Effects of urban living environments on mental health in adults

More than half of the world’s population lives in urban areas, a proportion projected to rise to two-thirds by 2050. Urban environments are characterized by denser built infrastructure, reduced access to green spaces, greater exposure to substances, and more stressful social conditions, but also potentially better infrastructure and work opportunities. Evidence on how urban living environments impact mental health is mixed: while some studies suggest higher risks of mental health conditions (especially depression and anxiety) in urban residents, findings are inconsistent and prior research has typically examined single environmental factors (for example, green space or socioeconomic deprivation) in isolation. The relationships between complex, co-occurring urban exposures and both psychiatric symptoms and brain structure are unclear, and responses to environmental adversity vary across individuals, in part due to genetic differences. This study aims to capture the multidimensional complexity of urban living by integrating physical and socioeconomic measures into environmental profiles, relating them to groups of psychiatric symptoms, identifying brain regions that mediate these relationships, and testing whether genetic variation moderates the brain-mediated effects.

The paper situates its work within literature showing urbanicity’s links to mental health outcomes, with emphasis on depression and anxiety as prevalent urban-related symptom domains. Prior studies often focused on single environmental exposures (for example, green spaces improving mental health; socioeconomic deprivation linked to poorer outcomes) and on specific disorders such as schizophrenia. Research has also highlighted individual differences in environmental sensitivity and gene–environment interplay, including stress-response pathways. However, comprehensive models integrating multiple macroenvironmental factors with brain structure and genetics to explain psychiatric symptoms have been lacking. The authors address this gap by modeling complex environmental profiles rather than isolated factors, incorporating genetic moderation and neurobiological mediation.

Design and cohorts: The study used UK Biobank (UKB) data. A total of 156,075 adults (age 41–77; mean ~59 years) with complete urban environmental and psychiatric symptom data were divided into two subsets: UKB-non-NI (no neuroimaging; n = 141,087) and UKB-NI (with neuroimaging; n = 14,988). Confounders (age, sex, assessment center; plus total intracranial volume for imaging analyses) were adjusted.

Urban environmental variables: 128 variables across 53 categories captured air and sound pollution, traffic, green and water proximity, coastal proximity, indices of multiple deprivation (IMD), building class, destination accessibility (distance to 33 services), land-use density (LD; 46 categories), normalized difference vegetation index (NDVI; greenness), street network (SN) accessibility (20 indices at 400 m radius), and terrain (slope). Outliers >4 MAD were removed; variables were screened for near-zero variance.

Category construction via CFA: Tenfold cross-validated confirmatory factor analysis (CFA) reduced redundancy in multi-item categories to 19 latent variables (in addition to 34 single-item categories), yielding 53 independent urban environmental categories. Model fit criteria included TLI > 0.95, CFI > 0.95, RMSEA < 0.06, SRMR < 0.08.

Psychiatric symptoms: From 44 UKB mental health questionnaire items, 21 items with acceptable completeness and distribution were retained (symptoms covering affective, anxiety, and personality-related domains).

Genomics: UKB imputed data underwent participant-level QC (sex mismatches, aneuploidy, kinship, heterozygosity/missingness, non-White ancestry, missing PCs excluded) and SNP-level QC (MAF ≥ 0.001; INFO > 0.3), yielding 275,988 participants and 13,918,727 SNPs for GWAS. Neuroimaging: T1-weighted MRI (3T Siemens Skyra) provided 139 regional gray-matter volumes (111 cortical/subcortical, 28 cerebellar) via standard UKB pipelines.

SCCA to link environment and symptoms: Sparse canonical correlation analysis (sCCA; mixOmics) related 53 environmental categories to 21 symptoms in UKB-non-NI with a 90%/10% train/test split. Stability selection used 100 half-sample resamples with a 90% non-zero loading threshold to retain stable variables; final models were refit without sparsity. Significance was assessed via 1,000 permutations and FDR control. Successive modes were obtained via projection deflation. Reliability checks included bootstrapping across sample sizes, resampling stability, sex-specific analyses, and sensitivity to household sharing (repeating sCCA in 122,516 unique-household participants identified via multiple household indicators).

Replication of sCCA: The sCCA pipeline was replicated in the independent UKB-NI sample (n = 14,988; 90%/10% split).

Pleiotropy of environmental factors: Non-sCCA regression quantified the fraction of explained variance (FEV) by each environmental profile for each symptom group.

GWAS and gene-set analyses: GWAS were performed on the canonical covariates (symptom-group scores) identified from sCCA (n = 76,508 with complete genetics/environment/symptoms). BGENIE additive models included age, sex, assessment center, batch, and 10 ancestry PCs; Bonferroni-corrected significance thresholds were applied. SNPs were mapped to genes via FUMA. Gene-set enrichment was tested with ToppGene; gene expression was examined using the Human Protein Atlas.

