Cross-trait GWAS in COVID-19 and systemic sclerosis reveals novel genes implicated in fibrotic and inflammation pathways

Abstract

Objectives

Coronavirus disease 2019 (COVID-19) and SSc share multiple similarities in their clinical manifestations, alterations in immune response and therapeutic options. These resemblances have also been identified in other immune-mediated inflammatory diseases where a common genetic component has been found. Thus, we decided to evaluate for the first time this shared genetic architecture with SSc.

Methods

For this study, we retrieved genomic data from two European-ancestry cohorts: 2 597 856 individuals from The COVID-19 Host Genetics Initiative consortium, and 26 679 individuals from the largest genomic scan in SSc. We performed a cross-trait meta-analyses including >9.3 million single nucleotide polymorphisms. Finally, we conducted functional annotation to prioritize potential causal genes and performed drug repurposing analysis.

Results

Our results revealed a total of 19 non-HLA pleiotropic loci, including 2 novel associations for both conditions (BMP1 and PPARG) and 12 emerging as new shared loci. Functional annotation of these regions underscored their potential regulatory role and identified potential causal genes, many of which are implicated in fibrotic and inflammatory pathways. Remarkably, we observed an antagonistic pleiotropy model of the IFN signalling between COVID-19 and SSc, including the well-known TYK2 P1104A missense variant, showing a protective effect for SSc while being a risk factor for COVID-19, along with two additional novel pleiotropic associations (IRF8 and SENP7). Finally, our findings provide new therapeutic options that could potentially benefit both conditions.

Conclusion

Our study confirms the genetic resemblance between susceptibility to and severity of COVID-19 and SSc, revealing a novel common genetic contribution affecting fibrotic and immune pathways.

Rheumatology key messages.

• Shared loci between SSc and coronavirus disease 2019 (COVID-19) suggest key roles for fibrotic and immune-related pathways.

• Certain risk factors show similar effects for COVID-19 and SSc (e.g. PPARG, BMP1 or NFKB1) while others exhibit opposite effects (e.g. TYK2, IRF8 or SENP7).

• Genetic evidence suggests new potential treatments for COVID-19 and/or SSc.

Introduction

Coronavirus disease 2019 (COVID-19) and immune-mediated inflammatory diseases (IMIDs) share similarities in their clinical manifestations, alterations in immune response and therapeutic options [1–4]. In fact, several studies have reported greater susceptibility of autoimmune development in patients with COVID-19 [5]. In addition, IMIDs patients are also at a higher risk of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and severe COVID-19 due to their compromised immune status [6, 7]. Particularly, severe COVID-19 manifests as acute respiratory distress syndrome, characterized by an immune imbalance that leads to a cytokine storm, marked by elevated levels of proinflammatory cytokines [3]. Of note, this phenomenon is also present in many IMIDs, such as SSc, where high levels of inflammatory cytokines such as IL-1, IL-6, IL-17, TNF-α and type I IFN contribute to its pathogenesis [3, 8, 9]. Notably, both SSc and severe COVID-19 patients also share pulmonary complications such as the development of interstitial lung disease (ILD), which can progress to pulmonary fibrosis [10, 11].

Genome-wide association studies (GWAS) have identified multiple genetic variants associated with susceptibility to or severity of COVID-19 [12, 13]. Interestingly, many of these associated-genes are related to inflammation and innate immunity, which are well-known altered pathways in IMIDs, such as RA, SLE and SSc, underscoring the genetic overlap between COVID-19 and IMIDs. In fact, previous studies have systematically evaluated the shared genetic architecture among COVID-19, RA and SLE, revealing new common risk factors for these conditions [14, 15].

Considering all the above, we aimed to explore for the first time the shared genetic component of COVID-19 and SSc by conducting a cross-trait meta-analysis of genomic data. We identified new pleiotropic genetic loci that improve our knowledge of the shared relevant pathways and nominate new therapeutic targets.

Methods

Study cohorts

A summary of the COVID-19 and SSc cohorts included in the present study is provided in Supplementary Table S1, available at Rheumatology online.

