Background
With the advent of high-throughput proteomic platforms, the identification of proteomic disease signatures has become increasingly attainable, as highlighted through recent publications from population cohorts around the world and resources such as the Human Protein Atlas. However, our understanding of the relationship between these proteomic markers and their associated diseases remains incomplete. While some markers may represent proteins directly involved in the disease process, others may instead reflect contributing risk factors and exposures such as smoking, obesity, stress and environmental factors. Repositories such as the Comparative Toxicogenomics Database (CTD) provide insight into the relationship between chemical exposures and proteins. The intersection of protein-cancer and protein-environment signatures has the potential to unveil putative causal pathways.

Objectives
Here, we aim to identify which protein-cancer associations may reflect exposure to known or suspected carcinogens, as classified by the IARC Monographs.

Methods
Study subjects were participants in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort.  In the current study, blood samples from a total of 10.118 individuals recruited in the UK, the Netherlands, Spain, and Italy underwent proteomic analysis by Somalogic using the SomaScan 7k Assay. The SomaScan 7k Assay uses 7596 aptamers to measure 6,432 unique proteins. We utilized a case-cohort study design consisting of 6,073 cancer cases and a randomly selected subcohort of 4,115  individuals. We estimated hazard ratios (HRs) and 95% confidence intervals (CI) for 24 cancer types using Prentice-weighted Cox regression models with age as the underlying time variable. The resulting FDR significant protein-cancer associations were then compared to protein-chemical associations obtained from the CTD using fisher exact tests. Only positive associations were included to preserve interpretability of the results. A Benjamin-Hochberg false discovery rate (FDR) control was applied for multiple testing (threshold for significance FDR-adjusted P < 0.1).

Results
We observed 36 chemicals with an FDR-significant association with a protein-cancer signature, with 24 of these associations being with hepatocellular carcinoma (HCC). Chemicals associated with HCC included: Tetrachlorodibenzodioxin, triclosan, aniline, perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), bisphenol A, F and S, diethylhexyl phthalate and arsenic.
We observed FDR-significant associations with lung cancer for 7 chemicals, including tobacco smoke pollution, Methylmercury Compounds, Nickel, chromium and Benzo(a)pyrene. Although the Cox regression models were adjusted for smoking, these results show there may be residual confounding. Squamous upper aerodigestive tract (UADT) cancer was associated with bisphenol A, bisphenol F and PFOS exposure. Non-Hodgkin lymphoma (NHL) was associated with nickel exposure and Esophageal squamous cell carcinoma was associated with aniline exposure.

Conclusions/Implications  
Our results highlight the potential of intersection of protein signatures as a hypothesis-generating approach to identify putative associations between carcinogens and cancer subtypes. Based on our results, intersection of protein signatures is particularly helpful for cancer subtypes that leave a notable imprint in the peripheral blood (HCC: 448 protein associations, Lung: 108 associations, UADT: 90 associations, NHL: 75 associations). More localized subtypes with a sparser imprint in the plasma are less likely to reach significance.