The relationship between circulating lipids and breast cancer risk: A Mendelian randomization study

Kelsey E Johnson; Katherine M Siewert; Derek Klarin; Scott M Damrauer; the VA Million Veteran Program; Kyong-Mi Chang; Philip S Tsao; Themistocles L Assimes; Kara N Maxwell; Benjamin F Voight

doi:10.1371/journal.pmed.1003302

. 2020 Sep 11;17(9):e1003302. doi: 10.1371/journal.pmed.1003302

The relationship between circulating lipids and breast cancer risk: A Mendelian randomization study

Kelsey E Johnson ^1,^#, Katherine M Siewert ^2,^#, Derek Klarin ^3,^4,⁵, Scott M Damrauer ^6,⁷; the VA Million Veteran Program^¶, Kyong-Mi Chang ^6,⁸, Philip S Tsao ^9,¹⁰, Themistocles L Assimes ^9,¹⁰, Kara N Maxwell ^8,¹¹, Benjamin F Voight ^6,^11,^12,^13,^*

Editor: Cosetta Minelli¹⁴

¹Cell and Molecular Biology Graduate Group, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States of America

²Genomics and Computational Biology Graduate Group, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States of America

³Boston VA Healthcare System, Boston, Massachusetts, United States of America

⁴Center for Genomic Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, United States of America

⁵Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, Massachusetts, United States of America

⁶Corporal Michael Crescenz VA Medical Center, Philadelphia, Pennsylvania, United States of America

⁷Department of Surgery, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States of America

⁸Department of Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States of America

⁹VA Palo Alto Health Care System, Palo Alto, California, United States of America

¹⁰Department of Medicine, Stanford University School of Medicine, Stanford, California, United States of America

¹¹Department of Genetics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States of America

¹²Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States of America

¹³Institute for Translational Medicine and Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States of America

¹⁴Imperial College London, UNITED KINGDOM

I have read the journal's policy and the authors of this manuscript have the following competing interests: SMD declares research support to institution from Renalytix AI and a patent application filed by VA on drug repurposing for lipid reduction.

¶ Membership of the VA Million Veteran Program is provided in S1 Text.

^✉

* E-mail: bvoight@pennmedicine.upenn.edu

Contributed equally.

Roles

Kelsey E Johnson: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing

Katherine M Siewert: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing

Derek Klarin: Data curation, Writing – review & editing

Scott M Damrauer: Data curation, Writing – review & editing

Kyong-Mi Chang: Data curation, Writing – review & editing

Philip S Tsao: Data curation, Writing – review & editing

Themistocles L Assimes: Data curation, Writing – review & editing

Kara N Maxwell: Conceptualization, Funding acquisition, Project administration, Supervision, Writing – original draft, Writing – review & editing

Benjamin F Voight: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Visualization, Writing – original draft, Writing – review & editing

Cosetta Minelli: Academic Editor

PMCID: PMC7485834 PMID: 32915777

Abstract

Background

A number of epidemiological and genetic studies have attempted to determine whether levels of circulating lipids are associated with risks of various cancers, including breast cancer (BC). However, it remains unclear whether a causal relationship exists between lipids and BC. If alteration of lipid levels also reduced risk of BC, this could present a target for disease prevention. This study aimed to assess a potential causal relationship between genetic variants associated with plasma lipid traits (high-density lipoprotein, HDL; low-density lipoprotein, LDL; triglycerides, TGs) with risk for BC using Mendelian randomization (MR).

Methods and findings

Data from genome-wide association studies in up to 215,551 participants from the Million Veteran Program (MVP) were used to construct genetic instruments for plasma lipid traits. The effect of these instruments on BC risk was evaluated using genetic data from the BCAC (Breast Cancer Association Consortium) based on 122,977 BC cases and 105,974 controls. Using MR, we observed that a 1-standard–deviation genetically determined increase in HDL levels is associated with an increased risk for all BCs (HDL: OR [odds ratio] = 1.08, 95% confidence interval [CI] = 1.04–1.13, P < 0.001). Multivariable MR analysis, which adjusted for the effects of LDL, TGs, body mass index (BMI), and age at menarche, corroborated this observation for HDL (OR = 1.06, 95% CI = 1.03–1.10, P = 4.9 × 10⁻⁴) and also found a relationship between LDL and BC risk (OR = 1.03, 95% CI = 1.01–1.07, P = 0.02). We did not observe a difference in these relationships when stratified by breast tumor estrogen receptor (ER) status. We repeated this analysis using genetic variants independent of the leading association at core HDL pathway genes and found that these variants were also associated with risk for BCs (OR = 1.11, 95% CI = 1.06–1.16, P = 1.5 × 10⁻⁶), including locus-specific associations at ABCA1 (ATP Binding Cassette Subfamily A Member 1), APOE-APOC1-APOC4-APOC2 (Apolipoproteins E, C1, C4, and C2), and CETP (Cholesteryl Ester Transfer Protein). In addition, we found evidence that genetic variation at the ABO locus is associated with both lipid levels and BC. Through multiple statistical approaches, we minimized and tested for the confounding effects of pleiotropy and population stratification on our analysis; however, the possible existence of residual pleiotropy and stratification remains a limitation of this study.

Conclusions

We observed that genetically elevated plasma HDL and LDL levels appear to be associated with increased BC risk. Future studies are required to understand the mechanism underlying this putative causal relationship, with the goal of developing potential therapeutic strategies aimed at altering the cholesterol-mediated effect on BC risk.

In a study of genetic data, Kelsey Johnson, Katherine Siewert, and colleagues use Mendelian randomization techniques to investigate the relationship between genetic variants associated with plasma lipid traits and risk for breast cancer.

Author summary

Why was this study done?

An individual’s lipid levels may affect their risk of breast cancer. However, previous studies disagree on whether a causal effect exists.
Mendelian randomization methods allow researchers to test whether genetically influenced lipid levels are associated with risk of breast cancer.

What did the researchers do and find?

We tested whether genetic variants that are associated with changes in lipid levels also have consistent associations with breast cancer.
We found that both high and low-density lipoprotein cholesterol (HDL and LDL) are associated with an increased risk of breast cancer.

What do these findings mean?

The techniques used in this study cannot rule out that our findings are due to the lipid-associated genetic variants being associated with breast cancer risk through mechanisms other than cholesterol level.
Further research will be needed to investigate the possibility that manipulation of LDL or HDL levels can influence risk of breast cancer.

Introduction

Breast cancer (BC) is the second leading cause of death for women, motivating the need for a better understanding of its etiology and more effective treatments [1]. Cholesterol is a known risk factor for multiple diseases that have reported associations with BC, including obesity, heart disease, and diabetes [2]. However, it is unknown whether cholesterol plays a causal role in BC susceptibility.

The body of epidemiological and clinical trial studies to date has yet to determine whether there is a causal relationship between cholesterol and BC. Observational epidemiological studies have reported positive, negative, or no relationship between lipid levels and BC risk [3–6]; however, these studies can suffer from confounding. A comprehensive meta-analysis found evidence that statin use may reduce BC risk [7], and cholesterol-lowering medications have been associated with improved outcomes in BC patients on hormonal therapy, suggesting an interaction of circulating cholesterol levels with estrogen-sensitive breast tissues [8]. These mixed findings motivate the need for a high-powered causal inference analysis of lipids on BC.

To try to resolve these discrepancies, recent studies have applied the framework of Mendelian randomization (MR) to determine whether genetically elevated lipid levels associate with BC risk. In a small sample of 1,187 BC cases, Orho-Melander and colleagues used multivariable MR to find suggestive evidence of a relationship between both triglycerides and HDL (high-density lipoprotein) cholesterol and BC, but no association between LDL (low-density lipoprotein) cholesterol and cancer [9]. In a second study, Nowak and colleagues [10] performed an MR analysis with genetic association data from large genome-wide association studies (GWASs) for lipids and BC [11,12]. They reported nominal positive associations between LDL-cholesterol levels and all BCs and between HDL-cholesterol levels and ER (estrogen receptor)-positive BCs. While compelling, this study also had limitations. First, they used relatively few variants in their genetic instrument because of the removal of pleiotropic variants in order to address confounding due to pleiotropy, resulting in a conservative analysis. Second, they analyzed each lipid trait separately rather than take advantage of multivariable methods to consider lipid traits together along with additional, potentially confounding causal risk factors. Third, the authors did not quantitatively assess heterogeneity to determine whether the observed lipid associations were statistically different across BC subtypes. Another recent study by Qi and Chatterjee applied a newly developed MR method and reported an association between HDL-cholesterol and BC that they defined as borderline statistically significant [13]. Like Nowak and colleagues, this paper also does not explicitly include correlated risk factors in their analysis and did not stratify BC by ER status.

These studies motivate an MR study that considers multiple lipid traits concurrently to delineate the independent effect of each lipid trait on BC susceptibility. Such an approach obviates the need to remove pleiotropic variants and the loss of statistical power that results from this removal. Therefore, an approach that considers the effects of genetic variants on known risk factors for BC, such as body mass index (BMI) and age at menarche [14–20], could increase power.

While MR assesses evidence for a causal relationship, genome-wide genetic correlation analysis determines whether 2 traits simply have a shared genetic basis. Local genetic correlation analyses test whether 2 traits have shared heritability that is localized to specific genomic regions. These loci may then harbor causal variants and genes that contribute to heritability of both traits. Jiang and colleagues recently estimated genome-wide genetic correlation between lipid traits and BC risk [21]. This study did not find a statistically significant association between any lipid trait and BC using lipid summary statistics, though a previous study with a smaller BC GWAS sample size did report a nominally significant (P < 0.05) negative genetic correlation between triglycerides and BC risk [16]. Both these studies used the same method to estimate genome-wide genetic correlations [15], and neither tested for local genetic correlations between lipid traits and BC risk.

In what follows, we apply the causal inference framework of MR to determine whether genetically elevated lipid traits modify BC susceptibility (both all BC and BC stratified by ER status). We take advantage of a recent GWAS for lipid levels performed in up to 215,551 individuals of European ancestry [22], which has not been previously applied to MR studies of BC, to provide power for our causal inference analyses. We utilize several MR techniques, including single-exposure, multivariable, and gene-specific approaches. Of chief concern in modern MR studies, including prior studies of BC and lipids, is confounding due to pleiotropy. For instance, a genetic variant may affect lipid levels indirectly through some other biomarker. If this biomarker directly affects BC risk, this could confound the MR analysis and cause an incorrect inference of a causal effect of lipid levels on BC. Our gene-specific approach utilizes only genetic variants near core HDL pathway genes to minimize this concern. Additionally, we use a multivariable approach that assesses the effects of lipid traits independent of one another and of BMI and age at menarche. Finally, we perform genetic correlation analyses to look for both genome-wide and locus-based correlation in effect sizes between lipids and BC.

Methods

Study populations

Lipid GWAS summary statistics were obtained from the Million Veteran Program (MVP) (up to 215,551 European individuals) [22] and the Global Lipids Genetics Consortium (GLGC) (up to 188,577 genotyped individuals) [12]. As additional exposures in multivariable MR analyses, we used BMI summary statistics from a meta-analysis of GWASs in up to 795,640 individuals and age at menarche summary statistics from a meta-analysis of GWASs in up to 329,345 women of European ancestry [17,23]. GWAS summary statistics from 122,977 BC cases and 105,974 controls were obtained from the Breast Cancer Association Consortium (BCAC) [11]. The MVP received ethical and study protocol approval from the Veteran Affair Central Institutional Review Board in accordance with the principles outlined in the Declaration of Helsinki, and written consent was obtained from all participants. For the Willer and colleagues [12] and BCAC [11] data sets, we refer the reader to the primary GWAS manuscripts and their supplementary material for details on consent protocols for each of their respective cohorts. More details on these cohorts are in the S1 Text.

Lipid meta-analysis

We performed a fixed-effects meta-analysis between each lipid trait (Total cholesterol [TC], LDL, HDL, and triglycerides [TGs]) in GLGC and the corresponding lipid trait in the MVP cohort [12,22] using the default settings in PLINK [24]. There is some genomic inflation in these meta-analysis association statistics, but linkage disequilibrium (LD)-score regression intercepts demonstrate that this inflation is in large part due to polygenicity and not population stratification (S1 Fig).

MR analyses

MR analyses were performed using the TwoSampleMR R package version 0.4.13 (https://github.com/MRCIEU/TwoSampleMR) [25]. For all analyses, we used a 2-sample MR framework, with exposure(s) (lipids, BMI, age at menarche) and outcome (BC) genetic associations from separate cohorts. Unless otherwise noted, MR results reported in this manuscript used inverse-variance weighting assuming a multiplicative random effects model. For single-trait MR analyses, we additionally employed Egger regression [26], weighted median [27], and mode-based [28] estimates. SNPs associated with each lipid trait were filtered for genome-wide significance (P < 5 × 10⁻⁸) from the MVP lipid study [12], and then we removed SNPs in LD (r² < 0.001 in UK10K consortium) [29] in order to obtain independent variants. All genetic variants were harmonized using the TwoSampleMR harmonization function with default parameters. Each of these independent, genome-wide significant SNPs was termed a genetic instrument. We estimated that these single-trait MR genetic instruments had 80% power to reject the null hypothesis, with a 1% error rate, for the following odds ratio (OR) increases in BC risk due to a standard deviation increase in lipid levels: HDL, 1.057; LDL, 1.058; TGs, 1.055; TC, 1.060 [30,31]. We tested for directional pleiotropy using the MR-Egger regression test [26]. To reduce heterogeneity in our genetic instruments for single-trait MR, we employed a pruning procedure (S1 Text). Genetic instruments used in single-trait MR are listed in S1 Table. For multivariable MR experiments [32,33], we generated genetic instruments by first filtering the genotyped variants for those present across all data sets. For each trait and data set combination (Yengo and colleagues [23] for BMI; Day and colleagues for age at menarche [17]; MVP and GLGC for HDL, LDL, and TGs), we then filtered for genome-wide significance (P < 5 × 10⁻⁸) and for linkage disequilibrium (r² < 0.001 in UK10K consortium) [29]. We performed tests for instrument strength and validity [34], and each multivariable MR experiment had sufficient instrument strength. We removed variants driving heterogeneity in the ratio of outcome/exposure effects causing instrument invalidity (S1 Text). Genetic instruments used in multivariable MR are listed in S2 Table. Because the MR methods and tests we employed are highly correlated, we did not apply a multiple testing correction to the reported P-values.

