A Mendelian randomization analysis of circulating lipid traits and breast cancer risk

Abstract

Background

Conventional epidemiologic studies have evaluated associations between circulating lipid levels and breast cancer risk, but results have been inconsistent. As Mendelian randomization analyses may provide evidence for causal inference, we sought to evaluate potentially unbiased associations between breast cancer risk and four genetically predicted lipid traits.

Methods

Previous genome-wide association studies (GWAS) have identified 164 discrete variants associated with high density lipoprotein-cholesterol (HDL-C), low density lipoprotein-cholesterol (LDL-C), triglycerides and total cholesterol. We used 162 of these unique variants to construct weighted genetic scores (wGSs) for a total of 101 424 breast cancer cases and 80 253 controls of European ancestry from the Breast Cancer Association Consortium (BCAC). Unconditional logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (CI) for associations between per standard deviation increase in genetically predicted lipid traits and breast cancer risk. Additional Mendelian randomization analysis approaches and sensitivity analyses were conducted to assess pleiotropy and instrument validity.

Results

Corresponding to approximately 15 mg/dL, one standard deviation increase in genetically predicted HDL-C was associated with a 12% increased breast cancer risk (OR: 1.12, 95% CI: 1.08–1.16). Findings were consistent after adjustment for breast cancer risk factors and were robust in several sensitivity analyses. Associations with genetically predicted triglycerides and total cholesterol were inconsistent, and no association for genetically predicted LDL-C was observed.

Conclusions

This study provides strong evidence that circulating HDL-C may be associated with an increased risk of breast cancer, whereas LDL-C may not be related to breast cancer risk.

Key Messages

• We conducted a large Mendelian randomization analysis to provide unbiased estimates of association with breast cancer risk for four lipid traits among 181 677 European-ancestry women from the Breast Cancer Association Consortium.

• One standard deviation increase (representing approximately 15 mg/dL) increase in genetically predicted high density lipoprotein-cholesterol (HDL-C) was associated with a 12% increased risk of breast cancer, whereas no consistent associations were found with low density lipoprotein-cholesterol, triglycerides or total cholesterol.

• This study suggests that circulating HDL-C levels may influence breast cancer susceptibility.

Introduction

Circulating lipids, including high density lipoprotein-cholesterol (HDL-C), low density lipoprotein-cholesterol (LDL-C), triglycerides and total cholesterol, have long been hypothesized to influence the risk of breast, colorectal and other common cancers.^1–4 Early prospective cohort studies reported inverse associations for total cholesterol and cancer risk.^2–4 However, these findings could be due to reverse causation, where disease development or progression leads to lower circulating cholesterol levels years before disease diagnosis.^5–7 It is also possible that confounding factors, such as smoking, alcohol consumption, and socioeconomic status may have biased associations reported in previous epidemiologic studies.^8,9 

The role of HDL-C in disease risk is controversial. Although it is an established risk factor for coronary heart disease,¹⁰ large Mendelian randomization analyses have suggested that the association between low HDL-C and heart disease may not be causal.^11,12 Furthermore, clinical trials designed to increase circulating HDL-C levels pharmacologically have not demonstrated overall benefits in heart disease prevention.^13,14 With regard to breast cancer, multiple studies have found inverse associations between HDL-C and risk.^15,16 Contrary to these findings, HDL-C was associated with increased breast cancer risk when repeated serum lipid measures were evaluated.¹⁷ Given the current controversy regarding the association between circulating lipid traits and cancer in general, and with breast cancer in particular, this finding has faced scepticism.¹⁸

Due to methodological limitations such as reverse causation and confounding, it is unlikely that conventional observational studies can resolve the longstanding debate about the role of circulating lipids in breast cancer development. Mendelian randomization analyses can potentially overcome some of the limitations inherent in conventional epidemiologic studies. Taking advantage of the random assortment of alleles which occurs during gametogenesis, thereby resembling randomized clinical trials, Mendelian randomization analysis uses genetic data (i.e. single nucleotide polymorphisms, or SNPs) as genetic instruments to estimate exposures of interest for association analyses with disease outcomes. Results from a Mendelian randomization analysis may provide strong evidence for causality, if the genetic instruments used are associated with the exposure, only affect the outcome via the exposure and are not associated with any of the confounders of the exposure-outcome relationship.¹⁹ To date, genome-wide association studies (GWAS) have linked circulating lipid traits to at least 157 genetic loci.^20,21 A recent Mendelian randomization analysis used only summary statistics approaches and reported that genetically increased LDL-C and HDL-C were associated with increased estrogen receptor (ER)-positive breast cancer risk.²² We independently conducted a Mendelian randomization analysis that leveraged both individual-level and summary statistics data for lipid-associated variants, and created instrumental variables to evaluate shared genetic components and associations between four circulating lipid traits and breast cancer risk.

