iPSC-endothelial cell phenotypic drug screening and in silico analyses identify tyrphostin AG1296 for pulmonary arterial hypertension

Abstract

Pulmonary arterial hypertension (PAH) is a progressive disorder leading to occlusive vascular remodeling. Current PAH therapies improve quality of life, but do not reverse structural abnormalities in the pulmonary vasculature. Here, we used high-throughput drug screening combined with in silico analyses of existing transcriptomic datasets to identify a promising lead compound to reverse PAH. Induced pluripotent stem cell-derived endothelial cells (iPSC-EC) generated from six patients with PAH were exposed to 4,500 compounds and assayed for improved cell survival after serum withdrawal using a chemiluminescent caspase assay. Subsequent validation of caspase activity and improved angiogenesis combined with data analyses using the Gene Expression Omnibus (GEO) and Library of Integrated Network-Based Cellular Signatures (LINCS) databases revealed that the lead compound AG1296 was positively associated with an anti-PAH gene signature. AG1296 increased abundance of bone morphogenetic protein receptors, downstream signaling and gene expression, and suppressed PAH smooth muscle cell proliferation. AG1296 induced regression of pulmonary arterial (PA) neointimal lesions in lung organ culture and PA occlusive changes in the Sugen/hypoxia rat model, and reduced right ventricular systolic pressure. Moreover, AG1296 improved vascular function and BMPR2 signaling, and showed better correlation with the anti-PAH gene signature than other tyrosine kinase inhibitors (TKIs). Specifically, AG1296 upregulated small mothers against decapentaplegic (SMAD)1/5 co-activators, cAMP-response element binding protein (CREB)3 and CREB5: CREB3 induced inhibitor of DNA binding 1 (ID1) and downstream genes that improved vascular function. Thus, drug discovery for PAH can be accelerated by combining phenotypic screening with in silico analyses of publicly available datasets.

One Sentence Summary:

Drug screening for improved function and transcriptomics of iPSC-derived endothelial cells from patients with pulmonary hypertension reveals AG1296.

Introduction

Pulmonary arterial hypertension (PAH) is a potentially fatal disease that has no curative therapy. It is characterized by dysfunction in both endothelial (ECs) and smooth muscle cells (SMCs) (1). Compared with donor control pulmonary arterial (PA) ECs, PAH PAECs show increased susceptibility to apoptosis under serum withdrawal or during reoxygenation after hypoxia, that could contribute to the loss of distal microvessels (2). Additionally, PAH ECs exhibit reduced adhesion to extracellular matrices (ECM), decreased migration, and impaired angiogenesis as judged by tubular network formation when placed in Matrigel (3, 4). PAH SMCs display heightened proliferation under serum or platelet-derived growth factor-BB (PDGF-BB) stimulation compared with healthy control SMCs (5). These cellular abnormalities, in addition to others that promote inflammation and fibroblast proliferation (6, 7), lead to the obliteration of the pulmonary microcirculation, causing a progressive increase in pulmonary vascular resistance (PVR) that culminates in right heart failure.

Human induced pluripotent stem cells (iPSCs) can be differentiated into a multiple vascular cell types including ECs (iPSC-EC) and SMCs (iPSC-SMC) (8, 9). These iPSC derivatives provide an unlimited source of patient-specific cell types for disease modeling and drug screening for both efficacy and toxicity (10–12). Our previous studies compared native PAECs removed from lungs during transplant with skin fibroblast-derived iPSC-ECs from the same patients (13). We found that iPSC-ECs expressed high expression of the arterial EC genes such as EFNB2 and ALK1, but low expression of the venous markers EPHB4 and NR2F2 when compared to human umbilical vein endothelial cells (HUVECs). This we attributed to the high concentration of vascular endothelial growth factor A (VEGFA) added in our iPSC-EC differentiation system (14). When we compared iPSC-EC and PAEC from the same set of patients with PAH vs. healthy controls we found that PAH iPSC-ECs recapitulated native PAH PAECs in impaired angiogenesis, cell migration, cell survival under stress, and BMP signaling as well as the angiogenic response to two emerging PAH therapies FK506 and elafin. Some differences were also observed in that PAH iPSC-ECs did not recapitulate the increased DNA damage observed in PAH PAECs when compared to the healthy control ECs, and RNA-seq analysis revealed differentially expressed genes in iPSC-ECs vs. PAECs related to collagen binding and non-integrin membrane-ECM interactions. In addition to the overall favorable profile of PAH iPSC-ECs in modeling disease, these cells offer the advantage of continuous availability for high-throughput screening of large libraries of compounds, as well as selective testing of emerging therapies in a patient-specific manner.

Current PAH therapies include vasodilators targeting the prostacyclin, endothelin, and nitric oxide-cGMP pathways (15, 16). These agents improve survival and quality of life, but they do not reverse cellular dysfunction; thus, their efficacy is frequently limited in preventing the progressive elevation in pulmonary vascular resistance that culminates in heart failure. The pro-proliferative and anti-apoptotic phenotype of PAH SMCs led to proposing inhibition of growth factors as a therapy. Tyrosine kinase inhibitors (TKIs) targeting a variety of growth factors showed promising results in reducing SMC proliferation and pulmonary hypertension (PH) in animal models (17, 18). However, a multi-center randomized trial of imatinib in patients with PAH found that despite an improvement in 6-minute walk distance and cardiac output, there was only a modest reduction of mean pulmonary arterial pressure, and no effect on clinical worsening of PAH. Moreover, there was a concerning and unexplained occurrence of subdural hematomas, resulting in an unfavorable risk-benefit ratio of imatinib for PAH (19–21). Another TKI, dasatinib, used in the treatment of refractory chronic myeloid leukemia (CML) by suppressing SRC and BCR-ABL, induced PAH. Guignabert et al. (22) demonstrated that an off-target effect of dasatinib, independent of inhibition of SRC family kinases, resulted in increased production of reactive oxygen species (ROS) and pulmonary EC apoptosis. Other TKIs that induce PAH include ponatinib (23), bosutinib (24, 25), and lapatinib (26), but, for the most part, PAH reverses after discontinuation of these TKIs.

Thus, there is an unmet medical need to develop a new therapy for PAH and to identify agents that might cause PAH. Due to considerable species differences, ninety percent of drugs that perform well in animal models fail to show efficacy in clinical trials. A recent drug screen identified FK506 as the top agent that activated the bone morphogenetic protein receptor (BMPR)2 signaling pathway. BMPR2 has homeostatic function in both PAECs and SMCs and mutations cause haploinsufficiency of BMPR2 in most patients with familial PAH. Moreover, BMPR2 function is impaired in most patients with idiopathic (I) PAH, and in patients for whom PAH is associated with a primary disorder (27). FK506 was compassionately repurposed for PAH, and a phase II dose-finding study has been carried out (28). However, the screen that identified FK506 as an activator of BMPR2 signaling was performed in a mouse myoblast cell line, and improved PAEC function with FK506 has been variable in PAH patient cell lines (4).