Core HDL and LDL pathway genetic instrument development

We defined sets of core genes for HDL or LDL that met the following criteria: (1) their protein products are known to play a key role in HDL or LDL biology (plus HMGCR and NPC1L1, 2 targets of LDL-lowering drugs, in the LDL gene set), and (2) there were conditionally independent lipid trait-associated variants within 100 kb upstream or downstream of the RefSeq coordinates for the gene (or locus, in the case of Apolipoprotein E [APOE]-Apolipoprotein C [APOC]1-APOC4-APOC2 and Apolipoprotein A [APOA]4-APOC3-APOA1) [22]. We then used the conditional HDL or LDL association statistics from Klarin and colleagues for those genes in gene-specific MR analyses [22]. The loci included in each set and the genetic instruments used in each locus-specific MR are listed in S3 Table. We performed a separate fixed-effects inverse-variance weighted MR with the conditionally independent genetic instruments at each gene and performed fixed-effects inverse-variance weighted meta-analysis of the results across HDL or LDL genes using the R package meta [35].

Genetic correlation analyses

We performed cross-trait LD-score regression using the LDSC toolkit, available at https://github.com/bulik/ldsc, with default parameters [15], with the BCAC association statistics for BC and our meta-analysis of GLGC and MVP for lipid associations. In addition, we used the ρ-Hess software both for whole-genome genetic correlation and for local genetic correlation analysis [36], using the UK10K reference panel, and the LD-independent loci published in Berisa and colleagues to partition the genome [37]. We used a Bonferroni significance threshold based on the number of these independent loci (1,704 loci). There was minor cohort overlap between the GLGC and BC GWASs because of the EPIC cohort [10]. We included this overlap when performing ρ-Hess, available at https://huwenboshi.github.io/hess/, using the cross-trait LD-score intercept to estimate phenotypic correlation [36]. The association of the lead BC and lipid SNPs at the ABO locus was obtained using the GTEx v8 data set [38].

Analysis plan

Our study did not develop a prospective analysis plan. We began by testing for a potential causal relationship between lipids and BC risk using single-trait two-sample MR with lipid genetic associations from the GLGC. After this experiment showed a significant relationship, we tested whether it persisted when we corrected for correlated phenotypes with multivariable MR. Following significant results with the GLGC summary statistics, we decided to confirm these findings using genetic associations from the larger MVP cohort. After our results with MVP confirmed our initial findings, we performed additional MR sensitivity analyses and locus-specific MR. In parallel to our MR experiments, we measured the cross-trait and local genetic correlations between these traits. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) and STROBE-MR guidelines (S1 STROBE Checklist, S1 STROBE-MR Checklist) [39,40].

Results

Single-trait MR in BC

We first performed single-trait MR analyses using summary statistics from MVP [22] for each of 4 lipid traits (i.e., TC, HDL-cholesterol, LDL-cholesterol, and TGs) as the intermediate biomarkers and risk for all BCs as the outcome (S2 Fig). We observed a significant relationship between genetically elevated HDL and BC risk (OR = 1.10 per standard deviation of lipid level increase, 95% confidence interval [CI] = 1.04–1.17, P = 2.1 x 10⁻³) and genetically decreased TG levels and BC risk (OR = 0.93, 95% CI = 0.88–0.99, P = 0.015; S4 Table). Sensitivity analyses identified heterogeneity (Methods, S5 Table), but there was no evidence of bias from directional pleiotropy (Methods, S6 Table). To mitigate concerns of instrument heterogeneity, we removed variants from our genetic instrument for each lipid trait that were responsible for instrument heterogeneity (S1 Text) and again observed a relationship with HDL-cholesterol (OR = 1.08, 95% CI = 1.04–1.13, P = 7.4 × 10⁻⁵) and TGs (OR = 0.94, 95% CI = 0.90–0.98, P = 2.6 × 10⁻³) (Fig 1, S3 and S4 Figs, S7 Table). Because HDL and TGs are inversely correlated [15,41], the opposing relationship between these 2 lipid traits and BC could be expected in single-trait analyses.

Fig 1 — Results plotted are after pruning for instrument heterogeneity. The forest plot on the right displays the OR of the effect of a 1-standard–deviation increase in genetically determined HDL-cholesterol on BC risk as a diamond, with error bars representing the 95% CI. The vertical dotted line delineates an OR of 1, i.e., no effect of the exposure on BC risk. BC, breast cancer; BMI, body mass index; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted MR; LDL, low-density lipoprotein; MR, Mendelian randomization; MVMR, multivariable MR; OR, odds ratio; TG, triglyceride.

To confirm that our results using lipid genetic associations from MVP were not due to heterogeneity between data sets, we also tested the relationship between lipid traits and BC using a meta-analysis of the 2 major lipid GWASs from MVP and GLGC and from GLGC alone. Overall, single-trait MR analyses with the meta-analysis and GLGC lipid associations produced consistent results to those with MVP alone (S5 Fig). In a reciprocal single-trait MR testing the effect of genetically determined BC risk on each lipid trait, we observed no relationship with HDL- or LDL-cholesterol (S8 Table) but did see a relationship with TGs. However, a Steiger test for directionality confirmed that using BC as the outcome was the correct causal direction for all lipid traits (S7 Table) [42]. We also performed genetic instrument pruning in the same manner as Nowak and colleagues: removing genetic instruments for LDL, HDL, and TGs that were associated with at least one of the 2 other lipid traits (P < 0.001) [10]. After this pruning, we did not find a significant relationship with LDL, HDL, or TG, and we note that this pruning procedure resulted in considerably larger CIs spanning OR = 1 for all traits, with reversed direction of effect estimates for LDL and TGs (S9 Table).

Multivariable MR with age at menarche and BMI as exposures

It has been previously observed that BMI and age at menarche are both genetically correlated and epidemiologically associated with both BC [20,43,44] and lipid traits [15,41]. To incorporate these potential confounders into our causal inference framework, we performed multivariable MR analyses using all 3 lipid traits (genetic effect estimates from MVP), age at menarche, and BMI as exposures and BC risk as the outcome (Fig 1). We observed significant relationships between genetically influenced HDL, LDL, BMI, and age at menarche with BC (HDL: OR = 1.06, 95% CI = 1.03–1.10, P = 4.93 × 10⁻⁴; LDL: OR = 1.04, 95% CI = 1.01–1.07, P = 0.02; BMI: OR = 0.90, 95% CI = 0.87–0.94, P = 1.15 × 10⁻⁶; age at menarche: OR = 0.96, 95% CI = 0.93–0.99, P = 2.44 × 10⁻³), but not TGs (OR = 0.98, 95% CI = 0.95–1.00, P = 0.10) (Fig 1, S10 Table). Our results were consistent before and after pruning for genetic instrument heterogeneity (S10 Table) and when using summary statistics from 3 independent subsets of the BC data set (S6 Fig, S10 Table). We also performed multivariable MR with pairs of lipid traits with genetic effect estimates from different data sets (GLGC or MVP), with and without BMI, and saw consistent results (S11 Table, S7 Fig). Considering the genetic correlation between HDL and TGs, the greater significance of the HDL association compared with the TG association with BC in multivariable analysis, and the consistent relationship between HDL and BC across BC data sets, we focused our further MR analyses on the relationship between HDL-cholesterol and BC, in addition to the previously reported association between LDL and BC [10].

MR with outcome stratified by ER status

We next performed an MR analysis of the relationship between genetically influenced lipids and BC risk stratified by ER-positive (ER+) or negative (ER‒) status. We observed similar effect size estimates of the 4 lipid traits on the BC subtypes as on BC not stratified by subtype (S8 Fig). A formal test for heterogeneity found no evidence to reject the null hypothesis of homogeneity between the cancer subtypes (e.g., HDL: Cochran’s Q = 6.6 × 10⁻⁵, P = 0.99; S12 Table). Thus, we observed no substantive difference in the relationship from any lipid trait to ER+ or ER− BCs, consistent with the strong genetic correlation between these 2 BC subtypes (cross-trait LD-score regression genetic correlation estimate = 0.62, P = 2.9 × 10⁻⁸³). When we used ER+ or ER‒ BCs as the outcome in multivariable MR, we also saw consistent effects as the analysis with all BCs as the outcome (S13 Table, S9 Fig).

HDL and LDL pathway gene-specific MR

We next examined associations for BC risk at genetic variants near core HDL or LDL genes. We identified conditionally independent associations at genes that were previously annotated with a core role in the metabolism of each lipid trait or an established drug target (HDL: ATP Binding Cassette Subfamily A Member 1 [ABCA1], APOA4-APOC3-APOA1, APOE-APOC1-APOC4-APOC2, Cholesteryl Ester Transfer Protein [CETP], Lecithin-Cholesterol Acyltransferase [LCAT], Lipase C Hepatic Type [LIPC], Lipase G Endothelial Type [LIPG], Phospholipid Transfer Protein [PLTP], Scavenger Receptor Class B Member 1 [SCARB1]; LDL: Apolipoprotein B [APOB], 3-Hydroxy-3-Methylglutaryl-CoA Reductase [HMGCR], LDL Receptor [LDLR], Lipoprotein(A) [LPA], Myosin Regulatory Light Chain Interacting Protein [MYLIP], NPC1-Like Intracellular Cholesterol Transporter 1 [NPC1L1], Proprotein Convertase Subtilisin/Kexin Type 9 [PCSK9]) (Methods, S3 Table). For each gene or locus with at least 2 conditionally independent genetic instruments (all except LCAT and MYLIP), we performed inverse-variance-weighted MR (fixed-effects model) with conditional HDL or LDL effect size estimates as the exposure and BC risk as the outcome (S10 and S11 Figs). We observed significant (P < 0.05) positive relationships between HDL and BC risk at 3 loci (ABCA1, APOE-APOC1-APOC4-APOC2, and CETP; Fig 2), and between LDL and BC risk at 1 locus (HMGCR, S12 Fig). Combining the effect estimates across core genes in a meta-analysis, we observed a significant positive relationship for HDL (OR = 1.11, 95% CI = 1.06–1.16, P < 0.001; Fig 2) and LDL (OR = 1.07, 95% CI = 1.01–1.14, P = 0.02; S12 Fig). There was no evidence of heterogeneity across loci in either meta-analysis (HDL: Q = 6.63, P = 0.47; LDL: Q = 5.53, P = 0.35).

Fig 2 — The forest plot on the right displays the OR of the effect of a 1-standard–deviation increase in genetically determined HDL-cholesterol for each locus on BC risk as a diamond, and the error bars represent the 95% CI of the effect estimate. The vertical dotted line delineates an OR of 1, i.e., no effect of the exposure on BC risk. For HDL gene-specific instruments, see S3 Table. *ABCA1*, ATP Binding Cassette Subfamily A Member 1; *APOC*, Apolipoprotein C; *APOE*, Apolipoprotein E; BC, breast cancer; *CETP*, Cholesteryl Ester Transfer Protein; CI, confidence interval; HDL, high-density lipoprotein; *LIPC*, Lipase C, Hepatic Type; LIPG, Lipase G, Endothelial Type; MR, Mendelian randomization; N SNPs, number of genetic instruments included in each locus’s MR analysis; OR, odds ratio; *PLTP*, Phospholipid Transfer Protein; *SCARB1*, Scavenger Receptor Class B Member 1.

Genome-wide and local genetic correlation

If cholesterol levels were a causal risk factor for BC, we might expect a correlation between the strength of genetic association with these 2 traits at genetic variants across the entire genome in addition to those at genome-wide significant loci. To answer this question, we utilized 2 approaches to estimate genetic correlation between BC and lipid traits. Using the ρ-Hess method, we found significant (P < 0.05) correlations between LDL (P < 0.001) and TC (P = 0.01) and BC, with directions consistent with our MR results (S14 Table). Cross-trait LD-score regression found positive genetic correlation estimates for TC, LDL, and HDL and a negative estimate for TGs (S13 Fig, S14 Table), consistent with our MR and ρ-Hess results [15]. However, the only significant association (P < 0.05) was with TC and ER-negative BC (P = 0.04).

To discover new loci that are enriched for genetic correlation between BC and lipids, we used the ρ-Hess method, which detects genomic regions harboring high genetic correlation between 2 traits [36]. ρ-Hess identified one region that surpassed Bonferroni test correction, with a positive correlation between both LDL and TC and BC (S15 Table). In this region, there are 2 SNPs in high LD (rs532436 and rs635634, r² = 0.99) that are genome-wide significantly associated with LDL (rs532436: P < 0.001), TC (P < 0.001), and BC (P < 0.001). These SNPs lie within an intron of the ABO blood group determining ABO gene. rs635634 moderately tags an SNP associated with ABO blood type [41]. However, this SNP is also associated with a change in gene expression of ABO in multiple tissues (P < 0.001 in breast mammary tissue) [38].

Discussion

Using MR, we provide evidence that genetically elevated HDL and LDL levels are associated with increased risk for BC, supporting a causal hypothesis. Previous meta-analyses of observational studies of BC risk and lipids reported a negative association with HDL and no relationship with LDL [4,5], whereas individual studies have reported a positive relationship with HDL [6] or no relationship with HDL or LDL [45,46]. Our analyses help clarify these mixed results and infer a direction of effect, which is not possible in observational studies because of potential reverse causation. Furthermore, we find evidence of genome-wide genetic correlation between some lipid traits and BC and local genetic correlation at the ABO locus. Although some studies have found an association between blood group and BC risk [47], haplotype patterns indicate that ABO gene expression, not blood group, may be the causal mechanism [37]. However, because of the pleiotropic nature of the ABO locus, it is unclear whether the BC association is caused by the lipid associations [41].

Although Nowak and colleagues previously used MR to discover associations between lipids and BC [10], our report presents a thorough reconsideration of these effects. Even after conditioning on the effects of HDL, BMI, and age at menarche, our MR analysis suggests a potential causal relationship between LDL and BC. Nowak and colleagues only found a relationship between HDL-cholesterol and ER+ BC, whereas we found a relationship between HDL-cholesterol and risk for all BCs. We also find a previously unreported association with TGs and BC, though our multivariable analysis suggests this may be explained by correlation between TGs and HDL and not an independent TG effect. In their analyses, Nowak and colleagues used a strict pruning procedure to isolate the effects of each lipid trait. However, this approach reduces power because of the high genetic correlation of these traits. The multivariable approach taken here is an alternative way to estimate the effect of an exposure while accounting for correlated exposures.

Our results largely agree with those reported in the recent MR study of BC and lipids by Beeghly-Fadiel and colleagues [48], published while this manuscript was under peer review. Both studies use multiple types of MR analyses, including approaches accounting for confounding by BMI and age at menarche, and report a positive association between HDL and BC and a negative association between TGs and BC. However, our report provides complementary analysis and data that support the central findings of both pieces of work. First, we took advantage of the recently reported MVP lipids GWAS [22], providing a larger number of genetic instruments for all lipid traits considered. Second, we explicitly considered age of menarche and BMI in multivariable models with all lipid traits. It is crucial to consider the collection of each of these risk factors together to estimate a causal effect estimate that is independent of these confounders, as well as across cancer risk strata (i.e., ER status). While Beeghly-Fadiel and colleagues took advantage of access to individual-level data to adjust for confounding factors in their single lipid MR, these confounders were not corrected for in their multivariable MR. Third, while both studies are consistent in their relationship between HDL and BC, we reported a nominal association (P < 0.05) with LDL levels when considering all risk and confounding factors jointly. Finally, we present unique, locus-specific MR analyses to show that conditionally independent associations at single loci implicated in HDL or LDL biology are significantly associated with BC risk.

