Falsification-aware triangulation of macrophage–stromal signaling in tendon repair prioritizes SPP1 across murine and human evidence layers
Description
Abstract Public single-cell data can nominate macrophage–stromal signals in tendon repair but may overstate temporal replication and causality. We asked which ligand candidates remained supported after replicated murine temporal testing, matched-null network perturbation, within-patient human analysis, and bounded external-human assessment. Methods We used a versioned plan frozen before candidate-restricted downstream testing. Four pooled murine tendon-injury single-cell libraries from GSE288443 (d0, d1, d7, and d28; one library per time point) were analyzed separately for state reconstruction and within-library CellChat–NicheNet prioritization. SPP1, TGFB1, OSM, and TNF were frozen for downstream assessment. Temporal evidence was evaluated in 26 independent GSE288444 bulk RNA-sequencing samples. Candidate-specific scTenifoldKnk analyses used five network seeds and empirical nulls of controls matched for expression, detectability, and network topology. Human evidence was assessed in 23 within-patient GSE26051 tendon pairs, including complete-pair exclusion sensitivity. Acute quadriceps and healthy Achilles datasets provided donor-aware contextual localization only. Final classes followed a non-weighted rule. Results Quality control retained 29,868 of 30,880 cells, resolving ten myeloid states and 14 mesenchymal states in seven receiver families. The myeloid system formed a resampling-stable but directionally unresolved chain; the full mesenchymal trajectory failed the prespecified concordance gate. Across 9,600 CellChat tasks, 390 canonical receiver-family ligand–receptor axes were assembled, and NicheNet retained 12 primary-core bundles in six reporting families. All four ligands showed replicated murine temporal effects, although receiver-program coherence varied. Perturbation profiles were stable across seeds, but none exceeded its candidate-specific matched null or achieved four-of-five-seed receiver-program enrichment. Within-patient results supported SPP1 most strongly and TGFB1 more moderately; OSM and TNF were opposite in direction and imprecise. External-human context did not change the classes: SPP1 Tier 1, TGFB1 Tier 2, OSM Tier 3, and TNF Tier 4. Conclusions SPP1 showed the strongest cross-layer directional coherence and is the leading candidate for experimental testing. TGFB1 remained context dependent, OSM exploratory, and TNF a falsification comparator. Negative matched-null results constrain claims of exceptional cell-intrinsic network centrality without excluding extracellular signaling hypotheses. Protein localization, receptor engagement, and replicated functional perturbation are required before causal or therapeutic conclusions.