Genetic determinants of risk in pulmonary arterial hypertension: international genome-wide association studies and meta-analysis

1Evidence before this study We searched PubMed for research articles published in English before Aug 23, 2018, with search terms including “pulmonary arterial hypertension”, “genetics”, and “GWAS”. Rare genetic variation, primarily in genes associated with transforming growth factor-β family members, including BMPR2, but also in the transcription factor SOX17, is known to cause pulmonary arterial hypertension. However, little is known about the contribution of common variation to this disorder. Additionally, both pulmonary vascular endothelial dysfunction and altered immune and inflammatory signalling are observed in pulmonary arterial hypertension, but the underlying genetic mechanisms are poorly characterised. Added value of this study To our knowledge, this is the largest genetic analysis of pulmonary arterial hypertension to date, comprising more than 2000 patients with pulmonary arterial hypertension from four international cohorts. We identified one locus near SOX17 and another in HLA-DPA1 and HLA-DPB1 that reached genome-wide significance, and we cross-validated these loci by meta-analysis. The SOX17 locus includes two independent signals, both of which identify enhancer regions that specifically regulate the expression of SOX17, which is essential for pulmonary vascular development. Allelic variation at HLA-DPB1 is associated with clinical outcomes, specifically survival, with more than two-thirds of patients harbouring genotypes associated with the poorest outcomes. Implications of all the available evidence Common variation near SOX17 is a risk factor for pulmonary arterial hypertension and dysregulation of SOX17 might be more common in pulmonary arterial hypertension than the occurrence of rare variants suggests. Further studies are needed to define whether HLA typing or rs2856830 genotyping improves risk stratification in clinical practice and in clinical trials.

2Evidence before this study

3We searched PubMed for research articles published in English before Aug 23, 2018, with search terms including “pulmonary arterial hypertension”, “genetics”, and “GWAS”. Rare genetic variation, primarily in genes associated with transforming growth factor-β family members, including BMPR2, but also in the transcription factor SOX17, is known to cause pulmonary arterial hypertension. However, little is known about the contribution of common variation to this disorder. Additionally, both pulmonary vascular endothelial dysfunction and altered immune and inflammatory signalling are observed in pulmonary arterial hypertension, but the underlying genetic mechanisms are poorly characterised.

4Added value of this study

5To our knowledge, this is the largest genetic analysis of pulmonary arterial hypertension to date, comprising more than 2000 patients with pulmonary arterial hypertension from four international cohorts. We identified one locus near SOX17 and another in HLA-DPA1 and HLA-DPB1 that reached genome-wide significance, and we cross-validated these loci by meta-analysis. The SOX17 locus includes two independent signals, both of which identify enhancer regions that specifically regulate the expression of SOX17, which is essential for pulmonary vascular development. Allelic variation at HLA-DPB1 is associated with clinical outcomes, specifically survival, with more than two-thirds of patients harbouring genotypes associated with the poorest outcomes.

6Implications of all the available evidence

7Common variation near SOX17 is a risk factor for pulmonary arterial hypertension and dysregulation of SOX17 might be more common in pulmonary arterial hypertension than the occurrence of rare variants suggests. Further studies are needed to define whether HLA typing or rs2856830 genotyping improves risk stratification in clinical practice and in clinical trials.

8Pulmonary arterial hypertension refers to an uncommon but devastating disorder characterised by obliterative pulmonary vascular remodelling, leading to a progressive increase in pulmonary vascular resistance and right heart failure. Annual mortality for idiopathic and heritable pulmonary arterial hypertension remains around 10%, despite the use of modern therapies., The high mortality partly reflects the limited effect of licensed treatments on the underlying pulmonary vascular pathology, which includes vascular smooth muscle and fibroblast hyperplasia, endothelial cell proliferation, and inflammation. Substantial variation between patients in their response to available treatments highlights underlying and inadequately characterised heterogeneity in the causes of pulmonary arterial hypertension.

