Previous studies for circulating glycemic traits (GTs), including fasting glucose (FG) and insulin (FI) concentrations, have shown inconsistent associations with breast cancer development (Gunter et al., 2009; Sieri et al., 2012; Boyle et al., 2013; Hernandez et al., 2014). This is partially due to selection bias, confounding, short time exposure to such metabolic biomarkers, measurement errors, and reverse causation. We tried to address those challenges by using a 2-sample Mendelian Randomization (MR) approach and examined whether genetically determined GTs are causally associated with breast cancer risk. The MR method may provide a relatively unbiased causal relationship between phenotype and cancer outcome because it reduces potential bias and confounding and prevents reverse causation by the random assortment of alleles at meiosis, resulting in random assignment of exposure, which precedes the phenotype and clinical outcomes (Merino et al., 2017).

For the GT instrumental variables, we used the recently updated publicly available data in 2019 from genome-wide association studies (GWASs) of the Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC) in non-diabetic European women¹. Detailed rationale and design of the studies have been described elsewhere (Scott et al., 2012). For breast cancer outcomes, we pulled 4 datasets from 2 independent consortia: Breast Cancer Association Consortium (BCAC²): (i) OncoArray; the Atlas of GWAS Summary Statistics (ATLAS³): (ii) CGEMS Breast Cancer GWAS, (iii) GEE adjusted for age, and (iv) GEE for age and body mass index. Study participants from each dataset provided written informed consent. Genetic instruments for each dataset were single-nucleotide polymorphisms (SNPs) associated with the trait at the genome-wide level (p < 5E-08).

We performed MR analysis using the inverse-variance weighted method (Burgess et al., 2013) which quantifies the genetically determined association between GTs and breast cancer risk. We assume that summarized data are available for multiple genetic variants in relation to the risk factor of interest X and the outcome Y; genetic variant k, k = 1,…, K is associated with an observed X_k mean change in the risk factor per additional variant allele with standard error σ_Xk and an observed Y_k changes in the log-odds or the log-probability of an outcome per allele with standard error σ_Yk. The inverse-variance weighted estimates combine the ratio estimates from each variant in a fixed-effect meta-analysis model:

The approximate standard error of the estimate is:

The results were reported as risk ratios and 95% confidence intervals for the change in breast cancer risk per unit increase in FG (mmol/L) or natural log-transformed FI (pmol/L). To determine the extent of pleiotropic signal, we applied Cochran’s Q test and the MR-Egger analysis. Given obesity and diabetes’s established role for breast cancer, we excluded those relevant SNPs from the analysis. R3.6.1 was used. The Institutional Review Board of the University of California, Los Angeles, approved this study.

Of 430 GWA-based SNPs related to GTs in MAGIC, 44 SNPs within linkage disequilibrium (r² < 0.1) were matched to either BCAC or ATLAS datasets (Table 1). The 38 FG-SNPs overall and stratified by cancer data source showed heterogeneous results but mostly showed a slightly increased effect on breast cancer risk without reaching statistical significance (Table 2 and Figure 1). After excluding top GWA-SNPs associated with type 2 diabetes and visceral obesity, the directions of the associations between GTs and breast cancer were changed in OncoArray but not in ATLAS-CGEMS. The 6 FI-SNPs showed similar patterns for the associations with breast cancer. No significant directional pleiotropy was observed (Table 3).

We analyzed the relatively large and most-updated GWA-datasets for causality between GTs and breast cancer. Given that associations between metabolic markers and breast cancer risk can differ by menopausal status, our findings may be confounded. However, data was not available on the menopausal status, thus warranting future studies that account for this difference. In addition, whereas MR is considered a conservative approach, it may be confounded when modeled SNPs independently affect breast cancer risk through intermediate traits other than GTs.

Our study results should be interpreted with caution because of population structure bias (i.e., results biased due to tagged environmental factors) and unmeasured confounding factors that could have introduced bias. MR analysis might also be subject to non-linearity between exposure and outcome, but potential violation of the linearity assumption tends to bias MR estimates toward the null, rather than generating a spurious association (Smith and Ebrahim, 2003). Moreover, our study may not be generalized to other races or ethnicity, in which the association between genetic instruments, GTs, and breast cancer risk may be different.

Our findings indicate a null association between genetically determined GTs and breast cancer risk among European women. Our study may contribute to more complete characterizing of molecular pathways in GTs and breast cancer. It also highlights the need to conduct a more comprehensive and individual-level analysis using more detailed trait information, including risk causing confusion in this field of research.

The datasets generated for this study are available on request to the corresponding author.

SJ, NM, SH, and Z-FZ designed the study. SJ and SH performed the genomic data QC and statistical analysis and interpreted the data. NM and Z-FZ supervised the genomic data QC and analysis and participated in the study coordination and interpreted the data. SJ secured funding for this project. All authors participated in the manuscript writing and editing, read and approved the submission of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.