Estimating the effect of lipids on IGF axis and subsequent breast cancer risk

Vanessa Y Tan, Caroline J Bull, Kalina M Biernacka, Alexander Teumer, Laura Corbin, Tom Dudding, Eleanor Sanderson, Qibin Qi, Robert C Kaplan, Jerome I Rotter, Nele Friedrich, Uwe Völker, Julia Mayerle, Claire M Perks, Jeff MP Holly, Nicholas J Timpson
doi: https://doi.org/10.1101/2020.06.04.20122630

Abstract

Circulating lipids have been associated with breast cancer (BCa). This association may, in part, be due to an effect of lipids on insulin-like growth factors (IGFs), which have been reliably associated with BCa. In two-sample Mendelian randomization (MR) analyses, we found that low density lipoprotein (LDL-C) was associated with IGFBP-3 (beta:0.08 SD; 95%CI:0.02,0.15; p = 0.01, per SD increase in LDL-C) and IGFBP-3 was associated with postmenopausal BCa (OR:1.09; 95%CI:1.00,1.19; p = 0.05, per SD increase in IGFBP-3). We also found that triglycerides were associated with IGF-I (beta:-0.13SD; 95%CI:-0.22,-0.03, per SD increase in triglycerides) and that IGF-I was associated with overall BCa (OR:1.10;95%CI:1.02,1.18, per SD increase in IGF-I). Taken together, these results suggest that IGFBP-3 may be a potential causal step between LDL-C and postmenopausal BCa and IGF-I a potential causal for triglycerides. Our two-step MR results build on evidence linking circulating lipids and IGFs with BCa, however, multivariable MR analyses are currently unable to support this relationship due to weak instruments.

Introduction

Breast cancer (BCa) is one of the leading causes of cancer-related deaths worldwide1,2. Circulating lipids including low density lipoprotein (LDL-C), high density lipoprotein (HDL-C) and triglycerides have been associated with known risk factors for BCa such as obesity3,4. However, evidence from observational studies investigating the relationship between lipid profile and BCa risk has been conflicting5,6. Statins competitively inhibit 3-hydroxy-3-methylglutaryl-coenzyme A (HMGCR) to reduce circulating LDL-C and are commonly prescribed for the prevention and treatment of cardiovascular disease, but have also been hypothesised to have anti-cancer effects7. Most observational studies have found no evidence of an association between statin use and BCa risk8. However, reverse causation, confounding and other biases may limit the interpretation of these observational findings. The associations, estimated from these observational studies, between statin therapy and breast cancer are particularly problematic, due to selection and so called immortal time biases9.

Mendelian randomization (MR) is an analytical approach where genetic variants reliably associated with risk factors of interest are used as “proxy” measures to investigate whether the effect of an exposure on an outcome is likely to be causal10. A recent MR study reported a positive association between LDL-C and HDL-C with overall or estrogen receptor (ER)-positive BCa risk11. It is becoming increasingly feasible to study drug effects in a MR framework using genetic variants located in the genes of the drug’s protein targets12. Genetic variants in HMGCR, associated with increased LDL-C levels, have been found to associate with increased overall and ER-positive BCa risk13,14. The mechanism by which statins protect against BCa and whether the effect differs by menopausal status remains to be elucidated.

Several randomised control trials have suggested that statin use is associated with alterations in circulating levels of insulin-like growth factors (IGFs), suggesting that perturbation of circulating lipids can alter levels of IGF axis traits15–17. The IGF axis includes IGF-I, IGF-II and six IGF binding proteins (IGFBPs), which are important for cancer initiation and progression through the regulation of cell growth, metabolism and survival18–21. A recent meta-analysis of individual data from observational studies found evidence that IGF-I was associated with increased BCa risk in both pre and postmenopausal women while IGFBP-3 was only associated with increased BCa risk in postmenopausal women22. However, it is unclear if a causal relationship between circulating IGFs and BCa exists, and whether estrogen receptor or menopausal status is important.

The role of the IGF axis in metabolic regulation is well-established23–25. Intervention studies in patients with growth hormone disorders support a link between IGF and lipid biology26–28. Population-based studies examining the relationship between circulating IGFs and lipid profile have yielded conflicting results29–31, however, these studies may be limited by their cross-sectional design. Hence, the direction of association and whether causation exists between circulating lipids and IGFs is still unclear. Given previous evidence implicating lipids and IGFs as potential modifiable risk factors for cancer32–34, there is motivation to assess the relationship between IGFs and lipids to investigate the mechanisms underlying BCa risk.

We aimed to investigate circulating IGFs as potential intermediates between lipid profile and BCa risk. We set out to examine the association between circulating lipid traits (LDL-C, HDL-C and TG), IGFs (IGF-I and IGFBP-3) and BCa (stratified by menopausal and ER status) using a two-step MR study design35.

Materials and Methods

Study design and data sources

To estimate the causal relationship between lipids and BCa using MR Figure 1,(i)), we obtained summary statistics from: i) a lipid GWAS conducted by the Global Lipids Genetic Consortium (GLGC, n = 188,577)36; ii) a BCa risk GWAS (overall (n = 228,951) and stratified by estrogen receptor status (n = 69,501 for ER positive BCa and n = 21,468 for ER negative BCa) conducted by the Breast Cancer Association Consortium (BCAC,37; and iii) a BCa risk GWAS (overall (n = 230,954) and stratified by menopausal status (premenopausal BCa, n = 59,124 for premenopausal BCa and n = 112,245 for postmenopausal BCa) conducted in UK Biobank (see Supplementary methods).

Figure 1.
Figure 1. Flow diagram of study design.

To investigate the observational relationship between lipids and IGFs (Figure 1, (ii)), we used individual-level data (n = 3,812) from the Study of Health in Pomerania (SHIP) where circulating lipids and IGFs were measured (these data are described in detail in the Supplementary methods).

To estimate the causal relationship between lipids and IGF-I or IGFBP-3 using MR (Figure 1, (iii)), we obtained summary statistics from: i) the lipid GWAS conducted by the GLGC36, as described above and ii) the IGF GWAS conducted by the “IGF working group” of the CHARGE consortium (n = 30,884)38 (see Supplementary methods).

To estimate the causal relationship between circulating IGFs and BCa (Figure 1,(iv)), we obtained summary statistics from: i) the IGF GWAS(29); ii) the BCa risk GWAS conducted by BCAC37 (overall and stratified by ER status); and iii) the BCa risk GWAS conducted in UKBB (overall and stratified by menopausal status)(see Supplementary methods).

To assess the potential causal effect of changes in LDL-C levels due to lipid lowering drugs such as statins on BCa and IGF levels (Figure 1,(v)), we constructed a genetic instrument to proxy statin treatment which comprised of 5 SNPs (rs12916, rs10515198, rs12173076, rs3857388, and rs7711235) within 100kb from the gene region for HMGCR and in low linkage disequilibrium (r2< 0.20)39. These SNPs were robustly associated (p< 5×10−8) with LDL-C in the lipid GWAS conducted by GLGC36. We obtained summary statistics for the genetic variants in HMGCR from: i) the lipid GWAS [29]; ii) the IGF GWAS(29); iii) the BCa risk GWAS conducted by BCAC37; and iv) the BCa risk GWAS conducted in UKBB (see Supplementary methods).

Genetic instruments for MR analyses

The GLGC identified 185 SNPs (r2 < 0.2) associated with at least HDL-C, LDL-C or TG (p< 5×10−8)36 (see Supplementary Table 1 for list of SNPs). Due to the complex overlapping nature of the lipid traits, genetic variants are commonly associated with more than one lipid trait. To go some way to disentangling the roles of LDL-C, HDL-C and TG, we used multivariable MR which was developed to estimate the causal effect of various correlated risk factors when conditioned on one another in a single model40,41 (see Supplementary methods for more details). For the multivariable MR analyses (Figure 1, (i)), all 185 SNPs were included in the model. Univariable MR methods included SNPs associated with a given lipid trait at a genome-wide level of significance (p< 5×10−8, 76 for LDL-C, 86 for HDL-C or 51 for TG). Sex-specific GWAS results were not available for lipid traits.

For the IGF MR analyses (Figure 1, (iii)&(iv)), we used 7 and 4 independent SNPs (r2< 0.01), associated (p< 5×10−8) with IGF-I and IGFBP-3, respectively, as identified by a IGFs GWAS (see Supplementary Table 2 for list of SNPs)38. IGF-II and IGFBP-2 were not investigated as GWAS summary results are not currently available for these traits. For IGFBP-3, marked differences in effect sizes by sex were found for 2 variants (with stronger associations in women). As differences in effect size could potentially bias MR estimates42, analyses estimating the causal relationship between IGF-I or IGFBP-3 and BCa were conducted using female-specific effect estimates for IGF-I and IGFBP-3.

Statistical analyses

Multivariable MR was used to estimate the effect of LDL-C, HDL-C and TG on the odds of BCa when conditioned on one another (Figure 1, (i)). As there were likely to be differences between the causal estimates between BCAC and UKBB, a random-effects meta-analysis was used to combine MR estimates using BCAC and UKBB summary statistics using the ‘metan’ package in Stata. Analyses stratified by ER status (in BCAC) and menopausal status (in UKBB) were also conducted.

Observational associations between lipids and IGF-I or IGFBP-3 were assessed using linear regression (Figure 1, (ii)). Fully adjusted models included age, sex, smoking status, body mass index (BMI) and diabetes status. Associations of lipids and IGFs with potential confounders were estimated using linear regression.

Bidirectional MR was performed to assess the causal effect of lipids on IGF-I and IGFBP-3 and vice versa (Figure 1, (iii)). SNP-exposure and SNP-outcome associations were combined using the inverse variance weighted (IVW) method43. MR-Egger44, weighted median45 and multivariable MR40,41 were performed as sensitivity analyses.

To assess the causal effect of IGFs on BCa, SNP-exposure (SNP-IGF) and SNP-outcome (SNP-BCa) associations were combined using inverse variance weighting (Figure 1, (iv))43. MR-Egger and weighted median were not conducted as these lack power for instruments with a small number of SNPs46. SNPs associated with IGF-I are also associated with other components of the IGF pathway38. To attempt to account for the overlapping associations for these SNPs, we employed multivariable MR by regression of the SNP-BCa estimates on SNP-IGF-I and SNP-IGFBP-3 estimates in a multivariable weighted regression model. A random-effects meta-analysis was used to combine BCAC and UKBB estimates. Analyses stratified by ER status (in BCAC) and menopausal status (in UKBB) were also conducted.

Multivariable MR47 was conducted as an extension of the IVW method to test the hypothesis that circulating IGF-I or IGFBP3 act as intermediate factors in any association between lipid profile and BCa, identified by the analyses outlined above (Figure 1, (i-iv)). Two-sample Conditional F-statistics48 were estimated to predict instrument strength in the multivariable model.

To estimate the causal relationship between statin use and BCa or IGF levels, we used a random-effect linear regression model weighted by the inverse variance of the SNP-BCa associations. The genetic correlation between the 5 SNPs in HMGCR was incorporated into the MR model using the weighted generalized linear regression method implemented in the ‘MendelianRandomization’ R package49,50.

All MR analyses were performed using the MR-Base “TwoSampleMR” package51. All other statistical analyses were performed using Stata version 14 (StataCorp, College Station, Texas, USA) or R version 3.2.4.

Results

Re-assessing the causal relationship between circulating lipids with breast cancer

The relationship between circulating lipids and BCa in BCAC and UKBB was estimated using multivariable MR. In BCAC, the odds ratio (OR) for overall BCa per SD increase in LDL-C, HDL-C and TG was 1.06 (95%CI 1.01, 1.12; p = 0.02), 1.07 (95%CI 1.01, 1.14; p = 0.03) and 0.93 (95%CI: 0.86, 1.00; p = 0.06), respectively (Figure 2). In UKBB, MR estimates of the association between LDL-C, HDL-C, and TG and BCa were consistent with results in BCAC (Figure 2). In a meta-analysis of BCAC and UKBB data, the OR for overall BCa per SD increase in LDL-C, HDL-C and TG was 1.07 (95%CI 1.02, 1.12; p = 0.01), 1.06 (95%CI 1.01, 1.10; p = 0.01) and 0.94 (95%CI 0.89, 1.00; p = 0.05), respectively (Figure 2).

Figure 2.
Figure 2. Estimates of the causal relationship between circulating lipid traits and breast cancer risk in BCAC and UK Biobank.

The forest plot shows the estimate of the causal effect of LDL-C, HDL-C and TG on overall breast cancer risk using summary data from the breast cancer association consortium (BCAC) (n=122,977 cases and 105,974 controls) and UK Biobank (UKBB) (n = 15,349 cases and 215,605 controls). Squares represent point estimates from individual analyses. Horizontal lines represent the 95% confidence intervals. Pooled estimates (represented by diamonds) were obtained using random-effects meta-analysis.

