Novel HLA class I and II insights into the pathogenesis of Systemic Sclerosis–associated Interstitial Lung Disease

Abstract

Objectives

Systemic sclerosis-associated interstitial lung disease (SSc-ILD) is the leading cause of mortality in systemic sclerosis (SSc), yet its genetic architecture remains incompletely understood. Therefore, given the key role of the MHC in SSc, we aimed to perform a comprehensive MHC-wide association study in the largest SSc-ILD cohort to date.

Methods

We analyzed 2,412 SSc-ILD⁺ patients, 3,550 SSc-ILD⁻ patients, and 15,076 controls of European ancestry from 10 international cohorts. After quality control, the MHC region was imputed, and inverse variance weighted meta-analysis was performed. Subsequently, conditional stepwise analyses, adjustment for anti-topoisomerase autoantibody (ATA) status, and functional annotation of significant single nucleotide polymorphisms (SNPs) were performed. Finally, we constructed a composite score combining genetic, clinical and demographic variables to predict SSc-ILD.

Results

After conditional analysis, we detected twelve significant associations within class I and class II HLA genes. ATA adjustment reduced the significance of class II HLA variants, while class I HLA variants remained unaffected. Finally, the built composite score had an area under the curve of 0.754, significantly outperforming the models including any of the variables alone.

Conclusions

In this study, we identify genetic mechanisms underlying SSc-ILD that support the potential implication of CD8⁺ T cells and ATA autoantibodies in its pathogenesis. Moreover, we also demonstrate the enhanced efficacy of integrating genetic information into predictive models to detect patients at high risk of SSc-ILD. These findings provide new insights into disease pathogenesis and suggest potential biomarkers and therapeutic targets for improved patient management.

INTRODUCTION

Systemic sclerosis (SSc) is an immune-mediated inflammatory disease (IMID) characterized by a strong immune imbalance, vasculopathy, and fibrosis affecting skin and internal organs [1]. Although it has low prevalence, SSc exhibits the highest mortality rate of all rheumatic diseases. In fact, pulmonary complications, such as SSc-associated interstitial lung disease (SSc-ILD), are the leading cause of death in these patients [2].

SSc-ILD is present in approximately 50% of SSc patients and it exhibits marked heterogeneity, with clinical presentations ranging from mild and stable to rapidly progressive forms [3,4]. Established risk factors for developing SSc-ILD include the presence of anti-topoisomerase autoantibodies (ATA), diffuse cutaneous SSc, and male gender [5,6]. Furthermore, although there are treatments that have shown to slow the progression of SSc-ILD, there is currently no disease-modifying treatment that improves SSc-ILD or reverses the established pulmonary fibrosis [3].

Both environmental and genetic factors play crucial roles in SSc pathogenesis, and human leukocyte antigen (HLA) genes have consistently proven to be the strongest genetic association for the disease [7–9]. Variants within HLA genes may influence the antigen presentation to CD4⁺ T helper cells, leading to autoreactive T-helper and B-cells and the production of autoantibodies [8,10]. Although multiple studies have identified genetic associations in SSc [7,8,11–13], few have analyzed how genetic risk variants relate with specific clinical manifestations, such as SSc-ILD, and existing studies are often limited by small sample sizes [14–25].

Considering this, we sought to comprehensively analyze the specific contribution of the major histocompatibility complex (MHC) region in SSc-ILD by conducting the largest HLA association analysis to date. We evaluated the association of HLA alleles, single nucleotide polymorphisms (SNPs) and polymorphic aminoacidic residues with SSc-ILD and functionally assessed its regulatory role in order to identify potential pathogenic mechanisms, biomarkers and/or therapeutic targets.

MATERIALS AND METHODS

Study population

This study included ten cohorts comprising 5,962 SSc patients (2,412 SSc-ILD⁺ and 3,550 SSc-ILD⁻) and 15,076 controls. Blood/saliva samples were collected from all individuals, and, after DNA extraction, genome-wide genotyping was carried out using the arrays specified in Supp Table S1. All individuals were of European ancestry and recruited from multiple centres across Spain, the United Kingdom, the United States of America, Canada, Australia, Italy, Germany, Switzerland, France, and the Netherlands. All SSc patients fulfilled the 2013 American College of Rheumatology/the European League Against Rheumatism classification criteria, or the criteria proposed by LeRoy and Medsger for early SSc [26,27]. Clinical characteristics and further information on the 10 cohorts analysed are summarised in Supp Table S1.

Presence or absence of SSc-ILD was evaluated using high-resolution computed tomography (HRCT), or by the presence of radiologic findings on chest x-ray, and abnormalities on pulmonary function tests. The study protocol was approved by participating centres and the ethics committee of the Consejo Superior de Investigaciones Científicas (CSIC), and written informed consent was obtained from all participants in accordance with the principles of the Declaration of Helsinki.

Quality control and MHC imputation

To ensure the high-quality of the genetic data in downstream analysis, we applied stringent quality control (QC) procedures prior to imputation. Regarding sample filtering, individuals with genotype missingness > 0.05, ambiguous sex annotation, or relatedness were excluded. Relatedness was assessed using identity-by-descent estimation, and one individual from each pair of relatives (Pi_Hat > 0.4) or duplicates (Pi_Hat > 0.99) was excluded. Regarding variant filtering, SNPs with low call rate (< 0.98), deviations from Hardy-Weinberg equilibrium (HWE) (p-value < 1 × 10⁻³), or a minor allele frequency (MAF) < 0.01 were excluded. In addition, palindromic SNPs with an A/T or C/G allele frequency > 0.4 were removed to avoid strand ambiguity.

After these QCs, the MHC region variants (SNPs, classical alleles and aminoacidic residues) were imputed using the Four-digit Multi-ethnic HLA v1 (2021) reference panel on the Michigan imputation server with default settings (https://imputationserver.sph.umich.edu/). After imputation, we carried out further QCs, including the exclusion of variants with imputation quality (squared correlation [Rsq]) < 0.9, MAF < 0.01, low call rate (< 0.9), or deviating from HWE (p-value < 1 × 10⁻⁶). Finally, to control for population structure, we conducted principal component analysis over genome wide data using PLINK and GCTA64 [28,29] for each of the ten cohorts. 10 principal components were calculated for each cohort, and outlier individuals deviating by more than four standard deviations from the cluster centroid were excluded.

Statistical analysis

Logistic regression analyses were conducted in PLINK 2.0 for each of the ten cohorts, adjusting for sex and the first five principal components. For each cohort, three pairwise comparisons were carried out: SSc-ILD⁺ vs SSc-ILD⁻, SSc-ILD⁺ vs controls, and SSc-ILD⁻ vs controls. Then, inverse variance weighted meta-analyses of the datasets were carried out using Metasoft [30] under a fixed-effect model for non-heterogeneous variants (Cochran’s Q p-value > 0.05), while for variants showing heterogeneity (Cochran’s Q p-value < 0.05), the random effect model (RE2) was used. Statistical significance threshold was declared at 4.6 × 10⁻⁶, and suggestive threshold at 9.2 × 10⁻⁵. These were calculated using the genetic type I error calculator (GEC v0.2) [31], based on the effective number of tests in the imputed data from the MHC region.

Additionally, to ensure that our results were specific for SSc-ILD, the associated variants had to meet at least one of the two following criteria: (1) significant in SSc-ILD⁺ vs SSc-ILD⁻ comparison and not reaching suggestive association in the SSc-ILD⁻ vs controls comparison p-value > 9.2 × 10⁻⁵), or (2) having opposite effects, defined by the odds ratio (OR), associated to SSc-ILD⁺ and SSc-ILD⁻ (p-value < 4.6 × 10⁻⁶ in the three comparisons and opposing OR between SSc-ILD⁺ vs controls and SSc-ILD⁻ vs controls). This filtering step allowed us to prioritize associations more likely driven by the presence of SSc-ILD rather than by the broader SSc phenotype.

Finally, to detect conditionally independent associations, we performed conditional stepwise analysis with the COJO tool in GCTA 1.92.1, applied separately to SNPs, classical alleles and aminoacidic residues. At each step, we selected the most associated variant, re-ran the analysis including it as a covariate, and retained variants that remained significant, iterating until no additional variant reached significance.

Stratification and adjustment of associations by ATA

Given the strong correlation between SSc-ILD and ATA positivity, we performed a stratification of ATA patients within SSc-ILD⁺ subgroup to assess the potential participation of the significant variants in SSc-ILD severity, as well as an ATA-conditioned analysis with the aim of evaluating whether our genetic associations were independent of its effect.

Functional annotation of SSc-ILD specific variants

With the aim to assess the potential regulatory role of our findings, we conducted functional annotation of the significant and SSc-ILD specific SNPs and their proxies (r² > 0.9), by using the SNP2GENE function of FUMA GWAS webtool (https://fuma.ctglab.nl/snp2gene) and GTEx v.8 database. The parameters used for FUMA GWAS are listed in Supp Table S2. Specifically, we used FUMA GWAS to annotate relevant genes based on expression quantitative trait loci (eQTL) and chromatin interaction (Chr_Int) mapping. Furthermore, we queried GTEx v.8 database for splicing QTLs (sQTL). We focused this QTL annotation on relevant tissues/cell types for SSc-ILD (whole blood, immune cells, lung and fibroblasts). Additionally, we manually annotated the nearest gene to each of the variants included.

