Background
Breast cancer is the most common cancer in women, with the lifetime risk of breast cancer for women in highly economically developed countries being 1 in 9. While breast cancer is a leading cause of death in women [
1], the exact mechanisms of breast cancer initiation and progression are not known [
2], necessitating a better understanding of disease aetiology.
The UK Biobank (UKB) study is a prospective cohort study that measured the genotypes and levels of 34 biochemical biomarkers of around 500,000 participants aged between 40 and 69 years, of which we sampled 194,174 women of white-British ancestry [
3]. The biomarkers are grouped into six categories, namely bone and joint, cancer, cardiovascular, diabetes, liver, and renal biomarkers, which were measured due to their established relevance in a range of diseases and their diagnostic value and because they characterise phenotypes that are otherwise difficult to assess.
A few observational studies have been performed to study the associations between some of the UKB biochemical biomarkers and overall breast cancer, and significant associations have been found for several biomarkers. However, observational studies are prone to residual confounding and reverse causation. Mendelian randomisation (MR) complements observational studies by using genetic variants as instrumental variables (IVs) to establish likely causal associations between exposures and outcomes. To our knowledge, fewer than half of the biochemical biomarkers in the UKB have been investigated for likely causal associations with overall breast cancer using MR, and even fewer studies have stratified breast cancer by oestrogen receptor (ER) presence, which influences the disease prognosis and type of therapy that will be most effective [
2]. See Additional file
1: Table S1 for a summary of the most recent observational and MR findings associating the UKB biochemical biomarkers with breast cancer in the literature.
This study aimed to use an MR framework to (1) explore univariable associations between genetically predicted levels of UKB biochemical biomarkers and genetic liability to overall, ER-positive, and ER-negative breast cancer; (2) investigate significant associations in detail through multivariable and bidirectional approaches; and (3) to rank the associated biomarkers by genetic evidence using a multivariable Bayesian MR approach. We achieved our aims by replicating and extending previous analyses to a bigger sample containing ~420,000 women and providing novel evidence for biomarkers not previously studied using MR.
Discussion
In this study, we used a hypothesis-generating two-sample summary-level MR approach to screen the UKB for biochemical breast cancer biomarkers. We found that increases of 1 standard deviation in the genetically predicted levels of testosterone, HDL cholesterol, IGF-1, and ALP were robustly and consistently associated with overall breast cancer liability in a variety of univariable, multivariable, bidirectional, and ranking methods based on MR. These associations remained for ER-positive breast cancer, but only HDL cholesterol remained associated with ER-negative breast cancer. To our knowledge, ALP has not been associated with breast cancer before. The summary of our findings and how these compared with the literature to the best of our knowledge can be found in Additional file
5: Table S1.
For bone and joint biomarkers, we observed a novel inverse association between genetically predicted levels of ALP and overall and ER-positive breast cancer liability that was robust in all MR analyses. One possible explanation for this finding is that ALP-prioritised genes are enriched in primary and secondary sexual organs, and crucially, gene sets enriched among ALP-associated variants included oestradiol 17-beta-dehydrogenase activity, which catalyses oestradiol to the less potent estrone, thus reducing the risk of breast cancer [
35]. We were unable to adjust our findings for oestradiol concentrations, as there are no large, high-quality GWASes for oestradiol. We instead adjusted for testosterone and SHBG in MVMR and did not observe an attenuation of the effect. Future research is required to clarify whether the ALP and breast cancer liability association is independent of oestrogens. A nominally significant negative association between serum levels of calcium and overall breast cancer risk was found in cohort studies [
36], but not in an MR study [
37], with which our study concurs. While vitamin D is negatively associated in observational analyses, no evidence of association could be found in MR [
38], in agreement with the current study. We found no evidence of an association between genetically predicted rheumatoid factor and breast cancer liability.
For cancer biomarkers, a previous observational and MR study found positive associations between levels of IGF-1 and overall breast cancer risk in women in the UKB [
39], which agrees with our findings. IGF-1 has long been implicated in breast cancer due to the role of IGF-1 receptors in activating the AKT and mitogen-activated protein kinase signalling networks in tumour growth [
40]. A meta-analysis of observational studies found positive associations between oestradiol and overall breast cancer risk in post-menopausal women [
41], which we could not confirm or dispute due to a lack of valid IVs for our MR analyses. This was likely due to imprecise measurements of oestradiol levels in the UKB [
42], which was also a problem in a different study that led to oestradiol being excluded [
43]. Positive associations between testosterone and breast cancer were found in a meta-analysis of prospective studies [
41] and an MR study [
42], which agree with our study. One hypothesis for our observed positive association of genetically predicted testosterone with overall, and ER-positive, but not ER-negative breast cancer liability is that the effect is in part mediated by the downstream conversion to oestradiol [
42]. A negative association between SHBG levels and breast cancer was observed in a meta-analysis of prospective studies [
44], and in an MR study, only after adjusting for BMI in an MVMR model [
30], in agreement with our study where we only found an association after adjusting for BMI for ER-positive breast cancer liability.
