Fig. 5: Structural equation model.

A total of 18,244 participants were utilized in structural equation model analyses. Standardized coefficients are shown. Lifestyle was a significant predictor of depression (β = −0.157, P < 1.0 × 10−20), immunometabolic function (β = −0.043, P = 4.7 × 10−7) and brain structure (β = 0.038, P = 2.9 × 10−6). PRS (β = 0.036, P = 1.5 × 10−5), brain structure (β = −0.023, P = 1.2 × 10−2) and immunometabolic function (β = 0.020, P = 3.5 × 10−2) were also significant predictors of depression. PRS was also a significant predictor of lifestyle (β = −0.022, P = 3.5 × 10−3) and immunometabolic function (β = 0.018, P = 3.4 × 10−2). All paths represent significant associations except for the association between brain structure and PRS, and the association between brain structure and immunometabolic function. Latent variables including brain structure, depression and immunometabolic function were estimated in the model and are shown adjacent to the model panel, respectively. All significant paths could pass the FDR correction (α = 0.05). *Two-sided unadjusted P < 0.05; **two-sided unadjusted P < 0.01; ***two-sided unadjusted P < 0.001.