A challenge of MR analyses that use hundreds of genetic instruments is that we do not know the mechanism of action of these instruments on the exposure trait. By focusing on loci with mechanistic evidence of a direct effect on lipid levels, we can remove uncertainty about potential pleiotropic effects on BC risk. Thus, the significant relationships observed in our locus-specific MR analyses across HDL or LDL pathway genes provide additional evidence for a direct effect of increased HDL or LDL levels on BC risk. Furthermore, these genes represent potential or established drug targets, and each locus-specific MR experiment provides preliminary evidence for the therapeutic potential of cholesterol modification on BC prevention.

Substantial effort has been spent developing HDL-raising therapies for cardiovascular disease prevention; however, recent studies have proposed an increase in all-cause mortality in individuals with high HDL levels [49,50]. Our results suggest that therapies that aim to reduce cardiovascular risk by raising HDL levels might have an unintended consequence of elevated BC risk. Specifically, our gene-based score using HDL-raising variation at the CETP locus predicted that CETP-based inhibition would elevate BC risk (OR = 1.11, 95% CI = 1.04–1.18, P < 0.001). Additionally, 2 recent MR studies reported causal evidence between elevated HDL and risk for age-related macular degeneration [51,52]. These potential disease-increasing consequences may not have been possible to identify in safety trials, given the limited window of study to monitor progression or incidence of disease, the putative causal effect estimates, and the demographics of the study population (i.e., a higher proportion of male participants) [53]. Our result supports the use of human genetics data as both a novel strategy for therapeutic targeting and for the discovery of potential drug complications to direct long-term post-clinical–trial follow-up [54].

We note several caveats to our analyses. The first is that MR makes several assumptions that must be met for accurate causal inference [55,56]. Although we used statistical methods that try to detect and correct for violations of these assumptions, these methods are not guaranteed to correct for all types of confounding, and alternative causal inference frameworks outside of MR are warranted. Secondly, MR is only able to make inferences about trait associations in the populations from which the GWASs are derived. We are unaware of evidence that BC or lipid genetic architecture varies significantly across populations, but if this was the case, our findings may not be generalizable to these different scenarios. However, the concordance between our results using the MVP and GLGC GWASs mitigate this concern with regards to potential heterogeneity in lipid genetic architecture. Thirdly, the estimated lipid/BC effect sizes represent only the population-averaged causal effect and may not generalize well to other populations or settings [57]. We note that our effect estimates may be attenuated because of association of lipid instrumental variables with the use of lipid-lowering medication, and that we cannot be certain that the true underlying causal exposure is lipid levels rather than another phenotype for which lipids are a proxy. However, we are not aware of any process for which lipids is a proxy through which BC would be affected, and our gene-specific approach minimizes this concern. Additionally, it is perhaps surprising that we did not find a significant genetic correlation between BC and lipids using cross-trait LD-score regression; however, our result corroborates a previous study that performed this analysis using smaller GWASs [16]. Our lack of significant results could be caused by limited polygenicity of either trait, which decreases the power of this method [15]. We do find significant cross-trait heritability between BC and 2 lipid traits (TC and LDL) using the ρ-Hess method. The discrepancy between LDSC and ρ-Hess may be due to a difference in the statistical power of these methods that has been previously reported [36].

The analyses presented here do not bring evidence on a specific mechanism for tumorigenesis, but they do bring renewed attention to potential mechanisms requiring future functional study. Cholesterol and its oxysterol metabolites, either in the circulatory system or in the local mammary microenvironment, may have direct effects on mammary tissue growth induction of breast tumorigenesis [58,59].

These findings support a causal relationship between increased HDL-cholesterol and increased BC risk, and this hypothesis warrants further exploration. Statins are widely prescribed to decrease LDL levels; however, statins also increase HDL levels. If further research substantiates the relationship between higher HDL levels and increased BC risk, the consensus that HDL is “good cholesterol,” or of benign effect, may require re-evaluation.

Supporting information

S1 STROBE checklist. Reporting document following the STROBE guidelines for our study.

STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

(DOCX)

Click here for additional data file.^{(32.6KB, docx)}

S1 STROBE-MR checklist. Reporting document following the preliminary STROBE-MR guidelines for our study.

MR, Mendelian randomization; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

(DOCX)

Click here for additional data file.^{(34.6KB, docx)}

S1 Fig. QQ plots for lipid association statistics.

Generated from a meta-analysis of Klarin and colleagues [22] and Willer and colleagues [12]. LD-score regression intercepts and standard errors from the meta-analysis association statistics were as follows: TC, 1.1293 (0.1143); TG, 1.0317 (0.0656); LDL, 1.0933 (0.1001); HDL, 1.1715 (0.0758); see also S14 Table. HDL, high-density lipoprotein; LD, linkage disequilibrium; LDL, low-density lipoprotein; TC, total cholesterol; TG, triglyceride; λGC, genomic inflation factor.

(PDF)

Click here for additional data file.^{(674.8KB, pdf)}

S2 Fig. Scatter plots of unpruned single-trait MR genetic instruments’ effect estimates on exposure and outcome.

Plotted are the genetic instruments included in unpruned single-trait MR analyses. Each plot contains effect estimates from MVP for one of 4 lipid traits (HDL, LDL, TC, TGs) on the x-axis and effect estimates for risk of all BCs (allBC) on the y-axis. Error bars represent the 95% CI, and regression lines represent the slope estimate from one of 3 MR tests: IVW (light blue), Egger regression (dark blue), and weighted median (green). BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; TC, total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(22KB, pdf)}

S3 Fig. Scatter plots of pruned single-trait MR genetic instruments’ effect estimates on exposure and outcome.

Scatter plots of genetic instruments included in pruned single-trait MR analyses. Genetic instruments were pruned to pass heterogeneity test. Each plot contains effect estimates from MVP for one of 4 lipid traits (HDL, LDL, TC, TG) on the x-axis and effect estimates for risk of all BCs on the y-axis. Error bars represent the 95% CI, and regression lines represent the slope estimate from one of 3 MR tests: IVW (light blue), Egger regression (dark blue), and weighted median (green). BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; TC, total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(18.8KB, pdf)}

S4 Fig. Results of single-trait MR analyses with a lipid trait as the exposure and risk for all BCs as the outcome.

The exposure association data for this analysis were from MVP. Genetic instruments were pruned to pass heterogeneity test. Error bars represent the 95% CI. Estimates were calculated using the IVW method. *P < 0.05; **P < 0.001. See S7 Table for ORs, CIs, and P-values. BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TC, total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(4.7KB, pdf)}

S5 Fig. Single-trait MR with lipid association statistics from MVP, GLGC, or MVP + GLGC meta-analysis.

Genetic association statistics for all BCs were used for the outcome. Genetic instruments were pruned to pass heterogeneity test. Error bars represent the 95% CI. Estimates were calculated using the IVW method. BC, breast cancer; CI, confidence interval; GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; TC, total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(51.8KB, pdf)}

S6 Fig. Multivariable MR analyses stratified by 3 independent subsets of the BCAC data set.

Results of multivariable MR analyses with 3 lipid traits (HDL, LDL, TG), BMI, and AaM as exposures and BC risk as the outcome. Each panel presents multivariable MR results using BC summary statistics from an independent subset of the BCAC data set (Oncoarray, iCOGS, or GWAS) or from the meta-analysis of all 3 together (BC meta-analysis, S1 Text). Results plotted are after pruning for instrument heterogeneity. Error bars represent the 95% CI. Estimates were calculated using the IVW method. *P < 0.05; **P < 0.001. See S10 Table for ORs, CIs, and P-values. AaM, age at menarche; BC, breast cancer; BCAC, Breast Cancer Association Consortium; BMI, body mass index; CI, confidence interval; GWAS, genome-wide association study; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; OR, odds ratio; TG, triglyceride.

(PDF)

Click here for additional data file.^{(6.3KB, pdf)}

S7 Fig. Multivariable MR analyses using lipid genetic associations from either MVP or GLGC.

Results of multivariable MR analyses including 2 lipid traits as exposures: (A) LDL and HDL or (B) TGs and HDL, with and without BMI as an additional exposure and with risk for all BCs as the outcome. Results plotted are after pruning for instrument heterogeneity. The lipid effect estimates were from one of 2 GWAS data sets (MVP or GLGC), and the results of each combination of lipid data sets are in a single plot. Error bars represent the 95% CI. Estimates were calculated using the IVW method. *P < 0.05; **P < 0.001. See S11 Table for ORs, CIs, and P-values. BC, breast cancer; BMI, body mass index; CI, confidence interval; GLGC, Global Lipids Genetics Consortium; GWAS, genome-wide association study; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TG, triglyceride.

(PDF)

Click here for additional data file.^{(299.7KB, pdf)}

S8 Fig. Single-trait MR with BC outcomes stratified by ER subtypes.

Results of single-trait MR with each lipid trait as an exposure, and one of 3 BC traits as the outcome: all BC, ER− BCs only, or ER+ BCs only. Error bars represent the 95% CI. Estimates were calculated using a fixed-effects IVW method after pruning for instrument heterogeneity. Lipid association statistics come from the MVP data. **P < 0.001, *P < 0.05. See S12 Table for ORs, CIs, and P-values. BC, breast cancer; CI, confidence interval; ER, estrogen receptor; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; OR, odds ratio; TC, total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(5.3KB, pdf)}

S9 Fig. Multivariable MR analyses with BC outcomes stratified by ER subtypes.

Results of multivariable MR analyses with 3 lipid traits (HDL, LDL, TGs), BMI, and AaM as exposures; and all BCs, ER‒, or ER+ BC as the outcome. Results plotted are after pruning for instrument heterogeneity. Error bars represent the 95% CI. Estimates were calculated using the IVW method. *P < 0.05; **P < 0.001. See S13 Table for ORs, CIs, and P-values. AaM, age at menarche; BC, breast cancer; BMI, body mass index; CI, confidence interval; ER, estrogen receptor; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; TG, triglyceride.

(PDF)

Click here for additional data file.^{(5.7KB, pdf)}

S10 Fig. Genetic instruments’ effect estimates on HDL and BC at each canonical HDL metabolism pathway loci.

Conditionally independent HDL-associated SNPs at canonical HDL metabolism pathway genes, plotted by their conditional effect estimates on HDL (from MVP) and effect estimates on all BCs. Error bars represent 95% CIs. The dashed green line represents the regression line from fixed-effects IVW MR. BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; MR, Mendelian randomization; MVP, Million Veteran Program.

(PDF)

Click here for additional data file.^{(9.7KB, pdf)}

S11 Fig. Genetic instruments’ effect estimates on LDL and BC at each canonical LDL metabolism pathway loci.

Conditionally independent LDL-associated SNPs at canonical LDL metabolism pathway genes, plotted by their conditional effect estimates on LDL (from MVP) and effect estimates on all BCs. Error bars represent 95% CIs. The dashed green line represents the regression line from fixed-effects IVW MR. BC, breast cancer; CI, confidence interval; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program.

(PDF)

Click here for additional data file.^{(7.9KB, pdf)}

S12 Fig. LDL locus-specific MR results with LDL as exposure and BC risk as outcome.

Forest plot of MR results for LDL gene-specific instruments (see S4 Fig) and meta-analysis of effect estimates across genes. Estimates were calculated using a fixed-effects IVW method. BC, breast cancer; CI, confidence interval; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; N SNPs, number of genetic instruments included in MR; OR, odds ratio.

(PDF)

Click here for additional data file.^{(54.1KB, pdf)}

S13 Fig. Genetic correlations between lipid and BC traits.

Results of LD-score regression testing for genetic correlation between each lipid trait and 3 BC traits: all BC, ER− BCs only, or ER+ BCs only. Error bars represent the 95% CI. Lipid association statistics were from a meta-analysis of GLGC and MVP. BC, breast cancer; CI, confidence interval; ER, estrogen receptor; GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; LD, linkage disequilibrium; LDL, low-density lipoprotein; MVP, Million Veteran Program; TC; total cholesterol; TG, triglyceride.

(PDF)

Click here for additional data file.^{(50.3KB, pdf)}

S1 Table. Summary statistics for genetic instruments used in single-trait MR analyses.

Lipid exposure summary statistics are from the MVP European data set and BC summary statistics from the BCAC consortium meta-analysis. SNP: rsID of genetic instrument; exposure: lipid trait for exposure statistics; effect_allele.exposure: allele used for lipid and BC effect estimates; other_allele.exposure: noneffect allele; beta.exposure: effect size estimate for lipid trait; se.exposure: standard error of lipid effect size estimate; pval.exposure: P-value of lipid trait effect estimate; beta.bc: effect size estimate of BC risk; se.bc: standard error of BC risk effect estimate; pval.bc: P-value for BC risk effect estimate; inclPruned: logical, was SNP included in pruned single-trait MR analysis. BC, breast cancer; BCAC, Breast Cancer Association Consortium; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(38.8KB, xlsx)}

S2 Table. Summary statistics for genetic instruments used in multivariable MR analyses.

SNP: rsID of genetic instrument; expZ: trait and data set used as first/second/… exposure (e.g., if exp1 is ldl_mvp, LDL summary statistics from GLGC were used as the first exposure); expZ_beta: effect size estimate for trait expZ; expZ_se: standard error of effect size estimate for trait expZ; expZ_pval: P-value of effect size estimate for trait expZ; bc_beta: BC risk effect size estimate; bc_se: standard error of BC effect size estimate; bc_pval: P-value of BC effect size estimate; test: unique identifier for each 2, 3, or 5-exposure MVMR experiment included in this table. AaM, age at menarche; BC, breast cancer; BMI, body mass index; GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVMR, multivariable MR; MVP, Million Veteran Program; TC; total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(913.8KB, xlsx)}

S3 Table. Conditionally independent summary statistics for HDL or LDL associations used in locus-specific MR analyses.

Data are from the conditional analysis of summary statistics from MVP and GLGC meta-analysis published in Klarin and colleagues [22]. CHR: chromosome; POS: base position (HG19); SNP: rsID; effect.allele: allele used for effect size estimate; other.allele: noneffect allele; effect.allele.freq: frequency of effect allele; conditional.beta: effect size estimate from conditional analysis; conditional.se: standard error of conditional effect size estimate; conditional.p: P-value of conditional effect size estimate; Locus: the HDL or LDL gene or locus for MR analysis; Trait: lipid trait used as exposure (HDL or LDL). GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program.

