Mendelian randomization analysis of the causal relationship between allergic rhinitis and depression and underlying mediation mechanisms☆ Authors: Shaojie Zhang, Rong Wang, Jingjin Weng, Min Li, Min Shi, Jiazhang Wei, Linsong Ye, Xueying Huang, Wenlin Huang, Shenhong Qu Published: 2026 ABSTRACT AbstractBackgroundEpidemiological studies have documented an association between allergic rhinitis (AR) and depression (DE), but causal interpretation is limited by confounding and reverse causation. We applied bidirectional two-sample Mendelian randomization to assess the direction and magnitude of this association and to explore candidate pathways.MethodsSummary-level data were obtained separately from 2 AR GWAS (Study AR1: n = 112 583; Study AR2: n = 83 529) and 2 DE GWAS (Study DE1: n = 484 598; Study DE2: n = 462 933) of European ancestry. Independent single-nucleotide polymorphisms (P < 5 × 10−8; LD r2 < 0.001) served as instruments. Forward and reverse estimates were derived using inverse-variance weighted MR, with MR-Egger and weighted median analyses for sensitivity. A two-step framework estimated separate indirect effects through sleep duration (SD) and anxiety symptoms (ANX). Multivariable MR simultaneously modeled AR and 8 additional traits—dynamic activity ratio, light-intensity physical activity, moderate-to-vigorous activity, loneliness (LON), neuroticism (NEU), immunoglobulin E (IgE), interleukin-6, and fasting insulin—to estimate conditional effects. Pleiotropy and heterogeneity were assessed using MR-Egger intercepts, Cochran’s Q and MR-PRESSO.ResultsAcross 2 AR instrument sets and 2 depression outcomes, genetic liability to AR was consistently associated with a small increase in depression risk (IVW ORs 1.0105–1.0125), whereas reverse-direction analyses provided no evidence that depression liability increased AR risk. AR liability was nominally associated with sleep duration (β = 0.079; 95% CI 0.005–0.153; P = 0.036) and anxiety symptoms (β = 0.031; 95% CI 0.004–0.058; P = 0.025), and was more strongly associated with total IgE (β = 1.981; 95% CI 1.200–2.762; P = 6.60 × 10−7). The separate indirect effects through sleep duration (β = 0.005; 95% CI −0.021 to 0.031; descriptive proportion 47.7%) and anxiety (β = 0.003; 95% CI −0.012 to 0.018; descriptive proportion 24.2%) were imprecise and were not combined. In MVMR, loneliness showed the strongest positive conditional association with depression (OR = 1.243; 95% CI 1.165–1.325; P = 3.40 × 10−11), whereas the conditional AR and IgE estimates were null. Because conditional instrument strength was unavailable, MVMR findings were interpreted as supportive pathway evidence.ConclusionsGenetic liability to AR was consistently associated with a small increase in depression risk across 4 dataset pairings. Although the magnitude alone is insufficient to support individual-level clinical decision-making, the concordant findings add etiological evidence for an AR–depression relationship. Sleep duration and anxiety remain plausible but unconfirmed candidate pathways, while the MVMR findings prioritize loneliness for further validation. Total IgE was more consistent with an AR-related biomarker than an independent mediator.Keywords: Allergic rhinitis, Depression, Mendelian randomization, Sleep duration, Anxiety, Loneliness, Mediation Introduction Allergic rhinitis (AR) is a chronic, immunoglobulin E (IgE) mediated inflammatory condition of the nasal mucosa characterized by rhinorrhea,1 sneezing, nasal congestion, and itching. Affecting up to 30% of adults and 40% of children worldwide,2 AR exacts a significant toll on individual well-being through sleep duration,3 daytime somnolence, and impaired cognitive function, which collectively reduce workplace productivity and quality of life.4 Beyond these somatic manifestations, mounting clinical evidence links AR to elevated rates of mood disturbances, particularly depression (DE), which itself is a principal contributor to global disability-adjusted life years and healthcare costs. Epidemiological surveys and cohort studies have consistently demonstrated that individuals with AR exhibit a 1.3–2.0 fold increased risk of developing depressive symptoms relative to non-AR counterparts.5 Proposed biological mechanisms implicate systemic inflammation, as reflected by elevated cytokines (eg, interleukin-6 [IL-6], tumor necrosis factor-α), which may breach the blood–brain barrier to activate microglia and perturb neurotransmitter networks.6 Additionally, AR-associated nasal obstruction precipitates nocturnal hypoxia and sleep disruption, exacerbating mood dysregulation via hypothalamic–pituitary–adrenal (HPA) axis activation and impaired neuroplasticity.7 However, these observational associations cannot definitively establish causation due to potential confounders socioeconomic status,8,9 comorbid atopic diseases, medication use—and the possibility of reverse causation,10,11 wherein depressive symptoms may enhance AR severity through stress-mediated immunomodulation.12,13 Mendelian randomization (MR) offers a genetically anchored approach to address these limitations,14 using single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs) that proxy for lifetime exposure to AR,15,16 thereby mitigating confounding and reverse causation.17, 18, 19 To date, MR has clarified causal relationships between immune biomarkers (eg, C-reactive protein, IL-6) and neuropsychiatric outcomes, but a targeted MR evaluation of AR and DE remains unexplored.20 Furthermore, conventional MR quantifies total causal effects without identifying intermediary pathways.21 Mediation MR extends this framework by decomposing total effects into direct and indirect components through proposed mediators,22 thereby illuminating the mechanistic route from exposure to outcome.23 Key behavioral and psychological candidate mediators in AR to DE linkage include sleep duration (SD) and anxiety (ANX),24, 25, 26 both of which are susceptible to modulation by chronic inflammation and sensory symptoms of AR.27, 28, 29 Additional factors physical activity patterns (dynamic activity ratio [DAR],30,31 light-intensity physical activity duration [LIPA-Dur],32, 33, 34 moderate-to-vigorous physical activity duration [MVPA-Dur]), psychosocial stressors (loneliness [LON], neuroticism [NEU]),35,36 and systemic biomarkers (IgE, IL-6, fasting insulin [FI]) may further channel the AR effect on DE but have not been simultaneously evaluated in an MR setting.37, 38, 39 In this comprehensive MR study, we addressed 3 objectives: (1) evaluate bidirectional associations between genetic liability to AR and DE; (2) estimate the potential indirect effects through SD and ANX using a two-step MR approach; and (3) estimate conditional associations of 8 additional candidate traits—DAR, LIPA-Dur, MVPA-Dur, LON, NEU, IgE, IL-6, and FI—within an MVMR framework. This integrated analysis was designed to clarify causal direction, quantify effect magnitude and prioritize pathways for further validation. Materials and methods Data sources and study design The overall conceptual framework of this Mendelian randomization study, linking genetic instruments, allergic rhinitis (AR), candidate mediators, potential confounders, and depression, is shown in Fig. 1. All analyses used de-identified summary-level GWAS statistics obtained from the IEU OpenGWAS platform and, where needed, the original consortium portals to ensure version alignment. AR instruments came from 2 UK Biobank GWAS datasets, ukb-b-7178 (n = 112,583; European ancestry; 2018 release) and ukb-a-254 (n = 83,529; 2017). Depression outcomes were drawn from ebi-a-GCST90038650 (n = 484,598; 2021) and the UK Biobank self-report phenotype ukb-b-12064 (n = 462,933; 2015). Mediators were selected for mechanistic relevance to both AR and depression and included sleep duration (ebi-a-GCST003839; n = 127,573), anxiety symptoms (ukb-d-20544_15; n = 117,751), 3 physical-activity metrics—dynamic activity ratio (ebi-a-GCST90061430; n = 88,411), light-intensity activity duration (ebi-a-GCST90061421; n = 88,411), and moderate-to-vigorous activity duration (ebi-a-GCST90061422; n = 88,411)—psychosocial traits loneliness (ukb-b-8476; n = 455,364) and neuroticism (ukb-a-230; n = 274,108), and immunometabolic biomarkers total IgE (prot-a-1456; n = 3301), interleukin-6 (ebi-a-GCST90012005; n = 21,758), and fasting insulin (ebi-a-GCST90002237; n = 29,792). Full dataset descriptions, accession IDs, sample sizes, ancestry and primary citations are provided in Supplementary Table S1. All GWAS summary statistics analyzed in this study are publicly available via the IEU OpenGWAS platform. Study design schematic. Stage 1 used bidirectional two-sample MR to estimate the association between genetic liability to allergic rhinitis (AR) and depression (DE). Stage 2 evaluated 10 candidate traits and estimated separate indirect effects through sleep duration and anxiety. Stage 3 used MVMR including AR and 8 additional candidate traits to estimate conditional associations. Sensitivity analyses assessed horizontal pleiotropy, heterogeneity and outliers Instrumental variable selection, harmonization, and code implementation We selected genome-wide significant variants for each exposure or candidate trait (P < 5 × 10−8) and enforced independence by LD clumping at r2 < 0.001 