Risk prediction with polygenic risk score

In order to explore a potential clinical application of our results, we built a polygenic risk score (PRS) using the PRSice-2 software (https://choishingwan.github.io/PRSice/) and subsequently integrated the genetic component of this score with clinical and demographical data to generate a composite score of predictors of SSc-ILD. In order to conduct this analysis, the study population for which ATA status information was available was divided into two independent datasets: a training cohort comprising seven of the study cohorts (4,081 individuals; 1,722 SSc-ILD⁺ and 2,359 SSc-ILD⁻), used for selecting the SNPs of the model, and a testing cohort including the remaining three cohorts (1,559 individuals; 553 SSc-ILD⁺ and 1,006 SSc-ILD⁻), used to assess the predictive capability of the model. Within the training cohort, logistic regressions adjusted by the first five PCs, sex, and ATA status were performed. Meta-analysis and joint analysis with COJO then independent variants to be included in the PRS model, resulting in a final set of ten SNPs. The PRS was then computed in the testing cohort using PRSice-2, including ATA as a covariate. Subsequently, with the aim of constructing the composite score, we quantified the contribution of ATA, ACA, disease subtype (limited cutaneous SSc [lcSSc] and diffuse cutaneous SSc [dcSSc]) and sex variables in the training cohort through a generalized linear model adjusted for the first five PCs. Then, the composite score was calculated by combining the genetic PRS and the score derived from the clinical and demographic variables, each weighted by their respective regression coefficients.

Subsequently, the predictive performance of the models was evaluated using the pROC R package. We compared the area under the curve (AUC) across models using DeLong’s statistical test, and individuals in the testing cohort were stratified into percentile-based risk categories for which the risk ratio (RR) for developing SSc-ILD was estimated.

Finally, in order to test the replicability/reproducibility of the model, we tested the model in a validation cohort from the UK Biobank using the same parameters optimized in the testing phase. This cohort consisted of 74 SSc-ILD⁺ patients and 202 SSc-ILD⁻ patients of European ancestry.

RESULTS

MHC associated variants

A total of 1,541 MHC genetic variants were significant in the meta-analysis of SSc-ILD⁺ vs SSc-ILD⁻ (1,399 SNPs, 12 classical alleles, and 130 aminoacidic residues) (Figure 1). After conditional analysis in the region, we found 5 SNPs, 3 classical alleles, and 4 aminoacidic residues jointly significant (Table 1).

Figure 1. Manhattan plot of the MHC-wide association analysis for SSc-ILD⁺ vs SSc-ILD⁻.

Association results across the major histocompatibility complex region. Each point represents a variant, plotted according to its genomic position and −log₁₀(p-value). The color of the dot indicates the variant type (blue: SNP; red: classical allele; yellow: aminoacidic residue). The red line indicates the significance threshold of 4.6 × 10⁻⁶, and the blue dashed line indicates the suggestive threshold of 9.2 × 10⁻⁵. Relevant HLA genes are plotted into their genomic positions below the x-axis. AA: “aminoacidic residue”, HLA: “human leukocyte antigen”, MHC: “major histocompatibility complex”, SNP: “single nucleotide polymorphism”, SSc-ILD: “systemic sclerosis associated interstitial lung disease”.

Table 1.

Summary statistics of the significant and SSc-ILD specific variants after conditional analysis for the comparison of SSc-ILD⁺ vs SSc-ILD⁻.