Methods

Study population

The Breast Cancer Association Consortium (BCAC) is an international collaboration initiated in 2005 to study genetic susceptibility to breast cancer. First, we included individual-level epidemiologic and genetic data from 62 846 breast cancer cases and 43 207 healthy controls of European ancestry from 67 BCAC studies; genetic data included more than 500 000 variants from a custom OncoArray platform that was designed to provide dense coverage across known cancer susceptibility loci as well as common variants.²³ Second, we included independent data from 38 578 cases and 37 046 BCAC controls that were genotyped on the Illumina iSelect genotyping Collaborative Oncological Gene-Environment Study (iCOGS) array [http://ccge.medschl.cam.ac.uk/research/consortia/icogs/].²⁴ Demographic and select patient characteristics were harmonized across BCAC studies according to a standardized protocol. All BCAC studies were approved by relevant institutional review boards, and all participants provided written informed consent.

Variant genotyping, imputation, and selection

Genetic variants associated with lipid traits were selected from the Global Lipids Genetics Consortium. The first lipid-trait GWAS included approximately 100 000 subjects of European ancestry and identified 102 genetic variants in 95 loci.²⁰ The second GWAS, conducted among 188 577 subjects predominantly of European ancestry, identified 83 additional variants in 62 loci, resulting in a total of 185 lipid-associated variants in 157 loci.²¹ However, among the 83 variants, 21 were associated with more than one lipid trait and were not discrete. Thus, we identified a total of 164 unique lipid trait associated variants, of which 87 were genotyped by OncoArray and 75 were imputed with high information quality scores (mean r² = 0.98, range = 0.86–0.99). In iCOGS data, 39 selected variants were genotyped, and 123 were successfully imputed (mean r² = 0.82, range = 0.35–0.99). Two variants (rs2247056 and rs3177928) were not imputed in either dataset, providing a total of 162 in our analysis (Supplementary Table S1, available as Supplementary data at IJE online). Except for two variants (rs2814982 and rs2814944) in moderate linkage disequilibrium (LD; r² = 0.51), all included variants were independent (r² < 0.1). Because rs2814982 was associated with total cholesterol and rs2814944 was associated with HDL-C, both were retained in our analysis, as no instrumental variable included both SNPs. Thus, based on information available from published GWAS, our instrumental variables for HDL-C, LDL-C, triglycerides and total cholesterol included 74, 57, 43 and 74 variants, respectively.

Mendelian randomization analyses

Our primary analysis used individual-level data from BCAC iCOGS and OncoArray to generate weighted-genetic scores (wGSs) for four lipid traits (HDL-C, LDL-C, triglycerides and total cholesterol). For each lipid trait, we constructed instrumental variables as follows: wGS=${\sum }_{i=1}^n{\beta }_{\mathit{gx}}\mathrm{*}{\alpha }_i$, where β_gx represents the effect for the genetic variant (g) associated with an increase in lipid levels (x) and α_i is effect allele dosage for each genetic variant (ranging from 0 to 2 for each individual), for n genetic variants from the Global Lipids Genetics Consortium GWAS.^20,21

Associations between lipid trait wGRs and breast cancer risk factors (conducted separately for iCOGS and OncoArray datasets) were assessed with linear or logistic regression for continuous or categorical variables, respectively (Supplementary Table S2, available as Supplementary data at IJE online). Associations between lipid trait wGSs and breast cancer risk were estimated by odds ratios (ORs) and 95% confidence intervals (95% CIs) from unconditional logistic regression using individual-level data. Analyses were conducted separately for BCAC participants with iCOGS and OncoArray data (Supplementary Table S3, available as Supplementary data at IJE online), and then combined by random-effects or fixed-effects meta-analysis (Supplementary Table S4, available as Supplementary data at IJE online); Cochran’s Q statistic was used to evaluate heterogeneity. Models were adjusted for age, principal components (PCs) for European ancestry (iCOGS: six PCs; OncoArray: 10 PCs), and either study site (iCOGS) or country (OncoArray), as previously described.^23,25 Additional adjustment included breast cancer risk factors that were associated with lipid trait wGSs. We assessed effect measure modification by menopausal status, age (dichotomized at 50 years) and body mass index (dichotomized at 30 kg/m²) using likelihood ratio tests (LRT) for multiplicative interaction terms in nested models. Polytomous regression was employed to evaluate associations with estrogen receptor (ER) positive (+) and ER negative (-) breast cancer subtypes; tests of equivalence of beta coefficients across subtypes were used to evaluate heterogeneity.

To reduce correlation between instrumental variables, we also constructed amended wGSs that included only genetic variants that were exclusively associated with HDL-C (55 variants), LDL-C (44 variants) or triglycerides (20 variants) at a genome-wide significance level (Supplementary Table S5, available as Supplementary data at IJE online) and then re-evaluated associations with breast cancer risk (Supplementary Table S6, available as Supplementary data at IJE online). Because total cholesterol includes other lipid traits, no such amended wGS was created. Analyses were completed using SAS (version 9.4) and Stata (version 12.1).

Sensitivity analyses

In addition to individual-level analyses, we also conducted Mendelian randomization analysis using inverse-variance weighted summary statistics (Supplementary Table S7, available as Supplementary data at IJE online).²⁶ Four additional sensitivity analyses were used to assess the influence of genetic pleiotropy and validity of our genetic instruments. First, Mendelian randomization Egger (MR Egger) regression was employed to evaluate the presence of directional pleiotropy by testing whether the intercept was statistically different from zero, and to estimate a bias-reduced Mendelian randomization estimate from the regression slope.²⁷ Second, a weighted multivariable regression-based approach was used to assess the influence of potential pleiotropic effects of genetic variants included in each instrument on other lipid traits; specifically, we regressed beta-coefficients for associations between genetic variants and breast cancer risk (β_BC) on beta-coefficients between genetic variants and lipid traits (HDL-C: β_HDL-C, LDL-C: β_LDL-C, triglycerides: β_TG, and total cholesterol: β_TC), thereby adjusting for the associations between genetic variants and other lipid traits.^28,29 Third, we estimated associations using a weighted-median Mendelian randomization approach where we assumed that 50% of the variants included in each genetic instrument were invalid instruments (i.e. did not meet at least one of the three assumptions necessary for a valid instrumental variable); standard errors were estimated by bootstrapping and were subsequently used to calculate 95% CIs.³⁰ Fourth, we conducted a leave-one-out analysis where the Mendelian randomization association was re-estimated after removing the strongest SNP (as determined by the largest change in magnitude in comparison with results from instruments with all variants). Sensitivity analyses were conducted using the ‘TwoSampleMR’ package curated by MR-Base³¹ using R version 3.5.1, R Foundation for Statistical Computing [https://www.r-project.org/]. Finally, visual representations of the IVW, MR Egger, and weighted-median approaches were created for comparison (Supplementary Figures S1–S4, available as Supplementary data at IJE online), and funnel plots for individual SNP MR estimates in relation to the inverse of the standard errors (Supplementary Figures S5–S8, available as Supplementary data at IJE online) were inspected for symmetry to indicate validity of our Mendelian randomization analysis.

Results

One genetically predicted standard deviation increase in HDL-C, LDL-C, triglycerides and total cholesterol was calculated to correspond to approximately 15, 37, 43 and 42 mg/dL increases, respectively. Associations between breast cancer risk factors and lipid trait wGSs were evaluated among all BCAC participants and among only controls (Supplementary Table S2). Several associations were identified; however, the only consistent association across the two populations and genotyping platforms was between increasing age and lower total cholesterol (iCOGS P = 4.0  ×  10⁻⁴ and OncoArray P = 0.01). Similarly, the only consistent association among controls was between increasing age and lower triglycerides (iCOGs P = 0.04 and OncoArray P = 0.03).

Associations for each standard deviation increase in genetically predicted lipid trait from iCOGS and OncoArray genotyped BCAC participants (Supplementary Table S3) were combined by random-effects meta-analysis (Table 1). Among all women, increased HDL-C levels were associated with increased breast cancer risk (OR: 1.12, 95% CI: 1.08–1.16) in models that included adjustment for age, study site or country and principal components for European ancestry. We found no association for LDL-C (OR: 1.00, 95% CI: 0.96–1.04) modest risk reduction was suggested for increasing triglycerides (OR: 0.93, 95% CI: 0.85–1.01) and a modest increase in risk was suggested for increasing total cholesterol levels (OR: 1.05, 95% CI: 0.99–1.11). Further, we found no significant interactions by menopausal status, age or body mass index (BMI) among either iCOGS or Oncoarray genotyped participants for any lipid trait. For the HDL-C wGS, increased breast cancer risk was observed per one standard deviation increase among postmenopausal women (1.11, 95% CI: 1.05–1.17), women less than 50 years of age (1.17, 95% CI: 1.01–1.34), women age 50 or greater (1.11, 95% CI: 1.06–1.16) and non-obese (BMI < 30 kg/m²) women (1.14, 95% CI: 1.08–1.20). Associations were also observed for both ER- (1.10, 95% CI: 1.03–1.18) and ER+ (1.11, 95% CI: 1.07–1.16) breast cancers. On the contrary, one standard deviation increase in triglycerides was associated with reduced breast cancer risk among postmenopausal women (OR: 0.93, 95% CI: 0.88–0.99), women age 50 or greater (OR: 0.93, 95% CI: 0.89–0.98) and non-obese women (OR: 0.90, 95% CI: 0.83–0.98); the association was also observed for ER+ breast cancer (OR: 0.91, 95% CI: 0.85–0.91). In these stratified analyses, total cholesterol was associated with breast cancer risk only among premenopausal women (OR: 1.08, 95% CI: 1.00–1.17). Results were materially unaltered when fixed-effect meta-analyses were conducted (Supplementary Table S4).

Table 1.

Overall and stratified analyses for the association between one standard deviation (SD) increase in genetically predicted lipid traits and breast cancer risk using individual-level iCOGS and OncoArray data summarized using random-effects meta-analysis in the Breast Cancer Association Consortium (BCAC)

