A senescent-cell transcriptional program in intervertebral disc degeneration replicates across cohorts and platforms, whereas the degeneration-grade contrast does not

Cellular senescence is implicated in intervertebral disc (IVD) degeneration, and single-cell reversal-signature analysis is increasingly used to nominate repurposing candidates — usually on the untested assumption that a signature derived from one cohort generalizes. This computational study tests that assumption directly in two independent human nucleus-pulposus single-cell RNA-seq datasets generated on different platforms (GSE244889, 10x Genomics; GSE165722, BD Rhapsody). After isolating the disc-cell population and restricting to clean protein-coding genes, two differential-expression contrasts were compared and their cross-cohort agreement quantified by the Spearman rank-correlation of per-gene log-fold-changes over the shared gene universe. A degeneration-grade contrast (severe vs mild) did not replicate (ρ = +0.015; p = 0.25; 5,937 shared genes) and in fact reversed direction on the collagen/matrix axis, consistent with the non-monotonic nature of disc degeneration. A within-cell senescent-state contrast (senescence-high vs -low nucleus-pulposus cells) did replicate (ρ = +0.287; p = 1.9×10⁻¹²³; 6,508 shared genes), yielding a consensus program of canonical senescence and SASP genes (CDKN1A, TP53, GLB1, SERPINE1, IGFBP3/7, CCL2, CXCL2, MMP3, ADAMTS5) up and chondrogenic nucleus-pulposus identity genes (ACAN, COL2A1, VCAN, FRZB) down. Querying the consensus signature against LINCS L1000 returned a reversal shortlist dominated by NF-κB/IKK inhibitors and HSP90 inhibitors. The contribution is methodological — a within-cell state contrast is a more portable basis for cross-cohort single-cell signature analysis than a disease-stage contrast, and replication should be assessed by rank-correlation over shared genes rather than top-N gene-set overlap. This is not a therapeutic-novelty claim: senotherapeutic approaches to disc degeneration are already well established in the literature, and the work is released openly with no patent claim. All code (Colab notebooks), the consensus signature, and the candidate tables are provided for reproduction.