Gene	Variant	Position (hg38)	EA	OR (95% CI)	P-value	P-value cond	SSc-ILD+ freq	SSc-ILD− freq	SSc-ILD+ vs Controls		SSc-ILD− vs Controls
									OR (95% CI)	P-value	OR (95% CI)	P-value
				Classical alleles
HLA-DQA1	HLA-DQA1*01:01:01:01	32637407	P	0.54 (0.47 – 0.61)	4.85E-22	-	0.116	0.201	0.72 (0.56 – 0.92)	5.04E-08	1.41 (1.30 – 1.53)	8.59E-16
HLA-DQA1	HLA-DQA1*03:01:01	32637415	P	0.71 (0.62 – 0.81)	1.87E-07	1.97E-10	0.131	0.182	0.80 (0.71 – 0.89)	1.06E-04	1.18 (1.08 – 1.29)	1.65E-04
HLA-DPB1	HLA-DPB1*13:01:01:01	33075955	P	2.70 (1.93 – 3.79)	7.68E-09	4.40E-08	0.060	0.024	3.80 (3.01 – 4.79)	3.91E-29	1.33 (0.99 – 1.78)	5.68E-02
				SNPs
HLA-DQB1	chr6:32663912:A:G	32663912	G	0.52 (0.46 – 0.60)	2.83E-23	-	0.118	0.205	0.71 (0.55 – 0.91)	1.17E-08	1.42 (1.31 – 1.54)	1.55E-16
HLA-DQA1	rs9272353	32636679	C	0.71 (0.62 – 0.80)	1.39E-07	8.37E-11	0.132	0.183	0.79 (0.71 – 0.89)	6.82E-05	1.18 (1.08 – 1.28)	2.41E-04
HLA-DPB1	chr6:33083222:G:A	33083222	A	2.53 (1.90 – 3.36)	1.72E-10	1.54E-09	0.061	0.025	4.13 (3.32 – 5.15)	1.71E-36	1.39 (1.06 – 1.83)	1.74E-02
HLA-B	rs2523558	31363260	G	1.37 (1.24 – 1.51)	6.01E-10	1.66E-06	0.377	0.297	1.35 (1.24 – 1.47)	3.15E-12	0.97 (0.86 – 1.10)	4.58E-01
NCR3	rs3132452	31595463	A	1.42 (1.22 – 1.64)	2.54E-06	3.60E-06	0.118	0.087	1.33 (1.18 – 1.50)	2.40E-06	0.97 (0.86 – 1.09)	6.00E-01
				Aminoacidic residues
HLA-DQA1	HLA-DQA1 Gln34	32641403	P	1.40 (1.28 – 1.55)	4.22E-12	-	0.630	0.546	1.39 (1.28 – 1.52)	6.96E-14	0.98 (0.92 – 1.06)	6.51E-01
HLA-DPB1	HLA-DPB1 Ile76	33080890	P	2.37 (1.81 – 3.12)	5.69E-10	3.32E-10	0.067	0.029	3.23 (2.63 – 3.96)	2.74E-29	1.17 (0.92 – 1.50)	2.07E-01
HLA-DQB1	HLA-DQB1 Gln224	32660998	P	1.95 (1.61 – 2.38)	2.36E-11	1.18E-07	0.867	0.774	1.48 (1.02 – 2.14)	2.70E-05	0.78 (0.68 – 0.89)	2.95E-04
HLA-C	HLA-C Asp9	31271844	P	1.27 (1.17 – 1.37)	6.36E-09	2.77E-06	0.435	0.372	1.20 (1.12 – 1.28)	1.59E-07	0.93 (0.88 – 0.99)	2.78E-02

Variants in this table meet the criteria for significance (p-value < 4.6E-06) and SSc-ILD specificity (opposing effect or p-value > 9.2E-05 in SSc-ILD⁻ vs controls comparison). For classical alleles and aminoacidic residues, the effect allele (P) indicates the presence of the variant. CI: Confidence interval; EA: Effect allele; OR: Odds ratio; SNP: Single nucleotide polymorphism; SSc-ILD: Systemic sclerosis-associated interstitial lung disease.

Regarding the SNPs, the strongest association corresponded to an intronic protective variant within HLA-DQB1 gene, chr6:32663912:A:G (OR = 0.52 and 95% confidence interval [95%CI] [0.46 – 0.60], p-value = 2.83 × 10⁻²³), which was in high linkage disequilibrium (LD) with HLA-DQA1*01:01:01:01 classical allele and the amino acid glutamine in the position 224 of HLA-DQB1 gene (Figure 2). Additionally, we detected another intronic SNP, located in the second intron of HLA-DPB1 gene and associated with higher risk of SSc-ILD (chr6:33083222:G:A, OR = 2.53 [1.90 – 3.36], p-value = 1.54 × 10⁻⁹). The remaining three SNPs significantly associated were intergenic and potentially involved in SSc-ILD through gene regulation.

Figure 2. Linkage disequilibrium among significant HLA variant types.

Circular plot representing the linkage disequilibrium (LD) structure (R² and D’) among the significant variants identified in the MHC region. The circle shows all LD interactions detected between SNPs (pink), classical HLA alleles (orange), and amino acid residues (yellow). Lines connecting variants represent pairwise LD, with color intensity proportional to R²/D’ values. HLA: “human leukocyte antigen”, LD: “linkage disequilibrium”, MHC: “major histocompatibility complex”, SNP: “single-nucleotide polymorphism”.

Concerning classical alleles, HLA-DQA1*01:01:01:01 was the most strongly associated with the phenotype (OR = 0.54 [0.47 – 0.61], p-value = 4.85 × 10⁻²²). Interestingly, another classical allele of HLA-DQA1 was also associated, HLA-DQA1*03:01:01, conferring protection as well (OR = 0.71 [0.62 – 0.81], p-value = 1.97 × 10⁻¹⁰). Furthermore, we found HLA-DPB1*13:01:01:01 as a risk variant for SSc-ILD (OR = 2.70 [1.93 – 3.79], p-value = 4.4 × 10⁻⁸). It is worth noting that all three classical alleles found were in high LD with three significant SNPs, and two of them also with a significant aminoacidic residue (Figure 2).

