Meta-analysis of blood genome-wide expression profiling studies in pulmonary arterial hypertension

Abstract

Inflammatory cell infiltrates are a prominent feature of aberrant vascular remodeling in pulmonary arterial hypertension (PAH), suggesting that immune effector cells contribute to disease progression. Genome-wide blood expression profiling studies have attempted to better define this inflammatory component of PAH pathobiology but have been hampered by small sample sizes, methodological differences, and very little gene-level reproducibility. The current meta-analysis (seven studies; 156 PAH patients/110 healthy controls) was performed to assess the comparability of data across studies and to possibly derive a generalizable transcriptomic signature. Idiopathic (IPAH) compared with disease-associated PAH (APAH) displayed highly similar expression profiles with no differentially expressed genes, even after substantially relaxing selection stringency. In contrast, using a false discovery rate of ≤1% and I² < 40% (low-to-moderate heterogeneity across studies) both IPAH and APAH differed markedly from healthy controls with the combined PAH cohort yielding 1,269 differentially expressed, unique gene transcripts. Bioinformatic analyses, including gene-set enrichment, which uses all available data independent of gene selection thresholds, identified interferon, mammalian target of rapamycin/p70S6K, stress kinase, and Toll-like receptor signaling as enriched mechanisms within the PAH gene signature. Enriched biological functions and diseases included tumorigenesis, autoimmunity, antiviral response, and cell death consistent with prevailing theories of PAH pathogenesis. Although otherwise indistinguishable, APAH (predominantly PAH due to systemic sclerosis) had a somewhat stronger interferon profile than IPAH. Meta-analysis defined a robust and generalizable transcriptomic signature in the blood of PAH patients that can help inform the identification of biomarkers and therapeutic targets.

INTRODUCTION

Pulmonary arterial hypertension (PAH) is a rare but fatal disease characterized by endothelial dysfunction, vascular cell proliferation, metabolic reprogramming, and perivascular inflammation of precapillary pulmonary arterioles (55). While PAH is classified as idiopathic (IPAH) in ~40% of cases, autoimmune disease is the most frequent predisposing condition among those patients with disease-associated PAH (APAH) (4). Notably, APAH can also develop in the setting of infection (e.g., HIV and schistosomiasis) as well as therapeutic interferon administration, supporting the role of inflammation as a critical modifier of PAH susceptibility (44, 47). In IPAH, perivascular inflammatory cell infiltrates associated with aberrantly remodeled vessels contain CD68⁺ macrophages, CD14⁺ macrophages/monocytes, dendritic cells, and T and B lymphocytes (43, 56). Importantly, the severity of perivascular inflammation correlates with the extent of vascular remodeling (49). Furthermore, IPAH patients have higher serum concentrations of circulating inflammatory markers compared with healthy controls (24, 48) and levels of IL-6, IL-8, IL-10, and IL-12p70 may be superior to traditional measures, such as 6-min walk distance, at predicting survival (48). Collectively, these findings suggest a sustained state of immune activation, even in the absence of underlying autoimmune disease or known infection that contributes to the development of PAH. Conversely, a recent immunoproteomics study described four distinct immune endophenotypes among PAH patients, including a subgroup with normal serum cytokine levels indicating that immune activation may not be the sine qua non of PAH (53).

Investigation of immune mechanisms that contribute to PAH pathobiology has been hampered by an inability to safely obtain lung tissue. This has limited most histopathologic, proteomic, and transcriptomic studies to specimens from end-stage disease at the time of lung transplantation. High-throughput functional genomic approaches have also been applied to whole blood or blood components as safe and easily obtained “liquid biopsies” in an attempt to understand PAH pathobiology at earlier stages of disease. Furthermore, accumulating recent evidence strongly implicates circulating bone marrow-derived cells in the development and progression of pulmonary vascular remodeling (1, 17, 19, 59), thus providing a renewed interest in examining the blood transcriptome in PAH patients. Unfortunately, prior blood expression profiling studies conducted over many years at different institutions have produced nonoverlapping results at the gene level. Variability in methodological approaches, small sample sizes, and patient heterogeneity may have negatively affected the reproducibility and generalizability of these results. Nonetheless, despite the lack of gene-level reliability, prior blood transcriptomic profiling studies exhibited some general thematic consistency and individual studies demonstrated the potential to distinguish PAH patients from healthy controls (8–10, 21, 35) and disease controls without PAH (35, 39). Notably, however, this approach has not detected differences in patients with IPAH compared with APAH (8), including the subset of APAH patients with systemic sclerosis (9). Whether this attests to the similarity of circulating immune cells across PAH etiologies or is due to limited statistical power is not entirely clear.

Meta-analysis of genome-wide expression studies can combine available data from independent studies to improve the statistical power and reliability of results (37). Here we performed a meta-analysis of seven genome-wide expression profiling studies of fresh peripheral blood mononuclear cells (PBMCs) or whole blood from PAH patients to assess the comparability of these seemingly disparate results. Representing the largest data set of circulating immune cell transcriptomic profiles from patients with PAH, the present study yielded sizable, stringently selected sets of differentially expressed genes across different centers and microarray platforms. A thematic analysis of these gene expression differences as well as an enrichment analysis of normalized data for over 22,000 transcripts supports the existence of an interferon-driven systemic immunologic response as a fundamental component of PAH pathobiology that was previously unrecognized in the individual blood expression profiling studies.

METHODS

Literature Search

Principal keywords included microarray, mononuclear, and pulmonary arterial hypertension. The searches were limited to English language articles but not limited by year of publication. The search strategies were adapted to accommodate the unique searching features of each database, including database-specific MESH and EMTREE vocabulary terms as appropriate (see Supplemental Methods; https://doi.org/10.6084/m9.figshare.8298023). The primary search strategy was completed on July 8, 2016. The search was updated on January 19, 2019 to capture newer studies that were published since the initial search. Likewise, a specific search strategy for RNA sequencing studies was completed on March 4, 2019. Of nine potential data sets identified, four were not available in either the Gene Expression Omnibus (GEO) or ArrayExpress. One of these studies was conducted at NIH (see Supplemental Methods; https://figshare.com/s/fbcc625a4e7bbcfbe65e), and the data are now available in GEO (GSE131793). Of the remaining, missing data sets (21, 41), one was obtained directly from the authors (21), providing a total of seven data sets available for analysis.