9Recent gene sequencing studies,, have revealed rare mutations in several genes, including BMPR2, genes encoding potassium channels, and most recently the transcription factor SOX17. Rare genetic variation is associated with both the risk of developing pulmonary arterial hypertension and survival, and it is found in up to 25% of patients with pulmonary arterial hypertension. In the majority of patients with pulmonary arterial hypertension, the extent of genetic contribution, including that attributable to common variation, remains largely unknown., Therefore, we aimed to test for genome-wide association for pulmonary arterial hypertension in large international cohorts and assess the contribution of associated regions to patient outcomes (panel). This is the first report of the associations found at SOX17 and HLA-DPA1 and HLA-DPB1 (collectively referred to as HLA-DPA1/DPB1 in this Article). Genome-wide association study A genetic analysis approach, typically using millions of common variants (eg, single-nucleotide polymorphisms [SNPs]) covering the genome, to test whether an allele of a genetic variant is associated with a disease or trait, or the levels of a continuous trait of interest. Common variant An SNP for which the frequency of the less frequent allele is at least 5% in a given population. Typically, a common variant has subtle biological effects, as opposed to rare variants (also known as mutations), which can cause diseases or extreme phenotypes. Genetic locus A position on the genome defined by the chromosome number and the genetic distance in centimorgans (cM) or physical distance in base pairs (bps) on the chromosome. A locus can refer to a gene or a non-coding region of varying length (eg, from one hundred to millions of bps). Credible set A set of variants that is statistically likely (eg, with 99% probability) to contain the causal variant for the disease or trait of interest at a genetic locus.

10Genome-wide association study

11A genetic analysis approach, typically using millions of common variants (eg, single-nucleotide polymorphisms [SNPs]) covering the genome, to test whether an allele of a genetic variant is associated with a disease or trait, or the levels of a continuous trait of interest.

12Common variant

13An SNP for which the frequency of the less frequent allele is at least 5% in a given population. Typically, a common variant has subtle biological effects, as opposed to rare variants (also known as mutations), which can cause diseases or extreme phenotypes.

14Genetic locus

15A position on the genome defined by the chromosome number and the genetic distance in centimorgans (cM) or physical distance in base pairs (bps) on the chromosome. A locus can refer to a gene or a non-coding region of varying length (eg, from one hundred to millions of bps).

16Credible set

17A set of variants that is statistically likely (eg, with 99% probability) to contain the causal variant for the disease or trait of interest at a genetic locus.

18We did two genome-wide association studies and a meta-analysis on pulmonary arterial hypertension. Pulmonary arterial hypertension was defined by haemodynamic criteria according to international guidelines. Unrelated individuals with idiopathic, heritable, or anorexigen-associated pulmonary arterial hypertension were included. Individuals with evidence of other known causes of pulmonary arterial hypertension were excluded; therefore, no patients were known to have pulmonary arterial hypertension associated with clinically diagnosable autoimmune diseases (appendix pp 2–3). All enrolled individuals provided written informed consent from their respective institutions or were included as anonymous controls under the DNA databank at Vanderbilt University, BioVU, opt-out policy (appendix p 2).

19Given the rarity of pulmonary arterial hypertension, four studies were used for the analyses. In the UK National Institute for Health Research BioResource (NIHRBR) Rare Diseases study, whole-genome sequencing (Illumina, San Diego, CA, USA; mean depth of around 35×; appendix p 2) was done in 5895 individuals of European descent, each with a rare disorder from 16 categories or their unaffected relatives, and 847 had pulmonary arterial hypertension (appendix p 4). The concept of this study was to sequence patients with rare diseases to identify genetic influences on the pathogenesis of one rare disorder using the other rare diseases as controls, assuming that distinct rare diseases are highly unlikely to share common genetic mechanisms. This assumption was tested by repeating analyses excluding each major control group (appendix p 6).

20Three studies used genome-wide genotyping arrays: the US National Biological Sample and Data Repository for Pulmonary Arterial Hypertension (also known as PAH Biobank [PAHB]) study included 694 individuals with pulmonary arterial hypertension and 1560 controls ascertained for a large pharmacogenomic study at Vanderbilt University (Nashville, TN, USA); the Pulmonary Hypertension Allele-Associated Risk (PHAAR) study included 269 individuals with pulmonary arterial hypertension and 1068 population-based controls from France; and the British Heart Foundation Pulmonary Arterial Hypertension (BHFPAH) study consisted of 275 individuals with pulmonary arterial hypertension and 1983 population-based controls from several European countries (appendix p 12). All genotyping studies were imputed (appendix p 12), and single-nucleotide polymorphisms (SNPs) with good imputation quality (r2≥0·3) were taken forward for testing. Individuals from NIHRBR, PHAAR, and BHFPAH were tested for relatedness to prevent inclusion of the same or related individuals across studies. Other quality-control steps are detailed in the appendix (pp 4, 5, 12).

