Answers are grounded in this paper and cite their sources.
Purpose: Integrative analyses of public transcriptomic datasets are widely used to nominate hub genes and therapeutic targets in intervertebral disc degeneration (IVDD), but whether such candidates replicate across datasets is rarely tested. We assessed the reproducibility of gene- and pathway-level signatures of IVDD using a discovery–validation design. Methods:Six Gene Expression Omnibus (GEO) series used in an overlap-and-network pipeline (four degeneration-versus-control comparisons and two cell-type or intervention contrasts) were re-analysed dataset by dataset (limma or DESeq2, with design adjustment where needed). Thirty-three candidate genes from that pipeline were audited, and a weighted Stouffer meta-analysis of the four discovery datasets was tested for replication in three independent validation cohorts (GSE70362, 48 samples from 18 donors; the non-overlapping samples of GSE23130; GSE34095). We also quantified genome-wide agreement between datasets, checked sample uniqueness, performed pre-ranked gene set enrichment analysis (Hallmark and Reactome) and analysed a combined protein–protein interaction network. Results:Samples 1–15 of GSE23130 were the specimens of GSE15227, and 23 PubMed Central articles cite both accessions. Genome-wide agreement between degeneration datasets was negligible (Spearman ρ −0.13 to 0.13). Of 1252 genes significant in the discovery meta-analysis, 49% changed in the same direction in validation (binomial p = 0.78); 15 replicated and two remained robust to leave-one-dataset-out analysis. None of the 33 candidates replicated robustly; ACAN, COL9A3 and COL11A2 changed significantly in the opposite direction in validation, and the apparent downregulation of MMP3 and MMP9 arose only from non-degeneration contrasts. A reproducible overlap step selected 14 mainly interferon-related genes driven by these contrasts. At the pathway level, E2F target and G2-M checkpoint (proliferation) gene sets were enriched in both meta-analyses and had positive enrichment scores in all nine datasets, whereas inflammatory and TNF-α/NF-κB signatures did not replicate. Conclusions:Gene-level IVDD signatures derived from small public datasets are not reproducible, and overlap-based hub genes should not be presented as therapeutic targets without independent validation. Within the datasets examined, a cell proliferation programme was the most consistent pathway-level signal; whether it is a feature of disc degeneration requires experimental validation. We provide a checklist for future secondary analyses of IVDD transcriptomes.
No takes yet. Share an insight, caveat, or question.
Muthusami et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: