Decreased Endothelial Cell Retinoic Acid Signaling Accelerates Progression of Single Ventricle Pulmonary Vascular Malformations

Visual Abstract

Highlights

• •Lung ECs are the primary affected cell type early after Glenn.
• •Single ventricle PVMs share differentially expressed genes with ECs in HHT AVMs.
• •Targeting ATRA signaling is a novel therapeutic strategy for single ventricle PVMs.

Summary

Pulmonary vascular malformations (PVMs) are vascular complications that universally develop in patients with single ventricle congenital heart disease after Glenn surgery, yet the pathogenesis of single ventricle PVMs remains unknown. Using a surgical rat model of Glenn circulation, single-cell RNA sequencing (scRNAseq) on whole lung samples identified dramatic endothelial cell (EC) gene expression changes post-surgery, including decreased all-trans retinoic acid (ATRA) signaling. Many gene expression differences (∼17% of all EC differentially expressed genes), including decreased ATRA signaling, were shared with 2 independent models of hereditary hemorrhagic telangiectasia arteriovenous malformations (AVMs). Subsequent dietary modification of vitamin A (Vit A) intake in rats, identified that dietary deficiency of Vit A increased PVM shunting severity, which was mitigated with excess dietary Vit A intake. Collectively, our study of post-Glenn PVMs identified shared patho-mechanisms with hereditary hemorrhagic telangiectasia AVMs, including decreased EC ATRA signaling, which is a novel potential target for treating AVMs.

Pulmonary vascular malformations (PVMs) are abnormal vascular structures in the lungs that universally develop in patients with single ventricle congenital heart disease.1, 2, 3, 4, 5 Single ventricle PVMs develop in patients after undergoing surgical palliation with a Glenn procedure, whereby pulmonary blood flow comes exclusively from the superior vena cava (SVC) and hepatic vein blood is excluded from pulmonary blood flow.^6,7 Post-Glenn PVMs cause progressive intrapulmonary shunting that exacerbates preexisting hypoxemia.⁸ However, the pathophysiology of single ventricle PVMs remains unknown, including the molecular drivers of PVM pathogenesis, affected cell types, and similarities and differences with other clinical vascular malformations, including fast-flow arteriovenous malformation (AVM) conditions.

Previous studies used large and small animal models of unilateral Glenn circulation to investigate candidate pathways potentially altered in single ventricle PVMs.7, 8, 9, 10, 11, 12, 13, 14, 15, 16 One previous study with a rat Glenn model used an unbiased approach to identify affected signaling pathways post-Glenn.¹⁵ Tipps et al¹⁵ performed an RNA microarray using whole lung lysate and identified multiple dysregulated vascular remodeling pathways at 8 months post-surgery (ie, later stages of PVM pathophysiology). To date, though, no studies have investigated whole lung transcriptional changes post-Glenn at single-cell resolution or in the early stages of PVM initiation post-Glenn. Recently, we developed and published our own surgical rat model of unilateral Glenn circulation.¹⁶ We validated that there is pathologically increased intrapulmonary shunting as early as 2 weeks post-Glenn and that shunting progressively increases over the next 6 months. We also confirmed that our unilateral Glenn model causes no major systemic effects, including no differences in ventilation and clinically insignificant oxygenation differences. Thus, with our validated surgical Glenn model and lack of early transcriptomic studies after Glenn surgery, the objective of this study was to identify the early cellular and molecular drivers of single ventricle PVMs using single-cell RNA sequencing (scRNAseq) in an animal model of Glenn circulation during the initiation phase of PVMs post-Glenn.

Methods

Complete details for all experimental methods are available in the Supplemental Appendix.

Animals

Adult Sprague-Dawley rats (male and female, 6-8 weeks of age) were used for all experiments (Taconic). All rats were housed in the Biomedical Research Center at the Medical College of Wisconsin. All experimental rat protocols were approved by the Medical College of Wisconsin Institutional Animal Care and Use Committee before initiation of experimental protocols (Animal Use Agreement #7731).

Smad4 endothelial cell (EC)-specific, tamoxifen-inducible knockout mice were generated as previously described¹⁷ using conditional Smad4^f/f mice¹⁸ and Cdh5-Cre^ERT2 mice.¹⁹ To induce Smad4 gene deletion, 300 μg of tamoxifen (Sigma, T5648) was delivered via intraperitoneal injection at postnatal day 1. Smad4 animal protocols were approved and performed in accordance with Tulane University's Institutional Animal Care and Use Committee policies and adhere to ethical standards.

Left-sided superior cavopulmonary anastomosis (left-sided Glenn)

To model Glenn circulation in rats and create unilateral PVMs, we performed an end-to-end unilateral cavopulmonary anastomosis between the left SVC and left pulmonary artery, as previously described.¹⁶ For a comparative control animal, a sham surgery was performed on an age- and sex-matched cage-mate, as previously described.¹⁶ After surgery, Glenn and sham rats were co-housed until endpoint experiments at 3 weeks post-surgery. This time point was chosen to evaluate early molecular changes after Glenn surgery during PVM initiation, while avoiding immediate postsurgical changes.

Immunohistochemistry—Histologic analysis of early post-Glenn PVMs

Based on previously published methodology for assessing early AVM microvascular dysplasia,²⁰ we quantified the number of normally sized microvessels (<3 μm) and abnormally enlarged microvessels (>3 μm but <10 μm) throughout the lung using CD31 (R&D Systems, #AF3628) immunohistochemistry (IHC).

scRNAseq library preparation and sequencing

Left lower lung samples from Glenn and sham rats were processed in pairs on 4 separate days (n = 4 pairs, n = 8 total) using 10x Chromium GEM-X single-cell 3′ reagent kit. Library preparation and next-generation sequencing were completed on all samples in one batch at the Medical College of Wisconsin Mellowes Center for Genomic Science and Precision Medicine (RRID: SCR_022926). 10x Genomics Chromium workflow was initiated with approximately 10,000 cells per sample and processed according to the Chromium Next GEM Single-Cell 3ʹ Reagent Kits using the 10x Genomics Chromium Controller. Sequencing was completed on the Illumina NovaSeq 6,000 targeting 50,000 reads per cell per condition. scRNAseq data were deposited in the Gene Expression Omnibus (GEO) for independent analysis (GSE312833).

scRNAseq bioinformatic analysis data processing and quality control

scRNAseq raw reads were assigned to Ensembl Rat Genome (reference version Rnor_6.0.98) transcripts using CellRanger (10x Genomic Cloud analysis) with default parameters. The processed data of each sample were then analyzed using the R statistical software package Seurat (v5.0.0).^21,22 For quality control, we excluded cells with unique feature counts of ≤200 or ≥2,500 and cells with a mitochondrial gene fraction ≥5%. Global scaling was applied to normalize counts across all cells in each sample. Approximately 2,000 most variable genes in the data set were used for dimensionality reduction by principal component analysis. Cell clusters were determined using FindNeighbours and FindClusters functions based on the first 17 principal components. The Uniform Manifold Approximation Projection (UMAP) algorithm was used to visualize the cell clusters. Global gene expression distances were calculated using Euclidean distance of the average expression within each cell type for each data set.^23,24 The optimal number of cell clusters was determined using the Modularity Optimizer (version 1.3.0) and Louvain algorithm. To annotate cell clusters visualized in the combined UMAP, we used the Wilcoxon rank-sum test to identify enriched gene expression within each cluster. Genes with average log2 fold change (log2FC) > 1 and expression in >10% of cells within each cluster were compared with published references of rat, mouse, and human lung.24, 25, 26, 27, 28, 29, 30 To compare the abundance of cells within each cluster, we quantified the number of cells within each cluster for each sample relative to the total cells for that sample.