GWAS summary statistics for COVID-19 data were obtained from the latest version (data release 7) of the public repository The COVID-19 Host Genetics Initiative (HGI) [16]. The HGI provides meta-analyses results from data of over 80 study cohorts across 35 countries stratified based on the ancestry of the individuals and COVID-19-related phenotypes. Each specific study conducted quality controls independently but according to consortium guidelines. These quality controls included filtering single nucleotide polymorphisms (SNPs) with low call rates (<0.95 or <0.98), missing difference <0.02 and Hardy–Weinberg equilibrium deviations (P-value <1 x 10⁻⁶ in controls or <1 x 10⁻¹⁰ in cases). Samples with call rates <0.98 were also excluded. To avoid biases, suggested covariates included age, sex, age × sex and the first 20 principal components. Specifically, we used summary statistics from three models of European ancestry: (i) severe COVID-19 (A2_eur model), analysing COVID-19-confirmed individuals who required respiratory support and/or died because of the disease (13 769 COVID-19 patients and 1 072 442 controls); (ii) hospitalized COVID-19 (B2_eur model), including COVID-19-confirmed individuals who were hospitalized regardless of whether they needed respiratory support or not (32 519 COVID-19 patients and 2 062 805 controls); and (iii) SARS-CoV-2 infection (C2_eur model), comprising all individuals who were infected by SARS-CoV-2 virus (122 616 cases and 2 475 240 controls). It is worth noting that these models are not mutually exclusive. Individuals included in the severe COVID-19 model are also part of the hospitalized and infection models, and those in the hospitalized model are included in the SARS-CoV-2 infection model. In order to assess potential bias due to population structure or other confounders, we estimated the genomic inflation factor (λ), after excluding the MHC region, for each model of COVID-19. We also rescaled it for an equivalent study of 1000 cases and 1000 controls (λ₁₀₀₀) (Supplementary Table S1, available at Rheumatology online).

The SSc cohort comprised GWAS data from a previous study of a total of 26 679 individuals (9 095 SSc patients and 17 584 controls) from 14 European cohorts [17]. All SSc patients fulfilled the 1980 ACR classification criteria, or the criteria proposed by LeRoy and Medsger for early SSc [18, 19]. The study was approved by the Ethics Committee of the ‘Consejo Superior de Investigaciones Científicas’ (CSIC) and all individuals included in the study voluntarily signed the written informed consent in accordance with the principles of the Declaration of Helsinki.

Imputation and analysis of the SSc dataset

To increase the number of SNPs to analyse, we applied genomic imputation, a statistical technique used to infer unobserved genotypes based on reference panels encompassing genetic information from a large number of individuals. We performed this analysis using the TOPMed imputation server and the TOPMed Imputation reference panel (https://imputation.biodatacatalyst.nhlbi.nih.gov/). As previously described [17], stringent pre-imputation quality controls were performed, including filtering SNPs with genotyping call rates <0.98, minor allele frequencies <0.01 and those deviating from Hardy–Weinberg equilibrium (P-value < 0.001 in both cases and controls). Samples with call rates <0.95 and those considered relatives and/or duplicates were excluded. To control for ancestry, principal components were calculated, and samples showed >4 s.ds from the cluster centroids of each cohort and were further removed from further analyses. After imputation, SNPs with a squared correlation (Rsq) <0.3 and a minor allele frequency <0.01 were excluded from further analyses. Next, association testing for allele dosages was performed using logistic regression with PLINK v2.0 (https://www.cog-genomics.org/plink/2.0/) and adjusting by sex and the first five principal components, as previously described [17]. Additionally, prior to the meta-analysis, each individual cohort was adjusted by their specific λ, without the extended MHC region. This was performed by multiplying each standard error by the square root of the calculated λ. Inverse variance weighted meta-analyses were performed using the software MetaSoft. The Cochran’s Q test was calculated to account for heterogeneity. A fixed effect model was applied for those SNPs without evidence of heterogeneity (Cochran’s Q P-value >0.05), and the conventional random effect model (RE) for SNPs displaying heterogeneity of effects between study cohorts (Cochran’s Q P-value <0.05). Finally, λ and λ₁₀₀₀ without the HLA region were also estimated (Supplementary Table S1, available at Rheumatology online).

Genetic correlation

To examine the level of genetic similarity among the diseases included in our study, we applied a cross-trait linkage disequilibrium (LD) score regression using LDSC v1.0.1 software (https://github.com/bulik/ldsc). This method uses genome-wide summary statistics and effectively accounts for LD. Precalculated LD scores were derived from the 1000 Genomes European reference population. We assessed pairwise genetic correlations (r_g) between each of the three COVID-19-related phenotypes and SSc using HapMap3 variants only, as recommended by the LDSC v1.0.1 software.

