Identifying novel risk targets in inflammatory skin diseases by comprehensive proteome-wide Mendelian randomization study

Abstract

Background

Despite advances in cancer biomarkers and targeted therapies, early diagnosis and treatment of inflammatory skin diseases remain challenging. This study aims to identify circulating proteins causally linked to inflammatory skin diseases, including acne, atopic dermatitis, systemic lupus erythematosus, psoriasis, rosacea, and urticaria, through a Mendelian randomization (MR) framework.

Methods

A large-scale MR analysis was performed to assess the causal effects of thousands of plasma proteins on common inflammatory skin diseases. Additional methods, including Steiger filtering, transcriptome-wide association studies, summary data–based MR, protein–protein interaction networks, pathway enrichment analyses, Bayesian colocalization, and drug target evaluation, were employed to validate MR findings and explore therapeutic targets.

Results

This study identified >100 circulating proteins that may be involved in inflammatory skin diseases. Tier 1 therapeutic targets include RARRES2, SERPINC1, GALK1, and ECM1 for atopic dermatitis and RARRES2, PPID, and IL1RL1 for acne, rosacea, and urticaria. These proteins represent promising avenues for developing new treatments, with the potential to improve diagnostics and therapeutic strategies in the future.

Conclusion

This MR analysis revealed numerous plasma proteins associated with inflammatory skin diseases, offering insights into protein-mediated mechanisms and highlighting promising therapeutic targets for future interventions.

Key message

What is already known on this topic 

Inflammatory skin diseases, including psoriasis, atopic dermatitis, and acne, are complex conditions linked to systemic factors such as alterations in circulating plasma proteins. Previous studies have identified certain proteins involved in skin immune responses; however, a comprehensive understanding of their causal roles remains lacking.

What this study adds 

This study utilized a large-scale proteome-wide Mendelian randomization analysis to identify >100 circulating proteins causally linked to inflammatory skin diseases. Notably, proteins such as RARRES2, SERPINC1, and ECM1 were highlighted as potential therapeutic targets for atopic dermatitis and acne, among others.

How this study might affect research, practice, or policy 

The findings provide novel insights into protein-mediated mechanisms underlying inflammatory skin diseases, suggesting new diagnostic and therapeutic avenues. Future research should focus on validating these protein targets in clinical settings and exploring their potential for therapeutic intervention.

Introduction

The skin, as the largest organ of the human body, functions as a critical protective barrier against external injury and plays an essential role in the regulation of immune responses [1]. While skin disorders such as psoriasis, atopic dermatitis (AD), urticaria, acne, and others have distinct pathogeneses, they are often collectively studied as inflammatory skin diseases, characterized by the infiltration of inflammatory cells and a significant increase in inflammatory cytokines [2, 3]. These disorders are highly prevalent and clinically significant, impacting a large number of population worldwide. Acne and AD are common in adolescents and young adults [4, 5], and psoriasis and rosacea primarily impact adults [6]. Conditions including acne, psoriasis, AD, systemic lupus erythematosus (SLE), and urticaria also have notable global impacts, which impose a substantial social and economic burden, contributing to long-term healthcare costs and reduced quality of life [7]. Furthermore, these inflammatory skin diseases exhibit common pathological features, particularly immune-driven inflammation, such as dysregulated T-cell activation and increased proinflammatory cytokine production [Tumor Necrosis Factor (TNF)-α, IL-17, etc.], leading to similar inflammatory responses despite their diverse clinical presentations [8]. The complexity of their molecular mechanisms and the variability in treatment efficacy and side effects among patients present significant challenges in clinical management.

Inflammatory skin diseases have a complex multifactorial etiology, in which genetic and environmental factors interact both in the genesis and development of the disease. Increasing evidence suggests that skin inflammation is frequently linked to systemic disorders, including alterations in plasma proteins, which exert effects beyond the skin [9–11]. For instance, circulating plasma proteins like C-reactive protein (CRP) [12] and intercellular adhesion molecule 1 (ICAM-1) [13] have been found to correlate with the severity of psoriasis. Serum biomarkers for inflammatory skin diseases have become a focus of current clinical research; however, most studies have employed low-throughput techniques to identify serum protein biomarkers, and the molecular insights into the pathomechanisms of these diseases primarily stem from skin-based investigations. A comprehensive proteomic analysis of the skin, integrated with blood data, remains lacking.

Circulating plasma proteins play a pivotal role in a wide array of biological processes underlying complex diseases. Notably, they represent a substantial reservoir of therapeutic targets, with ~800 plasma proteins currently being targeted by the Food and Drug Administration (FDA)-approved drugs [14]. Recent advancements in genome-wide association studies (GWASs) have identified numerous quantitative trait loci (QTL) significantly associated with protein levels and gene expression, referred to as protein QTLs (pQTLs) and expression QTLs (eQTLs) [15, 16]. These findings have fueled the development of sophisticated statistical methodologies designed to integrate multidimensional datasets [17]. Concurrently, transcriptome-wide association studies (TWASs) have been employed to examine the relationship between gene expression and phenotypic traits [18], while Mendelian randomization (MR) and Bayesian colocalization analyses have been widely adopted to identify candidate genes by integrating QTL data with disease-related GWAS findings [19, 20]. MR, akin to a randomized controlled trial, exploits the random allocation of genetic variants at conception to evaluate causal relationships between exposures and outcomes, thereby minimizing confounding factors and mitigating reverse causality [21–23]. Proteome-wide MR is a method that evaluates causal relationships between protein levels and disease, using genetic variants as instrumental variables to reduce confounding factors. Bayesian colocalization analysis further assesses the likelihood that two traits share a common causal genetic variant. By integrating GWAS data with multidimensional QTL datasets, this approach enhances the prioritization of specific pathways and candidate genes, aiding in the identification of those potentially contributing to the pathogenesis of inflammatory skin diseases.

In this study, we employed a comprehensive strategy to identify circulating plasma proteins as novel therapeutic targets for six prevalent inflammatory skin diseases—acne, AD, SLE, psoriasis, rosacea, and urticaria—by combining high-throughput proteomic data with genetic information and genomic architecture linked to protein expression. These diseases were specifically selected for this study due to their high prevalence, significant clinical impact, and shared immune-mediated inflammatory mechanisms. Systemic inflammatory processes often underpin these skin diseases, making them ideal for a comprehensive proteomic study. The detailed workflow is illustrated in Fig. 1.

Materials and methods

Data sources and instrumental variable selection

Based on previous research, we applied consistent screening criteria (only European ancestry background; sample size >500; measured proteins >50) to select 11 proteomic GWASs from European ancestry cohorts (details of all cohorts are available in Table S1). pQTLs were filtered according to the following criteria: (1) single-nucleotide polymorphisms (SNPs) associated with any protein were included if they reached the genome-wide significance threshold (P ≤ 5 × 10⁻⁸) in at least one of the 11 GWASs; (2) variants with cis-acting effects, defined as those located within 1 Mb upstream or downstream of the gene encoding the plasma protein, were selected; (3) pQTLs located within the major histocompatibility complex region (chr6, 26–34 Mb) were excluded due to the strong linkage disequilibrium and pleiotropic characteristics of this region; (4) SNPs were clumped to retain only independent variants, with a linkage disequilibrium threshold of r² < 0.001; (5) SNPs associated with five or more proteins were excluded to avoid pleiotropic effects. All instrumental variables (IVs) were required to meet the three core IV assumptions: strong association with the exposure (relevance assumption), influence on the outcome only through the exposure (exclusion restriction assumption), and independence from confounders (independence assumption).

The outcome selection criteria were established as follows: (1) inclusion of participants exclusively of European ancestry to minimize population stratification bias and (2) exclusion of datasets with potential sample overlap between exposure and outcome variables. Based on these criteria, we ultimately utilized GWAS summary statistics from the FinnGen database (Release 10, accessible at https://finngen.gitbook.io/documentation/) as the source of outcome data. Due to the limitations of available data, our final analysis focused on six inflammatory skin disease phenotypes from the FinnGen database: acne (AC, 3245 cases and 394 105 controls), atopic dermatitis (AD, 15 208 cases and 367 046 controls), systemic lupus erythematosus (SLE, 705 cases and 385 509 controls), psoriasis (PSO, 10 312 cases and 397 564 controls), rosacea (RS, 2455 cases and 394 105 controls), and urticaria (UR, 11 187 cases and 398 204 controls). All diagnoses were based on ICD-10 (International Classification of Diseases) criteria.

Statistical analysis

Mendelian randomization analysis

MR analysis was conducted using R software (version 4.1.3) with the TwoSampleMR package (version 0.5.6) [21]. Effect estimates for proteins were calculated using the Wald ratio for single SNPs [24] and the inverse variance weighted (IVW) method was applied for multiple SNPs [25]. To account for multiple comparisons, a false discovery rate (FDR) correction was applied, with a significance threshold set at P <.05 [26]. The results are presented as odds ratios (OR) per standard deviation increase in genetically predicted plasma protein levels.

