Genetically predicted childhood body mass index and lung cancer susceptibility: A two‐sample Mendelian randomization study

Abstract Background The association between adult body mass index (BMI) and lung cancer (LC) susceptibility have been reported, but the causal relationship with childhood BMI remains largely unclear. To evaluate the causal effect of childhood BMI on LC susceptibility, a two‐sample Mendelian randomization (MR) study was performed. Methods The two‐sample MR analysis utilized 25 single nucleotide polymorphisms (SNPs) as instrumental variables for childhood BMI. Genetic summary data from the International Lung Cancer Consortium and FinnGen databases were analyzed to estimate the causal effect of these SNPs on LC susceptibility. The IVW method was employed as the primary analysis, supplemented by the Weighted Median, MR‐Egger, and MR pleiotropy residual sum and outlier test. Results Our findings indicated that there was no causal association between childhood BMI and the susceptibility of LC (odds ratio [OR]: 1.03, 95% confidence interval [CI]: 0.90–1.17, p = 0.705), lung adenocarcinoma (OR: 0.99, 95% CI: 0.86–1.13, p = 0.832), lung squamous cell carcinoma (OR: 0.97, 95% CI: 0.84–1.13, p = 0.726), and small cell LC (OR: 1.09, 95% CI: 0.82–1.45, p = 0.554) based on the IVW as well as other methods employed. Furthermore, these findings indicated no causal effect of childhood BMI on the LC susceptibility in both ever smokers and never smokers. Conclusion This study did not conclude a causal effect between childhood BMI and LC susceptibility. However, given the complex nature of cancer development, further studies are needed to verify these findings.


| INTRODUCTION
Lung cancer (LC) remains the leading cause of cancerrelated deaths worldwide. 1While various risk factors, such as age, family history, and air pollution, have been studie, [2][3][4] they are not easily modified or controllable as effective surveillance strategies.Consequently, there is a critical need to identify other modifiable risk factors to enhance individualized prevention strategies and alleviate the global burden of LC.
6][7] The impact of adult BMI on LC susceptibility has been widely studied, with several studies indicating that higher adult BMI is associate with a reduced risk of LC. [8][9][10][11] However, the causal relationship between childhood BMI and LC susceptibility requires further exploration, as conducting large-scale and long-term randomized controlled trials poses significant challenges.Additionally, the presence of reverse causal effects and potential confounding factors, such as smoking and drinking, 12,13 can make it difficult to draw conclusive results from traditional retrospective studies.
To overcome these issues, we employed a novel epidemiology method named Mendelian randomization (MR) that reduces bias while saving time and cost. 14In the MR study, genetic variations such as single nucleotide polymorphisms (SNPs) are used as instrumental variables (IVs) to alternate the studied exposure.Due to their random assignment and occurrence prior to disease onset, genetic variants are largely independent of acquired or environmental factors. 15This allows MR to assess the causal effect of childhood BMI on LC susceptibility while avoiding biases caused by reverse causal effects or confounding factors.So far, one MR analysis related to this topic was reported. 16However, this study only examined potential causal relationships using a relatively small sample size, and it did not fully explore other LC subgroups, such as detailed histologic subtypes and LC with different smoking statuses, 16 which warranted further exploration.
In this context, this MR study combined large-size GWAS data to address these gaps by thoroughly investigating the potential causal associations between childhood BMI and LC susceptibility, specifically focusing on various LC subgroups.

| Study design
We conducted the two-sample the two-sample MR analysis using GWAS summary level data obtained from recently published GWAS meta-analyses and public database.The GWAS summary level data for the exposure and outcome were obtained from separate samples to ensure that research participants did not overlap.MR study adheres to three principal assumptions as follows: (i) the genetic instruments are associated with childhood BMI; (ii) the genetic instruments can only affect LC through childhood BMI without any other causal pathway; and (iii) the genetic instruments are not affected by any potential confounder, 17   approval was required for the study since it involved a reanalysis of the published data.

| Genetic variants for childhood BMI
We retrieved the GWAS summary level data for childhood BMI from the recently published GWAS study by the Early Growth Genetics consortium (EGG), which combined 41 studies enrolling 61,111 children aged 2-10 years in European ancestry. 18All parameters, including height and weight, were measured using standard international unit and BMI was calculated by dividing weight by the square of height (kg/m 2 ).Genetic instruments with significant genome-wide significance of p < 5 × 10 −8 were included, which were then clumped using the window of 500 kb and linkage disequilibrium of r 2 less than 0.2 to ensure that the variants were independent. 18SNPs that were not found in the outcome GWAS summary data or had palindrome structures were removed for a low loss proportion during matching and harmonizing.Finally, the instrumental variables consisted of 25 single-nucleotide polymorphisms (SNPs), which was demonstrated significantly associated with childhood BMI.Weak instruments resulted in inadequate ability to predict causal effect.Therefore, we calculated explained genetic variation (R 2 ) and F statistic to evaluate these instruments.The former was generated using the formula: R 2 = 2 × β 2 × EAF × (1 − EAF), with β being estimated genetic effect of each SNP and EAF being the effect allele frequency, while the latter was calculated using formula: with N being sample size and K being number of SNPs. 19The F statistic values, more than a conventional threshold of 10, indicating sufficient strength to predict childhood BMI. 20

| Genetic variants for LC
We obtained GWAS summary level data for LC from two databases, namely the International Lung Cancer Consortium (ILCCO), 21 with data updated in 2021, and FinnGen (www.finbb.fi),utilizing the latest version 9 for this study.The LC phenotype was defined as a binary trait, encompassing 29,266 LC cases and 56,450 controls from ILCCO, along with 5842 LC cases and 281,295 controls from FinnGen (Table S1).

