Early cancer interception rather than late-stage treatment represents medicine's most promising frontier, yet identifying who will develop lung cancer years in advance has remained elusive. A convergence of machine learning, multi-cohort proteomics, and single-cell transcriptomics now offers a concrete molecular framework for precisely that goal — one that could transform how prevention trials select participants and deploy anti-inflammatory therapies.
Researchers applied machine learning to plasma proteomics data and identified a 14-protein signature capable of predicting lung cancer more than five years before clinical diagnosis. Validated across eight independent cohorts, the signature was significantly elevated in active smokers and individuals with high particulate matter (PM) exposure. Single-cell analyses traced the signature's biological origins to lung myeloid and alveolar cell populations. In EGFR-driven adenocarcinoma, multiple epithelial lineages converged on a specific transitional cell state marked by keratin-8 and claudin-4 co-expression — termed KAC — whose transcriptional programs temporally aligned with signature appearance in blood. Critically, IL-1β inhibition suppressed PM-driven expansion of this KAC state and curtailed early tumorigenesis in experimental models. Re-analysis of the CANTOS trial — which showed canakinumab reduced lung cancer incidence — found that individuals with elevated signature scores at baseline derived greater benefit, substantially lowering the number needed to treat.
This work is significant on multiple levels. It bridges circulating proteomics with a specific tumor-promoting cellular mechanism, providing biological plausibility rather than pure statistical correlation. The eight-cohort validation is unusually robust for a biomarker discovery study, strengthening confidence in generalizability. That said, CANTOS was designed for cardiovascular endpoints, so the lung cancer findings remain secondary; a dedicated prevention trial using this 14-protein signature as an enrollment criterion is the logical and necessary next step. The KAC state also needs validation in non-EGFR-mutant tumors, which represent the majority of lung cancers in the broader population. Still, the combination of a validated blood test, an identified cellular mechanism, and an existing approved drug makes this finding genuinely actionable — incremental in some respects, but potentially paradigm-shifting for precision cancer prevention design.