Subgroup	Cases	Controls	High density lipoprotein (HDL-C) wGS a			Low density lipoprotein (LDL-C) wGS a			Triglycerides (TG) wGS a			Total cholesterol (TC) wGS a
			OR b	95% CI b	P	OR b	95% CI b	P	OR b	95% CI b	P	OR b	95% CI b	P
All women	101 424	80 253	1.12	1.08–1.16	1.7 × 10−9	1.00	0.96–1.04	0.88	0.93	0.85–1.01	7.5 × 10−2	1.05	0.99–1.11	0.11
Menopausal status
Premenopausal	20 782	17 902	1.14	0.96–1.35	0.13	1.00	0.92–1.08	0.89	0.90	0.75–1.08	0.24	1.08	1.00–1.17	0.04
Postmenopausal	43 787	38 847	1.11	1.05–1.17	3.2 × 10−4	0.99	0.94–1.05	0.80	0.93	0.88–0.99	0.03	1.00	0.95–1.05	0.96
Test for interactionc			P iCOGS = 0.04, POnco = 0.23			P iCOGS = 0.31, POnco = 0.41			P iCOGS = 0.41, POnco = 0.05			P iCOGS = 0.13, POnco = 0.37
Age
<50 years	24 572	22 944	1.17	1.01–1.34	3.4 × 10−2	0.96	0.90–1.03	0.30	0.90	0.78–1.05	0.17	1.05	0.98–1.12	0.20
≥50 years	71 098	51 700	1.11	1.06–1.16	2.3 × 10−6	1.01	0.97–1.05	0.65	0.93	0.89–0.98	6.7 × 10−3	1.04	0.99–1.08	0.12
Test for interactionc			PiCOGS = 0.07, POnco = 0.74			P iCOGS = 0.14, POnco = 0.68			P iCOGS = 0.78, POnco = 0.13			P iCOGS = 0.71, POnco = 0.99
Body mass index (BMI)
BMI <30 kg/m2	51 114	41 713	1.14	1.08–1.20	1.0 × 10−6	0.99	0.94–1.04	0.70	0.90	0.83–0.98	1.7 × 10−2	1.04	0.97–1.12	0.29
BMI ≥30 kg/m2	12 200	9507	1.06	0.95–1.18	0.31	0.95	0.85–1.05	0.31	1.07	0.95–1.21	0.29	0.96	0.87–1.06	0.41
Test for interactionc			P iCOGS = 0.67, POnco = 0.24			P iCOGS = 0.16, POnco = 0.99			P iCOGS = 0.22, POnco = 1.9 × 10−2			P iCOGS = 0.10, POnco = 0.68
Estrogen receptor (ER) status
ER- breast cancer	43 039	80 253	1.10	1.03–1.18	7.0 × 10−3	0.99	0.93–1.05	0.73	0.94	0.87–1.01	8.3 × 10−2	1.06	0.93–1.21	0.36
ER+ breast cancer	61 140	80 253	1.11	1.07–1.16	8.6 × 10−8	0.99	0.95–1.03	0.63	0.91	0.85–0.98	1.4 × 10−2	1.03	0.99–1.07	0.17
Test for heterogeneityc			P iCOGS = 0.95, POnco = 0.52			P iCOGS = 0.50, POnco = 0.61			P iCOGS = 0.97, POnco = 0.26			P iCOGS = 0.16, POnco = 0.69

^aTotal number of variants included in each lipid-specific weighted-genetic score (wGS): HDL-C = 74, LDL-C = 57, TG = 43 and TC = 74.

^bOdds ratios (OR) and 95% confidence intervals (CI) represent results from random-effects meta-analysis summarizing the associations from iCOGS and OncoArray datasets for one SD increase in wGS in relation to breast cancer risk among stratified groups. Polytomous regression was employed to estimate risk for ER+ and ER- breast cancer in the iCOGS and OncoArray datasets. All models were adjusted for age, BCAC study site (iCOGS) or study country (OncoArray), and either top six (iCOGS) or top 10 principal components for European ancestry (OncoArray).

^cInteractions were assessed on a multiplicative scale using the likelihood ratio test (LRT) for nested models. Differences in stratum-specific ORs and 95% CIs for tumour subtype were assessed using a test for heterogeneity.

In addition, we constructed amended instruments with reduced correlation by including only 55, 44 or 20 genetic variants that were exclusively associated with either HDL-C, LDL-C, or triglycerides (Supplementary Table S5); the amended HDL-C wGS was associated with increased breast cancer risk (OR: 1.14, 95% CI: 1.07–1.22) whereas the amended triglyceride wGS was not (OR: 1.00, 95% CI: 0.86–1.16) (Supplementary Table S6). Regardless of adjustment, or whether initial or exclusive variants were included, the LDL-C wGS was not associated with breast cancer risk in our analyses.