| Genetic variants for risk factors of LC
To explore whether childhood BMI potentially affected LC susceptibility through other risk factors, we performed inverse-variance weighted (IVW) analysis to assess the association between childhood BMI and these factors.Based on existing literature and available GWAS data, 12,22 we included four potential risk factors in our study, including smoking status, drinking status, education level, and total cholesterol.GWAS summary level data for smoking status (age of smoking initiation, ever vs. current smoker, ever vs. never smoker, and cigarettes smoked per day) and alcohol consumption (drinks per week) were obtained from the GWAS & Sequencing Consortium of Alcohol and Nicotine use (GSCAN). 23The GWAS data for education level were obtained from the Social Science Genetic Association Consortium (SSGAC). 24We evaluated the relationship of childhood BMI on total cholesterol using data from the Global Lipids Genetics Consortium (GLGC). 25The information of GWAS summary data is presented in Table S1.

| Mendelian randomization
The MR analyses were conducted independently within each outcome database, followed by a fixed-effect metaanalysis to combine the casual estimates.Several MR methods were employed to explore the potential causal impact of childhood BMI on LC susceptibility.First, we applied the IVW to estimate the causal impact of childhood BMI on LC susceptibility, which weights the regression model using the inverse of the outcome variance (the square of the standard error) and ignores the intercept of the regression model. 26Notably, when using IVW method, the existence of the horizontal pleiotropy may result in biased result.Therefore, we performed MR-Egger regression and weighted-median method to complement the IVW method.MR-Egger regression is similar to IVW method, otherwise, which allows the presence of horizontal pleiotropy. 27When dealing with invalid instruments, the weighted-median method is more robust MR approach than IVW and MR-Egger regression. 28Furthermore, we tested the horizontal pleiotropy using MR pleiotropy residual sum and outlier (MR-PRESSO) method and re-calculate the causal effect after removing outliers. 29A two-tailed p value <0.05 was considered significant.We conducted all MR analyses in R software (version 4.2.1)utilizing the packages "MRPRESSO" (version 1.0), "TwoSampleMR" (version 0.5.6), and "meta" (version 3.1.1).

| Sensitivity analysis
Several sensitivity analyses, including the pleiotropy test, heterogeneity test, and leave-one-out sensitivity test, were conducted to examine the robustness of the results.The horizontal pleiotropic effects were examined using intercept test of MR-Egger.Heterogeneity was measured by Cochran's Q statistic and I 2 .Furthermore, leave-one-out analysis was performed to evaluate the potential impact of individual SNPs on the IVW estimate by removing one SNP every time.The power analysis for our study was conducted employing the mRnd power calculator (https:// shiny.cnsgenomics.com/mRnd/). 30

| RESULTS
The genetic instruments explained 2.95% for childhood BMI and F statistic values ranged from 36 to 153, indicating strong instruments.The detailed information of SNPs is summarized in Table S2.

| Sensitivity and pleiotropy analysis
Regarding the heterogeneity analysis, we observed significant heterogeneity in the causal effect analysis of childhood BMI on susceptibility of LC, as well as LC in ever smokers (p < 0.001 in both the MR-Egger and IVW) using outcome data from ILCCO.However, no significant heterogeneity F I G U R E 2 Mendelian randomization using IVW method estimated the causal effects between childhood BMI and lung cancer susceptibility.

Lung cancer
Outcome was found in the other LC subgroups (p > 0.05 in both the Egger and IVW) (Table S4).Nevertheless, the causal effect of childhood BMI on susceptibility of LC, as well as LC in ever smokers, was consistent in direction and magnitude across different methods, indicating a robust result.Furthermore, no evidence of horizontal pleiotropy in other groups was found through MR-Egger regression analysis (Table S4).In the leave-one-out analysis, we did not identify any individual SNP that had a strong impact on the causal estimate (Figure S3 and S4), and the funnel plot revealed no evidence of bias in the instrumental variables (Figure S5 and S6).

| Causal relationship of childhood BMI on risk factors
We performed additional MR analyses using the IVW method to estimate the causal relationship between childhood BMI and potential risk factors of LC.The results revealed that each 1 kg/m 2 increase in childhood BMI was associated with a 6% decrease in the likelihood of ever smoking (OR: 0.94, 95% CI: 0.90-0.98,p = 0.006) and a decrease in smoking intensity (OR: 0.94, 95% CI: 0.90-0.98,p = 0.003).Nevertheless, the findings did not demonstrate significant causal effects between childhood BMI and the age of smoking initiation, smoking cessation, drinking per week, years of education, and total cholesterol (Table S5).

