Reversal of inflammatory reprogramming by vasodilator agents in pulmonary hypertension

Authors: Anna Kirillova, Meena Sethuraman, Xishuang Dong, Almina Kirdar, Gil Speyer, Yassmin Al Aaraj, Annie Watson, Lily K. Schneider, Michael D. Creager, Robert Lafyatis, Satoshi Okawa, Seungchan Kim, Stephen Y. Chan

Abstract

PAH: pulmonary arterial hypertension; PDE5i: phosphodiesterase-5 inhibitor; EDDY: Evaluation of Differential DependencY; NK: natural killer.

Introduction

Pulmonary arterial hypertension (PAH) is a progressive disease of the pulmonary vasculature without cure. The diagnosis is established after right heart catheterisation [1], an invasive haemodynamic procedure. There is a need for early and accurate non-invasive diagnostic and prognostic tools.

Systemic inflammation is increasingly associated with PAH pathogenesis [2]. Perivascular inflammation in the lungs of PAH patients contributes to vascular remodelling [3–5]. In particular, myeloid cells are recruited from the bone marrow contributing to the engraftment of immune cells from blood to lung [5–7]. Thus, inflammatory cells in the blood are likely to reflect initiation and progression of PAH in the lung and are translationally relevant to PAH pathogenesis.

Blood-based measures of immune activation in PAH could address a controversy in the field, namely whether current vasodilator drugs reverse immune reprogramming. It has been assumed that these drugs act through vasomotor tone pathways and do not prevent or regress the anti-apoptotic and pro-proliferative phenotypes of vascular cells in PAH. Phosphodiesterase-5 inhibitors (PDE5i) primarily control nitric oxide, a vasodilator, while endothelin-receptor antagonists, such as macitentan, inhibit endothelin vasoconstriction signalling [8]. PDE5i [9, 10] and endothelin-receptor antagonists [11, 12] may harbour anti-inflammatory properties. But this mechanism has not been well-defined in PAH patients, and whether such putative actions are due to direct drug effects on inflammatory cells or to indirect actions on systemic immunoactivation has not been fully explored. Furthermore, while transcriptomes in circulating immune cells have been reported [13–19], serial stages of PAH with or without treatment have not been evaluated. Such longitudinal profiling across untreated, single and combination drug treatments is especially challenging to study in the current era where immediate dual and sometimes triple combination vasodilator therapy is recommended for symptomatic PAH patients, based on the Ambrisentan and Tadalafil Combination Therapy in Subjects with Pulmonary Arterial Hypertension (AMBITION) trial [20]. Single-cell analyses could address these gaps and pave the way for clinical platforms. To do so, we leveraged orthogonal analyses beyond differential gene expression (DGE), which often underestimate intergenic relationships [21–23]. We further defined intergenic relationships by analysing single-cell RNA sequencing (scRNA-seq) data via the Evaluation of Differential DependencY (EDDY) [24, 25], which defines differential dependency networks (DDNs) based on the differential interconnectivity of genes in a pathway. Such connectivity can be perturbed in untreated versus treated disease. We sought to identify the effect of PDE5i and PDE5i+macitentan therapy on disease-based networks. A combination of DGE and EDDY analyses of scRNA-seq data could identify the comprehensive landscape of transcriptional inflammatory cell reprogramming in PAH.

Therefore, this systems-biology study in PAH incorporated both DGE and differential network connectivity analyses of longitudinal patient samples collected from untreated patients, from patients undergoing single and combination vasodilator therapy and from patients recruited prior to the initiation of combination therapy in PAH. In doing so, we sought to define at single-cell 1) any differences between idiopathic PAH (iPAH) and systemic sclerosis-associated PAH (sscPAH) patients in such responses, and 2) any evidence of immune cell reprogramming after either single or combination vasodilator therapy, as reflected by reversal of immune cell transcriptomic changes in PAH.

