Associations between TUBB-WWOX SNPs, their haplotypes, gene-gene, and gene-environment interactions and dyslipidemia

In this study, we investigated associations between single nucleotide polymorphisms (SNPs) in the tubulin beta class I (TUBB) and WW domain-containing oxidoreductase (WWOX) genes, gene-gene interactions, and gene-environment interactions and dyslipidemia in the Chinese Maonan ethnic group. Four SNPs (rs3132584, rs3130685, rs2222896, and rs2548861) were genotyped in unrelated subjects with normal lipid levels (864) or dyslipidemia (1129). While 5.0% of Maonan subjects carried the rs3132584TT genotype, none of the Chinese Han in Beijing subjects did. Allele and genotype frequencies differed between the normal and dyslipidemia groups for three SNPs (rs3132584, rs3130685, and rs2222896). rs2222896G allele carriers in the normal group had higher low-density lipoprotein cholesterol and lower high-density lipoprotein cholesterol levels. The rs3132584GG, rs3130685CC+TT, and rs2222896GG genotypes as well as the rs2222896G-rs2548861G and rs2222896G-rs2548861T haplotypes were associated with an elevated risk of dyslipidemia; the rs2222896A-rs2548861T and rs2222896A-rs2548861G haplotypes were associated with a reduced risk of dyslipidemia. Among the thirteen TUBB-WWOX interaction types identified, rs3132584T-rs3130685T-rs2222896G-rs2548861T increased the risk of dyslipidemia 1.371-fold. Fourteen two- to four-locus optimal interactive models for SNP-SNP, haplotype-haplotype, gene-gene, and gene-environment interactions exhibited synergistic or contrasting effects on dyslipidemia. Finally, the interaction between rs3132584 and rs2222896 increased the risk of dyslipidemia 2.548-fold and predicted hypertension.


INTRODUCTION
The average global prevalence of dyslipidemia is about 20% [1][2][3], and it is higher in patients with premature coronary arteriosclerotic diseases (CAD) [4] and chronic kidney disease [5]. Dyslipidemia plays crucial roles in the pathogenesis of hypertension, CAD, stroke, and chronic kidney failure that differ based on etiology, and identification of novel and safe treatments that decrease blood lipid levels, especially the level of low-density lipoprotein cholesterol (LDL-C), would be beneficial [6]. It is well known that lifestyle, environmental, and genetic factors all contribute to dyslipidemia. In the past decade, technological advances in biological sequencing AGING technology have enabled genome-wide association studies (GWASs) that have helped thoroughly characterize risk factors for dyslipidemia.
The tubulin beta class I gene (TUBB, gene ID: 203068) located on chromosome 6 in humans encodes a beta tubulin protein that is an essential component of microtubules. Microtubules perform many cellular functions, including chromosome segregation, maintenance of cell shape, transport and motility, and organelle distribution [7]. As a substrate of peptidylarginine deiminases, TUBB participates in citrullination, which is related to growth, infiltration, and drug resistance in some tumor cells. Thus, mutations in TUBB are involved in complex diseases [8] and carcinoma [9]. The WW domain-containing oxidoreductase gene (WWOX, gene ID: 51741) on human chromosome 16 encodes a member of the short-chain dehydrogenases/reductases protein family that is involved in a variety of important cellular processes, including induction of apoptosis, cell development, and steroid metabolism [10,11]. Mutations in WWOX are not only associated with multiple types of cancer [11], but also with lipid metabolism [12]. Both TUBB and WWOX are ubiquitously expressed in many tissues, including fat (reads per kilobase per million mapped reads (RPKM) 97.7 and 1.6, respectively), heart (RPKM 67.2 and 0.5) and liver (RPKM 58.1 and 0.9). Because of their functions and tissue distribution, single nucleotide polymorphisms in the TUBB and WWOX genes can have similar effects on lipid profiles. Recently, GWASs revealed that SNPs at locations rs3132584 and rs3130685 in TUBB are associated with LDL-C profile [13], while SNPs at rs2222896 and rs2548861 in WWOX are associated with LDL-C and high-density lipoprotein cholesterol (HDL-C) profiles [14], respectively. However, associations between TUBB and WWOX and lipid traits in the Chinese population have not been examined; we hypothesize that modifications of TUBB and WWOX affect blood lipid levels in this population as well.
The incidence of dyslipidemia in Chinese adults is as high as 34% [15]. Previous studies revealed that several lipid-related genes are associated with different lipid traits in the Maonan and Han populations [16][17][18]. The Maonan ethnicity is one of 55 recognized minorities in China, and seventy percent (64,500) of Maonan people live in Huanjiang Maonan Autonomous County of Guangxi Zhuang Autonomous Region according to a wire report from Xinhua News Agency (Beijing, May 20th, 2020). The Maonan people maintain a unique lifestyle and wedding culture characterized by a fat-rich diet, recreational use of alcohol and tobacco, and intraethnic marriages. As a result, the Maonan population has a singular genetic background and specific environmental risk factors that are particularly informative in genetic studies of dyslipidemia. In this study, we therefore explored associations between TUBB-WWOX SNPs, their haplotypes, and gene-gene (G × G) and geneenvironment (G × E) interactions and the prevalence of dyslipidemia in the Maonan population.

