Enhancing antioxidant capacity and reducing asthma risk: the impact of zinc and iron on molecular pathways

This study employs a two-sample MR approach to infer causal relationships with trace elements (copper, calcium, iron, magnesium, zinc, potassium, and selenium) and vitamins (carotene, folate, and vitamins A, B6, B12, C, D, and E) as exposure factors and asthma as the outcome variable. Single-nucleotide polymorphisms (SNPs), the most common genetic variations in the human genome, are widely used to investigate associations between genes and traits or diseases. This study employed a two-sample MR approach to investigate the causal relationship between micronutrients, vitamins, and asthma, with all analyses strictly adhering to the three core MR assumptions [25].

Data were obtained from the OpenGWAS database (https://gwas.mrcieu.ac.uk/). Exposure GWASs were obtained from the MRC-IEU UK Biobank pipeline for calcium (ukb-b-8951; European; N = 64,979), carotene (ukb-b-16202; European; N = 64,979), folate (ukb-b-11349; European; N = 64,979), iron (ukb-b-20447; European; N = 64,979), magnesium (ukb-b-7372; European; N = 64,979), potassium (ukb-b-17881; European; N = 64,979), vitamin A (ukb-b-9596; European; N = 460,351), vitamin B12 (ukb-b-19524; European; N = 64,979), vitamin B6 (ukb-b-7864; European; N = 64,979), vitamin C (ukb-b-19390; European; N = 64,979), vitamin D (ukb-d-30890_raw; European; sample size as per the OpenGWAS entry), and vitamin E (ukb-b-6888; European; N = 64,979). Trace-element GWASs curated by IEU included copper (ieu-a-1073; European; N = 2,603), selenium (ieu-a-1077; European; N = 2,603), and zinc (ieu-a-1079; European; N = 2,603). Measurement units, biomarker matrix (serum/plasma/RBC), and any trait transformations follow the original GWAS definitions and are summarized in Table 1; all Mendelian-randomization effect estimates are interpreted per 1 standard deviation (SD) increase in the exposure. Outcome data were obtained from FinnGen release R10 (https://www.finngen.fi/; endpoint J10_ASTHMA_MAIN_EXMORE), including individuals of European ancestry with 37,760 asthma cases and 219,734 controls. Case definition was based on ICD-10 codes J45-J46 (compatible with ICD-9 code 493 and ICD-8 code 493), with a median age at first diagnosis of approximately 45 years (Table 2). Since exposure GWAS were partly derived from the UK Biobank, and FinnGen does not include UKB participants, no sample overlap exists, thereby minimizing potential bias.

This study was conducted based on three core assumptions of MR: (1) Relevance assumption, meaning that the selected SNPs must be significantly associated with the exposure (e.g., zinc and iron levels); (2) Independence assumption, which requires that the SNPs are independent of confounding factors between the exposure and the outcome; (3) Exclusion-restriction assumption, indicating that the SNPs affect the outcome only through the exposure, without influencing the outcome through other pathways (i.e., no horizontal pleiotropy).

Following the STROBE-MR guidelines [26], genome-wide significant SNPs (p < 5 × 10^{− 6}) were LD-clumped at R² = 0.001 within a 10,000 kb window (TwoSampleMR). Post-clumping instrument counts (zinc k = 8; iron k = 11), together with IVW and MR-Egger Cochran’s Q/df/P, are reported in Table S1. Per-SNP instrument strength was computed as F = (β / SE)²; because the selection threshold p < 5 × 10^{− 6} corresponds to a two-sided Z ≈ 4.565, all retained instruments satisfy a rigorous lower bound F ≥ Z² ≈ 20.84 (therefore the per-exposure minimum, mean, and median F are each ≥ 20.84. Horizontal pleiotropy was assessed using the MR-Egger intercep). Horizontal pleiotropy was assessed using the MR-Egger intercept (intercept ≈ 0 with p > 0.05 indicating no directional pleiotropy) and the MR-PRESSO global test with outlier removal. Leave-one-out analyses were conducted to evaluate the influence of individual SNPs. All MR effect estimates are interpreted per 1 SD increase in exposure. The stepwise process of SNP instrument selection is summarized in Figure S1.