21We used logistic regression to test single-marker variants for genetic association with a diagnosis of pulmonary arterial hypertension assuming a log-additive genetic model and adjusting for sex, read length chemistry (NIHRBR only), and population structure using the first four (NIHRBR and PHAAR), three (PAHB), or ten (BHFPAH) principal components. We calculated the genomic inflation factor, which was verified to be between 1 and 1·05 for each study.

22We used two independent sets for discovery: whole-genome sequencing data from NIHRBR (n=5895, including 847 pulmonary arterial hypertension cases); and meta-analysis of genotyping studies PAHB, PHAAR, and BHFPAH (n=5849, including 1238 pulmonary arterial hypertension cases). We cross-validated findings and confirmed loci in a meta-analysis of all four studies using the inverse variance-weighted fixed-effects approach (which maximises power for discovery studies), implemented in the GWAMA software tool. Random-effects meta-analysis was subsequently applied to estimate generalisability of the results to different populations. We did a conditional analysis including the lead variant in each locus as a covariate to test for independent distinct signals reaching p<5×10–8.

23We used LDlink to assess linkage disequilibrium of variants in all European populations from the 1000 Genomes Project. Credible sets of variants considered 99% likely to include the functional causal variants were calculated by summing ranked posterior probabilities (appendix p 6).

24The locus near SOX17 was assessed against publicly available functional annotation datasets (including ENCODE, Factorbook Motifs, and Blueprint). The locus was investigated using CRISPR-mediated repression in human pulmonary artery endothelial cells (hPAECs; PromoCell GmbH, Heidelberg, Germany) by transduction with a lentivirus containing a plasmid encoding the nuclease-deficient Cas9 (dCas9) fused to the repressor KRAB and a 20 bp guide RNA (appendix pp 6–7). Cells were harvested following blasticidin selection, and the expression of SOX17 as well as neighbouring MRPL15 and TMEM68 was assessed by quantitative PCR.

25In-vitro enhancer activity of the loci and variants near SOX17 was investigated using a luciferase reporter assay. Specifically, genomic DNA was isolated from endothelial progenitor cells (also known as blood outgrowth endothelial cells) derived from a patient with pulmonary arterial hypertension who was heterozygous for the lead SNP at SOX17 and used to clone 100 bp putative enhancer regions containing the SOX17 pulmonary arterial hypertension variants. The cloned products were inserted into a luciferase reporter plasmid, which was subsequently used for transformation of stable bacteria. Picking various bacterial colonies allowed for isolation of luciferase reporter plasmids containing genomic DNA inserts differing only by the allele of the SNP of interest. Reporter plasmids were transfected into hPAECs by electroporation, and luciferase activity was measured to quantify the enhancer function of the inserts with the relevant haplotype.

26Loci associated with pulmonary arterial hypertension were tested for associations with clinical variables (appendix pp 17–18). All-cause mortality was the primary endpoint in survival analyses using Kaplan-Meier estimates and Cox regression in the survival package in R, version 3.3.0. Survival was calculated from diagnosis to date of death or censoring (Oct 31, 2016, for NIHRBR; Aug 1, 2017, for PAHB; Sept 27, 2017, for PHAAR; Oct 12, 2017, for BHFPAH), with left truncation using date of genetic consent, and patients were censored at lung or heart-and-lung transplantation. Age and sex were covariates to correct for their association with prognosis. NIHRBR and PAHB were analysed separately, and PHAAR and BHFPAH were combined before analysis because of their smaller sample sizes. Cox regression results from these three analyses were then meta-analysed using the default random-effects model restricted maximum-likelihood estimator method implemented in the metafor package in R, version 3.3.0. All cohorts were combined for Kaplan-Meier analysis. We did sensitivity analyses excluding pathogenic BMPR2 variant carriers, all pathogenic rare variant carriers, and patients diagnosed in previous decades who might have been exposed to different treatment regimens.

27HLA alleles and amino acids totalling 1873 features were determined by imputation from genotyped and high-quality imputed variants in the HLA region using the SNP2HLA software and the type 1 diabetes genetics consortium reference database. HLA alleles and amino acids were tested for association with the novel lead variants or case-control status by χ2 test with false discovery rate correction.

28The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. CJR, KB, MB, MHa, LSo, MG, MWP, CH, AA, KBH, JHK, MK, AU, LH, JA, EMS, and SGr had access to raw data for analyses. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

29In two separate GWAS discovery analyses (figure 1), we identified two loci associated with pulmonary arterial hypertension reaching genome-wide significance (p<5 × 10–8; table 1; appendix p 23). One locus was 100–200 kb upstream of the transcription factor SOX17. A second locus was within HLA-DPA1/DPB1, which encodes the MHC class II DP α and β chains.