Data Processing and Statistical Analysis

Preprocessed or raw data were directly downloaded from the GEO repository or obtained from the author. Data were reprocessed when necessary via background check and normalization in Bioconductor (https://www.bioconductor.org). Details of data processing specific to each study are available in the Supplemental Methods; https://figshare.com/s/fbcc625a4e7bbcfbe65e. Probe/probeset level data were annotated to Entrez Gene IDs using the appropriate annotation files as specified in the Supplemental Methods. Entrez Gene IDs that mapped to more than one probe/probeset within each study were resolved by selecting the probe/probeset with the smallest P value from an ANOVA F test across the patient groups: 1) healthy controls; 2) IPAH; and/or 3) APAH (see Supplemental Fig. S1; https://doi.org/10.6084/m9.figshare.8297984). Effect sizes and variances were calculated for each study using the R MetaDE package (https://www.rdocumentation.org/packages/MetaDE/versions/1.0.5). For each Entrez Gene ID present in two or more studies, overall effect estimate, standard error, P value, test of heterogeneity, and I² were generated using random effect models with the inverse variance method using the R meta package (https://cran.r-project.org/web/packages/meta) (37). False discovery rate (FDR) and I² cutoffs of 0.01 and 0.4, respectively, were used to control the number of false positives and heterogeneity across studies.

Bioinformatic Analysis

Ingenuity Pathway Analysis.

Differentially expressed genes in the combined PAH cohort (N = 1,269 transcripts) were uploaded into Ingenuity Pathway Analysis (IPA). Significance of enrichment is calculated using the right-tailed Fisher’s exact test. Significantly enriched (FDR < 0.0001), nonredundant diseases/functions were identified using the Bio Function application. Diseases/functions were considered redundant when >80% of the annotated gene transcripts overlapped with a similar, more significant disease/function. Significantly enriched cellular pathways (uncorrected P < 0.01) were identified using the Canonical Pathway application. The Upstream Regulator function and the Ingenuity Knowledge Base were used to construct a regulatory network based on the top 100 upregulated transcripts in all PAH patients compared with healthy controls [FDR ≤ 1%, fold change (FC) ≥ 1.7, and I² < 0.4]. Molecule type was restricted to genes, RNAs, and proteins. Additional connections were manually curated using PubMed (https://www.ncbi.nlm.nih.gov/pubmed/), Molecular Signatures Database (version 6.2; http://software.broadinstitute.org/gsea/msigdb/index.jsp; Broad Institute, MIT) (28), and Interferome (version 2.01; http://www.interferome.org/interferome/home.jspx) (40).

Connectivity Map.

The top 150 up- and downregulated genes from the combined PAH cohort were analyzed using Connectivity Map (https://clue.io) to compare the meta-analysis-identified gene list to the effects of PAH therapies on gene expression. Of the currently available PAH medications, bosentan, sildenafil, tadalafil, iloprost, treprostinil, and selexipag (listed as BMY-45778) are in the Connectivity Map database. Compounds are rank-ordered by similarity of differentially expressed genes to the PAH gene signature. Connectivity Map scores range from −100 to +100. A score of +90 or higher (gene signature of the drug is similar to the PAH gene signature) or −90 or lower (gene signature of the drug is opposite to the PAH gene signature) is considered significant and worthy of further study.

Gene set enrichment analysis.

All transcripts included in the meta-analysis (PAH, IPAH, and APAH patients vs. healthy controls) were rank ordered based on the log₂ fold change (FC). Gene sets from the hallmark collection (n = 50), the Reactome pathway database (n = 674), both from the Molecular Signatures Database (version 6.2; Broad Institute, MIT), and a gene set created from the PAH knowledgebase (60) were all tested for significant enrichment (27). Gene set permutation was performed to assess the statistical significance of the enrichment score.

ClueGO.

ClueGO (version 2.5.4), a Cytoscape (version 3.7.1) plug-in, was used to create a functionally organized network of immune and inflammatory annotation terms (pathways/functions) (6). The ClueGO network is created with kappa statistics and reflects the relationships between the terms based on the similarity of their associated genes.

Annotation and functional analysis of the top 100 upregulated transcripts in all PAH patients compared with healthy controls [FDR ≤ 1%, fold change (FC) ≥ 1.7, and I² < 0.4] was performed using Gene Ontology (GOImmuneSystemProcesses) and Reactome pathways.

Promoter analysis.

The promoter regions of the top 100 upregulated gene transcripts (FDR ≤1%, FC ≥1.7, and I² < 0.4) were analyzed using the F-Match module in TRANSFAC (version 2018.3) (31). A set of genes was selected from the expression data set that demonstrated no evidence of differential expression in PAH patients compared with healthy controls (“no” set) to examine the promoter structure of those transcripts upregulated in PAH patients (“yes” set). Applying an fold change equal to 1.0 for PAH compared with healthy controls yielded a no set of 421 genes. The “immune cell-specific” profile of position weighted matrices was chosen to identify the most promising potential binding sites relevant to circulating cells and PBMCs. Using only the best-supported promoters within the yes and no sets (n = 99 and n = 402, respectively), the promoter window region was set at −500 to +100 base pairs relative to the transcription start site. The threshold for significance was set at P < 0.01. To test the validity and stability of the output, the F-Match analysis was repeated using a randomly generated no set of genes (BIOBASE Knowledge Library) and then again using a profile of nonredundant vertebrate position weighted matrices rather than an immune cell-specific profile, and results were similar.

RESULTS

PubMed, EMBASE, Scopus, and the Web of Science Core Collection were queried to capture genome-wide blood expression profiling studies performed on freshly isolated PBMCs or whole blood samples in PAH patients. Two-hundred sixty-four references, including full text articles and abstracts, were identified and seven studies (n = 6 PBMCs; n = 1 whole blood) fulfilled selection criteria for meta-analysis (Fig. 1A). Prior genome-wide expression profiling studies that used Epstein-Barr virus-immortalized lymphocytes (2, 18, 23, 57) were not included because these cells were considered distinct from unaltered, circulating cells processed or preserved soon after collection.

Fig. 1.

Meta-analysis of genome-wide transcriptomic profiles reveals no differences in gene expression between patients with idiopathic pulmonary arterial hypertension (IPAH) versus associated PAH (APAH). A: study selection flow diagram. PBMCs, peripheral blood mononuclear cells; GEO, Gene Expression Omnibus. B: etiology of PAH among patients included in the meta-analysis. A total of 156 PAH patients were analyzed. Most subjects were diagnosed with either idiopathic PAH (IPAH; n = 65, 42% of all PAH subjects) or systemic sclerosis-associated PAH (SSc-PAH; n = 81, 52%). Meta-analysis of expression profiles of IPAH and APAH patients across 6 studies (N = 22,339 transcripts) yielded no differentially expressed genes at a false discovery rate (FDR) of ≤ 5%. C: the distribution of raw P values for each transcript was approximately uniform (0,1) consistent with the null hypothesis of no difference. Mean P value and FDR were 0.50 and 0.97, respectively. The lowest FDR was 0.093 (n = 2 transcripts), and only 6 transcripts had an FDR < 0.50. Based on the lack of differential gene expression between IPAH and APAH patients, these groups were combined into a single PAH cohort and expression profiles were compared with healthy controls. D: in contrast to the uniform distribution of raw P values for gene differences between IPAH and APAH, raw P values for the combined PAH cohort compared with controls illustrate significant expression differences between these two groups.

PAH patients (n = 156) were further classified as either having IPAH (65/156, 42%) or APAH (91/156, 58%; Fig. 1B). Systemic sclerosis (SSc) was by far the most common underlying cause of APAH (81/91, 89%). All seven studies included healthy controls and APAH patients, and six studies included IPAH patients (Table 1). SSc patients without PAH (9, 35, 39), SSc patients with pulmonary hypertension due to interstitial lung disease (9), as well as patients with cystic fibrosis or pulmonary veno-occlusive disease (10) were excluded from the meta-analysis (see Supplemental Methods for further details; https://figshare.com/s/fbcc625a4e7bbcfbe65e). The study from Risbano et al. (39) was included in our meta-analysis because concurrently processed microarray data from healthy controls and IPAH patients was available in the GEO repository (GSE22356), even though the published manuscript only reported on SSc patients with and without PAH.

Table 1.

Expression profiling studies included in the meta-analysis

Study	Cell Type	Platform	GEO Accession	N	Healthy Control	IPAH	APAH	Gene Selection Criteriaa	Unique Genes Selected (n)b
Bull et al. (8)	PBMCs	Affymetrix Human Gene FL Array (Hu6800)	GSE703	20	6	7	7	P < 0.001c	113 (PAH)
Grigoryev et al. (21)	PBMCs	Affymetrix Human Genome U133A 2.0	N/A	24	5	9	10	FDR < 1% and FC > 2.45 d	99 (IPAH) 286 (APAH)
Pendergrass et al. (35)	PBMCs	Agilent Human Gene Expression 4x44K G4112F	GSE19617	25	10	N/A	15	FDR < 0.14%	256 (APAH)
Risbano et al. (39)	PBMCs	Affymetrix Human Genome U133 Plus 2.0	GSE22356	28	10	8	10	N/Ae	N/Ae
Cheadle et al. (9)	PBMCs	Illumina HumanHT-12 V3.0	GSE33463	113	41	30	42	P ≤ 0.01 and FDR ≤ 10% and FC > 1.5f	583 (IPAH) 625 (APAH)
Chesné et al. (10)	Whole Blood	Agilent Human Gene Expression 8X60K G4851A	GSE38267	36	28	6	2	P < 0.01 and gene clusteringg	1353 (PAH)
NIH (14)	PBMCs	Affymetrix Human Gene 1.0 ST Array	GSE131793	20	10	5	5	FDR ≤ 20% and FC ≥ 1.2h	217 (PAH)

PBMCs, peripheral blood mononuclear cells; IPAH, idiopathic pulmonary arterial hypertension; APAH, associated pulmonary arterial hypertension; FDR, false discovery rate; FC, fold change.

^aSelection criteria corresponds to the gene list used for results reported in Fig. 2 and Supplemental Fig. S3.

^bThe original probesets meeting selection criteria were reannotated using the most recent annotation file available at the time the data were processed for meta-analysis. If a single probeset annotated to more than one gene, each unique gene is included in the total number. See Supplemental Methods for the annotation files used for each study.

^cThe number of probe sets was prefiltered based on signal “absolute call” in at least 11 of 21 samples (6,086 probesets reduced to 2,906). The gene list was then generated from these 2,906 probsets using a two-sample t test.

^dProbesets were prefiltered in two steps: 1) probesets with detectable hybridization signals in <5% of samples were flagged as nonfunctional; and 2) the remaining probesets considered “Present” by Affymetrix GeneChip Operating Software 1.4 and had a signal of at least twofold higher than background in ≥75% of samples in a given group (IPAH, APAH, or healthy control) were analyzed for differential expression.

^eThe study by Risbano et al. (39) was not included because the published analysis compared systemic sclerosis patients with and without PAH, a contrast that was not examined in our meta-analysis.

^fProbes were prefiltered based on whether ≥80% of samples in the group with higher average expression level for a particular probe had Illumina detection P < 0.01. Differentially expressed transcripts had to satisfy 3 criteria: 1) two-sided Welch t test P ≤ 0.01; 2) Benjamini-Hochberg FDR ≤ 10%; and 3) FC > 1.5 (calculated using geometric means).

^gThe number of probes was prefiltered based whether the signal in ≥50% of samples was less than the mean of all median signals (58,717 probes reduced to 30,146). Gene selection was based on Student’s t-test and whether genes clustered together.

^hTotal gene-level transcript clusters (33,297) annotated to 28,869 transcripts. The set of 28,869 transcripts were subsequently filtered based on gene-level P < 0.05 among at least half of chips in either group (PAH or healthy volunteer) resulting in 20,543 transcripts. Differentially expressed transcripts were selected from the prefiltered set of 20,543 transcripts using the paired t test from the limma package in Bioconductor.

The initial comparisons of interest were IPAH versus healthy controls, APAH versus healthy controls, and IPAH versus APAH. Selection of differentially expressed genes based on a FDR of ≤ 1% and I² < 40% for IPAH and APAH, each compared with healthy subjects, yielded 563 and 1,072 transcripts, respectively (see Supplemental Table S1, https://doi.org/10.6084/m9.figshare.8298032, and Supplemental Table S2, https://doi.org/10.6084/m9.figshare.8298035 for annotated gene lists). In contrast, direct comparison of 22,339 transcripts in patients with IPAH and APAH across 6 studies yielded no differentially expressed genes even when much less stringent selection criteria were applied (Table 2). The uniform distribution of raw P values illustrates the striking lack of transcriptomic differences between IPAH and APAH patients (Fig. 1C). Therefore, IPAH and APAH patients were merged into a combined PAH cohort, which was then compared with healthy controls. Applying the same stringent selection criteria to this analysis yielded 1,269 differentially expressed genes (Table 2; see Supplemental Table S3 at https://doi.org/10.6084/m9.figshare.8298038 for annotated gene list). In contrast to the uniform distribution of raw P values for gene expression differences between IPAH and APAH, raw P values for the combined PAH cohort compared with controls illustrate significant expression differences between these two groups (Fig. 1D). Notably, when the published lists of differentially expressed genes from each of the individual studies were compared with each other, no gene was found in common across all seven studies and only one gene was shared by four studies (see Supplemental Fig. S2 at https://doi.org/10.6084/m9.figshare.8298014 for the number of shared transcripts between study pairs). Conversely, despite differences in patient mix, treatment regimens, and expression profiling methodology across centers and studies, standardized data normalization followed by meta-analysis yielded a substantial, stringently selected set of differentially expressed genes, demonstrating the reproducibility and comparability of PBMC and whole blood mRNA transcript profiles in patients with PAH. Thus this meta-analysis revealed a remarkable consistency in the underlying data that was not apparent when comparing differentially expressed genes reported across the studies.

Table 2.

Between group contrasts included in the meta-analysis

Comparison	No. of Studies*	No. of Genes	FDR ≤ 0.05	FDR ≤ 0.01	FDR ≤ 0.01, I2 < 0 .4
IPAH versus healthy control	6	22,339	1,236	579	563
APAH versus healthy control	7	22,753	2,330	1,186	1,072
APAH versus IPAH	6	22,339	0	0	0
PAH versus healthy control	7	22,753	2,681	1,478	1,269

IPAH, idiopathic pulmonary arterial hypertension; APAH, associated pulmonary arterial hypertension; FDR, false discovery rate.

^*GSE19617 did not include patients with IPAH.

Comparison of Individual Studies with Results from the Meta-Analysis

Published gene lists from each individual study were compared with the corresponding results from the meta-analysis. Differentially expressed genes reported by studies that analyzed all PAH patients together (8, 10, 14) are represented by a volcano plot of FDR versus FC as calculated from the meta-analysis comparing all PAH patients to controls (Fig. 2A). Similarly, studies that reported differentially expressed genes in IPAH (9, 21) (Fig. 2B) or APAH (9, 21, 35) (Fig. 2C) patients separately are represented by a volcano plot of the FDR versus FC as determined from the meta-analysis of IPAH or APAH patients compared with healthy controls, respectively. The study by Risbano et al. (39) was not included because the published analysis compared SSc patients with and without PAH, a contrast that was not examined in our meta-analysis. Overall reproducibility between the original studies and the meta-analysis was at best moderate. Using a very liberal uncorrected, nominal P < 0.05 by meta-analysis, 23–77% of genes identified in the original studies were similarly expressed in the meta-analysis (Fig. 2). However, only the APAH gene list reported by Cheadle et al. (9) had more than 20% of transcripts that met the more stringent criteria (FDR of ≤ 1% and I² < 40%) used to define differential expression in our meta-analysis. The number of subjects in each of the individual studies moderately correlated (R = 0.76, P = 0.03) with the extent of overlap with the stringently selected gene lists generated by meta-analysis. The meta-analysis results tended to overlap better with previously published APAH (Fig. 2C) than IPAH gene signatures (Fig. 2B).

Fig. 2.

Reproducibility of published gene lists compared with the results of the meta-analysis. Differentially expressed genes as defined by each individual study’s original analysis are plotted using false discovery rates (FDR) and fold changes determined by meta-analysis. A: three studies, Bull et al. (8) pulmonary arterial hypertension (PAH), Chesné et al. (10) PAH, and NIH (14) PAH, included a comparison of all PAH with healthy controls. B: Grigoryev et al. (21) and Cheadle et al. (9) included a comparison of idiopathic pulmonary arterial hypertension (IPAH) patients with healthy controls. C: Grigoryev et al. (21), Pendergrass et al. (35), and Cheadle et al. (9) included a comparison of associated PAH (APAH) patients with healthy controls. Gene transcripts from each individual study were categorized into three groups: 1) concordantly expressed genes with FDR ≤ 1% and I² < 40% (red circles); 2) concordantly expressed genes with an uncorrected P < 0.05 (blue circles); and 3) concordantly expressed genes with an uncorrected P ≥ 0.05 and/or any gene whose expression was discordant with the results of the meta-analysis (gray circles). Of note, the Chesné et al. (10) PAH gene list was selected by t test (nominal P < 0.01) as well as applying a gene clustering approach. The subsequent gene list was almost exclusively composed of overexpressed genes relative to healthy controls.

Differentially expressed genes in PAH patients from each study were combined into a single list of 2,530 unique genes (see Supplemental Table S4, https://doi.org/10.6084/m9.figshare.8298041), regardless of study size, the rigor of selection criteria, or lack of overlap with any other study. Overall, only 294 (12%) of these genes were confirmed as differentially regulated and directionally concordant with the meta-analysis results (Fig. 3A). Notably, 975 (77%) of the differentially expressed genes identified by meta-analysis were either previously unrecognized by (n = 966) or directly discordant with (n = 9) the seven individual studies. All nine discordant genes, eight from a single study using whole blood, were previously reported as upregulated (8, 10) but instead were downregulated based on meta-analysis. Importantly, more than 25% of the 100 most highly upregulated transcripts were among the genes uniquely identified by meta-analysis (Fig. 3A; see Supplemental Table S3 for annotated gene list).

Fig. 3.

Previously unrecognized differentially expressed genes and thematic analyses of the combined pulmonary arterial hypertension cohort compared with healthy controls. A: the published lists of differentially expressed genes in pulmonary arterial hypertension (PAH) patients from each study were combined into a single list (n = 2,530) and compared with the differentially expressed genes identified by meta-analysis (n = 1,269). The Venn diagram reveals that 77% (n = 975) of the differentially expressed genes identified by meta-analysis were either previously unrecognized by (n = 966) or directly discordant with (n = 9) the prior 7 individual studies. B: Ingenuity Pathway Analysis identified tumorigenesis, autoimmunity, response to viral infection, cell death and survival and T cell development among the most highly significant specific diseases and functions associated with the PAH gene signature [false discovery rate (FDR) < 0.0001 for all]. C: canonical signaling pathways significantly overrepresented among genes differentially expressed in PAH patients (uncorrected P < 0.01 for all). D: gene sets from the hallmark collection (Molecular Signatures Database, version 6.2) were tested for significant enrichment among a rank ordered list of all transcripts meta-analyzed in PAH patients vs. healthy controls. Rank order was based on the log₂ fold change. Running enrichment scores (y-axis) and the rank ordered position of gene set members (x-axis) are shown. E: gene sets from the Reactome pathway database were tested for significant enrichment among rank ordered lists of all gene transcripts meta-analyzed in PAH, idiopathic PAH (IPAH), and disease associated PAH (APAH) patients versus healthy controls. Rank order was based on the log₂ fold change. Significantly enriched gene sets (FDR < 0.05) related to inflammation and immune responses and positively correlated with gene expression in the PAH cohort were selected. Normalized enrichment scores (NES) for the selected gene sets were then compared across the IPAH and APAH gene expression analyses and displayed as a heatmap. Each row represents a selected gene set in the Reactome pathway database, and each column represents a specific gene expression analysis based on the results of our meta-analysis. All of the values for NES are ≥1 because the data sets selected were all positively correlated with PAH, IPAH, or APAH. Some of the annotations in B–E were modified for clarity.