We also conducted Mendelian randomization analysis using summary statistics data and included several sensitivity analyses to assess the validity of our instrumental variables (Supplementary Table S7). The inverse-variance weighted Mendelian randomization estimate using summary statistics²⁶ per standard deviation (SD) increase in circulating lipids confirmed our initial findings: increased HDL-C was associated with increased breast cancer risk (OR_IVW: 1.12, 95% CI: 1.08–1.17) whereas increased LDL-C was not associated with breast cancer risk (OR_IVW: 0.99, 95% CI: 0.96–1.03). These associations were consistent for HDL-C and LDL-C regardless of dataset. On the contrary, triglycerides were associated with reduced breast cancer risk only in OncoArray data (OR_IVW: 0.88, 95% CI: 0.83–0.92) and total cholesterol was associated with increased risk only in iCOGS data (OR_IVW: 1.06, 95% CI: 1.01–1.12). The MR Egger regression intercept indicated that the IVW estimate for total cholesterol was potentially biased due to directional pleiotropy (iCOGS: β_intercept = 0.0105, P-value = 7.3 × 10⁻²; OncoArray: β_intercept = 0.0053, P value = 7.3 × 10⁻²). The bias-reduced estimate, derived from MR Egger regression, indicated a potential risk reduction for total cholesterol (OR_{MR Egger}: 0.92, 95% CI: 0.85–1.00). We also conducted weighted multivariable regression with mutual adjustment for other lipid traits.^28,29 Increasing HDL-C was associated with increased breast cancer risk in iCOGS data (OR_{weighted-regression}: 1.16, 95% CI: 1.08–1.25), and higher triglycerides were associated with decreased breast cancer risk in OncoArray data (OR_{weighted-regression}: 0.88, 95% CI: 0.81–0.95). Using a weighted-median approach, which assumes that half of included variants are invalid,²⁷ only HDL-C was associated with increased breast cancer risk after meta-analysis across our data sources (OR_{weighted-median}: 1.08, 95% CI: 1.02–1.14). Results from our leave-one-out analysis also yielded an association for HDL-C and breast cancer risk (OR_{Leave-one-out}: 1.13, 95% CI: 1.06–1.20). Finally, associations across approaches were compared visually (Supplementary Figures S1–S4), and symmetry of funnel plots supported the validity of our Mendelian randomization analysis (Supplementary Figures S5–S8).

Discussion

In this large-scale Mendelian randomization study using 162 lipid-associated GWAS variants, we found that higher levels of genetically predicted HDL-C were associated with an increased risk of breast cancer. This finding was robust and consistent across a variety of analytical approaches. Genetically predicted triglyceride and total cholesterol levels were also associated with breast cancer risk in some analyses, but these findings were not consistent and varied by data source and statistical adjustment. Genetically predicted LDL-C was not associated with breast cancer risk in any analyses. Traditional epidemiologic studies that have measured circulating lipids and evaluated breast cancer risk have had conflicting results, likely due to reverse causation, confounding and selection bias. By using a Mendelian randomization approach, we aimed to overcome limitations inherent in traditional studies and to provide strong evidence supporting a possibly causal association between high HDL-C levels and increased breast cancer risk.

Another Mendelian randomization analysis on lipids and breast cancer risk with BCAC data was recently published; their primary findings include an increased risk of ER-positive breast cancer risk per standard deviation of genetically raised HDL-C (OR: 1.13, 95% CI: 1.01–1.26) or LDL-C (OR: 1.14, 95% CI: 1.05–1.24).²² Several methodological differences may explain why these results differ from ours, most notably for LDL-C. First, in addition to using summary statistics approaches, our analysis included individual-level BCAC data, which enabled us to control for potential confounding by breast cancer risk factors and to conduct stratified analyses. Second, we selected GWAS-significant variants from primary tables in published GWAS, which had slight differences in information available from the Global Lipids Genetics Consortium.²¹ Third, although we both started with 185 variants in 157 loci, the number of SNPs included in the final genetic instruments differed considerably. Rather than a subset, we included all available independent variants in our primary instruments: for example, 57 versus 44 SNPs for LDL-C.

Other relevant publications include a study of serum lipids found that higher HDL-C was associated with increased breast cancer risk when serial measurements were assessed, but not when only one baseline measure was evaluated.¹⁷ This contrasts with a meta-analysis of prospective studies that found modest inverse associations with breast cancer risk for both total cholesterol and HDL-C.¹⁵ Given that circulating cholesterol levels are often decreased several years before cancer diagnosis, inverse associations for this trait could be attributable to bias from reverse causation. In addition, residual confounding from factors such as mammographic breast density or alcohol intake, and effect modification by menopausal status, may also likely influence associations between circulating lipids and breast cancer risk.¹⁸

Plasma lipoproteins transport triglycerides and cholesterol between the liver and tissues. HDL-C is the smallest and most dense lipoprotein, and accounts for approximately 30% of total cholesterol, with levels ranging between 40–60 mg/dL. Higher HDL-C concentrations are associated with better cardiovascular health and lower coronary heart disease risk.¹⁰ However, recent Mendelian randomization analyses have suggested that high HDL-C may not be causally related to reduced coronary heart disease risk.^11,12 Furthermore, pharmacological interventions to increase HDL-C levels have not consistently translated to improved health outcomes,^13,14 and a consensus statement from the National Lipid Association concluded that HDL-C is not currently a therapeutic target.¹⁰ Instead, measures of HDL functionality may be more important than absolute levels, as not all HDL-C functions the same way.¹⁰ For example, oxidized HDL-C and HDL-C from patients with type 2 diabetes had greater capacity to promote proliferation, migration and metastasis of breast cancer cells.³² Thus, in addition to a major role in reverse cholesterol transport and anti-atherogenic effects, HDL-C also seems to have other functions, including the potential to enhance proliferation of breast cancer cells.^32,33 These data provide possible biological mechanisms supporting the increased breast cancer risk seen with increasing levels of genetically predicted HDL-C in our study.

In the current study, associations for triglycerides and total cholesterol in relation to breast cancer risk were inconsistent. The total cholesterol wGS was associated with breast cancer risk only among iCOGS genotyped participants, and multivariate adjustment attenuated this association. Similarly, genetically predicted triglycerides were associated with reduced breast cancer risk only among OncoArray genotyped participants, and the exclusive variant instrument did not influence breast cancer risk. This suggests that some previously reported associations may be due to residual confounding, and that additional evaluation to understand these discrepant findings may be warranted.

