Reinterpreting Keratoconus Gene Expression in Light of Inherited Stromal Weakness

Purpose. To reinterpret keratoconus (KC) gene expression in light of inherited stromal weakness, and to define how expression should guide model and target selection. Methods. We combined genetic and chromatin data with expression from eight studies (182 donors). We compared the same KC samples with normal and pathological controls, built references matched to control source and tissue layer and tested them in independent tissues, and indexed published intervention-outcome records. Stimulated cells and genetic analyses assessed repair capacity and inherited risk. Results. Changing only the control reversed expression directions, flipping KC-like classification in 21-27 of 42 models and supplement-or-inhibit direction in 361-385 of 1,090 targets. Our references addressed this, predicting direction more accurately than published methods in independent spatial stroma and two bulk comparisons. Against these references, 41 of 46 target nominations required a different comparison, measurement, or use. A correct direction described only the stromal state: across control types, KC stroma showed lower inflammatory, immediate-early (stress-response), and transforming growth factor-{beta} (TGF-{beta}) programs, which signaled neither lost repair capacity (stimulated KC cells increased matrix production) nor disease cause (no genetic enrichment). Inherited risk converged on the matrix: collagen assembly was the only replicated pathway family, and genetically predicted higher corneal resistance lowered disease odds independently of thickness. Because these expression changes do not drive disease, reversing them does not ensure functional benefit: among 2,769 intervention-outcome records, the two were separate outcomes. Conclusions. Expression direction therefore reflects the choice of control and the stromal state, and does not indicate how to intervene. Because inherited risk lies in matrix structure and resistance to deformation, repair should restore organized, load-bearing matrix rather than reverse expression changes. Models should therefore be judged against matched references, and targets by measured functional outcomes.