Differential gene expression between Glenn and sham rats

Model-based analysis of single-cell transcriptomics was performed to identify differentially expressed genes (DEGs) between Glenn and sham rats within each cluster. We used a cutoff of |log2FC| >0.5 and adjusted P value <0.05 (q value) to identify DEGs within each cluster. Bonferroni correction was used to adjust P values. To identify biological pathways dysregulated within each cluster, we uploaded DEGs into DAVID (Database for Annotation, Visualization, and Integrated Discovery) and entire cluster data sets (gene list, log2FC, and adjusted P value) into Ingenuity Pathway Analysis (IPA).

Rat lung EC isolation for scRNAseq validation

We isolated CD31⁺ cells from left lung samples from independent Glenn and sham rats at 3 weeks post-surgery to validate scRNAseq data. After lung dissociation, we separated CD31⁺ lung cells via magnetic microbeads. First, we performed negative selection to remove CD45⁺ lung cells (anti-rat CD45-FITC; BD, #554877; and anti-FITC microbeads; Miltenyi, #130-048-701) using MS columns (Miltenyi, #130-042-201). After collecting the CD45⁻ population, we performed positive selection to capture CD31⁺ lung cells (anti-rat CD31-PE; BD, #555027; and anti-PE microbeads; Miltenyi, #130-048-801) using MS columns. After isolation of CD45^-CD31⁺ lung cells, we extracted RNA and quantitative polymerase chain reaction (qPCR) was performed on the CFX96 qPCR machine (Bio-Rad). Quantification was performed using the ΔΔCt method using Gapdh as the housekeeping gene. Primers for qPCR were previously published or custom-designed using Primer 3 (Supplemental Table 1).³¹

Secondary analysis of endothelial bulk RNA sequencing

To identify potential molecular overlap with AVMs in hereditary hemorrhagic telangiectasia (HHT) and single ventricle PVMs, we compared our EC cluster 1 dataset with EC-specific RNA sequencing (RNAseq) datasets of 2 independent HHT mouse models: a previously published Acvrl1/Alk1 brain AVM model¹⁷ and Smad4 lung AVM model.^32,33 RNAseq data were obtained from isolated brain and lung ECs harvested from EC-specific, tamoxifen-inducible Acvrl1 (Alk1^iBEC) and Smad4 (Smad4^iLEC) knockout mice, respectively. Both models isolated brain and lung ECs, as previously described.^17,32 Bulk RNAseq was subsequently performed as previously described using a NextSeq 550 System (Illumina, SY-415-1002).³⁴

The differential expression analysis was performed on raw gene counts using DESeq2 (v1.47.5) after prefiltering to include only genes that have a count of at least 10 for half of the samples.³⁵ DEGs were identified in both HHT AVM datasets by adjusted P value <0.05 (Bonferroni correction) and then compared with DEGs previously identified in our EC cluster 1 Glenn dataset (Glenn DEGs: |log2FC| >0.5 and adjusted P value <0.05 [q value]). After identifying DEGs common to all 3 datasets, we identified overlapping DEGs that were also differentially expressed in the same direction in all 3 datasets (ie, concordant). RNAseq data for re-analyzed public datasets are found in the GEO (GSE197105), and RNAseq data for newly analyzed datasets were deposited in the GEO for independent analysis (GSE312375).

Dietary modification of vitamin A intake

Glenn and sham rats were started on custom diets (Teklad Diets, Inotiv) 1 to 2 weeks before surgery and continued for 3 weeks after surgery. Vitamin A (Vit A) control diet contained 20 IU/g Vit A, Vit A–deficient diet contained 0 IU/g Vit A, and Vit A excess diet contained 200 IU/g Vit A (Supplemental Table 2). Custom diets were identical except for Vit A content. Tissue (lung, liver, and plasma) retinoid levels (retinol and retinyl esters) were analyzed to assess the Vit A status of animals on different diets. In addition, the diets contained 610 ppm aspirin with an estimated daily dosing of 5 to 10 mg/kg/d, which is equivalent to standard of care pediatric clinical dosing of aspirin for children with Glenn circulation.

Quantification of intrapulmonary shunting

To assess the functional impact of single ventricle PVMs, we objectively quantified intrapulmonary shunting of 10 μm fluorescent microspheres (ThermoFisher, F8834) within the left lung, as previously described.¹⁶ Microspheres and phosphate-buffered saline flush were aspirated from the left atrium ensuring to avoid negative pressure suction of microspheres through the pulmonary vasculature. Microspheres were quantified using a fluorescent plate reader (SpectraMax i3x, ex/em 580/605). Shunt fraction was calculated as a percentage relative to the fluorescence of noninjected microspheres: (collected spheres/noninjected spheres) × 100.

Flow cytometry quantification of intracellular retinoid levels

We used spectral cytometry to quantify the prevalence of retinoid-storing cells and intracellular retinoid levels, as previously described.36, 37, 38 After lung dissociation, we stained freshly isolated lung samples with anti-CD31-PE antibody (BD Biosciences, #555027), anti-CD45-APC-Cy7 antibody (BD Biosciences, #561586), and viability dye (Live/Dead Yellow, Thermo, #L64968). Using a Cytek Spectral Cytometer (Cytek Aurora) without autofluorescence extraction, we quantified the median fluorescent intensity (MFI) of all live cells at UV5 channel as a read-out for the amount of intracellular retinoids (retinol and retinyl esters). We also quantified the abundance of retinoid⁺ cells, MFI at UV5 channel within retinoid⁺ cells, abundance of retinoid⁺CD31⁺ cells, and MFI at UV5 channel within retinoid⁺CD31⁺ cells.

Bulk RNAseq analysis of sorted CD31⁺ lung cells

We performed bulk RNAseq on microbead-sorted CD31⁺ lung cells 3 weeks after Glenn or sham surgery on diet-treated rats. RNA was extracted from microbead-sorted CD31⁺ lung cells, as described previously. Library preparation was completed at the Medical College of Wisconsin Mellowes Center for Genomic Science and Precision Medicine, using the SMART-Seq Ultra Low v4 Kit (Takara). Sequencing was subsequently performed on the NovaSeqX platform at the University of Wisconsin-Madison Center for Genomic Science Innovation targeting 40 millions unique reads per sample. RNAseq output was aligned to the Norway rat genome (mRatBN7.2/rn7). RNAseq data were deposited in the GEO for independent analysis (GSE311758).