Cross-trait meta-analyses

We conducted three meta-analyses on the summary statistics of each of the COVID-19-related phenotypes and SSc using the software MetaSoft. Only SNPs shared by the meta-analysed summary statistics were considered. Due to the complexity of the MHC region and its known association with both traits, this region was excluded from the analyses. We employed a fixed-effect model for SNPs with no indication of variation across studies (Cochran’s Q P-value >0.05). For those SNPs showing diverse effects between studies (Cochran’s Q P-value <0.05) Han and Eskin’s random effect model (RE2), which is optimized to detect associations under heterogeneity, was applied. λ was also calculated for the three meta-analyses performed (Supplementary Table S1, available at Rheumatology online). Those SNPs showing genome-wide significance in the meta-analysis (P-value <5 × 10⁻⁸) and nominal significance (P-value <0.01) in each specific disease were considered shared variants for both traits. We applied this dual-threshold approach to ensure that the findings are corrected for multiple testing while excluding associations driven by only one of the diseases. Subsequently, to detect independent signals within the shared genomic regions identified, we conducted a conditional stepwise analysis using the COJO tool in GCTA 1.92.1 (https://yanglab.westlake.edu.cn/software/gcta/#COJO). In this analysis, we first identified the most statistically significant SNP in each region, termed the lead SNP. We then re-ran the association analysis including the lead SNP as a covariate in the model to control for its effect. This allowed us to detect any SNPs that remained significantly associated (P-value <5 × 10⁻⁸) after adjusting for the lead SNP, reflecting independent genetic associations. This process was repeated iteratively until no variant reached significance. In addition, to ensure independence, we only considered variants in low LD with the lead variant (r² < 0.1 and D′ < 0.5). Lead variants were mapped to the nearest gene, and novel associations were established if the shared genomic regions had not been previously associated with COVID-19 and/or SSc (P-value <5 × 10⁻⁸). This was determined by searching the lead SNP and the SNPs in LD with them (r² > 0.2) in the NHGRI-EBI GWAS catalogue, using the LDtrait search tool (https://ldlink.nih.gov/?tab=ldtrait).

Functional annotation of shared genetic variants and gene prioritization

To prioritize potential causal genes, we conducted functional annotation of the lead SNPs and their SNPs in high LD (r² > 0.9), namely their proxies, by using the SNP2GENE function of FUMA GWAS (https://fuma.ctglab.nl). The parameters used are listed in Supplementary Table S2, available at Rheumatology online. In summary, we annotated relevant genes based on positional mapping, expression quantitative trait loci (eQTL) mapping and chromatin interaction mapping. For positional mapping, a gene was considered physically near when the distance between the lead SNP and the starting/finishing position of the gene was lower than 10 000 base pairs, as the default parameter established in FUMA webtool. We also assessed if the lead SNP or its proxies were missense variants. Additionally, we queried GTEx v.8 database (https://gtexportal.org/home/) for splicing QTLs. We focused the QTL annotation on four relevant tissues/cell types in the diseases of interest (whole blood, immune cells, skin and lung). Moreover, we also retrieved information of the lead SNPs from the Open Target Genetics database (https://genetics.opentargets.org) for serum protein QTLs (pQTL) and the V2G score. Only coding genes with functional information for any positional and/or QTL category were considered candidate genes.

Drug repurposing

Finally, we conducted a drug repurposing analysis to identify potential candidate drugs for COVID-19 or SSc treatment. We used the DrugBank V.5.0 (https://go.drugbank.com) database to explore whether the proteins encoded by the prioritized candidate genes are target for approved drugs. To select the most potentially promising drugs, we prioritized approved drugs that were indicated and/or had clinical trials in COVID-19 or IMIDs. Additionally, we manually selected drugs based on literature search.

Results

Genetic correlation

First, with the aim to determine the extent of overlap in the genetic architecture between the three COVID-19-related phenotypes and SSc, we calculated pairwise genome-wide genetic correlations. Notably, these analyses revealed a modest, statistically significant genetic correlation between severe COVID-19 and SSc (r_g = 0.154 ± 0.070, P-value = 0.029). In addition, no statistically significant genetic correlation was observed between the hospitalized COVID-19 model and SSc (r_g = 0.131 ± 0.068, P-value = 0.057) or between SARS-CoV-2 infection model and SSc (r_g = 0.057 ± 0.076, P-value = 0.456).

Identification of shared loci

Subsequently, to identify shared loci between COVID-19 and SSc, we conducted three cross-trait meta-analyses encompassing >9.3 million variants, each involving one of the three phenotypes of COVID-19. No evidence of genomic inflation was observed in any of the meta-analyses; λ values are shown in Supplementary Table S1, available at Rheumatology online. In total, our analysis revealed 19 independent non-HLA pleiotropic loci associated with SSc and at least one of the three COVID-19-related phenotypes, including 9 for SARS-CoV-2 infection, 12 for hospitalization and 10 for severe illness (Fig. 1, Table 1, Supplementary Fig. S1 and Tables S3–S5, available at Rheumatology online). Interestingly, many of the genomic regions identified mapped to genes crucial for the immune response (NFKB1, IL12RB2, TYK2, IRF8 and IRF5) or implicated in fibrosis (PPARG and BMP1).

Figure 1.

