July 9, 2026 | Researchers in London have identified a 14-protein blood signature predictive of lung cancer risk more than five years before a clinical diagnosis. If the proteomics panel makes its way to routine clinical practice, it could prove invaluable in flagging individuals likely to benefit from preventive screening, according to Tej Pandya, a clinical Ph.D. student at the Francis Crick Institute.
In Europe as well as the U.S., eligibility for screening “rests predominantly on age and smoking history, and even within that group the event rate is too low to select people efficiently for prevention trials, and those criteria miss light- and never-smokers entirely,” he says. In a recently published exploratory study, the new signature-based model outperformed the best-performing established risk tools (Cell, DOI: 10.1016/j.cell.2026.05.005).
The most immediate role for the protein signature would therefore be to “sharpen risk stratification refining who enters low-dose CT surveillance, and, critically, enriching prevention trials for the people most likely to benefit,” says Pandya. It is not a diagnostic and does not replace that preventive screening tool, he adds.
Since there are currently no preventive drugs on the market for lung cancer, another logical direction is to pair the predictive test with preventive treatments that are or soon will be in clinical trials, Pandya continues. “The value of predicting risk years ahead is only realized if there is something to offer those at risk.”
The focus of Pandya and his colleagues is to repurpose anti-IL-1β therapy as an interception strategy, but the broader premise is to pair a risk signal with an intervention window, he says, which extends to other investigational approaches. Anti-IL-1β drugs, including canakinumab (Novartis), target the core drivers of inflammation.
The Canakinumab Anti-inflammatory Thrombosis Outcome Study (CANTOS), a heart attack prevention trial of canakinumab that Novartis ran between 2011 to 2017, incidentally found a dramatic reduction in lung cancer incidence and cancer mortality. Applying the 14-protein blood signature to 4,651 individuals in the CANTOS cohort in the latest study to identify individuals who would clearly benefit from anti-IL-1β therapy succeeded in lowering the “number needed to treat” (NNT) threshold that has limited use of the drug as a preventive strategy.
Among people with a high baseline signature, lung cancer risk almost halved whereas among those with a low signature, there was essentially no effect. It was estimated that the NNT to prevent one incidence of lung cancer was 55 (versus 1,500 in the low-signature group), on par with “statin-style” cardiovascular prevention strategies, says Pandya.
“This is, for me, the most striking result,” he states. “The open question was always: who benefits? When we applied the signature retrospectively to the CANTOS biomarker sub-cohort, the answer was clear.”
The study explored blood plasma protein data from more than 48,000 UK Biobank participants, using matched cancer registry records to identify those who later developed lung cancer. Along with age, smoking status, and previous history of lung disease, a machine learning algorithm identified the 14 key proteins predictive of a diagnosis of lung cancer within five years. The protein signature was then validated in eight datasets from across the world, establishing that “the associations are robust and reproducible, which is the hard part and genuinely reassuring,” says Pandya.
Analysis of the signature in patients and animal models suggested that the signature reflects an altered inflammatory lung environment that precedes cancer. A higher signature was seen in people who later developed idiopathic pulmonary fibrosis or chronic obstructive pulmonary disease, both of which are progressive lung diseases characterized by inflammation.
“The final model integrates the 14 proteins with four patient characteristics—age, smoking status, pack-years, and history of COPD—in a gradient-boosted [XGBoost classification model] framework,” Pandya says. “What makes the biology compelling is that the same proteins are induced by the known promoters of lung cancer, such as particulate matter and smoking, and map back to alveolar and myeloid cells in the lung.”
The 14-protein signature was determined to be the optimal size for a blood biopsy test. “Using smaller panels does help with cost and assay complexity, but the real bottleneck is quantification,” says Pandya. “Our current assays give relative, not absolute, protein levels, which limits comparison across cohorts and means we cannot yet set fixed, transferable thresholds.”
Work on the proteomic signature, including potentially reducing the size of the panel, is underway. “Robustness is not the same as a finished test,” he notes. “Many things are clearly improvable—moving to absolute quantification so thresholds transfer; strengthening performance in never-smokers, where our signal is currently more modest because that group was under-represented in the discovery cohort; and prospective validation with serial sampling and a potentially smaller panel.”
Current scientific priorities follow directly from the limitations, he notes, pointing to the need for an absolute-quantification assay so the signature is reproducible across sites, has actionable thresholds, and considers the biologically distinct lung cancers occurring in people who have never smoked where “the unmet need is high.” Clinically, he adds, “the headline gap is that the canakinumab evidence is retrospective.”
Logical next steps include a prospective prevention trial that uses the proteomic signature to enroll high-risk individuals, “which is precisely the inefficiency—very high NNT in unselected people—that has held lung cancer prevention back, and what biomarker enrichment is designed to overcome,” says Pandya. “This is currently a research tool, not an approved clinical test,” he stresses, “and the route runs through assay standardization, prospective validation and regulatory work before any clinical deployment.”
The protein signature has relevance beyond lung cancer, given that it is also enriched ahead of idiopathic pulmonary fibrosis and COPD at a level “significantly above other diseases such as ischemic heart disease,” says Pandya. “That fits the underlying idea [that] these conditions share a pre-disease state of chronic alveolar inflammation, and the signature appears to read out that shared at-risk lung niche rather than any one tumor.”
Pandya draws attention to the analysis done on the TRACERx (tracking cancer evolution through therapy) cohort study, where the 14-protein signature neither rose with tumor stage nor fell after being surgically removed. This suggests it is “reporting the susceptible field, not material shed by an established cancer. This is a meaningful distinction from conventional liquid biopsy.”
The panel itself is lung focused. But this is “just the beginning” with the framework Pandya describes as “a population-scale proteomic risk signal anchored in tumor-promoting biology, paired with an interception window.”
In principle, that framework could be “re-tooled with organ-specific signals for precision prevention in other cancers,” he says. “The larger story is molecular cancer prevention as an approach,” of which the 14-protein signature is an early, concrete example.
“Prevention is becoming molecular,” Pandya emphasizes. “We are moving from ‘you're over a certain age and have a certain risk factor’ towards a blood-based readout that identifies risk years before a tumor is detectable and, importantly, identifies who would benefit from a preventive intervention.”
What made the latest paper particularly exciting, he adds, is that it involved integrating data from epidemiology, machine learning, and experimental biology to dissect proteins that increase in the plasma prior to lung cancer diagnosis. “This approach has a lot of benefits, and I think it would be great if more people could take similar approaches.”