Differential gene expression was analyzed using DESeq2 (v1.47.5), as previously described.³⁵ We first identified DEGs (adjusted P value <0.05 using Bonferroni correction) between Glenn and sham rats on each diet. Next, we compared DEGs among all 3 diets to identify overlapping and unique DEGs. From the unique DEGs on each diet, we identified impacted biological pathways using DAVID.

Statistical Analysis

Data are expressed as mean and SD for continuous data and count (percentage) for categorical data unless otherwise stated. Bioinformatic approaches for scRNAseq and bulk RNAseq analyses are detailed previously. Statistical approaches for individual experiments are reported in each figure legend and detailed in the Supplemental Appendix. Normality was determined using the Shapiro-Wilk test. A log transformation was used when parametric assumptions were violated (eg, normality). A P value <0.05 was considered statistically significant unless otherwise specified. Analyses were performed using GraphPad Prism 11 (GraphPad Software) and SAS 9.4 (SAS Institute Inc).

Results

Histologic analysis of single ventricle PVMs

Histologic analysis of the pulmonary microvasculature at 3 weeks post-surgery using CD31 IHC (Figure 1) demonstrated that the left lung of post-Glenn rats had increased numbers of enlarged microvessels (>3 μm) compared with the left lung of post-sham rats (P < 0.001) and compared with the contralateral unaffected right lung of post-Glenn rats (P < 0.001) (Figure 1F). There were also fewer microvessels <3 μm in the left lung of post-Glenn rats compared with sham control (P = 0.003) and the contralateral lung of post-Glenn rats (P = 0.006) (Figure 1G). There were no differences identified in microvessel size in the lower half vs the upper half of the left lung from post-Glenn rats (Supplemental Figure 1).

Figure 1.

Single Ventricle Pulmonary Vascular Malformations Manifest With Diffusely Increased Microvascular Diameter

(A and B) Illustrations of experimental design showing the origin of different lung samples, including the control left lung from sham surgical rats that underwent left thoracotomy without microvascular anastomosis, control right lung from postsurgical Glenn rats (contralateral to left thoracotomy), experimental left lung from postsurgical Glenn rats (perfused by microvascular anastomosis after left thoracotomy). (C to E) Representative immunohistochemistry images of lungs of rats 3 weeks after sham surgery or after Glenn surgery. Scale bar indicates 50 μm. (F) Scatter plot with bars indicating mean and SD for the number of microvessels >3 μm in the smallest diameter, indicating abnormally increased microvascular diameter (ie, “dysplasia index”). Values represent averages over six ×80 magnification fields per animal. Glenn left lung values are combined from both the lower and upper left lung. Vessels larger than 10 μm in diameter were excluded from analysis. One-way analysis of variance with Sidak's multiple comparisons test; n = 6 sham and n = 6 Glenn rats. ∗∗∗P < 0.001. (G) Scatter plot with bars indicating mean and SD for the number of microvessels <3 μm in the smallest diameter. Values represent averages over six ×80 magnification fields per animal. Glenn left lung values are combined from both the lower and upper left lung. One-way analysis of variance with Sidak's multiple comparisons test; n = 6 sham and n = 6 Glenn rats. ∗∗P < 0.01.

scRNAseq dataset of rat lungs after Glenn or sham surgery

We performed scRNAseq in Glenn and sham rats using 8 samples (n = 4 Glenn and n = 4 sham) (Figure 2A) and a total of 79,846 cells (average of 9,980 cells/sample). After applying conservative quality control filters, we included a total of 26,454 cells (average of 3,307 cells/sample), which included 15,388 cells (58.2%) from Glenn rats and 11,066 cells (41.8%) from sham rats.

Figure 2.

Single-Cell RNA Sequencing of Rat Lung Cells Early After Glenn or Sham Surgery Identifies Transcriptionally Distinct Cell Clusters

(A) Schematic for experimental workflow for performing scRNAseq in Glenn (n = 4) and sham (n = 4) rats. (B) Combined uniform manifold approximation and projection plot with all samples (n = 8) showing distinct cellular clusters identified with unsupervised clustering and subsequent cluster annotation. Wilcoxon rank-sum test used to identify enriched gene expression within each cluster. (C) Uniform manifold approximation and projection plot color coded by surgical group (Glenn vs sham) demonstrates pronounced separation within cluster 1 (general lung endothelial cells) and cluster 3 (capillary lung endothelial cells) between Glenn and sham rats. (D) Dot plot of marker genes identifying each cluster. Dot size indicates percentage of cells within each cluster expressing the specific gene and dot color indicates the average log2 fold change of gene expression within each cluster compared with all other clusters. (E) Global heatmap of all clusters showing transcriptional gene expression differences for the top 5 genes within each cluster. Each column represents an individual cell and each row represents a gene.

Altered lung cell populations

Based on unsupervised cell clustering of our data set and annotation based on available lung references,24, 25, 26, 27, 28, 29, 30 we identified 13 distinct cell clusters, including 3 lung EC clusters (Figures 2B to 2E). Color coding the global UMAP by surgical group demonstrated transcriptional variation within clusters that was most pronounced within cluster 1, which is the general EC cluster (Figure 2C). Glenn/sham harvest pairs (Supplemental Figure 2A) and individual samples (Supplemental Figure 2B) were distributed across all clusters, suggesting no evidence of technical bias or artifact.

Cluster 1 showed the most discrepant transcriptional signatures with 16.7% (1,310 of 7,849) of all genes differentially expressed between Glenn and sham rats (Figure 3A). Cluster 0 and cluster 3 also showed marked differences with 4.9% (355 of 7,215) and 3.7% (321 of 8,303) of genes differentially expressed, respectively. All other clusters had ≤1.3% genes differentially expressed. Next, we quantified whole transcriptome gene expression differences within each cluster using a Euclidean distance–based statistical approach,^23,24 which validated that cluster 1 ECs were the most transcriptionally distinct cell type between Glenn and sham rats followed by cluster 0 fibroblasts and cluster 3 capillary ECs (Figure 3B). Last, there were significantly fewer lung capillary ECs (cluster 3) in Glenn rats compared with sham rats (Glenn 2.7 ± 0.4%, sham 6.9 ± 1.6%; q value = 0.025); however, there were no significant differences in cell numbers in any other cluster (Figure 3C). Altogether, these data indicate that ECs are the primarily affected cell type in the lung early after Glenn palliation, including that there is potential loss of capillary ECs as early as 3 weeks after Glenn.

Figure 3.

