Inclusivity in Research Matters: Variants in PVT1 Specific to Persons of African Descent Are Associated with Pulmonary Fibrosis

To the Editor:

Sarcoidosis is a systemic, granulomatous disease characterized by a dysregulated immune response. Its leading cause of mortality is pulmonary fibrosis (1). Less than 20% of patients with sarcoidosis develop fibrosis, but risk factors include limited access to care, extrapulmonary involvement, and African ancestry (2, 3). Unfortunately, despite studies highlighting ancestry-specific genetic effects in sarcoidosis and its manifestations (3, 4), genetic studies of sarcoidosis-related pulmonary fibrosis (SPF)—reviewed in Reference (5)—have been limited to patients who are persons of European descent (EUs). Although these studies found associations with genes implicated in fibrosis, inflammation, and immunoregulation, they included heterogeneous patient groups; included only candidate genes; had relatively small sample sizes; and, most important, did not include persons of African descent (AAs), the group most affected by SPF.

To meet the critical need of identifying biomarkers and mechanisms of pulmonary fibrosis in underrepresented populations, we performed the first whole-genome scan (WGS) of SPF in a non-EU population and pulmonary fibrosis in a large cohort of AAs. We identify African-derived risk variants constituting a haplotype that is not present in EUs in plasmacytoma variant translocation 1 (PVT1), which encodes for a PVT1, long-noncoding RNA associated with inflammation and fibrosis. We further implicate genes involved in profibrotic transforming growth factor beta (TGF-β) and antifibrotic bone morphogenic protein (BMP) signaling that encode well-characterized members of a receptor superfamily that utilizes canonical (SMAD) and noncanonical (phosphatidylinositol-3-kinase/protein kinase B, mitogen-activated protein kinases [MAPK], nuclear factor-κB) signaling pathways. Our findings support a model of SPF in which TGF-β/BMP signaling is dysregulated, ultimately inducing overproduction of collagen and proliferation of fibroblasts in both AAs and EUs. However, the genetic risk factors themselves are ancestry specific and may mediate ancestry-related differences in prognosis and treatment response.

Methods

Our WGS comprised 190 AAs with SPF with confirmed Scadding stage 4 on chest X-ray and 770 AAs with sarcoidosis with Scadding stages 1, 2, and 3 (non-SPF). The variant genotypes were obtained by means of whole-genome sequencing for 295 participants (69 [36%] SPF, 206 [27%] non-SPF) or were imputed from genotype data from ∼1.1 million observed SNPs (Illumina Human Omni1-Quad) using the TOPMed imputation server (Figure 1B). Pre- and postimputation methods were previously described (4) and resulted in 8,405,265 autosomal variants included in single-marker association tests using logistic regression in PLINK2, assuming an allelic (multiplicative) genetic model adjusting for sex and the first four ancestry principal components.

Figure 1.

Whole-genome scan of persons of African descent (AAs) with sarcoidosis-associated pulmonary fibrosis (SPF) compared with AAs without SPF and frequency of PVT1 risk variants. (A) Manhattan plot highlighting three regions reaching genome-wide statistical significance (upper line; P < 5 × 10⁻⁸) at LINC02982 (Chr5, rs11134383; P = 2.5 × 10⁻⁸; MAF_AA = 0.072), SPOP (Chr17, 17:49637910; P = 8.1 × 10⁻⁸; MAF_AA = 0.068), and PVT1 and replicating, at suggestive significance (lower line; P < 1 × 10⁻⁵), an effect reported in persons of European descent (EUs) at TGFB3 (Ch14, rs4252330; P = 1.2 × 10⁻⁶; MAF_AA = 0.36). The four statistically significant variants in PVT1 (Ch8, rs74730278, rs115311148, rs115529936, and rs115809470; P < 3.1 × 10⁻⁹; MAF_AA = 0.047) are in perfect linkage disequilibrium and form a haplotype with the same frequency as each of the alleles independently. (B) The number of individuals in this study with genome-wide genotyping (GWAS) and WGS by disease status. (C) The frequency of the PVT1 risk haplotype in AAs by disease status. The haplotype frequency for cases (n = 1,271) included SPF, non-SPF, and those missing Scadding stage but was not statistically significantly different from the frequency excluding those with missing data. ***The difference in the frequency of the risk haplotype in patients with sarcoidosis who have SPF (11.6%) compared with those patients without SPF (non-SPF) (4.0%) was significant at P = 9.99 × 10⁻⁹. Note that only one instance of the risk haplotype was found among EUs (haplotypic frequency, 0.018%). GWAS = genome-wide association study; MAF = minor allele frequency; WGS = whole-genome sequencing.

For comparison, we calculated haplotype frequencies for AAs with and without SPF—all cases with or without data for Scadding stage (1,273 AAs and 442 EUs)—and control subjects (1,645 AAs, and 2,284 EUs) using whole-genome sequencing, and we imputed data as described earlier. Significant differences in frequencies across groups were assessed using Haploview.

We screened for epistasis between SNPs in PVT1 and SNPs within 71 genes in the fibrosis pathway by fitting two-way interaction models using FastEpistasis (Bonferroni correction for 5,468 tests; P < 10⁻⁵) and then estimating interaction effect size for significant pairs of SNPs using PLINK2, adjusting for sex and four principal components.