30Both the SOX17 and HLA-DPA1/DPB1 loci reached genome-wide significance in the discovery analyses; our cross-validation strategy confirmed that the same alleles were more frequent in pulmonary arterial hypertension than in other disease or population controls in both analyses (table 1). The genome-wide meta-analysis of all four studies confirmed their associations with pulmonary arterial hypertension (rs2856830, odds ratio 1·56 [95% CI 1·42–1·71], p=7·65 × 10–20 for HLA-DPA1/DPB1; rs10103692, 1·80 [1·55–2·08], p=5·13 × 10–15 for SOX17; table 1; figure 2; appendix p 13). We detected no further loci at genome-wide significance. Allele frequencies in the different control groups were similar between studies and to non-Finnish Europeans in the public database gnomAD (table 1).

31The conditional analysis confirmed that the HLA-DPA1/DPB1 locus contained a single signal of association, but showed that the SOX17 locus was composed of two independent signals; signal 1 is 100–103 kb upstream of SOX17 (pconditional=9·82 × 10–9) and signal 2 is 106–200 kb upstream of SOX17 (pconditional=4·16 × 10–11; figure 2; appendix p 15). The lead SNPs for the two signals in the SOX17 locus were rs13266183 (for signal 1, odds ratio 1·36 [1·25–1·48], p=1·69 × 10–12) and rs10103692 (signal 2). A Bayesian credible set analysis to narrow the variants in these loci to those 99% likely to be causal (appendix p 15) showed that the HLA-DPA1/DPB1 locus included nine SNPs (all p<9·1 × 10–18), SOX17 signal 1 included four SNPs 100–103 kb upstream of SOX17 (all p<3·3 × 10–8), and SOX17 signal 2 included 31 SNPs 106–142 kb upstream of SOX17 (all p<5·7 × 10–10).

32Previous studies have reported the association of variants near CBLN2, and PDE1A and DNAJC10 with pulmonary arterial hypertension., These common variant signals showed no association with pulmonary arterial hypertension in the combined NIHRBR, PAHB, and BHFPAH cohorts (p=0·17 for CBLN2 and p=0·24 for PDE1A and DNAJC10; appendix p 14). Sensitivity analyses excluding pathogenic BMPR2 variant carriers, all pathogenic rare variant carriers, and controls from different disease groups yielded similar results to the main analyses (appendix p 9).

33To search for evidence of regulatory elements in relevant tissues at SOX17 signal 1 and signal 2, we examined publicly available epigenomic data (including histone modifications; figure 3; appendix p 24). We identified several putative enhancer elements active in both lung tissue and endothelial cells (figure 3). One of these elements (around hg19-chr8:55·270 Mb) contains a cluster of three of four credible variants from SOX17 signal 1 (figure 3). Another (around hg19-chr8:55·252 Mb) contains one credible variant from SOX17 signal 2. Of these variants, rs10958403 in signal 1 and rs765727 in signal 2 overlap a DNase I hypersensitivity signal, which indicates accessible chromatin (allowing binding of transcription factors), detected in hPAECs (figure 3).

34To study the effects of the pulmonary arterial hypertension risk variants on the putative enhancers defined by the epigenomic signals, we developed reporter constructs containing 100 bp of the regions containing either the risk allele or non-risk alleles at each of the four SNPs using genomic DNA from a patient heterozygous for both SOX17 signals. A haplotype-specific reporter assay in hPAECs confirmed that the regions containing either rs10958403 or rs765727 exhibited enhancer activity (between threefold and sixfold induction of luciferase:Renilla ratio, p<0·0001), whereas constructs containing rs12674755 or rs12677277 had no effect compared with the empty vector control (figure 4). We also observed haplotype-specific activity with the active constructs, which differed only by the alleles at pulmonary arterial hypertension-associated risk variants rs10958403 or rs765727 (both p<0·05; figure 4).

35DNA folding patterns determined by Hi-C data from lung tissue and endothelial cells (human umbilical vein endothelial cells [figure 3] and human microvascular endothelial cells [data not shown]) indicate that the SOX17 pulmonary arterial hypertension locus resides in a defined topologically associated domain in which the only gene found, and thus likely target of any regulatory elements in this region, is SOX17. CRISPR-mediated inhibition of the SOX17 signal 1 region in hPAECs resulted in selective downregulation of SOX17 expression but not the expression of neighbouring genes MRPL15 and TMEM68, suggesting that the enhancers in this locus specifically regulate SOX17 (figure 4; appendix p 26).