Recent advances in iPSC technology provide opportunities for biomedical and pharmaceutical research by permitting the use of differentiated tissue-specific human cells with the same genetic background as the patients while obviating the problem of accessibility of primary cells (29, 30). Publicly available databases that have been used to develop a PAH signature can also be used for drug discovery in silico (31), particularly as there is extensive drug transcriptomic information to find an agent with a PAH anti-signature. In this study, we screened a library of 4,500 compounds at different stages of clinical development in eight different doses using six PAH iPSC-EC lines, investigating improved function as assessed by survival after serum withdrawal. Of six compounds that improved cell survival, we identified the tyrosine kinase inhibitor (TKI) tyrphostin-AG1296, 6,7-Dimethoxy-2-phenylquinoxaline (C₁₆H₁₄N₂O₂), as the lead compound for further investigation based on the correlation of its drug signature with an anti-PAH gene signature. AG1296 is an ATP-competitive selective inhibitor of platelet derived growth factor receptors (PDGFR) with a molecular weight of 266.29 g/mol and a half maximal inhibitory concentration (IC₅₀) of 0.3-0.5 μM. AG1296 also inhibits fibroblast growth factor receptor (FGFR) and c-Kit with an IC₅₀ of 12.3 μM and 1.8 μM respectively (32). AG1296 activated BMPR2, improved signaling and gene expression in iPSC-ECs, and prevented hyperproliferation in six PAH SMC lines tested. Regression of PA neointimal lesions in lung organ culture and reversal of pulmonary hypertension (PH) and vascular remodeling in the Sugen/hypoxia rat model was observed with AG1296 treatment. AG1296 outperformed other TKIs, including imatinib, in improving PAH vascular function by increasing the amount of the transcription factors and SMAD 1/5 co-activators, CREB3 and CREB5. Our studies show the value of combining iPSC-derived patient-specific vascular cell functional assays with in silico analyses to identify promising PAH drugs.

Results

PAH iPSC-ECs recapitulate drug response in native PAECs

Our previous studies suggested that iPSC-ECs and PAECs from patients with PAH showed similar responses to potential PAH therapies as judged by an angiogenesis assay, and similar vulnerability to apoptosis after serum withdrawal as evident in the caspase3/7 glo assay (4). The caspase assay measuring cell apoptosis is helpful in explaining loss of distal arteries and is amenable to high-throughput drug testing, therefore we compared the apoptotic response in PAECs and iPSC-ECs from six patients after exposure to compounds in clinical development for treatment of PAH, e.g., BMP9 (33), Elafin (34), and FK506 (27). We noted a patient-specific degree of response to these therapies that was evident in both PAH PAECs and respective iPSC-ECs (fig. S1, A and B). This suggested that caspase activity in iPSC-ECs in response to serum withdrawal could be developed as a high-throughput assay to find agents that might improve this function in all PAH iPSC-EC lines and agents that could be patient-specific.

Functional screen in PAH iPSC-ECs identifies anti-apoptotic compounds

The high-throughput library used (NIH Clinic Collection, LOPAC 1280, Biomol, and MicroSource spectrum) consisted of 4,500 biologically annotated preclinical, clinical, and tool compounds. For the initial screen each compound was tested in duplicate in three different doses (20 μM, 10 μM, and 5 μM), in four PAH iPSC-EC lines [two idiopathic (I)PAH and two hereditary (H)PAH with a BMPR2 mutation], under serum withdrawal for 24 hours (Fig. 1). Demographic patient information related to the cell lines is provided in table S1. EC survival was measured by a caspase-3/7 activity assay which provides a reproducible luminescent readout as described in Fig. 1. Eighty compounds that improved EC survival by ≥50% compared with vehicle control were further screened in eight doses ranging from 0.156 μM to 20 μM. To eliminate the “false positive” compounds, where caspase activity was low owing to high cell death, a companion CellTiter-Glo assay was carried out to measure ATP production as a surrogate of cell viability. The remaining thirty-nine candidate drugs were then tested in two additional PAH iPSC-EC lines. Twenty out of the thirty-nine compounds improved EC survival in all six cell lines, whereas nineteen were only effective in a subgroup of PAH iPSC-EC lines (table S2). To eliminate compounds that interfered with the luminescent signal, i.e., to validate the caspase assay, the compounds in a 10 μM dose were re-tested for caspase activity using a CellEvent Caspase-3/7 green fluorescence dye (Fig. 2A, table S2). Eleven of 20 compounds with validated caspase activity in all six cell lines were further evaluated in a 10 μM dose in an angiogenesis assay, i.e., tube formation on growth factor reduced Matrigel (table S2). Six compounds improved tube formation in all six cell lines (Fig. 2B, table S2).

Fig. 1. Schema showing combined phenotype drug screen and bioinformatics to identify and validate AG1296 for pulmonary arterial hypertension (PAH).

(A) Six PAH induced pluripotent stem cell (iPSC)-endothelial cell (EC) lines were generated for the phenotypic drug screen. 4,500 bioactive compounds were tested for their ability to improve cell survival by more than 50% following serum withdrawal, using a luminescence assay measuring caspase 3/7 activity, an indicator of apoptosis. Compounds that reduced caspase activity but showed EC-specific toxicity were excluded. Twenty drugs were further selected based on (i) caspase validation using CellEvent Caspase-3/7 Green fluorescence dye, and improved tube formation in an angiogenesis assay; and (ii) Lead Compound selection based also on bioinformatic analysis of the correlation of drug signature based on the Library of Integrated Network-Based Cellular Signatures (LINCS) database, and PAH anti-signature generated by comparison of gene expression profiles of PAH vs. healthy control lung tissue. (B) Molecular structure of the lead compound, AG1296, and schematics showing validation of potential efficacy. AG1296 improved bone morphogenic protein (BMP) signaling and functions in PAH iPSC-ECs, suppressed SMC proliferation, and reversed vascular remodeling in PAH lung organ culture and animal model. AG1296 was superior to other tyrosine kinase inhibitors (TKIs) through a cAMP response element-binding protein (CREB)-dependent mechanism.

Fig. 2. Functional and bioinformatic analysis identifies AG1296 for further study.

