Genetic evidence for potential molecular mediators underlying the causal relationship between obesity and breast cancer: a two-step, two-sample Mendelian randomization study.

Introduction
Breast cancer has now surpassed lung cancer as the most commonly diagnosed cancer, accounting for approximately 24.5% of all female cases in 2020 [1]. Obesity, a widely recognized global public health challenge, plays a complex role in the development of female breast cancer [2]. Conventional wisdom has long posited a positive association between adult body mass index (BMI, a measure of adult general obesity) and breast cancer risk, especially in postmenopausal women [3, 4]. However, recent genetic analyses present an intriguing observation that genetically predicted BMI was inversely associated with susceptibility to breast cancer [5–9]. This counterintuitive finding has challenged our understanding of obesity-related cancer risk and sparked renewed interest in identifying potential pathways that support the complex relationship between BMI and breast cancer development.
Circulating biomarkers, such as adipocytokines and inflammatory factors, have been implicated in breast cancer risk and may act as mediators in the BMI-breast cancer relationship [10]. Several nested case-control studies have attempted to confirm this, but the results remain inconclusive [11–15]. On the basis of 188 breast cancer cases and 285 matched controls, Dashti et al. reported a statistically significant indirect effect (OR: 1.58; 95%CI: 1.17, 2.43) of insulin (INS) in the association between high BMI (> 30 kg/m2) and postmenopausal estrogen receptor-positive breast cancer risk [12]. However, in another study with a similar but larger sample size (149 cases and 1,029 controls), Dashti et al. observed an indirect effect of estradiol (E2) (OR: 1.56; 95%CI: 1.11, 2.19) but not INS (OR: 1.12; 95%CI: 0.68, 1.84) [13]. Similar findings were reported by Hvidtfeldt et al. [14]. As for other common adipocytokines and inflammatory biomarkers, such as adiponectin (ADP), leptin (LEP), and C-reactive protein (CRP), no statistically significant indirect effects were observed in only three of the available studies [11, 12, 15]. Despite these efforts, several gaps remain. First, the direction of the relationship was difficult to confirm in these conventional observational studies, as the exposures and mediators were collected simultaneously at baseline. Second, a single measurement of biomarkers captures short-term levels but may not be representative in the long term. Third, given the limited sample size, estimates of indirect effects may not be sufficiently robust, and the identification of dependable mediators to elucidate the underlying mechanistic pathway remains challenging.
Mendelian randomization (MR), which capitalizes on the random assortment of genetic alleles during gamete formation, is a powerful tool that uses genetic variants as instrumental variables (IVs) to make causal inference [16]. This approach provides an unprecedented possibility to determine the causal mediating role of circulating biomarkers in the relationship between BMI and breast cancer risk. By assessing whether genetically predicted exposures influence outcomes through affecting multiple mediators, MR mediation analysis can provide deeper insights into complex biological pathways [17, 18]. Additionally, using genetic variation as a surrogate for lifetime average biomarkers levels could also address the limitation of poor representation of single-time measurements in observational studies. Therefore, in this study, we employed an MR mediation approach to individually evaluate the causal mediating roles of 8 circulating biomarkers in the BMI-breast cancer relationship and to further elucidate the mediation network when these mediators are considered simultaneously.

Methods
This study adhered to the guidelines of the Strengthening the Reporting of Observational Studies in Epidemiology—Mendelian Randomization (STROBE-MR) (Supplementary file 1) [19, 20]. The summary-level data utilized in our study were publicly available, all of which had received ethical approval in the original studies. The overall study design is shown in Fig. 1.

Data sources

Exposures
The hitherto largest GWAS of BMI was conducted via a collaboration of the UK Biobank (UKB) and the Genetic Investigation of Anthropometric Traits (GIANT) consortium in 2019, including ~ 700 000 adults of European ancestry [21]. Specifically, the GIANT consortium performed a meta-analysis that included reintegration of height and weight data from approximately 450 000 European ancestry participants in the UKB into the dataset of 322 154 participants from the GIANT consortium.

