Integrative Meta-Analysis of Transcriptomic Networks Reveals Core Signatures and Master Regulators of Cellular Senescence

Cellular senescence appears as a stable, self-maintaining program rather than a transient stress reflex. To assess how this pattern persists across contexts, seven public human fibroblast RNA-seq datasets were assembled, encompassing both replicative and acute models. Raw count matrices from each dataset were batch-corrected with Surrogate Variable Analysis (SVA) to remove latent technical effects before differential expression modeling. DESeq2 was used to estimate differential expression with Wald statistics and Benjamini–Hochberg correction. Two complementary analytical routes tested the stability and depth of the signal. In the first, each dataset was modeled separately, and resulting effects were combined through random-effects meta-analysis with restricted maximum likelihood (REML) in metafor. It yields pooled log₂ fold changes and gene-level heterogeneity estimates. In the second, all count matrices were merged into a single batch-corrected dataset. DESeq2 contrasts were run for replicative and acute senescence within the same design. Only genes significant in both contrasts and moving in the same direction were retained as the consensus set. Primary inference relies on this unified, SVA-corrected analysis, while the meta-analysis provides an independent cross-study validation. Working in one corrected space standardizes size factors, dispersion trends, and surrogate variables, reducing nuisance variation and sharpening the common signal. Both routes converge on the same architecture, a nuclear lock suppresses DNA replication and mitosis, with LMNB1, MCM2–7, CDK1, PLK1, and EZH2 as consistent regulators. An inflammatory amplifier remains active, centered on STAT3 and RELA with cytokine outputs. While TP53 and CDKN1A provide the bridge that ties amplifier activity to arrest. Enrichment tests and the network map point to STAT3 and RELA at the crossroads where cytokine signaling meets chromatin control and checkpoint cues. TP53 sits as the hinge between the two arms. Across studies the spread is modest (I² ≈ 21.9 percent; τ² ≈ 0.013). Overlap between models is strong. The same themes keep showing up: DNA damage response, cell-cycle arrest, and the senescence-associated secretory phenotype. The interactome view makes the case stronger, since the shared genes lie on high-traffic routes and often bridge the modules that govern arrest and cytokine output. Taken together, these results outline a conserved dual-circuit system that locks the nucleus while sustaining inflammatory signaling. The balance between the two modules captures the logic of senescence and helps explain how molecular stability, together with persistent cytokine output, maintains an irreversible fate. The approach also explains why the consensus holds: a single SVA-corrected matrix for both contrasts, a strict intersection rule, and a network view that confirms central positioning of the shared hubs. The resulting map offers a practical basis for prioritizing targets and shaping future senotherapeutic strategies.