Dissecting the causal association between social or physical inactivity and depression: a bidirectional two-sample Mendelian Randomization study

Major depressive disorder (MDD) is a common, disabling condition with substantial global burden. Beyond established social and physiological risk factors (e.g., poverty, unemployment, family history, illness, life events), leisure/social and physical activities are potentially modifiable factors of growing interest. Observational evidence suggests physical activity reduces incident MDD across ages and regions, and social/leisure activities associate with fewer depressive symptoms. Conversely, MDD symptoms (anhedonia, low energy) may reduce participation in social and physical activities, implying bidirectionality that observational studies cannot disentangle. Physical activity’s links to obesity suggest adiposity traits could mediate effects on MDD. Mendelian randomization (MR), using SNPs as instrumental variables, can infer causality less affected by confounding and reverse causation. Prior MR showed protective accelerometer-based activity effects on MDD and exposure-wide analyses included activity traits, but inactivity per se has been underexplored. Imaging-derived brain phenotypes (IDPs) may also mediate relationships between inactivity and MDD. This study aimed to examine bidirectional causal effects between MDD and multiple types of leisure/social and physical activities, including “none of the above” inactivity measures, and to assess mediation by obesity-related measures and brain IDPs.

Design: Bidirectional two-sample Mendelian randomization (MR) using publicly available GWAS summary statistics, restricted to European-ancestry participants. Ethical approvals were from original studies. Data sources: Leisure/social activity traits (sports clubs/gyms, pubs/social clubs, religious groups, adult education classes, other group activities, and “none of the above” indicating leisure/social inactivity) from MRC-IEU UK Biobank (n=461,369). Physical activity traits over the last 4 weeks (walking for pleasure, strenuous sports, light DIY, heavy DIY, other exercises, and “none of the above” indicating physical inactivity) from MRC-IEU UK Biobank (n=460,376). MDD summary statistics from meta-analysis (PGC + UK Biobank; 170,756 cases, 329,443 controls). For reverse/subgroup analyses avoiding sample overlap when MDD was outcome, an alternative MDD GWAS excluding UK Biobank and 23andMe (45,591 cases, 97,674 controls) was used. Potential mediators included adiposity traits: body fat percentage (BFP; n=454,633), body mass index (BMI; n=461,460), hip circumference, waist circumference (both ~462k; MRC-IEU), and waist-to-hip ratio (WHR; GIANT, n=212,244). Brain IDPs: 3,144 structural/functional phenotypes from UK Biobank discovery dataset (n=8,428; BIG browser). Instrument selection and harmonization: Lead SNPs associated with exposures at P<5×10−8; when ≤10 instruments available, a relaxed threshold P<1×10−5 was used. LD clumping at r2<0.001 within 10,000 kb. Proxy SNPs (r2>0.8) substituted when needed. Minor allele frequency filter applied and exposure-outcome datasets harmonized to remove palindromic SNPs and align effect alleles. Instruments associated with potential confounders (e.g., smoking, alcohol) were removed using PhenoScanner queries. Instrument strength quantified by F-statistics; F<10 considered weak and excluded. Analyses limited to European ancestry. Primary MR analysis: Inverse-variance weighted (IVW) method as the main estimator. Complementary methods included MR-Egger, weighted median, simple mode, and weighted mode. Random-effects IVW used when heterogeneity (Cochran’s Q) was present. Sensitivity analyses and pleiotropy: MR-Egger intercept tested directional pleiotropy; leave-one-out IVW assessed influence of individual SNPs; MR-PRESSO detected/corrected outlier instruments indicating horizontal pleiotropy. Multiple testing: associations with P<0.004 (Bonferroni for 12 tests) considered significant; 0.004≤P<0.05 considered suggestive. Potential bias from sample overlap addressed using MRlap. Mediation analyses (two-step MR): For significant activity→MDD associations, two-step MR assessed mediation via adiposity traits (BFP, BMI, HC, WC, WHR) and, for inactivity exposures, via 3,144 brain IDPs. Step 1 estimated effects of inactivity/activity on mediators; Step 2 estimated effects of mediators on MDD. Indirect effects calculated using product-of-coefficients; for correlated adiposity traits, multivariable MR estimated direct effects to identify specific mediators. IDPs with significant effects in both steps and with heritability evidence were considered mediators or, when indirect and direct effects had opposite signs, masking effects.

- Activity→MDD (IVW, outliers removed): - Leisure/social: Sports clubs or gyms protective (OR 0.51; 95% CI 0.39–0.67; P=1.05×10−10). Leisure/social inactivity increased MDD risk (OR 1.64; 95% CI 1.29–2.08; P=5.14×10−5). Other leisure/social types showed no significant effects. - Physical: Walking for pleasure (OR 0.13; 95% CI 0.07–0.26; P=4.25×10−9), strenuous sports (OR 0.43; 95% CI 0.27–0.71; P=9.16×10−6), light DIY (OR 0.47; 95% CI 0.32–0.68; P=9.65×10−5), heavy DIY (OR 0.51; 95% CI 0.35–0.76; P=7.46×10−3), and other exercises (OR 0.44; 95% CI 0.35–0.55; P=3.30×10−12) were protective. Physical inactivity markedly increased MDD risk (OR 3.67; 95% CI 2.02–6.68; P=1.99×10−12). - Sensitivity: No evidence of horizontal pleiotropy (MR-Egger intercepts); results robust to leave-one-out; instruments strong (high F). Subgroup analyses avoiding overlap remained significant for most traits. - MDD→Activity (IVW, outliers removed): - Leisure/social: MDD associated with lower attendance at sports clubs/gyms (OR 0.96; 95% CI 0.95–0.98; P=1.82×10−7) and with increased leisure/social inactivity (OR 1.03; 95% CI 1.01–1.04; P=9.89×10−4). - Physical: MDD reduced heavy DIY (OR 0.97; 95% CI 0.96–0.99; P=1.54×10−3), other exercises (OR 0.96; 95% CI 0.94–0.98; P=1.88×10−4), and to a lesser extent strenuous sports (OR 0.99; P=0.044). MDD increased physical inactivity (OR 1.01; 95% CI 1.01–1.02; P=7.96×10−4). - Mediation by adiposity: - Multivariable MR identified BFP (OR 1.34; 95% CI 1.07–1.67; P=0.01) and BMI (OR 1.23; 95% CI 1.02–1.50; P=0.03) as causal risk factors for MDD. - Leisure/social inactivity→MDD: mediated by BFP (indirect effect ≈0.06; 12.99% of total) and BMI (≈0.06; 11.80%). - Physical inactivity→MDD: mediated by BFP (≈0.23; 17.82%) and BMI (≈0.19; 14.38%). - Mediation/masking by brain IDPs (two-step MR): - Leisure/social inactivity→MDD: partial mediation via grey matter volume in right Heschl’s gyrus (≈4.03%) and third ventricle volume (≈5.01%); masking effect via weighted-mean orientation dispersion index (OD) of left acoustic radiation (≈8.43%). - Physical inactivity→MDD: partial mediation via surface area of left precentral gyrus (≈3.29%), mean OD in right sagittal stratum (≈4.05%), and mean ISOVF in right superior corona radiata (≈3.86%); masking via right caudate volume (≈4.74%). Overall, findings support a bidirectional relationship where social/physical activities reduce MDD risk, while MDD diminishes engagement and increases inactivity; inactivity elevates MDD risk partly via adiposity and specific brain phenotypes.

This bidirectional two-sample MR provides genetic evidence that engaging in specific social and physical activities causally reduces MDD risk, while genetic liability to MDD causally reduces participation in several activities and increases inactivity. The results align with prospective observational findings on activity’s protective effects and with MDD symptoms (anhedonia, low energy) that discourage activity. Notably, inactivity (both social/leisure and physical) showed strong adverse effects on MDD risk, suggesting a potential vicious cycle. Mechanistically, mediation analyses indicate that part of inactivity’s effect operates through increased adiposity (BFP, BMI), consistent with prior MR linking sedentary behavior to higher BMI and adiposity to depression risk. Brain IDP analyses highlighted neuroanatomical and microstructural pathways: specific regional measures partially mediated or masked inactivity’s effects on MDD, suggesting complex brain-based adaptations. These neurobiological findings are consistent with literature showing exercise-induced brain plasticity that may protect against depression. Collectively, the results underscore the public health importance of promoting both social and physical activities to prevent MDD and suggest that targeting adiposity and brain-related pathways could enhance interventions.

The study identified potential bidirectional causal relationships between several activity types (sports clubs/gyms, strenuous sports, heavy DIY, other exercises) and MDD. Social and physical activities appear protective against MDD, whereas MDD reduces participation and increases inactivity. Social/physical inactivity increases MDD risk, with partial mediation by adiposity (BFP, BMI) and mediation/masking by specific brain imaging phenotypes. These findings support developing prevention and intervention strategies that promote activity and address adiposity to mitigate MDD risk. Future research should refine exposure measurement (including objective activity monitoring), explore temporal dynamics (first-episode vs. relapse), assess different ancestries and sexes, and further elucidate neurobiological mediators.

- Genetic instruments, though strongly associated, are imperfect proxies for complex behaviors; physical activity measured over the prior 4 weeks may not reflect long-term habits while MDD reflects longer-term pathology. - Predominantly self-reported activity measures can be influenced by mood and cognitive biases; objective actigraphy was not used here. - Analyses restricted to European ancestry; generalizability to other populations is uncertain. Sex-specific effects could not be examined due to summary-level data. - For some traits, instrument thresholds were relaxed (P<1×10−5) to ensure sufficient SNPs, potentially increasing pleiotropy risk; complete elimination of horizontal pleiotropy is difficult with summary data. - Potential sample overlap between exposure and outcome GWAS could bias estimates; addressed with strong instruments, overlap-free subgroup analyses, and MRlap corrections, but residual bias may remain. - Binary exposure coding limits inference on dose (frequency/intensity/duration) of activities; several reverse-direction effects had small magnitudes and warrant replication.