A cross-tissue POSTN+ fibroblast atlas links periodontal, tumor, and fibrotic stromal niches

Cross-tissue single-cell atlases have re-framed fibroblasts as a continuum of activation states, with universal Pi16+ progenitors giving rise to tissue-restricted activated populations[1] and shared pathological states recurring across inflammatory diseases[2]. Periostin (POSTN), a matricellular protein of injured, fibrotic, and tumor stroma, has been independently linked to activated fibroblasts in liver fibrosis[3], colorectal cancer[4], head-and-neck cancer[5], and dental contexts[6, 7], but cross-tissue conservation of a single POSTN+ program is untested. Here we built a Harmony-integrated atlas of 56,713 human and mouse fibroblasts from eight single-cell datasets spanning six organ contexts (periodontal ligament, periodontitis, oral squamous-cell carcinoma, colorectal cancer, temporomandibular-joint osteoarthritis, and bile-duct-ligation liver fibrosis). A conservative cluster-consensus definition (Wilcoxon padj < 0.05 and log2FC > 0.5 within an atlas-integrated leiden cluster, combined with per-cell POSTN > 0) identified 11,451 POSTN+ cells (20.2% of the atlas) recurring across all six contexts at frequencies from 6.2% (periodontal ligament) to 55.1% (liver fibrosis). Within-fibroblast differential expression yielded a 102-gene shared core program -- collagen biosynthesis, ECM crosslinkers, and matricellular markers including POSTN, SPARC, BGN, FN1, MMP2, and CTHRC1 -- interpreted as POSTN-specific transcriptional amplification of an activated ECM-remodelling module. KLF4, hypothesized a priori as a POSTN+ co-marker, was upregulated in only one of six contexts, consistent with its role as a quiescence brake released during activation[3]. Three pre-registered sensitivity analyses (Harmony parameter, three definitions, dataset exclusion) and an independent Puram-2017 OSCC cohort (1,422 fibroblasts; 101/102 core genes recovered; primary vs lymph-node-met Mann-Whitney p = 0.005) support robustness across integration parameters, definitions, dataset inclusion, and platform.