Lung Endothelial Cells Are the Predominantly Affected Cell Type Early After Glenn Surgery

(A) Scatter plot of all clusters showing all genes identified within each cluster. Colored dots indicate differentially expressed genes, defined as average |log2 fold change (log2FC)| >0.5 and adjusted P value <0.05 (q value). Gray dots indicate genes within each cluster that were not differentially expressed. Percent annotated along the x-axis indicates the percentage of genes within each cluster that were differentially expressed between Glenn and sham. Cluster 1 endothelial cells were again the most strongly affected cluster with 16.7% of all genes differentially expressed. (B) Whole transcriptome comparison of global gene expression differences between Glenn and sham rats within each cluster based on the Euclidean-distance–based statistical approach. Cluster 1 endothelial cells had the most distinct gene expression differences between Glenn and sham. (C) Stacked bar graph with accompanying table indicates the mean ± SD of percent cells within each cluster from each rat. The q value indicates the adjusted P value for multiple unpaired t-tests using Benjamini 2-stage step-up to compare the frequency of cells within each cluster. Cluster 3 capillary endothelial cells were significantly decreased in Glenn rats compared with sham rats; no differences between groups in all other clusters.

DEGs in rat Glenn lung ECs

Within cluster 1 ECs, we identified 755 (57.6%) up-regulated and 555 (42.4%) down-regulated genes (average |log2FC| >0.5, q value <0.05) (Figure 4A). Using all DEGs within cluster 1 (n = 1,310), we used DAVID pathway analysis to identify dysregulated biological processes (Figure 4B). Angiogenesis was the most significantly dysregulated pathway. Multiple pathways overlapping angiogenesis and vascular remodeling included positive regulation of cell migration, positive regulation of cell population proliferation, and vasculogenesis. In addition, several unexpected inflammatory response pathways were identified, including response to lipopolysaccharide (LPS), cellular response to tumor necrosis factor, cellular response to interleukin (IL)-1, and response to cytokine. Next, we used IPA to identify dysregulated canonical signaling pathways (Figure 4D). Multiple canonical pathways supported the DAVID analysis identifying angiogenesis, including increased phosphatidylinositol-3 kinase (PI3K)/AKT signaling, increased TGF-β signaling, decreased angiopoietin signaling, and decreased bone morphogenetic protein (BMP) signaling. We also re-identified pathways associated with inflammation, including pathogen-induced cytokine storm and macrophage migration inhibitory factor regulation of innate immunity. Finally, we identified decreased activity of key regulatory pathways, including signaling by retinoic acid, antioxidant action of vitamin C, and transport of vitamins. Altogether, cluster 1 DEGs and altered signaling pathways support that early PVM initiation is likely multifactorial and may be driven by pro-angiogenic/pro-inflammatory stimuli and derangement of regulatory pathways controlled by retinoic acid and vitamin C.

Figure 4.

General Lung Endothelial Cells Comprise Cluster 1 Cells and Are Significantly Dysregulated in Glenn vs Sham Rats

(A) Volcano plot showing up-regulated (red) and down-regulated (blue) genes identified in cluster 1 with annotation of notable differentially expressed genes. (B) Dot plot showing the top 25 biological pathways identified using DAVID (Database for Annotation, Visualization, and Integrated Discovery) pathway analysis and cluster 1 differentially expressed genes. (C) Heatmap of the top 25 up-regulated and top 25 down-regulated differentially expressed genes in cluster 1. Each column represents an individual rat (n = 4 sham, n = 4 Glenn), and each row represents a gene. (D) Dot plot of the 12 most activated and 12 most de-activated canonical pathways identified using Ingenuity Pathway Analysis (IPA) and the entire cluster 1 dataset. Positive activation z-score indicates increased activity of the associated pathway in Glenn rats, whereas negative activation z-score indicates decreased activity of the associated pathway in Glenn rats. CDX = caudal-type homeobox transcription family; LXR = liver X receptor; PTEN = phosphatase and tensin homolog; RHOA = Ras homolog family member A; RXR = retinoid X receptor.

Within cluster 3, we identified 165 (51.4%) up-regulated and 156 (48.6%) down-regulated genes (average |log2FC| >0.5, q value <0.05) (Supplemental Figure 3A). Using all DEGs within cluster 3 (n = 321), we used DAVID pathway analysis to identify dysregulated biological processes (Supplemental Figure 3B). Similar to cluster 1, angiogenesis was the most significantly dysregulated pathway along with multiple other vascular remodeling pathways, including positive regulation of angiogenesis, blood vessel remodeling, vasculogenesis, and more. Similarly, we again identified multiple inflammatory response pathways, including response to LPS, cellular response to IL-1, cellular response to cytokine stimulus, and response to cytokine. Next, we again used IPA to independently identify dysregulated canonical signaling pathways (Supplemental Figure 3D). Similar to cluster 1, we identified angiogenic/vascular remodeling pathways (activin inhibin signaling pathway and signaling by Notch3) and inflammatory pathways (interferonα/β signaling, pathogen-induced cytokine storm, but decreased IL-1 signaling).

We then further subdivided vascular lung ECs (clusters 1 and 3) to identify whether Glenn circulation preferentially affected arterial, venous, or capillary ECs. Based on unsupervised clustering, the 2 largest clusters in our subclustering analysis did not segregate based on classical arterial, venous, or capillary markers (Supplemental Figures 4A and 4B). Rather, the largest clusters (subclusters 0 and 1) differentiated based on previously identified DEGs between Glenn and sham ECs without evidence of technical bias based on Glenn/sham harvest pair or individual sample (Supplemental Figures 4C and 4D). Quantifying the contribution to each subcluster, there were significant differences (q value <0.001) in the number of cells from each surgical group within subclusters 0 and 1 (Supplemental Figure 4E). We again identified fewer capillary ECs (subcluster 2) in the Glenn group (q value = 0.009). Global heatmap of EC subclusters identifies the top 10 enriched genes within each subcluster (Supplemental Figure 4F). Additional feature plots of pan-EC markers and enriched genes within subclusters highlight differences among subclusters (Supplemental Figure 5).

Validation of scRNAseq endothelial gene expression with isolated rat lung ECs

To validate scRNAseq DEGs, we performed qPCR from isolated CD31⁺ lung cells from independent Glenn (n = 6) and sham (n = 6) rats (Figure 5; Supplemental Figure 6). Multiple genes in angiogenic pathways were confirmed to have significantly different RNA levels that were consistent with our scRNAseq data set, including increased Angpt2 (P = 0.015), Cxcr4 (P = 0.009), and Esm1 (P = 0.002), as well as decreased Acvrl1 (P = 0.004), Tmem100 (P = 0.041), and Vegfa (P = 0.009). We confirmed consistent gene expression differences in inflammation (increased Il6st [P = 0.009], increased Sparcl1 [P = 0.002], decreased Serpina1 [P = 0.002]), metabolism (decreased Lpl [P = 0.004], increased Pfkfb3 [P = 0.002]), and vasodilation (increased Nos2 [P = 0.026] and decreased Edn1 [P = 0.009]) (Figure 5C). We also confirmed decreased expression of multiple lung capillary markers, including both aerocyte (Cap1) capillary markers (Car4, P < 0.001; Ednrb, P = 0.001) and general lung (Cap2) capillary markers (Plvap, P = 0.002; Slc6a4, P < 0.001) (Supplemental Figure 7).²⁵ At the protein level, we also confirmed decreased levels of Slc6a4 in post-Glenn rats using IHC (Supplemental Figure 8).

Figure 5.

Glenn Circulation Affects Multiple Signaling Pathways in Lung Endothelial Cells

(A) Heatmaps of gene expression z-scores for differentially expressed genes in cluster 1 endothelial cells organized by general endothelial cell function: angiogenesis/vascular remodeling, inflammation and stress response, vasodilation, matrix remodeling, and metabolism. (B) Schematic for experimental methodology for isolating RNA from CD31⁺ sorted lung cells from sham (n = 6) and Glenn (n = 6) rats. (C) Scatter plot with bars (mean ± SD) shows validation of specific gene targets using quantitative reverse transcriptase polymerase chain reaction (qPCR); unpaired t-tests; ∗P < 0.05,∗∗P < 0.01. Genes are grouped by general biological processes, including angiogenesis, inflammation, metabolism, and vasodilation. Biological processes in (A) and (C) are not mutually exclusive as genes frequently perform multiple functions.

Comparison of endothelial gene expression in HHT and post-Glenn PVMs

Based on gene expression patterns observed in post-Glenn lung ECs that recapitulated gene expression patterns observed in HHT pathophysiology (Acvrl1, Angpt2, Esm1, Tmem100, and more),^17,34,39,40 we hypothesized that there may be shared endothelial dysregulation in post-Glenn PVMs and HHT AVMs. To identify potential shared gene expression patterns, we compared our list of endothelial DEGs to 2 independent data sets derived from Smad4 and Acvrl1 HHT mouse models using isolated ECs from lungs and brains, respectively (Smad4^iLECs and Alk1^iBECs) (Figure 6).¹⁷ We identified a total of 755 and 4,828 DEGs in the Smad4 and Alk1 data sets, respectively, in addition to the previously identified 1,310 DEGs in cluster 1 of our rat Glenn scRNAseq data set (Figure 6A). Comparing gene lists among all 3 datasets, we identified 136 (18.0%) overlapping DEGs and 88 (65% of overlapping DEGs) overlapping and concordant DEGs (Figures 6B and 6C). Among the 88 concordantly shared DEGs, we identified multiple expected angiogenic-related genes, including Angpt2, Cxcr4, Esm1, and Etv4. Surprisingly, we also identified shared decreased expression of retinoic acid receptor γ (Rarγ) and cytochrome P450 family 26 subfamily B member 1 (Cyp26b1), genes involved in and regulated by all-trans retinoic acid (ATRA) signaling, suggesting that decreased endothelial ATRA signaling may be a shared pathophysiologic feature of HHT AVMs and post-Glenn PVMs (Figure 6C).

Figure 6.

Conserved Endothelial Gene Expression Patterns in Transgenic Mouse Models of Hereditary Hemorrhagic Telangiectasia and Rat Model of Glenn Circulation

(A) Volcano plots for each model showing up-regulated (red) and down-regulated (blue) genes identified in endothelial-sorted bulk RNA sequencing (mouse models) and cluster 1 endothelial cells (ECs) in small-cell RNA sequencing (scRNAseq). Differentially expressed genes (DEGs) strictly defined based on adjusted P value <0.05. (B) Venn diagram showing unique and overlapping DEGs among each model. There were 136 overlapping DEGs common to all 3 models (of a possible 755 DEGs). Among the 136 overlapping DEGs, 88 (65%) were differentially expressed with log2FC in the same direction (ie, concordant differential expression) in all 3 models. (C) Heatmap of all 3 models identifying the 88 overlapping and concordant genes. Purple asterisks identify genes involved in all-trans retinoic acid signaling (RARγ and Cyp26b1), indicating that decreased all-trans retinoic acid signaling is common in all 3 models.

Validation of decreased retinoid levels in post-Glenn lung ECs

To determine the extent to which ATRA signaling may be dysregulated in lung ECs post-Glenn (Figure 7A), we further analyzed ATRA-related genes in our scRNAseq cluster 1 ECs and independently validated a subset of DEGs with CD31⁺ sorted qPCR (Figures 7A to 7C) and IHC (Supplemental Figure 9). In our scRNAseq cluster 1, we identified multiple DEGs associated with retinoid metabolism, including increased Adh6a (Log2FC = +1.2; q value = 1.7E-2; encodes a dehydrogenase enzyme that can oxidize retinol to retinaldehyde in the first step toward the active ATRA synthesis), increased Nceh1 (Log2FC = +1.0; q value = 6.1E-8; encodes a putative retinyl ester hydrolase that converts stored retinyl esters to retinol),⁴¹ decreased Cyp26b1(Log2FC = -3.2; q value = 1.6E-76; encodes a specific and potent catabolizing enzyme of ATRA), decreased Dhrs3 (Log2FC = -1.1; q value = 4.0E-28; encodes a retinaldehyde reductase to limit excess levels of ATRA), decreased Rarγ (Log2FC = -1.0; q value = 3.2E-9; encodes a γ isoform of retinoic acid nuclear receptor transcription factor that regulates ATRA-mediated gene expression), and decreased Rarres1 (Log2FC = -2.9; q value = 3.1E-3; encodes a transmembrane protein involved in cell proliferation and whose expression is retinoic acid responsive) (Figures 7A and 7B). Independent validation of a subset of scRNAseq DEGs with CD31⁺ sorted qPCR confirmed decreased gene expression of Cyp26b1 (P = 0.015) and Dhrs3 (P = 0.015) (Figure 7C). Among the canonical retinoic acid nuclear receptor transcription factors, only Rarγ was decreased in Glenn rats (P = 0.026) (Figure 7C).

Figure 7.

