Body mass index and the risk of basal cell carcinoma: evidence from Mendelian randomization analysis

Introduction

Skin cancer is one of the most common malignancies in the world and is an important public health concern. Malignant nonmelanoma skin cancers originate from keratinized epithelial cells, and these cancers include basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). Melanoma only accounts for about 2% of malignant skin cancer (Agbai et al., 2014). However, BCC, formerly referred to as basal cell epithelioma, is the most common form of skin cancer and most prevalent cancer in humans, particularly in elderly populations (Garcovich et al., 2017; Marzuka & Book, 2015). Age and sun exposure are the most significant and well-established risk factors for BCC in paler/white skin (van der Poort et al., 2020). Additional risk factors include ionizing radiation, chronic inflammatory skin diseases, and exposure to arsenic (Pellegrini et al., 2017; Surdu et al., 2013).

Obesity has been connected with a variety of cancer types in humans. In the context of skin cancer, obesity has been found to be positively related to melanoma in males but not in women (Dobbins, Decorby & Choi, 2013; Sergentanis et al., 2013). Similarly, epidemiologic evidence addressing the association between body mass index (BMI) and BCC was contradictory, with several studies suggesting an inverse relationship (Zhang et al., 2017; Zhou, Wu & Luo, 2016), and other studies demonstrated that there was no association between BMI and BCC risk (Milan et al., 2003; Olsen et al., 2006). The causes of the variability between these studies are unknown, and it remains to be determined whether there was a dose–response relationship or whether such an inverse relationship was simply explained by confounding factors of sun exposure and sensitivity assessments.

A relationship between being overweight and skin cancer may have both biological and behavioral origins. Biologically, a higher BMI is connected with increased circulating estrogen levels (De Pergola & Silvestris, 2013), which in mice models have been linked to a decreased incidence of BCC (Mancuso et al., 2009). From a behavioral perspective, ultraviolet radiation (UVR) exposure is a major risk factor for skin cancer, and differences in outdoor activity levels due to BMI (Tang et al., 2013; Wareham, Van Sluijs & Ekelund, 2005) may contribute to differences in UVR exposure. As a result, exposures to UVR and/or estrogen may serve as potential mediators of the connection between BMI and skin cancer. However, this conclusion may not always be true. Previous observational study conducted by Zhang et al. (2017) reported that BMI was negatively related to early-onset BCC (OR = 0.43, 95% CI [0.26–0.71]), and neither UVR nor estrogen-related exposures in women could account for this connection. Although useful, these observational studies are prone to confounding influences, resulting in incorrect causal findings. As a result, randomized controlled trials are required to determine whether the associations discovered in observational research are valid.

Randomized studies on skin cancer are complex because large sample sizes and prolonged follow-up are required. Therefore, drawing the causal connection between BMI and BCC becomes challenging. In order to determine whether or not the stated association between BMI and BCC is a causal relationship, an effective method need to be utilized. Mendelian randomization (MR) research has the potential to address this question in particular (Do et al., 2013; Emdin, Khera & Kathiresan, 2017; Frikke-Schmidt et al., 2008; Voight et al., 2012). Randomizing alleles during gamete production equalizes confounding factors, similar to a randomized trial. Furthermore, genetic variation can affect outcomes, but outcomes cannot affect genes, so no inference of reverse causality can be drawn. MR assumes the genetic variant is related with the exposure, not confounders, and impacts the result only through the exposure (Emdin, Khera & Kathiresan, 2017). Causality can be established through a genetic instrument that fits all MR assumptions. Previous research using MR approaches successfully predicted the causative effect of BMI on many cancers, including lung cancer (Zhou et al., 2021), colorectal cancer (Suzuki et al., 2021), and melanoma (Dusingize et al., 2020).

However, to our knowledge, there are currently no reports on the relationship between BMI and BCC using MR methods. Thus, we conducted a two-sample MR analysis and meta-analysis to investigate the causal relationship between BMI and BCC, utilizing the summary data from the largest BMI and BCC GWASs to date.

Methods

Assumptions of mendelian randomization study

In the MR research, three fundamental assumptions must be satisfied: (1) the genetic instrument variables (GIVs) must be significantly related to exposure; (2) the GIVs must not be linked to any possible exposure vs. outcome relationship confounders; (3) the GIVs should only impact outcome risk through exposure (Davey Smith & Hemani, 2014; Smith & Ebrahim, 2003). The assumptions and design of the MR study are shown in Fig. 1.

Results

Bidirectional univariable MR analysis

To increase the reliability of the results, we selected four different BMI-related GWAS datasets. All 468, 282, 394, and 405 independent SNPs associated with four different BMI datasets were available in the summary statistics for BCC, respectively. The F statistic of those SNPs was greater than 10 in four BMI datasets, indicating a low risk of weak-instrument bias (Supplementary File 1).

This work showed that genetically predicted BMI was inversely associated with BCC; the odds ratios were 0.884 (95% CI [0.815–0.959], P = 0.003), 0.904 (95% CI [0.831–0.983], P = 0.018), 0.898 (95% CI [0.832–0.969], P = 0.006), and 0.899 (95% CI [0.833–0.971], P = 0.007) in the IVW analysis, respectively (Table S1 and Figs. 2A–2D). Since the results of subsequent heterogeneity analysis showed substantial heterogeneity among GIVs (P-het < 0.05, Table S1), we used the random-effects model to estimate the above MR effect size. In addition, genetically predicted BMI was consistently associated with BCC across the different MR methods (Table S1) (Figs. 3A–3D, Fig. S1). The intercept term estimated from MR-Egger was centred at the origin (P-intercept > 0.05), suggesting that directional pleiotropy did not influence the results. Meanwhile, MR-PRESSO study found no outlier SNPs that increased MR pleiotropy. In addition, leave-one-out analysis showed no single SNP changed the total estimate (Fig. S2). Moreover, the SNPs explained 4.71%, 4.95%, 5.70%, and 5.76% of the variance of BMI traits, respectively. The causal assumption of BMI and BCC was verified via the MR Steiger test, and the result showed BMI’s influence on BCC was the correct causal direction (P = 0.000).

Moreover, when exposure and outcome were switched to see if reverse causation was possible, there was no indication of reverse causality from BCC to BMI (all P > 0.05, Fig. S3).

Meta-analysis

The results of meta-analysis were shown in Fig. 4. The results from the meta-analysis showed genetically predicted 1-SD (kg/m²) increase in BMI was significantly correlated with an average 10.4% decrease in the overall BCC risk; the OR was 0.896 (95% CI [0.861–0.933], P = 6.70E−8). Heterogeneity between the four datasets was evaluated using chi-squared-based Q and I² tests. And the results showed there were not any heterogeneity among the four datasets (Q = 0.15, I² = 0.00%, P = 0.98).

Discussion

We employed MR to explore genetically predicted BMI and BCC risk and replicated findings in four independent UK Biobank populations. Using summary data from the largest BMI and BCC GWASs to date, causation was investigated using a bidirectional univariable two-sample MR, with the results evaluated further using a variety of sensitivity measures. Bidirectional univariable MR demonstrated a negative causal effect of genetically predicted BMI on BCC risk and no clear evidence for reverse causation. Moreover, further meta-analysis results also indicated a significant inverse causal effect of the average genetically predicted effect of BMI on BCC risk.