Participant characteristics
As shown in Table 1, no statistically significant differences in mean age, gender ratio, height, cigarette smoking, and proportion over 65 years old were observed between the normal and dyslipidemia groups (P > 0.05). However, mean weight, waist circumference, body mass index (BMI), alcohol consumption, systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse pressure (PP), fasting blood-glucose (FBS), hypertension morbidity, and proportion with BMI greater than 24 kg/m 2 and with FBS of at least 7.0 mM were higher in the dyslipidemia group than in the normal group (P < 0.01). In addition, total cholesterol (TC), triglyceride (TG), LDL-C, and apolipoprotein (Apo) B levels were also significantly higher in the dyslipidemia group than in the normal group (P < 0.001). In contrast, HDL-C and ApoA1 levels and ApoA1/ApoB ratio were significantly lower in the dyslipidemia group than in the normal group (P < 0.001).

Genotype and allele frequencies
Information for four target SNPs (rs3132584, rs3130685, rs2222896, and rs2548861) in the TUBB and WWOX genes among the Chinese Han Beijing (CHB) population in NCBI dbSNP Build 132 is shown in Table 2; genotypic and allelic frequencies of these four SNPs in the Maonan population are presented in Table 3. The frequencies of point mutations at all four SNPs were similar between the CHB and Maonan populations. Mutations from G to T and from C to T were observed at the TUBB rs3132584 and rs3130685 SNPs, respectively, while mutations from A to G and from T to G were observed at the WWOX rs2222896 and rs2548861 SNPs, respectively. However, genotypic and allelic frequencies of the four SNPs differed between the ethnic groups; in particular, the frequency of the rs3132584TT genotype was 5.0% in Maonan population but was 0% in the CHB population. In this study, allelic and genotypic distributions of the four SNPs in both groups were consistent with the Hardy Weinberg Equilibrium (HWE, P > 0.05 for all). Additionally, allele and genotype frequencies of the rs3132584, rs3130685, and rs2222896 SNPs differed between the normal and dyslipidemia groups (P ≤ 0.01 for all). HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol. 1 Normally distributed quantitative data were assessed in a t-test and are described as means ± SD. 2 Qualitative data were assessed using the chi-square test. 3 Non-normally distributed quantitative data were assessed using the Wilcoxon-Mann-Whitney test and are described as medians (interquartile range). P < 0.05 indicated a statistically significant difference.