Decreased ATRA Signaling and Intracellular Retinoid Levels in Lung Endothelial Cells After Glenn

(A) Illustration of pathways for retinoid (retinol and retinyl ester) uptake, metabolism, and signaling. Red colored gene names are up-regulated genes in cluster 1 endothelial cells, blue colored gene names are down-regulated genes in cluster 1 ECs, and black colored gene names encode canonical enzymes in retinoid metabolism but are not detected (Stra6) or not differentially expressed (Lrat, Rdh10, Aldh1a1) in cluster 1 endothelial cells. (B) Heatmap of gene expression z-scores for significantly differentially expressed genes associated with ATRA signaling and metabolism in cluster 1 endothelial cells. (C) qPCR validation of a subset of ATRA genes in CD31⁺ sorted lung cells after Glenn or sham surgery validating the scRNAseq data (n = 6 sham and n = 6 Glenn rats; unpaired t-tests; ∗P < 0.05, ns = not statistically significant). (D-F) Spectral cytometry quantification of endogenous intracellular retinoid levels based on retinoid autofluorescence at UV5 spectral channel. There is no difference in the total amount of retinoid-containing (retinoid⁺ cells) in the lung (P = 0.17), although there are decreased intracellular retinoid levels based on median fluorescent intensity (MFI) in post-Glenn rats (P = 0.031). Scatter plot with bars indicates mean ± SD (n = 5 sham rats and n = 7 Glenn rats; unpaired t-test; ∗P < 0.05, ns = not statistically significant). (G-I) Spectral cytometry quantification of retinoid⁺CD31⁺ lung cells identified fewer retinoid⁺CD31⁺ cells after Glenn surgery (P = 0.012) and decreased retinoid levels (P = 0.018). Scatter plot with bars indicates mean ± SD (n = 5 sham rats and n = 7 Glenn rats; unpaired t-test; ∗P < 0.05).

Interestingly, we also identified significantly decreased expression of Lpl (scRNAseq cluster 1: Log2FC = -1.7; q value = 1.6E-76) (Figures 4A and 5A) (CD31⁺ sorted qPCR: P = 0.004) (Figure 5C), which encodes a lipase enzyme critical for retinyl ester hydrolysis from chylomicrons to facilitate lung EC uptake of free retinol.^37,42 In contrast, Stra6, which encodes the cell surface receptor for retinol uptake of liver-derived retinol-RBP4 complex, was not detected in cluster 1 ECs or via qPCR in CD31⁺ sorted cells. This is consistent with previously published data that lung ECs primarily uptake retinoids via chylomicron-derived retinoids and not liver-derived retinol-RBP4.³⁷

Looking beyond gene expression, we further quantified endogenous retinoid levels in lung cells 3 weeks after Glenn or sham surgery using spectral cytometry based on retinoid autofluorescence. Among cells with strongly detectable intracellular retinoids, there was no difference in the percentage of retinoid⁺ cells (P = 0.166) (Figures 7D and 7E); however, there were significantly decreased levels of intracellular retinoids detected in these cells after Glenn (P = 0.031) (Figure 7F). Looking more specifically at lung ECs, there were significantly fewer retinoid⁺CD31⁺ lung cells (P = 0.012) (Figures 7G and 7H) and significantly decreased retinoid levels in these cells after Glenn (P = 0.018) (Figure 7I). Using high-performance liquid chromatography to compare retinoid tissue concentrations in the lungs (retinol and retinyl esters), there were no statistical differences in whole lung Glenn samples compared with sham samples (Supplemental Figures 10A and 10B), which was similar to our spectral cytometry-based quantification of lung retinoids in all live lung cells (Supplemental Figure 10C). Collectively, these data suggest that retinoids are decreased in lung ECs after Glenn and are preserved in other non-EC cell types.

Dietary Vit A intake affects progression of post-Glenn PVMs

To investigate whether dietary intake of Vit A affects PVM pathogenesis, we first validated that dietary modification of Vit A intake decreases retinoid levels in the lung (Supplemental Figures 11A to 11C and 12A and 12B) and decreases retinoid levels specifically in lung ECs (Supplemental Figures 11D to 11F). As expected, high-performance liquid chromatography quantification confirmed that dietary modification significantly affected retinoid storage in the liver (Supplemental Figures 12C and 12D) but did not alter plasma retinol levels (Supplemental Figure 12E), supporting that circulating retinol levels do not reflect tissue-specific retinoid levels. In contrast, and as expected, dietary modification of Vit A significantly altered plasma retinyl esters levels (Supplemental Figure 12F) given that retinyl esters circulate in plasma in diet-derived chylomicrons.

We next assessed whether modifying Vit A dietary intake affects post-Glenn PVM progression (Figure 8) by quantifying intrapulmonary shunting (Figure 8B) and microvessel dysplasia (Figures 8C to 8J). Similar to our previously published data at 2 weeks post-surgery, Glenn rats on Vit A control diet had increased shunting compared with the diet-matched sham control rats (P = 0.001).¹⁶ Significantly, Glenn rats on a Vit A–deficient diet had increased shunting compared with diet-matched sham control rats (P < 0.001), as well as compared with Glenn rats on Vit A control diet (P < 0.001) and Glenn rats on Vit A excess diet (P < 0.001). In fact, Glenn rats on Vit A–deficient diet approached the progressive shunting we previously identified at 6 months post-Glenn on standard rat chow (1.54 ± 0.23%).¹⁶ Equally important, Vit A–deficient diet did not increase shunting in sham rats, whereas Glenn rats on Vit A excess diet did not exhibit increased shunting compared with diet-matched sham control rats (P = 0.23) (Figure 8B). Collectively, these data support that decreased retinoid availability may be critical for PVM progression but not PVM initiation, and that increasing lung EC retinoid availability may mitigate PVM pathogenesis.

Figure 8.

Decreased Dietary Vitamin A Promotes Progression of Post-Glenn PVMs

(A) Experimental schematic for modification of dietary vitamin A intake. Diets started 1 to 2 weeks before surgery at 6 to 7 weeks of age and continued for 3 weeks after surgery for assessment of microsphere shunting and histology microvessels. (B) Scatter plot with bars indicating mean and SD for quantification of intrapulmonary shunting with 10 μm fluorescent microspheres. There is increased shunting in Glenn rats on control diet compared with diet-matched sham control (P = 0.001), as well as increased shunting in Glenn rats on deficient diet compared with diet-matched sham control (P < 0.001) and Glenn rats on control diet (P < 0.001) and excess diet (P < 0.001). Glenn rats on excess diet did not have increased shunting compared with diet-matched sham control (P = 0.82) or any other group. n = 5 to 6 rats per group; 2-way analysis of variance with Tukey's post hoc test; ∗∗P < 0.01, ∗∗∗P < 0.001, ns = not statistically significant. (C-H) Representative immunohistochemistry images identifying CD31⁺ lung microvessels from sham and Glenn rats after dietary vitamin A modification. Scale bar indicates 50 μm. (I) Scatter plot with bars indicating mean and SD for the number of microvessels >3 μm in the smallest diameter, indicating abnormally increased microvascular diameter (ie, “dysplasia index”). Values represent averages over five ×80 magnification fields per animal. Vessels larger than 10 μm in diameter were excluded from analysis. n = 3 to 5 rats per group; 2-way analysis of variance with Tukey's post hoc test; ∗∗P < 0.01, ∗∗∗P < 0.001, ns = not statistically significant. (J) Scatter plot with bars indicating mean and SD for the number of microvessels <3 μm in the smallest diameter. Values represent averages over five ×80 magnification fields per animal. n = 3 to 5 rats per group; 2-way analysis of variance with Tukey's post hoc test; ∗∗P < 0.01, ∗∗∗P < 0.001, ns = not statistically significant.

In addition to functional intrapulmonary shunting, we also assessed whether modifying Vit A dietary intake affects microvessel dysplasia using CD31 IHC (Figures 8C to 8J). Similar to shunting, Glenn rats on Vit A control diet had increased numbers of enlarged dysplastic microvessels (>3 μm) compared with diet-matched sham control rats (P < 0.001). Glenn rats on Vit A–deficient diet also had increased numbers of enlarged dysplastic microvessels compared with diet-matched sham control rats (P < 0.001), as well as compared with Glenn rats on Vit A control diet (P = 0.005) and Glenn rats on Vit A excess diet (P < 0.001). Glenn rats on Vit A excess diet did not have increased numbers of enlarged dysplastic microvessels compared with diet-matched sham control rats (P = 0.41). These data further support that retinoid bioavailability is critical for PVM progression and that increasing lung EC retinoids may mitigate PVM pathogenesis.

Next, after identifying that a Vit A–deficient diet promotes post-Glenn PVM progression, we sought to determine how dietary Vit A modification affects EC gene expression by performing bulk RNAseq on sorted CD31⁺ lung cells (Figure 9). After identifying DEGs (q value <0.05) between Glenn and sham rats on each diet, we observed many reproducibly shared DEGs across all 3 diets (n = 2,383 shared DEGs) indicating a strong and reproducible Glenn effect on lung ECs (Figure 9B). We also identified many unique DEGs specific to each diet, particularly with n = 1,090 unique DEGs on the Vit A–deficient diet and n = 817 unique DEGs on the Vit A excess diet (Figure 9B). Analyzing these unique DEGs with DAVID, we identified multiple significantly enriched biological processes (Figures 9C and 9D). Most notably, unique deficient diet DEGs were associated with positive regulation of PI3K/AKT signaling, whereas unique excess diet DEGs were associated with negative regulation of PI3K/AKT signaling and phosphatidylinositol dephosphorylation. Deficient diet DEGs were also associated with altered lipid metabolism, lipid biosynthesis, increased ERK1/2 signaling, and others. Excess diet DEGs were additionally associated with multiple EC-nonautonomous biological processes, including macrophage differentiation, monocyte differentiation, and more. Altogether, these data suggest that decreased EC ATRA signaling may promote PVM progression through increased PI3K signaling, whereas increasing EC ATRA may mitigate PVM progression through decreased PI3K signaling and potentially EC-nonautonomous mechanisms involving crosstalk with innate immune cells.

Figure 9.

Dietary Vitamin A Modification Differentially Affects Lung Endothelial Cell Gene Expression Regulating PI3K Signaling

(A) Experimental schematic for performing RNA sequencing (RNAseq) on CD31⁺ sorted cells, including n = 5 sham rats and n = 5 Glenn rats per dietary group; total n = 30 rats. (B) Venn diagram showing differentially expressed genes (DEGs) for each Glenn vs sham diet-matched comparison and the corresponding overlapping and unique DEGs for each group. (C) Impacted biological processes for the unique DEGs from rats on Vit A–deficient diet. (D) Impacted biological processes for unique DEGs from rats on Vit A excess diet.

Discussion

This study is the first analysis of how Glenn circulation alters lung gene expression at single-cell resolution early after Glenn surgery in a surgical rat model. We identified significant transcriptional differences in lung ECs at this early stage of PVM initiation. We also identified that multiple molecular signaling pathways were dysregulated in ECs between Glenn and sham rats, including overlap with HHT pathophysiology, increased angiogenesis, and decreased ATRA signaling. Validation of our scRNAseq data confirmed decreased retinoid levels in lung ECs early after Glenn surgery. Importantly, we also showed that modification of dietary Vit A intake affects lung EC gene expression and progression of post-Glenn PVM shunting.

Glenn PVM pathophysiology

Previous groups reported their findings using animal models of unilateral Glenn circulation, although no studies have previously used next-generation sequencing, cell-specific sequencing, or scRNAseq.7, 8, 9, 10, 11, 12, 13, 14, 15 More than 15 years ago, Tipps et al¹⁵ reported their findings after performing a microarray using total RNA from the whole left lung lysate in rats 8 months post-surgery. They reported altered expression of genes associated with angiogenesis, vascular remodeling, vasodilation, and several other processes. Consistent with the seminal study by Tipps et al,¹⁵ many of these same expression patterns were observed in ECs in our scRNAseq experiment and confirmed with qPCR, including increased Angpt2 and Esm1, and decreased Acvrl1, Edn1, Glp1r, and Cyp26b1. Moreover, we expanded the scope and resolution of molecular profiling. We studied lung tissue early after surgery during the PVM initiation phase, and we confirmed that these gene expression changes are specifically occurring in lung ECs.

Glenn pathophysiology vs HHT pathophysiology

Similar to the finding by Tipps et al¹⁵ of decreased Acvrl1 expression, we identified differential expression of multiple genes associated with BMP-TGF-β signaling that overlap with the pathogenesis of HHT AVMs. Our scRNAseq data set identified that 2 causative genes of HHT (Acvrl1 and Eng) were down-regulated in rat Glenn lung ECs, consistent with loss of function of these genes in HHT AVM pathogenesis. In addition, we found multiple genes in this pathway were dysregulated in a similar manner to HHT AVM pathogenesis, including increased Angpt2 and decreased Id1, Id3, and Tmem100.^{17,34,39,40,43} This is consistent with a study by Bartoli et al⁴⁴ previously reporting that increased angiopoietin signaling may contribute to single ventricle PVMs. Collectively, these data establish that there is overlap in the pathogenesis of single ventricle PVMs and HHT AVMs. Accordingly, our bioinformatic analysis using IPA predicts with a high statistical probability (P = 7.5E-18) that GDF2 (also known as BMP9 and a known ligand for ALK1 and ENG) is a potential upstream regulator de-activated in Glenn circulation (activation z-score -3.7). Capasso et al⁴⁵ previously investigated patient blood samples to determine whether BMP9 might be enriched in hepatic vein blood; however, no differences in BMP9 plasma levels in hepatic vein blood vs SVC blood were found. This suggests that nonpulsatile and decreased flow in Glenn circulation may have a critical role in single ventricle PVM pathogenesis—potentially priming the vascular endothelium for disease or potentiating the effects of circulating biochemical factors in hepatic vein blood.⁴⁶

Beyond the conventional genes involved in BMP-TGF-β signaling, we identified conserved gene expression patterns in both HHT models and our Glenn model that are consistent with decreased ATRA signaling. Because ATRA signaling controls gene expression of hundreds of downstream genes and multiple downstream signaling pathways, decreased ATRA signaling may act as a central hub for coordinating vascular remodeling in AVM pathogenesis. Our dietary modification of decreased Vit A intake significantly increased PVM shunting in our rat Glenn model, potentially by promoting PI3K signaling, which is a common driver of HHT-associated AVMs,⁴³ whereas decreased Vit A intake did not initiate shunting in sham surgical rats. Collectively, these data suggest that ATRA signaling may act as a conserved and critical regulator of PVM pathogenesis such that decreased ATRA signaling promotes PVM progression. Whether ATRA can modify AVM pathophysiology in other vascular beds or AVM conditions remains unknown.