Using data from BCAC, we extended this to the investigation of the relationship between circulating lipids and BCa, stratified by ER status. The OR for ER-positive BCa per SD increase in LDL-C, HDL-C and TG was 1.06 (95%CI 0.99, 1.12; p = 0.07), 1.08 (95%CI 1.01, 1.15; p = 0.04) and 0.93 (95%CI 0.86, 1.02; p = 0.11), respectively. The OR for ER-negative BCa per SD increase in LDL-C, HDL-C and TG was 1.05 (95%CI 0.98, 1.13; p = 0.16), 1.08 (95%CI 0.99, 1.18; p = 0.09) and 0.92 (95%CI 0.83, 1.02; p = 0.11), respectively (Table 1).

View this table:
Table 1. Estimates of the causal relationship between lipids and breast cancer stratified by ER status or menopausal status.

Using data from UKBB, we investigated the relationship between circulating lipids and BCa, stratified by menopausal status. We found evidence that LDL-C was associated with increased odds of postmenopausal BCa (OR 1.14, 95%CI 1.04, 1.25; p = 0.004), but there was weak evidence for premenopausal BCa (OR 1.07, 95%CI 0.91, 1.26; p = 0.42). There was little evidence to suggest that HDL-C or TG associated with either premenopausal or postmenopausal BCa (Table 1).

Observational relationship between circulating lipids and levels of IGF-I and IGFBP-3 in the SHIP study

Study characteristics of the SHIP study are shown in Supplementary Table 3. The mean (SD) IGF-I and IGFBP-3 levels were 142.1 (57.6) ng/ml and 1884.9 (503.5) ng/ml, respectively. The mean (SD) LDL-C, HDL-C and TG levels were 3.57 (1.16) mmol/l, 1.45 (0.44) mmol/l and 1.82 (1.30) mmol/l, respectively. For IGF-I, a SD unit increase in LDL-C, HDL-C and TG was associated with a –0.11 (95%CI –0.14, –0.08; p = 2.27×10−12), 0.02 (95%CI –0.01, 0.05; p = 0.29) and –0.16 (95%CI –0.19, –0.13; p = 2.26×10−22) SD unit change in IGF-I levels, respectively. For IGFBP-3, a SD unit increase in LDL-C, HDL-C and TG was associated with a –0.01 (95%CI –0.04, 0.02; p = 0.47), 0.06 (95%CI 0.03, 0.10; p = 7.01×10−5) and 0.03 (95%CI –0.003, 0.06; p = 0.08) SD unit change in IGFBP-3 levels, respectively. Circulating lipids and IGF traits were associated with potential confounders of a lipid- or IGFBCa relationship, including age, sex, smoking status, diabetes status and body mass index(Supplementary Table 4). In the fully adjusted model, a SD unit increase in LDL-C, HDL-C and TG was associated with a 0.03 (95%CI 0.004, 0.06; p = 0.03), –0.05 (95%CI –0.08, –0.02; p = 0.001) and –0.06 (95%CI –0.09, –0.04; p = 1.5×10−5) SD unit change in IGF-I levels, respectively. For IGFBP-3, a SD unit increase in LDL-C, HDL-C and TG was associated with a 0.08 (95%CI 0.05, 0.11; p = 1.12×10−7), –0.02 (95%CI –0.05, 0.02; p = 0.30) and 0.13 (95%CI 0.10, 0.16; p = 2.69×10−15) SD unit change in IGFBP-3 levels, respectively (Table 2).

View this table:
Table 2. Beta estimates of SD unit change in IGF-I and IGFBP-3 per SD unit increase in HDL-C, LDL-C or TG based on observational analyses in SHIP.

Estimating the causal relationship between circulating lipids and IGFs

There was little evidence to suggest that LDL-C or HDL-C alter IGF-I levels. The MR estimate of the causal relationship between TG and IGF-I was –0.13 SD units (95%CI –0.22, –0.03; p = 0.01) per SD unit increase in TG. Estimates obtained using MR-Egger regression, weighted median and multivariable MR methods were of similar magnitude (Table 3). There was little evidence for directional pleiotropy using Egger regression (Intercept: 0.007; se: 0.004; directionality p = 0.11). In the reverse direction, there was little evidence to suggest that IGF-I altered circulating LDL-C, HDL-C or TG (Table 4).

View this table:
Table 3. Beta estimates of SD unit change in IGF-I and IGFBP-3 per SD unit increase in HDL-C, LDL-C or TG based on two-sample and multivariable Mendelian randomization analyses.
View this table:
Table 4. Beta estimates of SD unit change in LDL-C, HDL-C and triglycerides per SD unit increase in IGF-I or IGFBP-3 based on two-sample and multivariable Mendelian randomization analyses.

The MR estimate of the causal relationship between LDL-C and IGFBP-3 was 0.08 SD units (95%CI 0.02, 0.15; p = 0.01) per SD unit increase in LDL. There was little evidence for an association between HDL-C and IGFBP-3 (Table 3). The IVW estimate of the causal relationship between TG and IGFBP-3 was 0.10 SD units (95%CI 0.01, 0.19; p = 0.03) per SD unit increase in TG. Estimates obtained using MR-Egger regression, weighted median and multivariable MR methods were consistent with the IVW analyses for LDL-C and TG (Table 3). In the reverse direction, there was minimal evidence to suggest that IGFBP-3 altered circulating LDL-C, HDL-C or TG (Table 4).

Estimating the causal relationship between circulating IGFs and breast cancer

Using data from BCAC, the OR for overall BCa per SD increase in IGF-I and IGFBP-3 (using the IVW method) was 1.10 (95%CI 1.01, 1.20; p = 0.02) and 1.00 (95%CI 0.96, 1.03; p = 0.95), respectively. Estimates from multivariable MR analyses were consistent with the IVW analyses unadjusted for IGF-I (OR 1.12, 95%CI 1.03, 1.20; p = 0.01) and IGFBP-3 (OR 0.98, 95%CI 0.95, 1.02; p = 0.40). In UKBB, estimates for the effect of IGF-I on BCa were comparable. For IGFBP-3, the OR for overall BCa per SD increase in IGFBP-3 was 1.08 (95%CI 1.01, 1.15; p = 0.03) and 1.08 (95%CI 1.01, 1.16; p = 0.02), estimated by IVW and multivariable MR, respectively (Table 5). When results from BCAC and UKBB were meta-analysed, the OR for overall BCa for IGF-I and IGFBP-3 was 1.10 (95%CI 1.02, 1.18; p = 0.01) and 1.02 (95%CI 0.99, 1.06; p = 0.21), respectively, from IVW analyses (Figure 3).

