Germline and somatic variants in DNMT3A and other clonal haematopoiesis of indeterminate potential genes contribute to pulmonary arterial hypertension

Abstract

Background and Aims

Multiple germline gene variants promote familial and idiopathic pulmonary arterial hypertension (PAH); however, none are consistently identified in associated PAH with connective tissue disease (APAH-CTD). Moreover, the role of somatic variants in genes mediating clonal haematopoiesis of indeterminate potential (CHIP) in PAH is unknown. Here, somatic and germline DNMT3A variants and CHIP gene variants in PAH were evaluated.

Methods

Exome sequencing (ES) was compared between PAH Biobank participants (n = 1832 European ancestry/2572 total), vs. gnomAD controls (7509 European ancestry/141 456 total). Subsequently, targeted panel sequencing (TPS) of 22 CHIP genes, including DNMT3A, was performed in PAH (n = 1659) vs. controls (n = 3644). Somatic CHIP variants in the UK Biobank using ES (controls = 448 239; PAH = 2559) were also assessed. DNMT3A mRNA expression was measured in peripheral blood mononuclear cells (PBMCs) of patients with scleroderma APAH-CTD (n = 50), idiopathic PAH (n = 30), scleroderma without PAH (n = 19), and healthy controls (n = 41). Hemodynamic were evaluated in haematopoietic Dnmt3a-knockout mice.

Results

Predicted deleterious germline DNMT3A variants were increased in subjects of European ancestry (6/1832) vs. controls (6/7509) (relative risk [RR] = 4.1, P = .018). In the entire PAH Biobank cohort (n = 2572), DNMT3A germline and somatic variants were further enriched (PAH: 1.28% vs. controls: .43%, P = 1.65 × 10⁻¹⁰). Eight DNMT3A mutations (.39%) were likely germline (female/male: 7/1) and 25 (.82%) likely somatic (female/male: 21/4), including 13/33 APAH-CTD participants. TPS identified CHIP in 242 PAH subjects (48% DNMT3A). DNMT3A- and all-CHIP variants were associated with PAH after correcting for age, sex, and age–CHIP interactions (odds ratio [OR]: 25.44, P = 4.50 × 10⁻⁵; OR: 23.35, P = 2.87 × 10⁻⁸, respectively). In the UK Biobank, CHIP mutations were increased in PAH (PAH = 5.35% vs. Control = 3.45%, P ≤ .0001). DNMT3A was reduced in PAH-PBMCs (area under curve [AUC] = .82) (P < .0001). Haematopoietic Dnmt3a-knockout in mice caused inflammatory PAH, which was attenuated by IL-1β antibody therapy.

Conclusions

Germline DNMT3A variants and somatic variants of DNMT3A and CHIP genes increase the risk of PAH, including APAH-CTD, promote inflammation, and constitute potential biomarkers and therapeutic targets.

Structured Graphical Abstract

Structured graphical abstract.

Exome sequencing and targeted panel sequencing of large human biobanks showed that germline and somatic DNMT3A variants and clonal haematopoiesis of indeterminate potential (CHIP) variants are linked to increased risk of pulmonary arterial hypertension (PAH) in patients. Moreover, haematopoietic Dnmt3a depletion in mice is sufficient to promote PAH and inflammation. Thus, epigenetic dysregulation caused by abnormal DNMT3A expression increases inflammation (horizontal red bar), contributing to the development of PAH. This diagram demonstrates the infiltration of immune cells into the pulmonary vasculature, promoting adverse vascular remodelling and promoting PAH. Finally, DNMT3A mRNA expression was decreased in the peripheral blood mononuclear cells(PBMCs) of patients with PAH, suggesting its potential utility as a biomarker.

See the editorial comment for this article ‘From bone marrow to pulmonary arteries: the expanding role of clonal haematopoiesis in PAH’, by S. Fatima  et al., https://doi.org/10.1093/eurheartj/ehaf870.

Translational perspective.

Germline variants in the DNMT3A gene, which regulates DNA methylation, were enriched in pulmonary arterial hypertension (PAH) patients relative to controls. Somatic mutations in DNMT3A, TET2, and the composite of all clonal haematopoiesis of indeterminate potential (CHIP) mutations were associated with a >20-fold increased risk of PAH. There are five findings relevant to translational medicine: (i) This is the first report of germline DNMT3A variants in a human cardiovascular disease. (ii) Germline and somatic DNMT3A variants were also enriched in associated PAH with connective tissue disease, a form of PAH that had lacked a genetic underpinning. (iii) There was an association between somatic CHIP variants and PAH. (iv) DNMT3A mRNA was ubiquitously reduced in peripheral blood mononuclear cells in an independent cohort of patients with PAH, suggesting its potential as a biomarker. (v) Biologic plausibility for DNMT3A variants causing PAH is supported by the demonstration that mice lacking Dnmt3a only in haematopoietic cells spontaneously developed an inflammatory form of PAH which regressed with interleukin-1β blockade. Thus, DNMT3A germline and somatic variants and CHIP variants constitute novel risk factors for PAH and contribute to PAH’s pathophysiology. Human clinical trials would be required to determine whether anti-inflammatory strategies, including interleukin-1β blockade, are safe and effective in PAH patients with DNMT3A or CHIP variants.

Introduction

Pulmonary arterial hypertension (PAH) is an obstructive pulmonary vasculopathy that predominantly affects women, resulting in disability and death from right ventricular failure.¹ PAH is defined by mean pulmonary arterial pressure (mPAP) >20 mmHg, pulmonary vascular resistance (PVR) >2 Wood units, and pulmonary artery wedge pressure (PAWP) ≤15 mmHg,² in the absence of left heart, lung, or thrombo-embolic disease. Most PAH is idiopathic (IPAH, 55%) or associated with inflammatory connective tissue diseases (APAH-CTD, 21%),³ like scleroderma (SSc).^2,4

However, PAH often has a genetic basis and can be familial (FPAH). Most PAH-related genes are autosomal dominant germline mutations inherited with incomplete penetrance,¹ including BMPR2,⁵ the most prevalent of the ∼21 proposed PAH-associated genes. Mutations occur in ∼70% of subjects with FPAH and 10%–20% of IPAH⁶ but are rare in APAH-CTD.⁷

Epigenetic dysregulation, including altered DNA methylation and demethylation, can affect gene expression without changing DNA sequence and contributes to the pathophysiology of PAH.^8,9 Increased methylation of gene promoter regions reduces transcription. The methylome is regulated by methylome eraser genes, like Tet methylcytosine dioxygenase 2 (TET2), and methylome writer genes, like DNMT3A, a DNA methyltransferase that mediates de novo DNA methylation. Pathogenic variants in DNMT3A and TET2 generally lead to aberrant hypo- and hypermethylation, respectively; however, mutations in both genes promote inflammation.^10,11 In PAH, TET2 mutations increase cytokines¹² and result in pan-chromosomal DNA hypermethylation.^12,13  TET2 variants are classified as having moderate evidence for pre-disposing to PAH.⁵

Somatic DNMT3A and TET2 variants cause clonal haematopoiesis of indeterminate potential (CHIP) and are increasingly linked to cardiovascular diseases.^10,14 Unlike embryonic germline variants, CHIP variants are somatic variants that occur later in life, increase with age, only occur in haematopoietic cells, and are not heritable. CHIP may occur at low variant allele frequencies (VAF), with a defining threshold VAF ≥2%.¹⁵ The role of these somatic mutations in PAH, apart from TET2, is unknown.^1,12,16 The current study examines the impact of both germline and somatic variants in DNMT3A in PAH. The only human disease associated with germline DNMT3A variants is Tatton–Brown–Rahman syndrome, a congenital overgrowth syndrome reported in <300 individuals worldwide and not known to be associated with PAH.¹⁶