(A) PAH iPSC-ECs were incubated with vehicle control (DMSO), full media (positive control), or the indicated compounds (10 μM) under serum withdrawal (0.2% FBS) for 48 hours. Apoptosis was determined by the CellEvent Caspase3/7 fluorescence assay. (B) Representative images, with quantification below, of tube formation of PAH iPSC-ECs treated with 10 μM of indicated compounds in EC media with 0.2% FBS. Images show tubes formed 6 hours after seeding cells on growth factor reduced Matrigel. Scale bar = 100 μm. (C) Heatmap showing the PAH signature and anti-signature compared with the gene expression profiles of top candidate compounds that were available from the LINCS database. PAH lung signature was generated by multi-cohort analysis based on public datasets from Gene Expression Omnibus (GEO) by comparing 32 patients with PAH versus 28 healthy controls. PAH signature of 330 genes (FDR <1%, Effect size >0.6), of which 61 were up-regulated, and 269 were down-regulated vs. Control. The PAH anti-signature is the reversed gene expression pattern. (D) Pearson’s correlation coefficient between each compound’s gene signature from LINCS and the PAH anti-signature. Bars represent mean±SEM. n=6, *P < 0.05 vs. DMSO, by one-way ANOVA with Bonferroni multiple comparisons test (A and B).

In parallel we used an in silico bioinformatic approach, beginning with the 20 compounds that passed the initial caspase activity screen. A PAH gene expression signature was generated from applying multi-cohort analysis (35) to publicly available transcriptomic datasets from the GEO database (36). Datasets GSE69416, GSE15197, and GSE48149 comparing normal control (n=28) versus PAH lung tissue (n=32) were analyzed with the R package Metaintegrator. This analysis resulted in a signature comprising 330 genes that were differentially expressed in patients with PAH compared to normal controls. Based on the PAH disease signature, we computed a complementary anti-PAH signature by reversing the fold change of each gene. We then ranked the compounds based on the correlation with the anti-PAH signature by using the gene expression profiles of the compounds obtained from the Library of Integrated Network-Based Cellular Signatures (LINCS) database (Fig. 2, C and D). Within the nine compounds that have a LINCS profile, AG1296 and resveratrol were the only two compounds that correlated with an anti-PAH signature, indicating that these two compounds are more likely to be beneficial in reversing PAH gene expression abnormalities. Although resveratrol has a better anti-PAH signature than AG1296, it did not increase tube formation in iPSC-ECs (fig. S2). In addition, previous studies have reported that resveratrol can promote a rapid increase in reactive oxygen species (ROS) resulting in pro-oxidant activity (37, 38) and apoptosis of endothelial cells in a mitochondrial-dependent manner (39). AG1296 also induced BMPR2 signaling in a previous screen in myoblasts (27).

AG1296 improves endothelial function in PAH ECs

AG1296 is a tyrosine kinase inhibitor (TKI) targeting PDGF, c-Kit, and FGF receptor signaling. Based on its function as a potential BMP activator and a TKI, we hypothesized that AG1296 may have a beneficial effect in promoting EC regeneration (40) and in suppressing SMC proliferation (5). We verified the effect of AG1296 in the same six iPSC-EC lines from both patients with IPAH and HPAH over a broader dose range. AG1296 significantly (P < 0.05) improved cell survival under serum withdrawal in a dose-dependent manner (Fig. 3A), while cell viability determined by ATP amount remained unchanged (Fig. 3B). AG1296 (5 μM and 10 μM) also improved angiogenesis, judged by the number of capillary-like tubes formed when seeded onto growth factor reduced Matrigel (Fig. 3, C and D). However, AG1296 reduced cell migration in PAH iPSC-ECs measured by a wound closure scratch assay (fig. S3, A and B), and did not significantly alter cell adhesion to laminin (fig. S3C). This suggests that improved cell survival accounted for the beneficial effect of AG1296 on angiogenesis.

Fig. 3. AG1296 improves EC survival and tube formation.

(A) PAH iPSC-ECs were incubated with vehicle control (DMSO) or AG1296 at six different doses under serum withdrawal (0.2% FBS) for 24 hours. Apoptosis was determined by Caspase3/7 fluorescence assay. *P < 0.05 vs. 0 μM of AG1296, one-way ANOVA with Bonferroni multiple comparisons test. (B) Cell viability was measured under the same condition as (A) using CellTiter-Glow. (C-D) Representative images of tube formation of iPSC-ECs treated with AG1296 (AG) or vehicle (DMSO, Veh) with quantitative analysis, indicating the number of tubes formed 6 hours after seeding cells on growth factor reduced Matrigel. Scale bar = 100 μm. Bars represent mean±SEM. n=6, *P < 0.05, **P < 0.01 vs. Veh, one-way ANOVA with Bonferroni multiple comparisons test.

AG1296 activates BMP signaling and survival pathways

To further understand the genes and signaling pathway underlying the protective effect of AG1296, we examined the amount of expression of BMP receptors. Reduced quantity of BMPR1A and BMPR2 are observed in the lungs of patients with HPAH and IPAH (41, 42), and a missense mutation in BMPR1B is associated with the pathogenesis of IPAH (43). We found that AG1296 significantly (P < 0.05) increased BMPR1A, BMPR1B, and BMPR2 under conditions of serum withdrawal (Fig. 4A). This likely accounted for the activation of downstream BMP signaling pathways pAKT, pERK, and pSMAD1/5-ID1 under both serum withdrawal (Fig. 4, B and C), and hypoxic conditions (fig. S4). Additionally, AG1296 increased expression of previously reported BMPR2-related pro-survival and angiogenesis genes such as apelin (APLN) (44), Baculoviral IAP repeat containing 3 (BIRC3) (4), vascular endothelial growth factor A (VEGFA) (45), and follistatin (FST) (46) under serum withdrawal conditions (Fig. 4D). Unbiased RNA-seq analyses further revealed that AG1296 controls similar gene expression pathways in PAH iPSC-ECs and corresponding PAECs: 80 genes were up-regulated and 115 down-regulated upon AG1296 treatment in both cell types (table S3, Fig. 4, E and F). Gene ontology analysis based on the upregulated DEGs revealed enrichment in pathways regulating angiogenesis, response to hypoxia, response to oxidative stress, and positive regulation of p38, consistent with the EC functional assays carried out (Fig. 4G). Downregulated genes upon AG1296 treatment were found in pathways related to positive regulation of smooth muscle cell proliferation, monocyte activation, immune response, and apoptosis.

Fig. 4. AG1296 enhances BMPR gene expression, BMP, and VEGF signaling.