With regard to the aminoacidic residues, the most significantly associated with SSc-ILD was the HLA-DQA1 Gln34 (OR = 1.40 [1.28 – 1.55], p-value = 4.22 × 10⁻¹²), followed by aminoacidic residues in HLA-DPB1 (Ile76), HLA-DQB1 (Gln224) and HLA-C (Asp9), all of them conferring risk to SSc-ILD. Interestingly, all these residues, except for HLA-DQB1 Gln224, were in the peptide binding groove of the protein, highlighting their potential implication in the recognition of autoantigens by the HLA system.

It is worth mentioning that two of the significant risk variants were associated with class I HLA genes, the rs2523558 SNP (near HLA-B) and the aminoacidic residue HLA-C Asp9.

Finally, we compared the association of these 12 variants with those reported for SSc vs controls [8]. Interestingly, 8 of the 12 signals were not significantly associated with overall disease, supporting their SSc-ILD specificity. The remaining 4 variants showed smaller effect sizes in overall SSc despite the larger sample size of the study (Supp Table S3).

ATA adjustment maintains class I significance but reduces class II

Given the strong correlation of ATA with SSc-ILD, we decided to perform a complementary analysis including ATA as a covariate in the regression model in order to adjust for its effect. First, in order to check this association, we checked the proportion of ATA/ACA individuals in both SSc-ILD⁺ and SSc-ILD⁻ individuals. We found that the majority of patients in the SSc-ILD⁺ group had ATA antibodies (72.8%), while only 21.5% of SSc-ILD⁻ patients presented it (Supp figure 1). Subsequently, regarding the ATA adjusted analysis, the results showed that out of the 12 significant variants initially identified, four remained significant after adjustment. Two of them are class II HLA variants, and the other two are class I. Interestingly, while the class II associations substantially reduced their significance, the class I variants not only retained significance but showed a slight increase in association strength (Table 2).

Table 2.

Summary statistics of the singificant and SSc-ILD specific variants and their value after adjusting for ATA status as a covariate.

Gene	Variant	Position (hg38)	Effect allele	OR (95% CI)	P-value cond	ATA adjustment
						OR (95% CI)	P-value
Classical alleles
HLA-DQA1	HLA-DQA1*01:01:01:01	32637407	P	0.54 (0.47 – 0.61)	4.85E-22	0.64 (0.56 – 0.73)	4.93E-11
HLA-DQA1	HLA-DQA1*03:01:01	32637415	P	0.71 (0.62 – 0.81)	1.97E-10	0.80 (0.70 – 0.92)	1.69E-03
HLA-DPB1	HLA-DPB1*13:01:01:01	33075955	P	2.70 (1.93 – 3.79)	4.40E-08	1.40 (0.97 – 2.04)	7.30E-02
SNPs
HLA-DQB1	chr6:32663912:A:G	32663912	G	0.52 (0.46 – 0.60)	2.83E-23	0.63 (0.56 – 0.72)	7.90E-12
HLA-DQA1	rs9272353	32636679	C	0.71 (0.62 – 0.80)	8.37E-11	0.80 (0.70 – 0.92)	1.46E-03
HLA-DPB1	chr6:33083222:G:A	33083222	A	2.53 (1.90 – 3.36)	1.54E-09	1.32 (0.97 – 1.80)	7.26E-02
HLA-B	rs2523558	31363260	G	1.37 (1.24 – 1.51)	1.66E-06	1.33 (1.20 – 1.48)	1.06E-07
NCR3	rs3132452	31595463	A	1.42 (1.22 – 1.64)	3.60E-06	1.29 (1.10 – 1.52)	1.80E-03
Aminoacidic residues
HLA-DQA1	HLA-DQA1 Gln34	32641403	P	1.40 (1.28 – 1.55)	4.22E-12	1.25 (1.13 – 1.39)	2.22E-05
HLA-DPB1	HLA-DPB1 Ile76	33080890	P	2.37 (1.81 – 3.12)	3.32E-10	1.32 (0.94 – 1.84)	1.08E-01
HLA-DQB1	HLA-DQB1 Gln224	32660998	P	1.95 (1.61 – 2.38)	1.18E-07	1.57 (1.28 – 1.92)	1.30E-05
HLA-C	HLA-C Asp9	31271844	P	1.27 (1.17 – 1.37)	2.77E-06	1.24 (1.14 – 1.35)	1.32E-06

Variants that remain significant after adjusting are highlighted in boldface. For classical alleles and aminoacidic residues, the effect allele (P) indicates the presence of the variant. ATA: Anti-topoisomerase autoantibody; CI: Confidence interval; OR: Odds ratio; SNP: Single nucleotide polymorphism; SSc-ILD: Systemic sclerosis-associated interstitial lung disease.

Stratified analysis reveals potential progression/severity genetic markers

To investigate potential genetic determinants of SSc-ILD progression/severity, we performed a stratified analysis within SSc-ILD patients using ATA status as a surrogate marker of a more severe clinical phenotype. The stratified meta-analysis identified 3,814 variants reaching statistical significance, including 16 classical HLA alleles, 83 aminoacidic residues, and 3,715 SNPs. After performing conditional analysis within each variant category, we identified 6 classical alleles, 5 amino acid residues, and 7 SNPs as jointly significant associations (Supp Table S4).

Notably, the classical alleles HLA-DPB1*13:01:01:01 and HLA-DQA1*03:01:01, as well as the aminoacidic residue HLA-DPB1 Ile76, were among the jointly significant signals detected, overlapping with the variants identified as lead signals in the main comparison (SSc-ILD⁺ vs SSc-ILD⁻).

Functional assessment of SNPs

Since all the significant SNPs we identified were non-coding, we investigated whether they contribute to disease pathogenesis through regulatory roles. In order to assess this, we carried out functional annotation using FUMA GWAS webtool and GTEx v.8 databases. Two of the SNPs (chr6:32663912:A:G and chr6:33083222:G:A) were not present in the FUMA GWAS database, thus, they could not be included in the analysis. However, all the SNPs that were included showed overlap with at least one functional category such as QTLs or Chr_Int, suggesting they influence disease by affecting regulatory elements (Supp figure 2, Figure 3).

Figure 3. Summary of functional annotation of the significant SNPs.

Figure highlighting the functional role of three significant SNPs (and/or proxies) identified. Genes were prioritized based on positional mapping (orange), eQTL effects in relevant tissues/cell types (green), sQTL effects (blue), and chromatin interactions (purple). The presence of an annotation in a specific category is indicated by its corresponding color. eQTL: “expression quantitative trait loci”, SNP: “single nucleotide polymorphism”, sQTL: “splicing quantitative trait loci”.

In lung tissue, 25 genes showed altered expression associated with at least one of the three SNPs. These prioritized genes included HLA class I and class II genes, complement genes as C4A, or other genes as MICA and MICB. Notably, we also found that rs2523558 and rs3132452 SNPs, although they are independent in terms of LD (R² and D’ < 0.1), contributed jointly to the regulation of four genes: CCHCR1, PSORS1C1, PSORS1C2 and MICA. Finally, regarding sQTL annotation, rs9272353 and rs3132452 variants were associated with the modulation of the isoform proportions of seven genes, including HLA-DQA1, HLA-DRB1, and NCR3 (Figure 3). In total, our results prioritized 66 genes as potentially relevant for SSc-ILD (Supp figure 2, Figure 3).

Integration of genetic and clinical predictors improves SSc-ILD risk stratification

To develop a clinically relevant model, we combined PRS, autoantibody status, SSc subtype, and sex to predict SSc-ILD. Using the training cohort, PRSice-2 identified an optimal model including six SNPs out of the ten given as an input (Supp Table S5). Application of this PRS, or any of the clinical/demographic scores to the independent testing cohort resulted in AUCs below 0.7. However, interestingly, the composite score integrating all components achieved an AUC of 0.754 (Figure 4a). Remarkably, DeLong’s statistical test for comparing ROC curves showed that there was a statistically significant difference between the composite score and each of the individual scores considered (Supp Table S6). Furthermore, we tested the difference between this composite score and a score including all variables except the PRS, finding a significant difference (p-value = 8.64 × 10⁻³) and confirming the added value of integrating genetic information into predictive scores.

Figure 4. Predictive performance and risk stratification for SSc-ILD.

A) ROC curves and area under the curve (AUC) for the genetic PRS (green), ATA-only (blue), and composite score (red) models. B) Odds ratios (OR) and 95% confidence intervals (CI) for SSc-ILD across percentile thresholds of the composite score. Dashed red line indicates the 85th percentile, which was selected as the high-risk cutoff. ATA: “antitopoisomerase autoantibody”, PRS: “polygenic risk score”, SSc-ILD: “systemic sclerosis associated interstitial lung disease”.