CHIP variants, including TET2 and DNMT3A, increase the risk of myeloid neoplasms, like myelodysplastic syndromes and acute myeloid leukaemia,^17–19 and cardiovascular diseases, including atherosclerotic coronary artery disease.²⁰ Loss of DNMT3A or TET2 function triggers inflammation by various mechanisms, including eliciting a Type 1 interferon response in macrophages due to mitochondrial DNA damage²¹ and increasing macrophage cytokine production.¹¹ CHIP promotes inflammation in vascular diseases, like atherosclerosis and stroke.^14,22–24

We studied germline and somatic variants in 2572 PAH Biobank participants²⁵ using complementary sequencing techniques: exome sequencing (ES), optimal for the detection of germline variants, and targeted panel sequencing (TPS), optimal for detecting somatic CHIP variants. We also assessed somatic DNMT3A and CHIP variants using ES data from 2559 subjects with PAH in the UK Biobank (UKBB), a cohort comprised of APAH (n = 2114; ICD-10 I27.2) and IPAH or FPAH (n = 445, ICD-10 I27.0). In a third cohort, DNMT3A mRNA expression was assessed as a biomarker in subjects with APAH-CTD, IPAH, SSc without PAH, and controls. To establish biological plausibility, a mouse model, with conditional Dnmt3a depletion in haematopoietic cells, was studied for inflammation and PAH development, followed by IL-1β antibody therapy.

Methods

All supporting data are available in the article and Supplementary data online.

Human studies

Detailed methods are described in the Supplementary data online.

Genetic sequencing study participants

The initial genetic sequencing data cohort included participants from the PAH Biobank (Figure 1), maintained at Cincinnati Children’s Hospital Medical Center (CCHMC). It comprises biological samples, clinical data, and genetic data from over 2800 PAH participants enrolled in 38 North American PH centres. Subsequently, we interrogated the UKBB, which contains de-identified genetic data with accompanying health information and biological samples from ∼500 000 UK participants, including 2559 participants with PAH (Figure 1).

Figure 1.

Flow chart of PAH and UK Biobank participants and controls used for ES and panel sequencing. The PAH Biobank included 2572 PAH participants who underwent ES. Both germline and somatic DNMT3A variants were identified and compared with the prevalence of pathogenic variants in gnomAD control database (n = 141 456 individuals). A subset of the PAH Biobank participants underwent panel sequencing (n = 1659) to more sensitively identify somatic DNMT3A CHIP variants. Data were compared with a combined control cohort (n = 3644; 3401 participants from Vanderbilt University and 243 participants obtained from Queen’s University studies) that underwent identical panel sequencing. ES data were also mined from the UKBB to compare the prevalence of CHIP variants between patients with (n = 2559) and without PAH (n = 448 239).

Ethical approval was granted by IRBs at CCHMC, Queen’s University, and Vanderbilt University, with local IRBs approving Biobank data collection. Samples and data are available to the PAH research community, pending CCHMC IRB approval.

ES and genetic case–control comparisons

The PAH Biobank cohort included 43% IPAH, 48% APAH, 4% FPAH, and 5% other PAH cases. APAH-CTD (n = 722) constituted 28% of the cohort.²⁵ Most participants had adult-onset PAH, with a 3.73:1 female/male sex ratio. Genetically determined ancestries were 72% European, 12% Hispanic, 11% African, and 5% other. All 2572 PAH Biobank cohort participants underwent ES. Deleterious germline variants in DNMT3A detected by ES were initially compared in 1832 European PAH participants vs. 7509 non-Finnish European gnomAD controls using a gene-based, case–control association analysis (Figures 1, 2A and B). A REVEL score >.5 defined predicted deleterious missense (D-MIS) variants. Frameshift, non-sense, and splice variants were classified as likely gene-disrupting (LGD). Premature stop codons expressed in the canonical transcript outside the last 200 bp or last exon were considered loss-of-function variants. We tested for association in D-MIS or LGD. Rare variants were defined as having an allele frequency <.01% across all gnomAD ES samples (European and non-European). We classified variants according to the standard clinical criteria.²⁶ Variants that are pathogenic or likely pathogenic were considered disease-associated. Variants were excluded if they met any of the following criteria: missingness >10%, alternate allele read depth ≤4, alternative allele fraction ≤25%, or genotype quality <90.

Figure 2.

Prevalence of DNMT3A variants in PAH participants from the PAH Biobank using ES: a comparative analysis with the gnomAD database. A) The prevalence of germline DNMT3A variants was compared between 1832 European PAH participants and 7509 non-Finnish, European gnomAD controls (2017 release) using a gene-based, case–control association analysis. Association of D-MIS or D-MIS + LGD variants was investigated. An enrichment of D-MIS + LGD germline DNMT3A variants in the PAH Biobank (6/1832 cases vs. 6/7509 controls) was observed (RR = 4.1, P = .018). B) The association in D-MIS + LGD was primarily due to participants with APAH (4/843 cases: RR = 5.9, P = .013). One of the six PAH patients with a germline variant shown in (A) had FPAH, which is not shown in the graph in (B). C) ES data of the entire PAH Biobank (2572 participants) and whole-genome sequencing data of the gnomAD database (141 456 participants) were analysed for predicted deleterious germline and somatic DNMT3A variants using Integrative Genome Viewer. DNMT3A variants were more common in the PAH Biobank [1.28% (33/2572 participants)] than in the gnomAD version 4 group [0.43% (613/141 456 participants, P = 1.65 × 10⁻¹⁰)]. DNMT3A variants were significantly enriched in non-IPAH participants (which includes those with APAH-CTD; P = 1.26 × 10⁻⁵) but not in IPAH participants (P = .216). D) The age distributions of the gnomAD and PAH databases are displayed. The per cent of the cohort (y-axis) at each age range (x-axis) is shown

Subsequently, ES data from the entire PAH (2572 participants) and gnomAD GS (141 456 participants) databases were analysed for predicted damaging DNMT3A variants using Integrative Genome Viewer (Illumina, San Diego, CA, USA) (Figures 1, 2C and D). Nearly 100% of cases and controls had >10× sequencing coverage across the DNMT3A gene, and 90% had >15× coverage across most targeted regions.¹² The average read depth was ∼50×. Variants with a VAF ≤25% or in known CHIP hotspot regions were considered somatic. Truncating and splicing variants with a VAF >25% were considered germline. Data are available as BAM files and on DNANexus and can be accessed through consultation with W.C.N.

We also assessed CHIP and DNMT3A variants in the UK Biobank (UKBB; Supplementary data online, Table S1A) using their ES dataset, as described.²⁷ Subjects were classified based on ICD10 codes as IPAH or FPAH (I27.0, n = 445), APAH (I27.2, n = 2114), or controls (n = 448 293). Both prevalent and incident PAH cases were included.

TPS and case–control comparisons

ES has good specificity for identified CHIP variants but poor sensitivity for low VAF CHIP variants. Therefore, we performed TPS on 1659 samples from the PAH biobank. TPS was performed at Vanderbilt University Medical Center (VUMC) using a panel covering the most frequently involved CHIP genes (read depth ∼700×)²⁸ (Figure 1). CHIP variants with a VAF ≥.02 were identified using standard methodology.¹⁵ A pooled control cohort (n = 3644) was comprised of (i) 201 individuals from Queen’s Genomics Lab at the Ongwanada Autism registry who were older relatives of children with neurodevelopmental disorders, (ii) 42 randomly selected community controls from a prior CHIP study,²⁹ and (iii) 3401 individuals from the VUMC BioVu biorepository, sequenced as controls for a different study, based solely on the absence of chronic kidney disease.³⁰ The control cohort was sequenced on the same platform, and CHIP variants were identified using the same method as PAH samples. The TPS control group is, relative to the PAH group, significantly older, male predominant, includes a higher ratio of white/non-white people, and has significantly more systemic hypertension and type 2 diabetes mellitus (see Supplementary data online, Table S1B).

