Abstract
Circulating proteins can reveal key pathways to cancer and identify therapeutic targets for cancer prevention. We investigate 2,074 circulating proteins and risk of nine common cancers (bladder, breast, endometrium, head and neck, lung, ovary, pancreas, kidney, and malignant non-melanoma) using cis protein Mendelian randomisation and colocalization. We conduct additional analyses to identify adverse side-effects of altering risk proteins and map cancer risk proteins to drug targets. Here we find 40 proteins associated with common cancers, such as PLAUR and risk of breast cancer [odds ratio per standard deviation increment: 2.27, 1.88-2.74], and with high-mortality cancers, such as CTRB1 and pancreatic cancer [0.79, 0.73-0.85]. We also identify potential adverse effects of protein-altering interventions to reduce cancer risk, such as hypertension. Additionally, we report 18 proteins associated with cancer risk that map to existing drugs and 15 that are not currently under clinical investigation. In sum, we identify protein-cancer links that improve our understanding of cancer aetiology. We also demonstrate that the wider consequence of any protein-altering intervention on well-being and morbidity is required to interpret any utility of proteins as potential future targets for therapeutic prevention.
Introduction
Proteins govern cellular action in all human biological processes and are crucial for our defences against both the onset and progression of cancer. Identifying circulating proteins important to the aetiology of cancer may improve our understanding of pathways leading to cancer and highlight potential targets for therapeutic prevention. Circulating proteins are valuable candidate targets for drug development since drug-target engagement can be evaluated in the bloodstream during randomised control trials (RCTs), accelerating target development. Additionally, identifying circulating cancer risk proteins allows for the subsequent selection of future RCT participants with risk-inducing protein concentrations, which may improve RCT effectiveness. Developing therapeutic prevention strategies, either alone or as a complement to existing prevention programs, such as smoking cessation, are urgently needed given cancer burden is projected to double by the year 20401.
Therapeutic prevention is an effective and commonly used strategy for the primary prevention of some common chronic diseases. Prevention strategies have thus far been most successful for cardiovascular disease using statins that target the HMG-CoA reductase protein as a first-line treatment to lower low-density lipoprotein (LDL) cholesterol2,3. In contrast, efforts to identify targets for the therapeutic prevention of cancer have been less fruitful, hampered by a more complex aetiology and difficulty identifying potential targetable aetiological biomarkers4. Exceptions include therapeutically targeting the oestrogen receptor (ER) to prevent breast cancer5 and COX2 to prevent colorectal cancer in high-risk individuals6. Additional aetiological proteins for specific cancers have been identified, such as the role of higher levels of insulin-like growth factor-I in the development of breast7, colorectal8, and prostate9,10 cancers, and of higher microseminoprotein-beta with lower prostate cancer risk11. Together these examples highlight the opportunity that may result if aetiological proteins for cancer are identified and the feasibility of using these to develop therapeutic prevention tools where high-risk populations are well-defined.
Identifying candidate aetiological biomarkers for cancer risk has traditionally involved analysing specific hypothesis-driven markers for single cancer outcomes in pre-diagnostic blood samples and comparable controls from large prospective cohorts9,12,13. The advent of high-throughput platforms that can measure hundreds to thousands of biomarkers simultaneously using small sample volumes has enabled hypothesis-free discovery analyses, but costs remain prohibitively high. An alternative cost-effective approach, that also limits bias by confounding and reverse causation, is to use robust genetic proxies of blood biomarkers to evaluate their aetiological relevance along the lines of Mendelian randomisation (MR)14,15. Using such MR-based approaches facilitates simultaneously querying thousands of markers in relation to the risk of multiple cancers using genome-wide association data, which can identify risk markers and assess their association with one or multiple cancers. Proteins represent a particularly appealing application of MR as the blood concentrations of many proteins are regulated by genetic variants, many of which lie in or near a protein’s cognate gene (variants known as cis protein quantitative trait loci [cis-pQTL])16. Cis-pQTL likely influence biological processes directly, such as by transcription or translation, making them less prone to common sources of bias in MR studies like horizontal pleiotropy17. It is also possible to complement cis-pQTL-based MR analyses with colocalisation analyses to further exclude confounding by linkage disequilibrium18. These methodologies allow for the in-silico simultaneous evaluation of the role of thousands of proteins in the aetiology of common cancers with high specificity.
In the current study, we estimated the associations of 2074 circulating proteins with the risk of nine common cancers using data from a total of 337,822 cancer cases. We aimed to identify cancer-risk proteins and assess whether these proteins may cause multiple or specific cancers. Where possible, we mapped risk proteins to potential therapeutic interventions and used MR and colocalisation phenome-wide association analyses (PHEWAS) to describe the promise and complexities that may result from intervening on risk proteins in terms of potential adverse outcomes.
Results
Protein effects on cancer risk
In total, 4507 of the 4698 cis-pQTL were available for analysis with at least one cancer site [min: 3308 cis-pQTL for bladder cancer and max: 4303 cis-pQTL for endometrial cancer], which represented 2023 of 2074 proteins with cis-pQTL included in our study [min: 1692 for bladder cancer and max: 1934 proteins for skin cancer] (Fig. 1., Supplementary data 2). MR and colocalization analyses identified 40 proteins (Supplementary data 3, Fig. 2.) with an association with at least one cancer site [min: one protein for ovarian cancer, max: 21 proteins for breast cancer]. A further 428 proteins were identified to have evidence of colocalization [PP4 > 0.7] and at least a nominally significant MR association with the risk of cancer [min: 8 proteins for skin cancer and max: 241 for breast cancer] (Fig. 1, Supplementary data 2). We observed limited evidence for the association of proteins with risk for clear cell ovarian cancer, ever smoking lung cancers, HER2 enriched, luminal B, and luminal B-HER2 negative breast cancers. Additionally, we did not identify any proteins as a risk factor for cancer from multiple, independent cis-pQTL in MR analyses. Results by cancer site are summarised below. We did not find supportive evidence after correction for multiple tests for the association of cancer risk with protein levels for any protein identified in our main risk analyses (Supplementary data 3)…
