ERT2Alk1^flox/flox mice on postnatal day 7. Immunostaining showed a dramatic increase in arteriovenous shunts that markedly altered the retinal vascular networks (Figure S4). These results confirmed that AVMs occurred in VE-cadherin^cre/ERT2Alk1^flox/flox mice after endothelial Alk1 deletion.

Unexpectedly, we found large amounts of serous fluid in the chest and abdominal cavities (Figure 1D and 1E) of tamoxifen-injected VE-cadherin^cre/ERT2Alk1^flox/flox mice. The fluid appeared to be a transudate (Figure 1D and 1E), suggesting vascular leakage rather than vessel rupture. We examined the pulmonary capillaries of the VE-cadherin^cre/ERT2Alk1^flox/flox mice using TEM on days 2 and 5 after initiation of tamoxifen injections. The results showed dramatic changes in the endothelium compared with the Alk1^flox/flox mice (Figure 1F). On day 2, the thin and tight endothelium layer was replaced by a thick and disordered endothelium (Figure 1F, middle). On day 5, the endothelial morphology was barely recognizable with large empty spaces and ruptured junctions (Figure 1F, right). Histology also showed the excess fluid in alveoli cavity, and Evans blue injection indicated vascular leakage in the lungs of tamoxifen-injected VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 1G). Together, the results suggested that the alterations in the endothelium caused by Alk1 deletion led to vascular leakage and AVMs.

To determine whether the Alk1 deletion altered the EC characteristics, we isolated VE-cadherin⁺CD45^- pulmonary ECs from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after 5-day administration of tamoxifen and performed scRNA-seq. In both VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice, UMAP equipped with published cell-specific markers^34,35 identified 2 clusters of arterial ECs (arterial EC1 and EC2), a cluster of venous ECs, and 3 different clusters of capillary ECs (arterial capillary EC and capillary EC1 and EC2; Figure 2A and 2B; Figure S5). Based on the expression levels of cell markers,^34,35 the arterial EC1 cluster was identified as ECs of large pulmonary arteries, arterial EC2 was identified as ECs of small pulmonary arteries or arterioles, capillary EC1 was identified as ECs of general capillaries, capillary EC2 was identified as the aerocytes (the specific ECs in pulmonary capillaries), and arterial capillary EC was identified as ECs connecting arterioles with capillary networks (Figure 2A and 2B; Figure S5). A cluster of EC-like myofibroblasts was also present in both mice, as we have previously reported.³⁶ It is interesting that UMAP identified the appearance of a new cell cluster in the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 2A and 2B; Figure S5). This new cell cluster showed deletion of Alk1 but expressed the pan-endothelial markers VE-cadherin and CD31, and the arterial EC markers Ephrin B2, Sema3G, GJA5, and CXCL12, suggesting that the cluster was derived from arterial ECs (Figure 2A and 2B; Figure S5). However, a robust induction of the lymphatic endothelial markers Sox17, Sox18, Prox1, Lyve1, Ackr3, and Flt4 was also detected in this cluster (Figure 2A and 2B; Figure S5). Because this cluster coexpressed arterial and lymphatic endothelial markers, we referred to these cells as arterial-lymphatic-like ECs (Figure 2A and 2B).

Figure 2.

scRNA-seq of VE-cadherin⁺CD45^- pulmonary ECs from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration. A, UMAP for the cell populations subclustered from VE-cadherin⁺CD45^- pulmonary cells. Totals of 495 and 1482 cells were captured from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice, respectively, after quantity filtering. The dashed circle indicates a new cluster appearing in the VE-cadherin^cre/ERT2Alk1^flox/flox mice. The pink cell cluster was referred to as the arterial-lymphatic-like ECs. B, Violin plots of gene expression of lineage markers. C, Pseudotemporal trajectories of the cell clusters. D, Gene expression along the single cell trajectories according to the pseudotime. E, Cell compositions of different populations in VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration. F, Schematic comparison of cell compositions and structural alterations in A-V connections in VE-cadherin^cre/ERT2Alk1^flox/flox mice vs Alk1^flox/flox mice.

To determine from which type of ECs these arterial-lymphatic-like ECs were potentially derived, we analyzed the cell differentiation trajectory. In Alk1^flox/flox mice, the analysis projected that arterial ECs and arterial capillary EC1 differentiated into arterial EC2 and capillary EC1 and EC2 (Figure 2C, top). In VE-cadherin^cre/ERT2Alk1^flox/flox mice, this differentiation trajectory still existed with one strong additional direction for more arterial lineage ECs to differentiate into arterial-lymphatic-like ECs (Figure 2C, bottom). Following this trajectory, gene expression dynamics was determined by Moran’s I test, in which a q value <0.05 was considered to be a significant change. The results of the analysis showed a significant decrease in Alk1 expression with a significant induction of the lymphatic endothelial markers Lyve1, Flt4, and Prox1 across the cell differentiation trajectory (Figure 2D). The cell composition showed a decrease in the percentage of arterial ECs and capillary ECs with a dramatic increase in the percentage of arterial-lymphatic-like ECs (Figure 2E), suggesting that arterial ECs with Alk1 deletion might shift toward arterial-lymphatic-like ECs (Figure 2F).

