Integrative Mendelian randomization, single-cell, and bulk transcriptomic analyses reveal the landscape of druggable genome- and gut microbiome-associated biomarkers in osteoarthritis

Technical Summary: Integrative Mendelian Randomization, Single-Cell, and Bulk Transcriptomic Analyses in Osteoarthritis

Problem Statement
Osteoarthritis (OA) is a prevalent degenerative joint disease characterized by cartilage degradation and low-grade synovitis. Current clinical interventions primarily alleviate symptoms without halting disease progression, largely due to an insufficient understanding of the molecular mechanisms driving OA, particularly the regulatory interplay between the druggable genome (DG) and the gut microbiome (GM). While both DG genes and GM are individually linked to OA, their synergistic crosstalk mechanisms and specific biomarkers remain unclarified. This study aims to identify OA-related biomarkers linking the DG and GM, explore their underlying mechanisms, and define key cell types involved in this regulatory axis.

Methodology
The study employed a multi-omics integrative approach combining Mendelian randomization (MR), bulk transcriptomics, machine learning, and single-cell RNA sequencing (scRNA-seq).

1. Data Acquisition: Bulk transcriptomic data (GSE51588 and GSE114007) were obtained from the GEO database. OA-related druggable genome genes (DRGs) were sourced from literature, and OA genetic variation data were retrieved from the IEU OpenGWAS platform. Gut microbiome (GM) data were sourced from the MiBioGen database.
2. Mendelian Randomization (MR): Two separate MR analyses were conducted to establish causal relationships:
   • DRG-OA: Using eQTLs of DRGs as instrumental variables (IVs) to assess causal effects on OA.
   • GM-OA: Using eQTLs of GM taxa as IVs.
   • Analyses utilized the TwoSampleMR package, employing Inverse Variance Weighted (IVW) as the primary method, supplemented by MR Egger, weighted median, and sensitivity analyses (heterogeneity, horizontal pleiotropy, leave-one-out, and Steiger directionality tests).
3. Candidate Gene Identification:
   • Metabolite-Target Prediction: GM-related metabolites were retrieved from the gutMGene database. Target genes (GRGs) were predicted using SEA and STP platforms.
   • Intersection: Differentially expressed genes (DEGs) from bulk RNA-seq were intersected with OA-related DRGs and GRGs to identify candidate genes.
4. Biomarker Screening: Machine learning algorithms (XGBoost and Random Forest) were applied to rank candidate genes. Overlapping genes were validated across datasets using Wilcoxon tests and Receiver Operating Characteristic (ROC) curve analysis. A Back Propagation (BP) neural network was constructed to evaluate predictive performance.
5. Functional and Mechanistic Analysis:
   • Enrichment: Gene Set Enrichment Analysis (GSEA), Gene Ontology (GO), and KEGG pathway analyses were performed.
   • Molecular Interaction: Molecular docking and molecular dynamics (MD) simulations (using GROMACS) were conducted to assess the stability of interactions between core metabolites and biomarker-encoded proteins.
   • Immune and Subcellular Analysis: Immune cell infiltration was assessed via GSVA, and subcellular localization was predicted.
6. Single-Cell Analysis: scRNA-seq data were processed using Seurat for quality control, clustering, and annotation. Pseudotime trajectory analysis (Monocle), cell-cell communication (CellChat), and metabolic profiling (scMetabolism) were performed on identified key cell populations.

Key Results

• Causal Factors: MR analysis identified eight OA-related DRGs and four OA-related GMs. Notably, LDHA and GRIK1 were identified as protective factors (OR < 1), while Bifidobacterium (genus level) was identified as a risk factor.
• Biomarkers: Through intersection analysis and machine learning, two robust biomarkers were confirmed: LDHA (downregulated in OA) and GRIK1 (upregulated in OA). Both exhibited high diagnostic accuracy (AUC > 0.7) in independent datasets, with a combined BP neural network model achieving AUCs of 0.963 and 0.880.
• Metabolite-Biomarker Axis: The risk-associated Bifidobacterium was linked to 22 metabolites. Molecular docking and MD simulations revealed stable binding between core metabolites (e.g., folic acid, dihydrocaffeic acid) and the biomarkers. Specifically, folic acid showed strong binding to LDHA, and dihydrocaffeic acid to GRIK1.
• Functional Enrichment: Both biomarkers were significantly enriched in the unfolded protein response (UPR) pathway. LDHA expression correlated with activated dendritic cells and central memory CD4+ T cells.
• Key Cell Type: scRNA-seq analysis identified endothelial cells as the pivotal cell population expressing high levels of LDHA. These cells were further subdivided into four subtypes (Tip-like, Stalk-like, Lymphatic, and Tumor ECs). Strong intercellular communication was detected between endothelial cells and myeloid cells, particularly in OA samples.
• Regulatory Network: A comprehensive axis was constructed: Bifidobacterium (GM) → Metabolites (Folic acid/Dihydrocaffeic acid) → Biomarkers (LDHA/GRIK1) → Metabolic Pathways (Folate one-carbon cycle/Cysteine-methionine metabolism) → Endothelial cells.

Significance and Claims
The study claims to provide a systematic elucidation of the DG-GM-OA regulatory network. By integrating MR with multi-omics, the authors define LDHA and GRIK1 as novel protective biomarkers for OA diagnosis. The research highlights endothelial cells as critical stromal regulators in OA pathogenesis, bridging systemic metabolic signals (via gut metabolites) and local joint inflammation.

The authors posit that the identified biomarkers and the gut-derived metabolites (specifically dihydrocaffeic acid and folic acid) offer potential targets for microbial metabolome-based interventions and precise therapeutic strategies. The study emphasizes that these findings update the understanding of the gut-joint axis in OA, though it acknowledges limitations regarding the need for multi-ethnic validation and experimental verification (in vitro/in vivo) of the identified mechanisms.