Strengths of this study include a very large sample size, strong instrumental variables for all four lipid traits (F-statistics all >10: HDL-C = 190, LDL-C = 266, TG = 274 and TC = 364),³⁴ and multiple analytical approaches to assess instrument validity. We included 162 variants reported by the Global Lipids Genetics Consortium, which account for approximately 13.7%, 14.6%, 11.7% and 15.0% of the variance in HDL-C, LDL-C, triglycerides and total cholesterol, respectively.^20,21 Given the large number of variants used to construct our instruments, our wGSs are likely to be most strongly associated with lipids, and not as strongly associated with other traits, satisfying one of the assumptions for a valid Mendelian randomization analysis. Given that pleiotropy remains a concern for Mendelian randomization analyses, we carefully evaluated this possibility using several analytical approaches.^27,28 Associations from exclusive variant wGS, inverse-variance weighted Mendelian randomization and weighted-median regression analysis were consistent, showing associations between HDL-C and breast cancer risk. Furthermore, our estimates were consistent after considering the potential influence of pleiotropy via MR Egger regression and multivariable weighted-regression. In addition, our results were also unaltered whether fixed-effect or random-effect meta-analysis was conducted.

Limitations of our study include that we did not have direct measurements of circulating lipid levels from our study population to further confirm the validity of our instrumental variables. However, in Mendelian randomization analyses, it is preferable to use externally-derived weights for constructing genetic scores rather than internally-derived weights from the same study population.³⁵ We included external weights from a single large GWAS that was conducted predominantly among Europeans,^20,21 and included only women of European descent in the current analysis. Additional lipid trait genetic variants have also been reported;³⁶ however, weights from this multi-ethnic GWAS would not be applicable to our Caucasian study population. Additional limitations include incomplete information on all confounding factors, and that we could not evaluate or adjust for all such possible covariates, so whether our findings were influenced by residual confounding or another potential source of systematic bias cannot be determined. However, when we adjusted for known breast cancer risk factors that were associated with our lipid trait wGSs, our results for HDL-C were unaltered. In addition, many of the variants included in our analysis were not directly genotyped. However, we used only imputed variants of high quality (iCOGS: mean r² = 0.82, range = 0.35–0.99; OncoArray; mean r² = 0.98, range = 0.86–0.99). Any misclassification of genetic variants would be expected to be non-differential regarding our outcome, which would be more likely to attenuate rather than amplify an association. We also note that our genetic instruments may still include correlated SNPs (with r² threshold <0.1), so future analyses may consider more stringent LD thresholds or pruning methods. Finally, multiple comparisons were made in our analysis. When we amended our significance threshold using a Bonferroni correction, our primary finding regarding HDL-C and increased breast cancer risk remained unaltered.

As increasing levels of HDL-C are generally thought to be healthier, an association with increased breast cancer risk was somewhat surprising. Although we could identify potential mechanisms in support of this finding from the literature, additional research to elucidate underlying factors is needed. For example, if lipid trait associations vary by sex, then using sex-specific weights would be preferable. Similarly, associations with breast cancer risk may vary by clinical characteristics, such as stage or grade of disease; future studies should be undertaken to address these possibilities. Additional future directions include a bidirectional Mendelian randomization analysis to test whether breast cancer risk GWAS variants are associated with lipid traits, and mediation analysis to evaluate whether covariates such as BMI or smoking are confounding or mediating factors in the relationship between lipids and breast cancer risk.

Our results suggest that increased HDL-C levels are associated with a 12% increased risk of breast cancer. Given contradictory evidence in terms of the beneficial effects of modifying HDL-C to affect cardiovascular disease risk, our findings provide some additional support against the broad use of therapeutic approaches to increase HDL-C in the general population. Instead, our results may be most useful for precision medicine, such as identifying women at increased risk of breast cancer as predicted by HDL-C-associated SNPs. In addition, although Mendelian randomization analysis may provide evidence for causality, interpreting our results as causal is not recommended; it is impossible to confirm that all three assumptions for Mendelian randomization assumptions were met, or that our findings were not influenced by pleiotropy.

In conclusion, our Mendelian randomization analysis of circulating lipids demonstrated that genetically predicted levels of increasing HDL-C were associated with increased breast cancer risk. Given the strong methodology used in this study, our results may help to clarify the inconsistencies observed across previous conventional observational studies and to support the hypothesis that circulating lipids may influence breast cancer risk.

Funding

This work was supported by a National Cancer Institute at the National Institutes of Health research grant (R37CA070867) and endowment funds for the Ingram Professorship and Anne Potter Wilson Chair. N.K.K. was supported by a National Institutes of Health training grant (R25CA160056). BCAC is funded by Cancer Research UK (C1287/A16563, C1287/A12014) and by the European Communitýs Seventh Framework Programme under grant agreement number 223175 (grant number HEALTH-F2-2009–223175) (COGS). Funding for the iCOGS infrastructure came from: the European Community's Seventh Framework Programme under grant agreement n° 223175 (HEALTH-F2-2009–223175) (COGS), Cancer Research UK (C1287/A10118, C1287/A10710, C12292/A11174, C1281/A12014, C5047/A8384, C5047/A15007, C5047/A10692, C8197/A16565), the National Institutes of Health (CA128978) and Post-Cancer GWAS initiative (1U19 CA148537, 1U19 CA148065 and 1U19 CA148112 - the GAME-ON initiative), the Department of Defense (W81XWH-10–1-0341), the Canadian Institutes of Health Research (CIHR) for the CIHR Team in Familial Risks of Breast Cancer, Komen Foundation for the Cure, the Breast Cancer Research Foundation, and the Ovarian Cancer Research Fund. Genotyping of the OncoArray was funded by the NIH Grant U19 CA148065, and Cancer UK Grant C1287/A16563 and the PERSPECTIVE project supported by the Government of Canada through Genome Canada and the Canadian Institutes of Health Research (grant GPH-129344) and, the Ministère de l’Économie, Science et Innovation du Québec through Genome Québec and the PSRSIIRI-701 grant, and the Quebec Breast Cancer Foundation. The DRIVE Consortium was funded by U19 CA148065. The Australian Breast Cancer Family Study (ABCFS) was supported by the National Institutes of Health (USA) (UM1 CA164920). The content of this manuscript does not necessarily reflect the views or policies of the National Institutes of Health or any of the collaborating centres in the Breast Cancer Family Registry (BCFR), nor does mention of trade names, commercial products or organizations imply endorsement by the USA Government or the BCFR. The ABCFS was also supported by the National Health and Medical Research Council of Australia, the New South Wales Cancer Council, the Victorian Health Promotion Foundation (Australia) and the Victorian Breast Cancer Research Consortium. J.L.H. is a National Health and Medical Research Council (NHMRC) Australia Fellow and M.C.S. is a NHMRC Senior Research Fellow. The ABCS study was supported by the Dutch Cancer Society (grants NKI 2007–3839; 2009 4363) and BBMRI-NL, which is a Research Infrastructure financed by the Dutch government (NWO 184.021.007). The work of the BBCC was partly funded by ELAN-Fond of the University Hospital of Erlangen. The BBCS is funded by Cancer Research UK and Breakthrough Breast Cancer and acknowledges NHS funding to the NIHR Biomedical Research Centre, and the National Cancer Research Network (NCRN). E.J.S. (BIGGS) is supported by NIHR Comprehensive Biomedical Research Centre, Guy's & St Thomas' NHS Foundation Trust in partnership with King's College London, UK; I.T. is supported by the Oxford Biomedical Research Centre. The BSUCH study was supported by the Dietmar-Hopp Foundation, the Helmholtz Society and the German Cancer Research Center (DKFZ). The CECILE study was funded by Fondation de France, Institut National du Cancer (INCa), Ligue Nationale contre le Cancer, Agence Nationale de Sécurité Sanitaire (ANSES) and Agence Nationale de la Recherche (ANR). The CGPS was supported by the Chief Physician Johan Boserup and Lise Boserup Fund, the Danish Medical Research Council and Herlev Hospital. The CNIO-BCS was supported by the Spanish Network on Rare Diseases (CIBERER) and grants from the Asociación Española Contra el Cáncer and the Fondo de Investigación Sanitario (PI11/00923 and PI12/00070). The Human Genotyping-CEGEN Unit (CNIO) is supported by the Instituto de Salud Carlos III. The CTS was initially supported by the California Breast Cancer Act of 1993 and the California Breast Cancer Research Fund (contract 97–10500) and is currently funded through the National Institutes of Health (R01 CA77398). Collection of cancer incidence data was supported by the California Department of Public Health as part of the statewide cancer reporting programme mandated by California Health and Safety Code Section 103885. H.A.C. receives support from the Lon V Smith Foundation (LVS39420). The ESTHER study was supported by a grant from the Baden Württemberg Ministry of Science, Research and Arts. Additional cases were recruited in the context of the VERDI study, which was supported by a grant from the German Cancer Aid (Deutsche Krebshilfe). The GC-HBOC was supported by Deutsche Krebshilfe (107 352). The GENICA was funded by the Federal Ministry of Education and Research (BMBF) Germany grants 01KW9975/5, 01KW9976/8, 01KW9977/0 and 01KW0114, the Robert Bosch Foundation, Stuttgart, Deutsches Krebsforschungszentrum (DKFZ), Heidelberg, Institute for Prevention and Occupational Medicine of the German Social Accident Insurance and Institute of the RuhrUniversity Bochum (IPA), as well as the Department of Internal Medicine, Evangelische Kliniken Bonn gGmbH, Johanniter Krankenhaus Bonn, Germany. HEBCS was supported by the Helsinki University Central Hospital Research Fund, Academy of Finland (266528), the Finnish Cancer Society, and the Sigrid Juselius Foundation. The HMBCS was supported by the Rudolf Bartling Foundation. Financial support for KARBAC was provided through the regional agreement on medical training