within a 10,000-kb window. For each SNP, we calculated marginal instrument strength as F = β2/SE2. Effect alleles were harmonized across exposure, mediator and outcome datasets; strand-ambiguous palindromic variants at intermediate allele frequency and variants that could not be aligned unambiguously were excluded.40 For MVMR, genome-wide significant variants across the modeled traits were pooled, re-clumped using the same thresholds and used as a unified instrument set with trait-specific coefficients.41 Per-trait instrument profiles are provided in Supplementary Table S31. These marginal F-statistics do not establish instrument strength conditional on the other exposures in MVMR. Two-sample Mendelian randomization analysis and rationale Two-sample Mendelian randomization analyses were conducted using the TwoSampleMR R package to assess causal relationships in both directions. Our primary estimation relied on the inverse variance weighted (IVW) method under a multiplicative random effects model, chosen for its high efficiency when horizontal pleiotropy is balanced.42 To evaluate and guard against potential violations of the IV assumptions, we supplemented IVW with MR-Egger regression allowing for an intercept term to detect and adjust for directional pleiotropy and the weighted median estimator, which yields consistent effect estimates even if up to half of the instruments are invalid.43 Causal estimates are reported on the log-odds (log OR) scale with corresponding odds ratios (ORs) for binary outcomes; for continuous traits, estimates are reported as β per unit (or SD) increase in genetically predicted exposure. By triangulating results from multiple estimators, we strengthened causal inference and enhanced our ability to identify bias. Two-step mediation Mendelian randomization and interpretation We conducted two-step mediation MR to estimate potential indirect pathways from genetic liability to allergic rhinitis to depression through selected traits. In the first step, we estimated the effect of allergic rhinitis on each candidate trait; in the second, we estimated the effect of each trait on depression. The direct effect of allergic rhinitis conditional on each candidate trait was obtained from MVMR, and the indirect effect was calculated as the difference between the total and direct estimates. The proportion mediated was calculated as the indirect effect divided by the total effect. Because the pathways may overlap and the indirect-effect confidence intervals included zero, these proportions were interpreted as descriptive estimates and were not summed. Multivariable Mendelian randomization for multiple mediators We applied MVMR to estimate conditional effects of AR and 8 additional candidate traits while accounting for their genetic intercorrelations. Variants reaching genome-wide significance for any modeled trait were pooled and subjected to linkage disequilibrium pruning to create a unified instrument set. MVMR inverse-variance weighted coefficients were treated as supportive, hypothesis-generating estimates because conditional instrument-strength statistics were unavailable. Mediation MR and uncertainty quantification For two-step mediation, we estimated (i) AR → candidate-trait and (ii) candidate-trait → depression effects using the same MR framework. The direct effect of AR conditional on each candidate trait was obtained from MVMR. Indirect effects were computed using the difference approach (βindirect = βtotal − βdirect), and their standard errors were derived by normal-approximation propagation of the constituent MR standard errors. Proportions mediated were calculated as βindirect/βtotal and interpreted descriptively because both indirect-effect confidence intervals included zero. Multiple testing and correlated mediators Because the candidate traits are substantially and variably genetically correlated, we report unadjusted P values for the univariable MR screens and emphasize effect sizes with confidence intervals. Joint MVMR coefficients were treated as secondary, supportive estimates rather than definitive evidence of mediation. For transparency, all estimates and raw P values are provided together with marginal instrument-strength summaries (Supplementary Table S31) and sensitivity diagnostics (Supplementary Table S30). Sensitivity and heterogeneity assessments Primary causal estimates used inverse-variance weighted (IVW) MR under a multiplicative random-effects model, complemented by weighted median and MR-Egger to assess robustness to horizontal pleiotropy. We evaluated directional pleiotropy with the MR-Egger intercept, heterogeneity with Cochran’s Q for IVW and MR-Egger, and global outliers with MR-PRESSO. Full diagnostics (Egger intercept p, Q statistics/p, MR-PRESSO global p) are summarized in Supplementary Table S30. We tested for directional pleiotropy using the MR-Egger intercept, assessed heterogeneity using Cochran’s Q and evaluated global outliers using MR-PRESSO. For MVMR, no single blanket multiplicity correction was applied; all conditional estimates are reported with confidence intervals and raw P values and are interpreted in light of correlated exposures and unavailable conditional instrument-strength statistics. Analyses were performed in R using validated MR and pleiotropy-detection packages. Two-sided P < 0.05 was used as a nominal reference threshold rather than as evidence of clinical importance. Results 1.Consistent and Unidirectional Causal Effect of Allergic Rhinitis on Depression Across 8 Bidirectional Two-Sample MR Analyses Consistent and Unidirectional Causal Effect of Allergic Rhinitis on Depression Across 8 Bidirectional Two-Sample MR Analyses As shown in Fig. 2, we paired 2 AR instrument sets with 2 depression outcomes in both directions. Across the 4 forward analyses, IVW estimates ranged from β = 0.01049 to 0.01238, corresponding to ORs of 1.0105–1.0125 (P = 0.00520–0.02716). In contrast, all 4 reverse-direction IVW estimates had confidence intervals spanning the null (P = 0.14920–0.66371). The directionally concordant forward estimates across both AR definitions and both depression outcomes strengthen etiological evidence for an AR-to-depression relationship, while their proximity to the null indicates a small effect that is insufficient on its own for individual risk prediction. Full estimator-specific results are provided in Supplementary Table S26, and marginal instrument-strength summaries are provided in Supplementary Table S31. 2.Genetic Liability to Allergic Rhinitis Modulates Behavioral, Psychological, and Immunometabolic Traits Genetic Liability to Allergic Rhinitis Modulates Behavioral, Psychological, and Immunometabolic Traits Bidirectional two-sample MR analyses of allergic rhinitis and depression. The top 4 entries show AR-to-depression estimates using 2 AR instrument sets and 2 depression outcomes, with IVW ORs ranging from 1.0105 to 1.0125. The bottom 4 entries show reverse analyses, all with confidence intervals spanning the null. The forward estimates were directionally concordant but small in magnitude. Black squares, blue circles and red triangles represent inverse-variance weighted, MR-Egger and weighted median estimates, respectively; horizontal lines indicate 95% confidence intervals Our two-sample MR analyses of 10 candidate traits showed distinct evidence patterns (Fig. 3). Genetic liability to AR was nominally associated with longer sleep duration (IVW β = 0.079; 95% CI 0.005–0.153; P = 0.036) and higher anxiety symptoms (IVW β = 0.031; 95% CI 0.004–0.058; P = 0.025). Total IgE showed the strongest association with AR liability (47 SNPs; IVW β = 1.981; 95% CI 1.200–2.762; P = 6.60 × 10−7). Interleukin-6, fasting insulin, loneliness, neuroticism and the physical-activity measures were not detectably associated with AR liability (P > 0.10). These findings nominated sleep duration and anxiety as candidate pathways and total IgE as an AR-associated marker, without by themselves establishing mediation. 3.Differential Causal Impacts of Candidate Mediators on Depression Differential Causal Impacts of Candidate Mediators on Depression Associations of genetic liability to allergic rhinitis with 10 candidate traits. AR liability was nominally associated with sleep duration (β = 0.079; 95% CI 0.005–0.153; P = 0.036) and anxiety symptoms (β = 0.031; 95% CI 0.004–0.058; P = 0.025), and was more strongly associated with total IgE (β = 1.981; 95% CI 1.200–2.762; P = 6.60 × 10−7). Other traits were not detectably associated with AR liability. Continuous outcomes are interpreted on the β scale The two-sample MR analyses of candidate traits and depression are summarized in Fig. 4. Higher genetically predicted anxiety (OR = 1.055; 95% CI 1.034–1.076; P = 2.54 × 10−7), loneliness (OR = 1.196; 95% CI 1.155–1.239; P = 1.43 × 10−23) and neuroticism (OR = 1.015; 95% CI 1.013–1.017; P = 1.96 × 10−59) were associated with depression. Longer sleep duration (OR = 0.994; 95% CI 0.988–1.000; P = 0.043) and moderate-to-vigorous physical activity (OR = 0.995; 95% CI 0.992–0.999; P = 0.023) showed nominal inverse associations. Dynamic activity ratio, light-intensity activity, total IgE, interleukin-6 and fasting insulin were not detectably associated with depression. These univariable estimates informed pathway selection but did not alone establish mediation from AR. 4.Quantitative Estimation of Mediation by Sleep Duration and Anxiety Quantitative Estimation of Mediation by Sleep Duration and Anxiety Differential Causal Impacts of Candidate Mediators on Depression. Forest plot of inverse-variance weighted (IVW) causal estimates for 10 mediators on depression risk. Each row shows the IVW β coefficient per unit increase in the genetically predicted mediator—with 95% confidence intervals—to the left and the corresponding odds ratio with 95% confidence intervals to the right. Overlaid MR-Egger (blue circles) and weighted median (red triangles) estimates illustrate robustness. Mediators with IVW P < 0.05 are highlighted in bold. Psychological traits (sleep duration, anxiety, loneliness, neuroticism), physical activity measures (dynamic activity ratio, light-intensity and moderate-to-vigorous activity durations), and immunometabolic markers (IgE, IL-6, fasting insulin) are each shown, revealing that anxiety, loneliness and neuroticism have the strongest positive effects on depression, sleep duration and moderate-to-vigorous activity are modestly protective, and immunometabolic traits show null associations We used a two-step MR framework to estimate separate indirect effects through sleep duration and anxiety (Fig. 5). The difference method was applied to each pathway separately, and uncertainty was quantified using the delta method. Indirect-effect estimates through sleep duration and anxiety. The difference-method indirect effect through sleep duration was β = 0.005 (SE = 0.013; 95% CI −0.021 to 0.031; descriptive proportion 47.7%), and the indirect effect through anxiety was β = 0.003 (SE = 0.008; 95% CI −0.012 to 0.018; descriptive proportion 24.2%). Both confidence intervals included zero; the proportions were interpreted separately and were not summed The indirect effect through sleep duration was β = 0.005 (SE = 0.013; 95% CI −0.021 to 0.031), corresponding to a descriptive mediated proportion of 47.7%. The indirect effect through anxiety was β = 0.003 (SE = 0.008; 95% CI −0.012 to 0.018), corresponding to a descriptive mediated proportion of 24.2%. Both confidence intervals included zero. Neither pathway was statistically confirmed, and the 2 proportions were not combined. The point estimates therefore identify sleep duration and anxiety as plausible candidate pathways rather than confirmed mediators. Their main value is to prioritize specific behavioral and psychological processes for prospective replication and complementary longitudinal or interventional study. 5.Independent Direct Effects in Multivariable MR Highlighting Key Mediators Independent Direct Effects in Multivariable MR Highlighting Key Mediators In the joint MVMR model of AR and 8 additional candidate traits, loneliness showed the strongest positive conditional association with depression (OR = 1.243; 95% CI 1.165–1.325; P = 3.40 × 10−11) (Fig. 6). Neuroticism showed a nominal inverse conditional coefficient (OR = 0.995; 95% CI 0.991–0.999; P = 0.013), whereas the conditional estimates for AR (OR = 0.987; 95% CI 0.967–1.007; P = 0.192) and total IgE (OR = 0.999; 95% CI 0.995–1.004; P = 0.827) were null. The other conditional estimates were also not statistically significant. Because conditional instrument strength could not be assessed, these coefficients were interpreted as supportive pathway-prioritization evidence rather than definitive proof of independent mediation. 6.Sensitivity Analyses for Pleiotropy and Heterogeneity Sensitivity Analyses for Pleiotropy and Heterogeneity Conditional MVMR-IVW estimates for depression. The joint model included genetic liability to allergic rhinitis and 8 additional candidate traits: fasting insulin, interleukin-6, light-intensity physical activity duration, moderate-to-vigorous physical activity duration, dynamic activity ratio, total IgE, neuroticism and loneliness. Loneliness showed the strongest positive conditional association (OR = 1.243; 95% CI 1.165–1.325; P = 3.40 × 10−11), whereas AR and IgE were null. Conditional instrument-strength statistics were unavailable, so estimates were considered supportive As shown in Fig. 7, a comprehensive suite of sensitivity diagnostics underpins the credibility of our Mendelian randomization findings across all forward, mediation and multivariable analyses. We first examined MR-Egger intercepts to detect unbalanced horizontal pleiotropy: 9 of the 10 forward analyses yielded non-significant intercepts (P∖_Egger ranging from 0.2416 to 0.9200), indicating that directional pleiotropy was not a material concern. Only the MVPA-Dur model showed a borderline intercept (β∖_intercept = −0.0044; P∖_Egger = 0.0318), but its inverse-variance weighted and weighted median estimates remained closely aligned, suggesting that even this modest pleiotropic signal did not materially distort the causal inference. Sensitivity analyses for pleiotropy and heterogeneity. MR-Egger intercept P values evaluate directional pleiotropy, Cochran’s Q assesses heterogeneity and MR-PRESSO global-test P values indicate outlier signals. Although systematic directional pleiotropy was not observed, nominal pleiotropy, heterogeneity or outlier signals occurred in several individual pathway models. These diagnostics support cautious interpretation of the pathway findings Next, heterogeneity among SNP-specific estimates was quantified by Cochran’s Q under the IVW framework. Several downstream mediators exhibited nominal heterogeneity (MVPA-Dur: Q∖_IVW = 199.17, P∖_IVW <0.0001; loneliness: Q∖_IVW = 78.92, P∖_IVW = 0.0024; neuroticism: Q∖_IVW = 179.43, P∖_IVW = 0.0027; IgE: Q∖_IVW = 80.09, P∖_IVW = 0.0014; IL-6: Q∖_IVW = 149.55, P∖_IVW = 0.0936; FI: Q∖_IVW = 147.43, P∖_IVW = 0.0164), reflecting genuine variability in SNP effects. Crucially, when these same exposures were modeled jointly in multivariable MR, Q-statistics were markedly reduced (all P∖_IVW >0.10), demonstrating that adjustment for correlated pathways effectively harmonized variant contributions. Finally, MR-PRESSO global tests flagged outlier variants in 6 models—MVPA-Dur (P∖_PRESSO = 0.001), loneliness (P∖_PRESSO = 0.001), neuroticism (P∖_PRESSO = 0.015), IgE (P∖_PRESSO = 0.007), IL-6 (P∖_PRESSO = 0.008), and FI (P∖_PRESSO = 0.045). Removal of these outliers altered IVW β estimates by less than 0.01 in each case, confirming that no single SNP unduly influenced the results. Overall, the sensitivity analyses did not indicate systematic directional pleiotropy, but heterogeneity, outlier and nominal pleiotropy signals occurred in several individual pathway models. These findings do not negate the directionally concordant primary AR-to-depression estimates, but they support cautious interpretation of the mediation and MVMR results. Discussion Epidemiological studies have long documented comorbidity between allergic rhinitis (AR) and depression (DE), but causal direction and magnitude have remained uncertain. Using large-scale two-sample MR, we observed directionally concordant AR-to-depression estimates across 4 forward analyses and no evidence for the reverse direction. The OR range of 1.0105–1.0125 indicates a small association. For context, a published MR study reported an OR of 1.18 (95% CI 1.09–1.28) for depression per 1-SD higher body mass index.44 Because the exposure scales and outcome definitions differ, this comparison is contextual rather than direct. The smaller AR estimate is insufficient on its own to support individual-level clinical decision-making; however, its consistency across instrument sets and outcome definitions adds etiological evidence for an AR–depression relationship. The two-step analyses identified sleep duration and anxiety as plausible candidate pathways, but neither indirect-effect confidence interval excluded zero. The descriptive mediated proportions were 47.7% and 24.2%, respectively, and were not summed because the pathways may overlap and neither was statistically confirmed. This uncertainty does not remove the hypothesis-generating value of the pathway analysis; rather, it defines the level of evidence and the need for prospective validation. Indirect effects combine multiple causal estimates and can therefore be imprecise. Mediation inference also depends on adequate instrument strength, limited horizontal pleiotropy and correct model specification. Accordingly, these results support pathway prioritization rather than a claim that sleep duration and anxiety together explain a fixed proportion of the AR–depression relationship. In MVMR, loneliness showed the strongest positive conditional association with depression, whereas the conditional AR and IgE estimates were null. The strong AR-to-IgE association, combined with null IgE-to-depression evidence in univariable MR and a null conditional IgE estimate, is more consistent with IgE as an AR-related biomarker than as a supported mediator. Because conditional instrument-strength statistics were unavailable and the modeled traits are correlated, the reliability of individual conditional coefficients cannot be fully established. The MVMR results therefore prioritize pathways for further study but do not establish an independent biological mechanism. The translational relevance of these findings lies in etiological clarification and pathway prioritization rather than an immediate treatment recommendation. The consistent AR-to-depression estimates support greater awareness of mental-health comorbidity in AR, while anxiety, sleep-related phenotypes and loneliness can be evaluated prospectively as markers of vulnerability. If confirmed in longitudinal and interventional studies, these pathways may help inform integrated risk assessment and supportive care. The present MR estimates alone do not establish that modifying any single pathway will prevent depression. A key strength of this study is the layered evidence framework combining 2 AR instrument definitions, 2 depression outcomes, bidirectional analyses, two-step pathway estimates and MVMR. This design distinguishes a reproducible small total association from exploratory indirect and conditional effects. Important limitations remain. The GWAS datasets were predominantly of European ancestry, limiting transferability to other populations. Standard two-sample MR estimates an average linear effect and cannot characterize non-linear or threshold relationships without individual-level data. Several candidate traits were self-reported, and measurement error may attenuate their estimates. Finally, the indirect effects were imprecise and conditional instrument-strength statistics were unavailable. Replication in independent, multi-ancestry datasets and complementary longitudinal or interventional designs will be important for refining the clinical relevance of these findings. In conclusion, our integrative MR framework supports a small but directionally consistent association between genetic liability to allergic rhinitis and depression across multiple dataset pairings. Although the effect is not sufficient on its own for individual-level clinical decisions, it strengthens etiological evidence for an AR–depression relationship and highlights anxiety, sleep-related phenotypes and loneliness as priorities for further validation. Total IgE appears more consistent with an AR-related biomarker than a confirmed mediator. Conclusion Genetic liability to allergic rhinitis was consistently associated with a small increase in depression risk, with no evidence for the reverse direction. Although this effect size is insufficient on its own to support individual-level clinical decision-making, the concordant findings across instrument sets and outcome definitions add etiological evidence for the AR–depression relationship. Sleep duration and anxiety remain plausible but unconfirmed pathways, and the MVMR findings prioritize loneliness for further study; total IgE was not supported as an independent mediator. Disclosure statement of generative artificial intelligence (AI) and AI-assisted technologies The document’s grammar was corrected using an AI-assisted tool, Grammarly, which improved readability. Author contributions SZ and RW conceptualized the study and drafted the manuscript. JW, ML, and MS performed data curation and MR analyses. JW and LY contributed to code development and statistical interpretation. XH and WH managed data harmonization and sensitivity analyses. SQ supervised the project and finalized the manuscript. All authors reviewed and approved the final version. Data availability statement All GWAS summary statistics used in this study are publicly accessible via the IEU OpenGWAS platform, FinnGen, and the UK Biobank. Additional data and code supporting this study are provided in Supplementary File S1. Ethics statement This study used publicly available, de-identified summary data; ethical approval and informed consent were obtained in the original GWAS studies. Funding This work was supported by the National Natural Science Foundation of China [grant number: 81960186], the Natural Science Foundation of Guangxi [grant number: 2024GXNSFAA010060, 2017GXNSFAA198013]. Declaration of competing interest The authors declare no competing interests. Acknowledgments We thank the investigators and participants of the IEU OpenGWAS, FinnGen, and UK Biobank studies for making their data publicly available. We also acknowledge the contributions of the AR, DE, and mediator GWAS consortia. Footnotes Full list of author information is available at the end of the article. Supplementary data to this article can be found online at https://doi.org/10.1016/j.waojou.2026.101458. Supplementary data The following is the Supplementary data to this article: Multimedia component 1 References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement All GWAS summary statistics used in this study are publicly accessible via the IEU OpenGWAS platform, FinnGen, and the UK Biobank. Additional data and code supporting this study are provided in Supplementary File S1.