(XLSX)

Click here for additional data file.^{(17KB, xlsx)}

S4 Table. Single-trait MR results with unpruned lipid traits as the exposure and all BCs as the outcome for a range of MR methods.

Lipid association statistics are from MVP. Exposure: lipid trait used as the exposure; Method: MR method used; N SNPs: number of genetic instruments included in analysis; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test. BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(10KB, xlsx)}

S5 Table. Heterogeneity analyses by Cochran's Q of unpruned single-trait MR.

Lipid association statistics are from MVP. Estimates are from the IVW method, and the outcome trait was risk for all BC. Exposure: lipid trait used as the exposure; Q: Cochran’s Q statistic; Q_df: degrees of freedom in Cochran’s Q test; Q_P: P-value of Cochran’s Q test. BC, breast cancer; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(8.4KB, xlsx)}

S6 Table. Pleiotropy analysis using Egger regression of unpruned single-trait MR.

Lipid association statistics are from MVP, and the outcome trait was risk for all BC. Exposure: lipid trait used as the exposure; Egger intercept: intercept estimate from Egger regression; SE: standard error of intercept estimate; P: P-value of intercept estimate. BC, breast cancer; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(8.4KB, xlsx)}

S7 Table. Results of single-trait MR, heterogeneity analyses, and directionality analyses with pruned lipid IV sets.

Lipid summary statistics were from MVP, and MR tests used the IVW method. Heterogeneity analyses used Cochran's Q, and directionality analyses used the Steiger test. Exposure: lipid trait used as the exposure; N_SNPs: number of genetic instruments included in analysis; OR: OR of MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; MR_P: P-value of MR test; Q: Cochran’s Q statistic; Q_df: degrees of freedom in Cochran’s Q test; Q_Pval: P-value of Cochran’s Q test; SNP_r2_exposure: estimated variance in lipid trait explained by genetic instruments; SNP_r2_outcome: estimated variance in breast cancer risk explained by genetic instruments; Steiger_pval: P-value of Steiger test inference of causal direction; correct_causal_direction: logical, is the causal direction inferred by the Steiger test in the correct direction. CI, confidence interval; HDL, high-density lipoprotein; IV, instrumental variable; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(9.6KB, xlsx)}

S8 Table. Results of a reciprocal single-trait MR testing the effects of BC as the exposure on each lipid trait as the outcome.

Lipid summary statistics were from MVP, and MR tests used the IVW method. Exposure: BC used as exposure for all tests; Outcome: lipid trait used as the outcome; N_SNPs: number of genetic instruments used in MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test. BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(8.5KB, xlsx)}

S9 Table. Results of single-trait MR analyses after pruning for genetic instruments associated with other lipid traits.

For each exposure, genetic instruments that were associated (P < 0.001) with the other 2 listed lipid traits were removed before this MR analysis. Lipid summary statistics were from MVP, MR tests used the IVW method, and risk for all BCs was the outcome. Exposure: lipid trait used as the exposure; N_SNPs: number of genetic instruments used in MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test. BC, breast cancer; CI, confidence interval; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(8.5KB, xlsx)}

S10 Table. Results of multivariable MR with 3 lipid traits, BMI, and AaM, as exposures and BC risk as outcome.

Results are from 4 separate multivariable MR experiments: 3 with outcome summary statistics from independent subsets of the BC data set (BC_Onco, Oncoarray; BC_iCoGS, iCOGS; or BC_GWAS, GWAS), or using the BCAC meta-analysis summary statistics (BC_Meta). Lipid summary statistics are from MVP. Before/after pruning: results from MR before or after pruning for instrument heterogeneity; Exposure: trait used as exposure; Outcome: BC summary statistics used for the outcome; N_SNPs: number of genetic instruments used in MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test. AaM, age at menarche; BC, breast cancer; BCAC, Breast Cancer Association Consortium; BMI, body mass index; CI, confidence interval; GWAS, genome-wise association study; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(11.4KB, xlsx)}

S11 Table. Results of multivariable MR with lipid trait summary statistics from distinct cohorts.

Risk for all BCs was used as the outcome. Within a test, the effect estimates for each lipid trait are from different data sets (MVP or GLGC). Each test is separated by an empty row. Exposure_Dataset: lipid trait (or BMI) used as exposure and the data set for lipid summary statistics; N_SNPs: number of genetic instruments used in MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test; Before/after pruning: results from MR before or after pruning for instrument heterogeneity. BC, breast cancer; BMI, body mass index; CI, confidence interval; GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(11.7KB, xlsx)}

S12 Table. Results of Cochran's Q test for heterogeneity on single-trait IVW MR results for the effect of a lipid trait on ER+ versus ER− BC.

Trait: lipid trait used as the exposure; ERposBeta: MR effect estimate for ER+ BCs; ERnegBeta: MR effect estimate for ER− BCs; ERposSE: standard error of MR effect estimate for ER+ BCs; ERnegSE: standard error of MR effect estimate for ER− BCs; Q: Cochran’s Q test statistic; P: P-value of Cochran’s Q test. BC, breast cancer; ER, estrogen receptor; HDL, high-density lipoprotein; IVW, inverse-variance weighted; LDL, low-density lipoprotein; MR, Mendelian randomization; TC, total cholesterol; TG, triglyceride.

(XLSX)

Click here for additional data file.^{(8.7KB, xlsx)}

S13 Table. Results of multivariable MR with ER+ or ER− BC as the outcome.

For each test, 5 traits were used as the exposure traits and ER+ or ER− BC as the outcome. Results are after genetic instrument pruning. Lipid summary statistics are from MVP. N_SNPs: number of genetic instruments used in MR test; CI_95_L: lower bound of 95% CI; CI_95_U: upper bound of 95% CI; P: P-value of MR test. BC, breast cancer; BMI, body mass index; CI, confidence interval; ER, estrogen receptor; HDL, high-density lipoprotein; GLGC, Global Lipids Genetics Consortium; LDL, low-density lipoprotein; MR, Mendelian randomization; MVP, Million Veteran Program; OR, odds ratio; TG, triglycerides.

(XLSX)

Click here for additional data file.^{(9.2KB, xlsx)}

S14 Table. Genome-wide genetic correlation estimates.

Genetic correlation estimates between each lipid trait and risk of all BCs, calculated by 2 methods. Lipid Trait: lipid trait used, Method: method used to estimate genetic correlation; Lipid intercept: estimate of lipid intercept from LDSC; Lipid intercept se: standard error of lipid intercept estimate from LDSC; Cross-trait intercept with BC-all: estimate of cross-trait intercept with risk of all BCs from LDSC; Cross-trait intercept se with BC-all: standard error of estimate of cross-trait intercept with risk of all BCs from LDSC; Genetic covariance: estimate of genetic covariance between lipid trait and BC; Covariance SE: standard error of genetic covariance estimate; Genetic correlation: estimate of genetic correlation between lipid trait and BC; Correlation SE: standard error of estimate of genetic correlation; Correlation Z-score: standard normalized estimate of genetic correlation; Correlation p-value: P-value of genetic correlation. BC, breast cancer; HDL, high-density lipoprotein; Hess, ρ-Hess method; LDL, low-density lipoprotein; LDSC, linkage disequilibrium-score regression; TC, total cholesterol; TG, triglycerides.

(XLSX)

Click here for additional data file.^{(9.9KB, xlsx)}

S15 Table. Local genetic correlations calculated with the ρ-Hess method.

Genomic regions with significant local genetic correlation between BC and lipids using the ρ-Hess method. Only loci that passed Bonferroni correction (P = 0.05/1,703 partitions = 2.9 × 10⁻⁵) are shown. Lipid: lipid trait used; Position: genomic coordinates of loci tested (HG19); N_SNPs: number of SNPs in partition; K: number eigenvectors used; Local-rhog: local genetic correlation estimate; Variance: variance estimate; SE: standard error of genetic correlation estimate; Z: Z-score of genetic correlation estimate; P: P-value of genetic correlation estimate. BC, breast cancer.

(XLSX)

Click here for additional data file.^{(8.6KB, xlsx)}

S1 Text. Supplementary Text.

Included are details about the GWASs utilized, heterogeneity analysis for single-trait MR, instrument strength and validity assessment for multivariable MR, and a list of investigators associated with the VA MVP (banner author). GWAS, genome-wide association study; MR, Mendelian randomization; MVP, Million Veteran Program; VA, Veterans Affairs.

(DOCX)

Click here for additional data file.^{(54.8KB, docx)}

Abbreviations

ABCA1: ATP Binding Cassette Subfamily A Member 1
APOC: Apolipoprotein C1
APOE: Apolipoprotein E
BC: breast cancer
BCAC: Breast Cancer Association Consortium
BMI: body mass index
CETP: Cholesteryl Ester Transfer Protein
CI: confidence interval
ER: estrogen receptor
GLGC: Global Lipids Genetics Consortium
GWAS: genome-wide association study
HDL: high-density lipoprotein
HMGCR: 3-Hydroxy-3-Methylglutaryl-CoA Reductase
LCAT: Lecithin-Cholesterol Acyltransferase
LD: linkage disequilibrium
LDL: low-density lipoprotein
LDLR: LDL Receptor
LIPC: Lipase C, Hepatic Type
LIPG: Lipase G, Endothelial Type
LPA: Lipoprotein(A)
MR: Mendelian randomization
MVP: Million Veteran Program
MYLIP: Myosin Regulatory Light Chain Interacting Protein
NPC1L1: NPC1-Like Intracellular Cholesterol Transporter 1
OR: odds ratio
PCSK9: Proprotein Convertase Subtilisin/Kexin Type 9
PLTP: Phospholipid Transfer Protein
SCARB1: Scavenger Receptor Class B Member 1
STROBE: Strengthening the Reporting of Observational Studies in Epidemiology
TC: total cholesterol
TG: triglyceride

Data Availability

The summary statistics for the MR instrumental variables are available in S1, S2 and S3 Tables. Genome-wide summary statistics are available from the Global Lipids Genetics Consortium (GLGC) at http://csg.sph.umich.edu/abecasis/public/lipids2013/ and the Breast Cancer Association Consortium (BCAC) at http://bcac.ccge.medschl.cam.ac.uk/bcacdata/oncoarray/oncoarray-and-combined-summary-result/gwas-summary-results-breast-cancer-risk-2017/. The Million Veterans Program (MVP) lipid GWAS results are available in dbGAP. The dbGAP accession number for MVP overall is phs001672.v3.p1. The accession numbers for the European-specific MVP data are TC: pha004834.1, LDL: pha004831.1, HDL: pha004828.1, and TG: pha004837.1. BMI summary statistics from Yengo et al. are available at https://portals.broadinstitute.org/collaboration/giant/index.php/GIANT_consortium_data_files#2018_GIANT_and_UK_BioBank_Meta-analysis. Age of menarche summary statistics from Day et al are available at https://www.reprogen.org/data_download.html. The UK10K data utilized in the study cannot be shared publicly (per data use access agreement) but are available by Institutional Data Access request for researchers who meet the criteria for access at https://www.sanger.ac.uk/legal/DAA/MasterController.