View this table:
Table 5. Estimates of the causal relationship between IGF-I or IGFBP-3 and breast cancer stratified by ER or menopausal status.
Figure 3.
Figure 3. Estimates of the causal relationship between circulating IGF traits and breast cancer risk in BCAC and UK Biobank.

The forest plot shows the estimate of the causal effect of IGF-I and IGFBP-3 on breast cancer risk using summary data from the breast cancer association consortium (BCAC) (n = 122,977 cases and 105,974 controls) and UK Biobank (UKBB) (n = 15,349 cases and 215,605 controls). Squares represent point estimates from individual analyses. Horizontal lines represent the 95% confidence intervals. Pooled estimates (represented by diamonds) were obtained using random effects meta-analysis.

We also investigated the relationship between IGF traits and BCa by ER-status. The OR for ER-positive BCa per SD increase in IGF-I and IGFBP-3 (using the IVW method) was 1.10 (95%CI: 0.99, 1.20; p = 0.06) and 1.00 (95%CI: 0.96, 1.04; p = 0.94), respectively. The OR for ER-negative BCa per SD increase in IGF-I and IGFBP-3 (using the IVW method) was 1.05 (95%CI: 0.90, 1.23; p = 0.53) and 0.98 (95%CI: 0.92, 1.04; p = 0.43), respectively. For IGF-I and IGFBP-3, multivariable MR and IVW estimates were consistent for both ER-positive and ER-negative BCa (Table 5).

Using data from UKBB, we investigated the relationship between IGF traits and BCa stratified by menopausal status. The OR for postmenopausal BCa per SD increase in IGF-I and IGFBP-3 (using the IVW method) was 0.97 (95%CI 0.78, 1.20; p = 0.77) and 1.09 (95%CI 1.00, 1.19; p = 0.05), respectively. The OR for premenopausal BCa per SD increase in IGF-I and IGFBP-3 (using the IVW method) was 1.19 (95%CI 0.76, 1.87; p = 0.46) and 0.93 (95%CI 0.77, 1.12; p = 0.45), respectively. For IGF-I and IGFBP-3, multivariable MR and IVW estimates were consistent for both pre- and post-menopausal BCa (Table 5).

Multivariable MR analyses to estimate whether IGFs are on the causal pathway between lipids and BCa

As a sensitivity analysis, MVMR analyses were conducted to investigate if the effect of TG on overall BCa was attenuated following adjustment for IGF-I. Using data from BCAC, the multivariable MR OR for overall BCa per SD increase in TG, conditioned on IGF-I, was 0.92 (95%CI 0.86, 1.10; p = 0.03). In UKBB, estimates for the effect of TG on overall BCa after conditioning on IGF-I were comparable. When results from BCAC and UKBB were meta-analysed, the multivariable MR OR for overall BCa per SD increase in TG, conditioned on IGF-I, was 0.93 (95%CI 0.88, 0.99; p = 0.03), which did not differ in size or magnitude when compared to the IGF-I unadjusted model (OR 0.94; 95%CI 0.89, 1.00; p = 0.05) (Table 6).

View this table:
Table 6. Multivariable MR analysis of the association of LDL-C and TG with breast cancer conditioned on IGFs.

Using data from UKBB, MVMR analyses were conducted to investigate if the effect of LDL-C on post-menopausal BCa was attenuated following adjustment for IGFBP-3. The multivariable OR for postmenopausal BCa per SD increase in LDL-C, conditioned on IGFBP-3 was 1.11 (95%CI 1.02, 1.20; p = 0.02), which did not differ by size and direction of effect when compared to the IGFBP-3 unadjusted model (OR 1.13; 95%CI 1.04, 1.25; p = 0.004) (Table 6).

Estimating the causal relationship between LDL-C raising genetic variants in HMGCR with BCa and IGF levels

LDL-C raising variants in HMGCR (which simulate activation of HMGCR) were associated with increased odds of BCa in BCAC (OR 1.20, 95%CI 1.02, 1.40; p = 0.03). In UKBB, the estimated causal effect of LDL-C raising variants in HMGCR was consistent but less precise (OR 1.34, 95%CI 0.90, 2.02; p = 0.15). The meta-analysed OR for overall BCa in BCAC and UKBB was 1.22 (95%CI 1.04, 1.43; p = 0.01) (Table 7). Using data from BCAC, LDL-C raising variants in HMGCR were associated with increased odds of ER-positive BCa (OR 1.21, 95%CI 1.00, 1.47; p = 0.05) but not ER negative BCa (OR 1.20, 95%CI 0.90, 1.60; p = 0.23).

View this table:
Table 7. Estimates of the causal relationship between LDL-C raising variants in HMGCR with breast cancer or IGF levels.

Using data from UKBB, LDL-C raising variants in HMGCR were associated with increased odds of postmenopausal BCa (OR 1.63, 95%CI 1.06, 2.49; p = 0.03), but not premenopausal BCa (OR 1.47, 95%CI 0.57, 3.75; p = 0.42) (Table 7). There was weak evidence that statins could reduce IGF-I (beta: 0.02, 95%CI –0.02, 0.43, p = 0.07). There was little evidence that genetic variants in HMGCR were associated with IGFBP-3 (Table 7).

Discussion

In this study, we aimed to estimate the effect of circulating lipids and IGF traits on BCa and to investigate the role of circulating IGFs in the association between circulating lipids and BCa. We used complementary approaches to investigate circulating IGFs as causal intermediates between lipids and BCa. We found that TG was inversely associated with IGF-I and overall BCa and that IGF-I is positively associated with overall BCa. We also found that LDL-C was positively associated with IGFBP-3 and postmenopausal BCa and that IGFBP-3 is positively associated with postmenopausal BCa. Two-step MR results supported a hypothetical pathway between TG, IGF-I and overall BCa and also between LDL-C, IGFBP-3 and postmenopausal BCa.

A recent two-sample MR study assessed the relationship between circulating lipids and BCa in BCAC11. In an attempt to account for pleiotropic effects, the authors conducted sensitivity analyses by excluding pleiotropic variants. However, manual pruning of SNPs that are considered pleiotropic may result in an instrument that does not account for the underlying genetic architecture of the trait, resulting in findings that can be challenging to interpret52. Given that genetic variants included in the model are associated with multiple lipid fractions, the MR assumption that all genetic variants are valid instruments (i.e. not pleiotropic) is unlikely to hold. In this investigation, we have employed a multivariable MR approach, which allows genetic variants to have pleiotropic effects on other variables included in the model and jointly estimates the independent causal effect of each risk factor on the outcome. Our multivariable MR analysis found that TG is negatively associated with overall BCa, which was not observed in previous MR studies11,14. Care is needed in interpretation however as whilst these may reflect real differences, they could also be because of differences in analytical complexities. For example, MVMR model estimates the direct effect of each lipid conditional on the other lipid traits which differ from the total effect of the lipids as estimated by univariable MR analyses in the absence of pleiotropy.

Epidemiological evidence supports a relationship between lipids and IGFs15–17. In the context of this relationship, we set out to assess the relationship between circulating lipids and IGFs using a combination of observational and MR methods. Results from observational analyses and MR analyses highlight that TG decreases IGF-I levels and that TG and LDL-C increases IGFBP-3 levels. There was minimal evidence to suggest that circulating IGFs causally affect lipid levels, although this finding may be limited by the availability of robust genetic instruments for individual IGF traits53.

A meta-analysis of 17 studies and a case-control study showed that circulating IGF-I increases odds of ER-positive BCa in pre- and postmenopausal women but not odds of ER-negative BCa22,54. This finding is concordant with results of our study, suggesting that circulating IGF-I increases overall BCa, likely driven by ER-positive BCa. Our MR results suggest that circulating IGF-I is more influential in ER-positive BCa. However, preclinical evidence suggests that IGF signalling is mitogenic for both ER-positive and ER-negative BCa55. As we have used genetic variants that predict circulating levels of IGF-I, we cannot rule out important aspects of tissue-specific regulation which may contribute to BCa.

Menopause causes substantial disturbances to lipid metabolism56. Observational evidence suggests that the association between lipids or IGFs and BCa depends on menopausal status5,22. Using data from UKBB, our MR results suggest that circulating LDL-C is more influential in postmenopausal BCa compared to premenopausal BCa. In contrast to previous studies57–60, the association between HDL-C and BCa did not differ by menopausal status in our analysis. IGFBP-3 appeared to be more influential in post-menopausal BCa, consistent with findings from a meta-analysis of observational studies22. However, confidence intervals were overlapping for both LDL-C and IGFBP-3 estimates in premenopausal and postmenopausal BCa analyses and the difference could be due to low statistical power in subgroup analyses, in particular the small number of cases in the premenopausal analysis (n = 1,781).

Our two-step MR results suggested a pathway between TG, IGF-I and overall BCa and a pathway between LDL-C, IGFBP-3 and postmenopausal BCa. However, multivariable MR analyses, wherein either IGF-I or IGFBP-3 were included as a covariate in the model, did not attenuate the associations between TG and BCa or the associations between LDL-C and postmenopausal BCa. This is counter to our interpretation of results, however our main concern in this sensitivity analysis relates to instrument strength and the power of this analysis. Instrument strength in a MVMR model can be assessed using a conditional F statistic that summarises the relationship between genetic instruments and each exposure conditional on the other exposures included in the MVMR model48. We assessed the instrument strength in the multivariable MR models and found that the conditional F statistics for IGF-I and IGFBP-3 were low (Table 6). In addition, assessment of the Q-statistic for heterogeneity found evidence of potential pleiotropy in the MVMR model (Table 6), however, this could be due to presence of conditionally weak instruments for the IGF measures.

Given that our study found evidence that LDL-C increases both BCa and IGFBP-3 levels, it was appropriate to assess how statins (the most commonly prescribed LDL-lowering therapy) impact BCa and IGFBP-3 levels using LDL-C raising genetic variants in HMGCR. Increased LDL-C as proxied by genetic variants in HMGCR was associated with increased overall BCa, consistent with results from a previous MR study13. In our analysis the effect estimate for overall BCa using the LDL-C raising genetic variants in HMGCR was larger than the effect estimate using the 185 SNP genome-wide LDL-C score. It seems likely that the difference in the results could reflect properties of the instruments employed. The larger number of genetic variants used to proxy LDL-C levels could capture pathways beyond HMGCR, therefore pleiotropy could influence the effect estimate. As the 5 SNPs in the HMGCR used in the statin analysis are weakly correlated, the multiplicity of associations due to LD between SNPs could bias the MR results61. However, we attempted to account for the LD between the SNPs in the HMGCR instrument in our MR analyses by accounting for these49,50. Differences in the results using the HMGCR score and 185 SNP genome-wide LDL-C score could also be due to off-target effects of statins (i.e. not via LDL-C). For example, a recent study by Kar et al13 suggested that BMI, rather than LDL-C, could be the true underlying mechanism responsible for the protective association between genetic inhibition of HMGCR and BCa.

Randomised control trials have shown that statin use results in altered levels of IGF-I and IGFBP-315–17. Our study found weak evidence of an association between genetic variants in HMGCR and IGF-I levels and minimal evidence of an association with IGFBP-3 levels. Failure to detect an association between statins an IGFBP3 may be due to the relatively small size of the IGF GWAS.

Conclusion

This study employed complementary observational and MR analyses to confirm that LDL-C and HDL-C are positively associated with overall BCa and that TG is negatively associated with overall BCa. Our study found that TG is negatively associated with IGF-I and that IGF-I is positively associated with overall BCa, consistent with the TG-BCa association. We also found that LDL-C is positively associated with IGFBP-3 and that IGFBP-3 is positively associated with postmenopausal BCa, consistent with the LDL-C-postmenopausal BCa association. Our two-step MR results build on evidence linking circulating lipids and IGFs with BCa, however, multivariable MR analyses are currently unable to support this relationship due to presence of conditionally weak instruments for the IGF measures.

Data Availability

Data available on request from the authors

Funding

This work was specifically supported by the Medical Research Council (MRC) Integrative Epidemiology Unit (IEU) (MC_UU_12013/3) and by a Cancer Research UK Programme Grant [The Integrative Cancer Epidemiology Programme, ICEP] (C18281/A19169). VT, KMB, CMP, JMPH, NJT are supported by ICEP (C18281/A19169). NJT is a Wellcome Trust Investigator (202802/Z/16/Z) and works within the University of Bristol NIHR Biomedical Research Centre (BRC). LJC is supported by NJT’s Wellcome Trust Investigator grant (202802/Z/16/Z). NJT and LJC work in the MRC IEU at the University of Bristol which is supported by the MRC (MC_UU_00011) and the University of Bristol. TD received support from Wellcome (grant ref 201268/Z/16/Z) and is now funded by the NIHR as an Academic Clinical Fellow. QQ is supported by a Scientist Development Award (K01HL129892) from the NHLBI. JR is supported in part by the National Center for Advancing Translational Sciences, CTSI grant UL1TR001881, and the National Institute of Diabetes and Digestive and Kidney Disease Diabetes Research Center (DRC) grant DK063491 to the Southern California Diabetes Endocrinology Research Center. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health. This publication is the work of the authors who will serve as guarantors for the contents of this paper.

Tables and Figures

Supplementary Table 1. Genetic instruments used to construct genetic risk scores for lipid traits used in multivariable MR analyses

Supplementary Table 2. Genetic instruments used to construct genetic risk scores for IGF-I and IGFBP-3

Supplementary Table 3. Study characteristics of the SHIP cohort

Supplementary Table 4. Association of IGF axis and lipids with potential confounders in the SHIP study

Acknowledgements

Quality Control filtering of the UK Biobank data was conducted by R.Mitchell, G.Hemani, T.Dudding, L.Paternoster as described in the published protocol (doi:10.5523/bris.3074krb6t2frj29yh2b03×3wxj). SHIP is part of the Community Medicine Research net of the University of Greifswald, Germany, which is funded by the Federal Ministry of Education and Research (grants no. 01ZZ9603, 01ZZ0103, and 01ZZ0403), the Ministry of Cultural Affairs as well as the Social Ministry of the Federal State of Mecklenburg-West Pomerania, and the network ‘Greifswald Approach to Individualized Medicine (GANI_MED)’ funded by the Federal Ministry of Education and Research (grant 03IS2061A).