Following a stepwise method that combines filtering based on sequencing metrics, variant annotation, and population-based associations, we could accurately distinguish CHIP from false positives.²⁷ Binomial and age-association testing distinguished CHIP from possible germline variants, as described.²⁷

Statistical analysis of human data

Statistics were performed using R, version 4.4.2.³¹ Values for all human studies are stated as mean with 95% confidence interval (CI). A corrected P < .05 is considered statistically significant. The prevalence of CHIP in PAH cases compared with controls was evaluated using two-sided proportion testing and logistic regression. PAH was modelled by CHIP status (1 = CHIP, 0 = No CHIP), Age (Numeric Variable), and Sex (M: male, F: female). The predictors and interactions between each of these variables were tested and removed from the model if they were not significantly contributing to the logistic regression, with a cut-off of P < .05 The final model had the form: PAH∼CHIP + Age + Sex + CHIP:Age. To optimize our model’s sensitivity and specificity, we adjusted our classification threshold to .357 using Youden’s J Index,³² providing a balanced predictive accuracy of 66%. The Kruskal–Wallis test was used when comparing three or more independent groups that were not normally distributed. We utilized the Kaplan–Meier estimator for survival analysis and analysed the response to the acute vasodilator test² and haemodynamics to compare the predictive outcome based on the presence of CHIP (see Supplementary data online, Figures S2 and S3, Table S3). Refer to the ‘supplemental raw CHIP calls’ Excel file for a detailed version of the panel sequencing data.

Microarray and Gene Expression

The Gene Expression Omnibus was used to re-mine independent transcriptome data (GSE33463) from peripheral blood mononuclear cells (PBMCs) of participants at Johns Hopkins University with scleroderma-associated PAH (SSc-PAH, n = 50; 60 ± 13 years; females 79%), IPAH (n = 30; 50 ± 10 years; females 83%), SSc without PAH (n = 19), and healthy controls (n = 41; 45 ± 12 years; females 83%).³³ These adult-onset PAH cases had genetically determined ancestries: 82.1% European, 14.3% African, and 3.4% other. Receiver operating characteristic (ROC) analysis was performed, examining the expression of DNMT3A transcript variants 3 (NM_022552) and 4 (NM_175630.1).

Animal studies

Rodent experiments were conducted following the Canadian Council on Animal Care regulations approved by Queen’s University Animal Care Committee. Conditional, homozygous, and heterozygous haematopoietic Dnmt3a-knockout mice (Dnmt3a^−/−, Dnmt3a^+/−, respectively) were generated by crossing parental Dnmt3a floxed³⁴ and B6.Cg-Commd10^{Tg(Vav1−icre)A2Kio}/J mice from Jackson Laboratory (see Supplementary data online, Figure S7). We evaluated haemodynamics at ages 4.5 and 9 months with no additional stimuli, and in a cohort of mice studied at 4.5 months exposed to 3 weeks of hypoxia (10% oxygen), as a potential primer of PAH, followed by 3 weeks of normoxia before cardiac catheterization. A group of age-matched Dnmt3a  ^−/− mice were treated with a mouse IL-1β antibody (10 mg/kg/week or immunoglobulin G2a, i.p. for 6 weeks) (Novartis Pharma AG, Basel, Switzerland).¹² Heart function was assessed using ultrasound, as described.³⁵ Right ventricle (RV) and left ventricle (LV) pressure–volume loops were acquired using micromanometer-tipped catheters, in closed and open-chest conditions, respectively, as described.³⁶ Blood, lung, RV, LV, spleen, and thymus were used for cytokine studies, immunohistochemistry, flow cytometry, and western blotting.

Statistical analysis of rodent data

For murine studies, operators were blinded to genotype and experimental group. Results are presented as mean with 95% CI. Normally distributed data were analysed using ANOVA or Student’s t-test in Prism (GraphPad Software, La Jolla, CA, USA); otherwise, non-parametric testing was employed. A P-value <.05 was considered statistically significant. Mice of both sexes were used, but values were pooled due to the absence of sex-related differences.

Results

Predicted deleterious germline and somatic DNMT3A variants are increased in PAH using ES

Among European-ancestry PAH cases, we observed enrichment of D-MIS + LGD variants for DNMT3A (6/1832 PAH vs. 6/7509 controls; relative risk [RR] = 4.1, P = .018; Figure 2A) primarily due to participants with APAH (4/843 cases vs. 3/7509 controls; RR = 5.9, P = .013; Figure 2B). In the entire PAH Biobank (n = 2572), we identified 33 DNMT3A variants (8 germline; female:male ratio 7/1, Table 1; and 25 predicted somatic variants; female:male ratio 21/4, Table 2). Two-thirds (21/33) of participants with DNMT3A variants had APAH, and 13/33 had APAH-CTD. One subject (15-051, Table 2) had not only a somatic DNMT3A variant but also germline variants of BMPR2 and SMAD9. No other subject with a DNMT3A variant had abnormalities of known PAH genes.

Table 1.

Rare, predicted deleterious likely germline variants in candidate PAH risk gene DNMT3A among 2572 PAH Biobank cases

Patient ID	Nucleotide change	Amino acid change	MAF, gnomAD ES	VAF	CADD Phred	REVEL score	PAH class	Ancestry	Sex	Age enrolment	Age onset	mPAP (mmHg)	PAWP (mmHg)	CO Fick (L/min)	PVR (Wood units)	Mean SAP (mmHg)
1. 04-046	c.1471G>T	p.Glu491a		38%	38		IPAH	Hispanic	F	76	76	36	8	4.6	6.09	NA
2. 29-005	c.2083-1G>C	Splicing		31%	25		IPAH	EUR	F	67	66	25	7	NA	NA	97
3. 07-060	c.875T>C	p.Ile292Thr	4.08 × 10−6	65%	24	.72	FPAH	EUR	F	34	34	65	NA	3.4	NA	93
4. 13-025b	c.1937–2A>G	Splicing		66%	24		APAH-Porto	EUR	M	67	66	36	13	NA	NA	NA
5. 03-053	c.893G>A	p.Gly298Glu	4.07 × 10−6	39%	30	.87	APAH-CTD	EUR	F	70	67	43	5	2.2	17.27	94
6. 07-052	c.2032C>T	p.Gln678a		40%	40		APAH-CTD	EUR	F	78	74	26	14	3.9	3.08	115
7. 07-024	c.2655G>C	p.Arg885Ser		40%	27	.9	APAH-CTD	EUR	F	80	73	29	5	NA	NA	NA
8. 06-101	c.656A>G	p.Lys219Arg		43%	25	.53	APAH-CHD	African	F	7	0	37	5	4.3	7.44	87
Germline DNMT3A variant carriers, mean (95% CI)									7:1	60 (38 to 81)	57 (35 to 79)	37 (26 to 48)	8 (5 to 12)	3.7 (2.5 to 4.8)	8.47 (−1.31 to 18.3)	97 (84 to 110)
Germline DNMT3A variant carriers (n)										8	8	8	7	5	4	5
Cohort APAH + IPAH excluding all DNMT3A and TET2 carriers, mean (95% CI)										52 (52 to 53)	47 (47 to 48)	50 (49 to 50)	10 (10 to 10)	4.5 (4.4 to 4.6)	10.7 (10.3 to 11.0)	89 (88 to 90)
cohort APAH + IPAH excluding all DNMT3A and TET2 carriers (n)										2544	2476	2541	2474	1816	1765	1181
P-value										.075	.047	.006	.172	.314	.465	.238

APAH-CHD, pulmonary arterial hypertension associated with congenital heart disease; APAH-CTD, pulmonary arterial hypertension associated with connective tissue diseases; APAH-Porto, pulmonary arterial hypertension associated with porto-pulmonary hypertension; FPAH, familial pulmonary arterial hypertension; MAF, minor allele frequency; IPAH, idiopathic pulmonary arterial hypertension; PAP, pulmonary artery pressure; PAWP, pulmonary artery wedge pressure; PVR, pulmonary vascular resistance; SAP, systemic arterial pressure; ES, whole exome sequencing. Ancestry was genetically determined.