(A) PAH iPSC-ECs were treated with AG1296 (AG) or vehicle (DMSO, Veh) overnight under serum withdrawal, and gene expression of BMP receptors were quantified by real-time PCR. (B) Representative western immunoblots and quantification of pAKT and pERK activation after treatment of AG1296 under serum withdrawal condition overnight. (C) Representative western immunoblots and quantification of pSMAD1/5 and ID1 activation after treatment of AG1296 under serum withdrawal condition overnight. (D) PAH iPSC-ECs were treated with AG1296 overnight under serum withdrawal, and gene expression of APLN, BIRC3, VEGFA, and FST were quantified by real-time PCR. n=6. Bars represent mean±SEM. *P < 0.05, **P < 0.01, ***P < 0.001 vs. Veh by unpaired t-test (A and D) or one-way ANOVA with Bonferroni multiple comparisons test (B and C). (E) Heatmap displaying 195 differentially expressed genes (DEGs) commonly up- or down-regulated by AG1296 treatment on both iPSC-ECs and PAECs. FDR<0.1 and fold change>=1.5 for PAECs; FDR<0.1 and fold change >=1.2 for iPSC-ECs. Of these, 80 genes were up-regulated and 115 genes were down-regulated. Gene expression was normalized using DESeq2 variance stabilizing transformation and then calculated as the ratio between the treated and untreated samples of the same cell type from the same donor. (F) Dotplot of fold change values of the 195 common differentially expressed genes (DEGs) in iPSC-ECs (x-axis) and PAECs (y-axis). Fold change values were log2 transformed. Pink: DEGs (n=80) commonly up-regulated by AG1296 treatment in both iPSC-ECs and PAECs. Blue: DEGs (n=115) commonly downregulated. (G) Functional enrichment analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID, version 6.8). Pink bars: gene ontology biological processes (GO BPs) enriched by the up-regulated DEGs; blue bars: GO BPs enriched by the down-regulated DEGs.

Genes upregulated by AG1296 include ANGPT2, that enhances EC survival through the PI3K/Akt pathway (47) and inhibits vascular leak induced by inflammatory stimuli, and GADD45B, a DNA repair protein decreased in PAH PAECs compared with controls (3) and responsible for EC viability through NF-kappaB and JNK signaling (48). Genes downregulated by AG1296 in iPSC-EC and PAEC include YAP1 that mediates ECM stiffening in the monocrotaline rat model of PH (49), IL27RA that controls EC activation and myeloid cell recruitment in atherosclerosis (50), and CX3CL1 that promotes vascular inflammation and monocytic cell adhesion (51).

AG1296 does not function through FGFR, PDGFR, or cKIT in ECs

To understand the molecular pathway directly targeted by AG1296, we assessed its role as an inhibitor of its known target tyrosine kinases, FGFR, PDGFR, and cKIT, in regulating the pAKT cell survival pathway. We reduced these kinases using corresponding siRNAs and inhibitors, and found that despite the inhibition of FGFR, PDGFRA, PDGFRB, and cKIT, individually or together, AG1296 still activated the pAKT pathway (fig. S5). These results indicate that AG1296 might not function by inhibiting its target tyrosine kinase receptors, but through a non-canonical effect on BMPR2 signaling.

AG1296 inhibits SMC proliferation

Because pAKT is a cell survival pathway for SMCs, we next investigated the impact of AG1296 on SMC proliferation. The anti-apoptotic and hyper-proliferative phenotype of PA SMCs is well described in PAH and linked to the progressive formation of a neointima that occludes the lumen of the PA vasculature (52–54). Thus, we assessed the response to AG1296 in PA SMCs from six patients with PAH. AG1296 significantly (P < 0.01) suppressed SMC proliferation at concentrations of 10 μM and 20 μM (fig. S6A). In an EC-SMC contact co-culture system (55), we found that AG1296 at a lower concentration (2.5 μM) significantly (P < 0.01) decreased SMC proliferation (fig. S6B), indicating that the beneficial effect of AG1296 on ECs may also contribute to the secretion of factors, such as APLN, that suppress SMC growth. In contrast to the effect on iPSC-ECs, AG1296 reduced the activation of pAKT and pERK in the PAH SMCs (fig. S6C), accounting for its anti-proliferative effect in SMCs. Unbiased RNA-seq analysis of PAH PASMCs revealed 479 differentially expressed genes upon treatment with AG1296 (fig. S6D, table S4). In particular, we found that genes negatively regulating cell proliferation such as GADD45A and KLF15 were up-regulated (56, 57), and genes promoting SMC growth such as WNT7B and ANO1 (58, 59) were down-regulated (fig. S6E), indicating that AG1296 has an intrinsic inhibitory role in SMC proliferation. Gene ontology (GO) analysis revealed that pathways regulating intrinsic apoptotic signaling, small GTPase mediated signal transduction, and response to Interleukin-1 were upregulated, while processes regulating cell division and canonical Wnt signaling were downregulated in the AG1296 treatment group (fig. S6F).

AG1296 induces regression of neointima in PAH lung organ culture

We next examined the potential impact of AG1296 in inducing the regression of PA neointimal lesions in human lung tissue as in our previous studies with Elafin, an inhibitor of elastase and inflammation and an activator of BMPR2 signaling (34). We utilized lung explants from three patients with PAH removed at the time of lung transplantation. Daily administration of AG1296 (20μM) for eight days resulted in a reduction of neointima formation compared with vehicle controls, judged by the lumen-to-vessel diameter (Fig. 5, A, B, and C). The optimal effect of AG1296 at 20 μM on neointimal regression could be explained by the induction of BMPR2 co-receptors as well as survival- and angiogenesis-related genes APLN, BIRC3, and FST in human lung tissue (Fig. 5, D and E).

Fig. 5. AG1296 reverses vascular remodeling in lung explants of patients with PAH.

Cultured lung explants taken from four patients with PAH were treated with either vehicle (DMSO) or AG1296 for 8 days. Medium was changed daily. (A) Left: schematic diagram of PAH lung organ culture. Right: representative images of Movat-stained sections of PA from patients with PAH treated with vehicle or AG1296 at 20 μM, scale bar = 20 μm. (B) Ratios of lumen to vessel area and (C) lumen to vessel diameter, based on analysis of n=8 lung organ culture experiments. (D) Gene expression of BMP receptors, downstream ID1, and SMAD co-regulators was quantified by real-time PCR. (E) Gene expression of EC survival- and angiogenesis-related genes was quantified by real-time PCR. n=4. Bars represent mean±SEM. *P < 0.05, **P < 0.01 vs. Veh by unpaired t test (C) or one-way ANOVA with Bonferroni multiple comparisons test (D and E).

AG1296 reverses sugen/hypoxia induced pulmonary hypertension in rats

We then tested the efficacy of AG1296 in an experimental rat model of severe pulmonary hypertension caused by inhibition of VEGFR2 by a single subcutaneous injection of SU5416 (Sugen, 20 mg/kg), followed by exposure to chronic hypoxia for three weeks, and recovery in normoxia for three weeks (34). The VEGFR blockade causes initial EC apoptosis, followed by hyperproliferation of SMC like cells, leading to the severe occlusion of distal arteries. Eight-week-old male Sprague Dawley rats (180–220 g) were given a single subcutaneous injection of the VEGF receptor blocker SU5416, followed by hypoxia (10% O₂) for three weeks, and normoxia for an additional three weeks (Su/Hx). Male animals were used as there is high attrition of females subjected to this treatment protocol. The rats were divided at random into two groups: one treated with dimethyl sulfoxide (DMSO) as vehicle control, and the other with AG1296 (50 mg/kg/d) by subcutaneous injection every day for three weeks starting after three weeks of normoxia. Untreated age-matched control animals were kept in room air throughout the study period. Fig. 6A shows a schematic overview of the experimental design.