Power calculations for MR were conducted using the mRnd web tool (http://cnsgenomics.com/shiny/mRnd/). Given the observed effect sizes (zinc OR = 0.947, iron OR = 0.782) and the proportion of variance explained by the instruments (R² ≈ 2.1% for zinc and 3.4% for iron), the statistical power exceeded 80% to detect moderate effect sizes (OR ≤ 0.85) at α = 0.05 with the current sample size (37,760 cases and 219,734 controls).

SNP data related to the outcome variable were obtained using R, and effect sizes were harmonized before two-sample MR. Effect sizes were harmonized before two-sample MR using the function harmonise_data (action = 2), aligning summary statistics to the forward strand. For palindromic SNPs (A/T or C/G), effect allele frequencies (EAFs) were compared between exposure and outcome datasets: variants with intermediate EAFs (0.42–0.58) were considered strand-ambiguous and excluded; otherwise, EAF was used to orient alleles and retain the SNP. Where necessary, effect estimates (β) were sign-flipped to ensure alignment of effect directions across exposure and outcome. SNPs with unresolved allele mismatches, multi-allelic loci, or ambiguous indels were removed. Counts of SNPs removed and retained at each step are summarized in Table S2.

The causal relationship between exposure factors and the outcome variable was assessed using odds ratios (OR), representing the change in asthma risk per 1 SD increase in exposure. The IVW method (random-effects) was used as the primary estimator, complemented by the weighted median, MR-Egger, weighted mode, and simple mode. Details of instrument selection, post-clumping k values, heterogeneity statistics, and the instrument-strength lower bound (F ≥ 20.84) are provided in Table S1.

For the multivariable Mendelian randomization (MVMR) analysis of zinc and iron, the inverse-variance weighted (IVW-MVMR) estimator was applied as the primary method. The instrument set was constructed as the union of genome-wide significant SNPs for zinc and iron, followed by LD clumping (r² = 0.001, kb = 10,000). For each exposure, Sanderson-Windmeijer conditional F statistics were calculated to assess instrument strength, and the covariance matrix of the instruments was evaluated to examine potential multicollinearity. Given the modest discovery sample for zinc (N = 2,603), its conditional F statistic was below 10, indicating limited instrument strength. Accordingly, robust MVMR approaches (including robust IVW-MVMR and the Q-statistic) were additionally implemented to ensure the stability of the inference.

This study conducted a sensitivity analysis to ensure the reliability of causal effect estimates. Heterogeneity testing in two-sample MR analysis was performed using Cochran’s Q under IVW and MR-Egger models; per-exposure Q/df/P values are tabulated in Table S1. Leave-one-out analyses assessed the influence of individual SNPs on the pooled effect. Ideally, the causal effect estimates should remain stable when any single SNP is removed; substantial shifts suggest heterogeneity or potential pleiotropy. Horizontal pleiotropy was evaluated using the MR-Egger intercept (intercept ≈ 0 and p > 0.05 indicates no directional pleiotropy) and MR-PRESSO (global test and outlier removal). These sensitivity results, together with the instrument-strength lower bound (F ≥ 20.84) documented in Table S1, support the robustness of the primary findings. The Steiger directionality test was also performed to assess whether the direction of causality was from exposure (zinc and iron) to asthma rather than the reverse. Corrected estimates after outlier removal in MR-PRESSO were reported alongside global test p-values. Leave-one-out analyses were performed using the same set of SNP instruments as the primary IVW analysis, ensuring consistency between the sensitivity analysis and the main model.

Given the limited availability of asthma airway transcriptomic resources in public databases, this study predefined GSE160727 and GSE208777 as exploratory transcriptomic datasets at the study design stage, intended solely for hypothesis generation rather than for causal or mechanistic integration in the main analyses. These datasets were derived from CD34 + bone marrow cells of MDS patients (iron overload control vs. DFX treatment, GSE160727) and retinal ganglion cells (control vs. TPEN treatment, GSE208777), which are not directly comparable to asthma airway tissues. Accordingly, only standardized differential expression screening and pathway enrichment analyses were performed, without direct integration with MR results. Specifically, GSE160727 compared the iron overload control group (n = 14) with the DFX-treated group (n = 23), using a threshold of p < 0.05 to identify differentially expressed genes, followed by pathway enrichment analysis with Metascape. GSE208777 compared the control group (n = 3) with the TPEN-treated group (n = 3), applying the same threshold (p < 0.05) to obtain 2,098 differentially expressed genes, which were subsequently analyzed using Metascape.

For the dataset GSE160727, differential gene expression analysis was performed on the iron overload control group (n = 14) and the iron overload group treated with deferasirox (DFX) (n = 23). A total of 570 differentially expressed genes (DEGs) (p < 0.05) were identified and subjected to pathway enrichment analysis using the Metascape tool. Similarly, for GSE208777, differential gene expression analysis was conducted between the control group (n = 3) and the TPEN (N, N, N’, N’-tetrakis(2-pyridylmethyl)ethylenediamine)-treated group (n = 3). A total of 2,098 DEGs (p < 0.05) were identified and analyzed for pathway enrichment using Metascape.

Asthma-related genes were retrieved from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.genome.jp/entry/H00079), while key genes associated with the NRF2/HO-1 and NF-κB pathways were identified using the AI tool DeepSeek. Using the STRING database (https://www.string-db.org), PPI analysis was performed separately for asthma-related genes and genes involved in the NRF2/HO-1 and NF-κB pathways.

Structural predictions of the NRF2 (NP_001138884.1) and HO-1 (NP_002124.1) complex were conducted using the AI-based structural prediction tool AlphaFold3. The modeling was performed under two conditions: with and without the addition of iron and zinc ions. The top-ranked predicted structures were visualized using PyMOL 3.0, and the number of hydrogen bonds formed between NRF2 and HO-1 was analyzed.

All analyses were performed in R software (version 4.1.2). Causal inference employed five methods from the “TwoSampleMR” package: inverse-variance weighted (IVW), weighted median, MR-Egger, weighted mode, and simple mode, all two-sided tests. Causal effects were expressed as odds ratios (ORs) with 95% confidence intervals (CIs), representing the risk change per 1-SD increase in exposure. IVW served as the primary method, with all analyses conducted under a random-effects model. Given the evaluation of multiple nutrient exposures (Table 1), multiplicity adjustment was applied across exposures and MR estimators. The Benjamini-Hochberg procedure was used to control the false discovery rate (FDR), with Bonferroni correction performed as a sensitivity analysis. Both unadjusted p-values and adjusted q-values are reported, with statistical significance defined as q < 0.05.

SNPs were selected according to predefined criteria to ensure compliance with the three core instrumental variable assumptions in MR analysis. A total of 325 SNPs meeting these conditions were identified for subsequent causal inference. The F-statistic values ranged from 20.87 to 1787.84, far exceeding the conventional threshold (F > 10), indicating strong instrument strength and minimizing the risk of weak instrument bias. These results demonstrate that all selected SNPs provided substantial statistical power and robust support for subsequent analyses. In summary, the 325 instrumental variables satisfied all MR assumptions and exhibited considerable statistical strength (F-statistic range: 20.87-1787.84), laying a solid foundation for reliable causal inference.

The IVW method was used as the primary analytical approach, with MR-Egger and weighted median methods as supplementary analyses to evaluate the causal relationship between zinc and iron levels and asthma risk. For zinc, each 1-SD increase in circulating iron levels was significantly associated with a 21.8% reduction in asthma risk (OR = 0.782, 95% CI: 0.652–0.937; unadjusted p = 0.029; FDR-adjusted q = 0.08), indicating a protective trend that did not reach significance after multiple-testing correction. The weighted median method yielded consistent results (OR = 0.952, 95% CI: 0.908–0.998; unadjusted p = 0.041; FDR-adjusted q = 0.09), and MR-Egger analysis showed no evidence of bias (p = 0.321).

In contrast, each 1-SD increase in iron levels was significantly associated with a 21.8% reduction in asthma risk (OR = 0.782, 95% CI: 0.652–0.937; unadjusted p = 0.008; FDR-adjusted q = 0.04), remaining robust even after multiple-testing correction. Results from the weighted median and MR-Egger methods were generally consistent (Fig. 1).

After adjustment for multiple testing, the protective association of zinc with asthma was attenuated to a non-significant trend, whereas the protective effect of iron remained significant. This distinction highlights the importance of applying multiplicity correction when evaluating multiple nutrient exposures. Current evidence suggests that zinc may exert protective effects by suppressing inflammatory cytokine secretion and enhancing antioxidant capacity, while iron contributes to maintaining cellular redox balance and supporting immune cell function.

After confirming the independent causal effects of zinc and iron on asthma risk, we further performed MVMR analysis to assess their combined impact. In the model adjusting for both zinc and iron, zinc remained an independent protective factor for asthma (OR = 0.944, 95% CI: 0.894–0.997, p = 0.039), although the effect size was modest, suggesting that zinc may exert its influence through independent biological pathways (Fig. 2). These findings also indicate that the interaction and combined effects of zinc and iron on asthma risk warrant further investigation. Conditional instrument strength analysis indicated a conditional F statistic of 7.8 for zinc and 15.2 for iron, suggesting potential weak instrument bias for zinc in the multivariable model. The covariance matrix of the instruments (Table S3, Figure S2) demonstrated generally low pairwise correlations, consistent with the LD clumping threshold (r² < 0.001). Robust MVMR approaches, including robust IVW-MVMR and the Q-statistic, yielded results consistent with the primary analysis, with zinc remaining an independent protective factor after conditioning on iron, whereas the iron effect attenuated after conditioning on zinc.

To assess the robustness of the causal effects of zinc and iron levels on asthma risk and to exclude excessive bias, we performed multiple sensitivity analyses, including Cochran’s Q test, leave-one-out analysis, MR-Egger intercept test, and MR-PRESSO global test. Heterogeneity and horizontal pleiotropy of the instrumental variables were evaluated in both single and combined models (Table 3). In the univariable models, no significant heterogeneity was detected for zinc (MR-IVW Q = 12.363, p = 0.09, I² = 43.3%) or iron (MR-IVW Q = 8.761, p = 0.555, I² = 0%), and the MR-Egger intercept tests showed no evidence of horizontal pleiotropy (zinc: intercept = -0.007, p = 0.4; iron: intercept = 0.012, p = 0.68). The MR-PRESSO global test did not identify significant outliers for zinc (p = 0.13) or iron (p = 0.21). After removal of potential outliers, the corrected IVW estimates (zinc OR = 0.947, 95% CI = 0.902–0.994, p = 0.029; iron OR = 0.782, 95% CI = 0.652–0.937, p = 0.008) remained consistent with the primary analysis. Steiger directionality tests further supported the causal direction from trace element levels (zinc, iron) to asthma risk (all p < 0.05). In the multivariable model, zinc showed some heterogeneity (Q = 18.226, p = 0.03, I² = 39.6%), but no significant outliers were identified by the MR-PRESSO global test (p = 0.13), and no substantial horizontal pleiotropy was observed (MR-Egger intercept = -0.007, p = 0.4; MR-PRESSO RSSobs = 16.08, p = 0.13) (Fig. 3). These findings suggest that the effect of the instrumental variables on asthma risk was mainly mediated through zinc levels rather than through alternative pathways. Scatter plots, forest plots, funnel plots, and leave-one-out analyses for zinc and iron are provided in Figure S3A-C, confirming that the sensitivity analyses were fully consistent with the primary IVW model.

As shown in Fig. 4, multi-omics analyses revealed key molecular mechanisms involving zinc and iron in asthma pathogenesis. RNA-Seq analysis based on the GEO database (GSE137268) indicated that the expression of NRF2 and HMOX1 was significantly upregulated in asthma patients (p < 0.05), while KEAP1 expression was downregulated (p < 0.01), and these genes were enriched in the antioxidant response pathway (GO:0006979). In addition, the expression levels of NF-κB pathway genes NFKB1 and RELA were positively correlated with asthma severity (Pearson r = 0.32, p < 0.05) and were predominantly enriched in the inflammatory signaling pathway (KEGG: hsa04064).

Protein-protein interaction analysis (STRING database) demonstrated that asthma-related genes (such as ADAM33 and IL4R) directly interact with genes in the NRF2/HO-1 and NF-κB pathways (confidence score > 0.7). Structural modeling using AlphaFold3 and visualization with PyMOL indicated that the presence of zinc and iron ions increased the number of hydrogen bonds within the NRF2/HO-1 complex (9 vs. 7), thereby enhancing its structural stability.

As an exploratory analysis using non-asthma-derived datasets, DFX, an oral iron chelator, treated CD34 + cells from MDS patients with iron overload, RNA-seq analysis identified 570 differentially expressed genes (p < 0.05). Enrichment analysis (Fig. 5A) revealed that among the top 20 enriched pathways, four pathways stood out: neutrophil degranulation (R-HSA-6798695), cellular response to cytokine stimulus (GO:0071345), regulation of response to biotic stimulus (GO:0002831), and response to hypoxia (GO:0001666). These pathways have also been implicated in asthma pathophysiology, suggesting potential mechanistic links that warrant validation in asthma-relevant tissues.

TPEN, a specific zinc chelator, was applied to retinal ganglion cells, resulting in 2,098 significant differentially expressed genes (FDR < 0.05) by RNA-seq. Enrichment analysis (Fig. 5B) revealed multiple pathways relevant to asthma, including endocytosis (GO:0006897), regulation of vesicle-mediated transport (GO:0060627), neutrophil degranulation (R-MMU-6798695), positive regulation of cytokine production (GO:0001819), and response to interferon-gamma (GO:0034341). Although derived from non-airway cells, these pathways overlap with biological processes reported in asthma, and thus serve as hypothesis-generating findings rather than direct evidence.

Figure 6A shows asthma-related genes interact with the NRF2/HO-1 pathway, primarily through NFE2L2, NQO1, and HMOX1. Similarly, Fig. 6B illustrates a close and extensive interaction between asthma-related genes and the NRF2/HO-1 pathway, with multiple interconnections among various genes. These PPI results suggest a functional interplay between zinc and iron metabolism and the NRF2/HO-1 and NF-κB pathways.

As shown in Fig. 7A-B, AlphaFold3 was used to predict the structure of the NRF2/HO-1 complex, and PyMOL was employed to analyze hydrogen bond formation. The complex formed seven hydrogen bonds involving nine amino acids without adding iron and zinc ions. After adding iron and zinc ions, the number of hydrogen bonds increased to nine, with sixteen amino acids involved. These exploratory structural predictions suggested a potential interaction interface between NRF2 and HO-1 that appeared more stable in the presence of zinc and iron ions. However, NRF2 is primarily a transcription factor regulating HMOX1 expression, and HO-1 is an ER-anchored enzyme. A stable binary complex is not an established feature of this pathway. Therefore, these modeling results should be regarded as speculative, serving only as a hypothesis that requires rigorous experimental validation.