and clinical research (ALF) between Stockholm County Council and Karolinska Institutet, the Swedish Cancer Society, the Gustav V Jubilee foundation and Bert von Kantzows Foundation. The KBCP was financially supported by the special Government Funding (EVO) of Kuopio University Hospital grants, Cancer Fund of North Savo, the Finnish Cancer Organizations, the Academy of Finland and by the strategic funding of the University of Eastern Finland. kConFab is supported by a grant from the National Breast Cancer Foundation, and previously by the National Health and Medical Research Council (NHMRC), the Queensland Cancer Fund, the Cancer Councils of New South Wales, Victoria, Tasmania and South Australia and the Cancer Foundation of Western Australia. LMBC is supported by the ‘Stichting tegen Kanker’ (232–2008 and 196–2010); D.L.s is supported by the FWO and the KULPFV/10/016-SymBioSysII and by an ERC consolidator grant. The MARIE study was supported by the Deutsche Krebshilfe e.V. (70–2892-BR I, 106332, 108253, 108419), the Hamburg Cancer Society, the German Cancer Research Center (DKFZ), and the Federal Ministry of Education and Research (BMBF) Germany (01KH0402). MBCSG is supported by grants from the Italian Association for Cancer Research (AIRC) and by funds from the Italian citizens who allocated the 5/1000 share of their tax payment in support of the Fondazione IRCCS Istituto Nazionale Tumori, according to Italian laws (INT-Institutional strategic projects ‘5x1000’). The MCBCS was supported by National Institutes of Health grants (CA116167, CA176785, CA192393) a National Institutes of Health Specialized Program of Research Excellence (SPORE) in Breast Cancer (CA116201), the Breast Cancer Research Foundation and a generous gift from the David F. and Margaret T. Grohne Family Foundation. MCCS cohort recruitment was funded by VicHealth and Cancer Council Victoria. The MCCS was further supported by Australian NHMRC grants 209057, 251553 and 504711 and by infrastructure provided by Cancer Council Victoria. The MEC was support by National Institutes of Health grants (CA63464, CA54281, CA098758 and CA132839). The work of MTLGEBCS was supported by the Quebec Breast Cancer Foundation, the Canadian Institutes of Health Research for the ‘CIHR Team in Familial Risks of Breast Cancer’ program (grant # CRN-87521) and the Ministry of Economic Development, Innovation and Export Trade (grant # PSR-SIIRI-701). The NBCS was supported by grants from the Norwegian Research council, (155218/V40, 175240/S10 to A.L.B.D., FUGE-NFR 181600/V11 to V.N.K. and a Swizz Bridge Award to ALBD). The NBHS was supported by National Institutes of Health (R01CA100374); biological sample preparation was conducted by the Survey and Biospecimen Shared Resource, which is supported by National Institutes of Health (P30 CA68485). The OBCS was supported by research grants from the Finnish Cancer Foundation, the Academy of Finland (grant number 250083, 122715 and Center of Excellence grant number 251314), the Finnish Cancer Foundation, the Sigrid Juselius Foundation, the University of Oulu, the University of Oulu Support Foundation and the special Governmental EVO funds for Oulu University Hospital-based research activities. The OFBCR was supported by National Institutes of Health (UM1 CA164920). The content of this manuscript does not necessarily reflect the views or policies of the National Institutes of Health or any of the collaborating centres in the Breast Cancer Family Registry (BCFR), nor does mention of trade names, commercial products or organizations imply endorsement by the US Government or the BCFR. The ORIGO study was supported by the Dutch Cancer Society (RUL 1997–1505) and the Biobanking and Biomolecular Resources Research Infrastructure (BBMRI-NL CP16). The PBCS was funded by Intramural Research Funds of the National Cancer Institute, Department of Health and Human Services, USA. The pKARMA study was supported by Märit and Hans Rausings Initiative Against Breast Cancer. The RBCS was funded by the Dutch Cancer Society (DDHK 2004–3124, DDHK 2009–4318). The SASBAC study was supported by funding from the Agency for Science, Technology and Research of Singapore (A*STAR), the US National Institute of Health and the Susan G Komen Breast Cancer Foundation. The SBCS was supported by Sheffield Experimental Cancer Medicine Centre and the Breast Cancer Now Tissue Bank. SEARCH is funded by Cancer Research UK (C490/A10124, C490/A16561) and supported by the UK National Institute for Health Research Biomedical Research Centre at the University of Cambridge. The University of Cambridge has received salary support for P.D.P.P. from the NHS in the East of England through the Clinical Academic Reserve. SKKDKFZS is supported by the DKFZ. The SZBCS was supported by Grant PBZ_KBN_122/P05/2004. The TNBCC was supported by: a Specialized Program of Research Excellence (SPORE) in Breast Cancer (CA116201), a grant from the Breast Cancer Research Foundation, a generous gift from the David F and Margaret T Grohne Family Foundation, the Hellenic Cooperative Oncology Group research grant (HR R_BG/04) and the Greek General Secretary for Research and Technology (GSRT) Program, Research Excellence II, the European Union (European Social Fund – ESF), and Greek national funds through the Operational Program ‘Education and Lifelong Learning’ of the National Strategic Reference Framework (NSRF) - ARISTEIA. The UKBGS is funded by Breast Cancer Now and the Institute of Cancer Research (ICR), London. I.C.R. acknowledges NHS funding to the NIHR Biomedical Research Centre.