Fig. 6. AG1296 reverses obstructed distal PAs and pulmonary hypertension (PH) in sugen/hypoxia (Su/Hx) rats.

(A) Schematic of protocol to evaluate the therapeutic efficacy of AG1296 (AG) in a rat model of PH. Sprague Dawley rats were exposed to room air (normoxia, n=4) or to the Su/Hx protocol (n=12). The Su/Hx rats were then randomly divided into two groups: DMSO vehicle (Veh), or AG treatment (n=6). (B) Right ventricular systolic pressure (RVSP) by catheter study. (C) Pulmonary Artery Acceleration Time (PAAT)/Ejection Time (ET) by echocardiography. (D) Right ventricular (RV) hypertrophy (Fulton index), weight of RV relative to the weight of the left ventricle (LV) + septum (S). (E) Cardiac output (CO) by echocardiography, normalized to body weight (BW). (F) Representative histology of distal pulmonary arteries from normoxia-, vehicle-, and AG-treated rats; scale bars = 50 μm. (G) Ratios of lumen to vessel area and (H) lumen to vessel diameter, based on analysis of 15 vessels per lung section for each group. In B, C, D, G, and H, bars represent mean±SEM, n=6, *P < 0.05 (shown in G and H), **P < 0.01 vs. Normoxia, #P < 0.05 vs. vehicle (Veh), one-way ANOVA with Bonferroni multiple comparisons test.

Before initiation of the treatment at week 6, all rats exposed to the Su/Hx protocol showed signs of severe pulmonary hypertension compared with control rats maintained in room air, as affirmed by echocardiographic evidence of reduced PA acceleration time/ejection time (PAAT/ET) (Fig. 6C). After drug treatment for three weeks, we found that compared with vehicle control, AG1296 significantly (P < 0.05) increased PAAT/ET and decreased right ventricular systolic pressure (RVSP), indicating that PA function was significantly improved (Fig. 6, B and C). There were trends toward reduced RV hypertrophy and increased cardiac output, but values did not reach statistical significance (Fig. 6, D and E). Morphometric analysis of the percent occlusion of small precapillary arteries showed that AG1296 almost completely reversed this structural abnormality (Fig. 6, F, G, and H).

AG1296 is superior to other TKIs in restoring function in iPSC-ECs

Another TKI, imatinib, showed good potential for reversing PAH as judged by the monocrotaline-induced PH rat model (18), however a multicenter clinical trial resulted in potential benefit only in a small subset of patients that was offset by the complication of subdural hematoma (60). It has also been shown that dasatinib, used in the clinic to treat CML, can cause PAH in some patients due to the induction of pulmonary EC apoptosis, via increased production of ROS (22). Therefore, we compared AG1296 to imatinib and to other TKIs thought to have less potential toxicity (axitinib and pazopanib) (61), or that induced PAH (dasatinib). We found that AG1296 treatment was superior in reducing caspase activity under serum withdrawal compared to imatinib and axitinib, whereas dasatinib and pazopanib induced EC apoptosis compared with vehicle control (Fig. 7A).

Fig. 7. Comparison of AG1296 versus other TKIs in PAH iPSC-ECs.

PAH iPSC-ECs were incubated with vehicle (DMSO), AG1296, or other TKIs under serum withdrawal (0.2% FBS) for 24 hours. (A) Apoptosis was measured by Caspase3/7 fluorescence assay (doses indicated on x-axis). (B-D) Angiogenesis assays with 10 μM of compounds: (B) Representative images of tube formation in PAH iPSC-ECs treated with different TKIs (10 μM) with quantitative analysis (C, D). PAH iPSC-ECs were pretreated with vehicle or TKIs for 24 hours, and then cells were seeded on growth factor reduced Matrigel for another 6 hours under drug treatment. scale bar = 100 μm. (E) Representative western immunoblots and quantification of pAKT, pSMAD1/5, and ID1 activation under normal condition with full serum, or serum free condition or with different TKIs for 24 hours. (F) Representative western immunoblots and quantification of BMP receptors under conditions shown in (E). (G-H) Gene expression of BMP receptors and cell survival genes by real-time PCR. In A, C-H, bars represent mean±SEM, n=6, *P < 0.05, **P < 0.01, ***P < 0.001 vs. Vehicle (DMSO), #P < 0.05 vs. Full medium, &P < 0.05 vs. AG1296, one-way ANOVA with Bonferroni multiple comparisons test.

Additionally, AG1296 improved angiogenesis six hours after PAH iPSC-ECs were seeded onto Matrigel, while imatinib, axitinib, and pazopanib showed no changes in terms of number of tubes and tube length. In striking contrast, dasatinib, which induced PAH in patients with CML, showed severely impaired tube formation when compared to vehicle-treated iPSC-ECs (Fig. 7, B, C, and D). After 24 hours, when the tubes started to regress, iPSC-ECs treated with AG1296 maintained better tube structures compared with imatinib and pazopanib, although this feature was also improved by imatinib and pazopanib when compared to vehicle control (fig. S7).

Inflammation is associated with PAH; i.e., the PA lesions have elevated amount of inflammatory cytokines, such as interleukin (IL)-6 and granulocyte macrophage colony-stimulating factor (GM-CSF), and contain inflammatory cells including macrophages, T cells, B cells, and dendritic cells (6, 62, 63). In PAH iPSC-ECs, we showed by ELISA that TNFα induced a significant (P < 0.05) increase in IL-6 and GM-CSF production (fig. S8, A and B). Pretreatment with AG1296 reduced both IL-6 and GM-CSF secretion, but there was no distinction between AG1296 and other TKIs with the exception of dasatinib, which brought the amount of these inflammatory cytokines to the lowest quantity (fig. S8, A and B), possibly due to its toxic effect that shut down the protein production machinery.

We next investigated whether the superior effect of AG1296 when compared with other TKIs was related to non-canonical BMPR2 signaling. Compared with other TKIs, AG1296 showed superior activation of pAKT, pSMAD1/5 and ID1 (Fig. 7E). While the increase in pSMAD1/5 was variable with imatinib, the induction of the transcription factor ID1 was only seen with AG1296 treatment.

To explain this superior effect of AG1296 in activating pAKT and BMP signaling, we examined gene expression of different BMP receptors upon treatment with TKIs. We found that AG1296 significantly (P < 0.05) increased BMPR1A, BMPR1B and BMPR2 at both mRNA and protein abundance compared with vehicle, while other TKIs either decreased (desatinib), or increased only one or two out of the three BMP receptors (Fig. 7, F and G). Additionally, AG1296 promoted the expression of angiogenesis- and cell survival-related genes such as VEGFA and BIRC3 to a higher level than the other TKIs (Fig. 7H). To confirm that AG1296 functions through ID1, we reduced ID1 in PAH iPSC-ECs by siRNA and showed that AG1296 was unable to improve EC function as judged by the caspase apoptosis assay (fig. S9A) and angiogenesis tube formation assay (fig. S9, B and C).

