Abstract

Background and purpose

   Breast cancer (BC) is the primary cause of cancer-related deaths among
   women worldwide, with increasing evidence pointing to the effect of
   metabolic factors, particularly lipid levels, in its pathogenesis. In
   this research, Mendelian randomization (MR) was employed to explore the
   causality between four plasma lipid traits—total cholesterol (TC),
   triglycerides (TG), high-density lipoprotein cholesterol (HDL-C) and
   low-density lipoprotein cholesterol (LDL-C)—and the risk of BC.
   Additionally, we explored the potential mediating effects of coronary
   artery disease (CAD), total testosterone (TT) on these associations and
   possible mechanisms through bioinformatics analyses.

Methods

   Data of genome-wide association study (GWAS) on lipids, CAD, TT and BC
   were obtained from public sources and websites as part of a genome-wide
   association research. The inference of causality was primarily assessed
   through the inverse variance weighting (IVW) approach, with
   supplementary tests for horizontal pleiotropy and heterogeneity. To
   verify the directionality of causal relationships, the MR Steiger test
   was applied. Additionally, reverse causality was evaluated by regarding
   BC as the exposure. To adjust for confounders, multivariate MR (MVMR)
   was performed, followed by a two-step mediation analysis to investigate
   the mediating roles of CAD in the lipid-BC association, and of TT in
   the CAD-BC relationship. The intersecting SNP (rs11556924) between
   causal pathways was established through a Venn diagram and its
   associated gene (Zinc Finger C3HC-Type Containing 1, ZC3HC1) was
   identified through the g:Profiler database. The expression of ZC3HC1
   was further explored using the TIMER, GEPIA2 and HPA database. Finally,
   enrichment analyses of Gene Ontology (GO), Kyoto Encyclopedia of Genes
   and Genomes (KEGG) and protein–protein interactions (PPI) network
   analysis were conducted on ZC3HC1 and its related genes.

Results

   The random-effects IVW analysis yielded the following results: HDL-C on
   CAD (OR = 0.843, 95% CI 0.771–0.921, P < 0.001), CAD on BC (OR = 0.935,
   95% CI 0.892–0.980, P = 0.005), HDL-C on BC (OR = 1.127, 95% CI
   1.059–1.199, P < 0.001), CAD on TT (OR = 0.987, 95% CI 0.975–0.998,
   P = 0.020) and TT on BC (OR = 1.354, 95% CI 1.148–1.598, P < 0.001).
   The MR Steiger test results support the validity of the inferred causal
   direction (P < 0.001). There were no discernible causal relationships
   between BC and HDL-C/CAD according to reverse MR analysis (P > 0.05).
   Following MVMR adjustment, the causal effects of HDL-C, CAD, and TT on
   BC were still statistically significant (P < 0.05). Besides, the
   two-step mediation analysis indicated that CAD mediated 7.8% of the
   causal effect of HDL-C on BC, whereas TT mediated 6.1% of the causal
   effect between CAD and BC. The expression of ZC3HC1 showed no
   significant expression difference between normal and BC tissues
   (P > 0.05), which might indicate a carcinogenic effect independent of
   expression levels but driven by functional alterations induced by
   variants (C > T). Functional network analysis suggested that ZC3HC1 was
   associated with multiple signal pathways in cancers, such as PI3K-Akt
   and MAPK signal pathways.

Conclusions

   From a genetic perspective, our study reveals that there is causality
   between HDL-C levels and BC risk, with CAD and TT acting as partial
   mediators in this relationship. Moreover, our study firstly establishes
   a potential link between CAD-associated SNP (rs11556924), the
   corresponding gene (ZC3HC1) functional dysregulation, and the
   initiation of BC. These findings shed light on the biological links
   between lipids and BC, potentially contributing to future prevention
   and treatment strategies.

Supplementary Information

   The online version contains supplementary material available at
   10.1007/s12672-025-02597-x.

   Keywords: Breast cancer, Lipids, High-density lipoprotein cholesterol,
   Total testosterone, Coronary artery disease, Mendelian randomization,
   ZC3HC1

Introduction

   Breast cancer (BC) is the most prevalent malignant tumor among women
   globally and remains a major contributor to cancer-associated deaths in
   female populations [[28]1, [29]2]. Projections from the GLOBOCAN Cancer
   Tomorrow tool estimate that the incidence of BC will rise by over 54%
   by 2050. At the same time, despite advances in early diagnosis and
   therapy strategies, the prognosis of BC varies substantially,
   necessitating further study into novel biomarkers and possible
   therapeutic targets [[30]3]. The etiology of BC is multifactorial,
   encompassing genetic predispositions, environmental exposures, and
   lifestyle influences [[31]4, [32]5]. There is mounting evidence that
   some metabolic disorders (obesity, insulin resistance and dyslipidemia)
   are potential risk factors for BC because these conditions can further
   influence hormone levels and inflammatory pathways, contributing to
   BC's development and progression [[33]6, [34]7]. Numerous studies have
   indicated a positive association between dyslipidemia and BC risk
   [[35]8, [36]9]. However, most of the current studies on the association
   between dyslipidemia and BC incidence are observational studies, making
   it impossible to draw causal inference conclusions.

   Cholesterol is essential for a variety of physiological processes,
   including the formation/function of cell membranes and cellular
   homeostasis regulation [[37]10, [38]11]. Importantly, total cholesterol
   (TC) levels are not only correlated with the incidence of coronary
   artery disease (CAD) [[39]11] but also with adiposity and diabetes, all
   of which have been suggested as potential risk factors for BC [[40]12,
   [41]13]. However, the causal relationship between cholesterol BC
   susceptibility remains unclear, as studies have reported conflicting
   findings, including positive, inverse or absent associations between
   lipid levels and BC risk [[42]9, [43]14, [44]15]. In addition,
   triglycerides (TG) have complicated functions in cardiovascular health,
   and their significance in CAD has been extensively investigated and
   discussed [[45]16]. Concurrently, TG can also serve as an independent
   reservoir of fatty acid oxidation, which can further drive cell
   proliferation and cancer development [[46]17], suggesting the possible
   carcinogenicity of TG. However, the effect of TG on BC remains
   unidentified, with some studies suggesting that high TG levels increase
   BC risk [[47]18], whereas others have reported an inverse association
   [[48]15]. Low-density lipoprotein cholesterol (LDL-C) is broadly
   acknowledged as a significant risk factor for CAD. Some studies also
   indicated that lower LDL-C levels might increase the occurrence of
   cancer [[49]19]. However, in BC, the association appears to be the
   opposite [[50]20]. Moreover, high-density lipoprotein cholesterol
   (HDL-C) has gained prominence as a protective factor for CAD [[51]21].
   Previous studies have also demonstrated HDL-C as a powerful univariate
   predictor of CAD [[52]22]. Moreover, multiple investigations have
   suggested a negative correlation between HDL-C levels and tumor risk
   [[53]23, [54]24]. But for BC, observational studies have yielded
   conflicting results. Some studies concluded that HDL-C were positively
   linked to the risk of BC [[55]25, [56]26], whereas other studies found
   a negative association [[57]9, [58]27] or no association [[59]28].
   These contradictory findings suggest that HDL-C may exert a regulatory
   function in BC development, but its actual direction and mechanisms
   have not yet been clarified. In summary, there seems to be a connection
   between the risk of BC and these four categories of plasma lipids. But
   the research findings are inconclusive or contradictory, and the
   distinct causal relationships between these plasma lipid levels and BC
   have not been fully elucidated.