In addition to angiogenesis and ATRA signaling, we identified that lung ECs in Glenn circulation up-regulated cellular responses to potentially multiple inflammatory cytokines, including IL-1, TNF, and LPS (Figures 4 and 5). To our knowledge, inflammation has not been previously identified in single ventricle PVMs, although inflammation is a well-recognized aspect of HHT-related AVMs and single ventricle circulation.47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57 Decreased lung EC ATRA signaling specifically in lung ECs is associated with increased lung leukocyte infiltration even without exogenous inflammatory stimuli,³⁷ suggesting that ATRA signaling may in part be contributing to post-Glenn lung inflammation. In addition, we previously identified that SVC serum obtained from patients with Glenn circulation was enriched in multiple cytokines and chemokines compared with paired hepatic vein serum samples.⁵⁸ Whether broad or specific anti-inflammatory therapies can modify single ventricle PVM pathogenesis, or whether ATRA signaling is contributing to post-Glenn inflammation and potentially leukocyte recruitment, is currently untested and unknown but warrants further investigation.

Last, we identified gene expression changes in lung ECs associated with vasodilation, specifically down-regulation of a potent vasoconstrictor (Edn1) and up-regulation of a potent vasodilator (Nos2). The stimuli for these changes are unknown (flow-mediated, inflammation-mediated, or other); however, early vasodilation could be an initial physiologic cause of intrapulmonary shunting and the PVM phenotype. We speculate that this early vasodilatory response is the initial manifestation of single ventricle PVM shunting and sustained vasodilation combined with pro-angiogenic and pro-inflammatory signaling ultimately leads to architectural remodeling of the lung vasculature if hepatic vein blood perfusion is not restored. The specific factor(s) in hepatic vein blood remain unknown, but we are poised to identify causative factor(s) with our current animal model.

Clinical implications

The findings of our study have multiple potential clinical implications. First, identification of a shared molecular phenotype between single ventricle PVMs and HHT AVMs may enable new strategic comparisons and therapeutic strategies. For example, medical therapies currently used to treat patients with HHT may be considered for investigation in single ventricle PVMs. Second, identification of decreased ATRA signaling as a shared feature of single ventricle PVMs and HHT AVMs suggests that ATRA signaling may be a new therapeutic strategy for both conditions. Medical therapy aimed at directly increasing lung retinoids (ie, inhaled Vit A)⁵⁹ or targeting downstream effects of ATRA signaling may be viable next steps for preclinical and clinical trials. Last, inflammation appears to be a previously unidentified component of single ventricle PVMs. Because therapeutic strategies are already in place to target lung inflammation in pediatric medicine (eg, inhaled corticosteroids), anti-inflammatories may be a potential therapeutic strategy in the immediate future.

Study limitations

First, despite validation with independent methods, gene expression changes identified in this study may be related to tissue remodeling post-Glenn but indirectly or even independently of PVM pathogenesis post-Glenn. For example, formation of aortopulmonary collaterals is a universal phenomenon in patients post-Glenn,^8,60 so we anticipate that aortopulmonary collaterals similarly develop in our post-Glenn rats and may cause gene expression changes independent of PVMs. Our model alters both flow (nonpulsatile) and blood source (excluding hepatic vein blood flow) after Glenn surgery. Thus, we cannot differentiate changes that are mediated by flow, biochemical mediators, or both. In contrast to patients with single ventricle circulation and systemic hypoxemia, our rats with Glenn and sham surgeries do not experience significant hypoxemia.¹⁶ It is possible that systemic hypoxemia may contribute to relevant gene expression differences that our rat model does not fully recapitulate. Next, we validated many scRNAseq findings with additional animals and experiments; however, we included a limited number of biological replicates (n = 4 per surgical group) and cells (n = 10,000 per sample) in the original scRNAseq experiment. It is likely that non-EC cell types are still affected by Glenn circulation and may have important pathophysiologic roles in PVM initiation and progression. Finally, despite gene expression similarities to HHT AVMs, it remains unclear how closely post-Glenn vascular remodeling represents conventional AVMs (ie, fast-flow direct artery-vein connections) vs other vascular malformations such as capillary malformations or venous malformations. Thus, until more definitive data are available, we will conservatively use the term “pulmonary vascular malformation (PVM)” in this article as opposed to the more specific pulmonary arteriovenous malformation (PAVM).

Conclusions

In summary, we report the novel application of scRNAseq to study single ventricle PVMs in a surgical rat model of Glenn circulation. We identified that ECs are the primarily affected cell type in the lungs early post-Glenn, and multiple biological processes are dysregulated in rat lung ECs post-Glenn, including decreased ATRA signaling and conserved gene expression patterns with HHT mouse models. Dietary modification of Vit A intake altered post-Glenn shunting and represents a novel potential therapeutic strategy for single ventricle PVMs and HHT AVMs.

Data Availability

All sequencing data will be made publicly available on the GEO database.

Funding Support and Author Disclosures

Dr Spearman has received National Institutes of Health (NIH), National Heart, Lung, and Blood Institute (NHLBI) grant K08HL157510, funding from the Medical College of Wisconsin Department of Pediatrics and Herma Heart Institute Innovation Funds, and joint CTSI Pilot – Mellowes Center for Genomic Sciences and Precision Medicine award UL1TR001436. Dr Shmarakov has received NIH NHLBI grant R01HL171112. Dr Ramchandran has received NIH NHLBI grants R33HL154254 and R01HL179583. Dr Meadows has received grant R01HL139713 and R01HL163196 from NIH NHLBI and Department of Defense Investigator-Initiated Research award PR160198. Dr Zhou has received American Heart Association Predoctoral Fellowship 20PRE35120237. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.

PERSPECTIVES.

COMPETENCY IN MEDICAL KNOWLEDGE: PVMs universally develop in patients with single ventricle congenital heart disease following Glenn palliation, yet the fundamental pathophysiological mechanisms underlying single ventricle PVMs are unknown. Using scRNAseq and a surgical animal model, we identified multiple dysregulated biological processes in lung ECs. We also identified dysregulated signaling pathways that are shared between single ventricle PVMs and HHT AVMs. These data identify previously unknown similarities between single ventricle PVMs and HHT AVMs, which may lend shared insight into both clinical conditions.

TRANSLATIONAL OUTLOOK: There are no current medical treatments for single ventricle PVMs and only limited effective treatments for HHT AVMs. In this study, we identified that endothelial ATRA signaling is a shared and novel therapeutic target for modifying AVM disease progression. Future research aimed at safely modifying endothelial ATRA signaling has potential to deliver a new therapeutic modality to patients.