The effect of AG1296 and other TKIs on SMCs

AG1296 was superior to imatinib and axitinib in suppressing the proliferation of PAH PA SMC as judged by the BrdU assay, but was not as efficient as dasatinib and pazopanib (fig. S10A). Dasatinib, axitinib, and pazopanib also suppressed pAKT and pERK signaling to a greater extent than AG1296 (fig. S10B), indicating that the superior effect of AG1296 compared with other TKIs may depend on its combined beneficial impact on ECs and SMCs.

AG1296 showed a more robust anti-PAH gene signature than other TKIs

Because pSMAD1/5 was similarly activated in response to AG1296 and imatinib, we sought to explain whether ID1 expression that only occurred with AG1296 could be related to a pSMAD1/5 co-activator. We compared the AG1296 and imatinib anti-PAH signatures and found that the anti-PAH signature of AG1296 was more robust than that imatinib (Fig. 8A). Moreover, the gene expression profile for the deleterious TKI, dasatinib, closely resembled the PAH disease signature (Fig. 8B). This indicated that the LINCS database could be used to identify agents for further testing that could be deleterious in causing PAH. By applying this approach, we were able to identify additional compounds not included in our libraries that could be tested for either beneficial (fig. S11A) or detrimental (fig. S11B) effects on patients with PAH.

Fig. 8. AG1296 increases CREB3 gene expression to mediate increased ID1 in PAH iPSC-ECs.

Heatmap showing the PAH signature and anti-signature, as well as gene expression profile associated with beneficial TKIs (A) or toxic TKIs (B), based on LINCS database. (C) CREB3 and CREB5 expression quantified by real-time PCR in PAH iPSC-ECs. n=3, Bars represent mean±SEM. *P < 0.05, **P < 0.01, ***P < 0.001 vs. Vehicle, one-way ANOVA with Bonferroni multiple comparisons test. (D) ID1 gene expression in PAH iPSC-ECs treated with respective CREB3 and CREB5 siRNAs for 48 h, as determined by real-time PCR. n=3, Bars represent mean±SEM. **P < 0.01 vs. siControl with Vehicle (DMSO; 0 μM AG), #P < 0.05 vs. siCREB5 with Vehicle, one-way ANOVA with Bonferroni multiple comparisons test.

In comparing the LINCS profile of AG1296 to other TKIs as this relates to the PAH anti-signature, we identified a candidate gene that could explain improved BMPR2 mediated gene regulation (Fig. 8A, table S5). CREB5 (cAMP responsive element binding protein 5) was up-regulated only by AG1296, but not by other TKIs. KEGG pathway enrichment analysis based on these genes revealed that CREB is an upstream regulator of anti-apoptotic proteins, leading to the inhibition of apoptosis. Previous studies also showed that CREB and SMAD1 bind to the active promoter and enhancer regions of BMP-responsive genes including ID1, and CREB could therefore function as a pSMAD1/5 co-activator (64). Using qPCR, we verified that AG1296 increased the expression of CREB3 as well as CREB5 (Fig. 8C) in iPSC-ECs. As might be predicted, dasatinib significantly decreased CREB3 (P < 0.05). We then showed that reducing amounts of CREB3 in particular abolished the induction of ID1 expression by AG1296 (Fig. 8D). Additionally, AG1296 (20 μM) increased both CREB3 and CREB5 expression in PAH lung organ culture (fig. S12, A and B). These data suggest that AG1296 increases BMPR2, BMPR1A and CREB3/5, leading to the activation of pSMAD/ID1 and pAKT pathways, and accounting for the improved EC survival and angiogenesis in PAH iPSC-ECs (fig. S13).

Discussion

The current study highlights the effectiveness of functional high throughput studies with patient-specific iPSC-derived vascular cells combined with in silico analyses of publicly available datasets for drug discovery. Previous studies used a high-throughput drug screen with relevant PAH patient cell lines to identify potential therapies for PAH. For example, drugs were screened for their inhibition of PAH PASMC proliferation, and celastramycin (a benzoyl pyrrole-type compound originally found in a bacteria extract) was identified as a compound that also reduced cytosolic reactive oxygen species (ROS) and secretion of inflammatory cytokines (65). However, this screen used cells derived from patient lungs at explant or at postmortem, rather than those from individuals that could benefit from a novel therapy and did not assess the contribution of celestramycin to improving EC dysfunction (66). Our screening platform using iPSC-derived cells could be applied to iPSC-ECs as well as to iPSC-SMCs in a patient-specific manner. In addition to identifying compounds that improve function across all cell lines, our strategy would identify agents that might be particularly beneficial in a subset of patients with PAH.

FK506 was identified by screening compounds that specifically targeted the BMPR2-ID1 signaling pathway in a mouse myoblast cell line (27). However, human vascular cells were not used and agents that could improve vascular function independent of BMPR2 signaling would not be identified. This may explain the variable response to FK506 (28). Another study used a high-throughput short hairpin RNA (shRNA) screen to identify fragile histidine triad (FHIT) as a novel BMPR2 target (31), and the LINCS database to predict that enzastaurin, a compound that increases FHIT expression, might reverse vascular remodeling in the Sugen/hypoxia rat model. Our study represents an advance in combining three benefits of a high throughput screen: (i) phenotypic endpoints, (ii) patient-specific iPSC-vascular cells as phenotype/genotype surrogates for native PA cells, and (iii) an in silico approach to select a lead compound. Our studies could be further improved by generating LINCS gene expression signatures using vascular cells rather than cancer cells.

The beneficial effect of emerging PAH therapies, like elafin and FK506, is related both to their activation of BMPR2 signaling and to their anti-inflammatory activities. By increasing expression of both CREB3 and CREB5, AG1296 not only activated BMPR2 signaling but also would be expected to improve CREB-PGC1α signaling required for mitochondrial biogenesis (67). Mitochondrial dysfunction contributes to the pro-inflammatory and pro-apoptotic state of PAH PAECs (2). Additionally, CREB plays a central role in adipocyte survival (68), and down-regulation of CREB-binding protein is associated with decreased endothelial nitric oxide synthase (eNOS) expression and nitric oxide production, and increased vulnerability of ECs to apoptosis (69). More importantly, CREB suppresses RhoA activity by controlling p190RhoGAP-A expression and thereby maintains basal endothelial barrier function and reduces endothelial permeability in response to diverse agonists such as thrombin, lipopolysaccharide, and histamine (70). Notably, subdural hematoma associated with imatinib treatment in some patients with PAH might be related to endothelial gap junctions and increased endothelial permeability (71). Thus, stabilizing endothelial barrier function by increasing CREB could be critical in reducing the adverse events associated with imatinib.