Ingenuity Pathway Analysis Identifies Inflammation, Response to Infection, and Proliferation as Overrepresented Biological Processes and Pathways in the PAH Gene Signature

Next, we examined the meta-analysis-derived gene signature from the combined PAH versus healthy control analysis (N = 1,269). Of the 1,269 differentially expressed genes, 77% unique to the meta-analysis, 467 were upregulated and 802 were downregulated. IPA identified cancer, infectious diseases, immunological diseases, inflammatory response, and cell-mediated immune response as significant biological function and disease categories (FDR < 0.0001). The most highly significant specific diseases and functions included tumorigenesis, autoimmunity, response to viral infection, cell death and survival, and T-cell development (Fig. 3B). Top canonical pathways included eukaryotic initiation factor (specifically eIF2 and eIF4), interferon, p70S6K, and mammalian target of rapamycin (mTOR; uncorrected P < 10⁻⁶; Fig. 3C). Interferon signaling (activation z-score = 3.317), TREM1 signaling (z-score = 2.111), and the role of pattern recognition receptors (z-score = 2.111) were all predicted as activated while eIF2 (z-score = −3.838) was inhibited. Importantly, key components of mTOR/Raptor/S6K signaling involved in regulating eukaryotic translation (EIF3D, EIF3G, EIF3H, EIF3K, EIF4A2, and RPTOR), interferon responses (IDO1, IL15, IRF2, NMI, OASL, SP110, TAP1, and TRIM25), and pattern recognition receptor signaling (CASP1, DDX58, NOD2, PRKCD, TLR5, and TLR6) were among the concordant, differentially regulated genes uniquely identified by meta-analysis (966/1,269). Collectively, the meta-analysis-derived gene list (N = 1,269) highlighted a strong peripheral blood transcriptomic signature for interferon signaling and regulation of protein translation in PAH patients that was previously undescribed in the individual blood expression profiling studies.

In contrast, none of the IPA canonical pathways related to PAH therapies were significantly enriched in the combined PAH gene signature (N = 1,269) [endothelin-1 signaling (uncorrected P = 0.17), eicosanoid signaling (P = 0.49), cAMP signaling (P = 0.54), cellular effects of sildenafil (P = 0.54), NO signaling (P = 0.26), and endothelial nitric oxide synthase signaling (P = 0.21)]. Consistent with the results in IPA, the Connectivity Map scores for the available PAH medications were bosentan 0, sildenafil 24.72, tadalafil 0, iloprost −41.86, treprostinil 0, and selexipag (BMY-45778) 85. Therefore, none of these therapies demonstrated a strong similarity to the top 150 up- and downregulated genes from our meta-analysis. The score of 85 for selexipag is unexpected and not relevant since selexipag was not available to patients in these gene expression studies (Food and Drug Administration approved in 2015 and approved in Europe in 2016). Furthermore, in contrast to selexipag, the other prostacyclins analyzed demonstrated either no effect (treprostinil, score = 0) or an opposite effect (iloprost, score = −41.86) on the expression of the most highly up- and downregulated meta-analysis genes.

Gene Set Enrichment Analysis Reinforces the Central Role of Inflammatory Signaling in PAH

Next, genome-wide expression profiles of the combined PAH cohort and healthy controls were compared using gene set enrichment analysis (GSEA). GSEA is a computational method that determines whether an a priori defined set of genes demonstrates statistically significant, concordant differences between two biological phenotypes (32, 52). GSEA utilizes the complete gene expression data set without imparting gene selection filters (e.g., FDR thresholds or FC differences), in contrast to thematic analyses (e.g., IPA) that focus only on a subset of threshold-defined “differentially regulated” genes. Therefore, GSEA is complimentary to IPA and can be used to test the consistency of bioinformatic inferences. Top gene sets positively correlated with PAH were interferon-α [normalized enrichment score (NES) = 3.19, FDR < 0.0001] and interferon-γ (NES = 3.14, FDR < 0.0001) responses, heme metabolism (NES = 3.08, FDR < 0.0001), TNFα-induced NF-κB signaling (NES = 2.61, FDR < 0.0001), and IL6/JAK/STAT3 signaling (NES = 2.59, FDR < 0.0001) (Fig. 3D). Conversely, MYC target genes (NES = −2.15, FDR < 0.0001) were negatively correlated with PAH. Therefore, similar to results from IPA, activation of interferon signaling was a prominent feature of the PAH gene signature determined by meta-analysis that had gone unrecognized in individual blood expression profiling studies. Importantly, gene expression changes in the combined PAH cohort were also positively correlated (NES = 1.58, FDR < 0.0001) with an independently curated network of 341 human pulmonary hypertension-related genes (60).

Next, GSEA-generated signatures for PAH patients were compared directly to IPAH and APAH subpopulations to assess the relative magnitude of immunologic and inflammatory pathway activation among these groups. As before, this analysis included all of the transcripts in the meta-analysis without imparting arbitrary selection filters. Significantly enriched gene sets from the Reactome pathway database (FDR < 0.05) related to inflammation and immune responses and positively correlated with gene expression in the combined PAH cohort were selected for comparison. Not surprisingly, all of the selected inflammatory and immune-related Reactome pathways were either similarly or more significantly enriched in the APAH gene signature compared with the combined PAH or IPAH signatures (Fig. 3E). Although less pronounced, IFN-α, -β, and -γ signaling, the innate immune system, IL-1 signaling, Toll-like receptor (TLR) pathways including TLR4, antigen cross presentation, and p38 MAPK activation were all enriched in the IPAH gene signature (NES > 1.70 and FDR < 0.05).

