Genetics of circulating inflammatory proteins identifies drivers of immune-mediated disease risk and therapeutic targets

Abstract

Circulating proteins have important functions in inflammation and a broad range of diseases. To identify genetic influences on inflammation-related proteins, we conducted a genome-wide protein quantitative trait locus (pQTL) study of 91 plasma proteins measured using the Olink Target platform in 14,824 participants. We identified 180 pQTLs (59 cis, 121 trans). Integration of pQTL data with eQTL and disease genome-wide association studies provided insight into pathogenesis, implicating lymphotoxin-α in multiple sclerosis. Using Mendelian randomization (MR) to assess causality in disease etiology, we identified both shared and distinct effects of specific proteins across immune-mediated diseases, including directionally discordant effects of CD40 on risk of rheumatoid arthritis versus multiple sclerosis and inflammatory bowel disease. MR implicated CXCL5 in the etiology of ulcerative colitis (UC) and we show elevated gut CXCL5 transcript expression in patients with UC. These results identify targets of existing drugs and provide a powerful resource to facilitate future drug target prioritization.

Fig. 1. Genomic map of genetic determinants of inflammation-related proteins.

Circos plot linking the location of pQTLs to the gene encoding their associated proteins. Labels for the cis-pQTLs (red) indicate the gene encoding the target protein. For the trans-pQTLs (blue), the gene symbols of the target proteins are indicated, along with the putative mediating gene(s) at the trans-pQTLs in brackets where applicable. The −log₁₀(P) values are capped at 150 for readability. Two-sided P values are from meta-analysis of linear regression estimates.

Fig. 2. Genetic architecture of 91 inflammation-related proteins.

a, Circos plot showing the trans-pQTL ‘hotspot’ at the SH2B3 locus on chromosome 12, associated with six proteins. b, Distribution of the number of identified pQTLs per protein. The HLA was treated as a single region. c,d, Manhattan plots showing genetic associations with plasma abundance of IL-12B (c) and TNFSF10 (TRAIL) (d). The horizontal red line indicates statistical significance (P = 5 × 10⁻¹⁰). Two-sided P values are from meta-analysis of linear regression estimates. The nearest genes in the region of pQTL signals are annotated.

Fig. 3. Genetic regulation of the inflammasome affects plasma IL-18 levels.

a, Schematic illustrating the cleavage of pro-IL-18 by caspase-1 and subsequent secretion of mature IL-18 from the cell into the extracellular space. b, Regional association plots around NLRC4 showing: the trans-pQTL signal for plasma IL-18 protein (top) from the present study (n = 14,824) and the cis-eQTL signal for NLRC4 (bottom) in whole blood from the eQTLGen study (n = 31,684). The purple diamond shows the sentinel pQTL variant. Other variants are colored by LD to the sentinel pQTL. Two-sided P values are from meta-analysis of linear regression estimates.

Fig. 4. The LTBR–LTA axis in the etiology of multiple sclerosis.

a,b, Unconditioned (a) and conditioned (b) regional association plots at the TNFRSF1A-LTBR locus (rs2364485 ± 100 kb) for multiple sclerosis (top), plasma LTA protein levels (middle) and LTBR mRNA expression in whole blood from eQTLGen (bottom). Multiple sclerosis associations were conditioned on rs1800693 (the strongest disease signal in the region). LTBR mRNA expression levels were conditioned on the following independent eQTLs: rs3759322, rs1800692, rs2228576, rs10849448, rs2364480 and rs12319859. The purple diamond shows the sentinel pQTL variant. Other variants are colored by LD to the sentinel pQTL. Two-sided P values are from meta-analysis of linear regression estimates.

Fig. 5. MR analysis of circulating proteins in immune-mediated disease etiology.

GSMR analysis using cis-pQTLs as genetic instruments to test the causal role of plasma proteins across IMDs. Cells are colored according to the effect size and direction: red indicates that higher genetically predicted plasma protein levels are associated with increased disease risk; blue indicates that higher genetically predicted plasma protein levels are associated with reduced disease risk; and gray represents no result because fewer than three variants were available for GSMR analysis. Associations with FDR ≤ 0.01 are denoted with dots, with filled circles indicating those that were robust to confounding by LD and open circles indicating those that were not.

Fig. 6. CXCL5 in UC pathogenesis.

a, Genetic associations in the CXCL5 gene region. From top to bottom: plasma CXCL5 pQTL (n = 14,824 participants), whole-blood eQTL (from eQTLGen data, n = 31,684 participants), colon tissue eQTL (GTEx, n = 368 individuals), UC (cases = 12,366, controls = 33,609) and CD (cases = 12,194, controls = 28,072) (from the IBD Genetics Consortium). The purple diamond shows the sentinel pQTL variant. Other variants are colored by LD to the sentinel pQTL. P values are from linear regression for QTLs and logistic regression for case-control GWAS. b, Violin plots showing CXCL5 expression in gut mucosal samples from patients with UC or CD and healthy controls (HC) in IBD TaMMA. c, Volcano plot showing differential expression analysis comparing colonic tissue from UC with HCs (IBD TaMMA). Red and blue points represent significantly (5% FDR) up- and downregulated transcripts, respectively. Gray indicates nonsignificant. P_BH, Benjamini–Hochberg adjusted P values. P values in c and d are from Wald tests (two sided). d, Replication. Left, CXCL5 differential expression in colon biopsies in UC versus HCs from transcriptome-wide analysis across three cohorts. The GSE numbers are GEO accession numbers. Imperial is the Imperial UC cohort. Each lollipop represents a separate cohort: GSE16879 (n = 24 UC patients versus n = 6 HCs); GSE206285 (n = 550 UC patients versus n = 18 HCs); and Imperial (n = 16 UC versus 6 HCs). The circle color indicates the log₂(FC) in CXCL5 expression between UC and HCs. Right, CXCL5 expression in colon biopsies sampled at baseline during the UNIFI clinical trial. Each point represents an individual. e, Forest plot showing MR analysis for UC and CD. OR is the odds ratio for the risk associated with a 1 s.d. increase in the level of the protein. The center of the bar is the point estimate for OR and the whiskers are the 95% CIs.

Extended Data Fig. 1. Overview of the pQTL analysis.

Schematic of the analysis pipeline.

Extended Data Fig. 2. Plasma protein abundance and pQTL detection.

a) Proteins with low abundance are more likely to have no detectable pQTL. Y-axis: percentage of samples above lower limit of detection for each protein, calculated using the INTERVAL data (n = 4,896) for which we had individual-level protein data available. Blue and red points indicate presence or absence of at least 1 significant pQTL in the GWAS meta-analysis, respectively. b) Manhattan plot for genetic associations with plasma IL17C, where the red horizontal line indicates the statistical significance threshold (5 × 10⁻¹⁰). P-values from linear regression.

Extended Data Fig. 3. pQTL replication in the ARISTOTLE cohort.

Comparison of effect sizes between pQTLs from the discovery pQTL meta-analysis (n = 14,824) and the ARISTOTLE cohort (n = 1,585). Each point represents a genetic variant that was a significant pQTL in the discovery meta-analysis. Effect size = standard deviation (sd) increase in protein per allele. 174 of 180 genetic variants were available for testing in the ARISTOTLE data. Red= cis, Blue= trans.

Extended Data Fig. 4. Genetic architecture of circulating inflammation-related proteins.

a) Relationship between minor allele frequency (MAF), pQTL effect size and proportion of variance explained (2MAF(1-MAF)Effect²), for 227 conditionally independent pQTLs (red=cis, blue=trans). b) Proportion of variance explained (PVE) by the conditionally independent variants associated with each protein. Proteins are annotated using the gene symbol of their encoding genes. Protein names are coloured in red if over 80% of samples have levels below the lower limit of detection in the INTERVAL dataset.

Extended Data Fig. 5. Chemokine trans-pQTL hotspot.

Forest plot showing the associations for the pleiotropic trans-pQTL at rs12075 (GRCh37, 1:158175353-160525679) with plasma levels of chemokines and blood cell counts. Center of bar = effect size estimate, whiskers = 95% confidence interval (cI). WBC = white blood cell count. P = p-value, b= beta (effect size). SE = standard error. Blood cell association data from Chen et al.. P-values from linear regression.

Extended Data Fig. 6. Colocalisation of pleiotropic chemokine trans-pQTL and blood cell count trait signals.

Regional association plots in the region around rs12075 (GRCh37, 1:158175353-160525679). a, Association with plasma chemokine levels. b, Associations with basophil, monocyte and white blood cell (WBC) counts using data from Chen et al.. P-values from linear regression.

Extended Data Fig. 7. Interactions between the candidate mediators for multi-locus-regulated proteins.

a) TNFSF10 (also known as TRAIL), b) KITLG (also known as stem cell factor), and c) IL12B. The graphs were generated using the STRINGdb (v11.5) webtool. The colouring of the edges indicates the type of evidence supporting an interaction, as shown in the legend above.

Extended Data Fig. 8. Protein-disease connections from overlap of pQTLs and disease GWASs.

The protein and the corresponding pQTL sentinel variant are indicated in the format of protein-rsid. The nearest gene to the pQTL sentinel variant is shown in brackets. Red lettering= cis-pQTL, blue lettering= trans-pQTL. Asterix indicates the genetic variant lies in the HLA region. Red squares: genetic susceptibility to increased plasma levels of the protein is associated with increased disease risk. Blue squares: decreased disease risk.

Extended Data Fig. 9. Protein and immune-mediated disease (IMD) connections from overlap of pQTLs and disease GWASs.

The protein and the corresponding pQTL sentinel variant are indicated in the format of protein-rsid. The nearest gene to the pQTL sentinel variant is shown in brackets. Red lettering= cis-pQTL, blue lettering= trans-pQTL. Asterix indicates the genetic variant lies in the HLA region. Red squares: genetic susceptibility to increased plasma levels of the protein is associated with increased disease risk. Blue squares: decreased disease risk.

Extended Data Fig. 10. Mendelian randomisation analysis for CXCL5 and ulcerative colitis.

a) Scatterplot showing the 13 variants used in the GSMR analysis assessing the effect of CXCL5 on ulcerative colitis (UC) risk from the GWAS by de Lange et al (ref. ) Each point represents a genetic variant, and indicates the effect size of the variant on CXCL5 levels versus UC risk (log odds ratio). Vertical and horizontal lines represent 95% confidence intervals. b) Directional concordance between CXCL5 pQTL and blood and colon tissue eQTLs. Forest plots showing effect size estimates for rs450373 pQTL in plasma (from our discovery meta-analysis) and eQTLs in whole blood and transverse colon tissue (GTEx v8 data). OR= odds ratio, calculated from beta estimate (representing the change in inverse-rank normalised plasma protein level in standard deviations associated with each copy of the effect allele). CI = confidence interval. P = p-value. Centre of bar = OR estimate, whiskers = 95% CI.