36We investigated whether the HLA-DPA1/DPB1 and SOX17 variants affect clinical outcomes in pulmonary arterial hypertension, specifically all-cause mortality. The HLA-DPA1/DPB1 rs2856830 genotype, but not the SOX17 locus, was strongly associated with survival (figure 5). Median survival from diagnosis in patients with pulmonary arterial hypertension with the C/C homozygous genotype was double (13·50 years [95% CI 12·07 to >13·50]) that of those with the T/T genotype (6·97 years [6·02–8·05]). Cox regression survival analyses showed that the rs2856830 T/T genotype conferred an increased annual risk of death in pulmonary arterial hypertension (hazard ratio [HR] 1·94 [95% CI 1·08–3·51]; figure 5).

37Sensitivity analyses excluding pathogenic BMPR2 variant carriers, all pathogenic rare variant carriers, and patients diagnosed in previous decades who might have been exposed to different treatment regimens gave results similar to the main analyses (appendix p 27).

38We tested both loci for association with other clinical variables, including disease severity measures and comorbidities (appendix pp 17–18). The C allele at HLA-DPA1/DPB1 lead SNP rs2856830 was associated with younger age at diagnosis (figure 5), with C/C homozygotes presenting a decade earlier than T/T homozygotes (appendix p 17). The rs2856830 genotype was not associated with vasoresponder status.

39The HLA-DPA1/DPB1 locus included a missense variant rs1042140 in HLA-DPB1 reaching genome-wide significance (table 1) in partial linkage disequilibrium (r2=0·45 with lead rs2856830 in Europeans). The SNP rs1042140 determines a glutamic acid (Glu69) or a lysine at amino acid residue 69. To determine specific HLA alleles associated with the lead variant, rs2856830, we imputed HLA types from the genotype data. These types are represented by digit codes, where the first two digits represent related groups of similar alleles (eg, DPB1*02), and four digits represent specific proteins with distinct amino acid sequences (eg, DPB1*02:01). We found that the pulmonary arterial hypertension-enriched C allele of rs2856830 was associated with HLA-DPB1*02:01/02:02/16:01 (all p<1 × 10–9 after false discovery rate correction; table 2; appendix p 19), which all contain the Glu69 residue. The most numerous DPB1*02:01 and DPB1*04:01 alleles were associated with survival in patients with pulmonary arterial hypertension (HR 0·70 [95% CI 0·49–1·00] for DPB1*02:01 and 1·33 [1·04–1·70] for DPB1*04:01; appendix p 21).

40The risk alleles at both signals within the SOX17 locus are common (risk allele frequencies are 74% for rs13266183-C and 92% for rs9298503-C), such that 1230 (59%) of 2085 patients with pulmonary arterial hypertension were homozygous for the risk allele at both SOX17 SNPs, compared with 4443 (46%) of 9659 controls.

41The alleles at HLA-DPB1 associated with the poorest outcomes are also common (risk allele frequency of 86% for rs2856830-T), such that 1432 (69%) of 2085 patients with pulmonary arterial hypertension had the T/T genotype associated with the poorest outcomes and 1975 (95%) of 2085 patients had at least one T allele.

42Through a meta-analysis of 11 744 individuals, we have established loci at an enhancer upstream of SOX17 and at HLA-DPA1/DPB1 associated with pulmonary arterial hypertension disease risk. Common genetic variants in the enhancer region of SOX17 are biologically plausible candidates for susceptibility to pulmonary vascular disease. Polymorphic variation at the HLA-DPA1/DPB1 locus is strongly associated with both the age at diagnosis and prognosis in pulmonary arterial hypertension.