^a DNMT3A transcript: NM_175629.2.

^bParticipants carry more than one variant in the candidate risk genes. Exceptions to the mosaic pipeline variant filter (MAF and alternate allele fraction); detected as known mutation hotspot in cancer.

^cExceptions to the mosaic pipeline variant filter (MAF or minimum read depth); detection by the germline pipeline with posterior odds >10. Variant filter: allele frequency <.0001 and likely gene-disrupting (stop/gain, frameshift, or canonical splicing) or missense with REVEL score >.5.

Table 2.

Rare, predicted deleterious likely somatic variants in candidate PAH risk gene DNMT3A among 2572 PAH Biobank cases

Patient ID	Nucleotide change	Amino acid change	MAF, gnomAD ES	VAF	CADD Phred	REVEL score	PAH class	Ancestry	Sex	Age at enrolment (years)	Age onset (years)	mPAP (mmHg)	Mean PAWP at rest (mmHg)	Cardiac output, Fick (L/min)	PVR, Fick (Wood units)	Mean SAP (mmHg)
1. 12-147	c.2711C>T	p.Pro904Leu	3.25 × 10−5	29%	29	.94	FPAH	EUR	F	37	36	69	13	4	14	75
2. 21-044	c.2204A>G	p.Tyr735Cys	4.51 × 10−5	38%	27	.9	APAH-Porto	EUR	F	66	65	55	15	6.7	5.97	NA
3. 22-056	c.1903C>T	p.Arg635Trp	5.99 × 10−5	41%	35	.84	APAH-HIV	EUR	F	43	38	48	4	3.3	13.33	NA
4. 03-068	c.1226G>T	p.Trp409Leu		22%	34	.75	IPAH	EUR	F	57	50	50	7	3.8	11.44	97
5. 22-027b	c.2644C>T	p.Arg882Cys	1.00 × 10−4	42%	34	.89	IPAH	EUR	F	67	53	50	6	3.7	11.89	NA
6. 15-051	c.1258A>T	p.Lys420a		14%	38		IPAH	EUR	F	4	4	95	9	NA	NA	61
7. 03-061c	C.976C>T	p.Arg326Cys	4.47 × 10−5	18%	33	.77	IPAH	Hispanic	M	60	58	57	12	6.9	6.52	107
8. 05-202	c.2174-2A>G	splicing	9.37 × 10−6	23%	24		IPAH	EUR	F	59	45	35	14	6.8	3.09	102
9. 12-102	c.2200T>C	p.Phe734Leu		13%	32	.90	FPAH	EUR	M	66	55	42	4	NA	NA	NA
10. 13-067	c.2026C>T	p.Arg676Trp	4.49 × 10−5	13%	35	.91	DTOX	African	F	59	58	45	18	6.8	3.97	NA
11. 12-041	c.2116G>A	p.Gly706Arg		19%	33	.97	DTOX	EUR	M	58	48	50	12	NA	NA	97
12. 05-168b	c.2645G>A	p.Arg882His	2.00 × 10−4	28%	33	.74	APAH-Porto	EUR	M	61	60	66	9	5.1	11.18	98
13. 13-021b	c.2645G>A	p.Arg882His	2.00 × 10−4	28%	33	.74	APAH-CTD	EUR	F	61	61	35	11	6.9	3.48	NA
14. 16-033	c.920C>G	p.Pro307Arg	8.96 × 10−6	13%	27	.97	APAH-CTD	EUR	F	76	75	32	10	NA	NA	72
15. 22-067c	c.1742_1743delinsAT	p.Trp581Tyr	4.09 × 10−6	18%	43		APAH-CTD	Hispanic	F	60	55	35	13	NA	NA	NA
16. 22-013	c.2206C>T	p.Arg736Cys	3.28 × 10−5	20%	34	.92	APAH-CTD	EUR	F	60	49	32	8	7.1	3.38	67
17. 12-180	c.2221G>C	p.Arg741Pro		23%	25	.83	APAH-CTD	EUR	F	69	68	60	12	NA	NA	NA
18. 12-192	c.2402T>C	p.Met801Thr		12%	27	.93	APAH-CTD	EUR	F	72	69	55	6	4.0	12.31	99
19. 11-057c	c.2645G>A	p.Arg882His	.0002	19%	33	.74	APAH-CTD	EUR	F	67	66	60	11	4.9	10	122
20. 28-039	c.2695C>T	p.Arg899Cys	6.54 × 10−5	24%	34	.94	APAH-CTD	EUR	F	66	61	54	7	NA	NA	NA
21. 24-008	c.1258A>T	p.Lys420a		14%	38		APAH-CHD	EUR	F	5	0	49	9	1.8	21.98	47
22. 29-037b	c.2644C>T	p.Arg882Cys	1.00 × 10−4	39%	34	.89	APAH-CHD	EUR	F	48	47	88	16	3.9	18.46	93
23. 08-111c	c.1903C>T	p.Arg635Trp	5.99 × 10−5	10%	35	.84	APAH	Hispanic	F	62	49	NA	NA	NA	NA	NA
24. 17-004c	c.1937-2A>G	splicing		15%	24		APAH	EUR	F	66	62	NA	NA	NA	NA	NA
25. 20-010	c.2204A>G	p.Tyr735Cys	4.51 × 10−5	24%	26	.9	APAH-CTD	Hispanic	F	80	73	29	5	NA	NA	NA
Somatic DNMT3A variant carriers, mean (95% CI)									21:4	57 (50–65)	52 (45–60)	52 (45–59)	10 (8–12)	5.0 (4.1–6.0)	10.1 (6.9–13.2)	87 (75–100)
Somatic DNMT3A variant carriers (n)										25	25	23	23	15	15	13
Cohort APAH + IPAH excluding all DNMT3A and TET2 carriers, mean (95% CI)										52 (52–53)	47 (47–48)	50 (49–50)	10 (10–10)	4.5 (4.4–4.6)	10.7 (10.3–11.0)	89 (88–90)
Cohort APAH + IPAH excluding all DNMT3A and TET2 carriers (n)										2544	2476	2541	2474	1816	1765	1181
P-value										.061	.108	.727	.990	.157	.905	.952

APAH-CHD, pulmonary arterial hypertension associated with congenital heart disease; APAH-CTD, pulmonary arterial hypertension associated with connective tissue diseases; APAH-Porto, pulmonary arterial hypertension associated with porto-pulmonary hypertension; FPAH, familial pulmonary arterial hypertension; MAF, minor allele frequency; IPAH, idiopathic pulmonary arterial hypertension; PAP, pulmonary artery pressure; PAWP, pulmonary artery wedge pressure; PVR, pulmonary vascular resistance; SAP, systemic arterial pressure; ES, whole exome sequencing. Ancestry was genetically determined.

^a DNMT3A transcript: NM_175629.2.

^bParticipants carry more than one variant in the candidate risk genes. Exceptions to the mosaic pipeline variant filter (MAF and alternate allele fraction); detected as known mutation hotspot in cancer.

^cExceptions to the mosaic pipeline variant filter (MAF or minimum read depth); detection by the germline pipeline with posterior odds >10. Variant filter: allele frequency <.0001 and likely gene-disrupting (stop/gain, frameshift, or canonical splicing) or missense with REVEL score >.5.

The only differences between PAH participants with germline DNMT3A variants and non-carriers were the older age-of-onset and reduced mPAP compared with those with a variant (P = .006) (Table 1). Two-dimensional structures of predicted proteins showed that most of the DNMT3A missense variants (germline and somatic) localize to conserved protein domains, with 24/33 variants predicted to affect the DNA methylation domain. Specific mutation loci are provided in Tables 1 and 2. None of the DNMT3A variant participants identified by ES had a positive acute vasodilator test, whereas 13.4% of tested subjects in the PAH biobank were responders (see Supplementary data online, Table S3).