Previous studies utilizing the MR method have indicated that genetically predicted BMI was causally associated with several cancers (Liu et al., 2021; Suzuki et al., 2021; Zhou et al., 2021). To our knowledge, however, no studies have demonstrated a causal connection between BMI and BCC using MR methods. It has been shown in several earlier observational studies that BMI was linked to a lower incidence of BCC. For example, Zhang et al. (2017) reported that BMI was negatively related to early-onset BCC (OR = 0.43, 95% CI [0.26–0.71]). A prospective cohort research with 58,213 Caucasian participants revealed that the incidence of BCC decreased with rising BMI in both sexes, even after controlling for UVR exposures (Gerstenblith et al., 2012). In addition, Chan et al. (2019) found that a BMI ≥ 25 kg/m² was associated with lower nonmelanoma skin cancer (NMSC) hazard rates. A prospective study on Caucasians in the US found that participants with a BMI in the obese range had a 19 percent reduced chance of getting BCC than those with a BMI in the normal range, and those with a BMI in the severely obese group had a 29 percent lower BCC risk (Pothiawala et al., 2012). These results indicated that BMI might be a protective factor for BCC. Even though they are helpful, these observational studies are susceptible to confounding factors (such as sun exposure and outdoor activities), leading to inaccurate causal inferences. For instance, after adjusting for potential confounding variables like sun exposure, Olsen et al. (2006) discovered no significant correlation between any of the anthropometric parameters (such BMI) and the incidence of BCC. Moreover, a previous study involving twin pairs indicated no clear effect of BMI on decreasing BCC risk (Milan et al., 2003). MR methods can partially address the limitations of observational studies. Our results are consistent with those of most previous observational studies, and we found that BMI has a negative effect on the risk of BCC using MR analysis. At the same time, in order to ensure the reliability of the results, we selected four independent BMI-related datasets to study the causal relationship between BMI and BCC. Importantly, the analysis results presented by these four datasets maintain a high degree of consistency. Additionally, to enhance the accuracy of the genetically predicted effect of BMI on BCC, we also performed a meta-analysis; the analysis results still showed a strong negative causal relationship between BMI and BCC.

Previous researches have not studied the mechanism between obesity (BMI over 30 kg/m²) and BCC; we can only draw certain assumptions. Biologically, elevated estrogen levels due to obesity (Crosbie et al., 2010) may be associated with reduced BCC risk in obese people, which has been associated with a lower risk of BCC in mouse models (Mancuso et al., 2009). In addition, Cho et al. (2008) found that estrogen might protect mice’s epidermal cytokine production and immunological function from UVR-induced (the key skin cancer risk factor) damage. These results suggest that estrogen may indeed have some protective effect against skin diseases. In addition to biological factors, the behavior of obese people also shows certain characteristics (e.g., spending less time outdoors and avoiding sunbathing) (Gerstenblith et al., 2012). The observed inverse relationships between obesity and BCC risk may therefore be impacted in part by unmeasured confounding by UVR exposure.

Despite the fact that two-sample MR is an effective method for drawing causal inferences between exposures and outcomes using summary statistics, we should be prudent with our conclusions due to several limitations. First, although the leave-one-out test is a statistical method for assessing the reliance of MR results, it is more suitable for testing a good deal of SNPs rather than several SNPs. In this study, we identified a large number of SNPs from GWAS data with large samples and used these SNPs for MR analysis, thus avoiding this limitation. Second, our study was conducted utilizing populations from Europe, which limits its capacity to be generalized to a wider population. Third, it is also possible that additional factors, such as sun exposure and outdoor activities confound our results. However, it is difficult to account for the influence of other factors due to the limited availability of genetic data underpinning these traits. Finally, although our study shows that BMI negatively correlates with BCC, we still need to be cautious. Do not increase BMI excessively because the excessive increase in BMI may increase the risk of other diseases (Avgerinos et al., 2019). Therefore, in the future, we need to clarify other mediators of the causal relationship between BMI and BCC. Perhaps we can prevent BCC (the most common skin cancer in humans) by intervening with mediators to avoid side effects during the treatment of BCC by inducing an excessive increase in BMI. Of course, this requires the continuous efforts of subsequent researchers.

The study’s magnitude and the fact that BMI was measured rather than self-reported are two of its main strengths. In addition, to our knowledge, no MR has been conducted to evaluate the relationship between BMI and BCC. MR studies provided advantages over traditional observational investigations, such as reducing residual confounding risk. Additionally, we incorporated four independent BMI-related datasets to ensure the reliability of causal analysis. Critical sensitivity assessments were also undertaken to validate the assumptions of MR analyses. As a result, we were able to offer new perspectives that could aid in understanding the role of BMI in BCC incidence. In summary, using univariable MR analyses and meta-analysis, we found evidence to support BMI as a negative causal factor for BCC risk. More research is needed to determine how this possible cause-and-effect link works.

Supplemental Information