Supplementary Methods: Multi-Biofluid Metabolomics Analysis for Alzheimer's Disease and Dementia with Lewy Bodies

To elucidate metabolic alterations associated with Alzheimer's disease (AD) and , we developed and implemented an R-based computational pipeline to analyze targeted metabolomics data from both serum and urine using the Biocrates AbsoluteIDQ® p400 platform. This integrated pipeline was designed to identify disease-specific metabolic signatures, assess biofluid-specific differences, and support biomarker discovery across clinical phenotypes.

1. Data Preprocessing and Quality Control

All raw data from the Biocrates platform were processed using standardized nomenclature consistent with the AbsoluteIDQ® p400 assay. Metabolite names were harmonized across serum and urine matrices to ensure cross-biofluid comparability. Preprocessing steps included:

• Log₂ transformation of metabolite concentrations to normalize distributions and stabilize variance.
• Normalization within biofluid to control for batch effects and inherent differences in concentration ranges.
• Missing value filtering, removing metabolites with >30% missingness per group, followed by imputation or case-wise deletion as appropriate.
• Metabolite intensities were subjected to scaling and centering, ensuring compatibility with multivariate methods.

This data preprocessing facilitated accurate downstream statistical modeling and minimized technical artifacts.

2. Statistical Analysis

Univariate Analysis

We performed an analysis of variance (ANOVA) for each metabolite across diagnostic groups (AD, DLB, and cognitively normal controls). Where significant effects were observed, Tukey's HSD post-hoc tests were applied to identify pairwise group differences.

To correct for multiple hypothesis testing, Benjamini-Hochberg false discovery rate (FDR) correction was employed. Effect sizes were quantified using eta-squared (η²), providing insight into the magnitude of group-level differences beyond p-values.

Multivariate Analysis

We implemented:

• Principal Component Analysis (PCA) to reduce dimensionality and visualize clustering based on metabolic profiles.
• Partial Least Squares Discriminant Analysis (PLS-DA) for supervised classification of diagnostic groups using metabolomic features.

To ensure model robustness, 5-fold cross-validation was employed, and variable importance in projection (VIP) scores were computed to identify key discriminative metabolites.

3. Metabolite Classification and Chemical Categorization

All detected metabolites were systematically classified into biologically relevant chemical classes, facilitating functional interpretation and pathway mapping. Classes included:

• Acylcarnitines
• Amino acids
• Biogenic amines
• Phospholipids (e.g., phosphatidylcholines, lysophosphatidylcholines, sphingomyelins)
• Lipids (e.g., triglycerides, diglycerides, cholesteryl esters)

This classification was used to assess group-wise trends and to support metabolite set enrichment analysis.

4. Visualizations and Figures

The pipeline generates a suite of visualizations including:

• Demographic plots showing sample characteristics (age, sex, APOE status)
• PCA and PLS-DA plots highlighting group separation
• Volcano plots displaying log₂ fold-change versus –log₁₀ p-values for all metabolites
• Heatmaps of top-ranked discriminative metabolites, stratified by group
• All visual outputs are provided in PDF and PNG formats suitable for inclusion in manuscripts or presentations.

5. Biomarker Discovery Framework

To support the identification of potential metabolic biomarkers for AD:

• Univariate significance (FDR-adjusted p-values), effect sizes, and VIP scores from multivariate models were integrated.
• Metabolites consistently ranked across univariate and multivariate methods were prioritized as candidate biomarkers.
• Cross-biofluid comparisons were conducted to validate biomarker consistency and assess systemic vs. biofluid-specific metabolic signatures.

6. Pathway and Metabolite Set Enrichment

We conducted metabolite class enrichment analysis to identify pathways disproportionately represented among significantly altered metabolites. This included:

• Comparison of enrichment patterns between serum and urine
• Identification of metabolic pathways related to energy metabolism, lipid regulation, amino acid turnover, and neurotransmitter synthesis
• Emphasis on pathways previously implicated in neurodegeneration, oxidative stress, and mitochondrial dysfunction

Pathway-level interpretation was supported by curated databases and literature-based annotations.

7. Output Structure and Export

All results are exported into a structured output directory (metabolomics_results/), including:

• CSV files for all statistical results (p-values, effect sizes, VIP scores)
• Visual figures in high-resolution formats
• Cross-validation metrics for model performance
• Summary reports detailing key findings per biofluid

Clinical and Translational Relevance

This pipeline enables comprehensive metabolic profiling of clinical cohorts and is particularly suited for early-stage biomarker discovery in neurodegenerative diseases. It supports:

• Identification of disease-specific metabolic fingerprints (AD vs. DLB vs. CN)
• Exploration of biofluid-specific alterations and inter-biofluid correlations
• Construction of multi-metabolite panels for classification and diagnosis
• Functional interpretation through pathway-centric analysis

By combining statistical modeling with interpretability and visualization, this pipeline provides a framework for metabolomics research in clinical neuroscience.