Results and Discussion

Our WGS revealed three regions exceeding a genome-wide statistical significance of P < 5 × 10⁻⁸ (Figure 1A). The most significant region included four variants in PVT1 in perfect linkage disequilibrium (rs74730278, rs115311148, rs115529936, and rs115809470; each with P < 3.11 × 10⁻⁹). The frequency of these variants was virtually identical to the frequency of those in the Allele Frequency Aggregator (https://www.ncbi.nlm.nih.gov/snp/docs/gsr/alfa/). The frequency of the haplotype comprising these variants did not differ between AA cases and controls (5.2% and 5.4%, respectively, n = 5,836 haplotypes) and was only found once in the EU cohort (5,452 haplotypes; 0.01%), but it was found more frequently in AAs with SPF (11.6%) than in AAs without SPF (4.0%) (odds ratio, 3.28; 95% confidence interval, 1.87–5.78; P = 9.9 × 10⁻⁹) (Figure 1C).

PVT1 is an endogenous long noncoding RNA that acts as a sponge for microRNAs such as fibrosis-associated let-7 (6). In animal and cell culture models, PVT1 facilitates collagen production (7), fibroblast proliferation (7), and migration of lung fibroblasts (8). Knockdown or silencing of PVT1 attenuates fibrosis (7) and reduces the expression of major extracellular matrix proteins and their regulators (9). Additionally, PVT1 is proinflammatory by means of the canonically inflammatory MAPK and nuclear factor-κB pathways.

We verified the four associated variants in PVT1 as expression quantitative trait loci in immune cells in Ensembl (https://useast.ensembl.org/index.html) and QTLbase (http://www.mulinlab.org/qtlbase) and found epistasis between them and six variants within GREM2 (most significant: rs74511037, P = 4.3 × 10⁻⁶; minor allele frequency in AA = 0.05). We confirmed this association, showing significance at GREM2 lessened by two orders of magnitude (P = 4.3 × 10⁻⁴ to 1.2 × 10⁻²) when comparing carriers of the PVT1 risk haplotype to noncarriers. GREM2 encodes Gremlin-2, an extracellular BMP antagonist in the same family as Gremlin-1, encoded by GREM1, a gene associated with SPF in EUs (5). GREM2 is more highly expressed in the lung and blood of patients with idiopathic pulmonary fibrosis than in those of control subjects, and in fibrotic compared with nonfibrotic tissue, and elevated Gremlin-2 in human lung fibroblasts increases invasion and migration (10).

Although we acknowledge that our sample size was limited and that we do not have an AA replication cohort, our ability to replicate an effect that was previously identified in EUs within TGFB3 (5) and to identify novel associations—including known expression quantitative trait loci in an AA-specific haplotype within PVT1—in the first WGS of pulmonary fibrosis in AAs highlights the need for and potential impact of research in this area. As in studies of EUs, we found associations implicating dysregulated TGF-β/BMP signaling, but with distinct genetic risk factors. Specifically, for AAs, the SPF risk haplotype in PVT1 suggests an indirect effect on TGF-β signaling in addition to the effect seen at TGFB3, which encodes TGFβ-3, a cytokine involved in fibrosis and immune function (Figure 2). Likewise, our epistasis analysis suggests that SPF in EUs and AAs may uniquely inhibit antifibrotic BMP signaling through Gremlin-1 and Gremlin-2, respectively. Although dysregulation of the TGF-β/BMP signaling pathway may predispose patients to fibrosis, regardless of ancestry, the genetic influences on the mechanism of dysregulation appear to be ancestry specific and may mediate ancestry-related differences in prognosis. As both the first scan of SPF in a non-EU population and the first WGS of pulmonary fibrosis in a large cohort of AAs, our findings highlight the need for inclusion of underrepresented populations in research, as insights into mechanisms and potential treatments may otherwise remain undiscovered.

Figure 2.

Mechanisms of dysregulation of transforming growth factor β/bone morphogenic protein (TGF-β/BMP) signaling, the master regulator of fibrosis, include ancestry-specific genetic effects. A model, in fibroblasts, of our findings in persons of African descent, compared with those in persons of European descent, of TGF-β/BMP signaling dysregulation at the site of fibrosis formation in the lung. Fibroblasts express receptors for both TGF-β and BMP (TGFβR and BMPR, respectively) as well as PVT1. This model suggests ancestry-specific amplification of profibrotic TGF-β signaling and inhibition of antifibrotic BMP signaling pathways, ultimately leading to transcriptional effects and ECM production characteristic of fibrosis. PVT1, a long noncoding RNA, competes with other endogenous miRNAs, including let-7; promotes fibrosis through collagen production; and acts on the MAPK and NF-κB signaling pathways downstream of TGF-β/BMP signaling. Additional suggestive risk variants in TGFB3 and the epistatic effect of the BMP antagonist Gremlin-2 implicate multiple modulators of PVT1-mediated risk. In comparison, risk variants found in mixed cohorts of patients with progressive and fibrotic sarcoidosis reside in genes encoding TGF-β receptor agonists TGFβ-1 and TGFβ-3 and BMP antagonist Gremlin-1. ECM = extracellular matrix; miRNAs = microRNAs; NF-κB = nuclear factor-κB.