   For CAD, it is the leading cause of mortality in affluent nations,
   imposing a significant economic burden [[60]29]. Several studies have
   identified an association between CAD and cancer prevalence [[61]30,
   [62]31], including BC [[63]32], which prompts that these two diseases
   may have shared risk factors and biological mechanisms. Nonetheless,
   the current research indicates an absence of randomized controlled
   trials (RCTs) exploring the connection between CAD and BC and
   conducting such trials in the future presents significant ethical
   challenges. Thus, the probable causative link between CAD and BC has
   largely remained unexplored.

   Mendelian Randomization (MR) is a genetics-based approach that employs
   variations in the genome as instrumental variables (IVs) to reveal the
   causality between exposures and outcomes. It offers the advantages of
   reducing confounding bias, avoiding reverse causality and enabling the
   evaluation of exposures'long-term effects [[64]33]. Given the existing
   evidence indicating inconsistencies in the correlation between lipid
   levels, CAD and BC, as well as the uncertainty surrounding causality,
   we conducted two-sample MR (TSMR) analyses to clarify the relationship
   between genetically elevated levels of four lipid traits and BC, as
   well as the causal association between CAD and BC. Moreover, to better
   elucidate the mechanism through which CAD may influence the development
   of BC, we further selected total testosterone (TT) levels as a bridging
   variable to connect CAD and BC, owing to its distinct associations with
   both CAD [[65]34] and BC [[66]35, [67]36] based on current evidence. To
   adjust for confounding variables that could influence causal
   estimations [[68]37], we also employed Multivariable Mendelian
   Randomization (MVMR). In addition, the first mediation analysis was
   performed to evaluate the possible mediating function of CAD in the
   HDL-C-BC causative link [[69]38]. Subsequently, we conducted a further
   mediation analysis to clarify TT’s function in the causal pathway
   between CAD and BC to offer a deeper insight into the underlying
   biological mechanisms. Finally, we aimed to identify key SNPs and
   associated genes implicated in causal relationships through
   comprehensive bioinformatics analyses. We also integrated findings both
   from our study and existing studies to elucidate potential mechanisms
   at both the gene expression level and from a functional perspective.
   Our exploration, from the perspective of genetics, may provide new
   perspectives on the processes that drive the incidence and progression
   of BC, as well as identify novel targets for metabolic therapeutic
   approaches.

Materials and methods

Study design and data sources

   To evaluate the causative correlation between lipids and BC and
   determine if this relationship was mediated by CAD, different MR and
   mediation analyses were employed. Additionally, the effect of CAD on BC
   was further investigated by examining TT as a potential mediator.
   Briefly, we first explored the overall causal associations between four
   common lipid molecules (TC, TG, LDL-C, HDL-C) and BC through TSMR.
   Next, for positive associations, mediation effect analyses were
   conducted to quantify the mediating role of CAD and to assess its
   proportion in the overall causal relationship. Subsequently, we
   explored the function of TT (as a mediator) in the interaction between
   CAD and BC. The MVMR was employed to assess the mediating effect and
   proportion mediated by CAD and TT. The publicly accessible GWAS dataset
   served as the basis for this MR investigation as well, with all
   participants having submitted written informed consent. Concurrently,
   this study was conducted in keeping with the Strengthening the
   Reporting of Observational Studies in Epidemiology using MR (STROBE-MR)
   checklist (Additional file [70]1) [[71]39]. The study design schematic
   and used MR analyses are also illustrated in Fig. [72]1.

Fig. 1.

   [73]Fig. 1
   [74]Open in a new tab

   The flowchart of this MR study including the basic assumptions. The
   purpose of the UVMR, MVMR and bidirectional MR analyses was to
   investigate the potential genetic correlations between lipids and BC
   risk. Mediation analysis was also employed to validate the effects of
   the mediating factors

   Data for the two exposures, LDL-C (n = 70,814, European descent) and
   HDL-C (n = 77,409, European descent), were both extracted from the
   Family GWAS consortium data summarized in the IEU database, a
   comprehensive GWAS collecting resource
   ([75]https://gwas.mrcieu.ac.uk/). Data on TC, TG and TT was originated
   from GWAS conducted by Barton AR et al. (n = 437,878, European
   ancestry) [[76]40], Richardson TG et al. (n = 115,082, European
   ancestry) [[77]41] and Ruth KS et al. (n = 425,097, European ancestry)
   [[78]42]. Genetic data on CAD as a mediator were available from a
   comprehensive meta-analysis focusing on GWAS, which included 42,096
   cases and 361 controls of European ancestry [[79]43]. Moreover, genetic
   summary data on BC risk were acquired from the Breast Cancer
   Association Consortium (BCAC) [[80]https://bcac.ccge.medschl.cam.ac.uk/
   (accessed on 20 May 2024)] website, which was a multidisciplinary
   dynamic consortium formed in April 2005 and recruited over 100 groups
   to collect data from over 420,000 people. Sample overlap is minimal as
   data for exposures and outcomes come from completely different GWAS
   studies and databases. Additional introductions for diseases and lipids
   can be obtained from the IEU database, original studies [[81]40–[82]43]
   and the BCAC website. More detailed information for each GWAS summary
   data is shown in Table [83]1.

Table 1.