Funding Statement

This work was supported by the US National Institutes of Health (R01 DK101478 and HG010067 for BFV, T32 GM008216 for KEJ, T32 HG000046 for KMS) and a Linda Pechenik Montague Investigator award (to BFV). This research is based on data from the Million Veteran Program, Office of Research and Development, Veterans Health Administration and was supported by award #MVP000. This research was also supported by two additional Department of Veterans Affairs awards (I01-BX003362 [PST/KC], IK2-CX001780 [Damrauer]). This publication does not represent the views of the Department of Veterans Affairs or the United States Government. This study makes use of data generated by the UK10K Consortium, derived from samples from the Avon Longitudinal Study of Parents and Children (ALSPAC) and the Department of Twin Research and Genetic Epidemiology (DTR), the TWINSUK Cohort. A full list of the investigators who contributed to the generation of the data is available from www.UK10K.org. Funding for UK10K was provided by the Wellcome Trust under award WT091310. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1.Torre LA, Siegel RL, Ward EM, Jemal A. Global Cancer Incidence and Mortality Rates and Trends—An Update. Cancer Epidemiol Biomarkers Prev. 2016;25: 16–27. 10.1158/1055-9965.EPI-15-0578 [DOI] [PubMed] [Google Scholar]
2.Rose D, Gracheck P, Vona-Davis L, Rose DP, Gracheck PJ, Vona-Davis L. The Interactions of Obesity, Inflammation and Insulin Resistance in Breast Cancer. Cancers (Basel). 2015;7: 2147–2168. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Kuzu OF, Noory MA, Robertson GP. The role of cholesterol in cancer. Cancer Research. 2016;76(8): 2063–2070. 10.1158/0008-5472.CAN-15-2613 [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Touvier M, Fassier P, His M, Norat T, Chan DSM, Blacher J, et al. Cholesterol and breast cancer risk: a systematic review and meta-analysis of prospective studies. Br J Nutr. 2015;114: 347–357. 10.1017/S000711451500183X [DOI] [PubMed] [Google Scholar]
5.Ni H, Liu H, Gao R. Serum Lipids and Breast Cancer Risk: A Meta-Analysis of Prospective Cohort Studies. Singh S, editor. PLoS ONE. 2015;10: e0142669 10.1371/journal.pone.0142669 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Martin LJ, Melnichouk O, Huszti E, Connelly PW, Greenberg CV., Minkin S, et al. Serum Lipids, Lipoproteins, and Risk of Breast Cancer: A Nested Case-Control Study Using Multiple Time Points. JNCI J Natl Cancer Inst. 2015;107: djv032–djv032. 10.1093/jnci/djv032 [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Zhong S, Zhang X, Chen L, Ma T, Tang J, Zhao J. Statin use and mortality in cancer patients: Systematic review and meta-analysis of observational studies. Cancer Treatment Reviews. 2015;41(6): 554–567. 10.1016/j.ctrv.2015.04.005 [DOI] [PubMed] [Google Scholar]
8.Borgquist S, Giobbie-Hurder A, Ahern TP, Garber JE, Colleoni M, Láng I, et al. Cholesterol, Cholesterol-Lowering Medication Use, and Breast Cancer Outcome in the BIG 1–98 Study. J Clin Oncol. 2017;35: 1179–1188. 10.1200/JCO.2016.70.3116 [DOI] [PubMed] [Google Scholar]
9.Orho-Melander M, Hindy G, Borgquist S, Schulz C-A, Manjer J, Melander O, et al. Blood lipid genetic scores, the HMGCR gene and cancer risk: a Mendelian randomization study. Int J Epidemiol. 2018;47(2): 495–505. Epub 2017 Nov 20. 10.1093/ije/dyx237 [DOI] [PubMed] [Google Scholar]
10.Nowak C, Ärnlöv J. A Mendelian randomization study of the effects of blood lipids on breast cancer risk. Nat Commun. 2018;9(1): 3957 10.1038/s41467-018-06467-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Michailidou K, Lindström S, Dennis J, Beesley J, Hui S, Kar S, et al. Association analysis identifies 65 new breast cancer risk loci. Nature. 2017;551: 92–94. 10.1038/nature24284 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Global Lipids Genetics Consortium, Willer CJ, Schmidt EM, Sengupta S, Peloso GM, Gustafsson S, Kanoni S, et al. Discovery and Refinement of Loci Associated with Lipid Levels. Nat Genet. 2013;45: 1274–1283. 10.1038/ng.2797 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Qi G, Chatterjee N. Mendelian randomization analysis using mixture models for robust and efficient estimation of causal effects. Nat Commun. 2019;10: 1941 10.1038/s41467-019-09432-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Neuhouser ML, Aragaki AK, Prentice RL, Manson JE, Chlebowski R, Carty CL, et al. Overweight, Obesity, and Postmenopausal Invasive Breast Cancer Risk. JAMA Oncol. 2015;1: 611 10.1001/jamaoncol.2015.1546 [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh PR, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015;47: 1236–1241. 10.1038/ng.3406 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Lindström S, Finucane H, Bulik-Sullivan B, Schumacher FR, Amos CI, Hung RJ, et al. Quantifying the genetic correlation between multiple cancer types. Cancer Epidemiol Biomarkers Prev. 2017;26: 1427–1435. 10.1158/1055-9965.EPI-17-0211 [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Day FR, Thompson DJ, Helgason H, Chasman DI, Finucane H, Sulem P, et al. Genomic analyses identify hundreds of variants associated with age at menarche and support a role for puberty timing in cancer risk. Nat Genet. 2017;49: 834–841. 10.1038/ng.3841 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Burgess S, Thompson DJ, Rees JMB, Day FR, Perry JR, Ong KK. Dissecting causal pathways using mendelian randomization with summarized genetic data: Application to age at menarche and risk of breast cancer. Genetics. 2017;207: 481–487. 10.1534/genetics.117.300191 [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Feng Y, Hong X, Wilker E, Li Z, Zhang W, Jin D, et al. Effects of age at menarche, reproductive years, and menopause on metabolic risk factors for cardiovascular diseases. Atherosclerosis. 2008;196: 590–597. 10.1016/j.atherosclerosis.2007.06.016 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Hamajima N, Hirose K, Tajima K, Rohan T, Friedenreich CM, Calle EE, et al. Menarche, menopause, and breast cancer risk: Individual participant meta-analysis, including 118 964 women with breast cancer from 117 epidemiological studies. Lancet Oncol. 2012;13: 1141–1151. 10.1016/S1470-2045(12)70425-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Jiang X, Finucane HK, Schumacher FR, Schmit SL, Tyrer JP, Han Y, et al. Shared heritability and functional enrichment across six solid cancers. Nat Commun. 2019;10: 431 10.1038/s41467-018-08054-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Klarin D, Damrauer SM, Cho K, Sun Y V., Teslovich TM, Honerlaw J, et al. Genetics of blood lipids among ~300,000 multi-ethnic participants of the Million Veteran Program. Nat Genet. 2018;50: 1514–1523. 10.1038/s41588-018-0222-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Yengo L, Sidorenko J, Kemper KE, Zheng Z, Wood AR, Weedon MN, et al. Meta-analysis of genome-wide association studies for height and body mass index in ~700000 individuals of European ancestry. Hum Mol Genet. 2018;27: 3641–3649. 10.1093/hmg/ddy271 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet. 2007;81: 559–575. 10.1086/519795 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife. 2018;7: e34408 10.7554/eLife.34408 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44: 512–525. 10.1093/ije/dyv080 [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genet Epidemiol. 2016;40: 304–314. 10.1002/gepi.21965 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Hartwig FP, Smith GD, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46(6): 1985–1998. 10.1093/ije/dyx102 [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Walter K, Min JL, Huang J, Crooks L, Memari Y, McCarthy S, et al. The UK10K project identifies rare variants in health and disease. Nature. 2015;526: 82–90. 10.1038/nature14962 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Voight BF. MR-predictor: A simulation engine for Mendelian Randomization studies. Bioinformatics. 2014;30: 3432–3434. 10.1093/bioinformatics/btu564 [DOI] [PubMed] [Google Scholar]
31.Burgess S. Sample size and power calculations in Mendelian randomization with a single instrumental variable and a binary outcome. Int J Epidemiol. 2014;43: 922–929. 10.1093/ije/dyu005 [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Burgess S, Thompson SG. Multivariable Mendelian randomization: The use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181: 251–260. 10.1093/aje/kwu283 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Burgess S, Dudbridge F, Thompson SG. Re: “Multivariable Mendelian randomization: The use of pleiotropic genetic variants to estimate causal effects.” Am J Epidemiol. 2015;181: 290–291. 10.1093/aje/kwv017 [DOI] [PubMed] [Google Scholar]
34.Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. Int J Epidemiol. 2019;48: 713–727. 10.1093/ije/dyy262 [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Schwarzer G. meta: An R package for meta-analysis. R News. 2007;7: 40–45. [Google Scholar]
36.Shi H, Mancuso N, Spendlove S, Pasaniuc B. Local Genetic Correlation Gives Insights into the Shared Genetic Architecture of Complex Traits. Am J Hum Genet. 2017;101: 737–751. 10.1016/j.ajhg.2017.09.022 [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Berisa T, Pickrell JK. Approximately independent linkage disequilibrium blocks in human populations. Bioinformatics. 2016;32: 283–5. 10.1093/bioinformatics/btv546 [DOI] [PMC free article] [PubMed] [Google Scholar]
38.The GTEx Consortium, Aguet F, Ardlie KG, Cummings BB, Gelfand ET, Getz G, et al. Genetic effects on gene expression across human tissues. Nature. 2017;550: 204–213. 10.1038/nature24277 [DOI] [PMC free article] [PubMed] [Google Scholar]
39.von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies. PLoS Med. 2007;4: e296 10.1371/journal.pmed.0040296 [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Davey Smith G, Davies N, Dimou N, Egger M, Gallo V, Golub R, et al. STROBE-MR: Guidelines for strengthening the reporting of Mendelian randomization studies. PeerJ Preprints 27857 [Preprint]. 2019 [cited 2020 May 18]. https://peerj.com/preprints/27857/
41.Pickrell JK, Berisa T, Liu JZ, Ségurel L, Tung JY, Hinds DA. Detection and interpretation of shared genetic influences on 42 human traits. Nat Genet. 2016;48: 709–717. 10.1038/ng.3570 [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Hemani G, Tilling K, Davey Smith G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. Li J, editor. PLoS Genet. 2017;13: e1007081 10.1371/journal.pgen.1007081 [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Guo Y, Warren Andersen S, Shu XO, Michailidou K, Bolla MK, Wang Q, et al. Genetically Predicted Body Mass Index and Breast Cancer Risk: Mendelian Randomization Analyses of Data from 145,000 Women of European Descent. PLoS Med. 2016;13 10.1371/journal.pmed.1002105 [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Renehan AG, Tyson M, Egger M, Heller RF, Zwahlen M. Body-mass index and incidence of cancer: a systematic review and meta-analysis of prospective observational studies. Lancet. 2008;371: 569–78. 10.1016/S0140-6736(08)60269-X [DOI] [PubMed] [Google Scholar]
45.Borgquist S, Butt T, Almgren P, Shiffman D, Stocks T, Orho-Melander M, et al. Apolipoproteins, lipids and risk of cancer. Int J Cancer. 2016;138: 2648–2656. 10.1002/ijc.30013 [DOI] [PubMed] [Google Scholar]
46.Melvin JC, Seth D, Holmberg L, Garmo H, Hammar N, Jungner I, et al. Lipid profiles and risk of breast and ovarian cancer in the swedish AMORIS study. Cancer Epidemiol Biomarkers Prev. 2012;21: 1381–1384. 10.1158/1055-9965.EPI-12-0188 [DOI] [PubMed] [Google Scholar]
47.Zhang B-L, He N, Huang Y-B, Song F-J, Chen K-X. ABO blood groups and risk of cancer: a systematic review and meta-analysis. Asian Pac J Cancer Prev. 2014;15: 4643–50. 10.7314/apjcp.2014.15.11.4643 [DOI] [PubMed] [Google Scholar]
48.Beeghly-Fadiel A, Khankari NK, Delahanty RJ, Shu X-O, Lu Y, Schmidt MK, et al. A Mendelian randomization analysis of circulating lipid traits and breast cancer risk. Int J Epidemiol. Epub 2019 Dec 23. 10.1093/ije/dyz242 [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Madsen CM, Varbo A, Nordestgaard BG. Extreme high high-density lipoprotein cholesterol is paradoxically associated with high mortality in men and women: two prospective cohort studies. Eur Heart J. 2017;38: 2478–2486. 10.1093/eurheartj/ehx163 [DOI] [PubMed] [Google Scholar]
50.Hamer M, O’Donovan G, Stamatakis E. High-Density Lipoprotein Cholesterol and Mortality: Too Much of a Good Thing? Arterioscler Thromb Vasc Biol. 2018;38: 669–672. 10.1161/ATVBAHA.117.310587 [DOI] [PubMed] [Google Scholar]
51.Burgess S, Davey Smith G. Mendelian Randomization Implicates High-Density Lipoprotein Cholesterol–Associated Mechanisms in Etiology of Age-Related Macular Degeneration. Ophthalmology. 2017;124: 1165–1174. 10.1016/j.ophtha.2017.03.042 [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Fan Q, Maranville JC, Fritsche L, Sim X, Cheung CMG, Chen LJ, et al. HDL-cholesterol levels and risk of age-related macular degeneration: a multiethnic genetic study using Mendelian randomization. Int J Epidemiol. 2017;46: 1891–1902. 10.1093/ije/dyx189 [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Bonovas S, Filioussi K, Tsavaris N, Sitaras NM. Use of Statins and Breast Cancer: A Meta-Analysis of Seven Randomized Clinical Trials and Nine Observational Studies. J Clin Oncol. 2005;23: 8606–8612. 10.1200/JCO.2005.02.7045 [DOI] [PubMed] [Google Scholar]
54.Llewellyn-Bennett R, Edwards D, Roberts N, Hainsworth AH, Bulbulia R, Bowman L. Post-trial follow-up methodology in large randomised controlled trials: a systematic review. Trials. 2018;19: 298 10.1186/s13063-018-2653-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
55.VanderWeele TJ, Tchetgen Tchetgen EJ, Cornelis M, Kraft P. Methodological challenges in mendelian randomization. Epidemiology. 2014;25: 427–35. 10.1097/EDE.0000000000000081 [DOI] [PMC free article] [PubMed] [Google Scholar]
56.O’Connor LJ, Price AL. Distinguishing genetic correlation from causation across 52 diseases and complex traits. Nat Genet. 2018;50: 1728–1734. 10.1038/s41588-018-0255-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, Davey Smith G. Best (but oft-forgotten) practices: the design, analysis, and interpretation of Mendelian randomization studies. Am J Clin Nutr. 2016;103: 965–978. 10.3945/ajcn.115.118216 [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Silvente-Poirot S, Poirot M. Cancer. Cholesterol and cancer, in the balance. Science. 2014;343(6178): 1445–1446. 10.1126/science.1252787 [DOI] [PubMed] [Google Scholar]
59.Nelson ER. The significance of cholesterol and its metabolite, 27-hydroxycholesterol in breast cancer. Mol Cell Endocrinol. 2018;466: 73–80. 10.1016/j.mce.2017.09.021 [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS Med. doi: 10.1371/journal.pmed.1003302.r001

Decision Letter 0

Caitlin Moyer

24 Mar 2020

Dear Dr. Siewert,

Thank you very much for submitting your manuscript "Assessing a causal relationship between circulating lipids and breast cancer risk: Mendelian randomization study" (PMEDICINE-D-19-03886) for consideration at PLOS Medicine.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also sent to five independent reviewers. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers. In particular, please address in your response and through changes to the text and references, where appropriate, the concern mentioned by several of the reviewers regarding overlap between this study and recently published works (please see the first comments from Reviewers 1, 3, 4, and 5).

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Apr 14 2020 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

1. Data Availability: Thank you for pointing out that your data are freely available. In your data availability statement, please state the location of data both within the paper, and also provide the links (as you have done in the supporting information section).

2. Did your study have a prospective protocol or analysis plan? Please state this (either way) early in the Methods section.

a) If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant prospectively written document with your revised manuscript as a Supporting Information file to be published alongside your study, and cite it in the Methods section. A legend for this file should be included at the end of your manuscript.

b) If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and when/why any data-driven changes to analyses took place.

c) In either case, changes in the analysis-- including those made in response to peer review comments-- should be identified as such in the Methods section of the paper, with rationale.

3. Title: Please revise your title to: “The relationship between circulating lipids and breast cancer risk: a Mendelian randomization study” or similar.

4. Abstract: Line 10: The word “randomization” is missing at the end of the last sentence of the background.

5. Abstract: Please define abbreviations for BCAC, SD, OR, MR at their first use. Please also give some brief context for “ABO locus”, to indicate relevance for blood group, etc.

6. Abstract: In the last sentence of the Abstract Methods and Findings section, please describe the main limitation(s) of the study's methodology.

7. Abstract: Line 27: Please change “find” to “found”.

8. Abstract: Conclusions: Please modify the first sentence of the conclusion to: “Genetically elevated plasma HDL levels appear to be associated with increased breast cancer risk” or similar. The phrase "In this study, we observed ..." may be useful.

9. At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary

10. Introduction: Line 56: Please define the abbreviation for GWAS at first use.

11. Introduction: Line 58: Please define the abbreviation for “ER”

12. Introduction: Line 82-83: Please remove the description of the findings from the introduction.

13. Methods: Please include a statement to generally describe that institutional ethical approval and consent were obtained for those participating in the cohorts that provided the summary/ GWAS data.

14. Methods: Line 101: Should “plink” be capitalized?

15. Methods: Line 117-119: Please make sure that the referencing of these studies/criteria follow Vancouver style: https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references

16. Methods: Line 139: please define abbreviation for LD.

17. Results: Line 151: Thank you for spelling out these abbreviations- however, please spell them out where they are first used earlier in the manuscript.

18. Results: Line 248-251: For clarity, we suggest that you first mention only the significant interactions (and the method that produced them) and then list the methods/interactions that fall short of significant second. Please explicitly clarify where significance indicates statistical significance, and please clarify what is meant by “nominal association” - does this indicate statistical significance? If so, please state that.

