Formation, not degradation: the matrix-deposition axis of fibrotic lung is molecularly decoupled from forced vital capacity

The problem. Every approved and late-phase therapy for idiopathic pulmonary fibrosis (IPF) — pirfenidone, nintedanib, nerandomilast — is approved on the forced vital capacity (FVC) slope, a rate of functional loss. A therapy that regresses scar and one that merely slows new scarring can trace the same slope, so the endpoint can in principle be blind to the remodeling a reversal therapy would act on. We asked, pre-registered, which molecular axes of matrix remodeling the FVC level actually reflects. Approach. In the public Lung Genomics Research Consortium cohort (GEO GSE47460; two Agilent platforms; 157 UIP/IPF lungs), we scored a-priori ECM signatures with gene lists fixed before analysis: matrix formation (collagens, cross-linkers, chaperones), degradation machinery (MMPs, TIMPs, cathepsins), and their balance (formation minus degradation, the transcriptomic analogue of the serum PRO-C3-versus-C3M neoepitope balance). We tested, within fibrotic disease, how each relates to %-predicted FVC — with the full pre-registered robustness battery, a genome-wide positive control, and a cell-composition adjustment — then validated the arm structure in two independent RNA-seq cohorts (GSE150910, n = 288; GSE213001, n = 139) and resolved the signatures across 312,929 single cells (GSE136831). The finding. Within UIP/IPF, the two arms diverge. Active matrix formation is decoupled from FVC (Spearman ρ = −0.15, 95% CI −0.30 to +0.01), whereas matrix degradation/turnover tracks it (ρ = −0.39, −0.52 to −0.25; arm difference +0.24, +0.07 to +0.42). The composite balance score is correspondingly near-zero-to-weakly-positive (ρ = +0.13 mean-z; +0.16 ssGSEA; +0.17 broad matrisome) — i.e., it does not fall as lung function is lost. This is not insensitivity: genome-wide, 8–23% of genes correlate with FVC (|ρ|>0.3; max 0.73), a validated IPF-severity signature tracks it (ρ = −0.37) and a proliferation signature tracks it (ρ = −0.22), while a housekeeping negative control does not (ρ = +0.03). The formation-decoupling was stable across scoring methods, gene sets, leave-one-gene-out, covariate adjustment, and recovery of the canonical genes (COL1A1/POSTN/SERPINH1) by raw-Agilent reprocessing. Cell-composition adjustment sharpened the mechanism: formation stayed decoupled at every level of adjustment, whereas the degradation–FVC link attenuated from −0.39 to ≈ −0.10 — it is substantially carried by composition, not cell-intrinsic. The two-arm dissociation replicated in both independent cohorts (formation and degradation strongly disease-associated but distinct axes, ρ = 0.72; the arms dissociating along an independent apex→base severity gradient), and single-cell resolution revealed why: the formation program is concentrated in myofibroblasts/fibroblasts (COL1A1 4,542 CPM) while the degradation program is distributed across macrophages, aberrant-basaloid epithelium (MMP7) and other lineages — physically different cell populations. In the serum literature, remodeling-turnover biomarkers predict IPF outcome in real-world cohorts (PRO-C3 HR 1.8; C6M HR 2.3; serial CRPM adjusted HR 2.27) but not in the preserved-FVC INMARK trial — a population-dependent, manufacturer-conflicted signal without proven FVC-independence. Why it matters. The FVC endpoint reads the degradation/loss axis, not the formation/deposition axis a reversal therapy is designed to move — and there exist lungs, at preserved FVC, that are molecularly depositing matrix as hard as the rest. This is a pre-registered, positive-control-defended, tissue-level demonstration of a structural limit of the FVC slope, and the case for developing and qualifying a reversal-remodeling readout in the right population. The result is cross-sectional: it establishes that the formation axis is independent of FVC level, not yet that a readout tracks the FVC slope over time.