We conducted a two-sample MR analysis to determine the causal effect of circulating metabolites on chalazion. Genome-wide association study (GWAS) summary level data for circulating metabolites were obtained from two different datasets, and chalazion summary data were obtained from the FinnGen database. The robustness of these causal relationships was assessed by sensitivity and colocalization analyses. Only individuals of European origin were included in the analysis. All included studies received informed consent and local ethics committee approval. An overview of the study design is shown in Figure 1.

Summary-level GWAS datasets of 249 metabolic biomarkers for the primary analysis were derived from the UK Biobank and measured using the Nightingale Health Metabolic Biomarkers Phase 1 release study. This study included 115,078 randomly selected participants from the UK Biobank cohort, and analysed a total of 249 metabolites, with 168 metabolites measured in absolute concentrations (mmol/L) and 81 measurements presented in ratios (https://biobank.ndph.ox.ac.uk/ukb/label.cgi?id=220) (Mi et al., 2022). Summary-level datasets of 123 circulating metabolites used for secondary analyses were obtained from a previous publication by Kettunen et al. (2016), including 24,925 individuals of European ancestry. Summary-level statistics for chalazion were obtained from the FinnGen database (Release 9), which included 348,073 individuals of European ancestry (case: 3,389; control: 344,684). Confounders included eyelid inflammation, conjunctivitis, anxiety/panic attacks, alcohol consumption, BMI, smoking, and IBS, which were obtained from the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/datasets/) (Supplementary Table S1).

SNPs associated with circulating metabolites were selected according to a significance threshold (p < 5 × 10⁻⁸) and a reference panel (with a window size of 10,000 kb and r ² < 0.001) to estimate the linkage disequilibrium between SNPs and select independent genetic variants (Levin et al., 2020). The F-statistic was calculated to test for weak instruments (F-statistics < 10) as previously described (Burgess et al., 2011; Burgess and Thompson, 2011). Instrumental variables (IVs) for MR studies need to fulfil three hypotheses: 1) IVs should be strongly correlated with exposure; 2) IVs should not be associated with any confounders; and 3) IVs affect outcomes only through exposure and not through any other alternative pathways (Palmer et al., 2012).

The presence of shared causal variants between exposure and outcome was analysed using GWAS-GWAS colocalization, with a reference panel (GWAS gene window default of 50 kb, p < 5 × 10⁻⁸) (Venkateswaran et al., 2018). Colocalization analyses were performed using the “coloc” package, and a Bayesian framework was used to generate posterior probabilities for five mutually exclusive hypotheses about causal variance sharing between two traits: (H₀–H₄): H₀: the locus is not associated with circulating metabolites and chalazion; H₁: the locus is associated with circulating metabolites, but not with chalazion; H₂: the locus is associated with chalazion, but not with circulating metabolites; H₃: the locus is associated with circulating metabolites and chalazion (two independent SNPs) and H₄: the locus is associated with circulating metabolites and chalazion (one shared SNP) (Giambartolomei et al., 2014; Venkateswaran et al., 2018). A posterior probability (PP) of H₄ (PPH₄) > 0.8 was identified as evidence in favour of colocalization. The “locuscomparer” package was used for visualization of colocalization (Liu et al., 2019).

Causal associations between each metabolite and chalazion were determined using IVW methods. The IVW estimate is the mean of ratio estimates from two or more instruments, based on the assumption that all SNPs are valid instruments or that the overall bias is zero (Burgess et al., 2013). MR Egger regression, weighted median, and weighted mode methods were used to relax the IVW assumptions (Burgess et al., 2017a; Burgess et al., 2019). For single genetic instruments, causal effects were estimated by the wald ratio test (Burgess et al., 2017b). Phenoscanner (http://www.phenoscanner.medschl.cam.ac.uk/) was used to exclude the SNPs associated with confounders (Staley et al., 2016). Sensitivity analyses were performed to assess the validity and robustness of the MR assumptions, including Cochran’s Q to test for heterogeneity among IVs of metabolites (Burgess et al., 2013), MR-Egger regression for horizontal pleiotropy assessment (Bowden et al., 2015), and leave-one-out test. If heterogeneity was present, the multiplicative random effects model was chosen (Burgess et al., 2019). MR-Steiger was used to filter out SNPs suggesting opposite causal directions (Hemani et al., 2017).

We chose the Bonferroni method for multiple testing; associations with p-values below 0.0002 (0.05/249) in the primary analysis and 0.0004 (0.05/121) in the secondary analysis were considered strong evidence of an association, and associations with p-values between 0.0002 and 0.05 in the primary analysis, and between 0.0004 and 0.05 in the secondary analysis were considered suggestive. All analyses were performed on the R platform (version 4.1.0) using the “TwoSampleMR,” “Coloc,” and “forestploter” packages.

We first estimated the causal relationship of confounders, including inflammation of the eyelid, conjunctivitis, anxiety/panic attacks, alcohol consumption, BMI, current tobacco smoking, and IBS, with chalazion risk, using a two-sample MR method. The results showed that BMI [OR = 0.970; 95% CI, 0.826–1.138; P_(IVW) = 0.709], Current tobacco smoking [OR = 0.724; 95% CI, 0.214–2.454; P_(IVW) = 0.604], Alcohol consumption [OR = 1.334; 95% CI, 0.099–17.99; P_(IVW) = 0.828], IBD [OR = 2.49E-06; 95% CI, 9.06E-24-6.81704E+11; P_{(Wald ratio)} = 0.529], and conjunctivitis [OR = 0.945; 95% CI, 0.495–1.805; P_{(Wald ratio)} = 0.863] were not association with chalazion risk, and the remaining confounders failed to establish associations with chalazion due to the absence of SNPs (Supplementary Table S2).

In the primary analysis, we screened 249 circulating metabolites with a total of 13,275 SNPs when the screening threshold was (r ² < 0.001, kb = 10,000, p < 5 × 10⁻⁸). In the secondary analysis, we screened 123 circulating metabolites with a total of 1,375 SNPs when the screening threshold was (r ² < 0.001, kb = 10,000, p < 5 × 10⁻⁸), among them, two metabolites glycerol (met-c-861) and urea (met-c-939), had no SNP extracted. All IVs in the current analysis were robust instruments, with the range of F-statistics was 23-16414 (Supplementary Tables S3, S4). We used a two-sample MR to estimate the causal effect of 249 metabolites on chalazion in the primary analysis and 121 circulating metabolic biomarkers risk in the secondary analysis (Supplementary Tables S5, S6). Based on the IVW method, seven of the 249 metabolites were found to be suggestively associated with chalazion, with five unsaturation-associated biomarkers showing an association with a high risk of chalazion and two unsaturation-associated biomarkers showing a causal effect with a low risk of a chalazion (Figure 2A). There were as follows: alanine (OR = 0.680; 95% CI, 0.507–0.912; p = 0.010), acetone (OR = 1.726; 95% CI, 1.064–2.798; p = 0.027), degree of unsaturation (OR = 1.216; 95% CI, 1.055–1.401; p = 0.007), ratio of omega-3 fatty acids to total fatty acids (OR = 1.204; 95% CI, 1.054–1.377; p = 0.006), ratio of omega-6 to omega-3 fatty acids (OR = 0.850; 95% CI, 0.735–0.982; p = 0.027), ratio of polyunsaturated fatty acids to total fatty acids (OR = 1.214; 95% CI, 1.008–1.462; p = 0.041), and ratio of docosahexaenoic acid to total fatty acids (OR = 1.267; 95% CI, 1.092–1.471; p = 0.002) (Figure 2B; Table 1). In the secondary analysis, four of the 121 metabolites were found to be significantly associated with chalazion and three were suggestively associated with chalazion. Of these, three unsaturation-associated biomarkers significantly increased the risk of chalazion, and four unsaturation-associated biomarkers showed a negative correlation with chalazion susceptibility (Figure 3A). They were alanine (OR = 0.578; 95% CI, 0.439–0.759; p = 0.00008), the average number of methylene groups per double bond (OR = 0.768; 95% CI, 0.615–0.958; p = 0.020), the average number of double bonds in a fatty acid chain (OR = 1.380; 95% CI, 1.135–1.678; p = 0.001), ratio of bisallylic groups to double bonds (1.298 [95% CI, 1.145–1.471], p = 0.00005), ratio of bisallylic groups to total fatty acids (OR = 1.364; 95% CI, 1.163–1.600; p = 0.00013), 18:2, linoleic acid (LA) (OR = 0.827; 95% CI, 0.720–0.950; p = 0.007), and pyruvate (OR = 0.422; 95% CI, 0.265–0.674; p = 0.00029) (Figure 3B; Table 2). MR results of individual SNPs for all metabolites were shown in Supplementary Tables S7, S8.

In the sensitivity analysis of all the differential metabolites, heterogeneity (IVW method) was present only for unsaturation (p = 0.045), ratio of omega-6 to omega-3 fatty acids (p = 0.008), ratio of omega-3 fatty acids to total fatty acids (p = 0.018) in 249 metabolites (Supplementary Table S9), and average number of methylene groups per double bond (p = 0.016) in 123 metabolites (Supplementary Table S10), and we adopted a MR analysis with multiplicative random effects model. The horizontal pleiotropy analysis showed that there was no direct association between IVs and chalazion, and MR Steiger test results showed overall SNPs affected outcomes only by exposure and not by other pathways (Supplementary Tables S11, S12), suggesting that the results were relatively robust. Plots of the leave-one-out test showed that regardless of which SNP was excluded, it did not have any effect on the MR results for alanine, ratio of bisallylic groups to double bonds, and ratio of bisallylic groups to total fatty acids (All lines were on one side of 0). However, for other unsaturated fatty acids, there is a potentially influential SNP driving the causal link between these metabolites and chalazion. Therefore, we need to interpret this result with caution (Supplementary Figure S1 ).

Two colocalization analyses showed that there was no shared genetic variation region between alanine and chalazion (1.95% for PPH4 in the primary analysis and 25.3% in the secondary analysis), suggesting that the genetic variant SNPs were reducing chalazion risk through their impact on alanine. Shared regions of genetic variation were not observed for the differential metabolites in the primary analysis. In the secondary analysis, we identified two metabolites: the ratio of bisallylic groups to double bonds (PPH₄ = 0.817) and 18:2, linoleic acid (LA) (PPH₄ = 0.870), which shared one SNP with chalazion (rs174555), corresponding to the nearest genes FADS1 and FADS2 (Table 3; Supplementary Figure S2).