For cardiovascular biomarkers, a cohort study found an inverse association between ApoB, but not ApoA and breast cancer risk [
45]. However, our MR study found a positive association between genetically predicted ApoA, but not genetically predicted ApoB and overall breast cancer liability. This difference in findings may have arisen due to confounding or reverse causation in the prospective cohort study. A meta-analysis of 15 observational studies did not find evidence of an association between CRP levels and overall breast cancer liability [
46], in agreement with a previous MR study [
47] and the current MR study. No evidence of associations of cholesterol, HDL cholesterol, LDL cholesterol, and triglycerides with overall breast cancer risk was found in a meta-analysis of cohort studies [
48]. An MR study also found no evidence of associations of genetically predicted cholesterol, LDL cholesterol, or triglycerides, but found a positive association of HDL cholesterol with overall breast cancer [
49]. We also found no evidence of associations of genetically predicted cholesterol or LDL cholesterol with overall breast cancer. However, we observed that genetically raised HDL cholesterol was consistently significantly positively associated with all breast cancer outcomes. HDL cholesterol has been shown to stimulate breast cancer cell line proliferation in a dose-dependent relationship. The HDL receptor scavenger receptor class B type I, which contributes to tumour development via AKT and ERK1/2, has also been shown to be expressed more abundantly in human breast cancer tissue than in non-cancerous tissue [
50]. Triglycerides were associated with a decreased liability for breast cancer, although not significantly in MVMR including the other lipids, and not ranked highly in MR-BMA. There was no evidence of an association between genetically predicted lipoprotein A levels and breast cancer liability in our study.
For diabetes-related biomarkers, a meta-analysis of 10 cohort studies [
51] and a previous MR study [
52] found evidence of a positive association between serum glucose levels and risk or odds of overall breast cancer. However, we did not observe any evidence of association. We did not observe any associations between genetically predicted glycated haemoglobin levels and breast cancer liability.
For liver biomarkers, a case-cohort study found an inverse association between albumin and breast cancer risk [
53], while our MR analyses did not find any evidence of association, likely due to residual confounding in the case-cohort study. The results of a meta-analysis of two cohort studies showed a higher risk of breast cancer with higher gamma-glutamyltransferase concentrations [
54], whereas we did not find any evidence of an association in our MR study, possibly due to confounding in the cohort studies. No evidence of an association between total bilirubin concentrations and overall breast cancer liability was found in a case-cohort study [
53], in agreement with our MR results. We found evidence of an inverse association between genetically predicted aspartate aminotransferase concentrations and overall or ER-positive breast cancer liability. Yet, given the inconclusive evidence from our pleiotropy-robust approaches, possible bias from pleiotropy could not be excluded. We did not find any evidence of associations between genetically predicted alanine aminotransferase or direct bilirubin levels with breast cancer liability.
For renal biomarkers, a case-cohort study found an inverse association between uric acid levels and overall breast cancer liability. However, following adjustment for albumin, the association was attenuated [
53]. Our study found no evidence of an association between genetically predicted urate levels and breast cancer liability. We found evidence for an inverse association between genetically predicted urea levels and ER-positive breast cancer liability, but the evidence was inconclusive in the pleiotropy-robust approaches due to large uncertainties that included the null, meaning that our results were more suggestive of a lack of association. We could not find any evidence of associations of genetically predicted serum creatinine, enzymatic creatinine, cystatin C, microalbumin, phosphate, potassium, sodium, and total protein with breast cancer liability.
A limitation of our study was that the data was restricted to women of white-European ancestry to avoid heterogeneity issues, which hinders our ability to generalise to populations of other ethnic backgrounds. Another deficit of our study was that our exposure [
12] and outcome [
8] samples were predominantly post-menopausal, thus limiting generalisability to pre-menopausal women. Moreover, though we performed multiple MR sensitivity analyses, there is still the possibility of residual pleiotropy.
Our study’s strengths include applying many univariable sensitivity analyses to appraise the validity of IV assumptions and limit potential bias from pleiotropy. We also included several MVMR models in our study to adjust for potential risk factors. To investigate reverse causation, we also conducted bidirectional MR for the association between genetically predicted ALP concentrations and breast cancer liability. Biomarker samples were collected prospectively from a large sample, and we accounted for population stratification by restricting our study to participants of white-European ethnicity and adjusting for genetic principal components. We explored genetic associations in women, which excluded the potential for sex-specific effects that can arise for biomarkers such as testosterone [
55]. Most of our results supported findings from previous studies, which acted as positive controls for our methods. Our study allowed for the generation of hypotheses, enabling further studies to be targeted at biomarkers of interest with little prior evidence of association, such as ALP. Ranking biomarkers in an agnostic manner using MR-BMA reinforced our confidence in the strength of our findings and provided us with information about the importance of testosterone, HDL cholesterol, IGF-1, and ALP in breast cancer liability.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.