Gene scores and moderation: For significant genes, per-gene polygenic scores were computed using PLINK 2.0 with LD clumping (clump-p1 = 1, clump-r2 = 0.5, clump-kb = 250 kb); scores summed risk alleles weighted by GWAS betas across index SNPs. Gene-score associations and replication were evaluated in UKB-NI (n = 8,705 with complete data).

msCCA with brain volumes: Multiple sparse CCA related environmental profiles, brain volumes (139 ROIs), and symptom groups in UKB-NI (90%/10% split). Stability selection threshold was 85%; permutation tests (10,000) assessed significance, deriving brain components linked to each environment–symptom pairing.

Moderated mediation: Using PROCESS Model 59, moderated mediation tested whether brain components mediate environment–symptom relationships and whether gene scores moderate this mediation in 8,705 UKB-NI participants. Nonparametric bootstrapping (5,000) estimated indirect effects and 95% CIs; explained mediation effects (EME) were reported. Confounders were controlled.

• Three distinct urban environmental profiles correlated with specific psychiatric symptom groups and were replicated in an independent sample.

Affective symptom group (unenthusiasm, tiredness, loneliness, depressed mood, feeling fed-up):

• sCCA correlations: Training r = 0.20 (Pperm < 0.001; EV = 4.09%); Test r = 0.22 (Pperm < 0.001; PFDR < 0.001; EV = 4.71%). Replication in UKB-NI: Train r = 0.17 (Pperm < 0.001); Test r = 0.10 (Pperm < 0.001; PFDR < 0.001).
• Environmental correlates: Positive with IMD (deprivation), air and sound pollution, SN accessibility (radial/centrality), traffic, and dense land-use (factories, retail, offices, community). Negative with distance to facilities (services, factories, emergency, education, food stores, community, healthcare) and green space proximity (domestic gardens, natural environment, green space).
• Brain mediation (msCCA in UKB-NI): 13 regional volumes associated; the environmental profile negatively correlated with volumes and positively with symptoms. Regions included left amygdala, right ventral striatum, right frontal pole, right occipital fusiform gyrus, bilateral superior frontal cortex, cerebellar lobules VIIa/VIIb, and right Crus I/II—implicating reward processing circuits.
• GWAS: 3,436 significant SNPs across 22 protein-coding genes (Bonferroni). Lead SNP rs62062288 (MAPT, chr17q21.3; P = 6.09×10^-15). Additional loci include CRHR1, ARL17B, KANSL1, WNT3 (17q21.3), DCC and TCF4 (18q21.2), DCAF5/EXD2/GALNT16 (14q24.1), and STAG1/PPP2R3A/MSL2/PCCB (3q22.3). GSEA: enrichment for CRH/CRF receptor activity (Q = 5.23×10^-4), cellular response to CRH stimulus (Q = 0.02), axonal growth cone (Q = 0.002). Replication: 2,034 SNPs and 14/22 gene scores replicated in UKB-NI (for example, CRHR1, MAPT, DCC, TCF4, KANSL1, GALNT16, NRXN1, NSF).
• Moderated mediation: Gene scores CRHR1 (EME = 2.01%), MAPT (1.72%), TCF4 (1.71%), DCC (1.51%) significantly moderated the environment → brain volume → affective symptoms pathway. Higher genetic risk strengthened the indirect effect; for CRHR1, higher risk plus greater environmental exposure related to lower brain volume and more severe affective symptoms (β = 0.02, s.e. = 0.009, 95% CI 0.006–0.04).
• Pleiotropy (FEV): First environmental profile explained 68.43% of EV in affective group; second explained 29.34%; third 2.22%.

Anxiety symptom group (anxious feelings, seeing a psychiatrist, feeling tense, suffering from nerves, nervous feelings, worrying too long):

• sCCA correlations: Training r = 0.11 (Pperm < 0.001); Test r = 0.10 (Pperm < 0.001; PFDR < 0.001; EV = 1.03%). Replication in UKB-NI: Train r = 0.11 (Pperm < 0.001); Test r = 0.03 (Pperm < 0.001; PFDR < 0.001).
• Environmental correlates: Positive with dense urban build-up (leisure density, SN detour/shape), mean terrain, coastal proximity, NDVI variation, and mixed-use density (residential, transport, utility, animal center, storage, agriculture). Negative with mean NDVI (greenness), distance to waste/energy, and water proximity—indicating greenness and access to nature as protective.
• Brain mediation: 11 regional volumes associated, including left inferior frontal gyrus, left supplementary motor area, right amygdala, bilateral cerebellar lobules VIIa/VIIIb, bilateral Crus I, right lobule V, left lobule VI—implicating emotion regulation circuits.
• GWAS: 29 significant SNPs across 9 genes. Lead SNP rs77641763 in EXD3 (chr9; P = 9.53×10^-11). Other genes include CNNM2, GBF1, NOLC1, NT5C2, TRIM. Enrichment for small nucleolar RNP complex binding and serotonin metabolic processes. Replication: 18 SNPs and 6/9 gene scores (CNNM2, EXD3, GBF1, NT5C2, NOLC1, TRIM).
• Moderated mediation: EXD3 gene score moderated the pathway (EME = 1.65%), with higher EXD3 risk strengthening the indirect effect.
• Pleiotropy (FEV): Second environmental profile explained 64.24% of EV; first 25.62%; third 10.12%.