| DISCUSSION
In this two-sample MR study, we investigated the causal association between childhood BMI and LC susceptibility.The results did not support a causal effect of genetically predicted childhood BMI on LC susceptibility or any of the examined LC subgroups.These findings remained robust when considering pleiotropy and heterogeneity.Additionally, the consistency of results obtained from the Weighted median, MR-Egger, and IVW analyses further supported the robustness of these findings.
To our knowledge, one MR study investigated the correlation between childhood BMI and LC susceptibility, and our results aligned with its findings. 16Compared with the current study, the study by Gao et al. 16 used GWAS summary level data for LC with relatively small sample size (12,160 cases and 16,838), which may limit the evidence power.In this study, we obtained large sample size GWAS summary data from both currently largest meta-analysis for LC and FinnGen database.Also, a fixed-effect metaanalysis was employed to combine the casual effects from the two outcome databases, which provided more convinced conclusion.Second, the study conducted by Gao et al. 16 did not fully explore certain LC subgroups, such as detailed histologic subtypes and LC with different smoking statuses, which was further explored in this study.Our results provided renewed evidence that childhood BMI did not have a direct causal relationship with either SCLC or non-small cell lung cancer (NSCLC).Although we did not observe a significant causality, our MR analysis revealed that high childhood BMI might tend to be associated with an increased risk of SCLC, but a reduced risk of NSCLC.The difference in outcomes may be attributed to the inherent heterogeneity in the impact of childhood BMI on different LC subtypes or may be influenced by the relatively small sample size of SCLC.
Our findings suggested that no apparent causal relationship existed between childhood BMI and LC susceptibility.However, abnormal childhood BMI may still influence the morbidity of LC beyond the scope of our analysis.This is because adipose, considered as an active endocrine organ, can secrete various adipokines and interleukins, such as adiponectin and TNF-α, which may drive chronic inflammation and the subsequent development of cancer. 31dditionally, excessive adipose tissue can cause immune cell dysfunction and facilitate the formation of tumorpromoting environment. 32Meanwhile, growing clinical research has reported that abnormal BMI during adolescence may contribute to increased risk of several malignancies in adulthood, such as breast cancer and primary hepatic cancer. 16,33And in a 50-year cohort study, it was found that per standard deviation increase in childhood BMI could increase the overall risk of smoking-related cancer by 30%. 34owever, this study did not provide the detailed results in each cancer type, which precluded the establishment of a causal correlation between childhood BMI and smokingrelated LC.In this study, our results suggested that there existed little causal effect between childhood BMI and susceptibility of LC in ever and current smokers.While BMI is a useful tool for evaluating body fat, other indicators, for example waist-to-hip ratio, are also worth exploring.Combining BMI with waist-to-hip ratio or other indicators to assess the impact of childhood BMI on risk of LC may have promising clinical significance.
It was worth noting that prior study have established a correlation between smoking status and BMI, whereby smoking is associated with weight loss, while smoking cessation often leads to weight gain. 35These relationships can be explained through various mechanisms such as changes in neural pathways 36 and gut microbiota. 37nterestingly, in our analysis, we observed a causal relationship between increasing childhood BMI and a decreased likelihood of smoking.This finding underscores the complex interplay between smoking status and BMI.It highlights the need for further research to elucidate the underlying mechanisms driving this association.
Our study presents several strengths and practical implications.First, we adopted MR analysis to stimulate randomized controlled trails, which offers several advantages compared to traditional retrospective studies.For instance, this innovative method conserves physical, and financial resources, while also reducing potential confounding biases and reverse causal effects.Second, given the heavy burdens of LC and childhood obesity globally, revealing potential causality between childhood BMI and LC incidence may influence the future health care policymaking.These results may also provide valuable insights into the ideal timing for disease intervention, thereby contributing to the overall efforts aimed at reducing the burden of these two critical health issues.
However, as with any MR analysis, limitations are inevitable.First, the generalizability of our conclusions to other populations remains uncertain, as the participants included in our study were exclusively of European ancestry.Second, despite our efforts to combine available data, a larger sample size in the GWAS data would have further enhanced the statistical power of our MR analysis.In addition, BMI is intrinsically affected by the interactions of congenital and environmental factors such as dietary habits. 12Therefore, our study is constrained within the topic of genetically predicted childhood BMI.Regarding no positive evidence has been found in this current study, more explorations are needed to investigate the relation between environmentally determined childhood BMI and LC susceptibility. 38

| CONCLUSION
Our study provided the latest evidence indicating the absence of a causal relationship between childhood BMI and LC susceptibility.Therefore, utilizing childhood BMI as a screening tool for assessing LC susceptibility may not yield effective results.Instead, greater emphasis should be placed on uncovering the relationship between childhood BMI influenced by environmental factors such as dietary habits and LC susceptibility.
which are depicted in Figure 1.No ethical F I G U R E 1 Schematic illustration depicted Mendelian randomization assumptions.The assumptions including: (I) genetic instruments are significantly associated childhood BMI; (II) genetic instruments can only influence lung cancer through childhood BMI without any alternative causal pathways; (III) genetic instruments are independent of any cofounder.