Sensitivity analysis

For proteins with more than two SNPs, four additional methods were employed as sensitivity analyses. The MR-Egger method assesses the presence of pleiotropy through the regression slope and intercept, which tests whether the genetic variants have a directional effect on the outcome independent of the exposure. If the intercept significantly differs from zero, it indicates potential pleiotropy. Adjusted estimates from MR-Egger are used when significant pleiotropy is detected, as they provide bias correction for pleiotropic effects [27, 28]. Mendelian Randomization Pleiotropy Residual Sum and Outlier (MR-PRESSO) detects outliers that may cause pleiotropy and heterogeneity, and corrects the causal estimate by excluding these outliers. This method ensures that the analysis is not biased by outliers that distort the causal inference [29] The weighted-median method provides a robust estimate of the causal effect that is valid even if up to 50% of the instrumental variables are invalid. It is particularly useful in situations where the instrument strength varies significantly or where some instruments may be invalid [30]. Additionally, MR-Robust Adjusted Profile Score (MRAPS) is designed to enhance statistical power and provide robust estimates in the presence of weak instruments or horizontal pleiotropy. It adjusts for the variability in the associations between genetic variants and the exposure, thereby ensuring the reliability of the MR estimates.

Generally, the IVW method was prioritized [31], but if the MR-Egger pleiotropy test indicated significant pleiotropy, the MR-Egger estimates were preferred. When MR-PRESSO’s global test revealed outliers, the analysis focused on the MR-PRESSO-corrected results. In cases with three or fewer SNPs, the weighted-median method was considered more appropriate. To control for multiple comparisons, Bonferroni correction was applied, with a threshold of P <.01 deemed significant for sensitivity analyses. If the results of the sensitivity analysis are inconsistent with the primary MR analysis, the causal association of the protein is likely a false positive due to pleiotropy and will be excluded from subsequent analyses. We estimated the strength of IVs and assessed the relevance assumption using the F-statistic [24]. All proteins with an F-statistic <10 were considered at high risk of weak instrumental bias and excluded from subsequent analyses.

In our study, reverse causality specifically refers to scenarios where disease states may influence protein levels, rather than proteins influencing the disease. This is particularly relevant for inflammatory diseases, where the disease process itself may elevate levels of inflammation-related proteins, mistakenly suggesting that these proteins are causal factors in the disease onset. To robustly address this issue, we assessed reverse causality between the proteins identified by the primary MR analysis through reverse MR analysis and the MR Steiger test [26]. The five aforementioned methods (IVW/Wald ratio, MR-Egger, MR-PRESSO, weighted-median, and MRAPS) along with the same Bonferroni correction were applied in the reverse MR analysis. Plasma proteins that appeared significant in either the MR Steiger test (Steiger P > .05) or reverse MR analysis, suggestive of reverse causality, were excluded from further analysis.

Phenotype scanning

To further ensure the assumption of independence, potential associations between all identified proteins and confounding factors were assessed via phenotype scanning using the Open Targets Genetics database (https://genetics.opentargets.org/) [32, 33], with a genome-wide significance threshold set at P < 5 × 10⁻⁸. pQTLs associated with established risk factors for inflammatory skin diseases, indicating potential pleiotropic effects, were interpreted with caution to avoid bias.

Bayesian colocalization analysis

To evaluate whether a specific genetic variant affects both the exposure and outcome by regulating gene expression at common loci, colocalization analysis was performed [34]. Bayesian methods were employed to calculate the posterior probability that a shared causal variant influenced both traits, using the R package “coloc” (version 5.0, available at https://github.com/chr1swallace/coloc). Given the large sample sizes of both exposure and outcome datasets, conservative prior probabilities were set at 1e-5 for any single SNP being associated with either trait (P1 and P2) and 1e-5 for a SNP being associated with both traits simultaneously (P12), minimizing the risk of false positives [35]. In the context of assuming a single causal variant, four potential hypotheses were evaluated: H0, indicating the absence of causal variants for both traits; H1, suggesting a causal variant specific to the first trait; H2, implying a causal variant only for the second trait; and H3, proposing independent causal variants for both traits. The final hypothesis, H4, postulated the presence of a shared causal variant influencing both traits. Colocalization was deemed significant when the posterior probability of H4 exceeded 0.8, indicating strong evidence of a common genetic influence on both traits [36].

Extra validation analysis

Given that gene expression and protein synthesis are regulated by complex biological mechanisms beyond simple genetic variation, additional validation analyses were conducted to support the primary MR findings. These analyses were performed at both the tissue and protein levels, utilizing TWASs and summary data–based Mendelian randomization (SMR) methods to ensure robustness of the results.

TWAS was conducted to predict the influence of protein-coding genes on phenotype risk at the tissue level using Functional Summary-Based Imputation (FUSION) software (available at http://gusevlab.org/projects/fusion). This approach leverages outcome GWAS summary statistics and pre-computed eQTL reference panels from the GTEx8 (Genotype-Tissue Expression version 8) database. TWAS data for proteins were first obtained from the skin tissue reference panel; if unavailable, data were retrieved from the whole blood reference panel. Proteins with no information in either panel were marked as NA and considered not passing the TWAS analysis.

The SMR software (v.1.3.1) [37] was used to perform SMR analysis and the corresponding heterogeneity in dependent instruments (HEIDI) test. Using the same outcome summary data and the cis-eQTL data provided by the eQTLGen consortium [38], we evaluated the causal association between the expression levels of genes corresponding to the identified protein targets and skin diseases [39]. Additionally, the HEIDI test was employed to assess whether the observed association was due to pleiotropy. A causal association was considered significant only if the SMR analysis P-value was <.05 and the HEIDI test P-value was >.05, indicating that the association was not due to pleiotropy. An uncorrected P-value <.05 was considered the threshold of significance in the validation analysis. Specifically, for SMR, a causal association was considered significant only if the SMR analysis P-value was <.05 and the HEIDI test P-value was >.05; otherwise, the association was considered spurious due to pleiotropy.

A protein–protein interaction (PPI) network was constructed using the STRING (Search Tool for the Retrieval of Interacting Genes) database (version 11.5) [40] with a minimum required interaction score (IAS) threshold of 0.4 to indicate interactions among identified proteins and pre-existing drug targets, which were sourced from the DrugBank database [41].

Evidence-based grading of potential drug targets

Proteins were categorized based on the criteria established in prior studies [42]. Tier 1 Targets: Proteins in this category are supported by robust evidence (PPH4 > 0.8), validated through both TWAS and SMR, and directly associated with known drug targets within the PPI network. Tier 2 Targets: This group includes proteins that meet at least one of the following criteria: (1) directly linked to known drug targets within the PPI network and validated by both TWAS and SMR; (2) supported by substantial evidence (PPH4 > 0.8) and validated by both TWAS and SMR; (3) associated with known drug targets within the PPI network and supported by substantial evidence (PPH4 > 0.8) but not confirmed through external validations. Tier 3 Targets: Proteins in this tier are characterized by (1) a PPH4 >0.8 without confirmation from both TWAS and SMR or (2) a sole association with known drug targets within the PPI network, without confirmation from both external validations. Tier 4 Targets: Proteins that do not meet the criteria for the first three tiers are classified here.

Results

In conclusion, this research employed MR analysis to scrutinize the causal relationships between 2794 plasma proteins and 6 inflammatory skin diseases. There is an overview table summarizing the major proteins associated with each disease, together with their clinical relevance and integrated phenotype scanning (Table S2).

Primary Mendelian randomization analysis and sensitivity analysis

A total of 2794 proteins with 11 173 IVs (Table S3) were identified and included in the primary MR analyses. Chromosomal locations of the identified proteins are depicted in Manhattan plots (Fig. 2), while all significant associations are visualized in a volcano plot (Fig. 3).

Primary results for acne

After FDR adjustment, nine proteins were strongly linked to AC (Fig. S1). Among these, RARRES2 (OR_IVW: 0.8077, 95% CI: 0.7536–0.8657) showed the most significant association. Sensitivity analyses were consistent with the primary MR results. The F-statistic indicated weak instrumental bias for EPHB1 (F: 4.74). Further reverse MR analysis confirmed a significant reverse causal association between AC and DARS2 (beta_{MR-Egger ± SE}: 2.250 ± 0.550). Consequently, these two proteins were excluded from further analysis. Detailed results are presented in Table S4.

Primary results for atopic dermatitis

In the primary analysis, 68 proteins were strongly linked to AD (Fig. S2), with significant associations remaining stable in sensitivity analyses (Table S5). Fifteen proteins, including CD27, CTSH, DPEP2, FOLR2, GNLY, GP1BA, HLA-DRA, IL18, IL22, LGALS3, MLN, MMP13, NAAA, NPL, and THBS2, were excluded due to weak instrumental bias (F-statistic: 0.01–9.18). Ultimately, 48 proteins were included in further analyses.

Primary results for systematic lupus erythematosus

Sixteen proteins were associated with SLE (Fig. S3), and these associations remained stable in sensitivity analyses (Table S6). However, AGRP, HLA-DRA, LRPAP1, POSTN, and PTGR1 were excluded due to potential weak instrumental bias (F-statistic: 0.01–9.10).

Primary results for psoriasis

MR analysis identified 31 proteins significantly associated with PSO (Fig. S4 and Table S7). Thirteen proteins were excluded from further analysis. Among these, CXCL1, GGH, GP6, IL18, LGALS3, LILRB1, NPL, NPPB, and PZP showed potential weak instrumental bias (F-statistic: 0.16–8.15).

Primary results for rosacea

Thirteen proteins were associated with RS (Fig. S5), and these associations remained consistent across sensitivity analyses (Table S8). GFRA2, NPPB, and SERPINA1 were excluded due to potential weak instrumental bias (F-statistic: 0.01–0.15).

Primary results for urticaria

Nineteen proteins were associated with UR (Fig. S6), and these associations were consistent in sensitivity analyses (Table S9). CD58, CFH, NAAA, and PTPRU were excluded due to potential weak instrumental bias (F-statistic: 0.00–6.59).

Reverse causality findings

In our study, we have carefully considered the possibility of reverse causation by implementing systematic reverse MR analyses and Steiger filtering to assess potential causal relationships between inflammatory skin diseases and protein levels. Our findings identified reverse causal associations for several proteins with different diseases: DARS2 with acne; CD2AP, CD84, HLA-DRA, IL18, IL22, INPP5D, MMP13, NAAA, RCN1, and SEPTIN8 with AD; BTN3A3, DDR1, and HLA-DRA with SLE; ATP6V1G2, IFNLR1, IL18, IL1RL1, NPPB, PRSS8, and PZP with PSO; DAPK2, F7, GFRA2, NPPB, and TNFSF15 with RS; and CD58, CFH, FAP, LILRA3, LTB, and NAAA with UR. These proteins were found to be influenced by the diseases themselves and were consequently excluded from further analysis to prevent incorrect causal inference.

Phenome-wide association study analysis

Results of phenotype scanning revealed key gene associations with potential confounders across multiple conditions (Table S10). For AC, genes such as C1QTNF5, ENPP7, and RARRES2 were linked to testosterone levels. In AD, genes like CREB3L4, CSF2RB, ECM1, GALK1, and GIP showed strong associations with asthma, eczema, and allergic diseases. For SLE, genes including AKR1A1, BTN3A1, CST3, CTSB, HAVCR1, IL1RL1, SFRP4, and TMEM106A were associated with lipid levels, white blood cell counts, hypertension, diabetes, and bone mineral density, suggesting interactions with SLE-related phenotypes. In PSO, genes such as ASGR1, GCKR, and CRP were associated with lipid levels, obesity, and diabetes, while ICAM3, IL10, and LY9 were linked to autoimmune diseases, including ulcerative colitis, Crohn’s disease, and rheumatoid arthritis, indicating these phenotypes as potential confounders. For RS, BTD was associated with body mass index and type 2 diabetes, CD55 with white blood cell count, PSMB4 with ulcerative colitis and body mass index, and S100A4 with systemic inflammation. For UR, C1QTNF5, IL1RL1, IL6R, TPSB2, SPON2, and LHB were associated with allergic reactions, autoimmune traits, and infections. Additionally, C1QTNF5 was linked to physical factors, and IL6R was connected to stress.

Colocalization analysis

Colocalization analysis was conducted to identify shared genetic signals between the selected proteins and inflammatory skin diseases, using a stringent threshold of PPH4 ≥0.8 (Table S11). Four proteins exhibited colocalization with AC: BOC (PPH4 = 84.5%), KRT5 (PPH4 = 86.4%), RARRES2 (PPH4 = 93.6%), and SLAMF6 (PPH4 = 90.8%). In the case of SLE and UR, three proteins each demonstrated colocalization: BTN3A1 (PPH4 = 92.7%), IL1RL1 (PPH4 = 99.4%), and VAT1 (PPH4 = 89.6%) for SLE; and IL1RL1 (PPH4 = 82.1%), SPON2 (PPH4 = 84.0%), and TPSB2 (PPH4 = 97.1%) for UR. Four proteins colocalized with RS: BTD (PPH4 = 83.3%), PPID (PPH4 = 90.7%), PSMB4 (PPH4 = 97.8%), and S100A4 (PPH4 = 90.6%). For AD, 17 proteins passed the colocalization analysis: CD33, CHRDL2, ECM1, GALK1, GIP, IL6R, IL6ST, ITPKA, LACRT, LTA, MANBA, MDM1, NAGK, PLXND1, RARRES2, SERPINC1, and VPS4A, with PPH4 values ranging from 86.2% to 100%. Seven proteins were identified as colocalized with PSO, including ABO, CD7, GCKR, ICAM3, IL6R, LEAP2, and RMDN1, with PPH4 values ranging from 80.8% to 100%.

External validation

Only proteins that were significant in both TWAS and SMR were considered to have passed external validation (P_twas < .05, P_smr < .05, P_heidi > .05). In AD, 10 proteins passed external validation: CCM2, CREB3L4, ECM1, GALK1, ITPKA, MAPKAPK2, NIT2, RARRES2, SERPINC1, and TNFRSF14. For PSO, three proteins were validated: ABO, ECI2, and LEAP2. AC, SLE, RS, and UR each had one protein that passed external validation: RARRES2 for AC, VAT1 for SLE, PPID for RS, and IL1RL1 for UR. Detailed results are presented in Table 1 and Figs. S7–S8.

A PPI network was constructed using DrugBank data to illustrate interactions between known drug targets and proteins of interest (Table S12). In the PPI network for AC, interactions were identified between BOC, FCGR3B, and established drug targets, with an IAS >0.4 (Fig. S9). For AD, all proteins, except AES, CD164L2, CHRDL2, CREB3L4, CRYZ, CST6, ITPKA, LACRT, LEAP2, LRP11, MANSC1, MDM1, NAGK, NPW, PLXND1, and VPS4A, exhibited interactions with known drug targets in the AD-specific PPI network (Fig. S10). In the SLE-specific and PSO-specific PPI networks, only two and eight proteins, respectively, remained isolated (Figs. S11 and S12). Similarly, interactions were identified for CD55, CD7, PPID, and S100A4 in the RS-specific PPI network (Fig. S13), and for CD33, ECM1, IL1RL1, IL6R, and LRRC15 in the UR-specific PPI network (Fig. S14). Further gene-disease enrichment analysis revealed that the identified proteins were enriched in categories such as skin diseases, autoimmune diseases, connective tissue diseases, and diseases of anatomical entities. However, the proteins associated with RS did not show enrichment in strongly related diseases (Tables S13). We conducted KEGG enrichment analyses on the identified genes within the PPI network as a supplement. The analysis revealed that key immune-related pathways, such as cytokine–cytokine receptor interaction, JAK–STAT signaling, and complement and coagulation cascades, were significantly enriched across these diseases. Notably, proteins like IL6R, MAPK1, and CREB3L4 were recurrently involved in these pathways, underscoring their central role in the inflammatory and immune processes underlying these conditions (Tables S14).

Potential drug targets

Finally, 7 proteins for AC, 48 for AD, 9 for SLE, 18 for PSO, 7 for RS, and 12 for UR were evaluated for their potential as drug targets. Among these, RARRES2 was identified as an excellent Tier 1 potential drug target for AC, while RARRES2, SERPINC1, GALK1, and ECM1 were identified for AD. PPID and IL1RL1 were the only Tier 1 proteins for RS and UR, respectively. No Tier 1 proteins were identified for SLE and PSO. Tier 2 proteins included BOC for AC; CCM2, CD33, GIP, IL6R, IL6ST, ITPKA, LTA, MANBA, MAPKAPK2, and TNFRSF14 for AD; BTN3A1, IL1RL1, and VAT1 for SLE; ABO, CD7, ICAM3, IL6R, and LEAP2 for PSO; and S100A4 for RS. No Tier 2 proteins were identified for UR. Other proteins were categorized as Tier 3 or below (Table 1).

Discussion

This study employed an integrative approach, combining Mendelian randomization, Steiger filtering, transcriptome-wide association studies, summary data–based Mendelian randomization, protein–protein interaction networks, pathway enrichment analyses, Bayesian colocalization, and drug target evaluation to assess the causal effects of thousands of plasma proteins from large-scale proteomics GWASs on inflammatory skin diseases. In our primary analysis, 156 unique proteins were identified using all available pQTLs. Subsequent analyses verified these findings and explored the regulatory mechanisms and potential drug targets, identifying 101 unique proteins causally associated with inflammatory skin diseases. Among them, 2 proteins for acne (AC), 14 for atopic dermatitis (AD), 3 for systemic lupus erythematosus (SLE), 5 for psoriasis (PSO), 2 for rosacea (RS), and 1 for urticaria (UR) were categorized as top-tier drug targets.

Several Tier 1 proteins prioritized by MR analysis in our study have been previously reported to be associated with inflammatory skin diseases. For instance, retinoic acid receptor responder protein 2 (RARRES2), encoding chemerin, a multifunctional adipokine, has been shown to regulate immune functions such as inflammation and the chemotaxis of dendritic cells by binding to its receptors (e.g. ChemR23, CMKLR1). Chemerin expression is primarily localized to fibroblasts, mast cells, and endothelial cells within the skin [43–45]. Chemerin may contribute to disease pathogenesis by modulating immune responses and skin barrier function. Although direct associations between chemerin levels and skin diseases remain limited, its role in recruiting immune cells to inflammatory sites highlights its potential involvement in chronic skin conditions [46–48]. Additionally, chemerin inhibits Staphylococcus aureus–induced IL-33 expression in keratinocytes, impairing neutrophil recruitment and bacterial clearance [49, 50]. However, given chemerin’s context-dependent bioactivity, its systemic effects in plasma circulation may differ from those in local tissues, such as skin or adipose tissue [43–45]. Circulating chemerin has also been shown to cause neutrophils to leave inflammatory sites without inducing further neutrophil infiltration [51], suggesting that circulating RARRES2 may exert beneficial effects in conditions like AD and acne by mitigating inflammatory cytokine secretion and modifying relevant signaling pathways, which warrants further investigation.

Few studies have explored the relationship between SERPINC1 (also known as antithrombin III) and AD. Lotti et al. reported that antithrombin III levels in the plasma of AD patients were comparable to those of healthy controls, suggesting no significant alteration in antithrombin III levels [52]. However, the fibrinolytic system, of which antithrombin III is a part, plays a role in AD’s inflammatory processes, particularly during the acute phase, where increased cutaneous fibrinolytic activity has been observed [52]. This suggests that while antithrombin III itself may not be altered, the broader fibrinolytic system is involved in AD pathogenesis. Extracellular matrix protein 1 (ECM1) maintains skin homeostasis by regulating matrix protein expression and tissue repair under normal conditions. In AD patients, altered ECM1 expression may exacerbate skin barrier dysfunction and immune responses, worsening disease symptoms [53, 54]. Another protein, the overexpression of PPID (peptidylprolyl isomerase D, also called cyclophilin D, encoded by Ppid), has been implicated to increase the collagen secretion and promote fibroblast proliferation, contributing to skin wound healing [55], though direct links to inflammatory skin diseases remain sparse. IL1RL1 (also known as ST2) is part of the IL-33/ST2 pathway, which plays a role in allergic and inflammatory diseases. IL-33 binding to IL1RL1 activates immune cells such as mast cells and eosinophils, both crucial in the pathogenesis of urticaria [56, 57].

In this research, our findings build upon existing literature on the role of plasma proteins in inflammatory skin diseases while offering new insights into their clinical relevance and potential as biomarkers. Previous research has demonstrated the involvement of proteins such as RARRES2, ECM1, and SERPINC1 in immune regulation, skin barrier integrity, and inflammation. While these associations have been described, our findings extend this understanding by identifying novel causal relationships between Tier 1 proteins and disease progression. For instance, RARRES2, already recognized for its immune-modulating effects, is shown here to have broad applicability across multiple inflammatory skin diseases [50], including acne and AD, underscoring its potential as a shared therapeutic target. Similarly, within the epidermis, ECM1 traditionally has a role in the control of keratinocyte differentiation [58], but now is highlighted for its role in maintaining skin barrier function, providing new avenues for therapeutic intervention for barrier dysfunction in severe AD cases,

The clinical implications of these findings are significant. Tier 1 proteins like RARRES2 and SERPINC1 hold promise not only as diagnostic markers for stratifying patients by disease severity but also as tools to guide personalized treatment plans. For example, RARRES2 could support the early diagnosis of acne and AD by serving as a marker of inflammatory activity, while therapies targeting SERPINC1 might reduce systemic inflammation in patients with severe AD, and GALK1, linked to metabolic regulation, highlighting the importance of targeting both barrier dysfunction and metabolic pathways to optimize disease management. Similarly, PPID, implicated in oxidative stress regulation in rosacea, and IL1RL1, associated with mast cell activation in urticaria, emerge as promising candidates for biomarker development to monitor disease progression and predict therapeutic response. These findings align with prior studies while addressing critical gaps in biomarker discovery. By focusing on causal associations and identifying proteins with strong disease relevance, this study provides a framework for translating molecular research into clinical applications. The identification of Tier 1 proteins as biomarkers and therapeutic targets bridges the gap between research and practice, paving the way for more effective diagnostic tools and personalized therapies in inflammatory skin diseases.

For Tier 2 proteins, our findings are almost consistent with previous studies. For instance, IL6R, existing in both transmembrane and soluble forms, binds IL6 to initiate downstream signaling pathways vital for Th17 differentiation [59], a finding supported by our pathway enrichment analysis. A large multi-population GWAS identified IL6R as a novel risk locus for AD [60], with functional IL6R variants also implicated as risk factors for asthma and AD [61, 62]. Our study extends these findings by expanding the population sample and reinforcing the causal association between IL6R and AD. Other proteins identified in our analysis were classified as Tier 3 or 4 drug targets, suggesting their potential as therapeutic targets, though additional experimental validation is needed to confirm their clinical relevance.

Several limitations should be acknowledged in this study. First, our analysis focused on circulating proteins, which include both secreted and leaked proteins. The abundance of these proteins may differ between circulation and specific tissues, such as skin, limiting our ability to assess tissue-specific protein effects. Second, cis-pQTL coding variants that alter protein sequences may affect quantitative protein assays without influencing protein function, potentially leading to erroneous conclusions. Third, we were constrained by the lack of individual-level data from the GWAS, preventing the exclusion of non-European participants. However, given the predominance of European participants in the dataset, it is unlikely that this significantly impacted our findings. Meanwhile, our study primarily focused on populations of European ancestry, which may restrict the generalizability of the findings to other ethnic groups. Given the known genetic and environmental variability in inflammatory skin diseases across different populations, future research should include more diverse cohorts to ensure broader applicability of results and to uncover potential population-specific protein–disease associations. Lastly, while we relied on publicly available datasets, which may limit the novelty of the data, meticulous data analysis offered valuable new insights from existing resources. Besides, the study relied on aggregate data rather than individual-level data, which may limit the precision of associations and preclude detailed subgroup analyses. Future studies incorporating individual-level data could enable a more nuanced exploration of protein–disease relationships and their interactions with clinical variables such as age, sex, and comorbidities. Addressing these limitations in future studies will enhance the robustness and translational potential of findings, ultimately contributing to more equitable and effective personalized treatment strategies.

Conclusion

Our MR analysis identified several circulating proteins causally linked to inflammatory skin diseases, providing key insights into their clinical relevance. Notable Tier 1 proteins, including RARRES2, SERPINC1, GALK1, and ECM1 for AD, and RARRES2, PPID, and IL1RL1 for acne, RS, and UR, respectively, represent promising therapeutic targets due to their central roles in inflammatory skin diseases. These proteins also hold potential as diagnostic biomarkers, enabling earlier detection and stratification of inflammatory skin diseases, while guiding personalized treatment strategies such as targeting RARRES2 to modulate inflammation or leveraging ECM1 to restore skin barrier integrity.

Future studies should validate these targets in diverse populations through clinical trials to ensure broader applicability. While Tier 2 and lower-tier proteins suggest additional therapeutic opportunities, further experimental research is needed to clarify their roles in skin inflammation, potentially paving the way for personalized medicine in dermatology.

Author contributions

W.J. and C.Z. were responsible for conceptualization, data curation, validation, and writing—review and editing. L.Y. was responsible for conceptualization, methodology, formal analysis, data curation, validation, visualization, supervision, writing—original draft, and writing—review and editing. The authors read and approved the final manuscript.

Conflict of interest statement: None declared.

Funding

This work was supported the National Key R&D Program of China [No. 2019YFA0112100]. The study funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Ethics approval and consent to participate

No ethical approval was required for the present work as it relied solely on previously published research.

Data availability

The data sets used for the present study can be accessed in the full-text articles included in the systematic review and meta-analysis. The data underlying this article will be shared on reasonable request to the corresponding author.

References

Author notes