Outcomes
GWAS summary statistics for breast cancer were retrieved from the most updated GWAS conducted in 2020 involving 133 384 cases and 113 789 controls, combining results from 82 studies participating in the Breast Cancer Association Consortium (BCAC) and 11 large-scale studies from more than 20 countries [22]. All the subjects are European females and aged from 18 to 79 years old. This GWAS expanded upon a previous GWAS [23] from the Breast Cancer Association Consortium (2017) with an additional 10 407 cases and 7 815 controls (10% increase), and identified 32 novel susceptibility loci upon the 153 previously detected loci.

Mediators
The selection of circulating biomarkers was prioritized based on their biological relevance to obesity-related diseases, specifically focusing on the insulin/insulin-like growth factor (IGF) axis and chronic low-grade inflammation pathways [24]. Accordingly, we obtained publicly available GWAS summary statistics for 8 representative biomarkers associated with these pathways, including ADP [25], LEP [26], soluble leptin receptor (sOB-R) [27], resistin (RETN) [26], INS [28], insulin-like growth factor-1 (IGF-1) [29], insulin-like growth factor binding protein-3 (IGFBP-3) [27], and CRP [30]. Detailed information regarding data sources and the specific instrumental variables for each biomarker is provided in Supplementary Tables S1 and S2.

Instrument selection
Single nucleotide polymorphisms (SNPs) were selected as IVs based on the following criteria: (i) We included independent SNPs that were associated with risk factors at the genome-wide significance level (P < 5⋅10− 8). Given that the GWAS sample sizes for five biomarkers (ADP, LEP, sOB-R, RETN and IGFBP-3) were relatively small (N < 40 000) and that a limited number (< 10) of SNPs reached the conventional threshold, we applied a relaxed significance threshold at P < 5⋅10− 6 to select IVs for these biomarkers. (ii) We excluded SNPs associated with breast cancer at genome‐wide significance (P < 5⋅10− 8). (iii) IVs with an average F-statistic less than the threshold of 10 were not included to mitigate weak instrument bias. The proportion of variance in the phenotype explained by genetic variants (R2), sample size (Ν) and number of instruments (Κ) were used to calculate F-statistics via the formula F = .

Statistical analysis
A comprehensive mediation MR analysis was performed to identify molecular mediators of the relationship between BMI and breast cancer risk. A schematic overview of this study is shown in Fig. 1. First, we employed bidirectional univariable MR analysis to establish the relationships between genetically predicted BMI, circulating biomarkers, and breast cancer susceptibility. Second, based on the two-step MR mediation analysis framework, we conducted single-mediator models to examine the indirect effects of each biomarker in the BMI-breast cancer causal relationship. Third, based on the stepwise MR mediation analysis framework, we constructed multiple-mediator models to explore the mediation network of biomarkers, further elucidating the intricate pathogenic pathways linking genetically predicted BMI to breast cancer susceptibility.

Univariable Mendelian randomization analysis
To investigate the total effect of genetically predicted BMI on breast cancer risk, we used inverse-variance weighted (IVW) approach [31], MR-Egger (Mendelian randomization-Egger) regression [32], weighted-median approach [33] and MR-PRESSO (Mendelian Randomization Pleiotropy Residual Sum and Outlier) [34]. To evaluate the robustness of the results, we performed a series of sensitivity analyses, including leave-one-out analysis [35] and additional assessments that excluded palindromic SNPs with strand ambiguity or pleiotropic SNPs associated with potential confounding traits (age at menarche, age at natural menopause, smoking and drinking) according to the GWAS Catalog [36]. In addition, we performed bidirectional MR analysis to evaluate whether genetic predisposition to breast cancer influences BMI.

Two-step Mendelian randomization mediation analysis
To investigate the mediating effect of each biomarker on the relationship between BMI and breast cancer susceptibility, a two-step MR analysis was performed. A total of 281 female-specific index SNPs of BMI were utilized to estimate the causal effects of genetically predicted BMI on 8 biomarkers using univariable MR methods. Biomarkers with statistically significant associations with BMI were identified as candidate biomarkers. Composite IVs of BMI and each candidate biomarker were then clumped (r2 = 0.001, kb = 500) and included in a multivariable MR model to estimate the direct effects of each biomarker on breast cancer risk. The product and delta method were used to obtain indirect effects and confidence intervals. The proportion mediated was calculated as the indirect effect divided by the total effect.