43Both in-silico and experimental analyses of the common variants upstream of the SOX17 gene suggest that they affect susceptibility to pulmonary arterial hypertension through regulation of SOX17 expression. We provided direct evidence that inhibition of SOX17 signal 1 reduced SOX17 expression, and luciferase activity experiments showed a functional variant in both signals. Combined with the in-silico data, it seems highly likely that signal 2 would also be targeting SOX17 because no other gene is present in the topologically associated domain and promoter capture HiC data show that this area associates with the SOX17 promoter. We have recently reported enrichment and familial segregation in pulmonary arterial hypertension of causal rare deleterious variation in SOX17, implicating this gene in the pathogenesis of pulmonary arterial hypertension. SOX17 is involved in the development of the endoderm,,, vascular endothelium, haemopoietic cells, and cardiomyocytes., SOX17 also determines the endothelial fate of CD34 progenitor cells de-differentiated from fibroblasts. Deletion in the mouse leads to abnormal pulmonary vascular development, poor distal lung perfusion and biventricular hypertrophy. SOX17 is a pro-angiogenic transcription factor and interacts with well established endothelial molecular mediators;, reduction of SOX17 in endothelial cells through Notch activation (which is associated with BMPR2 signalling) restricts angiogenesis. Conversely, vascular endothelial growth factor (VEGF) upregulates SOX17 and, as part of a positive-feedback loop, SOX17 promotes expression of VEGF receptor 2.

44We report that HLA-DPB1 alleles are associated with pulmonary arterial hypertension and have a pivotal role in determining disease progression. The beneficial effect of the C/C genotype at rs2856830 on survival is clinically significant, extending average survival from about 7 to about 14 years, despite no apparent difference in baseline disease severity by standard clinical measures, including haemodynamics and exercise capacity. Patients with the C allele at rs2856830 presented at a significantly younger age than those with the T allele, but the association of the HLA-DPB1 SNP with survival remains significant after correction for both age and sex. The somewhat conflicting observation that the C/C genotype is associated with earlier, more frequent presentation but improved survival compared with the C/T or T/T genotypes perhaps suggests that there could be two different mechanisms involved; one that affects initial disease pathogenesis and another that alters the adaptation to the established disease state. A parallel in pulmonary arterial hypertension is the paradox of female prevalence contrasted with poorer outcomes for male patients, although the mechanisms for this still remain unclear. Further evaluation of this survival association in independent datasets would help to define how clinical HLA typing or rs2856830 genotyping could improve risk stratification in clinical practice and in clinical trials, in which over-representation of the C/C genotype in one treatment group could significantly affect outcomes.

45The mechanism of rs2856830 involvement in pulmonary arterial hypertension is probably through its association with specific HLA-DPB1 alleles. Class II (HLA-DRB1, HLA-DQB1, and HLA-DPB1) antigen-presenting proteins have crucial roles in the adaptive immune response., The HLA-DPB1 alleles associated with rs2856830 (HLA-DPB1*02:01/02:02/16:01) in the current study have also previously been linked to susceptibility to hard metal lung diseases, such as berylliosis., A number of individual amino acid residues in the peptide-binding pockets of the HLA-DPB1 molecule affect its function and T-cell recognition, either by changing peptide antigen binding or the conformation of the peptide-binding groove. HLA-DPB1*02:01/ 02:02/16:01 all contain a glutamate at position 69 and a valine at position 36 that reduce the risk of clinical deterioration. These same residues are essential for T-cell activation and cytokine production in berylliosis., The potential role of this modification in antigen binding, autoimmune response, and vascular damage in pulmonary arterial hypertension demands further investigation.

46To examine whether the associations observed were driven by trans effects of known rare pathogenic variants in pulmonary arterial hypertension, we did sensitivity analyses that demonstrated that the associations were independent of BMPR2 and other rare pathogenic variants. Although the sequencing and array platforms used in this study might perform differently across the genome, the signals detected for each platform remained strong.

47This study has some limitations. The majority of patients studied were prevalent cases, and the association with survival is only marginally significant and based on a relatively small sample size, from multiple studies with different ascertainment criteria. Thus, confirmation of the survival analysis in an independent sample of patients recruited at diagnosis would greatly increase confidence in this finding. Some variants displayed heterogeneity of effects between studies, which is most likely due to the limited sample size in the smaller genotyping studies. Variants in CBLN2 and other loci previously associated with pulmonary arterial hypertension, were not replicated by this study, suggesting that these previous findings were either false positives or only relevant to the specific subpopulations studied.

48We have shown in a rare disorder that common variation can drive significant clinical differences in presentation and outcomes. Furthermore, a common non-coding variant can regulate expression of a gene linked by rare, deleterious mutations to the same disease. HLA-DPB1, and wider immune regulatory pathways, should be considered a priority for patient stratification and investigation of new treatments in pulmonary arterial hypertension. SOX17 is a key endothelial regulator and its dysfunction in pulmonary arterial hypertension might be more common than suggested by the occurrence of rare pathogenic variants in heritable cases.

49Comprehensive affiliations listed in the appendix

50Comprehensive affiliations listed in the appendix