Regional Adiposity Shapes Brain and Cognition in Adults  

This repository contains the code used for the major analyses in MATLAB (R2017b; The MathWorks, Inc., USA), and R (version 4.3.1).  

Overview

The study investigated the associations between regional adiposity measures and brain morphology, function connectivity, white matter microstructure and cognition performance, using: 

1. Linear regression models
2. Mediation analysis  

Repository Structure

├── scripts/
│   ├── 01_two_step_cortical_demo.m      
│   ├── 02_mediation_analysis.R    
├── data/  
│   └── example_data.csv
└── README.md      

⚠️ The data in example_data.csv is simulated for illustration only.   It does not reflect real distributions, and the number/type of covariates is simplified.   For full methodological details, please refer to the published paper. 

Analysis Pipeline 

1. Data Preparation 

• Import and clean raw dataset (example_data.csv).
• Select targeted regional and general adiposity measures: arm fat, leg fat, trunk fat, VAT, and BMI. 
• Merge demographic and lifestyle covariates: age, sex, education, employment, smoking, alcohol, physical activity, metabolic health, etc. 
• Precompute cortical metrics (thickness, volume, area) using FreeSurfer; load them using SurfStat.
• Similar statistical models were applied to other modalities (e.g., subcortical volumes, functional connectivity, NODDI metrics VBM analysis using SPM12 ).   
• Below we illustrate the linear regression analysis using cortical morphology as an example.   

---

2. Two-Step Regression Analysis    

For each brain metric (e.g., thickness, volume, or other modality) and each adiposity measure:   

Step 1: Eliminate BMI masking effect   

We first remove the influence of overall obesity (BMI) from the brain metric, producing residual values: 

Model 1:    
 1.Brain_Metric = β₀ + β₁ * BMI + ε    
 2.Get residuals (BMI-adjusted brain metric): Residuals = Brain_Metric - predicted(Brain_Metric)  

Step 2: Test independent effect of regional adiposity  

Model 2:    
Residuals = γ₀ + γ₁ * Regional_Adiposity    
                    + γ₂ * Age  
                    + γ₃ * Sex  
                    + γ₄ * Education
                    + γ₅ * Smoking
                    + ... + η    

---

Standardized Beta Coefficient   

To make effect sizes comparable across regions and predictors,  we computed standardized beta values from the raw regression coefficients:   

# Standardized β (for regional adiposity predictor)
beta_standardized = (beta_raw * std(X)) / std(Y)

# In this context:
#     X = Regional_Adiposity
#     Y = Residuals from Step 1 (i.e., brain metric adjusted for BMI)

---

3. Mediation Analysis    

We tested whether the effect of regional adiposity on cognitive performance is mediated by system-level brain age gap (BAG), which was predicted by PLS. 

For each:

• Cognitive task (e.g., Symbol, Fluid Intelligence) 
• Regional adiposity measure (e.g., trunk fat, VAT) 
• Brain system (e.g., DMN, SMN)

We fit the following model: 

Step 1 (Path a):
    Mediator (BAG) = α₀ + α₁ × Adiposity + α₂ × Covariates + ε₁  

Step 2 (Paths b & c′):
    Cognition = γ₀ + γ₁ × Adiposity + γ₂ × Mediator + γ₃ × Covariates + ε₂

All variables are z-scored prior to modeling, except binary covariates (e.g., sex, employment).