Although the beneficial effect of imatinib as assessed in PAH cells and animal models by several groups predicted a favorable outcome in clinical trials, this was not realized. Here, we showed that there are non-canonical effects of TKIs that clearly distinguish AG1296 and imatinib. Not only was there evidence that AG1296 exhibited a more robust improvement of EC function, but the mechanism of this superior action was related to a stronger anti-PAH signature. Our data indicates that the beneficial function of AG1296 was unrelated to its targets as a TKI. It is uncertain how AG1296 increased gene and protein expression of BMP type 1 and type 2 receptors as well as CREB3, resulting in improved BMP signaling and functions in ECs. The direct target of AG1296 in ECs requires further investigation. It is possible that in SMCs, the anti-proliferative effect and the suppression of pAKT and pERK by AG1296 was mediated by its known targets PDGF, FGF, and KIT.

We have chosen to use iPSC to test the efficacy of AG1296. In addition to the six compounds with a positive response across all the cell lines tested, we identified seven other drugs that were effective in different subgroups of patients. AG1296 appears to act by a mechanism that could be complementary to other agents that improve BMPR2 function such as FK506 (27), elafin (34), or sotatercept (72), or that have a strong anti-PAH signature such as resveratrol. Future preclinical and clinical studies could consider assessing the independent as well as the combined effect of AG1296 with these agents.

AG1296 normalized function of cultured PAH iPSC-ECs and it partially reversed the PAH phenotype in the Sugen/hypoxia rat model. It is possible that a higher or more prolonged dosing regimen may have proven more effective. Our findings underscore the difficulties in optimizing a dose that could be most efficacious in a clinical study. For example, previous studies have shown that pre-clinical dosing can be problematic in extrapolating a clinically effective dose where it is necessary to follow drug amounts and to avoid human and patient-specific side-effects (28). Although our study provides a mechanistic explanation for the superiority of AG1296 relative to other TKIs in improving PAH cell function, and attests to the strength of the bioinformatic and high-throughput cell biology selection process, the ultimate test will be the performance of AG1296 in a clinical trial. In summary, our study highlights the utility of using patient-specific iPSC derived cells in a high-throughput screening platform to identify compounds that can then be further tested for potential efficacy or toxicity by combining functional assays with a bioinformatics approach to elucidate their mechanism of action.

Materials and Methods

Study design

To identify new compounds that could reverse the vascular remodeling in patients with PAH, we carried out high-throughput drug screening on six PAH iPSC-EC lines. Cells were treated with 4,500 compounds overnight under serum withdrawal condition, and cell survival was determined by a luminescence assay measuring caspase activity. All patient information was de-identified in accordance with the relevant Health Insurance Portability and Accountability Act (HIPAA) regulations, and tissues were collected with informed patient consent. Human iPSCs were generated under the Stem Cell Research Oversight protocol (SCRO #654). Animal studies to assess drug efficacy were conducted under the animal protocol (APLAC #31608). Measurements of cell survival, gene expression, and signaling pathways in six PAH patient cell lines were not blinded. Quantification of the number of tubes before and after drug treatment in the angiogenesis assay, and the animal studies were blinded. A minimum of n = 3 biological replicates were conducted for each experiment. Individual subject-level data are reported in data file S1.

High-throughput drug screen

Libraries of 4,500 Food and Drug Administration (FDA)-approved unique drugs and bioactive compounds (NIHCC, LOPAC, Biomol ICCB Known Bioactives, Microsource spectrum, Biomol FDA-approved drug library) were available from the High-Throughput Bioscience Center at Stanford University. We first optimized screening conditions such as cell number, time course for serum withdrawal, and incubation time for the luciferase reagent. We used Endothelial Cell Growth Medium-2 (EGM-2; Lonza) with 2% fetal bovine serum (FBS, Lonza) and 10 ng/mL vascular endothelial growth factor (VEGF; Gemini) as our positive controls, and high dose dimethyl sulfoxide (DMSO, Sigma) as a negative control in our preliminary studies. We developed the protocol that produced the most reproducible and sensitive readout for the caspase3/7 activity assay measuring cell apoptosis under serum withdrawal. For the high-throughput screen, PAH iPSC-ECs (n=6 lines) were dissociated with Accutase (Gemini) for 5 minutes to generate a single cell suspension. After enzyme neutralization with medium, cells were re-suspended in EGM-2 with 2% FBS, and were plated at 3,000 cells/50 μL medium in each well of a solid white 384-well plate (EK-30080, E&K Scientific) using the Matrix WellMate dispenser. These plates were then placed into the automated incubator at 37°C, 5% CO₂ overnight. The next day, the medium was replaced with serum-free media (EBM with 0.2% FBS) to induce apoptosis, except for the positive control wells that received fresh EGM-2 medium with 2% FBS. The plates were then returned to the Staccato System and 100 nL of compounds from aforementioned libraries were added using the Pin Tool. After an additional 24-hour incubation at 37°C, 5% CO₂, 10 μL of Caspase-Glo 3/7 Reagent (Promega) were added with the LabSystems Multidrop 384. After one-hour incubation, the luminescence signal was determined by Tecan Infinite M1000 PRO (0.2 sec/well). Positive results were defined as compounds that inhibited luciferase activity by more than 50% compared to no compound controls. The Z score of the screen was above 0.5, showing a robust and reproducible signal. Promising agents were subsequently tested in eight serial dilutions.

Endothelial differentiation of human iPSCs

We differentiated ECs from iPSCs based on a previous published protocol reported by our group (9). Briefly, iPSCs (over passage 15) were cultured to 70% confluence and placed in differentiation medium (RPMI and B-27 supplement minus insulin, Life Technologies) with 6 μM CHIR-99021 (Selleck Chemicals) for two days, followed by 3 μM CHIR-99021 for another two days. The medium was then changed and 50 ng/mL vascular endothelial growth factor (VEGF; Gemini), and 25 ng/mL basic fibroblast growth factor (FGFb; Gemini) were added for four additional days. On the eighth day, iPSC-ECs were sorted for CD144⁺ using antibody-coated beads and a magnetic-activated cell sorter (MACS) sorter (Miltenyi), and expanded on 0.2% gelatin coated plates. After sorting, iPSC-ECs were cultured in Endothelial Cell Basal Medium-2 (EBM-2) supplemented with the EGM-2 BulletKit (Lonza) at 37°C, 21% O₂, and 5% CO₂ in a humidified incubator with medium changes every 48 hours. Cells were passaged once they reached 80%–90% confluence. iPSC-ECs used for experiments were between passages 2 and 5.

Angiogenesis assay

PAH iPSC-ECs were cultured for 12 hours in serum free medium for synchronization. The angiogenesis assay was performed using an In Vitro Angiogenesis Assay Kit (Trevigen), following the manufacturer’s instructions. Briefly, 96-well plates were coated with basement membrane extract (BME) solution provided by the kit for 30-60 minutes at 37°C, and the cells stained with Calcein AM for fluorescent monitoring of the tubes. iPSC-ECs were seeded at 3x10⁴ cells per 100 μL EBM2 medium per well in pre-coated 96-well plates. Vehicle (DMSO) and other top candidate compounds (10 μM) were added to medium containing 0.2% FBS. Tube length and the number of tubes were measured after 6 hours and 24 hours in three microscopic fields using a Leica computer-assisted microscope with SPOT microscope digital imaging software. ImageJ was used to process and analyze the images.

Cell apoptosis/caspase activity validation

The caspase assay was modified from a previous publication (34). Cells were seeded in a 96-well plate (10,000 cells per well) in EGM-2 medium with full serum and allowed to adhere overnight. Cells were then washed and incubated with EC or iPSC-EC basal medium without FBS for 12 hours, followed by incubation with Caspase-Glo 3/7 Reagent (Promega), and total luminescence was measured in a plate reader.

Sugen/hypoxia PH rat model

We used the Sugen/Hypoxia (Su/Hx) rat model of PH (34). Briefly, eight-week-old male Sprague Dawley rats (180–220 g) were given a single subcutaneous injection of 20 mg/kg of the VEGF receptor blocker SU5416, followed by exposure to hypoxia (10% O₂) for three weeks, followed by three weeks of normoxia. Then the rats were divided at random into two groups, one treated with DMSO as vehicle control, and the other with AG1296 (50 mg/kg/d) by daily subcutaneous injection for three weeks. Untreated age-matched control animals were kept in room air throughout the study period. Echocardiographic measurements of cardiac function were obtained as previously described (27). Right ventricular (RV) systolic pressure was measured through right jugular vein catheterization, and RV hypertrophy (RVH) was assessed by the weight ratio of the RV to left ventricle and septum (LV+S). For histology, rat lung tissue was fixed in paraformaldehyde (PFA) for 48 hours and preserved in EtOH. Paraffin embedded lung slides were stained with a Movat pentachrome stain, where vessel loss and muscularization of pulmonary vessels could be visualized by light microscopy.

Lung organ culture

Tissue sections, containing PAs, were prepared from the explanted lungs of patients with PAH who had undergone lung transplantation. Mirror image sections were placed in organ culture and treated with either AG1296 in DMSO, or an equal volume of DMSO vehicle once daily for eight days. Movat staining was used to measure lumen diameter/external diameter and lumen area/vessel area. Lumen to vessel area and diameter were measured based on analysis of 15 vessels per lung section for each treatment group.

LINCS analysis

We used an integrated multi-cohort analysis to identify genes that were differentially expressed in patients with PAH compared to healthy controls. We identified six publicly available human PAH transcriptomic datasets from the NCBI Gene Expression Omnibus (GEO) (PBMC: GSE703, GSE22356, GSE33463, Lung: GSE15197, GSE48149, GSE69416). We downloaded and manually curated each dataset as previously described (31). Briefly, the data itself was normalized and converted to log2 using previously published methods. We used two different meta-analysis approaches: (i) combining fold changes and (ii) combining p-values as previously described. Differentially expressed genes were selected that had a specific false discovery rate threshold <1%, and a summary effect size (35) with absolute value greater than 0.6. These thresholds resulted in a PAH lung signature consisting of 330 differentially expressed genes.

The availability of the gene expression profile datasets for drugs allows identification of mechanisms of action of drugs towards PAH. We integrated a reference collection of gene-expression profiles from cultured human cells treated with bioactive small molecules. The database LINCS profiled a large number of drugs across many cell lines. LINCS is the largest database of gene expression profiles of cultured human cells treated with different drugs. At the time of analysis, there were 20,413 chemical perturbagens profiled on LINCS across 18 “gold” cell lines on the L1000 platform (www.lincscloud.org). By integrating the gene signatures found with the multi-cohort analysis with the drug profiles in LINCS, we were able to identify regulatory pathways affected by the drugs as well as assess how similar the expression profile of each drug was toward the PAH signature (or the anti-signature). In addition, this computational pipeline allowed us to compare the profiles of our compounds to establish differences in the mechanisms of action of each compound.

RNA-seq analysis

Quality assessment and pre-processing of RNA-seq reads were performed using FASTQC, Trim Galore, and samtools. Reads were then aligned to the human genome UCSC build hg38 using Bowtie2 (version 2.3.3.1). Low-quality alignments and PCR duplicates were removed using SAMtools and Picard MarkDuplicates tool. Gene expression was counted using HTSeq-count (73). DESeq2 (74) (version 1.26.0) was used for differential expression analysis using a paired design. The analysis was performed in R 3.6.1. Differential expression with at least a 1.5-fold change and false discovery rate (FDR) < 0.1 was considered significant for SMC and PAEC, and 1.2-fold change and FDR<0.1 for iPSC-EC.

Statistical analysis

Histograms represent the arithmetic mean±standard error (SEM). Statistical significance was determined by two-sided unpaired t-test for comparison of two groups, or by one-way ANOVA with Bonferroni’s post-hoc testing for comparisons of three or more groups. Analyses were carried out using GraphPad Prism 8.0.

Supplementary Material

table s3

Table S3. Differentially expressed genes regulated by AG1296 treatment in RNA-seq of both PAH iPSC-EC and PAEC.

data s1

Data file S1. Individual subject-level data.

table s4

Table S4. Differentially expressed genes regulated by AG1296 treatment in RNA-seq of PAH PASMC.

Supplementary material

Fig. S1. PAH PAEC and iPSC-EC show similar response to emerging PAH therapies.

Fig. S2. Resveratrol has no impact on tube formation.

Fig. S3. AG1296 decreases EC migration, and has no impact on cell adhesion to laminin.

Fig. S4. AG1296 activates cell survival under hypoxia.

Fig. S5. AG1296 induced pAKT is independent of FGFR, KIT and PDGFR.

Fig. S6. AG1296 reduces PAH SMC proliferation by suppressing pAKT, pERK, and Wnt pathways.

Fig. S7. AG1296 vs. other TKIs maintains better tube formation for 24 hours.

Fig. S8. AG1296 reduces the secretion of inflammatory cytokines in PAH iPSC-ECs.

Fig. S9. Suppression of ID1 in PAH iPSC-ECs prevents the benefits of AG1296.

Fig. S10. Comparison of AG1296 with other TKIs in SMC proliferation and signaling.

Fig. S11. Bait-correlation analysis predicts new beneficial or toxic agents for patients with PAH.

Fig. S12. CREB3 and CREB5 could be induced by AG1296 in PAH lung organ culture.

Fig. S13. Proposed model for the mechanism of action of AG1296.

Table S1. Characteristics of patients with PAH for high-throughput drug screen.

Table S2. Top candidate drugs that improved PAH iPSC-EC survival and tube formation.

Table S5. Differentially regulated genes by AG1296 vs. other TKIs that correlated with PAH anti-signature.