19. Results: Throughout the results section, where positive or negative relationships are noted, please clarify whether relationships are statistically significant, for example at Line 230: “We observed a positive relationship…” it seems such relationships reach statistical significance, while relationships described at line 248 do not. We suggest revising the sentence: “However, most P-values were not significant; the only nominal association (P < 0.05) was with total cholesterol and ER-negative breast cancer (P=0.04).” to: “However, most genetic correlation estimates between breast cancer and lipid traits were not statistically significant; the only nominal statistically significant association (P < 0.05) was with total cholesterol and ER-negative breast cancer (P=0.04).” or similar.

20. Figure 1: Some symbols in Figure 1 do not appear to be displaying properly, please check the formatting. Please define all abbreviations in the figure legend (i.e. for MR, OR, VW, HDL, LDL, TG).

21. Figure 2: Please define abbreviations “N SNPs” and “OR” in the figure legend. In the right-side panel, please describe the dotted line and bars.

22. Discussion: Line 308: Please define abbreviation for “IVS”.

23. Discussion: Line 323: Please edit the sentence to “These findings support a causal relationship between increased HDL cholesterol and increased breast cancer risk…” or similar.

24. References: For in-text citations, please use square brackets, like this [1].

25. Supporting Information: Please provide titles for supplementary figures and tables, and please ensure that all abbreviations present in the figures and tables are defined in the figure legends.

26. Supporting Information: For consistency, please use the "Vancouver" style for reference formatting, and see our website for other reference guidelines: https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references

Comments from the reviewers:

Reviewer #1: PMEDICINE-D-19-03886.

The authors conduct a nice, thorough Mendelian Randomization analysis of the relationship between blood lipids and breast cancer risk. Similar work has been done previously (Nowak and Ärnlöv, Nat Comm 2018; Qi and Chatterjee Nat Comm 2019). Nonetheless, the work presented here constitutes a more comprehensive MR analysis than those papers, but the authors should cite the work by Qi and Chatterjee. I have a few questions for clarification

1. It is not clear to me why the authors sometimes uses lipids summary statistics from MVP, sometimes from GLGC and sometimes from the meta-analysis between the two.

2. There is some evidence of sex difference in SNP-lipids associations (e.g. Asselbergs, AJHG 2012). If possible, it would be interesting to see MR analysis using SNPs associated with lipids specifically in women, although this might not be possible if the authors do not have access to the raw GWAS and lipids data for such analyses.

3. It is not clear to me why the authors present data from iCOGS, OncoArray and GWAS separately? Unless there are explicit reasons for this, I suggest only using the meta-analysis results for breast cancer only.

4. The authors pruned pleiotropic variants and reran the analyses similar to the Nowak paper (top page 10). Even though they note that the confidence intervals increased, it is also worth noting that the effect estimates changed quite a bit as well, so the null association is not only due to the increased confidence intervals. In particular, for TG the confidence intervals barely overlap.

5. As the authors pointed out, one of the things that distinguish this work from Nowak and Ärnlöv is the multivariable approach presented here. Another strength with this paper is the assessment of ER+ and ER- breast cancer. I think that the paper would gain from a more comprehensive analysis of ER+ and ER- breast cancer specifically, using some of the multivariable approaches described here.

6. Are the results from rho-HESS (Suppl Table 13) genetic covariances or genetic correlations? They seem small for being correlations. For the uninformed, a quick explanation about the differences in the overall (not local) genetic correlations as estimated from rho-HESS and LDSC would be helpful and help interpret the differences in results. Related to that, it would be interesting to expand on the discussion on the local genetic correlation results. Is the main conclusion from those analyses that ABO is casually associated with both blood lipids and breast cancer risk?

7. Minor point: Refs no. 5 and 7 seem to be the same.

Reviewer #2: This is a comprehensive Mendelian randomization study of lipids and breast cancer risk. Though I am not familiar with the more novel methods used in the study, it is clear that the authors apply both conventional and state-of-the-art methods on well-known large GWAS data to investigate the topic in depth. The study is well-written, and the authors generally use balanced language in relation to their methods and findings. I have few specific comments:

Introduction

1. Reverse causation has been frequently debated to be involved in the association between cholesterol and risk of several forms of cancer. I suggest to mention the potential for reverse causation also in breast cancer if this is of particular relevance also for breast cancer.

2. Reference 6 is about survival amongst cancer cases and not about cancer risk, as referred to in the introduction and which the present study is about. The reference also lacks authors in the reference list.

3. Reference 7 is not about statins, but is a meta-analysis of lipids and breast cancer risk. Reference 5 and 7 are the same.

4. In reference 9, an equally suggestive positive association between HDL and breast cancer was shown as the suggestive negative association between TG and breast cancer. Also, for TG, the authors should report the association in ref 9 to be inverse/negative. Reference 9 and 38 are the same. Reference 38 referred to as the study by Nowak et al (first sentence page 15) is wrong.

5. There are several strong risk factors of breast cancer in addition to BMI (BMI not even being amongst the most important ones) and age at menarche, so including particularly these in an MR analysis is not the "ideal approach", the authors should remove or reformulate the sentence on page 4, lines 69-71.

6. Please remove the plentiful results reported from the last paragraph of the introduction, as they are not essential to understand the aims.

Methods

1. Page 6, lines 107-108. Please revise the sentence if the lipid, BMI and age at menarche genetic associations were estimated in different cohorts to those of breast cancer genetic associations, i.e. the important part being no overlap with breast cancer cohorts.

2. Include a reference for the multivariable MR analysis performed.

Reviewer #3: This study uses genetic instruments defined in the Million Veterans Project, to assess the potential role of HDL and LDL in breast cancer risk data from the BCAC consortium.

This study ignores literature published during 2019, when 2 other papers on Mendelian randomisation provide essentially the same results. One of them, by Beeghly-Fadiel et al. on behalf of the BCAC group, had access to individual level data to stratify the analyses not only by ER status, but also for menopausal status, age, and BMI. In that study, only the genetic instrument for HDL was associated with BC. The second study, by Qi and Chatterjee also found the increased risk restricted to HDL, not to LDL or TG. The authors should revise their manuscript in light of these of the studies and explain the reason of their different findings.

The adjustment for BMI and age at menarche, using genetic instruments is an interesting approach when no observed data is accesible, but it should be commented what are the R2 of these instruments? if low, residual confounding may be high. The study bye the BCAC group analysed these associations among controls and only found consistent associations with age. Similarly, the R2 of the pruned genetic instruments finally used would be important to know. Could genetic ancestry bias the results? The qqplots of the genetic instruments in the MVP study showed some inflation.

It is not explicit if the heterogeneity pruning was performed at the level of genotyping array, or at the country/population level, since each of the 3 BCAC subsets comprises multiple smaller studies performed in diverse populations.

The discussion is very poor, with some emphasis on the potential adverse effects on increasing HDL by statins, but no discussion about the results of meta-analysis of cohort studies on BC risk that have measured plasma lipid levels.

One of the interesting analysis performed is the gene-level approach. The finding of ABO and BRCA is intriguing. The epidemiological evidence of such association has been explored previously, with a possible higher risk of the A group, but the discussion is missing.

Minor:

Abstract: end last sentence of Background

Reviewer #4: GENERAL COMMENTS

------------------

Johnson et al. conduct a thorough and comprehensive MR analysis of lipid traits and breast cancer risk. The methods applied are appropriate and well described, and their conclusions reasonable given the results. My only major concern relates to the novelty of the study, given that MR analyses of lipid traits and breast cancer risk have previously been published, including a recent analysis by Beeghly-Fadiel et al. (IJE, 2019) (although it is important to note that the paper by Beeghly-Fadiel may have been published after submission of this paper by Johnson et al., I am not sure).

MAJOR COMMENTS

------------------

1) Beeghly-Fadiel et al (https://www.ncbi.nlm.nih.gov/pubmed/31872213) have recently published an MR analysis of lipid traits and breast cancer risk, using similar data sets and arriving at similar conclusions. The paper by Beeghly-Fadiel et al should be referenced, and given the similarity of the papers, a comprehensive comparison of methods employed and conclusions drawn should be made.

MINOR COMMENTS

------------------

1) Line 10, "Randomization" is missing from the end of the sentence.

2) Line 109, authors should state whether they are using an additive or multiplicative random effects model.

3) The authors should state how they harmonized the lipid and breast cancer GWAS SNPs (I assume it was using TwoSampleMR). They should clarify that they checked that all SNP-exposure association estimates and the SNP-outcome association estimates used the same effect alleles, and that there existed no palindromic or contradictory alleles.

4) Lines 173-174, the authors state that they also tested the relationship between lipids and breast cancer also using GLGC along. I assume this was as a sanity check of heterogeneity between the data sets, but this should be made clearer.

5) The authors state that the meta-analysis of the Klarin and Miller GWAS "appears well-calibrated". However, the lambda values in Supplementary Figure 1 range from 1.29 to 1.51, indicating some inflation of the test statistics. The authors should state that such inflation exists, and discuss whether the inflation could affect the results of their MR analyses.

6) How was the set of genes with protein products with key roles in HDL or LDL biology chosen? Is it from prior knowledge of the authors?

7) How evidence of directional pleiotropy was assessed should be described in the methods section.

8) That causal effects were also estimated using MR-Egger, median-based, and mode-based approaches should be mentioned in the methods.

9) Estimates of study power should be provided for each of the lipids. I assume that the powers for the null results are good, but it would be useful to have estimates presented anyway.

10) Line 163, triglycerides should be abbreviated for consistency.

11) There have been a number of RCT of statins. Have any of these RCTs observed the predicted effects on breast cancer incidence?

Alex Cornish (ICR, London). I am happy to have my name made available to the authors.

Reviewer #5: My main concern with the manuscript is that the headline findings have all already been published - based on the same or similar datasets and using Mendelian randomisation methods - and therefore this manuscript does not add anything new to the literature at all.

In this manuscript, Johnson et al reports an association between HDL cholesterol and risk for all breast cancers. However, the same association (overlapping confidence intervals and similar point estimates) has already been reported by Beeghly-Fadiel et al International Journal of Epidemiology 23 December 2019 (doi.org/10.1093/ije/dyz242) using the (nearly) identical Breast Cancer Association Consortium (BCAC) data set. Both Johnson et al and Beeghly-Fadiel et al report results taking into account the effect of body mass index, menopausal status, and estrogen receptor status. Both papers perform multivariable Mendelian randomisation analyses for all three lipid traits evaluated - LDL, HDL, TG.

In this manuscript, Johnson et al also report genome-wide genetic correlation between cholesterol levels and breast cancer. However, these have already been reported by Jiang et al. Nature Communications 25 January 2019 using the same BCAC data set used in this manuscript (https://doi.org/10.1038/s41467-018-08054-4).

Both Johnson et al and Beeghly-Fadiel et al report an inconsistent relationship between triglycerides and breast cancer risk.

The only novel results presented by Johnson et al are gene-level analyses for HDL-associated genes (Figure 2) and local genetic correlation analyses that identify pleiotropic breast cancer risk- and LDL-associated SNPs near the ABO gene. However, this region is already known (published) to harbour genome-wide significant risk SNPs for LDL cholesterol and breast cancer risk in the publicly available data sets used by Johnson et al. The ABO locus itself is established as a highly pleiotropic region and this result is therefore not surprising (see for example: 10.1038/ng.3570 ). Further, the results in Figure 2 of Johnson et al can be worked out using supplementary data from Beeghly-Fadiel et al. Finally, Johnson et al use cholesterol SNPs from the Million Veteran Program but the use of these MVP SNPs does not make any qualitative difference to the results compared to the use of the SNPs from the Global Lipid Genetics Consortium (used by Beeghly-Fadiel et al).

Any attachments provided with reviews can be seen via the following link:

[LINK]

PLoS Med. 2020 Sep 11;17(9):e1003302. doi: 10.1371/journal.pmed.1003302.r002

Author response to Decision Letter 0

14 Apr 2020

Attachment

Submitted filename: Response_to_reviewers_2.pdf

Click here for additional data file.^{(254.5KB, pdf)}

PLoS Med. doi: 10.1371/journal.pmed.1003302.r003

Decision Letter 1

Caitlin Moyer

11 May 2020

Dear Dr. Voight,

Thank you very much for submitting your revised manuscript "The relationship between circulating lipids and breast cancer risk: a Mendelian randomization study" (PMEDICINE-D-19-03886R1) for consideration at PLOS Medicine.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with an academic editor with relevant expertise, and sent back to two of the reviewers for re-review. The comments from the academic editor are pasted below. Both of the reviewers were satisfied with the revised version and noted no further comments.

In light of the comments from the academic editor, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the comments. Obviously we cannot make any decision about publication until we and the academic editor have seen the revised manuscript and your response.

In revising the manuscript for further consideration, your revisions should address the specific points made by the editors and the academic editor. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

We expect to receive your revised manuscript by May 25 2020 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

We look forward to receiving your revised manuscript.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

1. Abstract: Line 18: Please use either Mendelian randomization or the abbreviation MR consistently.

2. Abstract: Line 29: Please revise this sentence to avoid implications of causality: “In addition, we found evidence that genetic variation at the ABO locus is associated with both lipid levels and breast cancer” or similar.

3. Abstract: Conclusions: Line 35: Would it make sense here to also mention the association between LDL and breast cancer risk that you identified with multivariable MR? (mentioned above at Line 22-23)?

4. Author Summary: Thank you for providing an author summary. Under “Why was this study done?” we suggest that you combine the first two bullet points. At line 35, under the third bullet point, we suggest changing “genetics” to “Mendelian randomization methods” or similar to enhance specificity.

5. Author Summary: Please use the first person perspective. For example, please use “We tested…” at line 49.

6. Under “What do these findings mean?” Please delete the first and third bullet point, and replace it with a sentence that can address the study implications without overreaching what can be concluded from the data; we suggest: "Further research will be needed to investigate the possibility that manipulation of LDL or HDL levels can influence risk of breast cancer"

7. Introduction: Line 93: Rather than "nominal", please change to "association of uncertain significance" or similar, to clarify.

8. Introduction: Line 101-105: This would be more appropriate in the Discussion section: “A recent study by Beeghly-Fadiel et al., published following the submission of this manuscript, performed an MR analysis of breast cancer risk that considered potentially confounding risk factors and draws some of the same major conclusions as we do [21]. We highlight the distinctions between that study and the present manuscript in the Discussion.”

9. Methods: Lines 133-135: Please provide details on the ethical approval/name the specific University IRBs that provided approval. Please specify whether informed consent was written.

10. Discussion: Lines 345-346: We suggest revising this sentence to read: “Using Mendelian randomization, we provide evidence that genetically elevated HDL and LDL levels are associated with increased risk for breast cancer…”

11. Discussion: Lines 371-372: We suggest not abbreviating Beeghly-Fadiel et al. “(hereafter B-F)” as this is only used once in the paragraph, and could cause confusion.

12. Checklist: Please ensure that the study is reported according to the STROBE guideline, and include the completed STROBE checklist as Supporting Information (or, please report your study according to the relevant guideline, which can be found here: http://www.equator-network.org/)

When completing the checklist, please use section and paragraph numbers, rather than page numbers.

Please add the following statement, or similar, to the Methods: "This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist)."

13. Supplementary Tables: Please provide titles and legends for each individual table and figure in the Supporting Information, rather than including these as a list in another supporting information file.

Comments from the Academic Editor:

This paper uses Mendelian randomization to assess the causal effects of lipids on breast cancer, but a much stronger case (in Introduction and Discussion) needs to be made to clarify what it adds to the existing knowledge from similar MR studies. The analyses reported are very comprehensive, and yet the rationale and practical implications of the different approaches used is not always clear to the reader – which makes it difficult to judge its added value compared with previous MR work. For example, the authors don’t discuss the implications of using a gene-specific vs. a conventional MR analysis, or a local vs. genome-wide genetic correlation analysis, and why using a gene-specific MR or a local genetic correlation analysis is important. Moreover, the MR and genetic correlation approaches answer different questions, and yet this is not clearly discussed.

Regarding the MR methods used, I think some justification (or reference to a method paper) is required for the method used to perform the “Heterogeneity analyses for single trait MR” described in the Supplement. Similarly, more justification is needed for the pruning used in the multivariable analysis (“stepwise post-hoc procedure to remove genetic instruments that contributed the most to QA”). The tests of instrument strength and validity reported in the Supplement also require some mention to correlation assumptions when applied to two-sample MR.

Comments from the reviewers:

Reviewer #4: The authors have addressed my comments.

Alex Cornish (ICR, London).

Any attachments provided with reviews can be seen via the following link:

[LINK]

PLoS Med. 2020 Sep 11;17(9):e1003302. doi: 10.1371/journal.pmed.1003302.r004

Author response to Decision Letter 1

26 May 2020

Attachment

Submitted filename: Response to editor-round 2.docx

Click here for additional data file.^{(12.6KB, docx)}

PLoS Med. doi: 10.1371/journal.pmed.1003302.r005

Decision Letter 2

Caitlin Moyer

6 Jul 2020

Dear Dr. Voight,

Thank you very much for re-submitting your manuscript "The relationship between circulating lipids and breast cancer risk: a Mendelian randomization study" (PMEDICINE-D-19-03886R2) for review by PLOS Medicine.

I have discussed the paper with my colleagues and the academic editor. I am pleased to say that provided the remaining editorial and production issues are dealt with we are planning to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Please take these into account before resubmitting your manuscript. In particular, please be sure to fully address the point below (point #1) regarding MR and homogeneity assumptions.

[LINK]

Our publications team (plosmedicine@plos.org) will be in touch shortly about the production requirements for your paper, and the link and deadline for resubmission. DO NOT RESUBMIT BEFORE YOU'VE RECEIVED THE PRODUCTION REQUIREMENTS.

In revising the manuscript for further consideration here, please ensure you address the specific points made by the editors. In your rebuttal letter you should indicate your response to the editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by Jul 13 2020 11:59PM.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

------------------------------------------------------------

Requests from Editors and the Academic Editor:

1. Please address the following note from the academic editor regarding Lines 177-179 in the Methods:

[The authors have addressed all the comments, but there is one sentence in Methods about multivariable MR which is misleading - p. 9, first paragraph: “Because MR assumes homogeneity between each instrument’s exposure to outcome effect, we then tested for instrument strength and validity [35], and removed instruments driving heterogeneity (Supplementary Methods).”.

In reference [35] that they cite, Sanderson et al. talk about: a) “good” heterogeneity – where a modified version of Cochran’s Q statistic is used to assess instrument strength; and b) “bad” heterogeneity – where the Cochran’s Q statistic is used for testing instrument validity (e.g. due to pleiotropy), similarly to what proposed for classical (single-exposure) MR (Greco M, Del F, Minelli C, Sheehan NA, Thompson JR. Detecting pleiotropy in Mendelian randomisation studies with summary data and a continuous outcome. Stat Med 2015;34: 2926–40). So it’s unclear what “Because MR assumes homogeneity between each instrument’s exposure to outcome effect” means in the sentence above. I guess what it means is that MR assumes homogeneity in the ratio of the genetic effect on the outcome on the genetic effect on the exposure (Wald estimator) across instruments, which means that all instruments are assumed to be valid - but if this is the case, the whole sentence needs to be re-written.]

2. Data analysis plan: Please update the link for the BCAC summary data: http://bcac.ccge.medschl.cam.ac.uk/bcacdata/oncoarray/gwas-icogs-and-oncoarray-summary-results/

3. Abstract: Line 14 (and anywhere else): Please replace "subject" with participant, patient, individual, or person.

4. Introduction: Lines 108-111: The term “nominal” is used twice here, and the meaning is not clear. If you mean “not statistically significant” please make that clear. Please revise the sentence using a more specific term to indicate your meaning: “This study did not find a nominal association between any lipid trait and breast cancer using lipid summary statistics, though a previous study with a smaller breast cancer GWAS sample size did report a nominal negative genetic correlation between triglycerides and breast cancer risk [16].

5. Methods: Lines 142-144: Please provide the references for these two datasets again here. “For the Willer et al and BCAC datasets, we refer the reader to the primary GWAS manuscripts and their supplementary material for details on consent protocols for each of their respective cohorts.”

6. Discussion: Lines 398-400: Similar to the above comment, can you please clarify the meaning of “nominal” in this sentence: Third, while both studies are consistent in their relationship between HDL and BC, we reported a nominal association with LDL levels when considering all risk and confounding factors jointly.”

7. Checklist: Thank you for including the STROBE-MR extension. We agree this would be the appropriate checklist for your study; however, can you please also include the STROBE checklist, as the STROBE-MR is still preliminary/unpublished?

8. References: Please use the "Vancouver" style for reference formatting, and see our website for other reference guidelines https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references

A few citations seem to be missing information: (e.g. # 9, 10, 29, 47)

9. Supporting information figures 4-9 and 13: Rather than indicating significance with *p<0.05 and **p<0.001, please report the exact p values and 95% CIs associated with the ORs.

Any attachments provided with reviews can be seen via the following link:

[LINK]

PLoS Med. 2020 Sep 11;17(9):e1003302. doi: 10.1371/journal.pmed.1003302.r006

Author response to Decision Letter 2

6 Aug 2020

Attachment

Submitted filename: Response_to_editors_08052020.docx

Click here for additional data file.^{(14.1KB, docx)}

PLoS Med. doi: 10.1371/journal.pmed.1003302.r007

Decision Letter 3

Caitlin Moyer

10 Aug 2020

Dear Dr. Voight,

On behalf of my colleagues and the academic editor, Dr. Cosetta Minelli, I am delighted to inform you that your manuscript entitled "The relationship between circulating lipids and breast cancer risk: a Mendelian randomization study" (PMEDICINE-D-19-03886R3) has been accepted for publication in PLOS Medicine.

PRODUCTION PROCESS

Before publication you will see the copyedited word document (in around 1-2 weeks from now) and a PDF galley proof shortly after that. The copyeditor will be in touch shortly before sending you the copyedited Word document. We will make some revisions at the copyediting stage to conform to our general style, and for clarification. When you receive this version you should check and revise it very carefully, including figures, tables, references, and supporting information, because corrections at the next stage (proofs) will be strictly limited to (1) errors in author names or affiliations, (2) errors of scientific fact that would cause misunderstandings to readers, and (3) printer's (introduced) errors.

If you are likely to be away when either this document or the proof is sent, please ensure we have contact information of a second person, as we will need you to respond quickly at each point.

PRESS

A selection of our articles each week are press released by the journal. You will be contacted nearer the time if we are press releasing your article in order to approve the content and check the contact information for journalists is correct. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact.

PROFILE INFORMATION

Now that your manuscript has been accepted, please log into EM and update your profile. Go to https://www.editorialmanager.com/pmedicine, log in, and click on the "Update My Information" link at the top of the page. Please update your user information to ensure an efficient production and billing process.

Thank you again for submitting the manuscript to PLOS Medicine. We look forward to publishing it.

Best wishes,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 STROBE checklist. Reporting document following the STROBE guidelines for our study.

STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

(DOCX)

Click here for additional data file.^{(32.6KB, docx)}

S1 STROBE-MR checklist. Reporting document following the preliminary STROBE-MR guidelines for our study.

MR, Mendelian randomization; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

(DOCX)

Click here for additional data file.^{(34.6KB, docx)}

S1 Fig. QQ plots for lipid association statistics.

(PDF)

Click here for additional data file.^{(674.8KB, pdf)}

S2 Fig. Scatter plots of unpruned single-trait MR genetic instruments’ effect estimates on exposure and outcome.

(PDF)

Click here for additional data file.^{(22KB, pdf)}

S3 Fig. Scatter plots of pruned single-trait MR genetic instruments’ effect estimates on exposure and outcome.

(PDF)

Click here for additional data file.^{(18.8KB, pdf)}

S4 Fig. Results of single-trait MR analyses with a lipid trait as the exposure and risk for all BCs as the outcome.

(PDF)

Click here for additional data file.^{(4.7KB, pdf)}

S5 Fig. Single-trait MR with lipid association statistics from MVP, GLGC, or MVP + GLGC meta-analysis.

(PDF)

Click here for additional data file.^{(51.8KB, pdf)}

S6 Fig. Multivariable MR analyses stratified by 3 independent subsets of the BCAC data set.

(PDF)

Click here for additional data file.^{(6.3KB, pdf)}

S7 Fig. Multivariable MR analyses using lipid genetic associations from either MVP or GLGC.

(PDF)

Click here for additional data file.^{(299.7KB, pdf)}

S8 Fig. Single-trait MR with BC outcomes stratified by ER subtypes.

(PDF)

Click here for additional data file.^{(5.3KB, pdf)}

S9 Fig. Multivariable MR analyses with BC outcomes stratified by ER subtypes.

(PDF)

Click here for additional data file.^{(5.7KB, pdf)}

S10 Fig. Genetic instruments’ effect estimates on HDL and BC at each canonical HDL metabolism pathway loci.

(PDF)

Click here for additional data file.^{(9.7KB, pdf)}

S11 Fig. Genetic instruments’ effect estimates on LDL and BC at each canonical LDL metabolism pathway loci.

(PDF)

Click here for additional data file.^{(7.9KB, pdf)}

S12 Fig. LDL locus-specific MR results with LDL as exposure and BC risk as outcome.

(PDF)

Click here for additional data file.^{(54.1KB, pdf)}

S13 Fig. Genetic correlations between lipid and BC traits.

(PDF)

Click here for additional data file.^{(50.3KB, pdf)}

S1 Table. Summary statistics for genetic instruments used in single-trait MR analyses.

(XLSX)

Click here for additional data file.^{(38.8KB, xlsx)}

S2 Table. Summary statistics for genetic instruments used in multivariable MR analyses.

(XLSX)

Click here for additional data file.^{(913.8KB, xlsx)}

S3 Table. Conditionally independent summary statistics for HDL or LDL associations used in locus-specific MR analyses.

(XLSX)

Click here for additional data file.^{(17KB, xlsx)}

S4 Table. Single-trait MR results with unpruned lipid traits as the exposure and all BCs as the outcome for a range of MR methods.

(XLSX)

Click here for additional data file.^{(10KB, xlsx)}

S5 Table. Heterogeneity analyses by Cochran's Q of unpruned single-trait MR.

(XLSX)

Click here for additional data file.^{(8.4KB, xlsx)}

S6 Table. Pleiotropy analysis using Egger regression of unpruned single-trait MR.

(XLSX)

Click here for additional data file.^{(8.4KB, xlsx)}

S7 Table. Results of single-trait MR, heterogeneity analyses, and directionality analyses with pruned lipid IV sets.

(XLSX)

Click here for additional data file.^{(9.6KB, xlsx)}

S8 Table. Results of a reciprocal single-trait MR testing the effects of BC as the exposure on each lipid trait as the outcome.

(XLSX)

Click here for additional data file.^{(8.5KB, xlsx)}

S9 Table. Results of single-trait MR analyses after pruning for genetic instruments associated with other lipid traits.

(XLSX)

Click here for additional data file.^{(8.5KB, xlsx)}

S10 Table. Results of multivariable MR with 3 lipid traits, BMI, and AaM, as exposures and BC risk as outcome.

(XLSX)

Click here for additional data file.^{(11.4KB, xlsx)}

S11 Table. Results of multivariable MR with lipid trait summary statistics from distinct cohorts.

(XLSX)

Click here for additional data file.^{(11.7KB, xlsx)}

S12 Table. Results of Cochran's Q test for heterogeneity on single-trait IVW MR results for the effect of a lipid trait on ER+ versus ER− BC.

(XLSX)

Click here for additional data file.^{(8.7KB, xlsx)}

S13 Table. Results of multivariable MR with ER+ or ER− BC as the outcome.

(XLSX)

Click here for additional data file.^{(9.2KB, xlsx)}

S14 Table. Genome-wide genetic correlation estimates.

(XLSX)

Click here for additional data file.^{(9.9KB, xlsx)}

S15 Table. Local genetic correlations calculated with the ρ-Hess method.

(XLSX)

Click here for additional data file.^{(8.6KB, xlsx)}

S1 Text. Supplementary Text.

(DOCX)

Click here for additional data file.^{(54.8KB, docx)}

Attachment

Submitted filename: Response_to_reviewers_2.pdf

Click here for additional data file.^{(254.5KB, pdf)}

Attachment

Submitted filename: Response to editor-round 2.docx

Click here for additional data file.^{(12.6KB, docx)}

Attachment

Submitted filename: Response_to_editors_08052020.docx

Click here for additional data file.^{(14.1KB, docx)}

Data Availability Statement

[pmed.1003302.ref001] 1.Torre LA, Siegel RL, Ward EM, Jemal A. Global Cancer Incidence and Mortality Rates and Trends—An Update. Cancer Epidemiol Biomarkers Prev. 2016;25: 16–27. 10.1158/1055-9965.EPI-15-0578 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref002] 2.Rose D, Gracheck P, Vona-Davis L, Rose DP, Gracheck PJ, Vona-Davis L. The Interactions of Obesity, Inflammation and Insulin Resistance in Breast Cancer. Cancers (Basel). 2015;7: 2147–2168. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref003] 3.Kuzu OF, Noory MA, Robertson GP. The role of cholesterol in cancer. Cancer Research. 2016;76(8): 2063–2070. 10.1158/0008-5472.CAN-15-2613 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref004] 4.Touvier M, Fassier P, His M, Norat T, Chan DSM, Blacher J, et al. Cholesterol and breast cancer risk: a systematic review and meta-analysis of prospective studies. Br J Nutr. 2015;114: 347–357. 10.1017/S000711451500183X [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref005] 5.Ni H, Liu H, Gao R. Serum Lipids and Breast Cancer Risk: A Meta-Analysis of Prospective Cohort Studies. Singh S, editor. PLoS ONE. 2015;10: e0142669 10.1371/journal.pone.0142669 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref006] 6.Martin LJ, Melnichouk O, Huszti E, Connelly PW, Greenberg CV., Minkin S, et al. Serum Lipids, Lipoproteins, and Risk of Breast Cancer: A Nested Case-Control Study Using Multiple Time Points. JNCI J Natl Cancer Inst. 2015;107: djv032–djv032. 10.1093/jnci/djv032 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref007] 7.Zhong S, Zhang X, Chen L, Ma T, Tang J, Zhao J. Statin use and mortality in cancer patients: Systematic review and meta-analysis of observational studies. Cancer Treatment Reviews. 2015;41(6): 554–567. 10.1016/j.ctrv.2015.04.005 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref008] 8.Borgquist S, Giobbie-Hurder A, Ahern TP, Garber JE, Colleoni M, Láng I, et al. Cholesterol, Cholesterol-Lowering Medication Use, and Breast Cancer Outcome in the BIG 1–98 Study. J Clin Oncol. 2017;35: 1179–1188. 10.1200/JCO.2016.70.3116 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref009] 9.Orho-Melander M, Hindy G, Borgquist S, Schulz C-A, Manjer J, Melander O, et al. Blood lipid genetic scores, the HMGCR gene and cancer risk: a Mendelian randomization study. Int J Epidemiol. 2018;47(2): 495–505. Epub 2017 Nov 20. 10.1093/ije/dyx237 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref010] 10.Nowak C, Ärnlöv J. A Mendelian randomization study of the effects of blood lipids on breast cancer risk. Nat Commun. 2018;9(1): 3957 10.1038/s41467-018-06467-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref011] 11.Michailidou K, Lindström S, Dennis J, Beesley J, Hui S, Kar S, et al. Association analysis identifies 65 new breast cancer risk loci. Nature. 2017;551: 92–94. 10.1038/nature24284 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref012] 12.Global Lipids Genetics Consortium, Willer CJ, Schmidt EM, Sengupta S, Peloso GM, Gustafsson S, Kanoni S, et al. Discovery and Refinement of Loci Associated with Lipid Levels. Nat Genet. 2013;45: 1274–1283. 10.1038/ng.2797 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref013] 13.Qi G, Chatterjee N. Mendelian randomization analysis using mixture models for robust and efficient estimation of causal effects. Nat Commun. 2019;10: 1941 10.1038/s41467-019-09432-2 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref014] 14.Neuhouser ML, Aragaki AK, Prentice RL, Manson JE, Chlebowski R, Carty CL, et al. Overweight, Obesity, and Postmenopausal Invasive Breast Cancer Risk. JAMA Oncol. 2015;1: 611 10.1001/jamaoncol.2015.1546 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref015] 15.Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh PR, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015;47: 1236–1241. 10.1038/ng.3406 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref016] 16.Lindström S, Finucane H, Bulik-Sullivan B, Schumacher FR, Amos CI, Hung RJ, et al. Quantifying the genetic correlation between multiple cancer types. Cancer Epidemiol Biomarkers Prev. 2017;26: 1427–1435. 10.1158/1055-9965.EPI-17-0211 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref017] 17.Day FR, Thompson DJ, Helgason H, Chasman DI, Finucane H, Sulem P, et al. Genomic analyses identify hundreds of variants associated with age at menarche and support a role for puberty timing in cancer risk. Nat Genet. 2017;49: 834–841. 10.1038/ng.3841 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref018] 18.Burgess S, Thompson DJ, Rees JMB, Day FR, Perry JR, Ong KK. Dissecting causal pathways using mendelian randomization with summarized genetic data: Application to age at menarche and risk of breast cancer. Genetics. 2017;207: 481–487. 10.1534/genetics.117.300191 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref019] 19.Feng Y, Hong X, Wilker E, Li Z, Zhang W, Jin D, et al. Effects of age at menarche, reproductive years, and menopause on metabolic risk factors for cardiovascular diseases. Atherosclerosis. 2008;196: 590–597. 10.1016/j.atherosclerosis.2007.06.016 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref020] 20.Hamajima N, Hirose K, Tajima K, Rohan T, Friedenreich CM, Calle EE, et al. Menarche, menopause, and breast cancer risk: Individual participant meta-analysis, including 118 964 women with breast cancer from 117 epidemiological studies. Lancet Oncol. 2012;13: 1141–1151. 10.1016/S1470-2045(12)70425-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref021] 21.Jiang X, Finucane HK, Schumacher FR, Schmit SL, Tyrer JP, Han Y, et al. Shared heritability and functional enrichment across six solid cancers. Nat Commun. 2019;10: 431 10.1038/s41467-018-08054-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref022] 22.Klarin D, Damrauer SM, Cho K, Sun Y V., Teslovich TM, Honerlaw J, et al. Genetics of blood lipids among ~300,000 multi-ethnic participants of the Million Veteran Program. Nat Genet. 2018;50: 1514–1523. 10.1038/s41588-018-0222-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref023] 23.Yengo L, Sidorenko J, Kemper KE, Zheng Z, Wood AR, Weedon MN, et al. Meta-analysis of genome-wide association studies for height and body mass index in ~700000 individuals of European ancestry. Hum Mol Genet. 2018;27: 3641–3649. 10.1093/hmg/ddy271 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref024] 24.Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet. 2007;81: 559–575. 10.1086/519795 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref025] 25.Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife. 2018;7: e34408 10.7554/eLife.34408 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref026] 26.Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44: 512–525. 10.1093/ije/dyv080 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref027] 27.Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genet Epidemiol. 2016;40: 304–314. 10.1002/gepi.21965 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref028] 28.Hartwig FP, Smith GD, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46(6): 1985–1998. 10.1093/ije/dyx102 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref029] 29.Walter K, Min JL, Huang J, Crooks L, Memari Y, McCarthy S, et al. The UK10K project identifies rare variants in health and disease. Nature. 2015;526: 82–90. 10.1038/nature14962 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref030] 30.Voight BF. MR-predictor: A simulation engine for Mendelian Randomization studies. Bioinformatics. 2014;30: 3432–3434. 10.1093/bioinformatics/btu564 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref031] 31.Burgess S. Sample size and power calculations in Mendelian randomization with a single instrumental variable and a binary outcome. Int J Epidemiol. 2014;43: 922–929. 10.1093/ije/dyu005 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref032] 32.Burgess S, Thompson SG. Multivariable Mendelian randomization: The use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181: 251–260. 10.1093/aje/kwu283 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref033] 33.Burgess S, Dudbridge F, Thompson SG. Re: “Multivariable Mendelian randomization: The use of pleiotropic genetic variants to estimate causal effects.” Am J Epidemiol. 2015;181: 290–291. 10.1093/aje/kwv017 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref034] 34.Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. Int J Epidemiol. 2019;48: 713–727. 10.1093/ije/dyy262 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref035] 35.Schwarzer G. meta: An R package for meta-analysis. R News. 2007;7: 40–45. [Google Scholar]

[pmed.1003302.ref036] 36.Shi H, Mancuso N, Spendlove S, Pasaniuc B. Local Genetic Correlation Gives Insights into the Shared Genetic Architecture of Complex Traits. Am J Hum Genet. 2017;101: 737–751. 10.1016/j.ajhg.2017.09.022 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref037] 37.Berisa T, Pickrell JK. Approximately independent linkage disequilibrium blocks in human populations. Bioinformatics. 2016;32: 283–5. 10.1093/bioinformatics/btv546 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref038] 38.The GTEx Consortium, Aguet F, Ardlie KG, Cummings BB, Gelfand ET, Getz G, et al. Genetic effects on gene expression across human tissues. Nature. 2017;550: 204–213. 10.1038/nature24277 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref039] 39.von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies. PLoS Med. 2007;4: e296 10.1371/journal.pmed.0040296 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref040] 40.Davey Smith G, Davies N, Dimou N, Egger M, Gallo V, Golub R, et al. STROBE-MR: Guidelines for strengthening the reporting of Mendelian randomization studies. PeerJ Preprints 27857 [Preprint]. 2019 [cited 2020 May 18]. https://peerj.com/preprints/27857/

[pmed.1003302.ref041] 41.Pickrell JK, Berisa T, Liu JZ, Ségurel L, Tung JY, Hinds DA. Detection and interpretation of shared genetic influences on 42 human traits. Nat Genet. 2016;48: 709–717. 10.1038/ng.3570 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref042] 42.Hemani G, Tilling K, Davey Smith G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. Li J, editor. PLoS Genet. 2017;13: e1007081 10.1371/journal.pgen.1007081 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref043] 43.Guo Y, Warren Andersen S, Shu XO, Michailidou K, Bolla MK, Wang Q, et al. Genetically Predicted Body Mass Index and Breast Cancer Risk: Mendelian Randomization Analyses of Data from 145,000 Women of European Descent. PLoS Med. 2016;13 10.1371/journal.pmed.1002105 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref044] 44.Renehan AG, Tyson M, Egger M, Heller RF, Zwahlen M. Body-mass index and incidence of cancer: a systematic review and meta-analysis of prospective observational studies. Lancet. 2008;371: 569–78. 10.1016/S0140-6736(08)60269-X [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref045] 45.Borgquist S, Butt T, Almgren P, Shiffman D, Stocks T, Orho-Melander M, et al. Apolipoproteins, lipids and risk of cancer. Int J Cancer. 2016;138: 2648–2656. 10.1002/ijc.30013 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref046] 46.Melvin JC, Seth D, Holmberg L, Garmo H, Hammar N, Jungner I, et al. Lipid profiles and risk of breast and ovarian cancer in the swedish AMORIS study. Cancer Epidemiol Biomarkers Prev. 2012;21: 1381–1384. 10.1158/1055-9965.EPI-12-0188 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref047] 47.Zhang B-L, He N, Huang Y-B, Song F-J, Chen K-X. ABO blood groups and risk of cancer: a systematic review and meta-analysis. Asian Pac J Cancer Prev. 2014;15: 4643–50. 10.7314/apjcp.2014.15.11.4643 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref048] 48.Beeghly-Fadiel A, Khankari NK, Delahanty RJ, Shu X-O, Lu Y, Schmidt MK, et al. A Mendelian randomization analysis of circulating lipid traits and breast cancer risk. Int J Epidemiol. Epub 2019 Dec 23. 10.1093/ije/dyz242 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref049] 49.Madsen CM, Varbo A, Nordestgaard BG. Extreme high high-density lipoprotein cholesterol is paradoxically associated with high mortality in men and women: two prospective cohort studies. Eur Heart J. 2017;38: 2478–2486. 10.1093/eurheartj/ehx163 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref050] 50.Hamer M, O’Donovan G, Stamatakis E. High-Density Lipoprotein Cholesterol and Mortality: Too Much of a Good Thing? Arterioscler Thromb Vasc Biol. 2018;38: 669–672. 10.1161/ATVBAHA.117.310587 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref051] 51.Burgess S, Davey Smith G. Mendelian Randomization Implicates High-Density Lipoprotein Cholesterol–Associated Mechanisms in Etiology of Age-Related Macular Degeneration. Ophthalmology. 2017;124: 1165–1174. 10.1016/j.ophtha.2017.03.042 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref052] 52.Fan Q, Maranville JC, Fritsche L, Sim X, Cheung CMG, Chen LJ, et al. HDL-cholesterol levels and risk of age-related macular degeneration: a multiethnic genetic study using Mendelian randomization. Int J Epidemiol. 2017;46: 1891–1902. 10.1093/ije/dyx189 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref053] 53.Bonovas S, Filioussi K, Tsavaris N, Sitaras NM. Use of Statins and Breast Cancer: A Meta-Analysis of Seven Randomized Clinical Trials and Nine Observational Studies. J Clin Oncol. 2005;23: 8606–8612. 10.1200/JCO.2005.02.7045 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref054] 54.Llewellyn-Bennett R, Edwards D, Roberts N, Hainsworth AH, Bulbulia R, Bowman L. Post-trial follow-up methodology in large randomised controlled trials: a systematic review. Trials. 2018;19: 298 10.1186/s13063-018-2653-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref055] 55.VanderWeele TJ, Tchetgen Tchetgen EJ, Cornelis M, Kraft P. Methodological challenges in mendelian randomization. Epidemiology. 2014;25: 427–35. 10.1097/EDE.0000000000000081 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref056] 56.O’Connor LJ, Price AL. Distinguishing genetic correlation from causation across 52 diseases and complex traits. Nat Genet. 2018;50: 1728–1734. 10.1038/s41588-018-0255-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref057] 57.Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, Davey Smith G. Best (but oft-forgotten) practices: the design, analysis, and interpretation of Mendelian randomization studies. Am J Clin Nutr. 2016;103: 965–978. 10.3945/ajcn.115.118216 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1003302.ref058] 58.Silvente-Poirot S, Poirot M. Cancer. Cholesterol and cancer, in the balance. Science. 2014;343(6178): 1445–1446. 10.1126/science.1252787 [DOI] [PubMed] [Google Scholar]

[pmed.1003302.ref059] 59.Nelson ER. The significance of cholesterol and its metabolite, 27-hydroxycholesterol in breast cancer. Mol Cell Endocrinol. 2018;466: 73–80. 10.1016/j.mce.2017.09.021 [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

The relationship between circulating lipids and breast cancer risk: A Mendelian randomization study

Kelsey E Johnson

Katherine M Siewert

Derek Klarin

Scott M Damrauer

Kyong-Mi Chang

Philip S Tsao

Themistocles L Assimes

Kara N Maxwell

Benjamin F Voight

Roles

Abstract

Background

Methods and findings

Conclusions

Author summary

Why was this study done?

What did the researchers do and find?

What do these findings mean?

Introduction

Methods

Study populations

Lipid meta-analysis

MR analyses

Core HDL and LDL pathway genetic instrument development

Genetic correlation analyses

Analysis plan

Results

Single-trait MR in BC

Fig 1. Results of MR analyses of the effects of lipids on BC risk.

Multivariable MR with age at menarche and BMI as exposures

MR with outcome stratified by ER status

HDL and LDL pathway gene-specific MR

Fig 2. MR results for HDL gene-specific instruments and meta-analysis of effect estimates across genes.

Genome-wide and local genetic correlation

Discussion

Supporting information

Abbreviations

Data Availability

Funding Statement

References

Decision Letter 0

Caitlin Moyer

Roles

Author response to Decision Letter 0

Decision Letter 1

Caitlin Moyer

Roles

Author response to Decision Letter 1

Decision Letter 2

Caitlin Moyer

Roles

Author response to Decision Letter 2

Decision Letter 3

Caitlin Moyer

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases