INTRODUCTION
Numerous studies indicate that tobacco is a significant contributor to the worldwide disease burden [1-3]. There is some indication that smoking can impact the structure and function of the brain [4-6]. Previous investigations have discovered a correlation between cigarette smoking and cortical thickness and surface area [7-11]. Although the causal relationship between the two remains unclear, it naturally leads us to associate smoking with changes in brain structure and function. While it is widely accepted that smoking adversely affects health and various bodily functions, the current available studies are observational in nature with small sample sizes, and observational studies cannot establish a causal relationship between smoking behavior and changes in brain structure [4-11]. Additionally, several studies have shown that linear regression analysis, adjusted for factors such as vascular, respiratory, substance use, and psychological characteristics, can significantly weaken the association between smoking and cortical thickness and surface area [9,12,13]. Other studies have indicated that brain structure and function are influenced by multiple factors, including genetic and environmental factors [14,15]. To clarify the causal relationship between smoking behavior and brain structure, we conducted a study analysis using Mendelian randomization (MR).
MR analysis approach is effective in mitigating the impact of confounding variables and evaluating reverse causality. This approach utilizes genetic variation as a tool variable to assess the causal relationship between exposure and outcome. Because genetic variation is randomly distributed during pregnancy, it is not influenced by diseases or other acquired factors [16].
The current study has utilized genome-wide association study (GWAS) to identify genetic variations associated with smoking behavior, cortical thickness, and surface area [17,18]. Building upon these GWAS findings, linkage disequilibrium score regression (LDSC) [19] was employed to investigate the genetic association between cortical thickness, surface area, and smoking behavior. Subsequently, MR analysis was conducted to explore the causal relationship between these variables.
METHODS
Data sources
The GSCAN website [17] (https://genome.psych.umn.edu/index.php/GSCAN) provided GWAS data on smoking behavior, which included information on 1) age of initiation of regular smoking (AgeSmk): “The age when an individual first becomes an official smoker and excluded outliers (e.g., those who said they smoked regularly at age 4.)”; 2) smoking initiation (SmkInit): “Whether the subject being a regular smoker in their life”; 3) smoking cessation (SmkCes): “A binary variable contrasting current versus former smokers”; 4) cigarettes per day (CigDay): “Excluding never-smokers and those who had smoked but for which we did not collect data, the samples that met the inclusion criteria were divided into different groups based on the number of cigarettes smoked per day,” comprehensive data on smoking phenotypes are listed in the Supplementary Material for further reference. PLINK (a tool set for whole-genome association and population-based linkage analyses) [20] was used to conduct GWAS on the aforementioned phenotypes, and linear regression was utilized for continuously variable phenotypes. Age, sex, genotype array, and genetic principal components were controlled as covariates during the analysis. The GWAS results that were ultimately included in the analysis did not incorporate UK Biobank data [18]. This study has been approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University (KY-2022-22).
The study utilized summary statistics of GWAS on cortical thickness and surface area obtained from source Oxford Brain Imaging Genetics (BIG) Server-version 2.0 [21] and brain imaging data along with genome-wide genetic data from source UK Biobank. The Desikan-Killiany-Tourville Atlas (denoted DKT39), which comprises of 62 brain regions in the left and right hemispheres, was used to model in FreeSurfer to obtain cortical thickness and surface area data. A total of 124 (62×2) traits originating from cortical thickness and surface area of each brain region were used as phenotypes along with whole-genome genetic data to perform the GWAS analysis. Covariates and other methods were implemented in the analysis to reduce the influence of potential confounding factors [22].
Genetic correlation analysis
LDSC analysis was performed to investigate the genetic correlation between the traits mentioned earlier [19]. Prior to analysis, we conducted quality control by excluding the following variants: variants with incomplete association statistics, non-biallelic variants with ambiguous alleles that did not match those in the 1,000 Genomes reference panel from European ancestry, variants with an info score <0.9, and variants with a minor allele frequency <0.01.
MR analysis
MR was used to assess the causal relationship between phenotypes with significant genetic correlation. Instrumental genetic variables single nucleotide polymorphisms (SNPs) were selected according to the clumping procedure in PLINK [20] based on an r-squared threshold of 0.05 within a 500-kb window at p<1×10-5 for four smoking behaviors, cortical thickness and surface area. The results of each screening should have at least 10 SNP before further analysis and these SNPs were screened based on 1,000 genomes of European ancestry.
In this study, we used MR to investigate the causal relationship between smoking behavior (smoking initiation, age at smoking initiation, cigarettes per day, and smoking cessation) and specific brain regions’ cortical surface area and thickness. The following methods are used: 1) the inverse-variance weighted approach [21], which obtains the causal estimate based on regression of the associations with the outcome on the associations with the risk factor, with the intercept set to zero and the weights being the inverse variances of the associations with the outcome; 2) the MR-Egger regression method [22], which allows the intercept to estimate whether genetic variants have an average pleiotropic bias; and 3) the median-based method [23], which selects the median MR estimate and the causal estimate. All methods run in R version 4.1.1 (R Foundation for Statistical Computing) or above.
Sensitivity analysis
To evaluate the reliability of our findings, we further examined horizontal pleiotropy and employed various analytical methods to test for heterogeneity of the results, including the modified Cochran Q statistic, the leave-one SNP-out analyses [24], the MR-Egger intercept test of deviation from the null [25], and MR-Pleiotropy Residual Sum and Outlier [26]. To further address potential pleiotropic effects in the MR analysis, we evaluated each instrument SNP using phenotype databases (PhenoScanner GWAS database [27]) to determine if it was associated with any potential confounding phenotypes.
RESULTS
Genetic association of four smoking behaviors with cortical thickness and surface area
We discovered significant genetic correlations between AgeSmk and specific brain regions: a negative correlation with the right precentral surface area (rg=-0.25, p=8.3×10-3), left caudal anterior cingulate surface area (rg=-0.22, p=0.025), left cuneus surface area (rg=-0.26, p=0.013), left inferior parietal surface area (rg=0.25, p=0.035), and right caudal anterior cingulate thickness (rg=0.29, p=0.028). Additionally, a correlation was observed between CigDay and the left caudal middle frontal surface area (rg=0.21, p=0.04), and between SmkCes and the left entorhinal thickness (rg=-0.36, p=0.019). Furthermore, associations were found between SmkInit and the left rostral anterior cingulate surface area (rg=-0.20, p=0.015), right rostral anterior cingulate thickness (rg=-0.24, p=0.011), and right superior frontal thickness (rg=-0.17, p=0.045) (Figure 1). No other genetic association was found between brain regions and four smoking behaviors.
MR analysis
Based on LDSC analysis, we obtained the genetic correlation results, and using this result, we conducted MR analysis, which showed a unidirectional causal relationship between only the left caudal middle frontal surface area and CigDay (βIVW=0.056, p=1×10-5, pBonferroni=2×10-4) (Figure 2A); in addition, we also used the weighted median and simple median analysis method to validate this result (βweightedmedian=0.063, p=0.007; βsimple median=0.062, p=0.006) (Table 1 and Figure 2A).
Furthermore, each SNP was removed and analyzed in turn, and no single SNP was found to have an impact on the results (Supplementary Figure 1). The modified Cochran Q analysis found no difference (Q=10.775, p=0.630) (Supplementary Table 1). MR-Egger analysis further showed that there was no pleiotropic bias (intercept<0.001, std error=0.003, p=0.990) (Supplementary Table 1). The 15 SNPs of left caudal middle frontal surface area were not significantly associated with other smoking behaviors (Supplementary Table 2).
However, when we used CigDay as the exposure and the left caudal middle frontal surface area as the outcome, we did not observe a significant causal relationship (βIVW=-0.002, p=0.972; βweighted median=-0.035, p=0.755; βsimple median=0.079, p=0.403) (Figure 2B). And no causal relationship was found between the other brain regions and the four smoking behaviors (Table 1, Supplementary Table 1, and Supplementary Figures 2-19).
DISCUSSION
In the study, we found a genetic association between smoking and brain structure, and a unidirectional causal relationship between left caudal middle frontal surface area and CigDay.
The LDSC analysis suggests that there is a certain degree of genetic association between all four smoking behaviors (AgeSmk, CigDay, SmkInit, and SmkCes) and either cortical thickness or surface area. This is consistent with previous research indicating a statistical association between smoking and brain structure [4,5,12]. This naturally leads us to associate smoking with changes in brain structure, as in other studies about the harmful effects of smoking. However, it is possible that differences in brain structure lead to differences in smoking behavior, as the brain is the root cause of all our behavior [4,5,12], and the observational studies cannot rule out the influence of confounding factors such as environment and reverse causality, so we explored this possibility.
We utilized a more precise MR analysis method to eliminate the effects of complex interactions and discovered a unidirectional causal relationship between left caudal middle frontal surface area and CigDay, and propose that the region in the brain responsible for decision-making and emotional regulation may affect an individual’s susceptibility to tobacco addiction [28-32].
We only found evidence for a unidirectional causal relationship between left caudal middle frontal surface area and CigDay. We did not find a clear unidirectional causal relationship between smoking and cortical surface area, which seems inconsistent with previous research. However, this does not necessarily mean that either study was incorrect. Inconsistent results between the two analysis methods do not imply that one method is wrong. Instead, this inconsistency may reflect the complex interplay between genetic and environmental factors in human disease and traits. Therefore, we speculate that the relationship between smoking and cortical thickness and surface area is not a direct correlation, but rather involves some mediating factors or interference. For example, several studies have found that linear regression analysis, adjusted for factors such as vascular, respiratory, substance use, and psychological characteristics, can greatly weaken the association between smoking and cortical thickness and surface area [9,12,13]. Other studies have also shown that brain structure and function are influenced by multiple factors, including genetic and environmental factors [14,15]. Furthermore, it is important to note that brain damage is not only related to cortical thickness and surface area, but also to functional changes in neurotransmitter synaptic connections [33-35]. Those studies have found that smoking can affect the neurotransmitter system and brain function [12,36,37], and even be related to mental disorders [38-41]. Smoking does not directly cause changes in brain structure, but affects multiple physiological processes such as blood circulation and release of neurotransmitters, which may further lead to changes in cortical surface area and volume [9,12,13,42,43]. These effects may result in the existence of genetic correlations, but the absence of a bidirectional causal relationship.
In addition, it is possible that the surface area of the left caudal middle frontal surface area is associated with tobacco addiction, leading to changes in smoking behavior. Rodent studies have shown that the inter-modular connectivity index and participation coefficient of the insula-frontal cortex module can predict the severity of future nicotine dependence [28]. The frontal cortex is also associated with smoking behavior, addiction, and smoking cessation [29-31]. Stimulation of the frontal cortex can reduce the desire to smoke [32]. From a developmental perspective, a larger left caudal middle frontal surface area suggests more synaptic connections, more neurons, and richer neural functions [44,45], which may explain the increased CigDay resulting from a larger brain surface area. Furthermore, existing evidence suggests that inhibitory control is related to nicotine addiction [46,47], and the frontal cortex is a key area involved in inhibitory control [48,49], indicating that the frontal cortex plays an important role in nicotine addiction. Similarly, several studies using Braincloud mRNA phenotype data have shown that multiple SNPs in the frontal cortex are specifically and stably associated with nicotine withdrawal [50], further emphasizing the important role of the frontal cortex in nicotine addiction and supporting our findings.
Currently, there are many observational studies on the relationship between smoking and brain structure. Our innovative approach utilizes LDSC and MR analysis to investigate the genetic basis of this relationship. We aim to understand the association between cortical thickness and surface area and smoking behavior from both genetic correlation and causal association perspectives. This provides a genetic understanding of the individual differences in smoking behavior and the brain changes caused by smoking, while accounting for confounding factors. This method represents a novel approach to exploring the relationship between smoking and brain structure and is worth further exploration in research related to human diseases and tobacco addiction.
Indeed, no study is ever absolutely perfect: 1) although SNPs of left caudal middle frontal surface area were not significantly associated with smoking-related phenotypes, however, the biological functions of the SNPs we analyzed are not completely clear at present, we cannot rule out pleiotropic bias; 2) data of this study was based on European population only, so generalizing the results to the whole population may require broader data; 3) the genetic data obtained in this study on different smoking phenotypes and cortical thickness and surface area of brain regions were not stratified by age, so it is not known whether there are differences across age groups; 4) although we explored the causal relationship between smoking and cortical thickness and surface area at the genetic level, which effectively controlled for confounding factors that might exist in the process of gene expression, sample overlap of the two MR samples may still bias the results towards observational estimates [51].
In summary, this study represents the first use of large-scale GWAS data in exploring the genetic association between smoking behavior and brain structure using LDSC and MR analysis. We have also validated the use of genetic tools for investigating the relationship between smoking behavior and brain structure. These findings provide a foundation for future research on the physiological mechanisms underlying smoking behavior.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
