1. Introduction

Pulmonary arterial hypertension (PAH; OMIM #178600) is a rare disease with a prevalence of 15–50 cases per one million individuals [1, 2]. Symptoms of PAH are nonspecific including shortness of breath particularly during exertion, palpitations, and, in advanced cases, edema, chest pain, and dyspnea at rest. While noninvasive assessments such as echocardiography, electrocardiography, laboratory tests, pulmonary function tests, and cardiopulmonary exercise testing can support a diagnosis of PAH, the definitive diagnosis is made by right heart catheterization. PAH is hemodynamically defined by an elevated mean pulmonary artery pressure > 20 mmHg, pulmonary vascular resistance > 2 Wood units, and a pulmonary artery wedge pressure ≤ 15 mmHg [2]. PAH is currently classified as idiopathic (IPAH), heritable (HPAH), or associated with other diseases or exposures, including connective tissue disease, congenital heart disease, HIV infection, or drugs and toxins. By definition, patients with HPAH have a positive family history and/or carry a pathogenic variant in a definitive PAH risk gene. Thus, apparently idiopathic patients may be reclassified as HPAH following genetic analysis.

The first gene identified as causative for PAH encodes the Bone Morphogenetic Protein Receptor Type 2 (BMPR2; HGNC: 1078) [3, 4], and it remains the gene with the highest frequency of Pathogenic variants in HPAH patients [5–7]. BMPR2 is a cell surface receptor and part of the Transforming Growth Factor β superfamily responsible for vascular homeostasis of cell proliferation and apoptosis. Heterozygous Pathogenic or Likely Pathogenic (LP) BMPR2 variants lead to PAH in an autosomal dominant manner with reduced penetrance of approximately 15%–40% and show sexual dimorphism, with higher penetrance in females [8].

Subsequently, several additional genes have been implicated in the etiology of IPAH and HPAH, especially with the advent of next-generation sequencing of large cohorts [9–11]. A recent study carried out by our Clinical Genome Resource (ClinGen) pulmonary hypertension gene curation expert panel (PH GCEP) thoroughly characterized 27 genes and classified 12 of them to have a definitive gene–disease relationship with IPAH/HPAH (ACVRL1, ATP13A3, BMPR2, CAV1, EIF2AK4, ENG, GDF2, KCNK3, KDR, SMAD9, SOX17, and TBX4), three to have moderate evidence (ABCC8, GGCX, and TET2), six with limited evidence (AQP1, BMP10, FBLN2, KLF2, KLK1, and PDGFD), and one to have no evidence for the relationship with PAH (TOPBP1) [12]. Notably, four genes in the BMPR2 pathway (BMPR1A, BMPR1B, SMAD1, and SMAD4) were disputed as causative PAH genes [12].

ClinGen, funded by the National Institutes of Health (NIH), serves as a central resource with information on the clinical significance of genes and variants [13]. Curation of genes and variants focuses on gene disease validity, variant pathogenicity, dosage sensitivity, and clinical actionability. ClinGen also offers the opportunity to form expert panels to curate not only genes but also variants [14]. To this end, following our gene curation efforts, we founded a PH variant curation expert panel (VCEP) to adapt the American College of Medical Genetics and Genomics (ACMG)/Association of Molecular Pathology (AMP) guidelines with a focus on BMPR2. We modified the general variant curation guidelines by the ACMG/AMP framework [15] for BMPR2 variants specific to HPAH and/or IPAH. Since a significant number of BMPR2 variants have an unclear impact on gene expression or function we aimed to resolve conflicting classifications to improve the interpretation of genetic test results. We tested our refined specification criteria on a pilot subset of 28 representative BMPR2 variants, the results of which are described herein.

2. Methods

Within the framework of our Genetics Task Force of the International Consortium for Genetic Studies in PAH (PAH-ICON, https://pahicon.com/), we founded a ClinGen Pulmonary Hypertension VCEP with 16 members from nine institutions across the United States and Europe, with representation from six countries (https://www.clinicalgenome.org/affiliation/50071/). Members included a chairperson (WKC), scientific lead/coordinator (CLW), expert reviewers, and biocurators. Expert reviewer and biocurator roles were assigned based on members’ experience, but most members had dual roles. Seven members hold professional certifications in clinical genetics and/or regularly use ACMG/AMP guidelines to classify variants for clinical laboratory case sign-out. Most members had a history of ClinGen activities including curation of PAH gene relationships [12]. All members received additional training in the ClinGen variant curation standard operating procedure V3, the 2015 ACMG/AMP guidelines, the ClinGen sequence variant interpretation working group recommendations for applying ACMG/AMP criteria, sequence variant literature searches, and use of ClinGen’s variant curation interface (VCI) [16]. We followed the ClinGen four-step process (protocol Version 9) to obtain VCEP approval (February 2024) with resulting variant classifications meeting US Food and Drug Administration (FDA) recognition. A flow chart for the VCEP formation and activities is shown in Figure 1.