Associations between genotypes and alleles, serum lipid levels, and dyslipidemia
Associations between the four TUBB and WWOX SNPs and serum lipid levels are shown in Figure 1. Serum levels of HDL-C, LDL-C, and ApoA1 differed among the three rs2222896 genotypes in the normal group (P ≤ 0.001 for all); rs2222896G allele carriers had higher LDL-C and lower HDL-C levels than those without rs2222896G. However, there were no differences in serum lipid levels depending on genotype for the other three SNPs. In the dyslipidemia group, some lipid parameters differed significantly depending on genotype for the rs3132584, rs2222896, and rs2548861 SNPs (P < 0.0125 for all). Furthermore, higher TC levels and lower ApoA1/ApoB ratio were associated with rs2222896G and rs2548861T alleles, higher LDL-C and ApoB levels and lower HDL-C levels were associated with the rs3132584G, rs2222896G, and rs2548861T alleles, lower ApoA1 levels were associated with the rs2222896G allele, and lower ApoA1/ApoB ratio was associated with the rs2548861T and rs2222896G alleles.
Because environmental factors can affect lipid phenotypes, we adjusted for the risk factors of age, sex, drinking, smoking, FBS, and BMI, to clarify relationships between genotypes and lipid phenotypes (Table 4). Overall, in the normal group, serum HDL-C levels were lower in rs2222896G allele carriers than in rs2222896G non-carriers (P ≤ 0.001), LDL-C was higher, but ApoA1 was lower, in subjects with the rs2222896GG genotype compared to those with the rs2222896AA genotype (P < 0.001), ApoB levels were higher in subjects with the rs3132584GG genotype than in those with the rs3132584TT genotype (P = 0.006), and ApoA1/ApoB ratio was lower in subjects with the rs2222896GG and rs3132584GT genotypes than in those with the rs2222896AA and rs3132584TT genotypes, AGING respectively (P = 0.012 and P = 0.007). In the dyslipidemia group, serum TC and ApoB levels were higher, but ApoA1/ApoB ratio was lower, in subjects with the rs2548861GG genotype than in those with the rs2548861TG genotype (P ≤ 0.005 for all), serum HDL-C and ApoA1 levels and ApoA1/ApoB ratio were lower, but LDL-C levels were higher, in subjects with the rs2222896GG genotype than in those with the rs2222896AA genotype, and serum HDL-C levels were lower in subjects with the rs3132584GT, rs3132584GG, and rs2548861TT genotypes than in those with the rs3132584TT and rs2548861GG genotypes (P = 0.001 for all).
Associations between the four SNPs and dyslipidemia are shown in Table 5. rs3132584, rs3130685, and rs2222896 were associated with dyslipidemia (P < 0.05).

AGING
Associations between the four haplotypes and serum lipid levels are shown in Figure 3. rs2222896G-rs2548861G AGING haplotype carriers in the normal group had lower HDL-C levels than those without rs2222896G-rs2548861G haplotype (P = 0.001), rs2222896A-rs2548861G haplotype carriers had higher ApoA1 and HDL-C levels and ApoA1/ApoB ratio, but lower TG levels, than those without rs2222896A-rs2548861G haplotype (P ≤ 0.002), and rs2222896G-rs2548861T haplotype carriers had higher TC levels than those without rs2222896G-rs2548861T haplotypes (P < 0.001). In the dyslipidemia group, rs2222896A-rs2548861T haplotype carriers had lower TC and ApoB levels, but higher HDL-C and ApoA levels and ApoA1/ApoB ratio than those without corresponding haplotype (P < 0.001), rs2222896A-rs2548861G haplotype carriers had lower LDL-C and higher ApoA1 and TG levels than those without corresponding haplotype (P ≤ 0.004), and rs2222896G-rs2548861T haplotype carriers had lower ApoA1 and HDL-C levels and ApoA1/ApoB ratio than those without corresponding haplotype (P < 0.001).
As shown in Table 7, after adjusting for sex, age, BMI, FBS, smoking, and drinking in the normal group, the rs2222896G-rs2548861T haplotype was associated with increased levels of serum TC and LDL-C (P < 0.001), the rs2222896A-rs2548861G haplotype was associated with increased serum HDL-C and ApoA1 levels and ApoA1/ApoB ratio, but decreased serum TG levels (P ≤ 0.006), and the rs2222896G-rs2548861G haplotype was associated with decreased serum HDL-C levels (P < 0.001). In the dyslipidemia group, the rs2222896A-rs2548861T haplotype was associated with decreased serum TC and ApoB levels and increased serum HDL-C and ApoA1 levels and ApoA1/ApoB ratio (P < 0.001), the rs2222896A-rs2548861G haplotype was associated with decreased serum LDL-C levels and increased serum ApoA1 (P ≤ 0.006) and TG levels (P < 0.001), and the rs2222896G-rs2548861T haplotype was associated with decreased serum ApoA1 and HDL-C levels and ApoA1/ApoB ratio (P < 0.001).
Interactions between the four haplotypes and environmental factors on dyslipidemia are shown in Figure 4. Dyslipidemia morbidity was increased in those with the rs2222896G-rs2548861G and rs2222896G-rs2548861T haplotypes compared to those without corresponding haplotypes (OR = 1.44, 95% CI = 1.18−1.76, P < 0.001 and OR = 15.08, 95% CI = 9.24−24.60, P < 0.001, respectively); environmental factors such as BMI > 24 kg/m 2 and FBS ≥ 7.0 mM also increased dyslipidemia morbidity. In contrast, dyslipidemia morbidity was decreased in those with the rs2222896A-rs2548861T and rs2222896A-rs2548861G haplotypes compared to those without corresponding haplotypes (OR = 0.47, 95% CI = 0.39−0.56, P < 0.001 and OR = 0.70, 95% CI = 0.56−0.86, P < 0.001, respectively). Interactions between haplotype and environmental factors had different effects on the incidence of dyslipidemia than those observed for each individual locus. Interactions between the rs2222896G-rs2548861G and rs2222896G-rs2548861T haplotypes and smoking, high BMI, or FBS increased the risk of dyslipidemia, as did the interaction between the rs2222896G-rs2548861T haplotype and sex or age (P < 0.05 for all). Furthermore, interactions between the rs2222896A-rs2548861T haplotype and sex, age, BMI, drinking, or smoking, and between the rs2222896G-rs2548861T haplotype and BMI or FBS decreased the effects of the environmental factors on dyslipidemia (P < 0.05 for all).

Associations between G × G interactions and dyslipidemia
As shown in Table 8, thirteen TUBB-WWOX G × G interactions were observed in the normal and dyslipidemia groups (LEF for gene interactions > 0.03).
Frequencies of nine of these G × G interactions, namely and T-C-G-T, differed significantly between the two groups (P < 0.05). Among them, only the T-T-G-T interaction was associated with increased incidence of dyslipidemia (OR = 1.371, 95% CI = 1.112-1.689, P = 0.003), while the other eight interactions, particularly T-T-G-G, were associated with reduced incidence of dyslipidemia (OR = 0.795, 95% CI = 0.640-0.988, P = 0.038).

Risk of hypertension based on different interactive models of dyslipidemia
The effects of traditional and potential genetic risks associated with dyslipidemia on hypertension are shown in Tables 11, 12. In this study, dyslipidemia, female sex, age > 65 years, BMI > 24 kg/m 2 , and FBS ≥ 7.0 mM were clear risk factors for hypertension (P < 0.05).    AGING Moreover, interactions between the rs3132584 and rs2222896 SNPs increased the risk of hypertension 1.523-fold (P < 0.006) after adjustment for BMI, sex, age, and FBS.

DISCUSSION
Dyslipidemia, a multifactorial non-communicable chronic disease, contributes to pathogenesis of arteriosclerotic diseases and is characterized by high TC, TG, and LDL-C and low HDL-C levels [6]. LDL-C and other ApoB-rich lipoproteins accumulate in the arterial wall, playing a key role in atherosclerotic plaque formation and subsequent related cardiovascular events [19]. Studies have shown that the risk of developing carotid plaque increases by 62% for every 1 mM increase in serum LDL-C concentration [20]. Lipid levels in the plasma are influenced by many factors, and genetic variation in lipid-related genes leads to diverse phenotypes in different countries and ethnicities. Previous GWASs demonstrated associations between the TUBB and WWOX genes and LDL-C and HDL-C levels, and differences in the expression of these two genes in different populations contributes to diversity in lipid traits. further demonstrating the differences in gene distribution between these two ethnic groups. Furthermore, although genotypic frequencies for rs3132584, rs3130685, and rs2222896 in the Maonan population differed significantly between the normal and dyslipidemia groups, only rs2222896 genotype was correlated with various serum lipid parameters. Normal group subjects with the rs2222896GG genotype had higher LDL-C and lower HDL-C and ApoA1 concentrations than those with other genotypes. This association between rs2222896 and serum LDL-C level was consistent with a previous study [21]. Additionally, Yamada et al. identified two novel loci in TUBB, rs3132584 and rs3130685, that were associated with high LDL-C in an exome-wide association study of early-onset dyslipidemia [13].
Another study of 2911 Japanese subjects revealed that the reference alleles rs3132584A and rs3130685C were associated with high LDL-C (P < 0.0001). In this study, we found that a rs3132584 mutation from G to T was present at a higher frequency. In the dyslipidemia group, rs3132584G was associated with high LDL-C levels; however, we observed little association between rs3130685C and LDL-C concentration. Two previous studies [14,22] indicated that the rs2548861 SNP in WWOX was associated with HDL-C concentration in the Young Finns Study population (n = 6,728, P = 6.9 × 10 -7 ), but another [23] failed to replicate this result in the Spanish population (n = 801). In our study, an association  H1, rs2222896A-rs2548861T haplotype; H3, rs2222896A-rs2548861G haplotype; I1, rs3132584G-rs3130685C-rs2222896A-rs2548861T; I4, rs3132584G-rs3130685C-rs2222896G-rs2548861G; I8, rs3132584T-rs3130685T-rs2222896G-rs2548861T; I10, rs3132584G-rs3130685T-rs2222896A-rs2548861G; I11, rs3132584T-rs3130685C-rs2222896G-rs2548861G; OR, odds ratio; CI, confidence interval. Different types of interactions were analyzed by logistic regression. P < 0.006 indicated a statistically significant difference after Bonferroni correction and adjusting for body mass index and fasting blood-glucose. BMI, body mass index; FBS, fasting blood-glucose; OR, odds ratio; CI, confidence interval. P < 0.05 indicated a statistically significant difference. AGING between rs2548861 and HDL-C concentration was observed in the dyslipidemia group (n = 1129, P = 0.001) but not in the normal group (n = 864, P = 0.597). We also found that HDL-C concentration in the dyslipidemia group might be affected by other non-genetic factors. Thus, we speculated that mutations at rs2548861 were not correlated with HDL-C concentration in these subjects. A correlation analysis between genetic models and disease risk was then conducted to understand the relationship between alleles at these four SNP sites and dyslipidemia. We found that codominant, dominant, recessive, and log-additive models of the rs3132584, rs3130685, and rs2222896 SNPs identified significant differences in dyslipidemia. Using the log-additive models, we identified the direction of the associations between each minor allele frequency (MAF) of these three SNPs and dyslipidemia. The rs3132584T and rs2222896A alleles were potential protective factors, while rs3130685C was a risk factor, for dyslipidemia. Various lipid-related SNPs and the frequencies of different alleles in the TUBB and WWOX genes can therefore lead to different clinical lipid phenotypes in different populations.

AGING
Genetic mechanisms are influenced by many types of interactions [24][25][26], including G × G and G × E interactions, which can have synergistic or contrasting effects on gene expression [27,28]. Interactions among genes can include SNP-SNP interactions between individual genes, haplotype-haplotype interactions, and the combined effects of mutations in genes on different chromosomes [29,30]. Detailed information on the four target SNPs was obtained from the 1000 Genome database. It indicated that there was a large gap between rs3132584 and rs3130685 at chromosomal positions 30720650 and 31238429, respectively, that prevented LD between them. In contrast, rs2222896 and rs2548861 at chromosomal positions 78058601 and 78624496, respectively, were about 500 kb from each other, resulting in a weak LD between them. Of the four common haplotypes observed for rs2222896 and rs2548861 alleles, rs2222896A-rs2548861T and rs2222896A-rs2548861G were associated with reduced risk of dyslipidemia while rs2222896G-rs2548861G and rs2222896G-rs2548861T were associated with elevated risk for dyslipidemia. In addition, multi-site mutations in TUBB and WWOX altered the incidence of dyslipidemia in the Maonan population. Thirteen types of multi-site mutation combinations for the two genes were screened using SHEsisPlus online software (http://shesisplus.bio-x.cn/SHEsis.html). Among these 13 combinations, rs3132584T-rs3130685T-rs2222896G -rs2548861T increased the risk of dyslipidemia 1.371fold; all of the remaining 12 combinations could potentially decrease the risk of dyslipidemia. Although TUBB and WWOX are located on two different chromosomes, interaction models were successfully screened using the GMDR method (https://sourceforge. net/projects/gmdr/), and the effects of G × G and G × E interactions on dyslipidemia were examined further. The fourteen highest-performing models of interactions linked to dyslipidemia were identified; all had a 10 of 10 CV, training and testing balanced accuracy values greater than 0.5, and sign and permutation test P values less than 0.05. Furthermore, an intuitive interactive dendrogram and logistic regression analysis indicated that the rs3130685-rs2548861, rs3130685-rs2222896, H1-H3, I4-I8, and I4-110 interactions exhibited strong synergistic effects increasing the risk of dyslipidemia, while the rs3132584-rs2222896 and I1-I11 interactions decreased the risk of dyslipidemia. This study therefore demonstrated that WWOX haplotype and the combined effects of TUBB × WWOX have larger impacts on serum lipid levels and dyslipidemia than single SNPs in those genes [31]; additional experiments with larger samples sizes are needed to confirm this finding.
Many environmental factors, including aging, sex, unhealthy lifestyle (high-fat diet [32,33], smoking, drinking), high blood glucose, and weight and obesity, are directly or indirectly involved in the pathogenesis of dyslipidemia. Moreover, these environmental factors can interact with lipid-related genes to alter lipid expression profiles [34,35]. Several studies demonstrate that the incidence of hyperlipidemia is higher in elderly people (1259/1657, 76%, mean age 69 years) than in young people (203/494, 41%, mean age 29 years); this difference was attributed to age-associated declines in ribosome coverage in the vicinity of start codons and increases near stop codons, alterations in expression of genes associated with lipid metabolism, loss of hepatic LDL receptors, and decreases in sex hormone levels [36,37]. Furthermore, the risk of dyslipidemia was higher in postmenopausal women and elderly men [38,39]; food pickling processes can increase concentrations of secondary metabolites in lemons and onions, thus endangering health [40]; and heavy (8 or more drinks per week for women or 15 or more drinks per week for men) rather than low or moderate alcohol consumption was also linked to dangerous increases in lipid levels [41]. The Maonan ethnic group is located mainly in the Huanjiang region, which is known as the "hometown of cattle and grain" due to the prevalence of agriculture and beef cattle industries. Beef, duck, half-cooked chicken, animal offal, and pickled sour pork, snails, and vegetables, are common dishes in this region. In addition, consumption of wine made from rice, corn, sweet potatoes, pumpkins, etc., three times per day with meals and smoking cigarettes made from local dried tobacco leaves is common among Maonan men. Many high-risk environmental factors may therefore affect the Maonan population. We used random stratified sampling in our examination of dyslipidemia risk in this population to eliminate systematic differences in age and gender structure between the two groups. We found no differences in either proportion of subjects > 65 years old or severity of smoking between subgroups, but severity of alcohol consumption, BMI > 24 kg/m 2 , and FBS ≥ 7.0 mM were significantly higher in the dyslipidemia group than in the normal group, and these factors interacted with the TUBB and WWOX SNPs to affect the prevalence of dyslipidemia. Pairwise multiple regression interaction models and optimal interaction models from GMDR revealed that the rs2222896G-rs2548861G haplotype interacted with smoking, overweight/obesity, or FBS ≥ 7.0 mM, while the rs2222896G-rs2548861T haplotype interacted with old age, female sex, smoking, FBS ≥ 7.0 mM, or overweight status/obesity to exacerbate the risk of dyslipidemia. In addition, age > 65 years increased the synergistic effect of rs2222896-rs3130685, and FBS ≥ 7.0 mM increased the synergistic effects of H1-H3 or rs2222896-rs3132584 on dyslipidemia. In contrast, interactions between the rs2222896A-rs2548861T haplotype and female sex, old age, smoking, and drinking, and between the rs2222896A-rs2548861G haplotype and overweight status/obesity or FBS ≥ 7.0 mM, decreased the risk of dyslipidemia. Finally, overweight status/obesity decreased the protective effects of H1-H3 or I1-I11 interactions against dyslipidemia, and drinking weakened the increased risk of dyslipidemia associated with rs2222896, possibly because most subjects were low-moderate drinkers.
Blood pressure is also modified by multiple genetic and environmental factors and their interactions [42][43][44], and dyslipidemia is a common independent risk for hypertension [45]. We therefore further explored differences in the incidence of hypertension between the two groups and found that the prevalence of hypertension was associated with gender, age, BMI, FBS, smoking, drinking, and dyslipidemia. Interestingly, we also found that the rs3132584 and rs2222896 SNP interaction that affected risk of dyslipidemia also significantly increased the risk of hypertension; this might be a novel explanation of how dyslipidemia contributes to the pathogenesis of hypertension.
Some important strengths and weaknesses of this study should be considered when interpreting the results. Notable strengths include the following: (1) identification of differences in TUBB and WWOX mutation frequencies between the Maonan ethnic group and the CHB population provides novel data for human genomic studies; (2) detection of TUBB-WWOX interactions and G × E interactions that affected the prevalence of dyslipidemia might provide new therapeutic targets for dyslipidemia; and (3) this study characterized an integrated effect of the TUBB rs3132584 and WWOX rs2222896 SNPs on dyslipidemia that could also predict the risk of hypertension, identifying a possible novel pathogenic mechanism by which dyslipidemia can lead to hypertension. However, the following limitations should also be considered: (1) this study did not include other variable risk factors, such as daily exercise habits and anti-hyperlipidemia interventions; (2) despite the unique diet (high salt and sour, high fat and alcohol) associated with the Maonan population, intake of these foods was not accurately characterized and their effects on dyslipidemia could not be examined; (3) larger sample sizes are needed to verify the results of this study; and (4) biological function studies are needed to further examine the effects of interactions between these two genes on dyslipidemia.
In summary, this study identified associations between TUBB, WWOX, environmental factors, serum lipid levels, and dyslipidemia in the Maonan population. Our findings revealed that, compared to single-locus effects, TUBB-WWOX-environment interactions can result in synergistic or contrasting effects on incidence of dyslipidemia, thereby increasing or reducing the risk of dyslipidemia.

Participants
A total of 1993 unrelated subjects were selected from our previous sample library in 2015 through stratified random sampling [46]; all subjects were from three generations of the

Laboratory values
Serum lipid levels were measured in peripheral blood (3 mL) collected in the morning after fasting for more than 8 hours using an autoanalyzer (type 7170A; Hitachi Ltd., Tokyo, Japan). Serum TC, TG, HDL-C, and LDL-C levels were determined using commercially available enzymatic assay kits, and serum ApoA1 and ApoB levels were determined by turbidimetric immuno-assay [46][47][48].

Variable definitions
Dyslipidemia is defined by elevated levels of TC, LDL-C, and TG and decreased HDL-C levels [6]. In this study, normal ranges for serum TC, TG, HDL-C, LDL-C, ApoA1, and ApoB levels and ApoA1/ApoB ratio were 3.10−5.17 mM, 0.56−1.70 mM, 0.90−1.81 mM, 2.70−3.10 mM, 1.00−1.78 g/L, 0.63−1.14 g/L, and 1.00−2.50, respectively. Dyslipidemia was therefore defined by any of the following either alone or in combination: TC > 5.17 mM, TG > 1.7 mM, LDL-C > 3.10 mM, and/or HDL-C < 0.9 mM [49]. Blood pressure was measured after the subject had been sitting for at least 5 minutes. An average SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg for three measurements over two days was defined as hypertension [50]. BMI was determined by dividing weight (kilograms) by height (meters squared); overweight and obese categories in China [51] are defined as BMI > 24 kg/m 2 and BMI ≥ 28 kg/m 2 , respectively.

Statistical analyses
SPSS software, version 25 (SPSS Inc., Chicago, IL, USA) was used to perform statistical analyses. Normally distributed data are presented as means ± standard deviation (SD) and differences between groups were identified using Student's unpaired t-test and one-way analysis of variance (ANOVA). Non-normally distributed data are shown as interquartile ranges and medians and differences between groups were identified using Mann-Whitney nonparametric tests. Differences in qualitative variables, including different ratios and HWE between the two groups, were analyzed using Chi square tests. Associations between genotypes or haplotypes and continuous serum lipid level data were assessed by multivariable linear regression. Differences in serum lipid levels associated with genotypes and haplotypes were considered statistically significant at P < 0.0125, and G × G or G × E interactions were considered significant at P < 0.006 (corresponding to P < 0.05 after Bonferroni correction for four or eight independent tests). OR values and 95% CIs were calculated by multiple logistic regression after adjustment for stratified risk factors including age, sex, tobacco and alcohol consumption, BMI, and FBS. HWE, genotypic and haplotypic frequencies, and D′ and R 2 values used to describe LD were calculated using SHEsisPlus online software. GMDR online software was used to screen optimal SNP-SNP, haplotype-haplotype, G × G, and G × E interaction models. All other data visualizations were generated using GraphPad Prism (version 8.0.0).

AUTHOR CONTRIBUTIONS
C.-X.L. conceived and designed the study, collected data, performed statistical analyses, and drafted the manuscript. R.-X.Y. improved the design of study, organized the epidemiology investigation, collected samples, and further revised the manuscript.