Using the entire PAH Biobank cohort, DNMT3A variants (germline + somatic) were more common [1.28% (33/2572 participants)] vs. gnomAD controls [.43% (613/141 456 participants, P = 1.65 × 10⁻¹⁰) (Figure 2C). DNMT3A variants were significantly enriched in APAH (notably APAH-CTD) (P = 1.26 × 10⁻⁵) but not in IPAH (P = .216) (Figure 2C). APAH-CTD subjects accounted for 14/33 (42%) of combined germline or somatic DNMT3A variants by ES (Table 1). The ages of the PAH and gnomAD cohorts were comparable (Figure 2D).

In the UKBB I27.0 + I27.2 cohort, ES identified 137 CHIP variants (including 65 DNMT3A somatic variants, Supplementary data online, Table S1A). By proportion testing, CHIP variants were increased in PAH (Control = 3.45%; PAH = 5.35%; P ≤ .0001; Figure 3B); however, after multivariable analysis, statistical significance is lost (odds ratio [OR]: 2.35, 95% CI: .25–22.07), due to sample size, older age, and male sex in the PAH group (Figure 3A, Supplementary data online, Table S1A). Interestingly, proportion testing by decade, identified significant enrichment of CHIP in PAH patients aged 50–59 years vs. controls (P = .013), with a similar trend in 60–69 year olds (P = .176; Figure 3C).

Figure 3.

ES data of CHIP prevalence in PAH and control patients in the UKBB I27.0 + I27.2. A-i) The age distribution curves of all control (n = 448 239) and PAH patients (n = 2559) are shown. The median age is 58 years for controls and 64 years for PAH patients. A-ii) The age distribution curves of controls with CHIP (n = 15 487) and PAH patients with CHIP (n = 137) are shown. The median age is 62 years for controls and 65 years for PAH patients. B) The proportion of CHIP in PAH (5.35%) and control patients (3.45%) is shown. A significantly higher proportion of PAH patients had CHIP compared with controls on univariate analysis (P = 2.183 × 10⁻⁷). C) The proportion of CHIP in PAH and control participants by decade is shown. A significantly higher proportion of PAH participants were found to have CHIP variants compared with controls in the 50–59 age range (P = .013), with a similar trend in the 60–69 age range (P = .176). There was no difference in 70–79 year olds (P = .749)

Somatic DNMT3A variants are associated with PAH using TPS

DNMT3A variants were identified in 7% (116/1659) of PAH Biobank participants and 7.3% (73/998) of APAH participants. The median age of the control cohort for TPS (62 years) was 5 years older than that of the PAH cohort (57 years; Figure 4A). We also evaluated CHIP in PAH vs. controls by decades. A significantly higher proportion of PAH participants than controls aged 60–69 years had CHIP variants (P = .037, Figure 4B).

Figure 4.

CHIP and somatic DNMT3A variants are associated with PAH. A) The age distribution curves of 1659 PAH Biobank participants and a pooled control cohort (n = 3644) are shown. The median age of the control cohort used for TPS (62 years) was 5 years older than that of the PAH cohort (57 years, top). CHIP mutations were found at an earlier median age in PAH participants (63.5 years) vs. control participants (70 years, bottom). B) The proportion of CHIP in PAH and control participants by decade is shown. A significantly higher proportion of PAH participants were found to have CHIP variants compared with controls in the 60–69 age range (P = .0387). C) Mutation prevalence by gene in PAH participants is shown. DNMT3A was the most mutated CHIP gene, accounting for 48% of CHIP mutations in PAH participants (116/242), followed by TET2 (19%, 46/242) and PPM1D (12%, 28/242). D) The female:male ratio of mutation prevalence is shown for control and PAH participants. While there was a 3.6:1 female:male ratio for PAH Biobank patients without CHIP, there was a greater (P = .0781) female:male preponderance for participants with DNMT3A variants (5.8:1). There was no female preponderance of DNMT3A variants in control participants (P = .7115). E) Logistic regression was used to model PAH given CHIP status. Covariates included age, sex, and the age–CHIP interaction. The number of PAH participants with each variant is listed under ‘Count of PAH’ of the 1659 participants. A significant association between All CHIP (P = 2.87 × 10⁻⁸), DNMT3A CHIP (P = 4.50 × 10⁻⁵), Non-DNMT3A CHIP (P = 4.18 × 10⁻⁵) and TET2 CHIP (P = .0299) with PAH were identified

DNMT3A was the most frequently mutated CHIP gene in PAH, accounting for 48% of variants (116/242), followed by TET2 (19%, 46/242) and PPM1D (12%, 28/242) (Figure 4C). The predominance of DNMT3A variants was also seen in the control cohort (51%, 314/620) (see Supplementary data online, Figure S1). CHIP variants were found at an earlier median age in PAH (63.5 years) vs. control participants (70 years) (Figure 4A).

Logistic regression analysis identified a significant association between DNMT3A CHIP and PAH vs. controls (OR: 25.44, 95% CI: 5.37–120.40, P < .0001; Figure 4E). All CHIP (OR: 23.35, 95% CI: 7.67–71.03, P < .0001), TET2 CHIP (OR: 13.69, 95% CI: 1.29–145.30, P = .03), and non-DNMT3A CHIP (OR: 26.80, 95% CI: 5.56–129.23, P < .0001) were each significantly associated with PAH (Figure 4E). Specific DNMT3A variant loci are described in Supplementary data online, Table S2.

The overall PAH Biobank female:male ratio is 3.73:1. Interestingly, this sexual dimorphism (female > male) is even stronger in those with somatic DNMT3A variants (5.8:1) compared to those without CHIP (3.6:1; P = .0781; Figure 4D) and stronger still in those with germline DNMT3A variants (7:1) (Table 1). This constitutes a 1.61-fold (somatic vs. no CHIP) and 1.94-fold (germline vs. no CHIP) further enrichment in female sex amongst PAH patients with DNMT3A variants in the PAH Biobank. For unknown reasons, the female-dominant sexual dimorphism seen in all PAH Biobanks was not found in the UKBB (female: male ratio is .89:1; Supplementary data online, Table S1A).

The presence of a DNMT3A variant by TPS did not predict disease severity (see Supplementary data online, Table S2) or survival, whether subjects were under (P = .38; Supplementary data online, Figure S2A) or over (P = .39; Supplementary data online, Figure S2B) age 50 years. The proportion of responders to the acute vasodilator test was similar between participants with and without CHIP in the TPS cohort (P = .2035; Supplementary data online, Figure S3A) or DNMT3A (P = .359; Supplementary data online, Figure S3B), unlike the ES cohort, where those with DNMT3A mutations lacked response (see Supplementary data online, Table S3). Interestingly, we identified an increase in mPAP in patients with TPS called somatic DNMT3A variants compared to those with germline DNMT3A variants (P = .023) (Table 1 and Supplementary data online, Table S2). Neither higher VAF (see Supplementary data online, Figure S4), the presence of multiple variants within DNMT3A (n = 14 subjects), nor the occurrence of multiple mutated CHIP genes (n = 23 subjects) predicted worse haemodynamics (see Supplementary data online, Figure S5).

Subjects with APAH-CTD taking mycophenolate mofetil had higher cardiac output (CO) (P = .014), and those receiving a monoclonal antibody had a trend towards reduced PVR (P = .172), compared with those not on an immunosuppressant (Figure 5A). APAH-CTD subjects taking an immunosuppressant drug had increased CO (P = .012) and lower PVR (P = .035) vs. untreated subjects (Figure 5B). CHIP did not predict response to immunosuppressant treatment (see Supplementary data online, Figure S6). PAH Biobank participants were retrospectively screened for blood disorders. Twelve of 116 participants with DNMT3A CHIP variants had anaemia, 1 had pancytopenia, and none had leukaemia.

Figure 5.

Haemodynamics in PAH Biobank patients with APAH-CTD on immunosuppressants. A) The haemodynamics of patients with APAH-CTD on one of seven immunosuppressant drugs, another unlisted immunosuppressant, or no immunosuppressant, are shown. The Kruskal–Wallis test was not significant for mPAP (P = .1228) or PAWP (P = .9036) so no post hoc testing was conducted. A significantly higher CO is seen in patients taking mycophenolate mofetil compared with patients not on an immunosuppressant drug using Dunn’s test of multiple comparisons (P = .014) following a significant Kruskal–Wallis test. Further, a trend towards reduced PVR is seen in patients receiving a monoclonal antibody compared with patients not on an immunosuppressant drug using Dunn’s test of multiple comparisons (P = .172) following a significant Kruskal–Wallis test. B) The haemodynamics of patients with APAH-CTD on any immunosuppressant drug compared to those without immunosuppressive drugs are shown. All tests were conducted using a Wilcoxon Rank-Sum Test. A significantly higher CO (P = .013) and lower PVR (P = .035) are seen in patients on immunosuppressants compared with those not on immunosuppressant, although the use of these drugs was not randomly assigned

DNMT3A expression is reduced in PAH-PBMCs

DNMT3A variant 3 expression was decreased in 80% of SSc-PAH and 73.3% of IPAH participants (SSc-PAH .93 and IPAH .92 of control expression, respectively, P < .05). DNMT3A variant 4 was decreased in 88% of SSc-PAH and 93.3% of IPAH participants (SSc-PAH .80 and IPAH .77 of control expression, respectively; P < .0001; Figure 6A and B). DNMT3A expression was not altered in SSc participants without PAH. ROC analysis suggests that DNMT3A variants 3 (area under curve [AUC]: .67; P < .002) and 4 (AUC: .82; P < .0001) are putative predictive biomarkers of PAH (Figure 6C and D). The genotypes of these participants are unknown.

Figure 6.

DNMT3A expression is reduced in PBMCs) of participants with PAH in an independent cohort. A and B) The mRNA expression of DNMT3A isoforms 3 and 4 was measured in PBMCs of control, IPAH, SSc-PAH, and scleroderma only (SSc) participants. Expression of other isoforms was not significant. Expression of DNMT3A variants 3 and 4 were decreased in 80% of SSc-PAH and 73.3% of IPAH participants (IPAH 0.92, SSc-PAH 0.93) and 88% of SSc-PAH and 93.3% of IPAH participants (IPAH 0.77, SSc-PAH 0.8), respectively. DNMT3A expression was not altered in SSc participants without PAH vs. control. C and D) ROC analysis on a cohort consisting of 41 healthy controls and 80 PAH participants (IPAH/SSc-PAH) was performed for DNMT3A isoforms 3 and 4. Sensitivity % is shown on the y-axis while the x-axis represents 100% − specificity %. The results indicate that DNMT3A variant 3 (AUC: 0.67, P = .0024) and variant 4 (AUC: 0.82, P < .0001) could serve as potential predictors of PAH

Dnmt3a depletion in hematopoietic cells promotes PAH in mice

We confirmed targeted, haematopoietic cell Dnmt3a deletion (see Supplementary data online, Figure S7). Compared with control mice, 9-month-old Dnmt3a^−/− mice exhibited pulmonary hypertension with reduced RV function (see Supplementary data online, Figure S8). Even at age 4.5 months, mice spontaneously developed PAH, evident by increased right ventricular systolic pressure (RVSP) (P = .0015) and mPAP (P = .0015), shortened pulmonary artery acceleration time (PAAT) (P = .0030), reduced tricuspid annular plane systolic excursion (TAPSE) (P = .0001), and a trend towards reduction in CO (P = .2061; Figure 7A). In Dnmt3a^−/− mice, transient exposure to hypoxia for 3 weeks, followed by a 3-week recovery period pre-catheterization, had elevated RVSP (P = .0132), mPAP (P = .0132), and reduced PAAT (P = .0044), TAPSE (P = .0109), and CO (P = .0107), compared with transient hypoxic Dnmt3a^f/f controls (Figure 7A). No significant differences were observed in LV haemodynamics in Dnmt3a^−/− vs. Dnmt3a^f/f (see Supplementary data online, Figure S9). In heterozygous mice (Dnmt3a^+/−), there was a trend towards increased RVSP (P = .07) and mPAP (P = .07), while PAAT (P = .0015) and TAPSE (P = .0085) were significantly reduced vs. Dnmt3a^f/f (see Supplementary data online, Figure S10).

Figure 7.

Dnmt3a depletion in mice haematopoietic stem cells promotes inflammatory PAH, which is attenuated by a mouse IL-1β antibody. About 4.5-month-old control and Dnmt3a^−/− mice were maintained in normoxic conditions for 6 weeks or were exposed to 3 weeks of hypoxia followed by 3 weeks of normoxia. Another group of Dnmt3a^−/− mice exposed to 3 weeks of hypoxia, followed by 3 weeks of normoxia, was also treated with the IL-1β antibody. A) RVSP and mPAP are elevated in Dnmt3a^−/− mice that were maintained in normoxia (P = .0015 and P = .0015) or underwent a second hit (hypoxia) (P = .0132 and P = .0132, respectively), compared with their controls. PAAT and TAPSE are significantly reduced in Dnmt3a^−/− mice that were kept in normoxia, compared with the normoxic controls (PAAT: P = .0030; TAPSE: P = .0001). They are also reduced in Dnmt3a-knockout mice that underwent a second hit (hypoxia) compared to controls (PAAT: P = .0044, TAPSE: P = .0109). CO showed a trend towards being reduced in the normoxic Dnmt3a^−/− mice, though not statistically significant (P = .2061), and it was significantly reduced in the hypoxic Dnmt3a^−/− mice compared with controls (P = .0107). Treatment of hypoxic Dnmt3a^−/− mice with a mouse IL-1β antibody (green) improved haemodynamic measurements and cardiac function (n = 6–7; RVSP: P = .0117, mPAP: P = .0117, PAAT: P = .0003, TAPSE: P = .0839, CO: P = .1967). B) Histological assessment via haematoxylin and eosin staining was performed on lung tissue slides using small pulmonary arteries <50 um diameter. There is an increase in pulmonary artery medial area (%) in the lungs of haematopoietic Dnmt3a^−/− mice that were maintained in normoxia or exposed to second-hit hypoxia compared with their controls (P < .0001 and P < .0001; n = 3 per group). C) Immunofluorescence and confocal microscopy were used to measure leucocyte infiltration in the lungs of Dnmt3a^−/− mice with PAH compared with controls. CD45, a marker of leucocytes, is shown in green. DAPI, a nuclei stain, is shown in blue. There is a significant increase in the number of CD45⁺ cells in the lungs of the knockout mice that have developed pulmonary hypertension compared with controls (n = 5\ per /group; normoxia: P = .047, hypoxia: P = .0211). D) Flow cytometry was used to investigate the subpopulations of leucocytes seen on confocal microscopy. Macrophages are increased in Dnmt3a^−/− mice (n = 5–8 per group, P = .0092). E) IL-13 was increased in the plasma of Dnmt3a^−/− mice compared with controls (P = .047). *P < .05.

In the 4.5-month-old Dnmt3a^−/− mice, the percentage medial area of small PA (diameter <50 µm) was increased vs. controls, both in normoxic mice (P < .0001, Figure 7B) and in the group that received transient hypoxia (P < .0001; Figure 7B). Total RV collagen was increased in normoxic and hypoxia-accelerated Dnmt3a^−/− mice vs. respective Dnmt3a^f/f controls (P = .05, P = .004, respectively; Supplementary data online, Figure S11).

Dnmt3a^−/− mice had increased pulmonary leucocyte infiltration (P < .05, Figure 7C). Immunofluorescence microscopy (Figure 7C) and flow cytometric analysis of live leucocytes (CD45⁺ cells) from lung-derived, single-cell suspensions (n = 5–8 per group) (Figure 7D) revealed increased leucocytes in Dnmt3a^−/− lungs, with macrophages (F4/80⁺) being predominant (P < .01; Figure 7D). T cells were decreased in the Dnmt3a^−/− lungs (CD3; P = .03) (see Supplementary data online, Figure S12). There was a trend towards an increase in neutrophils (Ly6G⁺; P = .08), but no difference in B cells between groups (CD19; P = .46) (see Supplementary data online, Figure S12). Interleukin-13 (IL-13) was increased in the plasma of Dnmt3a^−/− mice (P = .046; Figure 7E). A complete cytokine profile of Dnmt3a^−/− mice is shown in Supplementary data online, Figure S13.

Dnmt3a^−/− mice exposed to transient hypoxia were treated with a mouse IL-1β antibody or vehicle for 6 weeks. Treatment reduced RVSP (P = .0117) and mPAP (P = .0117), and improved PAAT (P = .0003), also causing a trend towards increased TAPSE (P = .0839) and CO (P = .1967; Figure 7A).

Discussion

This study identifies a role for germline and somatic mutations in DNMT3A in PAH (including APAH-CTD) and has six major findings. First, we show that rare predicted deleterious germline variants of DNMT3A are enriched in PAH Biobank participants. This is the first demonstration of germline variants in DNMT3A in a human cardiovascular disease. Second, somatic DNMT3A variants and CHIP in aggregate are enriched in PAH. Third, using TPS, which is more sensitive than ES for detecting somatic variants, we show that people with DNMT3A or other CHIP variants have an ∼25-fold higher likelihood of having PAH. Interestingly, 41.4% of total DNMT3A variants were in APAH-CTD participants. Likewise, in the UKBB, CHIP variants (measured by ES) were increased in PAH, although this was only significant in univariate analysis. Fourth, DNMT3A expression is reduced in the PBMCs of participants with IPAH and APAH-CTD, highlighting its potential utility as a PAH biomarker. Fifth, in the PAH Biobank, both somatic and germline variants in DNMT3A occur predominantly in females. Sixth, mice with haematopoietic-specific Dnmt3a deletion developed spontaneous PAH, supporting the human genetic findings. The PAH in this murine model was associated with lung macrophage infiltration and was attenuated by IL-1β antibody therapy, suggesting an underlying inflammatory mechanism.

Our findings are robust because of the size and diversity of the cohorts we studied, especially since PAH is an orphan disease with an annual incidence of only 2–7.6 per million.³⁷ This study includes 2572 PAH Biobank and 2559 UKBB PAH participants, a separate biomarker cohort from Johns Hopkins University, and uses a haematopoietic Dnmt3a^−/− murine model to establish biological plausibility. The PAH Biobank also reflects the ethnic diversity of PAH, including individuals of Hispanic, African, and other ancestries, in addition to European ancestry (72%) and a good representation of the understudied APAH-CTD population (which comprises 28% of our cohort).

Germline DNMT3A variants were enriched 4.1-fold amongst European origin PAH Biobank participants vs. controls. When expanding across the entire PAH Biobank and including both germline and somatic variants, DNMT3A variants (8 germline and 25 somatic) were present in 1.2% of PAH subjects (vs. .43%, of controls, P = 1.65 × 10⁻¹⁰). None of the 13 DNMT3A variant carriers subjected to acute vasodilator testing were responsive vs. 13.4% (140/1043) of other biobank participants with available data (see Supplementary data online, Table S3), a clinically relevant finding given the poor prognosis associated with vasodilator non-responsiveness.

In our North American cohort 45.5% of DNMT3A variants detected by ES occurred in subjects with APAH-CTD, representing the first report of a prevalent gene mutation in this PAH subpopulation. APAH-CTD is the second most common form of PAH and had not been associated with a consistent underlying gene variant, although rare cases with variants in candidate PAH genes have been reported.^7,25 In an independent cohort, DNMT3A in the PBMCs was reduced in subjects with APAH-CTD patients compared with control subjects or participants with scleroderma without PAH. The AUCs noted for PBMC DNMT3A expression suggest a potential role as a biomarker.

Because somatic mutations can be pathogenically relevant at low VAFs (≥2%), which are not readily detectable by ES, we also studied a PAH Biobank subset using TPS. Panel sequencing provided a read depth 14 times greater than ES and demonstrated that both somatic DNMT3A variants and the aggregate of all CHIP variants are significantly linked to PAH (Figure 4). Individuals with CHIP were 23.35 times more likely to have PAH, supporting a strong association between CHIP and PAH. DNMT3A was the most frequently mutated CHIP gene, while TET2 and PPM1D were the second and third most prevalent, reinforcing the potential role of somatic mutations in PAH pathogenesis. Although somatic DNMT3A or other CHIP variants were a PAH risk factor, it conferred no significant difference in survival (see Supplementary data online, Figure S2) nor did it predict response to acute vasodilator testing (see Supplementary data online, Figure S3) or severity of haemodynamics (see Supplementary data online, Table S2). Thus, somatic DNMT3A and CHIP variants lack the prognostic impact of BMPR2 germline variants, which confer high PAH risk and poor prognosis in patients under age 50 years.³⁸

The role of DNMT3A variants and CHIP variants as risk factors for PAH is consistent with the risk attribution of somatic variants in other diseases. For example, as a precursor to acute myelogenous leukaemia, CHIP variants increase disease risk 10-fold; however, the absolute risk of leukaemia remains low at .5%–1% per year.^17,18 Likewise, CHIP variants are risk factors for adverse outcomes post-myocardial infarction.^39,40

One might have predicted a higher VAF would identify patients with a worse prognosis; however, neither the severity of VAF, nor a VAF >.1 vs. <.1 for DNMT3A R882 variants, nor the presence of multiple CHIP or DNMT3A variants predicted greater mortality or worse pulmonary haemodynamics (see Supplementary data online, Figures S4 and S5). This is consistent with the CANTOS genotyping sub-study, in which there was no statistically significant increase in MACE in those with a VAF >.1 vs. <.1.²⁰ It’s possible that somatic variants cause only mild loss of function, not enough to determine disease severity independently. Alternatively, CHIP-mutant cells might influence nearby healthy cells through paracrine signals, allowing even a small clone to drive widespread changes, making VAF less reflective of overall disease impact.⁴¹

ROC analysis demonstrated the potential of reduced DNMT3A variant 3 and 4 expression as a predictive biomarker for PAH, but this requires confirmation in an independent cohort. Ricard et al. reported an increase in CHIP, particularly DNMT3A, in 90 people with scleroderma vs. 44 healthy controls⁴²; however, they did not evaluate for PAH. Although DNMT3A germline and CHIP variants can decrease DNMT3A mRNA or protein,⁴³ the >90% prevalence of DNMT3A mRNA suppression suggests that non-genetic processes are in play, mirroring the down-regulation of TET2 expression we previously observed in this cohort.¹² Similarly, BMPR2 protein expression is reduced in the small pulmonary arteries of IPAH patients without BMPR2 mutation⁴⁴ and in genetically normal rats with monocrotaline-induced PAH.⁴⁵ We suspect the mechanism of down-regulation of DNMT3A (and TET2 and BMPR2) in such circumstances is epigenetic.