There were only 2 arterial-lymphatic-like ECs identified in the Alk1^flox/flox mice through the clustering process (Figure 2A and 2E). The pseudotime programming did not recognize these 2 cells as typical clusters when projecting the differentiation trajectory (Figure 2C). Therefore, we did not include the gene expression of these cells in Figure 2B.

To validate the occurrence of arterial-lymphatic-like ECs in the tamoxifen-treated VE-cadherin^cre/ERT2Alk1^flox/flox mice, we examined the lung tissues. We first performed immunostaining to examine the areas where the vessel radii were <50 µm in the lungs of the Alk1^flox/flox control mice, then the corresponding area in the VE-cadherin^cre/ERT2Alk1^flox/flox mice. These were the research areas of interest to show the possible nidus formation and alteration in the cell characteristics (Figure 1B and 1C). Immunostaining showed obvious coexpression of the arterial endothelial marker Ephrin B2 with the lymphatic endothelial markers SOX17, PROX1, and LYVE1 in VE-cadherin^cre/ERT2Alk1^flox/flox mice compared with Alk1^flox/flox mice (Figure 3A). The results also revealed colocalization of Ephrin B2, SOX17, and LYVE1 in similar areas (Figure 3B). FACS with quantification confirmed that a large cell population coexpressing Ephrin B2 and LYVE1 emerged in the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 3C). In isolated Ephrin B2⁺ pulmonary ECs from these mice, real-time PCR revealed a reduction of Alk1 combined with an induction of the lymphatic endothelial markers Sox17, Prox1, and Lyve1 (Figure 3D). The results validated that ALK1-deficient arterial lineage ECs acquired lymphatic endothelial markers to transition to arterial-lymphatic ECs. To determine if the arterial-lymphatic ECs caused vascular malformations, we used a model of hindlimb ischemia in nude mice. We isolated Ephrin B2⁺LYVE1⁺ cells from VE-cadherin^cre/ERT2Alk1^flox/flox mice and transplanted them into nude mice, in which the proximal and distal femoral arteries were ligated. Ephrin B2⁺LYVE1^- cells were used as controls. Two weeks after transplantation, latex dye staining with tissue clearing showed enlarged and twisted small vessels at the repair sites in mice transplanted with the Ephrin B2⁺LYVE1⁺ cells (Figure 3E). Immunostaining with anti-CD31 antibodies confirmed the presence of nested vessels with large diameter at the repair sites, supporting that the Ephrin B2⁺LYVE1⁺ pulmonary ECs from VE-cadherin^cre/ERT2Alk1^flox/flox mice gave rise to vascular malformations (Figure 3E).

Figure 3.

Arterial-lymphatic-like ECs appearing in A-V shunts in the lungs of VE-cadherin^cre/ERT2Alk1^flox/flox mice and patients with HHT2, and contributing to vascular malformations in the revascularization of ischemic hindlimbs. A and B, Immunostaining of lung tissues from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=6). Scale bars=50 µm. C, FACS with percentage analysis of lung cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration. Three mice were used in each group, and experiments were performed ≥3 times. D, Gene expression in Ephrin B2⁺ cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration (n=6). E, Latex dye staining (top) and immunostaining for CD31 (bottom) with diameter analysis in nude mice after transplantation of Ephrin B2⁺ (left) and Ephrin B2⁺LYVE1⁺ (right) cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice after tamoxifen administration (n=11). Scale bars=0.2 mm (top) and 100 µm (bottom). F and G, Immunostaining with cell quantification in pulmonary AVMs from patients with HHT2 (n=3). Scale bars=50 µm. C through G were analyzed for statistical significance by unpaired 2-tailed Student t test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values. *P<0.01; ***P<0.0001.

To show the relevance of our results to human HHT2, we costained Ephrin B2 with SOX17, PROX1, or LYVE1 in pulmonary AVMs from patients with HHT2. Immunostaining followed by quantification of the stained cells showed an enhancement of cells that coexpressed Ephrin B2 with SOX17, PROX1, or LYVE1 in the pulmonary AVMs (Figure 3F and 3G). The results supported that arterial-lymphatic-like ECs may appear in AVMs from patients with HHT2.