Stepwise Mendelian randomization mediation analysis
To evaluate the mediatory effects of multiple candidate biomarkers simultaneously and to identify a mediation network, stepwise MR mediation analysis was conducted. Hypothetical models incorporating multiple candidate mediators were first constructed based on a systematic review by Nimptsch et al. [24]. This review suggests that the INS/IGF axis and chronic low-grade inflammation pathway might be the two major pathways linking obesity to cancer risk, and that specific adipocytokines, such as LEP and ADP, act through these major pathways. Therefore, a key issue arises as whether the INS/IGF axis pathway precedes the inflammation (CRP) pathway or it is the other way around.
To test these alternatives, step-by-step MR mediation was performed. We first examined the direct effect of exposure (BMI) and all mediators with an outcome (breast cancer) through multivariable MR. After that, according to the hypothetical model, each mediator was sequentially designated as the temporary secondary outcome, prioritizing the most distal mediator at each analytical step. The same multivariable MR was performed to estimate the direct effect of upstream factors (i.e., the exposure and remaining candidate mediators) on the temporary outcome until the procedure was iterated through all mediators. The product method was used to obtain the indirect effect by sequentially multiplying the direct effects along the same mediation pathway, and the delta method was used to obtain the confidence interval. The proportion mediated by that pathway was calculated as the indirect effect divided by the total effect.

Secondary analysis
To validate the robustness of the results, a stringent genome-wide significance threshold (P < 5⋅10− 8) was also employed to identify proxy SNPs for all biomarkers. Two-step and stepwise MR mediation analyses were re-ran.

Genetic correlation analysis
To understand the shared genetic background of BMI with biomarkers and breast cancer, linkage disequilibrium score regression (LDSC) analysis was conducted [37]. Full-set GWAS summary data were used to estimate genome-wide genetic correlations (), which quantify the intrinsic average sharing of genetic effects between pairs of traits that are independent of environmental factors [38].
All MR analyses were conducted using the “TwoSampleMR” package (version 0.5.6) and the “MendelianRandomization” package (version 0.5.1) in R (version 4.2.2). Statistical significance was set at P < 0.05.

Results
281 and 184 index SNPs explained 4% and 30.2% of the phenotypic variance of BMI and breast cancer, respectively. For biomarkers, the percentage of phenotypic variance explained by the index SNPs ranged from 0.2% (INS with 19 index SNPs) to 42.4% (sOB-R with 11 index SNPs). The F-statistics for all IVs ranged from 10.7 to 608.6, suggesting strong instruments (Table S1).

Genetic correlations between BMI, circulating biomarkers and breast cancer
LDSC analysis revealed positive genetic correlations of BMI with LEP ( = 0.785; P = 1.20 × 10− 13), RETN ( = 0.115; P = 1.54 × 10− 2) and CRP ( = 0.546; P = 1.13 × 10− 47), as well as negative genetic correlations with INS ( = -0.114; P = 2.00 × 10− 4) and IGF-1 ( = །0.135; P = 1.15 × 10− 17). Additionally, a positive genetic correlation between CRP and breast cancer ( = 0.042; P = 2.70 × 10− 2) was also observed. These results indicated a shared genetic background among these traits (Figure S1).

Causal effects between BMI and breast cancer risk
Univariable analysis using IVW method revealed a negative association between genetically predicted BMI and breast cancer susceptibility (OR: 0.894, 95%CI: 0.832 to 0.960; P = 2.06 × 10− 3). The directions of the associations were consistent across the weighed-median approach, MR-Egger regression and MR-PRESSO method. There was no evidence to support a reverse association (genetically predicted breast cancer risk → BMI) (Figure S2). Sensitivity analyses excluding 19 pleiotropic SNPs or 34 palindromic SNPs, as well as leave-one-out analysis, revealed similar findings (Figure S3-4, Table S3). Multivariable analysis, which adjusted for genetic liability to the biomarkers, showed that the association of genetically predicted BMI with breast cancer susceptibility remained statistically significant even after conditioning on leptin (OR: 0.911, 95% CI: 0.834 to 0.994), insulin (OR: 0.854, 95% CI: 0.783to 0.931), IGF-1 (OR: 0.913, 95% CI: 0.850 to 0.980) or CRP (OR: 0.861, 95% CI: 0.795 to 0.932) (Table 1).

Causal effects between BMI and circulating biomarkers
Univariable analysis using IVW method showed a positive association of genetically predicted BMI with genetically predicted circulating LEP (β: 0.453, 95%CI: 0.380 to 0.526; P = 4.10 × 10− 34), INS (β: 0.198, 95%CI: 0.166 to 0.229; P = 9.44 × 10− 35) and CRP (β: 0.429, 95%CI: 0.360 to 0.498; P = 4.32 × 10− 34) levels, as well as a negative association with genetically predicted circulating IGF-1 levels (β: −0.118, 95%CI: −0.160 to − 0.076; P = 3.11 × 10− 8) (Fig. 2). These results were also supported by the other three univariable methods and the sensitivity analyses, with estimates consistent in direction. No statistically significant associations were found between genetically predicted BMI and genetically predicted circulating ADP, RETN, LEPR, or IGFBP-3 levels (Figure S5). Therefore, LEP, INS, IGF-1 and CRP were identified as candidate mediators in mediation analysis.

Mediation effects of each candidate mediator in the BMI-breast cancer relationship
Two-step MR mediation analysis revealed that only the direct effect of genetically predicted circulating CRP levels on breast cancer susceptibility was statistically significant (OR: 1.110, 95%CI: 1.035 to 1.191; P = 3.65 × 10− 3) when adjusted for genetically predicted BMI. The indirect effect of CRP was 1.046 (95%CI: 1.014 to 1.079; P = 4.83 × 10− 3), with a mediation proportion of − 39.91% (Table 1). In contrast, no statistically significant indirect effect on breast cancer susceptibility was observed for the other three candidate mediators, including LEP (OR: 0.998, 95%CI: 0.961 to 1.036; P = 0.90), INS (OR: 1.033, 95%CI: 0.987 to 1.081; P = 0.16), and IGF-1 (OR: 0.994, 95%CI: 0.988 to 1.001; P = 0.10). Using a stringent genome-wide significance threshold (P < 5⋅10− 8) for all IVs, the results of secondary analysis of each candidate biomarker’s indirect effect were consistent with the findings of the main analysis (Tables S4-5).

Mediation network of BMI, candidate mediators and breast cancer risk
Two hypothetical multiple-mediator models with different relationships between the INS/IGF axis and the chronic low-grade inflammatory pathway were respectively verified using the stepwise MR mediation analytical framework (Figure S6).

Multiple-Mediator Model 1
In Model 1 (changes of INS/IGF axis preceding changes of chronic low-grade inflammation), multivariable MR analysis showed direct effects of genetically predicted BMI (OR: 0.838, 95%CI: 0.766 to 0.917; P = 1.18 × 10− 4), CRP (OR: 1.119, 95%CI: 1.037 to 1.206; P = 3.61 × 10− 3), and IGF-1 levels (OR: 1.062, 95%CI: 1.002 to 1.124; P = 4.10 × 10− 2) on breast cancer risk (Fig. 3, panel a). In the following steps of the mediation analysis, we found strong evidence for the direct effects of genetically predicted BMI (β: 0.353, 95%CI: 0.274 to 0.432; P = 1.66 × 10− 18) and IGF-1 (β: −0.145, 95%CI: −0.200 to − 0.090; P = 2.60 × 10− 7) on the CRP levels (Fig. 3, panel b). Step-wisely, we also observed direct effects of BMI (β: −0.166, 95%CI: −0.231 to − 0.101; P = 5.91 × 10− 7) and INS (β: −0.145, 95%CI: −0.200 to − 0.090; P = 2.60 × 10− 7) on the IGF-1 levels (Fig. 3, panel c), a direct effect of BMI on the INS levels (β: 0.203, 95%CI: 0.165 to 0.242; P = 2.47 × 10− 25) (Fig. 3, panel d), and a positive association between BMI and the LEP levels (β: 0.453, 95%CI: 0.380 to 0.526; P = 4.10 × 10− 34) (Fig. 3, panel e).

The indirect effects of the five potential pathways in Model 1 were estimated respectively (Fig. 4). In the association between genetically predicted BMI and breast cancer susceptibility, we identified an indirect effect solely mediated by CRP (θ4 × θ2) (OR: 1.040, 95%CI: 1.012 to 1.070; P = 5.73 × 10− 3) and an indirect effect sequentially mediated by IGF-1 and CRP (θ6 × θ5 × θ2) (OR: 1.003, 95%CI: 1.000 to 1.005; P = 2.38 × 10− 2), with mediation proportions of − 35.25% and − 2.39%, respectively. No statistically significant indirect effects were observed for the remaining three pathways.

Multiple-Mediator Model 2
However, in Model 2 (changes of chronic low-grade inflammation preceding changes of INS/IGF axis), the indirect effects of only two potential pathways were estimated. Detailed estimates of Model 2 can be found in Figure S7. We only observed an indirect effect mediated by CRP alone (θ7 × θ2) (OR: 1.048, 95%CI: 1.014 to 1.084; P = 5.05 × 10− 3), whereas no statistically significant indirect effect was observed for the sequential mediation of CRP and IGF-1 (Figure S8).
Taken together, two important mediation pathways were identified in the relationship between genetically predicted BMI and breast cancer risk, including the indirect pathway mediated by CRP individually and the indirect pathway mediated by IGF-1 and CRP sequentially. Results of the secondary analysis supported the findings of the main analysis (Tables S6-7).

Discussion
Our MR mediation study evaluated the mediating effects of eight circulating biomarkers both individually and simultaneously on the relationship between genetically predicted BMI and breast cancer risk. Utilizing data from the hitherto largest GWAS(s) available for each trait, we observed positive associations between genetically predicted BMI and genetically predicted circulating LEP, CRP, and INS levels, as well as a negative association with genetically predicted circulating IGF-1 levels. Further analyses restricted to these four candidates revealed an indirect effect mediated solely by genetically predicted circulating CRP on the relationship between genetically predicted BMI and breast cancer susceptibility. Moreover, multiple-mediator models, accounting for the complex interplay among these candidates, demonstrated that in addition to the indirect pathway mediated by CRP, the indirect pathway sequentially mediated by IGF-1 and CRP was also recognized. The proportions mediated by these two pathways were − 35.25% and − 2.39%, respectively.
Adipocytes are indispensable components of systemic metabolic homeostasis and function as endocrine organs. Obesity is usually correlated with abnormal function of adipocytes, resulting imbalanced levels of biologically active adipocytokines [2]. In the present study, we observed only a positive association between genetically predicted BMI and genetically predicted circulating LEP levels, and failed to detect any associations with other adipocytokines, including ADP, RETN and sOB-R. In addition, our analyses confirmed the associations between genetically predicted BMI and altered levels of key biomarkers related to the INS/IGF axis and inflammatory signaling pathway, including reduced IGF-1 and elevated INS and CRP levels, which aligns with prior MR studies [39–43].
Consistent with previous MR studies [5–9], we replicated the inverse association between genetically predicted BMI and breast cancer susceptibility. However, our further mediation analyses provided evidence supporting a mediating role of CRP in this relationship. The single-mediator model indicated that genetically predicted BMI was positively associated with higher genetically predicted CRP levels, which subsequently increased breast cancer susceptibility. To the best of our knowledge, this mediating role of CRP in the relationship between genetically predicted BMI and breast cancer susceptibility has not been reported previously using a Mendelian Randomization framework. Although several observational studies have investigated this mediation pathway, none have observed statistically significant indirect effects [11, 12, 15]. Jung et al. conducted a one-sample MR study among 10,179 women, revealing an association between elevated CRP polygenic risk scores and increased hormone-positive breast cancer risk; however, they were unable to identify the indirect effect of CRP due to insufficient statistical power [44]. To date, only two MR studies have quantitatively estimated the indirect effect of CRP in the relationship of genetically predicted BMI with endometrial or pancreatic cancer [42, 45]. Compared to these studies, our study utilized 266 genetic variants associated with circulating CRP levels as IVs, the number of which expanded fourfold, and the phenotypic variation increased twofold (from 7% to 16.3%) [30, 42, 45]. Notably, the observed positive indirect effect of BMI on breast cancer risk via CRP may be counterbalanced by stronger negative mediating pathways, such as the negative indirect effect of IGF-1. This suggests that the overall inverse association may reflect the net effect of multiple mediators through distinct biological mechanisms. Nevertheless, these findings, in conjunction with prior observational evidence, underscore the pivotal role of chronic low-grade inflammation in breast cancer pathogenesis [46]. A plausible underlying mechanism involves oversecretion of proinflammatory mediators from dysfunctional adipocytes, leading to increased hepatic CPR production. This process may induce endocrine abnormalities and chronic low-grade systemic inflammation, potentially promoting DNA damage, tumor angiogenesis, and breast carcinogenesis [47–49].