Additionally, to ensure that the risk estimates were consistent in an independent cohort, we ran a validation analysis in a SSc-ILD cohort from UK Biobank. Using the same parameters as in the testing cohort, we were able to validate the predictive capacity of our genetic model. For instance, the AUC in the testing cohort was 0.585, and in the validation cohort 0.589.

Finally, using our composite score, we stratified individuals in the testing cohort into percentile-based risk categories to define a high-risk subgroup for SSc-ILD. The 85^th percentile group corresponded to a RR of 3.54 (95% CI: 2.77 – 4.55) for SSc-ILD development, indicating that patients in this high-risk group present more than three times higher risk of developing this clinical manifestation (Figure 4b).

DISCUSSION

In this study, we performed the largest MHC-focused genetic analysis of SSc-ILD to date, including 2,412 SSc-ILD⁺, 3,550 SSc-ILD⁻ and 15,076 controls. By analysing classical alleles, aminoacidic residues and SNPs, we show that the MHC plays a key role in SSc-ILD susceptibility and identify novel class I and class II signals with potential pathogenic relevance. We also examined the impact of ATA⁺ status on these associations, functionally annotated the associated SNPs and developed a composite score to identify patients at increased risk of SSc-ILD.

We detected 12 significant variants comparing SSc-ILD⁺ against SSc-ILD⁻, eight of which were not significant in the overall disease [8], with the remaining showing greater effects in SSc-ILD. Nine of the 12 variants localized to HLA class II, including HLA-DQA1*01:01:01:01 and HLA-DPB1*13:01:01:01, which have previously been associated with SSc-related pulmonary fibrosis [19], and ATA positivity [8,11,19,32]. Interestingly, HLA-DPB1*13:01:01:01, HLA-DQA1*03:01:01, and the aminoacidic residue HLA-DPB1 Ile76 were also significant in the ATA-stratified analysis within SSc-ILD, supporting a contribution of these variants to disease severity in the ILD subset. However, because ATA is an imperfect surrogate of progression, longitudinal studies will be needed to determine whether these variants influence fibrotic trajectories. Furthermore, to disentangle direct effects on SSc-ILD from those mediated by ATA, we adjusted the analysis on ATA status. After adjustment, all HLA class II associations were markedly attenuated, whereas HLA class I associations remained significant and became slightly stronger. This pattern is consistent with a model in which HLA class II variation contributes to SSc-ILD mainly through ATA-related mechanisms, whereas HLA class I variation acts through a distinct, ATA-independent pathway. This would also be consistent with ATA autoantibodies participating in the pathogenic processes of SSc-ILD, which would explain why only HLA class II variants lose significance. This interpretation is supported by experimental evidence showing that ATA-containing immune complexes can incorporate nucleic acids and activate Toll-like receptors that induce profibrotic programs [33,34], as well as by clinical evidence linking B-cell activity in ATA⁺ patients with pulmonary fibrosis severity [35].

Beyond HLA class II, we detected two significant HLA class I signals: rs2523558 near HLA-B and Asp9 in HLA-C. As HLA class I molecules present peptides to CD8⁺ T cells and are expressed on essentially all nucleated cells [36], these associations point to a role for CD8⁺ T-cell-mediated antigen recognition in SSc-ILD. Variation in HLA class I molecules could affect the self-peptide repertoire, promoting a cytotoxic response. In fact, class I HLA variants have previously been reported for overall SSc [8]. Furthermore, CD8⁺ T-cell enrichment has been observed in skin from early diffuse SSc patients [37], together with reduced numbers in peripheral blood [38], suggesting redistribution to affected organs. Single-cell RNA-seq studies further support this mechanism. Shimagami et al. [39] reported an enrichment of a CD8⁺ effector-memory population in peripheral blood and lung from SSc-ILD patients, characterized by a strong type-II interferon signature, and expression of migration markers as CXCR3 or CCR5. Moreover, Padilla et al. [40] found increased CD8⁺ tissue-resident memory T-cells in SSc-ILD lungs showing upregulation of TCR signaling, exhaustion markers, and fibrosis-associated pathways. Altogether, this evidence supports a model linking altered HLA-I presentation of self-peptides to persistent CD8⁺ T-cell activation leading to fibrotic remodeling.