Interferon Pathway Activation, Proinflammatory Cytokine, and Toll-Like Receptor Signaling Orchestrate the PAH Gene Signature in Circulating Cells

To explore the interrelatedness among inflammatory response pathways in PAH, a functionally organized network of immune related annotation terms (pathways/functions) (6) was generated from the top 100 upregulated transcripts in ClueGO (version 2.5.4). Using Gene Ontology (GOImmuneSystemProcesses) and Reactome pathways, a strong interferon gene signature was again found among the most highly upregulated transcripts in PAH patients (Fig. 4A). TLR signaling and an antiviral response were also identified, paralleling the thematic analysis in IPA. Furthermore, the Interferome database identified 80 of the top 100 transcripts and 62% of the entire combined PAH list (786/1269) as interferon regulated.

Fig. 4.

Regulatory network of the top 100 upregulated gene transcripts from the combined pulmonary arterial hypertension (PAH) data set. A: a functionally organized network of immune and inflammatory annotation terms (pathways/functions) of the top 100 upregulated transcripts in all PAH patients compared with healthy controls [false discovery rate (FDR) ≤ 1%, fold-change (FC) ≥ 1.7 and I² < 0.4] was generated in ClueGO (version 2.5.4) using the Gene Ontology (GO_ImmuneSystemProcesses) and Reactome Pathway databases. Node size corresponds to the enrichment significance of the terms and central nodes are depicted in colored text. B: the top 100 upregulated genes were uploaded into Ingenuity Pathway Analysis (IPA) and examined using the Upstream Regulator module to construct a central network of potential therapeutic targets. Upstream regulator “molecule type” was restricted to genes, RNA and protein. Regulatory nodes were chosen based on the following criteria: 1) IPA-predicted upstream regulators present among the top 100 upregulated PAH gene transcripts were selected (DDX58, IFIT1, IRF9, JAK2, MAP2K3, NCF4, PARP9, and TLR4); and 2) significant IPA-predicted upstream regulators (P < 0.05 for significance of enrichment) linked to PAH pathobiology and involved in biological functions or pathways consistently recognized across our bioinformatic analyses [interferon (IFN), Toll-like receptor (TLR), NF-κB, and cytokine signaling] were selected (IFN-α; IFN-β; IFN-γ; GM-CSF; IL-1β; IL-6; TNFα; TNFRSF1A; TGF-β1; IFNAR1 and 2; IFNGR1 and 2; TLR2, 3, 7, and 9; NF-κB; IRF1, 3, 5, and 7; FOXO1 and 3; and STAT1, 2, 4, and 6). These upstream regulators connected 72 of the top 100 transcripts based on the Ingenuity Knowledge Base. Connections for the 28 remaining transcripts were manually curated using PubMed, the Molecular Signatures Database (MSigDB; version 6.2) and the Interferome (version 2.01). Some of the annotations in A and B were modified for clarity.

Next, the promoter regions of the top 100 upregulated genes from the combined PAH cohort were compared with the promoter regions of a set of genes that were not differentially expressed among PAH patients and healthy controls to determine which transcription factors and regulatory response elements were associated with the most highly upregulated genes in PAH. Using only the best supported promoter sequences in the TRANSFAC database, the interferon regulatory factor (IRF) family of transcription factors were prominently represented among the top enriched promoter binding sites (matched promoter P < 0.001, Table 3; see Supplemental Table S5 at https://doi.org/10.6084/m9.figshare.8298044 for a full list of enriched promoter binding sites). Binding sites for several other transcription factors previously associated with PAH were also notably enriched within the promoter regions of the top upregulated PAH genes, including AP-1 (3, 30, 58), NF-κB (16, 36), NFAT (7, 34), FOXO1 (42), and C/EBPβ (3, 13, 58) (Table 3 and Supplemental Table S5). Transcription factor binding sites for IRF, AP-1, and NF-κB family members comprised > 60% of the enriched promoter regions identified among the top 100 upregulated genes in the PAH signature.

Table 3.

Transcription factor binding site enrichment among top upregulated PAH genes

Transcription Factor Family	Transcription Factor Name(s)	Matrix Name*	Yes/No Score†	Matched Promoters P Value
Interferon regulatory factor	IRF-1	V$IRF1_Q5	44.6667	2.84 × 10−6
		V$IRF1_03	40.6061	2.84 × 10−6
		V$IRF1_05	32.4848	6.62 × 10−5
		V$IRF1_08	32.4848	6.62 × 10−5
		V$IRF1_07	28.4242	3.09 × 10−4
	IRF-2	V$IRF2_06	28.4242	3.09 × 10−4
		V$IRF2_05	24.3636	3.09 × 10−4
		V$IRF2_03	16.2424	9.51 × 10−8
	IRF-4	V$IRF4_Q5	28.4242	6.62 × 10−5
		V$IRF4_05	22.3333	1.30 × 10−5
		V$IRF4_03	13.5354	4.36 × 10−5
	IRF-1, 2, 3, 4, 5, 7, 8, 9	V$IRF_Q4	32.4848	1.38 × 10−5
		V$IRF_Q6_01	16.2424	2.39 × 10−6
		V$IRF_Q6	14.2121	2.49 × 10−4
Signal transducer and activator of transcription	STAT1	V$STAT1_11	4.2743	2.66 × 10−4
	STAT2	V$STAT2_01	44.6667	2.84 × 10−6
		V$STAT2_02	32.4848	6.62 × 10−5
Activator protein 1	c-Fos	V$CFOS_Q6	8.1212	1.77 × 10−4
		V$FOS_12	6.497	9.68 × 10−4
	JunB	V$JUNB_01	6.7677	5.78 × 10−4
		V$JUNB_06	6.381	3.42 × 10−4
		V$JUNB_04	6.381	3.42 × 10−4
	c-Fos, c-Jun	V$AP1_C	4.3144	3.62 × 10−4
NF-κB	P50	V$P50_Q6	4.4298	3.62 × 10−4

^*Matrix name: transcription factor binding site matrix from TRANSFAC database.

^†Yes/No Score: ratio of transcription factor binding sites of a particular matrix in the set of upregulated genes in pulmonary arterial hypertension (PAH) patients (Yes-set) compared with the background set of genes (No-set).

IRF, interferon regulatory factor; STAT, signal transducer and activator of transcription.