Comprehensive evaluation of multiple mediating pathways involved in breast cancer etiology has long been a significant challenge in large prospective cohort studies [50], and our MR mediation analysis provides new insights. The multiple-mediator model, accounting for the indirect effects of LEP and the INS/IGF pathway, revealed an additional indirect pathway sequentially mediated by IGF-1 and CRP (genetically predicted BMI → IGF-1 → CRP → breast cancer susceptibility). These findings suggest that genetically predicted higher BMI contributes to elevated circulating CRP levels—potentially via a pathway involving reduced IGF-1 levels—thereby increasing breast cancer susceptibility. Corroborating these results, experimental studies have demonstrated the anti-inflammatory effects of IGF-1, showing that elevated circulating IGF-1 attenuates vascular inflammatory responses and oxidative stress [51]. Furthermore, IGF-1 has been shown to counteract CRP-induced endothelial cell activation by stimulating the PI3K/Akt signaling pathway while inhibiting the JNK/c-Jun and MAPK p38/ATF2 signaling pathways [52].
In the present study, we were unable to confirm conclusive evidence regarding the individual mediating roles of IGF-1, INS, or LEP. Using the coefficient product method, we found that genetically predicted IGF-1 levels mediated approximately 5% of the association between genetically predicted BMI and breast cancer susceptibility, albeit without statistical significance. Moreover, we also observed an inverse association between genetically predicted BMI and IGF-1 levels, this may partly explain the negative association between genetically predicted BMI and breast cancer risk. Loh et al. reported a similar mediation effect of IGF-1 [43], however, their use of the coefficient difference method may have introduced bias owing to the dichotomous nature of the outcome and the high prevalence of cases (53.71%) in the BCAC samples [53, 54]. We did not observe evidence supporting the indirect effect of INS or LEP in genetically predicted BMI-breast cancer relationship, despite this mediating role being reported in conventional observational studies [12, 13, 55], warranting further investigation in future studies with larger sample sizes [12, 55].
Our MR mediation study provides novel insights into the potential molecular pathways that link genetically predicted BMI to breast cancer susceptibility. However, this study has several limitations that warrant consideration. First, while we comprehensively examined potential mediators previously associated with obesity-related breast cancer pathways, our hypothesized models may not encompass all possible mechanisms because of the null associations observed for some biomarkers with genetically predicted BMI. Future GWASs with larger sample sizes for these biomarkers are required to complement our findings. Second, the use of a relaxed significance threshold (P < 5⋅10− 6) for IV selection of several biomarkers may have introduced false-positive variants and potential bias. However, the F-statistics of all IVs were > 10, suggesting less possibility of weak instrument bias [56]. Moreover, secondary analyses using a more stringent genome-wide significance threshold (P < 5⋅10− 8) for all IVs yielded robust results. Third, the genetic effects of some biomarkers were derived from sex-combined populations, potentially introducing bias in causal estimates for two-sample MR studies [57]. Female-specific biomarker data could provide a more comprehensive view of the mediation pathways involved in the obesity-related breast cancer relationship among women. Finally, the possibility of horizontal pleiotropy cannot be completely ruled out, despite the various ‘pleiotropy-robust’ methods we applied to minimize such bias, including MR-PRESSO, MR-Egger and Multivariable MR. This is because the full biological functions of the included SNPs are not yet completely understood, and there may remain other unmeasured pleiotropic pathways distinct from the specific biomarkers adjusted for in our models.

Conclusions
Our study integrated multivariable MR and mediation analyses to identify chronic low-grade inflammation (elevated CRP) as a noteworthy pathway in the association between genetically predicted BMI and breast cancer risk. Genetically predicted BMI is associated with higher genetically predicted CRP levels, potentially through a pathway involving reduced IGF-1 levels, which may attenuate the inverse association between genetically predicted BMI and breast cancer risk. These findings shed light on how genetic determinants of BMI influence the risk of breast cancer and provide novel insights into potential targets for cancer prevention.

Supplementary Information