We limited the scope of this work to BMPR2 variants because of the large number of loss-of-function variants, recurrent variants, and the relatively large body of experimental evidence defining critical domains/residues and functional effects of individual variants compared to other PAH risk genes [12]. Monthly virtual meetings were held over a 14-month period (June 2021 to August 2022) to develop variant classification rules for BMPR2 based on the ACMG/AMP guidelines (Figure 1). We then tested our adapted specifications on a pilot group of 28 variants, representing all ACMG classes, including those with conflicting interpretations of pathogenicity. The selected variants included insertions, deletions, 5 ^′-untranslated region, missense, splice, nonsense, and synonymous variants. Only BMPR2 variants identified in nonsyndromic HPAH and/or IPAH cases were considered for pilot curation. Thus, one additional variant (c.319T>C p.Ser107Pro) was removed from the pilot curation list as it had been described in a proband with congenital heart disease–associated PAH and was therefore outside the scope of this work. The BMPR2 MANE Select transcript with accession number NM_001204.7 was used as the reference for all variant curations.

We assigned six variants to each curator and teams of three curators per variant. Provisional classifications were presented and discussed at monthly meetings over another 14-month period (August 2022 to October 2023). Modifications of our BMPR2 criteria specifications were made over this period and following guidance of the ClinGen sequence variant interpretation working group. The specifications are available in the ClinGen Criteria Specification Registry (https://cspec.genome.network/cspec/ui/svi/doc/GN125). Final curations were uploaded to the VCI. Following the final PH VCEP approval, variant curations were published to the ClinGen Evidence Repository and submitted to ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/?term=bmpr2%255Bgene%255D%26redir=gene). All members disclosed conflicts of interest, and the curation assignments were mindful of conflicts. Our ACMG/AMP specifications are updated periodically. To find the most current information, please visit https://cspec.genome.network.

3. Results

3.2. Pilot Testing of PH VCEP Specifications

The results of our pilot curation of 28 BMPR2 variants are detailed in Table 2 and Figure 2. The classification of seven variants (25%; six pathogenic and one VUS) did not change from the existing ClinVar classification. These included well-established Pathogenic variants that were included as “positive controls” to test our specifications. If the P/LP and B/LB classifications are grouped per their clinical utility, a further seven variants were unchanged, including four with mixed B/LB classifications in ClinVar. Three of these were resolved to Likely Benign and one to Benign when our specifications were applied, although this is not consequential for their clinical interpretation. Of the remaining 14 variants (50%), nine had discrepant classifications in ClinVar. Eight of these were VUS/LB or VUS/LB/B, of which three were resolved to Benign, three to Likely Benign, and two remained as VUS. Two variants changed from Likely Benign to VUS or vice versa. The most pronounced and clinically consequential changes were seen with intronic splice region variants, with three variants (c.247+1_247+7del, c.529+2T>C, and c.968-3C>G) reclassified from pathogenic or P/VUS to VUS and one from VUS to Pathogenic (c.968-5A>G). As discussed further below, mRNA analysis to confirm the predicted consequences of splice site variants is critical for accurate classification, especially for variants outside of the canonical positions.

Table 2. Detail of the BMPR2 variants curated in the pilot study.

Varianta	Variant type	ClinVar classification	PH VCEP classification	ACMG/AMP criteria and points applied	Scoreb
c.-924A>G	5 ′-Untranslated region	VUS/likely benign	Benign	BA1 −8, BS2_supp −1, BP4 −1	−10
c.-669G>A	5 ′-Untranslated region	Likely benign/benign	Benign	BA1 −8	−8
c.218C>G (p.Ser73 ∗)	Nonsense	Pathogenic	Pathogenic	PVS1 +8, PS4_supp +1, PM2_supp +1	+10
c.240_241insT (p.Lys81 ∗)	Nonsense	Pathogenic	Likely Pathogenic	PVS1 +8, PM2_supp +1	+9
c.247+1_247+7del	Splice region/intronic	Pathogenic	VUS	PVS1_strong +4, PM2_supp +1	+5
c.251G>A (p.Cys84Tyr)	Missense	Pathogenic	Pathogenic	PS4_mod +2, PM1_strong +4, PM2_supp +1, PM5 +2, PP3 +1	+10
c.354T>G (p.Cys118Trp)	Missense	Pathogenic (3)	Pathogenic	PS3 +4, PS4 +4, PM1_strong +4, PM2_supp +1, PM5_supp +1, PP1 +1, PP3 +1	+16
c.419-38del	Splice region/intronic	VUS/likely benign/benign	Benign	BA1 −8, BS2 −4, BP4 −1, BP7 −1	−14
c.529+2T>C	Canonical splice site	Pathogenic	VUS	PVS1_strong +4, PM2_supp +1	+5
c.545G>A (p.Gly182Asp)	Missense	VUS (3)/likely benign (1)	Likely benign	BS3 −4, PP3 +1	−3
c.797G>C (p.Arg266Thr)	Missense	VUS	VUS	PM2_supp +1	+1
c.901T>C (p.Ser301Pro)	Missense	Likely pathogenic	Pathogenic	PS3 +4, PS4 +4, PM1 +2, PM2_supp +1, PM6 +2	+13
c.968-5A>G	Splice site/intronic	VUS (3)	Pathogenic	PVS1 (RNA) +8, PS4_supp +1, PM2_supp +1	+10
c.968-3C>G	Splice site/intronic	Pathogenic/VUS	VUS	PS4_supp +1, PM2_supp +1, PP3 +1	+3
c.1040G>A (p.Cys347Tyr)	Missense	Pathogenic	Likely pathogenic	PS3_supp +1, PS4 +4, PM2_supp +1, PM5_supp +1, PP3 +1	+8
c.1042G>A (p.Val348Ile)	Missense	Likely benign/benign	Likely benign	BS1 −4, PS3_supp +1	−3
c.1413+1G>A	Canonical splice site	Pathogenic	Pathogenic	PVS1 +8, PS4_mod +2, PM2_supp +1	+11
c.1424C>A (p.Ser475 ∗)	Nonsense	Pathogenic (2)	Pathogenic	PVS1 +8, PS4_supp +1, PM2_supp +1	+10
c.1472G>A (p.Arg491Gln)	Missense	Pathogenic	Pathogenic	PS2 +4, PS3_supp +1, PS4 +4, PM1_strong +4, PM2_supp +1, PM5 +2, PP1 +1, PP3 +1	+18
c.1481C>T (p.Ala494Val)	Missense	Likely benign	VUS	PM1 +2, PP3 +1, BS1 −4	−1
c.1509A>C (p.Glu503Asp)	Missense	VUS	Likely benign	BS3 −4	−4
c.1698T>A (p.Ile566=)	Synonymous	VUS/likely benign	VUS	PM2_supp +1, BP4 −1, BP7 −1	−1
c.1766A>G (p.Tyr589Cys)	Missense	Likely benign/benign	Likely benign	BS1 −4	−4
c.2186G>C (p.Gly729Ala)	Missense	VUS/likely benign	Likely benign	BS1 −4	−4
c.2352C>T (p.Val784=)	Synonymous	VUS/likely benign	VUS	PM2_supp +1, BP4 −1, BP7 −1	−1
c.2618G>A (p.Arg873Gln)	Missense	VUS/likely benign	Benign	BS1 −4, BS3_supp −1, BS4 −4	−9
c.2887G>T (p.Gly963Cys)	Missense	Likely benign/benign	Likely benign	BS1 −4	−4
c.2948G>A (p.Arg983Gln)	Missense	VUS (3)/likely benign (1)/benign (1)	Likely benign	BS1 −4	−4

• Abbreviations: mod, moderate; supp, supporting; VUS, variant of uncertain significance.
• ^aVariant nomenclature refers to transcript NM_001204.7.
• ^bScoring modified from Tavtigian et al. [38]: pathogenic (P), ≥ 10 points; likely pathogenic (LP), 6 to 9; VUS, −1 to 5; likely benign (LB), −2 to −6; benign (B), ≤ −7.

4. Discussion

We adapted the general ACMG/AMP variant specification criteria specifically to BMPR2, the gene most commonly implicated in PAH, with loss or reduced function as the mechanism of action. This guidance should be valuable to improve harmonization in reporting variants identified in patients with heritable or idiopathic PAH across clinical laboratories. Of note, we did not consider variants identified in patients with associated PAH; therefore, these specifications may require modification in such cases.

Unlike null variants such as canonical splice site or frameshift variants, it is often challenging to classify novel and rare missense variants as Likely Pathogenic/Pathogenic with currently available data. In many cases, functional data are lacking, and previous reports of the same variant in other PAH patients are usually limited. This often leaves only PM2 (low population frequency) and PP3 (pathogenicity predictions) to support missense variant interpretation. Hence, it is crucial for patients and clinicians that all variants are uploaded to ClinVar to potentially identify recurrent variants or amino acid changes to apply additional criteria (PS4, PS1, and PM5). Accurate predictive testing for the familial disease-causing variant is only possible for Likely Pathogenic or Pathogenic variants, and genetic risk prediction is critical as there are increasingly effective therapies for PAH that are likely more effective when applied earlier in the disease course.

Of the 28 BMPR2 variants curated in this pilot study, only eight (28.6%) had functional analyses that could be scored, of which five provided evidence for PS3, two for BS3, and one for BS3_supp. Access to tissues, cell lines, and nucleic acids to perform the functional analyses considered in this study is not always possible, and, when available, the materials and resources may be limited. The lack of functional analyses presents challenges when determining the impact of a given reported genomic variant. In this regard, the assessment of mRNA in patients with suspected splice site variants located more than two base pairs within the intron is relatively straightforward and provides a key experimental indication of pathogenicity. The importance of this is highlighted by three splice site variants in our pilot panel, including one at the +2 position of a splice donor site, that were downgraded from Pathogenic to VUS due to the lack of supporting mRNA data. Conversely, one variant in the −5 splice acceptor position was upgraded from VUS to Pathogenic on the basis of mRNA analysis, performed in the lab of a VCEP member, that provided evidence for PVS1 (RNA).

Our pilot testing of AI-based prediction algorithms, including AlphaMissense, suggests that such programs may be highly accurate in predicting the pathogenicity of missense variants in the absence of functional data. However, this is an evolving field, with ongoing development of new programs and calibration of threshold values for established tools. Of note, while this manuscript was in revision, a new posterior probability-based calibration was proposed for AlphaMissense [41]. The minimal threshold to apply PP3 was determined as ≥ 0.792, considerably higher than the original 0.564, while the threshold for BP4 moved from 0.340 to < 0.170. As a result, the agreement with our VCEP classification for BMPR2 missense variants was reduced to 7 of 15 variants, with five previously concordant variants changing from Likely Benign to uncertain (Supporting Information 2: Table S1). Nonetheless, Bergquist et al. emphasize that these tools should not be considered the sole classifier; rather, an integrated analysis per the ACMG framework will almost always be needed [41]. With this in mind, we reframed our comparison of these algorithms with our integrated VCEP curation to focus on whether the score produced by each program was concordant, neutral, or discordant with the final VCEP classification, rather than applying a definitive in silico classification. Uncertain calls were considered neutral. The updated thresholds rendered AlphaMissense more conservative in calling evidence for benignity (BP4), with only 2/8 Benign/Likely Benign variants achieving BP4 and the remaining six having indeterminate scores. Importantly, all Pathogenic/Likely Pathogenic variants still met the minimal threshold for PP3, and no AlphaMissense calls were discordant (Supporting Information 2: Table S1). In contrast, CADD scores, even when recalibrated to the more stringent thresholds proposed by Pejaver et al. [40], were discordant for five variants, while the recalibrated thresholds for REVEL and BayesDel [41] were discordant in three and five cases, respectively. While these results suggest that AlphaMissense and REVEL are promising tools, the use of in silico prediction programs is a dynamic landscape that will require ongoing reevaluation. Therefore, the composite approach adopted in our BMPR2 guidelines remains the most robust method of determining variant pathogenicity or benignity [40, 41].

In this study, we encountered a large Iberian family in which 32 individuals were genotyped and 22 were heterozygous for c.1472G>A p.Arg491Gln cosegregating with PAH [42]. However, segregation analyses are typically challenging due to small family sizes, the work needed to organize family studies, and incomplete penetrance. Due to the need for cardiac catheterization for clinical diagnosis, pulmonary hemodynamics may not be available for some family members. In addition, without enough affected heterozygotes, the number of meioses observed may not be sufficient to use the segregation criteria PP1 or BS4. The presence or absence of this evidence can significantly influence the final classification of a given variant. Therefore, it is essential to develop protocols or policies to collect relevant data from large families when there is suspicion of HPAH. Moreover, clinical guidelines recommend offering cascade testing to potentially identify at-risk heterozygotes and offer clinical screening to diagnose PAH as early as possible [5].

ACMG guidelines were developed for Mendelian genetic disorders with the goal of dichotomizing variants into Pathogenic/Likely Pathogenic and Benign/Likely Benign categories to define their clinical utility (or lack thereof). Most heritable PAH cases, including those related to BMPR2, show autosomal dominant inheritance, and thus, our VCEP appropriately sought to adapt ACMG guidelines for this gene. However, PAH shows reduced penetrance, suggesting that other genetic and/or environmental modifiers also play a role. In this regard, the publication by Masson et al. is thought provoking, recognizing that some VUS may demonstrate disease relevance based on population genetics or functional data but do not satisfy the stringent criteria for Pathogenic or Likely Pathogenic classification [43]. Thus, they proposed new terminology of “predisposing” and “likely predisposing.” This concept may be helpful in classifying lower penetrance variants that could contribute to PAH in a polygenic model and should be considered in the future.

In conclusion, these newly adapted ClinGen specification guidelines to assess BMPR2 variants for PAH patients will allow more reliable and accurate variant classification. Although the overall proportion of VUS increased, this reflects that approximately one-third of the chosen variants had discrepant classifications in ClinVar. Notably, we were able to resolve six of the eight variants with a discrepant VUS/LB, reclassifying them as Benign or Likely Benign. The remaining VUS may be reclassified with increasing evidence in the future. Benign or Likely Benign variants are most likely unrelated to the disease and thus do not increase the genetic risk of developing PAH. In contrast, Pathogenic or Likely Pathogenic variants allow for risk stratification of other family members. Thus, the consensus specification document substantially aids in providing patients and caregivers with an accurate genetic diagnosis which may play an increasing role in therapy.

Ethics Statement

All data were obtained from ClinVar and/or the published literature and were fully deidentified. Therefore, IRB approval and informed consent was not required.

Disclosure

This manuscript was submitted as a preprint at the following link [44]: https://www.medrxiv.org/content/10.1101/2024.11.24.24317862v1

Conflicts of Interest

CAE received honoraria for lectures and presentations from OMT and MSD, consulting fees from MSD. CAE is coinventor of the issued European patent “Gene panel specific for pulmonary hypertension and its uses” (EP3507380). CAE works at a laboratory that performs fee for service testing in genes that have been specified by the ClinGen PH VCEP. FC works at a French healthcare hospital (Pitié-Salpêtrière Hospital in Paris) laboratory which performs noncommercial fee for service testing in BMPR2 and other PAH genes. DD works at a Dutch healthcare hospital laboratory which performs noncommercial fee for service testing in BMPR2 and other PAH genes. ME worked at a French healthcare hospital laboratory which performs noncommercial fee for service testing in BMPR2 and other PAH genes. DM works for GeneDx, a laboratory that performs fee for service genetic testing. WKC serves on the Board of Directors of Prime Medicine. GMV and SB were graduate students at the time the study was performed but are now employed by fee for service genetic testing companies, Tempus and Baylor Genetics, respectively. MAA, SB, KMD, and GMV received salary support from the National Institutes of Health. All other authors declare no conflict of interest.

Author Contributions

Conceptualization, methodology, and data curation: all authors; supervision: C.L.W. and W.K.C.; writing—original draft: C.A.E., G.M.-V., R.D.M., S.G., M.S., C.L.W., and M.A.A.; writing—review and editing: all authors

Funding

This research is supported by the National Heart Lung and Blood Institute (R35HL140019) (MAA, SB, and KMD) and Diversity Supplement (R35HL140019-S1) (GMV).

Supporting Information

Additional supporting information can be found online in the Supporting Information section.