Methods

Patient data

We isolated peripheral blood mononuclear cells (PBMCs) from patients seen at the University of Pittsburgh Medical Center via peripheral venous blood draw. sscPAH patients (n=4) and iPAH patients (n=5) were recruited to the study. Serial samples were obtained at the untreated PAH stage, after receiving stable PDE5i for 3 months and 3 months after the addition of macitentan treatment. A similar population of cells was isolated from non-diseased matched control persons (non-PAH control patients (n=3): a 66-year-old female, a 33-year-old male and a 28-year-old female). Blood was analysed from untreated PAH patients (sscPAH, n=2; iPAH, n=2; total untreated PAH, n=4), PAH patients receiving PDE5i treatment (sscPAH, n=3; iPAH, n=4; total PAH+PDE5i, n=7) and PAH patients receiving PDE5i and macitentan treatment (sscPAH, n=3; iPAH, n=3; total PAH PDE5i+macitentan, n=6). sscPAH and iPAH patient samples were combined to increase the statistical power of the analyses for the PAH group.

The inclusion criteria for sscPAH and iPAH patients were as patients aged 18–75 years; a confirmed diagnosis of sscPAH or iPAH by right heart catheterisation; mean pulmonary arterial pressure >25 mmHg; pulmonary capillary wedge pressure <15 mmHg; pulmonary vascular resistance >3 Wood Units; New York Heart Association Class I–III; current pulmonary vasodilator therapy with PDE5i; and ability to take 10 mg of macitentan p.o. daily, for at least 3 months, as per standard of care. The exclusion criteria included group II–V pulmonary hypertension, treatment-naïve pulmonary hypertension patients, patients on combination pulmonary vasodilator therapy, patients on prostacyclin, riociguat or endothelin-receptor antagonist therapy, inability to maintain dosing of macitentan (10 mg p.o. daily) for the 3-month trial period. Supplementary table S1 lists the demographics of PAH patients and control individuals recruited for this study. The number of recruited patients was determined primarily by the availability of clinical samples. iPAH and sscPAH patient PBMCs were utilised for the single-cell analysis of monocytes, B cells, CD8+ T cells, CD4+ T cells and natural killer (NK) cells. Experimental procedures involving human tissue were approved by institutional review boards at the University of Pittsburgh (IRB STUDY19070123). Ethics approval for this study and informed consent conformed to the standards of the Declaration of Helsinki.

Single-cell RNA sequencing

Blood samples collected in EDTA purple-top Vacutainers were centrifuged at 2000 rpm for 15 min and plasma was removed for banking. Buffy coats were collected and used for Ficoll gradient to isolate PBMCs, which were used for the scRNA-seq analysis. RNA libraries were prepared using Chromium Single Cell 3' Reagent Kits v.2 per manufacturer's guidelines (10x Genomics manual CG00052, catalogue no. PN-12237). scRNA-seq was performed using the GemCode Single Cell Instrument (Chromium, 10x Genomics), which included at least 800 cells (∼50,000 reads per cell) per each patient sample. Lung samples were processed and analysed as described previously [26].

Bioinformatics analysis

Raw data from scRNA-seq were processed using 10x Genomics’ Cell Ranger v.8 to generate raw counts data for individual cells. Raw counts data were first analysed for normalised counts and potential batch corrections among different patient samples using SCTransform in Seurat v.3 [24, 27, 28]. Cell types were determined with SingleR [25] using the Blueprint ENCODE reference [29, 30]. Normalised counts were then analysed for differential gene dependency networks and for DGE analysis, followed by pathway enrichment analysis. For DGE analysis of each cell type between different conditions, the normalised expression matrices were analysed using the Wilcoxon rank-sum test implemented in the FindMarker function in Seurat [24, 27, 28] with 10 minimum percent expression threshold and 0.1 logfc threshold. The Benjamini–Hochberg-adjusted p-value cut-off was set to 0.05. Gene set enrichment analysis (GSEA) was conducted on differentially expressed genes (DEGs) with the Enrichr R package using the “Reactome_2022” databases [27]. The false discovery rate (FDR)-adjusted p-value cut-off was set to 0.05. Statistical testing of cell proportions was conducted using propeller [28].

EDDY analysis to identify pathways with differential dependency

Traditional pathway enrichment analysis following differential expression analysis often underestimates the systems-level interactions underlying complex biological processes present in drug response. To complement DGE-based pathway analysis, higher-level network analyses were undertaken. That is, we engineered a computational strategy that identifies DDNs, which are crucial to defining the landscape of transcriptional network reprogramming expected in cells’ response to therapies across the sampled cell types. We built upon the capabilities of a previously developed computational method, EDDY [24, 25], a highly specific and sensitive computational pipeline that defines DDNs based on the rewiring of interactions among genes in a network when comparing two conditions. Previously, we applied EDDY to identify rewired DDNs in PAH lung tissue versus control lung tissue [31]. We leveraged the principles of EDDY for use in single-cell transcriptomic data, defining the comprehensive gene expression rewiring across the sampled blood-borne cell types, in the context of PAH disease states and vasodilator treatment.

Gene–gene interactions are highly context-dependent; hence, identification of the rewiring of such interactions may reveal complex biological processes involved in the therapeutic responses of cells, as well as the development of disease. EDDY4 was developed to capture such context-dependent gene interactions by combining robust statistical analysis of gene expression data and pathway databases, such as Reactome.

Given a gene set, EDDY constructs graphs for gene dependencies in each condition, for example, B cells of untreated PAH, where edges between nodes were defined by a pairwise independence test (chi-square test) of gene expression, with known interactions (edges) were given priority. By repeated resampling of each group (B cells of untreated PAH versus non-PAH control), multiple unique networks were constructed for each group; upon scoring, each group was characterised with a network likelihood distribution. Divergence between two network likelihood distributions was computed via Jensen–Shannon entropy and the statistical significance (p-value) of the divergence between the two distributions was assessed via permutation test. Thus, each gene set in a pathway database, for example Reactome, was assessed for Jensen–Shannon entropy and corresponding p-value, and gene sets with statistically significant differential dependency among gene networks between two conditions were catalogued. A pathway with differential dependency can be visualised as a DDN. For visualisation, each characteristic line in DDNs indicates the identified relationship between nodes (genes). A line indicates a gene-to-gene connection.

Cell culture

THP-1 (human leukaemia monocytic cell line) monocytes (ATCC, TIB-202) were grown to confluency and serum-starved in 0.1% FBS+RPMI-1640 (Gibco, 11875093) either with or without exposure to 10 ng⋅mL^−1^ of lipopolysaccharide (LPS) (Sigma, L4391-1MG) for 24 h. After 24 h, cells were re-plated at 5×10^5^ cells per well in a six-well plate in 10% FBS+RPMI-1640. Cells with exposure to LPS were treated with 25 μM of sildenafil+25 μM of macitentan (Sigma, S-010-1ML; Sigma, SML3386-10MG); cells without exposure to LPS were treated with drug vehicles. Methanol or methanol+dimethyl sulfoxide (DMSO) were used as vehicle controls.

RNA isolation, reverse transcription and quantitative real-time PCR

Total RNA from cells was isolated using QIAzol lysis reagent (QIAGEN) according to the manufacturer's protocol. 500 ng of total RNA was reverse transcribed into cDNA using a high-capacity cDNA reverse transcription kit (Thermo Fisher Scientific). Quantitative real-time PCRs were performed on a QuantStudio 6 Flex Real-time PCR System (Applied Biosystems). TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific, 4444965) and TaqMan primers were used for GNG5 (Hs00893832_g1), GNG11 (Hs00914578_m1) and GNB2 (Hs00929275_g1). PGK1 (Hs00943178_g1) was used as a housekeeping gene control to normalise the data.

Results

scRNA-seq identified transcriptomic and connectome changes in PBMCs in PAH (naïve to PAH treatment) versus non-PAH individuals

To define the transcriptional landscape across blood inflammatory cells, we compared scRNA-seq datasets from PBMCs of PAH patients (combined iPAH and sscPAH) naïve to vasodilator treatment (untreated PAH, n=4) with those of non-PAH control individuals (n=3) (figure 1a and supplementary tables S1 and S2). Alterations of single gene expression were assessed through DGE analysis, and alterations of network connectivity were defined via EDDY.

We utilised post hoc identification of cell type-specific transcripts to generate uniform manifold approximation and projection (UMAP) plots displaying clustering of major cell types (supplementary figure S1a–d). Upon cell type identification, modest differences were noted in untreated PAH versus non-PAH controls in terms of cell proportions across the major cell type populations (supplementary figure S1e). Based on consistent identification of predominant cell types across samples, further analysis focused on B cells, CD4+ T cells, CD8+ T cells, monocytes and NK cells.

To define cell type-specific transcriptome changes, we compared untreated PAH with non-PAH conditions in these cell types and investigated the top 20 up- or downregulated DEGs (figure 1b and c, and supplementary table S3). Notably, genes related to the human leukocyte antigen (HLA) clusters, such as HLA-A, HLA-DPA1 and HLA-B, were upregulated (figure 1b). These findings are consistent with those of genome-wide association studies whereby a gene locus HLA-DPA1 is associated with risk and severity of PAH [32–34]. The S100 superfamily was upregulated in multiple cell types, consistent with its known role in immune and vascular regulation in PAH [35]. Several cell types had prominent downregulation of ribosomal protein (RP) superfamily genes, which are associated with fundamental processes such as protein translation (figure 1c and supplementary table S3). Furthermore, monocytes displayed prominent downregulation of HLA-DQB1, -DRA and -DQA1, while HLA-C was downregulated in B cells, CD4+ T cells, CD8+ T cells and NK cells, highlighting the dynamic role of immune processes and antigen presentation to the vascular endothelium in PAH.

To further define regulatory pathways represented by the gene expression alterations in untreated PAH versus non-PAH control in each major cell type, we performed GSEA [27]. Overall, GSEA identified enrichment of 44 pathways in B cells, 45 in CD4+ T cells, 70 in CD8+ T cells, 177 in monocytes and 54 in NK cells (supplementary table S4). Top 20 pathways as defined by the Reactome pathway database [36] were overrepresented in all five immune cell types and were associated with transcription and translation, cell cycle and infection (figure 2a and supplementary table S7).

Beyond the DGE and GSEA analytics, EDDY offered orthogonal evidence into connectivity within intergenic networks across immune cells in each condition, annotated by the Reactome database. Disease states or response to treatment can alter intergenic connectivity, resulting in loss or gain of connections (figure 1a). EDDY revealed 57 differentially connected networks (p<0.05, as annotated by Reactome) in the untreated PAH condition (compared with non-PAH individuals) (figure 2b and supplementary table S5). To define particularly significant immune reprogramming in PAH, we focused on pathways that were identified by both GSEA and EDDY analyses. Comparisons between these modalities revealed no overlapping pathways in B cells, CD4+ T cells, NK cells or CD8+ T cells, but four unique overlapping pathways (GSEA: adjusted p<0.05 and EDDY: p<0.05) were observed in monocytes, suggesting a particularly active reprogramming of this pro-inflammatory cell type in PAH (figure 2b and supplementary tables S3 and S4). Specifically, the pathways entitled “antigen folding assembly and peptide loading of class I MHC” (GSEA: adjusted p=4.7589×10^−5^; EDDY: p=0.003147), “complement cascade” (GSEA: adjusted p=0.03495598; EDDY: p=0.004407), “downstream TCR signalling” (GSEA: adjusted p=0.00468134; EDDY: p=0.006211) and “SPRY regulation of FGF signalling” (GSEA: adjusted p=0.00982496; EDDY: p=0.03102) were identified. Thus, overall, combining a GSEA and EDDY analytic approach offered a more comprehensive landscape of transcriptional reprogramming across the PBMC cell types involved, revealing a shared signature of enriched pathways, particularly in monocytes. The full analysis generated via EDDY is accessible at https://chan.vmi.pitt.edu/PBMC_scRNAseq_EDDY/.

iPAH and sscPAH patients display both similarities and key differences in immune response to vasodilator therapy

To offer increased granularity of the analysis, we investigated the comparisons between the iPAH and sscPAH patient subgroups. sscPAH patients have a worse prognosis and response to treatment, partially due to the systemic involvement and the inflammatory nature of the disease [37]. Therefore, we hypothesised that iPAH and sscPAH patients may display different immune responses in the setting of PAH. To track the responses of iPAH and sscPAH patients to therapy, we utilised scRNA-seq data from iPAH or sscPAH patients prior to treatment (iPAH: n=2, sscPAH: n=2), 3 months post PDE5i (iPAH: n=4, sscPAH: n=3) and 3 months post PDE5i+macitentan (iPAH: n=3, sscPAH: n=3) (supplementary tables S1 and S2). After performing DGE analysis comparing untreated iPAH and sscPAH patients, we identified differentially up- and downregulated genes, emphasising the differences in immune and vascular disease processes (supplementary figure S2a and b). Furthermore, we performed GSEA comparing iPAH with sscPAH. The top 50 pathways were related to immune system regulation, response to stress, metabolism, amino acid, transcription and translation mechanisms (figure 3a). In addition, we compared iPAH with sscPAH in the setting of PDE5i treatment alone (supplementary figure S3a) and in the setting of PDE5i+macitentan treatment (supplementary figure S3b). After the addition of treatments, the pathways became more uniformly enriched in all immune cell types.

When comparing the proportions of each cell type in iPAH and sscPAH, monocytes were significantly enriched prior to treatment (FDR<0.01); however, no differences were observed following PDE5i and PDE5i+macitentan treatments [28] (figure 3b). Consistent with the more pro-inflammatory sscPAH programme, we observed significantly more upregulated DEGs in sscPAH immune cell types than in iPAH immune cell types (figure 3c). After the addition of PDE5i, the number of DEGs was substantially attenuated in sscPAH, followed by an increase after the addition of macitentan. These findings underscore the differences in therapy responses in these two PAH subtypes and provide evidence for the complex processes occurring at the level of immune cells.

Finally, we investigated individual longitudinal iPAH and sscPAH patient samples to track their response to each treatment. We established the following longitudinal 1) iPAH or sscPAH prior to treatment versus non-PAH control, 2) iPAH or sscPAH after PDE5i versus iPAH or sscPAH prior to treatment and 3) iPAH or sscPAH after PDE5i+macitentan versus iPAH or sscPAH after PDE5i alone. The top 50 pathways were represented in each comparison, and enrichment or depletion of pathways was analysed. Compared with iPAH (supplementary figure S4a), sscPAH (supplementary figure S4b) displayed more pathway enrichment following PDE5i+macitentan therapy, consistent with the increase in DEGs. Overall, these findings highlight the differences in transcriptional inflammatory responses between iPAH and sscPAH, as well as in responses to vasodilator therapy.

Pathway and network analyses revealed potentially therapeutic reversal of inflammatory cell reprogramming in PAH patients by vasodilators

We sought to determine whether vasodilator therapy controls and/or reverses inflammatory reprogramming in the blood of PAH patients. Specifically, we searched for pathways across inflammatory cell types that displayed DGE in untreated PAH versus non-PAH control groups but reversed directionality with single (PDE5i) or combinatorial (PDE5i+macitentan) treatment, indicating a therapeutic reprogramming of these immune cells with vasodilator treatment. Thus, we compared scRNA-seq of PBMCs in the combined iPAH and sscPAH patient cohort across therapeutic time 1) untreated PAH patients (prior to therapy, n=4); 2) 3 months after PDE5i therapy (n=7); and 3) 3 months after PDE5i+macitentan therapy (n=6) (supplementary tables S1 and S2). Across these datasets, we applied the systems-biology analytic pipeline, including GSEA and EDDY analyses (figure 1a).

Among those who received treatment, significant transcript changes were identified in nearly every inflammatory and immune cell type screened, with a substantial proportion displaying reversal of gene expression directionality, compared with untreated PAH versus non-PAH controls (individual DEG reversal adjusted p-value <0.05). Reversal of gene expression was observed after PDE5i alone, while other DEGs displayed reversal with PDE5i combined with macitentan (figure 4a and supplementary table S6). The reversal DEGs obtained from the dataset for each cell type were highly significant, based on empirical null distribution (B n=24, p-value <2.2×10^−16^; CD4+ T n=67, p-value <2.2×10^−16^; CD8+ T n=108, p-value <2.2×10^−16^; n=183, p-value <2.2×10^−16^; NK n=66, p-value <2.2×10^−16^) (supplementary figure S5a–e). Of relevance for macitentan, a specific cohort of genes displayed reversal of disease expression only after addition of macitentan, but not with PDE5i alone. Similarly, enhanced fold change was noted in some gene subsets, particularly after macitentan therapy, compared with treatment with PDE5i only. These findings define extensive transcriptional alterations in inflammatory cells triggered by PAH, but reversed with PDE5i or combinatorial drug treatment. To identify the regulatory pathways reversed by PAH treatment, we performed GSEA of reversed DEGs and constructed a representative heat map of the top 20 Reactome terms for each immune cell type (figure 4b and supplementary table S7). Most of the pathways significantly overlapped in each cell type. Fundamental cellular processes were represented consistently across all cell types, including DNA damage, RNA processing, metabolism and adaptive and innate immune functions.

In parallel, by application of EDDY, we identified a complementary set of networks that were differentially connected in any of the following 1) untreated PAH versus non-PAH control, 2) PDE5i versus untreated PAH and 3) PDE5i+macitentan versus untreated PAH (supplementary figure S6a–e). Differential connectivity implicated the activity of these gene networks in response to disease or to vasodilator treatment. As with the DGE analysis, we were specifically interested in identifying pathways that were reprogrammed with treatment – that is, pathways that showed altered connectivity in PAH and displayed reversal to baseline non-diseased connectivity after PDE5i or PDE5i+macitentan treatment. In untreated PAH, we identified reprogramming in four pathways in B cells, seven in CD4+ T cells, seven in CD8+ T cells and seven in NK cells. As with the DGE analysis, monocytes carried the most pathways (24) with altered connectivity in PAH (supplementary figure S6a–e). We then sought to further define reversal of individual connections (presence or absence) between genes in serial stages of untreated PAH, after PDE5i and after PDE5i+macitentan. Reversal of gene connectivity was defined as “ON” when a connection was present in non-PAH individuals, absent in untreated PAH and present again after the addition of either PDE5i or PDE5i+macitentan. Conversely, the “OFF” reversal state was computed, defined by a connection absent in non-PAH, present in untreated PAH and absent again with either PDE5i or PDE5i+macitentan use. After quantification of the number of “ON” and “OFF” states (supplementary figure S7), such patterns were identified in every cell type. However, they were most prominently observed in monocytes, for which more significant “ON”/“OFF” reversals were acquired after the addition of macitentan, highlighting its potential additive effect to PDE5i (figure 5a).

Notably, “glucagon signalling in metabolic regulation” in monocytes was the lone network found to be reversed by vasodilator therapy via both GSEA and EDDY pipelines (figure 5a and b). Specifically, DEGs such as GNG5, GNB2 and GNG11 within this pathway were reversed by vasodilators (figure 5b). Via EDDY analytics, this network contained 14 connections in the “ON” state – present in the non-PAH state, absent in the PAH state and regained with PDE5i+macitentan use; 15 connections were present in the “OFF” state – absent in the non-PAH state, gained in untreated PAH and absent again with PDE5i+macitentan use (figure 5c). No reversal of intergenic connections was detected with PDE5i treatment alone.

To determine if such vasodilator therapy carries direct versus indirect effects on “glucagon signalling in metabolic regulation” in monocytes, we performed additional validation of the DEGs identified in the “glucagon signalling in metabolic regulation” pathway in THP-1 monocytes (supplementary figure S10). We stimulated THP-1 monocytes with LPS to induce inflammation and subsequently treated them with PDE5i+macitentan. Pro-inflammatory LPS upregulated a known inflammation-responsive IRF7 transcript, which was also present in our DEG dataset. Notably, sildenafil and macitentan did not alter transcript levels of GNG5, GNB2 or GNG11, suggesting that there is not a direct vasodilator effect on “glucagon signalling in metabolic regulation” in monocytes.

Taken together, these findings converge on a notion that these molecules, and their modulation by PDE5i and macitentan, carry roles in regulating the dynamics of monocytes and their role in mediating metabolism in PAH – events that may be the drivers of insulin and glucagon signalling and could go far beyond control of simply vasomotor tone in PAH.

Classical and non-classical monocytes may have distinct roles in PAH

Since monocytes were strongly activated in the setting of PAH, we further investigated the differences between classical (CD14^high^ CD16^low^) and non-classical monocytes (CD14^low^ CD16^high^) subtypes. Classical monocytes are known to differentiate into tissue resident macrophages in the lung, while non-classical have a pro-inflammatory role; both play roles in PAH pathogenesis [7, 38]. In the setting of PAH, classical and non-classical monocytes displayed similar top 20 pathway activations (supplementary figure S8a). Treatment with PDE5i+macitentan activated different classical and non-classical monocyte pathways, suggesting that combinatorial treatment has more distinct effects on both cell types than PDE5i alone (supplementary figure S8b and c). Furthermore, we investigated the activation of the key “glucagon signalling in metabolic regulation” pathway in classical and non-classical monocytes (supplementary figure S9a and b). Classical monocytes most comprehensively recapitulated the regulatory pattern observed in the combined monocyte group (figure 5b). Overall, these findings suggest that the addition of macitentan has a unique role in activating classical and non-classical monocyte cell populations.

Discussion

In this study, we performed DGE/GSEA and EDDY analytics on serial scRNA-seq PAH patient samples to investigate inflammatory reprogramming and its relationship to vasodilatory agents. We observed 1) baseline and post-treatment differences in iPAH and sscPAH; 2) reversed DEGs and networks after either single or combination vasodilator use in the combined PAH group (iPAH+sscPAH); and 3) a key reversed pathway, indicating a role for glucagon signalling in PAH. These results are especially unique since contemporary recruitment of patients undergoing both single and dual therapy longitudinally is limited by current PAH treatment recommendations for upfront combination therapy [20]. Ultimately, these findings address a long-standing question regarding the anti-inflammatory activity of vasodilator therapy in PAH. In doing so, our findings set the stage for the possibility that some of the observed efficacy of vasodilators may extend beyond vasodilation alone and depend upon their anti-inflammatory properties. Future studies should be geared towards determining the contribution of the anti-inflammatory nature of these drugs on overall disease progression. In doing so, this work also establishes the foundation for single-cell “liquid biopsy” tools that could track inflammatory reprogramming in PAH.

Single-cell technology is accelerating rapidly, necessitating advances in large-scale analytics of complex data [39]. This study offers evidence of significant expanded transcriptional reprogramming, otherwise missed from traditional approaches, by applying EDDY to scRNA-seq datasets from individual PAH patients alongside DGE and GSEA. At a translational level, the DGE/GSEA and EDDY analytics pipelines pave the way to monitor real-time evolution of disease, as well as identification of new druggable pathways and reversal after delivery of therapies, thus informing clinical decision-making.

Through this process, firstly, we found fundamental differences between iPAH and sscPAH and their response to treatments, thus expanding our understanding of the immune response in PAH. Our results are consistent with the notion that sscPAH represents a pro-inflammatory subtype of PAH [40], potentially mediated by expansion of monocytes. Compared with iPAH, sscPAH patients also displayed more robust reprogramming after treatment with PDE5i+macitentan, which could shed light on the proposed unique benefits of PDE5i and/or macitentan for each PAH subtype [41].

Secondly, our findings also support the paradigm whereby vasodilator treatment can reprogramme the inflammatory landscape of PAH. At a more granular level, several pathways were activated across all major immune cell types with disease and treatment, suggesting the potential existence of multiple master regulators that orchestrate an inflammatory state. Our in vitro data (supplementary figure S10) suggest that the anti-inflammatory effect of vasodilators in PAH is likely to be due to an indirect mechanism resulting from vasodilation, with potential consequent effects on myeloid stimulation at the level of the bone marrow [5–7]. It is possible that improved haemodynamics via vasodilator treatment attenuate the stress responses from the pulmonary vasculature, dampening the pro-inflammatory stimulation on the bone marrow. Therefore, in response to less inflammatory stimuli, the bone marrow could suppress the activation of CD34+/CD133+ pro-angiogenic myeloid progenitor cells, thus leading to less pro-inflammatory circulating myeloid lineage cells in the blood [42, 43].

Furthermore, the reversal of immune pathways only by the addition of macitentan, but not PDE5i alone, indicates a unique immunomodulatory activity of each drug. The response to PDE5i and macitentan in combination corresponds with recent findings presented in the clinical study to compare the efficacy and safety of macitentan and tadalafil monotherapies with the corresponding fixed-dose combination therapy in subjects with PAH (A DUE) [44], which showed significant improvements in outcomes in patients with combination PDE5i and macitentan therapy, compared with monotherapy. However, given that macitentan was administered in combination with PDE5i here, future work will be necessary to decipher the precise immunomodulatory role of macitentan alone.

At a molecular level, via GSEA and EDDY, we identified monocyte alterations in PAH [45]. Perhaps most intriguingly, we identified reversal of the pathway “glucagon signalling in metabolic regulation” in monocytes after therapy. This pathway carries numerous G-protein subunit genes, such as GNG5, GNG11 and GNB2, that are linked to glucagon and glucagon-like peptide (GLP)-1 signalling, but have never previously been linked to PAH, to our knowledge; however, increasingly, they are being targeted by GLP-1 agonists to improve cardiovascular mortality [46]. Pharmacologic GLP-1 activation in preclinical models of PAH have been explored [47–49], and a recent clinical trial has indicated the haemodynamic benefits of such therapy for patients with PAH and those with chronic thromboembolic pulmonary hypertension [50]. Furthermore, GLP-1 agonists harbour anti-inflammatory properties [51], which may disrupt monocyte recruitment and activation to the perivascular space in the setting of pulmonary hypertension. Our findings now set the stage for future studies to define whether glucagon signalling alterations in immune cells in PAH may be a key cellular target for GLP-1 agonism.

The limitations of this study should be acknowledged. First, our study included a modest sample size, which could limit the full generalisability of our findings across all PAH patients and across all therapies of the same vasodilator classes. While we would have preferred to recruit more participants, our ability to offer sequential single and dual combination vasodilator therapy in a single PAH patient became ethically challenging as the study progressed, given the advancing standard of care to initiate immediate dual combination therapy, based on the results of the AMBITION trial [20]. Nonetheless, given the longitudinal nature of the study, whereby each patient could be used as their own control, and the multimodal analytic pipeline available for each individual subject, the results still hold substantial merit, especially considering the granularity of the findings and the now unique nature of molecular profiling across longitudinal untreated and single and dual vasodilator therapy contexts. Second, although we identified major inflammatory gene networks that were reversed with vasodilator use, future work will be needed to determine if these immune changes constitute causative pathogenic mechanisms in PAH [52].

In sum, we leveraged a systems-biology approach by combining DGE/GSEA and EDDY analyses of scRNA-seq data to define immune pathways dysregulated in the PBMCs of PAH patients and to identify pathways that were reversed by vasodilator use. These results offer new insights into the inflammatory mechanisms involved in PAH and emphasise the utility of PBMCs in predicting the pro-inflammatory processes in the pulmonary vasculature. Our findings now serve as a foundation for clinical tools of scRNA-seq for the diagnosis and monitoring of disease progression in real time.

Supplementary material