To give functional insight into the associated non-coding SNPs, we annotated them using FUMA GWAS and GTEx v.8. In total, we prioritized 66 genes, and the expression of 25 of them was regulated in lung tissue by at least one of the three SNPs included in the analysis. Additionally, the expression levels of several genes, including HLA-DRB1, HLA-DQA1 and HLA-DQB1, were predicted to be regulated across other relevant tissues/cell types considered, as blood, immune cells and fibroblasts. Although some prioritized genes have been associated with other ILDs as rheumatoid arthritis-associated ILD (RA-ILD) and myositis-associated ILD, the specific variants differed, suggesting partly distinct pathogenic mechanisms despite their clinical overlap [41,42]. Furthermore, we found that rs2523558 is a lung eQTL for C4A, with the risk allele associated with reduced expression. This result aligns with previous associations of C4 variation with SSc, systemic lupus erythematosus (SLE) or primary Sjögren's Syndrome [43–46]. Moreover, given the role of C4 in clearing immune-complexes [47], this also supports their aforementioned pathogenic role in SSc-ILD. Finally, regulation of MICA and MICB further supports cytotoxic involvement in SSc-ILD [48]. Both genes have also been associated with SLE and RA [49–53], and a recent study using exome sequencing and machine learning highlighted MICB as a genetic risk locus for SSc [54].

Finally, our composite score integrating the effects of six HLA SNPs, the presence of ATA/ACA, SSc subtype and sex achieved an AUC of 0.754, significantly outperformed models based on any single variable. The improvement observed after adding genetic factors to clinical and demographic variables (p-value = 0.0086) further supports the value of combining these dimensions for SSc-ILD risk prediction, which is consistent with previous observations in SSc [55]. Since current guidelines already recommend baseline HRCT screening for all SSc patients [6], the main clinical utility of this score lies in follow-up. In particular, it could help clinicians identifying SSc-ILD⁻ patients at high risk of future SSc-ILD development who may benefit from closer surveillance, especially in late-onset SSc-ILD cases [56]. Beyond risk prediction, the distinction between ATA-linked HLA class II associations and ATA-independent HLA class I signals also has potential mechanistic and therapeutic implications. For instance, our ATA-dependent HLA class II associations pointing towards the role of immune complexes, supports the rationale for B cell–directed therapies, and provide an opportunity for therapies targeting TLR signaling, as currently explored in SLE (NCT05540327, NCT04895696) or plaque psoriasis (NCT01899729). Additionally, the ATA-independent HLA class I associations, together with the regulatory signals at MICA and MICB, point towards CD8⁺ T cells and NK cells, suggesting that modulation of cytotoxic lymphocyte activity may also hold therapeutic promise in SSc-ILD.

Despite its strengths, this study has several limitations. First, less than 10% of patients from earlier cohorts were not evaluated by HRCT because imaging was performed selectively according to clinical risk factors. This may have led to some underdiagnosis of SSc-ILD, although the impact on statistical power is likely limited given the small proportion affected. Second, as this study focuses on the MHC region, the derived analyses, including the PRS incorporated into the composite score, only include variants from this locus. Incorporating risk variants from other genomic regions in future studies may further improve predictive performance and provide a more broad view of the genetic architecture of this clinical manifestation. Finally, our analyses were restricted to individuals of European ancestry, limiting generalizability. Validation in higher-risk populations such as African Americans, will be essential [6].

In summary, this comprehensive MHC analysis identifies loci specifically associated with SSc-ILD and supports distinct contributions of HLA class I and class II variation to disease pathogenesis. HLA class I signals point towards CD8⁺ T-cell participation, whereas HLA class II associations are consistent with a contribution of ATA-related mechanisms. We also developed a composite score integrating genetic, clinical and demographic variables to help identify patients at high risk of SSc-ILD. Together, these findings refine the immunogenetic architecture of SSc-ILD and may inform biological stratification and clinical monitoring.

KEY MESSAGES.

What is already known on this topic

• Systemic sclerosis–associated interstitial lung disease (SSc-ILD) is the leading cause of mortality in SSc.

• The MHC region is the strongest genetic risk locus for SSc, but its specific contribution to SSc-ILD has not been clarified and previous genetic studies in this clinical manifestation are often underpowered.

What this study adds

• This study provides the largest MHC genetic analysis of SSc-ILD to date, identifying twelve specific associations in class I and class II HLA genes.

• Class II HLA associations lose significance after adjustment for ATA, whereas class I signals remain robust, revealing distinct ATA-dependent and ATA-independent pathogenic pathways.

• Adding genetic information to SSc-ILD risk scores improves its ability to identify patients at high risk of developing SSc-ILD.

How this study might affect research, practice, or policy

• These findings provide genetic evidence supporting pathogenic mechanisms in SSc-ILD, involving autoantibody-driven processes and CD8⁺ T cell-mediated pathways.

• Our composite risk score results encourage the inclusion of genetic data in predictive scores for early SSc-ILD screening and monitoring.

• The identified genetic loci and pathways provide potential biomarkers and therapeutic targets relevant to the management of SSc-ILD.

DATA AVAILABILITY

Summary statistics for the 3 comparisons performed in this study are publicly available in the following Figshare repository: https://doi.org/10.6084/m9.figshare.30773531