References

  1. ↵
    Torre, L. A. et al. Global cancer statistics, 2012. CA Cancer J Clin 65, 87–108, doi:10.3322/caac.21262 (2015).
    OpenUrlCrossRefPubMed
  2. ↵
    Siegel, R., Naishadham, D. & Jemal, A. Cancer statistics, 2012. CA Cancer J Clin 62, 10–29, doi:10.3322/caac.20138 (2012).
    OpenUrlCrossRefPubMedWeb of Science
  3. ↵
    Garcia-Estevez, L. & Moreno-Bueno, G. Updating the role of obesity and cholesterol in breast cancer. Breast Cancer Res 21, 35, doi:10.1186/s13058-019-1124-1 (2019).
    OpenUrlCrossRef
  4. ↵
    Blucher, C. & Stadler, S. C. Obesity and Breast Cancer: Current Insights on the Role of Fatty Acids and Lipid Metabolism in Promoting Breast Cancer Growth and Progression. Front Endocrinol (Lausanne) 8, 293, doi:10.3389/fendo.2017.00293 (2017).
    OpenUrlCrossRef
  5. ↵
    Touvier, M. et al. Cholesterol and breast cancer risk: a systematic review and meta-analysis of prospective studies. Br J Nutr 114, 347–357, doi:10.1017/S000711451500183X (2015).
    OpenUrlCrossRef
  6. ↵
    Ni, H., Liu, H. & Gao, R. Serum Lipids and Breast Cancer Risk: A Meta-Analysis of Prospective Cohort Studies. PloS one 10, e0142669, doi:10.1371/journal.pone.0142669 (2015).
    OpenUrlCrossRef
  7. ↵
    Chan, K. K., Oza, A. M. & Siu, L. L. The statins as anticancer agents. Clin Cancer Res 9, 10–19 (2003).
    OpenUrlAbstract/FREE Full Text
  8. ↵
    Islam, M. M. et al. Exploring association between statin use and breast cancer risk: an updated meta-analysis. Arch Gynecol Obstet 296, 1043–1053, doi:10.1007/s00404-017-4533-3 (2017).
    OpenUrlCrossRef
  9. ↵
    Suissa, S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf 16, 241–249, doi:10.1002/pds.1357 (2007).
    OpenUrlCrossRefPubMedWeb of Science
  10. ↵
    Davey Smith, G. & Hemani, G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Human molecular genetics 23, R89–98, doi:10.1093/hmg/ddu328 (2014).
    OpenUrlCrossRefPubMedWeb of Science
  11. ↵
    Nowak, C. & Arnlov, J. A Mendelian randomization study of the effects of blood lipids on breast cancer risk. Nature communications 9, 3957, doi:10.1038/s41467-018-06467-9 (2018).
    OpenUrlCrossRef
  12. ↵
    Walker, V. M., Davey Smith, G., Davies, N. M. & Martin, R. M. Mendelian randomization: a novel approach for the prediction of adverse drug events and drug repurposing opportunities. International journal of epidemiology 46, 2078–2089, doi:10.1093/ije/dyx207 (2017).
    OpenUrlCrossRefPubMed
  13. ↵
    Kar, S. P. et al. Body mass index and the association between low-density lipoprotein cholesterol as predicted by HMGCR genetic variants and breast cancer risk. International journal of epidemiology, doi:10.1093/ije/dyz047 (2019).
    OpenUrlCrossRef
  14. ↵
    Orho-Melander, M. et al. Blood lipid genetic scores, the HMGCR gene and cancer risk: a Mendelian randomization study. International journal of epidemiology 47, 495–505, doi:10.1093/ije/dyx237 (2018).
    OpenUrlCrossRef
  15. ↵
    Bergen, K., Brismar, K. & Tehrani, S. High-dose atorvastatin is associated with lower IGF-1 levels in patients with type 1 diabetes. Growth hormone & IGF research : official journal of the Growth Hormone Research Society and the International IGF Research Society 29, 78–82, doi:10.1016/j.ghir.2016.06.001 (2016).
    OpenUrlCrossRef
  16. Narayanan, R. P. et al. Atorvastatin administration is associated with dose-related changes in IGF bioavailability. European journal of endocrinology 168, 543–548, doi:10.1530/EJE-12-0844 (2013).
    OpenUrlAbstract/FREE Full Text
  17. ↵
    Szkodzinski, J. et al. Effect of HMG-CoA (3-hydroxy-3-methylglutaryl-CoA) reductase inhibitors on the concentration of insulin-like growth factor-1 (IGF-1) in hypercholesterolemic patients. Pharmacological reports : PR 61, 654–664 (2009).
    OpenUrlPubMedWeb of Science
  18. ↵
    Perks, C. M. & Holly, J. M. IGF binding proteins (IGFBPs) and regulation of breast cancer biology. J Mammary Gland Biol Neoplasia 13, 455–469, doi:10.1007/s10911-008-9106-4 (2008).
    OpenUrlCrossRefPubMedWeb of Science
  19. Perks, C. M. & Holly, J. M. Insulin-like growth factor binding proteins (IGFBPs) in breast cancer. J Mammary Gland Biol Neoplasia 5, 75–84 (2000).
    OpenUrlPubMed
  20. Biernacka, K. M., Perks, C. M. & Holly, J. M. Role of the IGF axis in prostate cancer. Minerva Endocrinol 37, 173–185 (2012).
    OpenUrlPubMed
  21. ↵
    Pollak, M. The insulin and insulin-like growth factor receptor family in neoplasia: an update. Nat Rev Cancer 12, 159–169, doi:10.1038/nrc3215 (2012).
    OpenUrlCrossRefPubMedWeb of Science
  22. ↵
    Endogenous, H. et al. Insulin-like growth factor 1 (IGF1), IGF binding protein 3 (IGFBP3), and breast cancer risk: pooled individual data analysis of 17 prospective studies. Lancet Oncol 11, 530–542, doi:10.1016/S1470-2045(10)70095-4 (2010).
    OpenUrlCrossRefPubMedWeb of Science
  23. ↵
    Clemmons, D. R. Metabolic actions of insulin-like growth factor-I in normal physiology and diabetes. Endocrinology and metabolism clinics of North America 41, 425–443, vii-viii, doi:10.1016/j.ecl.2012.04.017 (2012).
    OpenUrlCrossRefPubMed
  24. Kim, H. S. Role of insulin-like growth factor binding protein-3 in glucose and lipid metabolism. Ann Pediatr Endocrinol Metab 18, 9–12, doi:10.6065/apem.2013.18.1.9 (2013).
    OpenUrlCrossRefPubMed
  25. ↵
    Russo, V. C., Azar, W. J., Yau, S. W., Sabin, M. A. & Werther, G. A. IGFBP-2: The dark horse in metabolism and cancer. Cytokine Growth Factor Rev 26, 329–346, doi:10.1016/j.cytogfr.2014.12.001 (2015).
    OpenUrlCrossRefPubMed
  26. ↵
    Mauras, N., Martinez, V., Rini, A. & Guevara-Aguirre, J. Recombinant human insulin-like growth factor I has significant anabolic effects in adults with growth hormone receptor deficiency: studies on protein, glucose, and lipid metabolism. J Clin Endocrinol Metab 85, 3036–3042, doi:10.1210/jcem.85.9.6772 (2000).
    OpenUrlCrossRefPubMedWeb of Science
  27. Mauras, N. et al. Insulin-like growth factor I and growth hormone (GH) treatment in GH-deficient humans: differential effects on protein, glucose, lipid, and calcium metabolism. J Clin Endocrinol Metab 85, 1686–1694, doi:10.1210/jcem.85.4.6541 (2000).
    OpenUrlCrossRefPubMedWeb of Science
  28. ↵
    Mauras, N. & Haymond, M. W. Metabolic effects of recombinant human insulin-like growth factor-I in humans: comparison with recombinant human growth hormone. Pediatr Nephrol 10, 318–323 (1996).
    OpenUrlCrossRefPubMedWeb of Science
  29. ↵
    Friedrich, N. et al. Cross-sectional and longitudinal associations between insulin-like growth factor I and metabolic syndrome: a general population study in German adults. Diabetes/metabolism research and reviews 29, 452–462, doi:10.1002/dmrr.2412 (2013).
    OpenUrlCrossRefPubMed
  30. Lam, C. S. et al. Circulating insulin-like growth factor-1 and its binding protein-3: metabolic and genetic correlates in the community. Arteriosclerosis, thrombosis, and vascular biology 30, 1479–1484, doi:10.1161/ATVBAHA.110.203943 (2010).
    OpenUrlAbstract/FREE Full Text
  31. ↵
    Eggert, M. L. et al. Cross-sectional and longitudinal relation of IGF1 and IGF-binding protein 3 with lipid metabolism. European journal of endocrinology 171, 9–19, doi:10.1530/EJE-13-1017 (2014).
    OpenUrlAbstract/FREE Full Text
  32. ↵
    Melvin, J. C., Holmberg, L., Rohrmann, S., Loda, M. & Van Hemelrijck, M. Serum lipid profiles and cancer risk in the context of obesity: four meta-analyses. J Cancer Epidemiol 2013, 823849, doi:10.1155/2013/823849 (2013).
    OpenUrlCrossRefPubMed
  33. Chandler, P. D. et al. Lipid biomarkers and long-term risk of cancer in the Women's Health Study. Am J Clin Nutr 103, 1397–1407, doi:10.3945/ajcn.115.124321 (2016).
    OpenUrlAbstract/FREE Full Text
  34. ↵
    Pollak, M. Insulin and insulin-like growth factor signalling in neoplasia. Nat Rev Cancer 8, 915–928, doi:10.1038/nrc2536 (2008).
    OpenUrlCrossRefPubMedWeb of Science
  35. ↵
    Relton, C. L. & Davey Smith, G. Two-step epigenetic Mendelian randomization: a strategy for establishing the causal role of epigenetic processes in pathways to disease. International journal of epidemiology 41, 161–176, doi:10.1093/ije/dyr233 (2012).
    OpenUrlCrossRefPubMedWeb of Science
  36. ↵
    Global Lipids Genetics, C. et al. Discovery and refinement of loci associated with lipid levels. Nature genetics 45, 1274–1283, doi:10.1038/ng.2797 (2013).
    OpenUrlCrossRefPubMed
  37. ↵
    Michailidou, K. et al. Association analysis identifies 65 new breast cancer risk loci. Nature 551, 92–94, doi:10.1038/nature24284 (2017).
    OpenUrlCrossRefPubMed
  38. ↵
    Teumer, A. et al. Genomewide meta-analysis identifies loci associated with IGF-I and IGFBP-3 levels with impact on age-related traits. Aging cell 15, 811–824, doi:10.1111/acel.12490 (2016).
    OpenUrlCrossRef
  39. ↵
    Yarmolinsky, J. et al. Association Between Genetically Proxied Inhibition of HMG-CoA Reductase and Epithelial Ovarian Cancer. JAMA 323, 646–655, doi:10.1001/jama.2020.0150 (2020).
    OpenUrlCrossRef
  40. ↵
    Burgess, S., Dudbridge, F. & Thompson, S. G. Re: “Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects”. American journal of epidemiology 181, 290–291, doi:10.1093/aje/kwv017 (2015).
    OpenUrlCrossRefPubMed
  41. ↵
    Burgess, S. & Thompson, S. G. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. American journal of epidemiology 181, 251–260, doi:10.1093/aje/kwu283 (2015).
    OpenUrlCrossRefPubMed
  42. ↵
    Lawlor, D. A. Commentary: Two-sample Mendelian randomization: opportunities and challenges. International journal of epidemiology 45, 908–915, doi:10.1093/ije/dyw127 (2016).
    OpenUrlCrossRefPubMed
  43. ↵
    Burgess, S., Butterworth, A. & Thompson, S. G. Mendelian randomization analysis with multiple genetic variants using summarized data. Genetic epidemiology 37, 658–665, doi:10.1002/gepi.21758 (2013).
    OpenUrlCrossRefPubMed
  44. ↵
    Bowden, J., Davey Smith, G. & Burgess, S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International journal of epidemiology 44, 512–525, doi:10.1093/ije/dyv080 (2015).
    OpenUrlCrossRefPubMed
  45. ↵
    Bowden, J., Davey Smith, G., Haycock, P. C. & Burgess, S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genetic epidemiology 40, 304–314, doi:10.1002/gepi.21965 (2016).
    OpenUrlCrossRefPubMed
  46. ↵
    Burgess, S. et al. Using published data in Mendelian randomization: a blueprint for efficient identification of causal risk factors. Eur J Epidemiol 30, 543–552, doi:10.1007/s10654-015-0011-z (2015).
    OpenUrlCrossRefPubMed
  47. ↵
    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. International journal of epidemiology 48, 713–727, doi:10.1093/ije/dyy262 (2019).
    OpenUrlCrossRefPubMed
  48. ↵
    Sanderson, E., Spiller, W., Bowden, J.. Testing and Correcting for Weak and Pleiotropic Instruments in Two-Sample Multivariable Mendelian Randomisation. bioRxiv, doi:https://doi.org/10.1101/2020.04.02.021980 (2020).
  49. ↵
    Yavorska, O. O. & Burgess, S. MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data. International journal of epidemiology 46, 1734–1739, doi:10.1093/ije/dyx034 (2017).
    OpenUrlCrossRefPubMed
  50. ↵
    Burgess, S., Dudbridge, F. & Thompson, S. G. Combining information on multiple instrumental variables in Mendelian randomization: comparison of allele score and summarized data methods. Stat Med 35, 1880–1906, doi:10.1002/sim.6835 (2016).
    OpenUrlCrossRefPubMed
  51. ↵
    Hemani, G. et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife 7, doi:10.7554/eLife.34408 (2018).
    OpenUrlCrossRefPubMed
  52. ↵
    Holmes, M. V., Ala-Korpela, M. & Smith, G. D. Mendelian randomization in cardiometabolic disease: challenges in evaluating causality. Nat Rev Cardiol 14, 577–590, doi:10.1038/nrcardio.2017.78 (2017).
    OpenUrlCrossRefPubMed
  53. ↵
    Bonilla, C. et al. Assessing the role of insulin-like growth factors and binding proteins in prostate cancer using Mendelian randomization: Genetic variants as instruments for circulating levels. International journal of cancer 139, 1520–1533, doi:10.1002/ijc.30206 (2016).
    OpenUrlCrossRef
  54. ↵
    Kaaks, R. et al. Insulin-like growth factor I and risk of breast cancer by age and hormone receptor status-A prospective study within the EPIC cohort. International journal of cancer 134, 2683–2690, doi:10.1002/ijc.28589 (2014).
    OpenUrlCrossRefPubMedWeb of Science
  55. ↵
    Surmacz, E. & Bartucci, M. Role of estrogen receptor alpha in modulating IGF-I receptor signaling and function in breast cancer. J Exp Clin Cancer Res 23, 385–394 (2004).
    OpenUrlPubMedWeb of Science
  56. ↵
    Auro, K. et al. A metabolic view on menopause and ageing. Nature communications 5, 4708, doi:10.1038/ncomms5708 (2014).
    OpenUrlCrossRefPubMed
  57. ↵
    Agnoli, C. et al. Metabolic syndrome and postmenopausal breast cancer in the ORDET cohort: a nested case-control study. Nutr Metab Cardiovasc Dis 20, 41–48, doi:10.1016/j.numecd.2009.02.006 (2010).
    OpenUrlCrossRefPubMedWeb of Science
  58. Furberg, A. S., Veierod, M. B., Wilsgaard, T., Bernstein, L. & Thune, I. Serum high-density lipoprotein cholesterol, metabolic profile, and breast cancer risk. J Natl Cancer Inst 96, 1152–1160, doi:10.1093/jnci/djh216 (2004).
    OpenUrlCrossRefPubMedWeb of Science
  59. Kucharska-Newton, A. M. et al. HDL-cholesterol and incidence of breast cancer in the ARIC cohort study. Ann Epidemiol 18, 671–677, doi:10.1016/j.annepidem.2008.06.006 (2008).
    OpenUrlCrossRefPubMedWeb of Science
  60. ↵
    Moorman, P. G. et al. Association between high-density lipoprotein cholesterol and breast cancer varies by menopausal status. Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology 7, 483–488 (1998).
    OpenUrl
  61. ↵
    Swerdlow, D. I. et al. Selecting instruments for Mendelian randomization in the wake of genome-wide association studies. International journal of epidemiology 45, 1600–1616, doi:10.1093/ije/dyw088 (2016).
    OpenUrlCrossRefPubMed
Back to top
Posted June 07, 2020.
Download PDF

Supplementary Material

Data/Code
Email
Estimating the effect of lipids on IGF axis and subsequent breast cancer risk
Vanessa Y Tan, Caroline J Bull, Kalina M Biernacka, Alexander Teumer, Laura Corbin, Tom Dudding, Eleanor Sanderson, Qibin Qi, Robert C Kaplan, Jerome I Rotter, Nele Friedrich, Uwe Völker, Julia Mayerle, Claire M Perks, Jeff MP Holly, Nicholas J Timpson
medRxiv 2020.06.04.20122630; doi: https://doi.org/10.1101/2020.06.04.20122630
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area