Emotional instability symptom group (highly strung, feeling miserable, mood swings, neuroticism, risk-taking, irritability and sensitivity, hurt feelings, grief and stress):

• sCCA correlations: Training r = 0.05 (Pperm < 0.001); Test r = 0.03 (Pperm < 0.001; PFDR < 0.001). Replication in UKB-NI: Train r = 0.10 (Pperm < 0.001); Test r = 0.02 (Pperm = 0.004; PFDR = 0.027).
• Environmental correlates: Positive with density of education facilities, terrain variability, building classes (high-rise flats, terraced houses), SN link characteristics, density of accommodation, medical and emergency facilities. Negative with density of unused land, water, open space, amenities, parks, allotments, information stations; and with distance to food stores.
• Brain mediation: 13 regional volumes including bilateral frontal pole, amygdala, precentral gyrus, insula, and left lateral occipital cortex—implicating emotion response networks.
• GWAS: 10 significant SNPs; lead SNP rs77786116 in IFT74 (chr9; P = 4.16×10^-10). Other genes: LDHC, SLC9A7P1, TMPO; enrichment for cerebellar development processes. Replication: 3 SNPs and 3/6 gene scores (IFT74, LDHC, TMPO).
• Moderated mediation: IFT74 gene score moderated the pathway (EME = 1.52%).

Additional results:

• Replication of multivariate sCCA relationships in UKB-NI mirrored discovery findings across all three symptom groups.
• Sensitivity analyses (bootstrapping, resampling, sex-stratified, and excluding potential household-sharing) supported robustness.
• Pleiotropy patterns indicated strongest specificity for the affective and anxiety profiles; the third profile explained less covariance (typical for successive CCA modes).

The study demonstrates that complex, real-world urban environmental profiles are differentially associated with specific psychiatric symptom groups and that these associations are mediated by distinct brain structural substrates and moderated by genetic variation. The first profile—characterized by deprivation, pollution, high traffic and limited green space—aligns with increased affective symptoms and involves brain regions central to reward processing (ventral striatum, amygdala, frontal and cerebellar areas). Genetic moderators enriched for stress-response and neurodevelopmental pathways (CRHR1, MAPT, TCF4, DCC) strengthen the brain-mediated effects in those with higher genetic risk, supporting a stress-related mechanism linking adverse urban exposures to affective symptoms. The second profile highlights greenness, water proximity, and distance from waste/energy as protective features against anxiety symptoms, with brain correlates in prefrontal–limbic–cerebellar circuits implicated in emotion regulation; EXD3 variation moderates this pathway. The third profile shows weaker but significant associations with emotional instability, engaging frontal, insular, and cerebellar regions and moderated by IFT74 linked to neuronal migration. Together, these findings provide an integrative framework connecting macroenvironmental features with transdiagnostic symptom dimensions via neurobiological pathways, offering targets for urban planning and neurobehavioral interventions. The work advances beyond single-factor studies by quantifying each factor’s contribution within comprehensive environmental profiles and by linking environment, brain, and genetics in a unified model.

This study identifies three distinct urban-living environmental profiles that differentially relate to affective, anxiety, and emotional-instability symptom groups, each mediated by specific brain volume patterns and moderated by biologically plausible genetic pathways. By moving from isolated environmental measures to comprehensive profiles, the approach explains more variance in symptoms and reveals mechanistic insights. The findings suggest actionable targets for public health and urban planning—such as increasing greenness and access to nature—and inform individualized prevention strategies guided by genetic susceptibility and neural substrates. Future research should pursue longitudinal and mechanistic studies to establish causality, identify biomarkers of risk and resilience, and test interventions (for example, neurofeedback-guided approaches). Broader validation across diverse ethnicities and non–high-income settings is also needed to assess generalizability.

• Potential attrition bias and sample selection differences across analytical subsets of the UK Biobank, though most demographic distributions were similar.
• Cross-sectional mediation precludes causal inference; observed associations may reflect alternative explanations (for example, selective migration to deprived urban areas, potentially with genetic influences; unmeasured familial or contextual confounders).
• The third environmental profile explained a smaller proportion of covariance after orthogonalization, limiting interpretability.
• Generalizability to other ethnicities and to low- and middle-income or non-industrialized settings remains uncertain.
• Biological pathways mediating specific environmental adversities were not directly characterized; deeply phenotyped longitudinal datasets are required for causal and mechanistic validation.