Endothelial Lineage Tracing Showed That Arterial-Lymphatic-Like ECs Derived From Alk1-Deficient ECs, Potentially Arterial ECs

To perform endothelial lineage tracing in ALK1 deficiency, we crossbred VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice and VE-cadherin^cre/ERT2Rosa^tdTomato control mice. At 10 weeks of age, we administered tamoxifen to the mice by intraperitoneal injection for 5 consecutive days, isolated tdTomato⁺ cells from the lungs, and performed scRNA-seq. UMAP showed that all cell clusters expressed tdTomato and endothelial markers (Figure 4A and 4B; Figure S6), confirming the endothelial origin of all cell clusters, including the arterial-lymphatic-like ECs. Based on the same analysis with the same group markers as we used in Figure 2,^34,35 UMAP again identified the different cell clusters as arterial ECs, arterial capillary ECs, capillary EC1 and EC2, and venous ECs in both VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice and VE-cadherin^cre/ERT2Rosa^tdTomato control mice (Figure 4A and 4B; Figure S6). UMAP also showed that the arterial-lymphatic-like ECs only appeared in the tdTomato⁺ cells from the VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox lungs (Figure 4A and 4B; Figure S6). It is interesting that because the lineage tracing was able to trace ECs with tdTomato labeling during the early stages of the cell transition, it allowed for clarification of the cell contents and the differentiation trajectories of the arterial-lymphatic-like ECs. UMAP found that the arterial-lymphatic-like ECs, which were identified in Figure 2, contained 2 groups with similar characteristics, arterial-lymphatic-like EC1 and EC2 (Figure 4A and 4B; Figure S6).

Figure 4.

Endothelial lineage tracing and immunostaining showed that arterial-lymphatic-like ECs were derived from Alk1-deficient ECs, potentially arterial ECs. A, UMAP for the cell populations with tdTomato expression subclustered from tdTomato⁺CD45^- pulmonary cells. Totals of 7949 and 13 022 cells were captured from VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice and VE-cadherin^cre/ERT2Rosa^tdTomato control mice, respectively, after quantity filtering. The dashed circles indicate new clusters that appeared in the tdTomato⁺ cells of VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice. The circled pink and red cell clusters were referred to as the arterial-lymphatic-like ECs. B, Violin plots of gene expression of lineage markers. C, RNA velocity analysis of differentiation trajectories of the cell clusters using stream plots and partition-based graph abstraction plots. D and E, Immunostaining with cell quantification in lung tissues from VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice and VE-cadherin^cre/ERT2Rosa^tdTomato control mice after tamoxifen injection (n=10). Scale bars,=50 µm. E was analyzed for statistical significance by ANOVA with the post hoc Tukey test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values (B). Error bars are mean±SD (A). ***P<0.0001.

RNA velocity analysis, which identifies temporal dynamics in single-cell gene expression, showed 2 different trajectories derived from arterial ECs toward these 2 groups. One trajectory went from arterial ECs to capillary EC1 to arterial capillary EC1 and last to arterial-lymphatic-like EC1. The other trajectory went from arterial ECs to capillary EC1 and then to arterial-lymphatic-like EC2 (Figure 4C). Partition-based graph abstraction velocity graphs further mapped the 2 trajectories from arterial ECs toward the arterial-lymphatic-like ECs (Figure 4C). These results showed a more detailed derivation of arterial-lymphatic-like ECs and again supported the potential arterial origin of arterial-lymphatic-like ECs after Alk1 deletion.

Furthermore, we performed immunostaining to examine the corresponding areas where the vessel radii were <50 µm in the lungs of VE-cadherin^cre/ERT2Rosa^tdTomato control mice and VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice. Immunostaining showed an enhanced expression of Sox17 and Prox1 in Ephrin B2⁺tdTomato⁺ cells in the VE-cadherin^cre/ERT2Rosa^tdTomatoAlk1^flox/flox mice compared with the VE-cadherin^cre/ERT2Rosa^tdTomato control mice (Figure 4D and 4E). Together, the lineage tracing revealed the endothelial origin of arterial-lymphatic-like ECs and suggested that arterial-lymphatic-like ECs might be derived from arterial ECs.

In the Absence of ALK1, Induction of Sox17 Caused HPAECs to Acquire Lymphatic Endothelial Characteristics

Our results suggested that the arterial-lymphatic-like ECs were potentially derived from arterial ECs after Alk1 deletion. We examined the transcription factors involved in the development of lymphatic ECs. scRNA-seq identified Sox17 induction in the cluster of arterial ECs, especially in the arterial EC1 and EC2, and the arterial-lymphatic-like ECs from the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 2B, far left). To determine if the induction of Sox17 is the causative factor for arterial ECs gaining lymphatic endothelial characteristics, we used HPAECs and examined the time-course SOX17 expression after depletion of ALK1 by siRNA. Real-time PCR showed that ALK1 depletion robustly induced SOX17 (Figure 5A). Then, we depleted ALK1 in HPAECs in combination of either SOX17 deletion using siRNA or SOX17 overexpression using lentiviral vectors, where SOX17 cDNA was driven by the cytomegalovirus promoter. Immunoblotting showed that SOX17 depletion abolished LYVE1 expression in the ALK1-depleted HPAECs (Figure 5B), and SOX17 overexpression enhanced the LYVE1 induction in the ALK1-depleted HPAECs (Figure 5B). We administered Sox17 small hairpin RNA to VE-cadherin^cre/ERT2Alk1^flox/flox mice by tail vein injection together with tamoxifen. Real-time PCR showed a reduction of both Prox1 and Lyve1 in the Ephrin B2⁺ pulmonary cells in VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 5C), in which micro-CT revealed an improvement of pulmonary vasculature and a reduction of AVMs in the lungs (Figure 5D; Figure S7). The results suggested that SOX17 could be a key driving factor for arterial ECs to potentially gain lymphatic endothelial characteristics and cause AVMs in ALK1-deficient mice. We also depleted ALK1 in the human umbilical vein ECs and examined Sox17 expression. Real-time PCR did not show the change in SOX17 expression in human umbilical vein ECs after ALK1 deletion (Figure S8), suggesting that the impact of the ALK1 deletion was different between HPAECs and human umbilical vein ECs.

Figure 5.

In the absence of ALK1, induction of Sox17 causes arterial lineage ECs to acquire lymphatic endothelial characteristics. A, Time-course expression by real-time PCR of Alk1 and Sox17 in HPAECs after transfection of Alk1 siRNA (n=5). B, Immunoblotting of HAECs transfected with Alk1 siRNA in combination with Sox17 siRNA or infection of lentiviral vectors overexpressing Sox17 (n=3). β-Actin was used as loading control. C, Gene expression in Ephrin B2⁺ cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after Sox17 small hairpin RNA administration (n=8). D, Micro-CT imaging with analysis of nidus formation, vessel radius, and total number of branches of the lungs of VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after Sox17 small hairpin RNA administration (n=5). Scale bars=1 mm. E, Immunoblotting of HPAECs transfected with combinations of Alk1 siRNA and Flt4 or Flk1 siRNA (n=3). F, Sox17 expression in HPAECs after transfection of Alk1 siRNA treated with or without VEGF-C, VEGF-C (Cys156Ser), or VEGF-165 (n=7). SCR, scrambled siRNA. G, Immunoblotting after coimmunoprecipitation of Ephrin B2⁺ cell lysates isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration using anti-ALK1 (α-ALK1) or anti-FLT4 (α-FLT4, ab15079) (n=3). H, Immunoblotting after coimmunoprecipitation of Ephrin B2⁺ cell lysates isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration using anti-ALK1 (α-ALK1), anti–VEGF-C (α-VEGF-C), anti-FLT4 (α-FLT4, ab15079), or anti-FLT4 (α-FLT4, AFL4) (n=3). A, C, D, and F were analyzed for statistical significance by ANOVA with post hoc Tukey test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values (B). Error bars are mean±SD (A). *P<0.01; ***P<0.0001.

To explore how SOX17 was induced by ALK1 deletion, we examined factors known to regulate lymphatic endothelial differentiation. We transfected HPAECs with Alk1 siRNAs in combination with specific siRNAs to FLT4 (VEGFR3) or FLK1 (VEGFR2). The immunoblotting showed that depletion of FLT4, but not FLK1, abolished the induction of SOX17 in ALK1-depleted HPAECs (Figure 5E), suggesting that the activation of FLT4 was responsible for the induction of SOX17. Then, we treated ALK1-depleted HPAECs with VEGF-C or VEGF-C (Cys156Ser), which is a recombinant mature VEGF-C with cysteine-156 replaced by serine to specifically interact with FLT4, but not FLK1.³⁷ We also treated the cells with VEGF-165, which specifically interacted with FLK1.³⁸ The results showed that VEGF-C and VEGF-C (Cys156Ser) had similar effects on the enhancement of SOX17 induction in ALK1-depleted HPAECs. However, VEGF165 did not affect the SOX17 induction (Figure 5F). The results suggested that VEGF-C activated FLT4 and induced Sox17.

Previous studies have shown that ALK1 interacts with other receptors to modify cell signaling.³⁹ To explore the possibility that ALK1 interacts with FLT4, we isolated Ephrin B2⁺ pulmonary cells from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after 2 and 5 days of tamoxifen administration. Using the lysates of these cells, we performed coimmunoprecipitation with anti-ALK1 and anti-FLT4 antibodies followed by immunoblotting, which showed an interaction between ALK1 and FLT4 in the Ephrin B2⁺ pulmonary cells from the Alk1^flox/flox mice (Figure 5G). Two days after administration of tamoxifen, immunoblotting showed a decrease in ALK1 and less interaction between ALK1 and FLT4 in the Ephrin B2⁺ cells from the VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 5G). Five days after administration of tamoxifen, immunoblotting showed deletion of ALK1 and no interaction between ALK1 and FLT4 (Figure 5G, far right). It is interesting that FLT4 remained at similar levels after 2 and 5 days of tamoxifen administration in both VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice (Figure 5G, bottom). Together, the results suggested that ALK1 bound to FLT4 in normal arterial lineage ECs but not in ALK1-depleted arterial lineage ECs, thereby allowing VEGF-C to induce SOX17.

To explore which domain of FLT4 bound to ALK1, we performed coimmunoprecipitation to examine the complex of VEGF-C, FLT4, and ALK1 using different antibodies. We collected the Ephrin B2⁺ pulmonary cells from VE-cadherin^cre/ERT2Alk1^flox/flox and Alk1^flox/flox control mice after 5 days of injection of tamoxifen. We performed immunoprecipitation using anti–VEGF-C, anti-ALK1, and 2 different anti-FLT4 antibodies, the polyclonal ab15079 antibodies as well as the AFL4 antibodies, where the binding epitope was mapped to the sequence in the FLT4 immunoglobulin homology domain 5 (D5), not the epitope in the ligand binding region (D1–D3).^40,41 Subsequently, immunoblotting was used to examine the precipitated complex. In the cells from the VE-cadherin^cre/ERT2Alk1^flox/flox mice, coimmunoprecipitation followed by immunoblotting showed that the complex of VEGF-C and FLT4 was pulled down by all of anti–VEGF-C, anti-FLT4 (ab15079) and anti-FLT4 (AFL4) (Figure 5H), suggesting that the release of FLT4 after Alk1 deletion allowed activation of VEGF-C. It is interesting that in the control mice, the results showed that the complex of VEGF-C, FLT4, and ALK1 was only pulled down by anti–VEGF-C, anti-ALK1, and anti-FLT4 (ab15079; Figure 5H). Anti-FLT4 (AFL4) was unable to precipitate VEGF-C, FLT4, and ALK1 (Figure 5H), suggesting that ALK1 binds to the FLT4 immunoglobulin homology D5, not the ligand binding region D1 to D3, and causes a conformational change resulting in poor signaling transduction of VEGF-C/FLT4.

MDM2 Inhibition Prevents Arterial ECs From Gaining Characteristics of Lymphatic ECs

In our studies, the scRNA-seq revealed a clear trajectory from arterial ECs toward arterial-lymphatic-like ECs. By analyzing the differential gene expression between arterial-lymphatic ECs and arterial ECs, we identified 1544 genes with increased expression and 1566 genes with decreased expression using adjusted P value <0.05 cutoff with 2-fold difference. The alterations in gene expression detailed the genetic signature of the transition from arterial ECs to arterial-lymphatic-like ECs. To identify a compound that prevented this transition, we input the differential expression profile into the CMap platform. The CMap platform was created to connect genetic perturbations with compound treatments.⁴² Using a connectivity score from designed measurements of expression profiles, a query of CMap can be generated to search compounds that cause similar genetic perturbations.⁴³ With this advantage, the CMap platform provides a possible approach to identify compounds that redirect the transcriptional landscape toward a desired direction. To find a compound that prevents the genetic alterations from arterial ECs to arterial-lymphatic ECs, we input the top 100 differential expression profiles between arterial ECs and arterial-lymphatic ECs with an opposite direction of alteration to generate a novel query, which allowed the platform to find compounds that were capable of correcting the genetic alteration. CMap identified the small molecule HLI373, an inhibitor of MDM2, as a potential candidate.

We treated VE-cadherin^cre/ERT2Alk1^flox/flox mice with tamoxifen in combination with HLI373 (5 µg/g daily) for 5 days. Micro-CT showed that HLI373 significantly reduced AVMs in the lungs of tamoxifen-treated VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 6A), suggesting that the inhibition of MDM2 counteracted the AVMs caused by the Alk1 deletion. Furthermore, TEM revealed a normalization of the capillary endothelium after HLI373 treatment (Figure 6B), and FACS showed a decrease in cells that coexpressed Ephrin B2 and LYVE1 in the lungs after HLI373 treatment (Figure 6C and 6D). We also isolated Ephrin B2⁺ pulmonary cells from treated mice and found a reduction of Prox1 and Lyve1 in these cells as determined by real-time PCR (Figure 6E). It is interesting that HLI373 did not alter the induction of Sox17 in the lungs of the tamoxifen-treated VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 6E), suggesting that MDM2 affected the activity of SOX17 rather than its expression.

Figure 6.

Inhibition of MDM2 prevents the Ephrin B2⁺ cells, potentially arterial ECs, from acquiring lymphatic endothelial characteristics and reduces AVMs in VE-cadherin^cre/ERT2Alk1^flox/flox mice. A, Micro-CT imaging with the analysis of nidus formation, vascular radius and total number of branches of the pulmonary vasculature of VE-cadherincre/ERT2Alk1flox/flox mice treated with tamoxifen in combination with or without HLI373 (n=5). Scale bars=1 mm. B, TEM of small blood vessels in VE-cadherin^cre/ERT2Alk1^flox/flox mice after tamoxifen administration in combination with or without HLI373. Alk1^flox/flox mice were used as controls (n=6). Red arrowheads represent the endothelium. RBC, red blood cells. Magnification for TEM, 3.7×10³. Scale bars=800 nm. C and D, FACS with percentage analysis of lung cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and Alk1^flox/flox mice after tamoxifen administration in combination with or without HLI373. Three mice were used in each group, and the experiments were performed at least 3 times. E, Gene expression in Ephrin B2⁺ cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice after tamoxifen administration in combination with treatment with or without HLI373 (n=4). A and D were analyzed for statistical significance by ANOVA with post hoc Tukey test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values (B). *P<0.01; **P<0.001; ***P<0.0001.

We also examined the expression of Sox17, Prox1, and Lyve1 in the retinal vasculature. The immunostaining showed that the increased expression of SOX17, PROX1, and LYVE1 were exclusively identified in the retinal arteries and arterial sides of arteriovenous shunts in VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 7A and 7B). We showed that the induction of the lymphatic endothelial markers and arteriovenous shunts were abolished by HLI373 treatment (Figure 7A and 7B).

Figure 7.

Inhibition of MDM2 prevents arterial ECs and arterial-capillary ECs from acquiring lymphatic endothelial characteristics and reduces AVMs in the retinas of VE-cadherin^cre/ERT2Alk1^flox/flox mice. A, Immunostaining of retinal vasculature with quantification of AVMs of VE-cadherincre/ERT2Alk1flox/flox mice after tamoxifen administration in combination with or without HLI373 (n=5). Scale bars=100 µm. B, Immunostaining of A-V shunts in the retinal vasculature of VE-cadherin^cre/ERT2Alk1^flox/flox mice after tamoxifen administration in combination with or without HLI373 (n=5). A, artery. V, vein. Scale bars=100 µm. A was analyzed for statistical significance by ANOVA with the post hoc Tukey test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is the median. Error bars are maximal and minimal values. ***P<0.0001.

To further investigate the role of MDM2 in AVMs as a result of ALK1-deficiency, we bred VE-cadherin^cre/ERT2Alk1^flox/flox mice with Mdm2^± mice. Because Mdm2^-/- mice are embryonic lethal,⁴⁴ VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice were used for this study. We treated VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice with tamoxifen for 5 days and examined the lung vasculature using latex blue staining with tissue clearing, which has been used to examine arteriovenous shunts in previous studies.⁴⁵ The latex blue staining combined with the Z-stack imaging function of the fluorescence stereo microscope efficiently visualized the pulmonary vasculature and nidus formation in the mouse lungs (Figure 8A; Figure S9), similar to micro-CT. Using this method, we quantified nidus formation, total number of branches and vascular density in areas with different vessel radii in the lungs. The results showed increased niduses formation, decreased branching, and increased vascular density in areas with the vessel radius <50 µm in the VE-cadherin^cre/ERT2Alk1^flox/flox mice. The results were consistent with the micro-CT (Figures 1 and 8B). We then examined the mice with Mdm2 deletion and showed that loss of Mdm2 reduced the nidus formation, increased the branching, and improved the vascular density of VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice (Figure 8A and 8B; Figure S9). TEM showed a structural correction of the endothelium in the pulmonary capillaries in VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice (Figure 8C). We also showed a large decrease in the LYVE1-expressing cell population in Ephrin B2⁺ cells from VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice, as determined by FACS (Figure 8D). This result was confirmed by real-time PCR, which showed that limiting MDM2 prevented the induction of Prox1 and Lyve1 in Ephrin B2⁺ cells from VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice (Figure 8E). Again, the results did not detect any changes in Sox17 in VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice or VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure 8E).

Figure 8.

Gene deletion of MDM2 prevents arterial lineage ECs from acquiring lymphatic endothelial characteristics and reduces AVMs. A and B, Latex dye staining with analysis of nidus formation, total number of branches, and vascular density of the lungs from VE-cadherin^cre/ERT2Alk1^flox/flox mice and VE-cadherin^cre/ERT2Alk1^flox/flox Mdm2^± mice after tamoxifen administration. Alk1^flox/flox mice were used as controls (n=6). Scale bars=2 mm. C, TEM of small pulmonary blood vessels in VE-cadherin^cre/ERT2Alk1^flox/flox mice and VE-cadherin^cre/ERT2Alk1^flox/flox Mdm2^± mice after tamoxifen administration. Alk1^flox/flox mice were used as controls (n=6). Red arrowheads represent the endothelium. RBC, red blood cells. Magnification for TEM, 3.7×10³. Scale bars=800 nm. D, FACS of LYVE1⁺ cells derived from pulmonary Ephrin B2⁺ cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and VE-cadherin^cre/ERT2Alk1^flox/flox Mdm2^± mice after tamoxifen administration. Alk1^flox/flox mice were used as control. Three mice were used in each group, and experiments were performed ≥3 times. D, Gene expression by real-time PCR in pulmonary Ephrin B2⁺ cells isolated from VE-cadherin^cre/ERT2Alk1^flox/flox mice and VE-cadherin^cre/ERT2Alk1^flox/flox Mdm2^± mice after tamoxifen administration. Alk1^flox/flox mice were used as controls (n=8). B and E were analyzed for statistical significance by ANOVA with post hoc Tukey test. The bounds of the boxes are upper and lower quartiles with data points. The line in the box is median. Error bars are maximal and minimal values. ***P<0.0001.

MDM2 Is Required for SOX17 Activity to Induce Lymphatic Endothelial Characteristics in Arterial ECs

To investigate the relationship between MDM2 and SOX17 required for arterial ECs to acquire lymphatic endothelial characteristics, we depleted ALK1 in HPAECs in combination with transfection of SOX17 siRNA or HLI373 treatment. Time-course experiments showed an induction of SOX17 and LYVE1 starting after depletion of ALK1 (Figure S10A). Depletion of SOX17 or HLI373 treatment abolished the LYVE1 induction in ALK1-depleted HPAECs, and HLI373 treatment had no effect on the SOX17 induction (Figure S10A). We then overexpressed SOX17 in HPAECs using a lentiviral vector that contained SOX17 cDNA driven by the cytomegalovirus promoter. We also transfected the cells with PROX1 siRNA or treated with HLI373. Immunoblotting showed that overexpression of SOX17 induced PROX1 and LYVE1 in HPAECs (Figure S10B). HLI373 treatment abolished induction of both PROX1 and LYVE1, and the transfection of PROX1 siRNA abolished the LYVE1 induction (Figure S10B). On the other hand, we depleted ALK1 in HPAECs in combination of transfection of SOX17 siRNA or MDM2 siRNA as well as HLI373 treatment. The immunoblotting showed that knockdown of SOX17 or MDM2, or HLI373 treatment abolished the induction of PROX1 and LYVE1. Knockdown of SOX17 did not affect MDM2 expression, and limiting MDM2 did not affect SOX17 expression (Figure S10C). The results suggested that SOX17 and MDM2 cooperated to induce lymphatic endothelial characteristics in arterial lineage ECs.

Because alterations in MDM2 affected SOX17 activity rather than its expression, we hypothesized that MDM2 interacted with SOX17. To test this hypothesis, we isolated Ephrin B2+ pulmonary cells from the Alk1^flox/flox mice, VE-cadherin^cre/ERT2Alk1^flox/floxMdm2^± mice, and VE-cadherin^cre/ERT2Alk1^flox/flox mice that had been treated with or without HLI373. We performed coimmunoprecipitation from cell lysates using anti-SOX17 or anti-MDM2 antibodies. Subsequent immunoblotting showed that elevated SOX17 interacted with MDM2 in the Ephrin B2⁺ pulmonary cells from VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure S10D). Blocking MDM2 by HLI373 or deleting MDM2 abolished the interaction between SOX17 and MDM2 (Figure S10D). We then performed chromatin immunoprecipitation from the same cell lysates using anti-SOX17 antibodies. We examined the enrichment of SOX17 around its binding sites, which were reported to be located at –1135 bp and –822 bp in the Prox1 promoter,⁴⁶ and uncovered a strong SOX17 enrichment in this area of the Prox1 promoter in VE-cadherin^cre/ERT2Alk1^flox/flox mice (Figure S10E). HLI373 or deletion of Mdm2 reduced this SOX17 enrichment (Figure S10E). To further explore the activation by PROX1, we located the PROX1-DNA binding site in the Flt4 promoter as in previous studies.⁴⁷ The chromatin immunoprecipitation assay showed enrichment of PROX1 around its binding site in the Flt4 promoter, and HLI373 or deletion of MDM2 reduced this enrichment (Figure S10E). Together, the results suggested that MDM2 was required for enhanced SOX17 to induce lymphatic endothelial characterizes in ALK1-deficient arterial ECs (see Figure S11 for schematic working model).

DISCUSSION

In this study, we took advantage of mouse models of endothelial specific deletion of Alk1 and demonstrated that pulmonary arterial-lymphatic-like ECs appeared after ALK1 deletion. We showed that these arterial-lymphatic-like ECs caused vascular malformations. The results also provided evidence that arterial-lymphatic-like ECs potentially were derived from arterial ECs after loss of ALK1. This information appears to be a previously unknown mechanism of HHT2, which might enhance our understanding of the causative role of dysregulated vascular cell differentiation in the progression of HHT2.

This study suggested that pulmonary arterial ECs that acquired lymphatic endothelial characteristics contributed to the source of pulmonary AVMs. Advanced studies also demonstrated the important role of venous ECs in AVMs of the retinal vascular bed, and showed that venous ECs transitioned into arterial-like cells or migrated toward arterioles to contribute to retinal AVMs.^48–50 Differing from the expression in the retinal vascular bed, ALK1 is predominantly expressed in arterial ECs in major organs, including the lungs.¹⁹ Our results from scRNA-seq did not detect arterial or lymphatic endothelial characteristics in pulmonary venous ECs (Figures 2 and 4). We argue that pulmonary venous ECs may not contribute to the alterations in the pulmonary arterial ECs or initiate AVMs. However, the changes in flow dynamics and cell signaling from altered arterial ECs after initiation of AVMs may drive the venous and capillary ECs in a pathogenic direction that coordinates with the altered arterial ECs and form pulmonary AVMs. Further studies using specific arterial or venous lineage tracing would be of interest to study how pulmonary venous ECs might join forces with altered arterial ECs in the formation of AVMs.

EC proliferation in response to lack of ALK1 could be another important contributor to HHT2. Interestingly, the reports on the effect of ALK1 deficiency on EC proliferation are diverse. Previous studies showed that ALK1 deletion caused an increase in EC proliferation,^51,52 whereas other studies suggested that loss of ALK1 had no effect on the proliferation of arterial ECs but altered their migration.⁵³ In our study, we observed the change in characteristics of arterial ECs and how these changes altered their capacity to expand toward lymphatic ECs. Because there is no detailed report about the proliferation of arterial ECs in pulmonary AVMs in HHT2, measuring the growth of pulmonary arterial ECs with lymphatic endothelial characteristics in HHT2 could be an interesting question for future studies.

In human HHT, pulmonary AVMs can be classified as simple, complex, or diffuse type.⁵⁴ The simple type has only 1 feeding artery with a single or multiple branches that feed the malformation. The complex type has multiple feeding arteries, whereas the combination of simple and complex AVMs forms a diffuse type lesion.⁵⁴ Because a systematic analysis of changes in small vessels in lungs of patients with HHT2 has not been reported, enlarged feeding arteries connecting with enlarged capillaries are considered the common features of pulmonary AVMs.⁵⁴ HHT2 develops from Alk1 haploinsufficiency, and only a certain percentage of patients develops pulmonary AVMs.⁴ In our mouse model with endothelial Alk1 deletion, >80% of Alk1 was deleted in pulmonary ECs and allowed us to examine the overall changes in the lungs. The results showed a robust change in entire pulmonary endothelium with multiple niduses fed by multiple arteries (Figure 1). We examined the changes in the different vessels and showed an increase in vessels with radii <50 µm, specifically the range of 25 to 50 µm. Micro-CT showed that this vessel size range mainly included small arteries and enlarged capillaries that composed the niduses. The spatial distribution of the vessels was consistent with the clinically observed lesion formation. Immunostaining showed coexpression of arterial and lymphatic endothelial markers in cells located in the niduses, suggesting that arterial-lymphatic-like cells appeared in these areas (Figure 3). Transcriptomics showed that the arterial-lymphatic-like cells could be derived from arterial or arterial-capillary ECs, which again points to the locations around the connections between feeding arteries and capillary networks (Figures 2 and 4). Together, the results supported that the cells with arterial characteristics located around the connections between small arteries and capillaries acquire lymphatic endothelial characteristics after loss of ALK1 and contribute to AVMs.

ALK1 regulates lymphatic endothelial differentiation through BMP9 activity.^20–23 In this study, we found several factors that involved how Alk1 deletion led to the emergences of lymphatic endothelial differentiation in arterial ECs. These included a strong induction of Sox17, a modifier of lymphangiogenesis.⁴⁶ Elevated SOX17 induced PROX1, which in turn introduced lymphatic endothelial characteristics into arterial ECs. ALK1 bound directly to FLT4, thereby preventing VEGF-C/FLT4 signaling in arterial ECs. In ALK1 deficiency, FLT4 is available for activation by VEGF-C to induction of Sox17 and Prox1.

Interestingly, we found that MDM2 is required for SOX17 to induce PROX1 in ALK1-deficient arterial ECs. MDM2 was initially identified as a regulator of the p53 tumor suppressor gene, able to promote tumor growth.⁵⁵ MDM2 also executes p53-independent activity through various mechanisms.^56,57 We found that inhibition of MDM2 prevented SOX17 from inducing PROX1 and ultimately reduced pulmonary AVMs. The results suggested a previously unknown role for MDM2 in AVMs associated with ALK1 deficiency and might lead to new therapeutic strategies. However, recent clinical trials using MDM2 inhibitors for cancer treatment showed a potential for hematological toxicity, specifically thrombocytopenia.^58,59 Thus, the duration and dose will have to be carefully adjusted if MDM2 inhibitors were to be tested in patients with HHT to minimize potential off-target effects.

ARTICLE INFORMATION

Acknowledgments

Y. Yao and K.I.B supervised the experiments, analyzed data, and wrote the article. Y. Yang, X.W., Y.Z., D.Z., L.Z., X.C., J.J., and Z.J. performed experiments and data analysis.

Sources of Funding

Funding for this work was provided in part by National Institutes of Health grants NS79353 (to Y. Yao), HL139675 (to Y. Yao), HL162643 (to Y. Yao), HL81397 (to K.I.B.), HL158053 (to K.I.B.), and HL168147 (to X.C.).

Disclosures

None.