   A detailed summary of Genome-Wide Association Studies (GWAS) included
   in the Mendelian randomization analysis
   Exposures/Outcomes GWAS ID Consortium Ethnicity Sample sizes Number of
   SNPs Year
   TC ebi-a-GCST90025953 NA European 437,878 4,232,052 2021
   TG ebi-a-GCST90092992 NA European 115,082 11,590,399 2022
   LDL-C ieu-b-4846 Family GWAS consortium European 70,814 7,892,997 2022
   HDL-C ieu-b-4844 Family GWAS consortium European 77,409 7,892,377 2022
   TT ebi-a-GCST90012114 NA European 425,097 16,132,861 2020
   CAD ebi-a-GCST003116 NA European 141,217 8,597,751 2015
   BC ieu-a-1129 BCAC European 106,776 10,680,257 2017
   [84]Open in a new tab

   TC: Total cholesterol levels; TG: Triglycerides levels; LDL-C:
   Low-density lipoprotein cholesterol; HDL-C: High-density lipoprotein
   cholesterol; TT: Total testosterone levels; CAD: Coronary artery
   disease; BC: Breast cancer

Selection of genetic IVs

   We identified single nucleotide polymorphisms (SNPs) significantly
   related to exposure using a threshold of P < 5 × 10^–8. To minimize
   bias from linkage disequilibrium (LD), a standard (r^2 < 0.001, kb
   > 10,000) was set to filter IVs and to ensure that they were
   distributed independently [[85]44]. Palindromic sequences with moderate
   allele frequencies (0.42–0.58) were eliminated during the harmonization
   step. In addition, SNPs associated with the results (P value for BC
   risk < 5 × 10^–8) were removed from the MR analysis to reduce potential
   confounding effects. The F-statistic value was calculated as well using
   the formula: F = [(N − k − 1)/k] × [R^2/(1 − R^2)] (N—the sample size
   in the GWAS analysis, k—the number of IVs, R^2—the proportion of
   exposure variance explained by these variables). And R^2 was calculated
   by 2 × (1 − MAF) × MAF × β^2 (MAF—minor allele frequency, β—effect size
   of SNP on exposure). SNPs receiving an F value ≥ 10 were thought to be
   powerful IVs [[86]45].

Mendelian randomization and evaluations of statistics

   The “TwoSample MR” package and R (version 4.3.2) were primarily used to
   conduct the TSMR analysis. To elucidate the causation, we chose the
   random effects model within the inverse variance weighting (IVW)
   framework [[87]46]. Also, MR-Egger [[88]47], weighted median [[89]48],
   simple mode, and weighted mode [[90]49] were performed as additional
   supporting analytic approaches. The risks of BC were interpreted
   through the odds ratio (OR) and 95% confidence interval (CI). MVMR was
   also implemented using the “MendelianRandomization” and “TwoSample MR”
   packages. Additionally, the validity of the causative direction was
   evaluated by conducting the MR Steiger test [[91]50]. P < 0.05
   indicated that a significant causal link was identified. Besides, IVW
   method results are the main elements presented in this article and the
   results of other methods for the main causal pathways were included in
   the Additional file [92]2 (Tables S1-5).

Sensitivity analysis

   Three approaches for sensitivity analyses were chosen: heterogeneity
   test, horizontal pleiotropy test and leave-one-out method. In addition,
   the IVs'heterogeneity was evaluated through the Cochran Q test.
   Significant heterogeneity was indicated by a P value of Q less than
   0.05 and IVW with random effects would be performed to get more
   accurate estimations. Furthermore, horizontal pleiotropy—the genetic
   variants might influence multiple phenotypic traits independently—was
   evaluated by performing the MR-Egger intercept method. If P > 0.05 for
   the intercept, then the possibility of horizontal pleiotropy of IVs was
   considered negligible [[93]51]. Finally, the leave-one-out strategy
   evaluates the overall impact of the remaining SNPs by sequentially
   removing each SNP and observing the changes in results.

Mediation analyses

   The effect values for HDL-C on CAD (βA), CAD on BC (βB), HDL-C on BC
   (βC), CAD on TT (βD) and TT on BC (βE) were sequentially calculated
   using the TSMR analysis firstly to test whether the mechanism could be
   passed through. Then MVMR method was used to demonstrate the
   relationship between CAD/TT and BC by adjusting for HDL-C (βB') or CAD
   (βE'). The calculation of the mediating effect was performed as
   βA*βB'(CAD as mediator) and βD*βE'(TT as mediator). The ratio of the
   mediating effect to the overall effect was established through the
   formulas: R = βA*βB'/βC*100% and βD*βE'/βB*100%. After controlling for
   confounding factors, the impact of exposure was regarded as a direct
   effect (βC − βA*βB', βB − βD*βE').

MR Data visualization

   To illustrate the influences of sequentially excluded single SNP on the
   research results, the leave-one-out method was conducted. Scatter plots
   were graphical depictions of causal effect estimates, with the x-axis
   displaying the effect of SNPs associated with exposure and the y-axis
   indicating the effect of SNPs associated with outcome. Forest plots
   were utilized to evaluate the effect estimates of genetic variations.
   Furthermore, funnel plots were utilized to investigate the possibility
   of significant heterogeneity among individual genetic variants. R
   software (version 4.3.2) was used to generate all the visualization
   graphs.

Identification of the common SNP and related gene expression analysis

   The intersecting SNPs in different causal relationships were analyzed
   using the “VennDiagram” R package. In addition, the most relevant
   coding gene (ZC3HC1) for the common SNP was identified from the
   g:SNPense module in the g:Profiler database
   ([94]https://biit.cs.ut.ee/gprofiler/) [[95]52]. The g:Profiler is a
   public database for characterizing and manipulating gene lists and is
   g:SNPense a tool that allows mapping of human SNPs to gene names,
   chromosomal locations, and variant consequence terms in the sequence
   ontology [[96]52]. Subsequently, pan-cancer expression analyses of the
   relevant genes were performed via The Tumor IMmune Estimation Resource
   (TIMER) ([97]https://cistrome.shinyapps.io/timer/) database [[98]53].
   The ZC3HC1 mRNA and protein expression levels were analyzed through The
   Gene Expression Profiling Interactive Analysis 2 (GEPIA2)
   ([99]http://gepia2.cancer-pku.cn/#index) [[100]54] and Human Protein
   Atlas (HPA) ([101]https://www.proteinatlas.org/) [[102]55] databases.

Analysis of ZC3HC1 co-expressed genes in BRCA

   The co-expressed genes of ZC3HC1 in BRCA were explored utilizing the
   LinkedOmics ([103]http://www.linkedomics.org/) [[104]56] database that
   contains multi-omics data from various cancers. The results of all
   co-expressed genes were analyzed using Spearman's correlation test and
   illustrated as heat maps. The “clusterProfiler” R package was employed
   to analyze and visualize the Gene Ontology (GO) categories, including
   cellular components (CC), biological processes (BP), and molecular
   functions (MF), as well as the Kyoto Encyclopedia of Genes and Genomes
   (KEGG) pathways associated with the co-expressed genes.

Protein–protein interactions network analysis of ZC3HC1

   STRING ([105]https://cn.string-db.org/) is a comprehensive database
   that compiles both known and predicted interactions between proteins
   [[106]57]. The Protein–Protein Interaction (PPI) network of ZC3HC1
   co-expressed proteins in BRCA was constructed and analyzed using the
   STRING database. The “clusterProfiler” R package was also applied to
   conduct GO term and KEGG pathway analyses.

Results

Selection of IVs

   SNPs were evaluated for independence after LD was eliminated (r^2 =
   0.001, kb = 10,000) and using the threshold (P < 5 × 10^–8). After
   removing palindromic sequences with intermediate allele frequencies,
   the following numbers of IVs were identified to investigate the causal
   relationships between lipids, TT and BC: TC-CAD/BC: 185, 181;
   TG-CAD/BC: 62, 62; LDL-C-CAD/BC: 42, 42; HDL-C-CAD/BC: 67, 63;
   CAD-BC/TT: 37, 40; TT-BC: 164; BC-HDL-C: 51. In addition, following the
   computation, the strength of all IVs was deemed robust (F statistics
   > 10). Detailed IV data for relationships of HDL-C/CAD/BC and CAD/TT/BC
   are included in the Additional file [107]2 (Tables S6-10).

Two-sample MR analysis between lipids, CAD, TT and BC

   The MR results obtained using the IVW approach are shown in
   Fig. [108]2. The results mainly revealed the negative causal effect of
   HDL-C on CAD (OR = 0.843, 95% CI [0.771, 0.921], P < 0.001); CAD on BC
   (OR = 0.935, 95% CI [0.892, 0.980], P = 0.005); CAD on TT (OR = 0.987,
   95% CI [0.975, 0.998], P = 0.02). Also, significant positive causal
   effect of HDL-C on BC (OR = 1.127, 95% CI [1.059, 1.199], P < 0.001);
   TT on BC (OR = 1.354, 95% CI [1.148, 1.598], P < 0.001) and modest
   inverse causality between TG and BC (OR = 0.922, 95% CI [0.866, 0.983],
   P = 0.012) were observed. Moreover, scatter plots demonstrate that
   genetically driven elevation of HDL-C is related to a higher risk of
   developing BC and a lower risk of developing CAD (Fig. [109]3A,
   [110]B). An elevated level of TT is associated with an increased BC
   risk (Fig. [111]3C). However, as the incidence of CAD increases, both
   the TT levels and the risk of BC decrease (Fig. [112]3D, [113]E).

Fig. 2.

   [114]Fig. 2
   [115]Open in a new tab

   Results utilizing the IVW method as the primary analytical method for
   two-sample MR analysis between lipids, CAD, TT and BC. TC, total
   cholesterol; TG, triglycerides; LDL-C, low-density lipoprotein
   cholesterol; HDL-C, high-density lipoprotein cholesterol; CAD, coronary
   artery disease; TT, total testosterone. ^*P < 0.05; ^**P < 0.01;
   ^***P < 0.001. ^aSignificance for reverse Mendelian randomization
   between BC and HDL-C/CAD. P > 0.05 indicating the absence of reverse
   causal associations

Fig. 3.

   [116]Fig. 3
   [117]Open in a new tab

   Scatter plots illustrating the genetic correlation among HDL-C, CAD, TT
   and BC using various MR analysis methods. A, B HDL-C on CAD/BC. C TT on
   BC. D, E CAD on TT/BC

   The results of the heterogeneity and horizontal pleiotropy tests are
   presented in Table [118]2. Heterogeneity was observed to varying
   degrees in all MR analysis results (Cochrane’s Q: 54.131–308.828, P:
   0.027- < 0.001). Thus, IVW with random effects analysis for all results
   were performed. In addition, our analysis did not demonstrate
   significant horizontal pleiotropy, indicating that the IVs were
   unlikely to substantially influence the outcomes through pathways
   unrelated to the exposures (Table [119]2). Forest and funnel plots,
   which are consistent with the heterogeneity test results, are included
   in Additional file [120]3 (Figs. S1–10). Furthermore, the robustness of
   the MR analysis results was confirmed by the leave-one-out approach,
   which revealed a minor change in the error line (Additional file
   [121]3, Figs. S11-15). Besides, as evidenced by the MR-Steiger
   directionality test findings, our assessment of causal direction was
   accurate (both P for HDL-C/CAD as exposure, BC as outcome: < 0.001).

Table 2.

   Sensitivity analysis for two-sample MR results: heterogeneity and
   horizontal pleiotropy tests
   Exposure Outcome Heterogeneity test (IVW) Horizontal pleiotropy test
   (MR‒Egger)
   Cochrane’s Q Q_df P value Egger_Intercept SE P value
   HDL-C CAD 223.848 66 < 0.001 < 0.001 0.005 0.985
   HDL-C BC 113.876 62 < 0.001 < − 0.001 0.004 0.904
   CAD BC 54.131 36 0.027 0.008 0.005 0.133
   CAD TT 183.208 39 < 0.001 − 0.002 0.001 0.110
   TT BC 308.828 163 < 0.001 − 0.001 0.002 0.783
   BC HDL-C 92.063 50 < 0.001 < 0.001 0.002 0.797
   BC CAD 278.416 58 < 0.001 − 0.010 0.007 0.180
   [122]Open in a new tab

Reverse two‑sample MR analysis between BC, HDL-C and CAD

   In the TSMR analysis between BC and HDL-C/CAD, a total of 51 and 59
   SNPs were identified respectively (Fig. [123]2). The IVW results
   demonstrated no causal relationships between BC and HDL-C (P = 0.968)
   or between BC and CAD (P = 0.279). Similarly, owing to the
   studies’ heterogeneity, we performed a random effects model for IVW
   methods and did not observe significant horizontal pleiotropy in these
   studies as well (Table [124]2).