Finally, a PBMC regulatory gene network was created using the Upstream Regulator module in IPA. Similar to the analysis of transcription factor binding sites, the Upstream Regulator module was applied here to better understand key mediators and signaling pathways (i.e., “upstream regulators”) associated with the top 100 upregulated transcripts from the combined PAH cohort. Among the top 100 transcripts, IPA identified 8 as significant upstream regulators (DDX58, IFIT1, IRF9, JAK2, MAP2K3, NCF4, PARP9, and TLR4). Additional core regulators significantly enriched in the IPA Upstream Regulator module were selected based on their consistency across several bioinformatic analyses and/or their known relevance to PAH pathobiology. IFN-α, -β, and -γ, the IFN-α/β receptor, transcription factors important for IFN signaling (IRF1, 3, 5 and 7; STAT1 and 2), TLR 3, 7, and 9, IL-1β, IL-6, TNFα, and granulocyte-macrophage colony-stimulating factor (GM-CSF) were among the mostly significantly enriched upstream regulators (P ≤ 10⁻⁵ for all). The IFN-γ receptor TNFRSF1A, TGF-β1, TLR2, NF-κB, FOXO1 and 3, and STAT4 and 6 were also identified as significantly enriched (P < 0.05). Together these upstream regulators from the Ingenuity Knowledge Base connected >70% of the top 100 transcripts (Fig. 4B). Therapeutic or disease-monitoring strategies based on the networks, pathways, and gene products identified here may be useful in the future management of PAH but require more study.

DISCUSSION

Combining available data from independent studies, a large PAH cohort was assembled of genome-wide blood transcriptomic profiles. Meta-analysis across these studies yielded a large, stringently defined set of differentially expressed genes despite differences in patient mix and methodology between these investigations. The current study confirmed previous findings that blood expression profiles in both IPAH and APAH differ from healthy controls with the combined PAH cohort yielding 1,269 differentially expressed, unique gene transcripts. In contrast, IPAH and APAH patients had highly similar blood expression profiles with no differentially expressed genes. Notably, the combined PAH cohort identified a large number of differentially expressed genes compared with healthy controls that went unrecognized by any of the seven individual studies. Importantly, inflammatory networks, in particular interferon responses, and cell signaling pathways including eIF2/eIF4, stress kinases, mTOR, and p70S6K identified by meta-analysis are consistent with several established theories of PAH pathogenesis but were not previously recognized in any of the individual blood expression profiling studies.

Global expression studies of PBMCs and whole blood have attempted to better define the contribution of inflammation to PAH pathobiology. Although pathways and cellular functions involved in the inflammatory response were enriched, and typically activated, among the differentially expressed genes reported in prior PBMC expression profiling studies (8, 9, 21, 35), reproducible results across studies have been hampered by small sample sizes and methodological differences. The identification of abnormalities in interferon signaling as a unifying theme in the PAH gene signatures determined by meta-analysis is unique to our study and was highly consistent across multiple bioinformatic approaches. While interferon-related signaling was particularly strong in the APAH and combined PAH cohorts, it was also evident in the IPAH gene signature, albeit less prominently. Surprisingly, despite clear links to PAH pathogenesis (20) and recent recognition that interferon therapy can trigger the development of PAH (45, 47), interferon signaling was not uncovered in any of the individual blood expression profiling studies included in our meta-analysis. In a follow-up study that included a reanalysis of their original PBMC expression data set included here (35), Christmann et al. (11) explored an IFN-regulated gene cluster common to SSc patients with and without PAH. Consistent with their results, >60% of differentially expressed IFN-regulated genes were significantly increased in our APAH gene list and >50% were significantly increased in our combined PAH signature.

Biallelic mutations of the eukaryotic translation initiation factor 2 alpha kinase 4 (EIF2AK4) gene leading to an absence of general control nonderepressible 2 (GCN2) protein expression is a cause of familial and some sporadic cases of pulmonary veno-occlusive disease (PVOD) (5, 15). Interestingly, reduced (but not absent) GCN2 protein expression has also recently been observed in lung tissue from PAH patients (33). GCN2 is a serine/threonine kinase that phosphorylates the α-subunit of eIF2, thereby inactivating the eIF2 complex and promoting preferential translation of stress response mRNA transcripts (12). Enrichment of genes involved in regulation of protein translation during cellular stress (e.g., nutrient depletion, oxidant and endoplasmic reticulum stress, and viral infection) including eIF2, mTOR, eIF4, and p70S6K signaling was uniquely uncovered by meta-analysis. Based on the directional changes of the combined PAH gene signature, eIF2 signaling and general protein translation are predicted to be downregulated, consistent with activation of an integrated stress response. Decreased GCN2 protein expression, as seen in PVOD and PAH patient lung samples, would be expected to reduce eIF2-α phosphorylation thereby preserving general protein translation. The observed transcriptional changes detected by meta-analysis may represent a compensatory response to dysregulated global protein translation due to an absence of GCN2. Evidence directly linking proper GCN2 function to TLR responses (29), as well as controlling autoimmunity (38), further highlights its relevance and that of eIF2 signaling to PAH pathobiology.

Increased serum levels of IL-1β, IL-6, TNFα, GM-CSF, IFN-α, and IFN-γ (24, 48, 53) as well as higher local expression of TNFα and GM-CSF in the pulmonary vasculature (25, 46) are evidence of an activated inflammatory response in patients with PAH. Notably, these inflammatory cytokines were all predicted regulators of the most highly expressed PAH genes identified here by meta-analysis. Likewise, several transcription factors implicated in PAH pathobiology and central to regulating inflammation, such as NF-κB (16, 36), AP-1 (3, 30, 58), and STAT (34, 51), were identified bioinformatically as key regulators of the top upregulated transcripts in the combined PAH cohort. Pattern recognition receptor signaling was also significantly enriched in the PAH gene signatures identified by meta-analysis. Notably, TLR pathway enrichment was previously seen in patients with IPAH, SSc-PAH, and SSc without PAH compared with controls in the largest of the six prior PBMC expression profiling studies (9). In particular, increased TLR4 expression in PAH patients had been highlighted in two other blood expression profiling studies (10, 35). Our meta-analysis found an even broader activation of genes involved in TLR signaling, and as supported by recent lung transcriptomic findings (50), this pathway appears important in both IPAH and APAH patients. Mirroring the blood transcriptomic signature described here, several immune-related and inflammatory pathways were recently found to be significantly enriched in explanted PAH lung tissue (50). Likewise, key upstream mediators of the PAH lung transcriptome overlap with those identified here including TNFα, IFN-γ, lipopolysaccharide (a TLR4 ligand), and TGF-β1.

Meta-analysis of IPAH and APAH genome-wide blood expression profiles suggest shared immunologic mechanisms and is consistent with their comparable histopathologies (49). The present study, combining data from previous investigations, improved the statistical power and reliability of the notion that PAH, regardless of differences in genetics and the absence or presence of contributing diseases, has a core pathobiology that might be targeted with broadly applicable therapeutics. Similar to our findings, Bull et al. (8) did not detect any statistically significant differences in gene expression between IPAH and APAH. Despite larger and more homogeneous cohorts, Cheadle et al. (9) also found no significant differences in PBMC gene expression between patients with IPAH and those with APAH due to SSc. In the three remaining published studies, a direct comparison was either not possible (35, 39), or was not specifically examined (10). In the largest transcriptomic study of lung tissue from PAH patients to date, IPAH and APAH were compared with control donors but not directly to each other (50). Since nearly 90% of our APAH group was composed of patients with SSc-PAH, the similarities in gene expression with IPAH are more confidently ascribed to this group of APAH patients. Whether PAH patients with other associated conditions (e.g., congenital heart disease, portal hypertension, drug, or toxin exposure) have blood expression profiles similar to IPAH patients remains less clear. It is also possible that similarities in gene expression between IPAH and APAH reflect similar exposure to PAH medications. However, using unbiased machine learning, a recent large scale immunoproteomics study reported distinct immunophenotypes that were independent of PAH clinical subtypes as well as the number (including treatment naïve) or type of PAH background medications (53). Furthermore, the consistent detection of a proinflammatory gene signature in our data set compared with the paucity of evidence suggesting enrichment for vasodilator pathways argues against a major influence from PAH treatments aimed at reducing pulmonary pressures. Nevertheless, a comparison of blood gene expression profiles in treatment naïve PAH patients compared with healthy controls, as well as changes in expression patterns after treatment initiation, would be the most rigorous way to determine how pulmonary vasodilator therapy impacts gene expression in circulating cells.

Previously published, differentially expressed genes confirmed by meta-analysis included TYMP (ECGF-1) (8), VEGFB (21), ALAS (9, 35), JAK2 (35), CCR1 (35), TLR4 (10, 35), and AHSP (9). ALAS and AHSP, identified by Cheadle et al. (9) as part of their erythroid differentiation signature, were among the most highly upregulated transcripts in the combined PAH cohort. AHSP, but not ALAS, was also highly upregulated in IPAH patients but neither met our selection criteria in the APAH cohort. JAK2, CCR1, and TLR4 are among the top 100 upregulated transcripts in the PAH and APAH signatures, and TLR4 is upregulated in all three gene signatures (IPAH, APAH, and PAH). In contrast, several genes of pathologic importance highlighted by the individual blood expression profiling studies (e.g., ADM, TNFRSF14, MMP9, IL1B, IL8, IL7R, MDK, and LGALS3) did not meet the stringent criteria used for selection in our meta-analysis, suggesting either biologic or methodologic variability in their expression. Recently, Hemnes et al. (23) reported that PCR gene expression differences in whole blood reliably predict whether or not PAH patients are responsive to vasodilators. Only two gene transcripts, B4GALT5 and EPS8, identified by meta-analysis overlapped with those differentially regulated in vasodilator “responders” compared with “nonresponders.” B4GALT5 was upregulated in vasodilator responders as well as in the APAH and PAH cohorts, while EPS8 was downregulated in vasodilator responders but upregulated in the APAH cohort. As most PAH patients are vasodilator nonresponders, these results are not directly comparable to our meta-analysis.

There are specific features and limitations of the current study that may impact the interpretation of our findings. Due to differences in microarray platforms and the relatively small number of transcripts in early studies, expression data were only available for 2,413 genes across all 7 APAH studies and 2,421genes across all six IPAH studies. However, expression data for 19,035 and 16,253 genes was available for 4 or more APAH and IPAH studies, respectively and 84.5% of the 1,269 unique genes identified as differentially expressed in PAH were supported by 5 or more studies (see Supplemental Fig. S3; https://doi.org/10.6084/m9.figshare.8298017). A moderate amount of heterogeneity (I² < 40%) across studies was permitted for gene selection. Nonetheless, bioinformatic approaches independent of gene selection criteria produced comparable results. Furthermore, imparting a more stringent threshold for heterogeneity (I² < 10%) retains nearly 80% of the combined PAH gene list and nearly identical IPA-generated thematic analyses. A subgroup analyses based on severity of disease or treatment effects could not be performed due to the inconsistent availability of patient level clinical data. Lastly, no studies were found that used RNA sequencing, which can identify known and unknown coding transcripts as well as noncoding transcripts including microRNAs. RNA sequencing is also more sensitive, has a larger dynamic range compared with microarray and allows for detection of splice variants as well as allele-specific expression (26). Despite these limitations, standardized data normalization followed by meta-analysis yielded large, stringently-selected sets of differentially expressed genes.

In conclusion, application of meta-analysis has defined a robust and generalizable PAH blood transcriptomic signature that can be used along with existing data as a roadmap to inform the development of biomarkers and new therapeutics. For example, the prominent and consistent activation of interferon signaling identified by meta-analysis supports the testing of JAK/STAT inhibitors in preclinical models of PAH. In addition to future clinical trials of immune modulators, notable pathways and molecular targets identified here that are already being investigated as part of ongoing or recently completed clinical trials include mTOR activation (ClinicalTrials.gov Identifier: NCT02587325), IL-6 (NCT02676947), and IL-1β (54). Lastly, our findings may inform the design of future large-scale, multi-institution collaborative efforts utilizing high-throughput RNA sequencing methods to further refine our understanding of the PAH blood transcriptome.

GRANTS

This work was supported by the National Institutes of Health Clinical Center intramural funding.

DISCLOSURES

The NIH Clinical Center PAH Program receives support from Aadi Bioscience for a research coordinator but the authors have no personal financial relationship with Aadi Bioscience or any other entity.

AUTHOR CONTRIBUTIONS

J.M.E., M.A.S., and R.L.D. conceived and designed research; G.G., B.H., and G.A.F. performed experiments; J.M.E., A.J.M., R.C., J.S., and R.L.D. analyzed data; J.M.E. and R.L.D. interpreted results of experiments; J.M.E., A.J.M., and M.L. prepared figures; J.M.E. and A.J.M. drafted manuscript; J.M.E., M.A.S., and R.L.D. edited and revised manuscript; J.M.E., A.J.M., M.L., G.G., B.H., G.A.F., J.S., M.A.S., and R.L.D. approved final version of manuscript.