The blood proteome of imminent lung cancer diagnosis

Abstract

Identification of risk biomarkers may enhance early detection of smoking-related lung cancer. We measured between 392 and 1,162 proteins in blood samples drawn at most three years before diagnosis in 731 smoking-matched case-control sets nested within six prospective cohorts from the US, Europe, Singapore, and Australia. We identify 36 proteins with independently reproducible associations with risk of imminent lung cancer diagnosis (all p < 4 × 10^-5). These include a few markers (e.g. CA-125/MUC-16 and CEACAM5/CEA) that have previously been reported in studies using pre-diagnostic blood samples for lung cancer. The 36 proteins include several growth factors (e.g. HGF, IGFBP-1, IGFP-2), tumor necrosis factor-receptors (e.g. TNFRSF6B, TNFRSF13B), and chemokines and cytokines (e.g. CXL17, GDF-15, SCF). The odds ratio per standard deviation range from 1.31 for IGFBP-1 (95% CI: 1.17-1.47) to 2.43 for CEACAM5 (95% CI: 2.04-2.89). We map the 36 proteins to the hallmarks of cancer and find that activation of invasion and metastasis, proliferative signaling, tumor-promoting inflammation, and angiogenesis are most frequently implicated.

Fig. 1. Identification of 36 protein biomarkers associated with risk of imminent lung cancer diagnosis among 731 cases and 731 matched controls in the Lung Cancer Cohort Consortium.

The volcano plot depicts the lung cancer odds ratio per standard deviation increment in relative protein concentrations (log-base-2 transformed) (x axis) and the −Log10 p value (y axis). The 36 identified markers of imminent lung cancer are labeled (see Methods; markers were identified through a resampling process that measured the association of each protein with lung cancer risk in a discovery set and a replication set. The risk markers were required to have a p < 0.05/effective-number-of-tests in the discovery set and p < 0.05 in the replication set in at least 50% of the resampling iterations). Source data are provided as a Source Data file.

Fig. 2. Lung cancer odds ratios for the 36 proteins associated with imminent lung cancer diagnosis before and after detailed adjustment for smoking intensity, duration, and years since cessation.

Data for 95% confidence intervals are presented as$e^{\left(\beta \pm 1.96\times sd\right)}$. β is the estimate from each conditional logistic regression, and sd is their respective standard deviation. Number of samples used are presented in Supplementary Data 10. Source data are provided as a Source Data file.

Fig. 3. Biological context of the 36 proteins associated with risk of imminent lung cancer diagnosis.

a Relationship between our 36 proteins and the 10 hallmarks of cancer described by Hanahan and Weinberg, based on their descriptions and functions available on GeneCards, the Human Protein Atlas, and Uniprot. Each hallmark is represented by a different color. b Network analysis among the 36 proteins, the figure depicts partial correlation networks (accounting for sex, age, cohort, and all other identified proteins) and stable protein associations. In lung cancer cases, no stable connections were found for ANGPT2, CDCP1, CEACAM5, CFHR5, CXCL13, IGFBP-1, IGFBP-2, IL6, MUC-16, SCF, SFTPA1, TFPI-2. In controls, no stable connections were found for ANGPT2, CEACAM5, CFHR5, CXCL13, CXCL9, IL6, MMP12, MUC-16, SCF, SFTPA1, and SYND1. Source data are provided as a Source Data file.

Fig. 4. Gene expression of 36 protein biomarkers associated with risk of imminent lung cancer diagnosis in normal and tumor tissue.

Proteins are listed in order of their relative expression in non-cancerous lung cell. a mRNA expression in normal tissue (gtex). b mRNA expression in tumor tissue (TCGA). Source data are provided as a Source Data file.