Evaluation of circulating plasma proteins in breast cancer using Mendelian randomisation

Abstract

Biomarkers for early detection of breast cancer may complement population screening approaches to enable earlier and more precise treatment. The blood proteome is an important source for biomarker discovery but so far, few proteins have been identified with breast cancer risk. Here, we measure 2929 unique proteins in plasma from 598 women selected from the Karolinska Mammography Project to explore the association between protein levels, clinical characteristics, and gene variants, and to identify proteins with a causal role in breast cancer. We present 812 cis-acting protein quantitative trait loci for 737 proteins which are used as instruments in Mendelian randomisation analyses of breast cancer risk. Of those, we present five proteins (CD160, DNPH1, LAYN, LRRC37A2 and TLR1) that show a potential causal role in breast cancer risk with confirmatory results in independent cohorts. Our study suggests that these proteins should be further explored as biomarkers and potential drug targets in breast cancer.

Fig. 1. Flowchart of study design, data analyses and main results.

A total of 2929 unique proteins (2949 assays) were measured using the Olink PEA Explore assay in plasma samples taken from 598 women. Protein levels were correlated with baseline clinical characteristics using linear regression. Genetic association analysis of protein levels was performed which led to the identification of 812 cis-pQTL for 737 proteins, which were used in Mendelian randomisation analysis of breast cancer in the Breast Cancer Association Consortium (BCAC) case-control meta-analysis of breast cancer risk, followed by replication in independent studies of breast cancer risk. To follow up on significant proteins, the genetic signals for protein levels and breast cancer risk were visualised and evaluated using Mirror plots and were also tested for causal relationships with established and emerging breast cancer risk factors, also using Mendelian randomisation.

Fig. 2. Volcano plots showing estimated effect sizes (x-axis) and the corresponding non-adjusted –log10(p-value) (y-axis) for each of the 2476 proteins analysed in relation to KARMA baseline characteristics.

The plots show estimated effect sizes (x-axis) and the corresponding non-adjusted 2-sided −log10(p-value) (y-axis) with the dashed line marking p = 1 × 10⁻⁵ for visual support. Effect sizes are given by a linear regression model per protein, including all 7 baseline characteristics. Each panel shows one of the investigated baseline characteristics, corresponding to one term in the regression model. The names of up to ten significant proteins per clinical parameter are indicated in each panel according to FDR < 0.05 corrected statistical significance (unadjusted p < 0.0037). The number of proteins reaching FDR-adjusted significance were for age: 459, Alcohol consumption: 172, Birth times: 7, BMI: 684, HRT: 93, Menopause pre vs. peri: 18, Menopause pre vs post: 127, Current smoking: 213.

Fig. 3. Volcano plot of effect sizes (X-axis) and −log10p (Y-axis) for the 730 proteins tested for breast cancer risk in the Mendelian randomisation analysis.

Mendelian randomisation analysis on breast cancer risk in the BCAC study was performed by modelling exposure to genetically higher plasma levels of 730 proteins with at least one cis-pQTL. The Y-axis shows the −log10 p-value of the Wald-score or IVW and the X-axis shows the beta-estimates of the MR result for each protein that was tested.