This study adhered to the guidelines stipulated by the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian randomization (STROBE-MR) (20). The STROBE-MR checklist for the reporting of MR studies was showed in Supplementary Table 1. We utilized the two-sample MR method to investigate the potential causal relationships between circulating micronutrients and the risk of SLE. Given that our study harnessed data extracted from pre-existing published literature, it circumvented the need for further ethical approval or informed consent. The architecture of our study was based on the three core assumptions underpinning MR (Figure 1).

A systematic search of PubMed was conducted to identify observational studies published on circulating micronutrients in relation to SLE. This resulted in an initial list of such micronutrients, comprising vitamin A, vitamin B6, vitamin B9, vitamin B12, vitamin C, vitamin D, vitamin E, sodium, phosphorus, calcium, magnesium, copper, iron, zinc, and selenium (4). Although several MR studies have evaluated the role of vitamin B9 (21), vitamin B12 (21), vitamin D (22), and iron status (16) in SLE, the Genome-Wide Association Study (GWAS) data for both exposures and outcomes used in our research exhibit slight variations. As a result, we have undertaken a replication of these analyses. Following this, we explored the GWAS catalog and PubMed for published GWAS centered on circulating micronutrients in European populations (the search was last updated in September 2023). Owing to the lack of relevant studies on sodium, it was excluded from our analysis. Ultimately, our research encompassed GWAS of 14 different circulating micronutrients: vitamin A (23), vitamin B6 (24), vitamin B9 (25), vitamin B12 (25), vitamin C (26), vitamin D (27), vitamin E (28), phosphorus (29), calcium (30), magnesium (31), copper (32), iron (33), zinc (32), and selenium (32). In this research, Single Nucleotide Polymorphisms (SNPs) linked to these 14 circulating micronutrients were designated as instrumental variables (IVs) adhering to the following standards: (1) The SNP demonstrates significant association with circulating micronutrient (P < 5e-08) and lacks linkage disequilibrium (r² < 0.001, KB = 10,000) (34); (2) The SNP with a minor allele frequency (MAF) of ≥ 5% (35); (3) The SNP showing no evidence of reverse causality, as determined by the Steiger filtering test (36); (4) In cases where the SNP is not found in the results dataset, a closely associated SNP (r² > 0.8) is chosen as a proxy in the 1000 Genomes database. If proxy SNP was unavailable, it was excluded from the analysis (37). (5) The chosen SNP is confirmed to be unassociated with confounding factors through inspection via the PhenoScanner database¹ (P < 5e-08,r² = 0.8) (38). Furthermore, we calculated the R² to denote the variance explained by the SNP and the F-statistic to signify potential weak IV bias in MR analysis. The R² was calculated as follows (39): R² = 2 × Beta² × (1-EAF) × EAF/SD², and the F-statistic was calculated as (40): F = (Beta)²/(SE)², where Beta is the per allele effect size of the association between each SNP and phenotype, EAF is the effect allele frequency, SE is the standard error, SD is the standard deviation. The IV is deemed strong when the F-statistic ≥ 10 (40). Ultimately, we identified 56 SNPs correlated with 14 circulating micronutrients, serving as IVs. The summary statistics of these SNPs utilized for MR analysis are presented in Table 1 and Supplementary Table 2.

The GWAS summary data (GCST90018917) for SLE were sourced from a recent large-scale GWAS in the IEU OpenGWAS database. This dataset comprises 647 cases of European ancestry (from Finland and the UK) and 482,264 control subjects of European ancestry. Then, cases of non-European ancestry (from Japan) have been excluded.

Following the harmonization of SNPs in both exposure and outcome using identical alleles, a two-sample MR analysis was conducted. When the MR estimate contained only one single SNP, the Wald ratio method was adopted as the primary analysis method (41); when the number of SNPs was ≥ 2, we employed the inverse variance weighted (IVW) method as the primary analysis method (42). When the number of SNPs was ≥ 3, the MR-Egger and Weighted median methods were applied for supplementary approaches to test the robustness of the primary analysis (43, 44). Furthermore, an MR analysis was conducted separately for each SNP associated with exposures. In addition, to ensure that the MR effects were oriented in the correct direction (from exposure to SLE), we conducted the MR Steiger test to confirm that each instrumental variable (IV) explained more variance in the exposure than in the outcome (36).

The degree of heterogeneity amongst the IVs was evaluated using Cochrane’s Q test (45). When P < 0.05, it signifies the presence of heterogeneity. In cases of observed heterogeneity, the random effects IVW method is deployed to ascertain the causal relationship between exposure and outcome, thereby mitigating bias from heterogeneous IVs. The MR-Egger intercept detected horizontal pleiotropy in the IVs, with P < 0.05 indicating its presence (44). The MR-Pleiotropy Residual Sum and Outlier method (MR-PRESSO) was employed to identify outlying SNPs and rerun the analysis after outlier removal (46). Finally, the leave-one-out analysis was implemented to ascertain the MR analysis’s robustness and determine whether a specific SNP drove any association (47).

In this study, P < 0.05 was considered statistically significant. All analyses were performed using R software, with the “TwoSampleMR” and “MR-PRESSO” packages facilitating the two-sample MR analysis.

We utilized the NHANES III (1988–1994) data as the external validation dataset for this study. The NHANES III participants were restricted to adults aged 17 years and older. After excluding 5 participants with unknown SLE status, a total of 20,045 participants with the completed household interview and physical examination were included in the analysis. The participant’s SLE status was determined by the item in the questionnaire: “Doctor ever told you had: lupus?” The other variables including age, gender, and race were also derived from the household interview data, while BMI was calculated using the formula: BMI = weight (kg)/[height (m)]². The serum levels of the 6 circulating micronutrients (vitamin A, vitamin C, vitamin D, serum calcium, iron, and selenium) were obtained from the laboratory examination data.

Considering the complex survey design, the weight variables were provided in the NHANES data files and t-tests, chi-square tests, and rank-sum tests were utilized to compare demographic disparities between the SLE group (N = 40) and the control group (N = 20,005). In the univariate regression analysis, we first constructed a preliminary rude model using only age, gender, race, and BMI. Then 6 circulating micronutrients were individually analyzed based on this rude model. Finally, variables with a significance level of P < 0.10 in above univariate regression analysis were included in the next multivariate regression analysis, which aimed to identify the most significant factors associated with SLE. And both the multivariate and univariate regression analyses were adjusted for the same covariates.

The MR estimates obtained from the IVW method revealed potential negative correlations between circulating calcium (OR: 0.06, 95% CI: 0.01–0.49, P = 0.009), iron levels (OR: 0.63, 95% CI: 0.43–0.92, P = 0.016) and the risk of SLE (Figure 2). Concurrently, the weighted median method also derived similar results regarding the causal relationship between circulating iron level and the risk of SLE (OR: 0.60, 95% CI: 0.39–0.92, P = 0.020). The directional consistency of the causal relationship between circulating calcium level and the risk of SLE was maintained in the weighted median analysis (OR: 0.08, 95% CI: 0.00–1.18, P = 0.066) and IVW analysis, albeit without statistical significance (Supplementary Table 3). However, we did not observe significant correlations between vitamin A, vitamin B9, vitamin B12, vitamin C, vitamin D, vitamin E, phosphorus, magnesium, copper, zinc, selenium and the risk of SLE, as detailed in Supplementary Table 3.

As indicated in Supplementary Table 4, both Cochrane’s Q test and the MR-Egger intercept suggest no heterogeneity and horizontal pleiotropy present in our MR analyses (P > 0.05). In the MR-PRESSO analyses, the rs1697421 (P = 0.01), rs17265703 (P = 0.005) and rs1801725 (P = 0.005) were identified as outliers. Then, no outlier SNPs were detected after removing and re-testing (P > 0.05) (Supplementary Table 4). When conducting the MR analyses using individual SNPs for either circulating calcium or iron levels, the results aligned with those obtained through the MR-PRESSO method (Supplementary Figure 1). The scatter plots, funnel plots, and leave-one-out plots all showed that the MR analysis results for the relationship between circulating calcium and iron levels with the risk of SLE remained robust, even under various pairs of sensitivity analyses (Supplementary Figures 2–4). Furthermore, as shown in Supplementary Table 5, the F-statistics for all 56 SNPs exceed 10, indicating no weak instrumental bias in our MR analyses.

To further validate the findings in our MR analysis, we compared and analyzed the levels of circulating micronutrients in the serum of patients with SLE and those without SLE which sourced from a large cohort, known as the NHANES III cohort.

A total of 20,045 participants were ultimately included in this study. The demographic characteristics of the NHANES III participants by SLE status are presented in Supplementary Table 6. After applying appropriate weighting for the analysis, we observed that the mean age of the SLE group [52.06 (13.72)] was significantly higher than that of the control group [57.80 (13.92)] (P < 0.01). However, there were no significant differences observed in other demographic characteristics, including BMI, gender, and race. In the comparative analysis of 6 circulating micronutrients between the two groups, the SLE group exhibited significantly lower levels of vitamin D (P < 0.01), serum total calcium (P = 0.01), and serum iron levels (P = 0.04) (Supplementary Table 7). Consistent with the results of univariate Logistic regression analyses, multivariate Logistic regression analyses also found that vitamin D (OR: 0.98, 95% CI: 0.97–1.00, P = 0.01) and serum total calcium (OR: 0.03, 95% CI: 0.00–0.58, P = 0.02) had a protective effect against SLE (Supplementary Table 8).