Although validated worldwide, purely clinico-epidemiological scores (usually based on age, BMI, ethnicity, education, history of COPD, smoking habits and the personal or family history of cancer) are insufficient to predict the transition from (apparent) health to cancer diagnosis at the individual level in our PREVALUNG cohort. Thus, 30% of the CVD patients enrolled in the PREVALUNG cohort who eventually developed cancer were not included in the at-risk category, based on the purely clinical NLST, NELSON and PLCOm2012 scores. Hence, risk assessment to advise LDCT LC screening programs remains imprecise. Here, we propose functional biomarkers to identify the health-to-cancer transition, based on high-dimensional unbiased multi-omics technologies.
Machine learning guided dimension reduction led to a predictive algorithm based on blood borne soluble factors measured in a first retrospective cohort (FLEMENGHO) that we applied and validated in external cohort, one with a similar risk epidemiology (ACVD and tobacco for PREVALUNG). The risk score was computed in CVD tobacco users from 27 soluble factors that were validated in two other cohorts and a prospective blinded FU of controls with high-risk scores at study entry. Interestingly, this 27-plex fingerprint was also able to detect history of cancer, related or not to tobacco consumption. Finally, this 27-plex fingerprint allowed to categorize patients into three major functional clusters (relying on cholesterol metabolism and CHIP, gut barrier permeability/IL-6/TH2, and maladaptive immunity with cumulative immune inhibitory checkpoints and CHIP), amenable to specific interceptive measures.
CHIP mutation analysis revealed the presence of multiple CHIP mutations in some tobacco users with CVD that developed LC. This accumulation of CHIP variants correlated with elevated IL-1β levels. Fueling this notion, high-dimensional spectral blood flow cytometry-based immune profiling revealed an increase of specific subsets of inflammatory monocytes in tobacco users with CVD who developed LC. These data suggest a dysregulated epigenetic control of clonal myeloid progenitors leading to systemic IL-1β-associated inflammation and LC, in accord with recent reports. Individuals combining several CHIP variants or exhibiting high variant allelic frequencies (VAFs) might benefit from recombinant IL-1R antagonist or colchicine. Indeed, IL-1R antagonist may be particularly useful for patients with cancer-associated CHIP variants, which are causatively linked to increase circulating IL-1β.
Several limitations affect our study. The value of the PREVALUNG prospective study is limited by the low number of cases. Moreover, the absence of longitudinal sampling precludes the elucidation of the temporal order of biological deviations affecting each subject. Despite the strength of machine learning-based algorithms used in this study, and its blinded confirmation in the prolonged follow up of the CVD controls, further prospective validation is needed to strengthen our conclusions. Regardless of such shortcomings, distinct biomarkers identified in this work have already been unveiled in other reports as candidate biomarkers of early diagnosis of lung cancer or other malignancies. Pending further validation, the new biological signatures we found may permit a comprehensive and personalized screening program and prepare the grounds for adapted cancer interception.