Haematopoietic Dnmt3a depletion in mice induced spontaneous pre-capillary pulmonary hypertension associated with adverse pulmonary vascular remodelling and inflammation, key features of PAH (Figure 7). Since this mouse lacks Dnmt3a only in haematopoietic cells, we conclude that the pulmonary circulation is a paracrine target of this inflammatory shift in bone marrow-derived inflammatory cells. Somatic DNMT3A variants can promote in vivo immortalization of affected haematopoietic stem cells, decreasing DNA methylation at regulatory regions of self-renewal genes, leading to age-related clone expansion and progressive inflammation.⁴⁶ Depletion of Dnmt3a in haematopoietic cells led to heightened lung inflammation, notably the accumulation of lung macrophages, which may contribute to adverse vascular remodelling in PAH by stimulating pulmonary artery smooth muscle cell (PASMC) proliferation (Figure 7C and D).^47,48 Plasma IL-13 levels (but not IL-1β) were elevated in Dnmt3a^−/− mice, consistent with increased IL-13 reported in APAH-CTD patients and IL-13’s role in monocyte/macrophage activation in PAH.⁴⁹ Treatment with IL-1β antibody regressed PAH and reduced lung macrophage infiltration in Dnmt3a^−/− mice, even though plasma IL-1β levels were not elevated. This is consistent with the CANTOS study in which canakinumab was most effective in patients with TET2 mutations but also conferred benefit irrespective of CHIP status.²⁰ Thus, while CHIP is a key inflammatory driver, it is not the only source of canakinumab-sensitive inflammation. For example, Woo et al. found that inflammation-related anaemia was increased by CHIP in CANTOS, but canakinumab prevented anaemia even in subjects lacking a CHIP variant.⁵⁰ Even established germline mutations, like BMPR2, display incomplete penetrance. For example, the annual risk of an asymptomatic BMPR2 variant carrier developing PAH is only 1% for males and 3.5% for females.⁵¹ We suspect that risk factors for PAH, including age, female sex, exposure to anorexigens or amphetamines, and connective tissue diseases, are more likely to cause PAH when superimposed on an abnormal genetic background.

Further evidence of the importance of inflammation in PAH comes from 641 patients with APAH-CTD in the PAH Biobank 40% of whom (240) (see Supplementary data online, Figure S6) received anti-inflammatory therapy. Those receiving anti-inflammatory therapies had less severe pulmonary haemodynamics than those who did not, with significantly higher CO and lower PVR (Figure 5B). Although assignment to these agents was not randomized, we note that compared with no therapy, CO was highest with the use of mycophenolate mofetil and PVR lowest in those receiving a monoclonal antibody therapy (Figure 5A).

We previously demonstrated that excessive DNA methylation of specific target genes plays a role in the development of PAH.¹³ Moreover, pathogenic variants of the demethylating gene, TET2, predispose to PAH in humans and mice.¹² However, the consequence of TET2 mutations in PAH is pan-chromosomal hypermethylation and inflammation.¹³ Thus, it may appear counterintuitive that both mutation of TET2 and DNMT3A (which have opposing effects on the methylome) concordantly promote inflammatory PAH. However, Rauch et al¹¹ found both TET2 and DNMT3A variants cause a shift towards a pro-inflammatory macrophage phenotype.⁵² This paradox likely reflects that TET2 and DNMT3A target different genes,²³ with TET2 regulating genes associated with cell differentiation, while DNMT3A regulates genes related to stem cell renewal.¹¹

Study limitations

The study has potential limitations. First, while ES lacks sensitivity for low VAF CHIP variants, it has good specificity for detected somatic variants and allows comparison to gnomAD and the UKBB, which also used ES. Consequently, we also used TPS which, as expected, increased the incidence of DNMT3A variants from 25/2572 (.97%), measured using ES, to 116/1659 (7%), by TPS.

Second, the TPS control group was selected solely based on normal renal function³⁰ and relative to the PAH group was older, male predominant, and had significantly more systemic hypertension and Type 2 diabetes mellitus (see Supplementary data online, Table S1B). However, a control group with substantial cardiovascular comorbidities¹⁴ and diabetes⁵³ would likely be associated with increased CHIP making our finding of increased somatic variants in PAH more robust. Using the PAH Biobank, we examined subjects under age 50 years with no comorbidities. CHIP was more prevalent (5/54, 9.3%) in this subset of patients with PAH vs. control subjects who were also under 50 years and void of diabetes and hypertension (8/207, 3.9%; P = .1512; Supplementary data online, Figure S14). This further supports the finding from the larger age-adjusted analysis that CHIP variants are increased in PAH.

Third, we did not have access to germline analysis within the UK biobank. Also, in the UKBB, PAH is defined only by ICD-10 codes, whereas enrolment in the PAH Biobank involved diagnosis by a PAH specialty clinic, enhancing case veracity. Moreover, the female:male ratio is .89:1 in the UKBB vs. 3.73:1 in the PAH Biobank. The UKBB cohort sex ratio in PAH is not in keeping with other PAH specific registries, all of which have >3/1 female predominance, reviewed in Thenappan et al.³⁷ In addition, the median age of PAH patients is higher in the UKBB vs. the PAH Biobank (64 vs. 57 years).

Fourth, we acknowledge a 2025 paper suggesting a lack of role for DNMT3A in PAH.⁵⁴ They used two-sample Mendelian randomization (MR) on genome-wide association studies (GWAS) data sets and confirmed our 2020 finding¹² that TET2 variants are a risk factor for PAH, but found no causal relationship for DNMT3A or other CHIP genes. They did not assess for germline DNMT3A variants. Moreover, the genetic basis for CHIP remains incompletely understood (and is poorly MR). There is substantial environmental influence on CHIP development, making it difficult to draw conclusions regarding the association of DNMT3A CHIP and PAH from MR alone. In addition, methodologic differences likely contribute to the discordant findings (GWAS, which uses microarrays of SNPs vs. direct gene sequencing to precisely define variants).

Fifth, we used mice with floxed Dnmt3a alleles (f/f) lacking Cre as control mice. While cardiac Cre recombinase expression, driven by an αMyHC promoter, can cause cardiac toxicity, the targeted deletion in our study was exclusive to Cre-targeted haematopetic cells (thus, there was no cardiac Cre). Moreover, we identified no haemodynamic differences in the LV haemodynamics in Dnmt3a^−/− mice (see Supplementary data online, Figure S6).

Conclusion

Human genetic studies and the use of a relevant murine model indicate that somatic and germline DNMT3A and CHIP variants are associated with PAH, including APAH-CTD. The link between DNMT3A and CHIP variants and PAH is likely through inflammation. These findings deepen our understanding of the genetic and epigenetic mechanisms underlying PAH, highlight the potential role of DNMT3A variants and CHIP as PAH risk factors, and suggest potential utility of DNMT3A expression as a biomarker. A randomized clinical trial would be required before considering the use of canakinumab in PAH.

Supplementary data

Supplementary data are available at European Heart Journal online.

Declarations

Disclosure of Interest

Nothing to declare.

Data Availability

The authors declare that all supporting data are available within the article and the online Supplemental Data.

Funding

This study received support from US National Institutes of Health (NIH) grants NIH 1R01HL113003-01A1 and NIH 2R01HL071115-06A1 (S.L.A), NIH R24HL105333 and NIH R01HL160941 (W.C.N.), Canada Foundation for Innovation 229252 and 33012 (S.L.A.), Tier 1 Canada Research Chair in Mitochondrial Dynamics and Translational Medicine 950-229252 (S.L.A.), Canadian Institutes of Health Research (CIHR) Foundation Grant 143261 (S.L.A.), the William J. Henderson Foundation (S.L.A.), Canadian Vascular Network Scholar Award (F.P.), JPB Foundation (W.K.C.), CIHR Canada Graduate Scholarship—Master’s (I.M.E.), an Ontario Molecular Pathology Research Network (OMPRN)/Ontario Institute for Cancer Research (OICR) Cancer Pathology Translational Research Grant (M.J.R.), CIHR Project Grant 451147 (M.J.R.), and a Canada Foundation for Innovation Grant (M.J.R.).

Ethical Approval

Human data: Ethical approval of this study was obtained from the IRBs from CCHMC, Queen’s University, and Vanderbilt University, while individual enrolling centres’ IRBs approved local data collection for the Biobank. The samples and data are available for use by the PAH research community, as approved by CCHMC’s IRB. Rodent experiments were conducted following the Canadian Council on Animal Care (CCAC) regulations approved by Queen’s University Animal Care Committee (Protocol 2021-2128 and 2022-2130).

Pre-registered Clinical Trial Number

Not applicable.