Multivariate MR analysis and mediation analysis

   After adjusting for CAD, HDL-C still had a substantial causative
   influence on BC (OR = 1.126, 95% CI [1.059, 1.198], P < 0.001), as
   indicated in Table [125]3. Similarly, after adjusting for HDL-C levels,
   CAD was still significantly inversely associated with BC risk (OR
   = 0.947, 95% CI [0.899, 0.997], P = 0.039). When adjusting for each
   other's presence, TT levels and CAD still had substantial causal
   influences on the risk of BC respectively. The aforementioned results
   suggested that HDL-C levels, CAD, and TT levels might collectively
   contribute to the initiation and progression of BC.

Table 3.

   Multivariate MR results of the relationship between HDL-C/CAD/TT and BC
   Exposure Outcome SNPs (n) MVMR-IVW
   Beta (95%CI) OR (95%CI) P value
   CAD as a mediating factor (HDL-C-CAD-BC)
    HDL-C BC 58 0.119 (0.058, 0.180) 1.126 (1.059, 1.198) < 0.001
    CAD BC 30 − 0.054 (− 0.106, − 0.003) 0.947 (0.899, 0.997) 0.039
   TT as mediating factor (CAD-TT-BC)
    TT BC 141 0.304 (0.136, 0.472) 1.355 (1.146–1.604) < 0.001
    CAD BC 24 − 0.070 (− 0.126, − 0.015) 0.932 (0.882, 0.985) 0.013
   [126]Open in a new tab

   Mediation studies were conducted to investigate if CAD mediated the
   causal effect of HDL-C on BC and whether the effect of CAD on BC was
   mediated by TT (Fig. [127]4). The results indicated that with HDL-C as
   the exposure and CAD as the mediator, the total effect was 0.1193 (Beta
   C) and the mediation effect was 0.0093 (Beta A * Beta B', 95% CI:
   0.0004–0.0210), accounting for 7.8% of the total effect. Furthermore,
   when the exposure was CAD and the mediator was TT, the total effect was
   −0.0673 (Beta B) and the mediation effect was − 0.0041 (Beta D * Beta
   E', 95% CI: 0.0001–0.0023). The percentage of mediating effect was
   6.1%. In addition, the direct effect was 0.11 and − 0.0632,
   respectively. The results above indicate that CAD might mediate the
   causal effect of HDL-C on BC, and TT could act as a mediator between
   CAD and BC.

Fig. 4.

   [128]Fig. 4
   [129]Open in a new tab

   Diagram of effect values for causal pathways in this study. Univariable
   MR results: βA effect of HDL-C on CAD; βB effect of CAD on BC; βC
   effect of HDL-C on BC; βD effect of CAD on TT. βE effect of TT on BC.
   Multivariable MR results: β B', effect of CAD on BC adjusted HDL-C;
   βC', effect of HDL-C on BC adjusted CAD; βE', effect of TT on BC
   adjusted CAD

Common SNP in the causal pathways and related gene expression

   Next, after performing intersection analyses among two causal pathways
   (HDL-CAD-BC, CAD-TT-BC), one overlapping SNP (rs11556924, effect allele
   = T) was identified in the causal relationships between CAD and BC
   (Fig. [130]5A). Then we identified the most relevant coding
   gene—ZC3HC1, for this SNP through the g:SNPense online tool in the
   g:Profiler database. Compared to normal controls, the expression of
   ZC3HC1 was significantly upregulated (all P < 0.05) in several cancer
   types including bladder, bile duct, colon, esophageal cancer,
   glioblastoma and other malignancies, as shown in Fig. [131]5B. However,
   there was no significant difference in the expression of ZC3HC1 in
   Breast invasive carcinoma (BRCA) and its subtypes compared to normal
   tissue (Fig. [132]5B). Meanwhile, the ZC3HC1 mRNA expression of BRCA in
   the GEPIA database was used as evidence for validation (Fig. [133]5C).
   Additionally, immunohistochemistry (IHC) results from the HPA database
   did not demonstrate notable differences in ZC3HC1 protein expression
   (Fig. [134]5D). This implies that it is the functional alterations in
   the protein encoded by the gene, rather than its expression level, that
   may be more pivotal in the causal relationship between CAD and BC.

Fig. 5.

   [135]Fig. 5
   [136]Open in a new tab

   Identification of the hub SNP in two causal pathways and the expression
   of associated coding gene (ZC3HC1) in BC. A The common SNP between
   causal pathways (A-B-C, SNPs associated with causal links of HDL-C on
   CAD, CAD on BC and HDL-C on BC; D-E, SNPs associated with causal links
   of CAD on TT and TT on BC). B A comprehensive pan-cancer analysis of
   the common SNP-associated gene (ZC3HC1) expression using the TIMER
   database. The expression data in BC are displayed within the area
   highlighted by the red frame. Besides, red and blue denote tumor and
   normal tissue samples, respectively. C ZC3HC1 mRNA expression in BRCA
   tissues in the GEPIA2 database. D ZC3HC1 protein expression in both
   normal breast tissue and tumor, as summarized in the HPA database.
   ZC3HC1, Zinc Finger C3HC-Type Containing 1, also named NIPA; BRCA:
   breast invasive carcinoma

Functional enrichment of ZC3HC1 co-expressed genes and PPI network analysis
of ZC3HC1 co-expressed proteins

   To further elucidate the association network of this gene, the
   co-expression profiling of ZC3HC1-associated genes was performed on
   BRCA mRNA sequencing data utilizing the functional module of
   Linkedomics. The heat map displays the top 50 genes that are positively
   (Fig. [137]6A) or negatively (Fig. [138]6B) correlated with ZC3HC1
   respectively. GO analysis of ZC3HC1co-expressed genes indicates that
   they were mainly implicated in small GTPase mediated cellular signal
   transduction (Fig. [139]6C). The CC comprised neuronal cell body,
   cell-substrate junction and basal plasma membrane and the primary MF is
   characterized by activities including DNA-binding transcription
   activator activity and transcription factor binding (Fig. [140]6C).
   KEGG pathway analysis identified significant enrichment in the PI3
   K-Akt and MAPK signaling pathways, with additional notable pathways
   including the breast cancer signaling pathway (Fig. [141]6D). In
   conclusion, the GO and KEGG analyses of co-expressed genes suggest that
   ZC3HC1 plays a pivotal role in intracellular signal transduction
   pathways. And the correlation analysis of the top 10 co-expressed genes
   related to ZC3HC1 was shown in Additional file [142]3 (Fig. S16).

