Senescence Geometry Persists Beyond Canonical Markers and Converges Across Morphology and Transcriptomics Evidence for a marker-robust, cross-modal state geometry in independent human fibroblast datasets

Cellular senescence is commonly identified using canonical markers, yet individual markers vary across cell types, induction paradigms, and temporal states. Here we tested a different hypothesis: whether a senescence-associated state geometry remains detectable after canonical senescence markers are deliberately removed, and whether independently measured cell morphology and transcriptomics converge on a similar progression geometry. We performed a secondary analysis of two public human fibroblast resources: the SenSCOUT single-cell morphology/biomarker dataset (Dryad DOI: 10.5061/dryad.rbnzs7hp8) and the GSE63577 replicative-senescence RNA-seq series. In SenSCOUT untreated controls, a morphology-only state coordinate constructed without donor-age labels or molecular biomarkers separated young and old donor populations (old-like fraction 18.3% versus 33.5%; enrichment 1.83; AUC 0.601) and was independently associated with higher SA-beta-gal fluorescence, p16, and p21. The continuous morphology coordinate also revealed a broad intermediate region with partial molecular senescence burden. In GSE63577, we then removed a broad panel of canonical senescence, cell-cycle, DNA-damage, and SASP genes before constructing a blind transcriptomic coordinate from 9,704 remaining genes. Despite this marker depletion, the coordinate remained strongly associated with normalized population-doubling progression (Spearman rho=0.800, P=6.8x10^-5; stratified permutation P=0.0006). Random removal of 50% of the remaining genes preserved a positive progression association in 100/100 resamples (median rho=0.800; empirical 95% interval 0.790-0.834), whereas several individual canonical markers were not directionally consistent across HFF and MRC-5. Finally, independently derived morphology and transcriptomic trajectories showed quantitative three-dimensional concordance: trajectory-shape rho=0.848, endpoint cosine=0.950, and Procrustes disparity=0.0506. The geometry correlation was significant against an exact 48-transformation axis null (P=0.0417), although an exact three-stage order test was intrinsically underpowered and non-significant. The distinctive result is therefore not merely senescence heterogeneity or continuum behavior, but the persistence of distributed senescence-associated organization after canonical-marker depletion together with quantitative convergence across independent morphology and transcriptomic modalities. We term this testable hypothesis marker-robust cross-modal senescence geometry. Paired single-cell multimodal and multi-donor validation will be required to determine whether this convergence reflects a shared biological manifold. NOVELTY STATEMENT. Rather than proposing another senescence marker or classifier, this study asks whether the higher-order state structure of senescence survives removal of canonical markers and reappears in an independent measurement modality. The principal novelty hypothesis is the joint observation of canonical-marker-depleted transcriptomic progression and quantitative 3D convergence with an independently derived morphology trajectory. Keywords: cellular senescence; morphology; transcriptomics; latent state; fibroblast; heterogeneity; manifold; biomarker; replicative senescence