Associated signals for the meta-analyses of COVID-19 and SSc and its effects. (A) Circular Manhattan plots of the three meta-analyses. Chromosomes are indicated in the inner layer and the genome-wide significance threshold is marked (5 × 10⁻⁸). Shared SNPs (P-value for each trait <0.01) are highlighted. (B) Effect of the shared loci reaching significance in the meta-analyses. Each circle represents a disease model separately, with the reference line indicating an OR of 1. COVID-19: coronavirus disease 2019; SNP: single nucleotide polymorphism; OR: odds ratio

Table 1.

Summary statistics for the three meta-analyses of COVID-19 with SSc

Locus	rsID	EA	Nearest gene	Comparison	Meta-analysis		SSc		COVID-19
					P-value Q	P-value	P-value	OR (95% CI)	P-value	OR (95% CI)
1p31.3	rs755826	C	IL12RB2 COV	SSc—COVID-19 hosp	1.16E-04	1.58E-08	1.30E-08	1.18 (1.11–1.25)	9.10E-03	1.04 (1.01–1.07)
3p25.2	rs310751	C	PPARG	SSc—COVID-19 sev	1.41E-01	1.42E-09	9.31E-08	0.83 (0.78–0.89)	1.34E-03	0.89 (0.83–0.96)
3p21.31	rs13086080	T	SACM1L SSc	SSc—COVID-19 sev	4.80E-10	1.39E-11	3.00E-03	1.09 (1.03–1.15)	1.23E-11	0.88 (0.85–0.91)
				SSc—COVID-19 hosp	3.57E-08	8.46E-12	3.00E-03	1.09 (1.03–1.15)	7.65E-12	0.92 (0.89–0.94)
3q12.3	rs79181492	A	SENP7 SSc	SSc—SARS-CoV-2	1.37E-05	3.98E-15	8.27E-03	1.07 (1.02–1.12)	1.17E-15	0.96 (0.95–0.97)
4p16.3	rs9799610	T	DGKQ COV	SSc—SARS-CoV-2	5.31E-06	4.74E-08	3.69E-08	1.14 (1.09–1.19)	4.84E-03	1.02 (1.01–1.03)
4q24	rs57463114	G	NFKB1 COV	SSc—COVID-19 hosp	1.14E-04	1.08E-08	1.08E-08	1.13 (1.08–1.18)	7.72E-03	1.03 (1.01–1.05)
				SSc—SARS-CoV-2	2.91E-06	2.50E-09	1.08E-08	1.13 (1.08–1.18)	5.82E-04	1.02 (1.01–1.03)
7q32.1	rs3778753	G	IRF5	SSc—COVID-19 hosp	2.26E-13	1.47E-21	6.97E-22	1.23 (1.18–1.29)	2.46E-03	1.03 (1.01–1.05)
8p21.3	rs73225841	A	BMP1	SSc—COVID-19 hosp	3.22E-01	4.36E-08	9.52E-03	0.84 (0.73–0.96)	8.51E-07	0.90 (0.86–0.94)
8q12.1	rs9298039	C	RAB2A	SSc—COVID-19 sev	1.56E-01	6.78E-11	1.17E-06	1.11 (1.06–1.16)	4.65E-06	1.07 (1.04–1.10)
				SSc—COVID-19 hosp	3.50E-02	3.84E-12	1.17E-06	1.11 (1.06–1.16)	5.08E-08	1.06 (1.03–1.08)
				SSc—SARS-CoV-2	1.32E-04	1.16E-08	1.17E-06	1.11 (1.06–1.16)	3.24E-05	1.02 (1.01–1.03)
9q34.2	rs7849887	A	ABO SSc	SSc—SARS-CoV-2	2.05E-01	1.31E-11	9.92E-03	1.07 (1.02–1.12)	1.73E-10	1.03 (1.02–1.04)
11p13	rs11604331	G	CAT SSc	SSc—COVID-19 sev	1.26E-06	2.60E-08	8.64E-03	0.94 (0.90–0.99)	1.40E-08	1.09 (1.06–1.12)
				SSc—COVID-19 hosp	9.00E-08	5.45E-09	8.64E-03	0.94 (0.90–0.99)	2.78E-09	1.06 (1.04–1.08)
15q24.1	rs4886410	C	CSK	SSc—COVID-19 hosp	2.57E-08	2.07E-15	8.90E-15	0.85 (0.81–0.88)	4.94E-04	0.97 (0.95–0.98)
				SSc—SARS-CoV-2	1.63E-11	1.47E-15	8.90E-15	0.85 (0.81–0.88)	1.22E-04	0.98 (0.97–0.99)
16p11.2	rs71389500	T	ITGAM COV	SSc—SARS-CoV-2	1.27E-08	3.78E-08	8.24E-07	1.17 (1.10–1.24)	7.64E-05	0.97 (0.96–0.99)
16q24.1	rs11644034	A	IRF8 COV	SSc—COVID-19 sev	4.73E-15	7.18E-14	1.30E-14	0.81 (0.77–0.86)	9.25E-03	1.05 (1.01–1.09)
17q21.1	rs2305479	T	GSDMB	SSc—COVID-19 sev	6.87E-03	3.69E-10	1.78E-08	1.12 (1.08–1.17)	2.90E-04	1.05 (1.02–1.08)
				SSc—COVID-19 hosp	5.00E-04	1.54E-10	1.78E-08	1.12 (1.08–1.17)	5.42E-05	1.04 (1.02–1.06)
				SSc—SARS-CoV-2	1.12E-06	2.12E-08	1.78E-08	1.12 (1.08–1.17)	2.68E-03	1.01 (1.00–1.02)
17q21.31	rs63750417	T	MAPT SSc	SSc—COVID-19 sev	1.48E-01	1.08E-15	9.15E-04	0.92 (0.88–0.97)	9.95E-14	0.88 (0.85–0.91)
				SSc—COVID-19 hosp	7.22E-01	5.53E-18	9.15E-04	0.92 (0.88–0.97)	1.37E-15	0.91 (0.89–0.93)
19p13.2	rs34536443	C	TYK2	SSc—COVID-19 sev	2.81E-16	7.33E-26	1.88E-04	0.77 (0.68–0.88)	3.03E-25	1.46 (1.36–1.57)
				SSc—COVID-19 hosp	2.14E-11	7.10E-21	1.88E-04	0.77 (0.68–0.88)	3.83E-20	1.26 (1.20–1.33)
				SSc—SARS-CoV-2a	9.10E-06	2.92E-08	4.16E-04	0.89 (0.83–0.95)	1.55E-07	1.03 (1.02–1.05)
19p13.11	rs10410017	T	MPV17L2 COV	SSc—COVID-19 sev	2.03E-04	2.72E-09	2.50E-09	0.87 (0.83–0.91)	9.83E-03	0.96 (0.94–0.99)
19q13.33	rs2248949	G	NR1H2 SSc	SSc—COVID-19 sevb	4.23E-01	3.32E-08	4.21E-04	1.09 (1.04–1.14)	1.61E-05	1.07 (1.04–1.10)
				SSc—COVID-19 hosp	2.15E-01	5.81E-10	5.44E-04	1.09 (1.04–1.14)	1.24E-07	1.05 (1.03–1.07)

Genes in bold represent novel associations for both diseases.

^aStatistical data for this comparison correspond to the SNP rs78295726.

^bStatistical data for this comparison correspond to the SNP rs788401. 

COVID-19: coronavirus disease 2019; EA: effect allele; OR: odds ratio; P-value Q: P-value associated with the Cochran’s Q test for heterogeneity; _COV: novel variant only for COVID-19; _SSc: novel variant only for SSc; COVID-19 sev: severe COVID-19; COVID-19 hosp: hospitalized COVID-19; SARS-CoV-2: severe acute respiratory syndrome coronavirus 2 infection.

Notably, two loci annotated to BMP1 and PPARG genes were identified as novel associations for both COVID-19-related phenotypes and SSc. In addition, 12 genomic regions represented novel associated loci for COVID-19 or SSc, thus emerging as new pleiotropic loci. Interestingly, six of the common associations showed opposite effects across traits. Among these, three signals annotated to SACMIL, SENP7 and ITGAM genes comprise risk factors for SSc and protective factors for COVID-19-related phenotypes. Conversely, the signals annotated to CAT, IRF8 and TYK2 genes represent susceptibility factors for COVID-19-related phenotypes while showing a protective effect for SSc (Fig. 1, Table 1).

Functional annotation of pleiotropic loci

Functional annotation of the pleiotropic variants revealed that three of the lead SNPs were non-synonymous variants, those mapping to GSDMB (rs2305479_G304R), MAPT (rs63750417_P277L) and TYK2 (rs34536443_P1104A) genes. Interestingly, these three variants are predicted to be damaging to the protein structure or function with high confidence according to PolyPhen-2 scores (0.999, 0.996 and 0.999, respectively) [20]. Additionally, >20 proxy variants were also missense, such as those mapping to RAB2A (rs2981277_L68P) and ITGAM (rs1143679_R77H) genes (Fig. 2).

Figure 2.

Summary of functional annotation of the identified shared variants. The figure highlights lead SNPs and/or proxies overlapping with the regulatory features analysed. Variants were categorised based on positional mapping, eQTL effects, other QTL effects, and chromatin interactions. The presence of an annotation in a specific category is indicated within the corresponding column. *Lead SNP or proxies are missense variants for the gene. CI: chromatin interaction; eQTL: expression quantitative trait loci; pQTL: protein quantitative trait loci; sQTL: splicing quantitative trait loci; V2G: variant to gene; A2: severe COVID-19 model; B2: hospitalized COVID-19 model; C2: SARS-CoV-2 infection model; SNP: single nucleotide polymorphism; COVID-19: coronavirus disease 2019; SARS-CoV-2: severe acute respiratory syndrome coronavirus 2

Considering that most of the pleiotropic variants and their proxies reside in non-coding regions, we also checked the overlap of the lead variants and their proxies with regulatory functional annotations to prioritize potential causal genes. Notably, 95% of the pleiotropic loci showed overlap with any functional categories such as QTLs in relevant tissues or cell types or chromatin interactions, suggesting they might influence disease by disrupting regulatory elements. We observed that most of the pleiotropic variants act as eQTLs in blood, immune cells, skin or lung tissue. For instance, the genetic variants annotated to the integrin ITGAM modify the gene expression levels of another integrin family member, ITGAX, in whole blood. Interestingly, this eQTL showed a physical interaction between the genetic variant and the ITGAX promoter, providing further evidence of its potential causal role. We also observed that rs10410017, annotated to MPV17L2 gene, acts as an eQTL for nine different genes, including IL12RB1 and IFI30, supporting the implication of immune response genes in the diseases. In addition, six of the lead variants affect serum protein levels (pQTLs). Specifically, the TYK2 missense variant (rs34536443_P1104A) altered the protein levels of ICAM1 and ICAM5, two intercellular adhesion molecules part of the immunoglobulin superfamily [21]. In total, our results prioritized 99 potentially pleiotropic causal genes for COVID-19-related phenotypes and SSc (Fig. 2 and Supplementary Fig. S2, available at Rheumatology online).

Candidate drugs for repurposing

In order to find potential new drugs that could be used for the treatment of COVID-19 and/or SSc, a drug repurposing analysis was carried out using the genes from the functional annotation. Eleven out of the 99 proteins encoded by the functionally relevant candidate genes were targeted by approved drugs with indication and/or clinical trials for COVID-19 or IMIDs (Supplementary Fig. S2, available at Rheumatology online). In total, we identified 25 drugs meeting these criteria (Table 2 and Supplementary Table S6, available at Rheumatology online).

Table 2.

Summary of relevant drugs from the drug repurposing analysis

Locus	Associated gene	Target	Drug	Type	Current indication	Current clinical trials for other diseases
3p25.2	PPARG	PPARG	Indomethacin	Small molecule	RA, OA, AS, acute pain	COVID-19, UC
15q24.1	CSK	SRC	Nintedanib	Small molecule	IPF, progressive fibrosing ILD including SSc patients, cell lung carcinoma	COVID-19, SSc, ILD, pulmonary fibrosis
		CSK	Dasatinib	Small molecule	Leukaemia	COVID-19, SSc
19p13.11	IL12RB1	IL12B	Ustekinumab	mAb	CD, Pso, PsA, UC	RA, SLE
19p13.2	TYK2	TYK2	Ruxolitinib	Small molecule	Atopic dermatitis, vitiligo, GVHD, myelofibrosis	RA, Pso, cytokine storm
			Tofacitinib	Small molecule	RA, PsA, UC, AS	COVID-19, SSc, SLE, CD
			Fostamatinib	Small molecule	Chronic immune thrombocytopenia	COVID-19, RA, SLE
			Deucravacitinib	Small molecule	Pso	SLE, CD, UC
19p13.2	ICAM1	ICAM1	Natalizumab	mAb	CD, MS	RA

CD: Crohn’s disease; GVHD: graft-vs-host disease; ILD: interstitial lung disease; IPF: idiopathic pulmonary fibrosis; MS: multiple sclerosis; Pso: psoriasis; UC: ulcerative colitis.

Discussion

We systematically explored the shared genetic architecture between susceptibility to and severity of COVID-19 and SSc by using genomic data from large European cohorts of COVID-19 and SSc. Our findings revealed a total of 19 non-HLA pleiotropic loci, including 2 novel associations for both conditions and 12 emerging as new pleiotropic loci. Moreover, functional annotation of these shared regions highlighted their potential regulatory role and allowed us to prioritize potential causal genes, many of which are implicated in fibrotic and inflammatory pathways.

Notably, despite the clinical presentation of lung involvement in SSc and COVID-19 patients may differ, we have identified potential mediators of the underlying fibrotic mechanisms common to both diseases. Specifically, we observed two pleiotropic loci annotated to BMP1 and PPARG genes as novel for COVID-19-related phenotypes and SSc. The BMP1 gene encodes a metalloprotease involved in the extracellular matrix maintenance through its crucial role in collagen biosynthesis. In addition, BMP1 is implicated in TGF-β activation, which is essential to balance the differentiation of endothelial cells to myofibroblasts in fibrotic processes in lungs [22]. In fact, a recent study showed that the inhibition of BMP1 resulted in a reduction of pulmonary fibrosis in a murine model [23]. This is consistent with our results indicating that the lead variant rs73225841 with a protective effect for both COVID-19 and SSc is associated with reduced expression levels of BMP1 in lung tissue. These results suggest that BMP1 might be a potential therapeutic target for fibrotic processes related with COVID-19 and SSc.

On the other hand, PPARG gene encodes for the peroxisome proliferator-activated receptor gamma, the expression/function of which has been inversely linked with fibrosis [24, 25]. Moreover, it has been stated that PPARG agonists effectively block TGF-β-induced fibrotic responses in murine models and attenuate fibrosis of skin and lung [26, 27]. It is noteworthy that the lead variant annotated by proximity to PPARG also acts as a pQTL for TIMP4, an inhibitor of matrix metalloproteinases also implicated in fibrotic processes with statistically significantly higher levels observed in the serum of SSc patients [28, 29]. Considering all this evidence, this variant might play a relevant role in the fibrotic processes associated with COVID-19 and SSc by regulating both PPARG and TIMP4 genes.

Our results indicate an enrichment of immune-related genes in the shared genetic architecture of both diseases, such as IL12RB1, IL12RB2 and NFKB1. The well-known genetic factors for SSc IL12RB1 and IL12RB2 encode the subunits of the IL-12 receptor, which promotes the JAK/STAT and IFN responses. Remarkably, these two loci represent novel susceptibility factors for COVID-19. These results are consistent with previous evidences reporting higher serum levels of the inflammatory IL-12 observed in COVID-19 patients due to the cytokine storm [30]. On the other hand, NFKB1 gene encodes a key component of the NF-κB family of transcription factors, considered master regulators of the pro-inflammatory response to infection and immune function. In this context, our functional annotation indicates that the lead variant rs57463114 annotated to NFKB1 gene affects NFKB1 expression levels in neutrophils and T lymphocytes. Indeed, it is well known that this gene is implicated in the pathogenesis of SSc and other IMIDs [31]. These results support our findings of NFKB1 as a novel risk factor for SARS-CoV-2 infection and COVID-19 hospitalization.

Interestingly, our results revealed several pleiotropic loci exhibiting opposing effects between COVID-19 and SSc. Among these, the TYK2 missense variant P1104A results in near-complete loss of TYK2 function leading to impairment of type I IFN, IL-12 and IL-23 signalling pathways [32]. This genetic variant is a well-known example of the antagonistic pleiotropy hypothesis, being a protective factor against multiple IMIDs while increasing susceptibility to infectious diseases [15, 33, 34]. The antagonist effect of this genetic variant is consistent with the divergent role of type 1 IFN in SSc and COVID-19. Recent studies have reported that both inborn errors of type I IFN immunity and autoantibodies neutralizing type I IFNs, mechanisms leading to low IFN levels, underlie severe, life-threatening COVID-19 pneumonia [35, 36]. In contrast, a strong type I IFN signature has been consistently implicated in SSc pathogenesis [37, 38]. Supporting these opposite roles, we identified two additional novel loci linked to type I IFN immunity that also show antagonistic effects. Specifically, the lead variant rs11644034 annotated to IRF8 gene showed a protective effect against SSc while increasing susceptibility to severe COVID-19. IRF8 encodes a member of the IFN regulatory factor family that plays essential roles in host defence and immune homeostasis [39]. Consistent with our results, the same genomic region has been described to be a protective factor for SSc and other IMIDs, such as SLE and idiopathic inflammatory myopathies [40]. Moreover, a candidate gene study identified suggestive associations between IRF8 genetic variants and increased risk for tuberculosis [41]. Remarkably, several studies have described that different mutations in IRF8 impair the type I IFN–mediated antiviral immune response causing an immunodeficiency syndrome [42–44]. Our results also showed that the rs79181492 SNP annotated to SENP7 gene comprises a susceptibility factor for SSc but shows a protective effect for SARS-CoV-2 infection. SENP7 encodes a SUMO-specific protease that enhances IFN signalling and is involved in innate defence against herpes simplex virus 1 infection in mice [45]. These two novel associated genes, in addition to the well-known TYK2 variant, collectively support a model of antagonistic pleiotropy of the IFN signalling between COVID-19-related phenotypes and SSc, further supporting the genetic evidence for this phenomenon in both diseases [34]. Nevertheless, further functional studies are needed to elucidate the exact mechanism by which IRF8 and SENP7 affect the autoimmunity–infection balance.

Genetic evidence serves as a highly valuable resource for identifying and prioritizing potential drug targets or providing additional support for ongoing clinical trials [46]. Based on the prioritized genes from our study, we identified several drugs that could be repurposed for the treatment of COVID-19 and/or SSc, many of which have not yet been tested for these conditions and emerge as promising candidates. Among these, we found drugs such as ustekinumab or fostamatinib, which are currently indicated for IMIDs other than SSc, exhibiting potential for treating the immune consequences of severe COVID-19 or SSc itself. It is worth noting that our analysis also highlighted deucravacitinib, a highly specific inhibitor of TYK2 currently indicated for plaque psoriasis [47] that presents a promising option for the management of SSc. Supporting the validity of this analysis, our findings also comprised drugs such as nintedanib and tofacitinib, which are currently under trial or approved for COVID-19 and/or SSc. Nintedanib is a kinase inhibitor that has been recently approved to slow the rate of decline in pulmonary function in patients with SSc-associated ILD [48], and it is currently in clinical trials for COVID-19 patients who develop ILD or pulmonary fibrosis (NCT04619680, NCT04856111 or NCT04541680). Similarly, while tofacitinib is approved for the treatment of RA or ulcerative colitis, this kinase inhibitor is being studied for the treatment of COVID-19 (NCT04469114) and skin-related manifestations in SSc [49, 50]. Altogether, these findings underscore the potential of genetically supported drug repurposing to expand treatment options and address unmet therapeutic needs in both COVID-19 and SSc.

We acknowledge several limitations to this study. First, the public nature of the COVID-19 data (summary statistics) limited our ability to render the three COVID-19 categories mutually exclusive, which could have provided more comprehensive findings. Additionally, data availability prevented us from evaluating the effects of specific demographic factors (e.g. sex and age) or clinical variables such as SSc-ILD status. The inclusion of this data could have yielded valuable insights as we recognize that it is the most severe form of the disease and challenging to clinicians, especially if there is rapid progressive pulmonary fibrosis, a feature shared with severe COVID-19. Furthermore, while our study reveals several novel genetic associations for both diseases, functional studies are needed to confirm their potential role in the pathogenic mechanisms shared by both diseases. Lastly, although our drug repurposing analysis identified promising therapeutic candidates for SSc and COVID-19, further experimental and clinical validation will be necessary to establish the efficacy and safety of each proposed treatment.

Our findings have shed light on the shared genetic architecture of COVID-19 and SSc. We have identified 19 pleiotropic loci, including BMP1 and PPARG genes, representing novel genetic associated factors for both conditions. Additionally, we observed that these pleiotropic loci converge on perturbing fibrotic and immune pathways. Of particular relevance, we detected a model of antagonistic pleiotropy of the IFN signalling between COVID-19-related phenotypes and SSc. Finally, our results provide new candidate therapeutic options that could potentially benefit both conditions.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

The summary statistics data for the three meta-analyses are publicly available through a Figshare repository, at https://figshare.com/articles/dataset/Summary_Stats_CrossTraits_SSc_COVID19/26317693.

Contribution statement

C.R.-B.: Data curation, Formal analysis, Visualization, Writing—original draft, Writing—review & editing. G.B.-Y.: Data curation, Formal analysis, Methodology, Visualization, Writing—review & editing. I.R.-M.: Data curation, Writing—review & editing. M.K.: Data curation, Writing—review & editing. M.A.-H.: Conceptualization, Methodology, Writing—review & editing. J.M.: Conceptualization, Funding acquisition, Supervision, Writing—review & editing. L.O.-F.: Conceptualization, Methodology, Supervision, Writing—original draft, Writing—review & editing. All authors critically revised and approved the manuscript.

Funding

This work was supported by Redes de Investigación Cooperativa Orientadas a Resultados en Salud (RICORS) [RD21/0002/0039]. C.R.-B. was supported by the program FPU: [FPU22/01652], funded by the Spanish Ministry of Science Innovation and Universities. I.R.-M. was supported by the program FPU: [FPU21/02746], funded by the Spanish Ministry of University. G.B.-Y.’s contract is part of the grant PREP2022-000712, funded by the MCIN/AEI/10.13039/501100011033 and the ESF+. M.A.-H. is a recipient of a Miguel Servet fellowship [CP21/00132] from Instituto de Salud Carlos III. L.O.-F. is recipient of a Ramon y Cajal fellowship [RYC2022-036635-I] funded by MICIU/AEI/10.13039/501100011033 and by ESF+.

Disclosure statement: The authors have declared no conflicts of interest.