Fig. 6.

   [143]Fig. 6
   [144]Open in a new tab

   Analysis of co-expressed genes of ZC3HC1 in BRCA and ZC3HC1
   co-expressed proteins. A, B Heat maps depict the top 50 genes
   exhibiting positive (A) and negative (B) correlations with ZC3HC1 in
   BRCA. Red: positive correlation, blue: negative correlation. C GO
   enrichment analysis identified the most significantly enriched
   biological processes, cellular components, and molecular functions of
   the co-expressed genes. D KEGG pathways enriched by co-expressed genes.
   E PPI network of ZC3HC1 co-expressed proteins. The color of the lines
   represents the different interaction information

   The PPI network was established using STRING, encompassing 11 highly
   interconnected proteins (Fig. [145]6E). GO analysis indicated that
   these proteins are mainly localized in the Cul7-RING/SCF ubiquitin
   ligase complex (Additional file [146]3, Fig. S17 A) and are primarily
   involved in the SCF-dependent proteasomal ubiquitin-mediated protein
   catabolic process (Fig. S17B). Reactome and KEGG pathway enrichment
   analysis from STRING database revealed that these proteins participate
   in nuclear processes activated by ALK signaling in cancer, as well as
   in cell cycle regulation and cancer-associated pathways (Fig. S17 C,
   D).

Discussion

   In this TSMR study, we demonstrated a causal association between
   genetically determined HDL-C levels and BC within the European
   population. MVMR and mediation analysis indicated that CAD contributes
   a mediating function (proportion of mediating effect: 7.8%) in the
   causal connection between HDL-C and BC. Further mediation study also
   emphasized the significant role of TT as a mediator (Proportion of the
   mediating effect: 6.1%) in the causal pathway between CAD and BC.
   Finally, bioinformatics analyses identified a putative key SNP
   (rs11556924) implicated in the causal association, along with its
   corresponding coding gene, ZC3HC1. Notably, no significant differences
   in ZC3HC1 expression were observed between normal and cancerous
   tissues, suggesting that alterations in the function of its encoded
   protein caused by genetic variation may play an essential role in the
   causal correlation between CAD and BC.

   Evidence from previous meta-analyses and observational research
   suggested that there was a negative relationship between HDL-C/TC
   levels and BC risk [[147]9, [148]58], whereas TG/LDL-C levels were
   positively associated with BC [[149]59, [150]60]. Yet, TC/HDL-C levels
   have been found to positively correlate with BC in certain studies
   [[151]61], while TG/LDL-C levels were inversely correlated with BC risk
   [[152]15, [153]58]. In addition, multiple studies together showed that
   none of these four lipids appeared to significantly correlate with BC
   [[154]9, [155]59, [156]62]. In contrast to the aforementioned studies,
   our MR analyses efficiently clarify these contradictory results and
   indicate probable directions of effect, which is typically impossible
   to accomplish in other observational studies given the influences of
   reverse causation and the existence of confounding factors.
   Additionally, according to Nowak and colleagues’ MR study, genetically
   elevated LDL-C levels might increase BC risk, while HDL-C and TG levels
   were not related to BC development [[157]63], which contradicted the
   findings of this study. This may be explained by the fact that they
   used a rigorous pruning procedure to separate the influence of
   different lipid traits, which would diminish statistical power owing to
   the resilient genetic interrelation between these traits. The
   multivariate method employed in this research is a different approach
   for accessing exposure effects while adjusting for other relevant
   exposures. Moreover, MR studies from Johnson et al. and Beeghly-Fadiel
   et al. suggested that higher HDL-C levels were correlated to an
   increased risk of BC, which aligned with our primary finding [[158]20,
   [159]64]. However, they also noted a positive relationship between
   LDL-C levels and BC risk, which might stem from the fact that they used
   genetic data for a much smaller defined range (Core LDL pathway genetic
   instrument). Besides, as also indicated in our analysis, their
   investigation discovered a previously unreported inverse connection
   between TG and BC (OR = 0.94, 95% CI [0.90, 0.98], P = 2.60 × 10^–3),
   although this might be attributed to the correlation of TG with HDL-C
   rather than the independent effect of TG. Moreover, our findings
   broadly align with those of the recent MR research on lipids and BC by
   Beeghly-Fadiel et al. [[160]64]. Multiple MR analyses were performed in
   both studies and a positive association of HDL-C with BC was observed
   as well as other lipid traits (TC/TG/LDL-C) did not appear to be
   related to the incidence of BC. However, our study goes a step further
   by analyzing the possible mediating role of CAD in the HDL-C-BC causal
   relationship, as well as the role of TT as a mediator between CAD and
   BC, providing deeper insights into the superficial causal
   relationships.

   HDL, representing approximately 25–30% of the circulating lipoproteins,
   plays a crucial role in reverse cholesterol transport [[161]65]. HDL-C
   is an important indicator of HDL. Many previous studies have discussed
   HDL-C as the protective factor for CAD, citing its antithrombotic,
   anti-inflammatory, reverse cholesterol transport and antioxidant
   properties as possible mechanisms [[162]66, [163]67]. But recent
   growing explorations also suggested a complicated relationship between
   genetically determined HDL-C levels and BC risk [[164]64, [165]68].
   More importantly, the specific mechanisms through which HDL-C
   influences the development and progression of BC remain largely
   unelucidated. According to current research, HDL-C tends to promote the
   proliferation of BC cells [[166]69, [167]70]. Furthermore, the
   association between elevated HDL-C levels and BC risks seems to be
   prompted by the heterogeneous populations of HDL-C particles that may
   have vital functions beyond their levels, such as the capabilities and
   status of different HDL particles [[168]71]. It appears that HDL-C
   levels alone may not sufficiently explain the biological mechanisms
   underlying its causal association with BC. Instead, the status and
   function of HDL—the particle's functional core, likely play a more
   crucial role. For instance, oxidized HDL and diabetic HDL are more
   capable of facilitating the tumorigenic processes of distinct BC cells
   [[169]69, [170]72]. More specifically, different HDL particles could
   stimulate BC cells to synthesize and release more matrix
   metalloproteinase (MMP)−2/9 and vascular endothelial growth factor-C
   (VEGF-C) [[171]72]. Additionally, they likely orchestrated BC cell
   migration/invasion via the p38 mitogen-activated protein kinase (MAPK)
   pathway, as well as the extracellular signal-regulated kinase (ERK) and
   Akt pathways [[172]72]. In addition, some HDL particles significantly
   stimulate the expression of cell surface integrins and protein kinase C
   (PKC) to augment metastatic potential and invasive capability
   [[173]69]. Furthermore, studies have reported that the receptor for
   HDL, Scavenger receptor class B type I (SR-BI), might play a pivotal
   role by facilitating the selective uptake of HDL-cholesteryl ester
   (HDL-CE, the primary components of HDL-C) and activating oncogenic
   signaling pathways [[174]73, [175]74]. To be specific, SR-BI interacts
   with PDZK1 to trigger PI3K, c-Src, and Erk1/2 signaling, promoting BC
   cell proliferation and migration [[176]75–[177]77]. Elevated SR-BI
   expression in BC cell lines [[178]75, [179]78], tissues [[180]78,
   [181]79] and mouse models [[182]80] further underscores its oncogenic
   potential. Therefore, in light of evidence suggesting a positive
   correlation between HDL-C levels and the number of HDL particles
   [[183]81], it is sensible to conclude that elevated HDL-C levels may
   indirectly impact the occurrence of BC through the aforementioned
   mechanisms. Besides, HDL-C particles can directly shift from
   anti-inflammatory to pro-inflammatory under oxidative stress and might
   promote cancer initiation and progression [[184]82, [185]83]. Likewise,
   cholesterol and its oxysterol metabolites, whether circulating or
   within the tumor microenvironment, may directly influence the
   proliferation of mammary tissue, thereby promoting breast tumorigenesis
   [[186]84, [187]85]. In conclusion, the aforementioned studies provide
   potential biological mechanisms linking elevated HDL-C levels with the
   higher risk of BC. But further investigations are necessary to
   comprehend the detailed mechanisms by exploring not only the change of
   HDL-C levels, but also the functionality of diverse key apolipoprotein
   components within HDL-C.

   Although the causal association between CAD and BC and specific
   mechanisms behind the causality receive limited attention, our study
   explores the causal relationship between CAD and BC by using TSMR and
   determines that CAD mediates the causation between HDL-C and BC through
   MVMR and mediation analyses. The sensitivity analyses also confirm that
   our findings are consistent and robust. Notably, for better
   understanding the mechanism behind the effect of CAD on BC risk, we
   identified TT as a genetically plausible mediator in the causality
   through further mediation analysis as observational studies have
   demonstrated an inverse relationship between CAD and TT levels among
   female populations [[188]86, [189]87], MR study (OR = 1.14, 95% CI
   [1.09, 1.21], P = 4.07 × 10^–7) and another study (OR = 1.14, 95% CI
   [1.08, 1.20], P = 1.60 × 10^–6) have indicated that high TT levels were
   linked to the risk of BC [[190]35, [191]42]. It therefore seems
   reasonable to believe there is a possibility that TT mediates the
   causal association between CAD and BC on the basis of these findings.
   Moreover, their studies revealed that this relationship was
   statistically significant in the estrogen receptor-positive (ER+) BC
   subtype. The correlation with ER+ neoplasms involving testosterone
   might be interpreted by the following proposed mechanisms. The first
   possible answer is that high levels of testosterone are converted into
   more estradiol [[192]88], which in turn binds to ER and activates the
   transcription of pro-growth genes while suppressing the expression of
   growth-inhibitory regulators, thus driving the proliferation of BC
   cells [[193]89]. An alternative explanation is that the expression of
   the androgen receptor (AR) is positively correlated with ER expression
   in tumors, implying that ER expression could serve as a possible
   indicator of AR expression and potential interactions of
   androgen-estrogen signaling pathway in cancer development [[194]90,
   [195]91]. This is further evidenced by the discovery that AR expression
   is present in only 20–30% of ER− BC cells [[196]92]. In order to
   unravel the mechanism by which testosterone functions in BC, further
   detailed research focusing on BC subtypes stratified by the expression
   status of AR/ER and the specific effects of fluctuations in
   testosterone levels on key components of estrogen-related signaling
   pathways may be helpful.

   ZC3HC1 encodes an F-box-containing protein that serves as a crucial
   component of an SCF-type E3 ubiquitin ligase complex, which
   orchestrates cell cycle progression by modulating the initiation of
   mitosis [[197]93]. The G2/M transition necessitates the formation of a
   regulatory complex between Cyclin-B1 and cyclin-dependent kinase 1
   (CDK1). Upon activation, the ubiquitin ligase complex facilitates the
   proteasomal degradation of cyclin-B1, thereby preventing premature
   entry into mitosis and ensuring proper cell cycle control [[198]93].
   The coding SNP rs11556924 (C > T) within the ZC3HC1 gene results in an
   arginine-to-histidine polymorphism at amino acid residue 363 in the
   Nuclear Interaction Partner of ALK (NIPA) protein [[199]93]. It has
   been reported that the Arg-363 allele (C) represents the predominant
   variant (allele frequency = 0.62) and is linked to a 9% increase in CAD
   risk per allele [[200]94], while in our study the effect allele is T.
   Jones et al. further demonstrated that the His-363 (T) variant exhibits
   a 44% increase in phosphorylation at the Ser-395 site of NIPA induced
   by Cyclin-B1-CDK1 compared to the Arg-363 allele, which would cause
   higher accumulation of Cyclin-B1 in the nucleus of these cells, thereby
   facilitating the progression of mitosis [[201]95]. Although the study
   did not provide enough evidence that the difference in cell
   proliferation phenotype between two genotypes was significant, this may
   be explained by the compensation of cells’ inherent regulatory
   mechanisms under normal conditions, or the small difference in
   proliferation time (0.2% of the entire cell cycle) being too small to
   allow precise detection of changes in cell number within this timescale
   [[202]95]. It is important to emphasize that, within the intricate
   genetic landscape of BC progression, even minor alterations in cell
   cycle regulation can exert a profound influence on the trajectory of
   the disease over months or years of development.

   Notably, our study is distinguished by several advantages. First, to
   the best of our knowledge, this study is the first MR analysis that
   performs a two-step mediation analysis to explore in depth the
   possibility of a complete causal chain between HDL-C and BC, which may
   provide more potential targets for intervention in the incidence of BC.
   Secondly, the application of MR, utilizing IVs to rigorously assess the
   causal association, could reduce the bias induced by confounding
   factors and reduce the likelihood of reverse causality relative to
   observational studies. Furthermore, this study features an extensive
   sample size and powerful IVs for each lipid trait (all F-statistics
   > 10), and instrument validity was also carefully assessed. In
   addition, data on the exposures and outcomes analyzed were sourced from
   large, non-overlapping GWAS datasets, which made the results more
   robust. Besides, our results remained consistent even after adjusting
   for potential pleiotropic effects through the application of MR-Egger
   regression and multivariable weighted regression. Moreover, our study
   represents the first indication that CAD-associated SNP (rs11556924)
   may be implicated in the BC risk, thereby offering new perspectives on
   the mechanistic pathways of BC development and potential avenues for
   targeted therapeutic interventions.

   The association with a heightened risk of BC is unexpected given that
   greater HDL-C levels are typically thought to be beneficial. Although
   we can discuss possible mechanisms in the literature that underpin this
   finding, the influence of the underlying factors remains largely
   unclarified. Therefore, the following directions should be the focus of
   future studies. First, the relationship between lipids and BC may vary
   based on the clinical statuses of the patient's condition, including
   the disease's stage or grade and the presence of any coexisting
   metabolic disorders. Secondly, the details and exact nature of the
   associations between BC subtypes and lipid levels, along with the
   specific mechanisms underlying this relationship, remain elusive and
   warrant further rigorous investigation. Furthermore, as indicated by
   the research, if the indicators of lipid-metabolite combinations are
   identified as risk factors for BC as reported [[203]96], it may become
   imperative to thoroughly investigate the genetically determined
   interactions between the levels of lipids and various metabolites.
   Besides, the genetically determined function of HDL may merit as much
   attention as HDL-C levels. Also, given the diversity of genetic
   backgrounds associated with lipid levels and metabolism, investigating
   the causal relationships across different ethnicities is essential for
   cross-population comparisons and validation of our findings. Notably,
   the paradoxical roles of HDL-C in modulating the risk of CAD and BC
   presents significant research challenges. Mechanistically, HDL-C exerts
   cardioprotective effects primarily through reverse cholesterol
   transport, antioxidative activity, and anti-inflammatory properties.
   However, HDL-C may facilitate BC development by disrupting cholesterol
   homeostasis, enhancing tumor cell membrane fluidity, and activating
   oncogenic signaling pathways. Moreover, our findings indicate that CAD
   and TT levels may serve as mediators in this association, reinforcing
   the context-dependent functionality of HDL-C and underscoring the
   necessity for a nuanced understanding of its role across different
   pathological states. Therefore, we propose that future research should
   aim to define the optimal range of HDL-C levels and explore the
   appropriate interventions (e.g. pharmacological treatments and
   lifestyle modifications) to maintain HDL-C within this range, thereby
   achieving better control for both CAD and BC. Finally, while our study
   primarily focuses on the genetic determinants of lipid metabolism in BC
   etiology, we acknowledge the necessity of integrating non-genetic
   influences to achieve a more comprehensive understanding. Environmental
   and lifestyle factors, such as diet, physical activity or hormonal
   exposures, may modulate lipid metabolism and, in turn, influence BC
   susceptibility. More importantly, these factors likely interact with
   genetic variants, potentially amplifying or mitigating the initial
   effects on cancer risk. Future investigations incorporating intricate
   gene-environment interactions will be crucial for refining BC risk
   stratification and developing targeted prevention strategies.

Limitations

   However, there are also several important caveats to consider in our
   analyses. First, MR must satisfy several assumptions to make accurate
   causal inferences [[204]97]. Although statistical approaches were
   employed to identify and correct for breaches of these assumptions, it
   is necessary to use additional causal inference frameworks or different
   sorts of verification since these approaches cannot be guaranteed to
   adjust for all types of confounders. Second, MR is constrained in its
   ability to make inferences about trait information among the population
   where the GWAS originated (in this study is the European population).
   Therefore, given the potential differences in the genetic architecture
   of lipid profiles [[205]98], the applicability of our conclusions to
   other populations may be restricted. Moreover, it is necessary to
   consider that our results could be attenuated due to the discovery of
   BC’s correlation with the usage of lipid-lowering medications
   [[206]99]. This raises the possibility that the real underlying
   causative factor may not be restricted to lipid levels themselves, but
   may include another trait for which lipids serve as a proxy. Finally,
   our two-step mediation analyses suggest that CAD mediates partial HDL-C
   effects on BC, whereas TT mediates a portion of CAD effects on BC,
   which is certainly an exciting finding. However, the design of the
   two-step mediation study may introduce the possibility of more
   undiscovered mediators that need to be explored in more aspects or
   sessions. Finally, our study investigates causality exclusively from a
   genetic perspective and does not extend to experimental validations (in
   vitro or in vivo functional assays) to corroborate the observed
   associations. Therefore, further studies are urgently needed to
   complete the puzzle of lipids-BC interactions and explore the unique
   function of the CAD-associated SNP (rs11556924) in BC development.

Conclusions

   To conclude, this MR analysis provided genetic evidence that higher
   HDL-C levels are associated with increased BC risk, with CAD mediating
   this effect. Further mediation analysis indicated that TT mediated the
   causal effect of CAD on BC. Importantly, the findings provide further
   evidence against the broad application of HDL-C-raising therapies in
   the general population while highlighting their potential value in
   precision medicine for identifying women at higher risk of BC based on
   special SNP (rs11556924) and the function of its associated gene
   ZC3HC1, which may further contribute to the mechanistic understanding
   about the initiation and advancement of BC as well as the potential
   novel metabolic targets for therapeutic intervention.

Supplementary Information

   [207]Supplementary file1 STROBE-MR checklist^ (44.2KB, docx)
   [208]12672_2025_2597_MOESM2_ESM.xlsx^ (77.1KB, xlsx)

   Supplementary file2 Table S1-5. MR results for causal effect between
   exposures and outcomes. Table S6-10. Information of SNPs for causal
   association analysis between exposures and outcomes
   [209]12672_2025_2597_MOESM3_ESM.docx^ (8.6MB, docx)

   Supplementary file3 Figs.1-5. Forest plots for causal effect of
   exposures on outcomes. Figs.6-10. Funnel plots for causal effect of
   exposures on outcomes. Figs.11-15. Leave-one-out analyses for causal
   effect of exposures on outcomes. Fig.16. ZC3HC1 gene correlation
   analysis. Fig.17. GO/KEGG and Reactome enrichment analysis of
   co-expressed protein

Acknowledgements