A total of 1,436 unrelated people (668 males, 46.52% and 768 females, 53.48%; 749 normal, 52.16% and 687 HLP, 47.84%) of the Han ethnic group and 1,393 unrelated individuals (624 males, 44.80% and 769 females, 55.20%; 631 normal, 45.30% and 762 HLP, 54.70%) of the Maonan ethnic group were randomly selected from our previously stratified randomized specimen bank (Wang et al., 2016; Bin et al., 2017). All participants were occupants of the same location in the Guangxi Zhuang Autonomous Region, which is the Huanjiang Maonan Autonomous County of China. All individuals were farmers, ages from 20 to 88. Both Han and Maonan ethnic groups possessed similar sex ratio, age distribution, and average age (55.38 ± 12.46 vs. 56.35 ± 15.34; respectively). All subjects had no previous history of type-2 diabetes mellitus (T2DM), ischemic stroke, CAD, and myocardial infarction. None of the subjects was taking any medication that impacted blood lipid profiles. All protocols were vetted by the Ethics Committee of the First Affiliated Hospital, Guangxi Medical University (No. Lunshen-2014 KY-Guoji-001, March 07, 2014). All participants provided signed informed consent prior to the study.

Epidemiological investigation was performed in compliance to protocols previously established (Tao et al., 1992). Questionnaires allowed for documentation of relevant demographic and lifestyle information. Three separate subgroups each of smoking status [0 (non-smoker), <20 cigarettes/day, and ≥20 cigarettes/day] and alcohol consumption [0 (non-drinker), <25 g/day, and ≥25 g/day] were formed. The waist circumference, height, blood pressure, and body mass index (BMI) were assessed based on previous research (Guo et al., 2015a).

A fasting venous blood sample of 5 ml was collected from each participant. A portion of the sample (2 ml) was collected in a test tube and used to measure serum lipid levels. The rest of the sample (3 ml) was collected into a test tube containing anticoagulants (13.20 g/l tri-sodium citrate, 14.70 g/l glucose, and 4.80 g/l citric acid) and utilized to extract deoxyribonucleic acid (DNA). Methods for measuring serum LDL-C, apolipoprotein A1 (ApoA1), apolipoprotein B (ApoB), TC, TG, and HDL-C were described in detail in a previous study (Sun et al., 2015). All measurements were performed using an autoanalyzer (Type 7170A; Hitachi Ltd., Tokyo, Japan) in the Clinical Science Experiment Center of the First Affiliated Hospital, Guangxi Medical University (Guo et al., 2015b).

Six SNPs located on the SYTL3 and SLC22A3 were chosen using a number of pre-determined criteria: (1) SYTL3–SLC22A3 cluster was selected from previous GWASes associated with blood lipid levels. (2) Tagging SNPs were identified via Haploview (Broad Institute of MIT and Harvard, Cambridge, MA, United States, version 4.2), and potential lipid metabolism-associated functional SNPs were predicted using the latest version of the 1000 Genomes Project database. (3) More completely, details of the selected SNPs were collected from the NCBI dbSNP Build 132. (4) Regarding the SNP selection, we also referenced to a previous study by Ober et al. (2009), and the minor allele frequency (MAF) of all SNPs was more than 1% and associated with blood lipid profiles in previous study. (5) Six SNPs of SYTL3 rs9364496, rs6455600, rs2129209, and rs9456350 and SLC22A3 rs446809 and rs539298 were selected using the block-based method, which involves marking the association of linkage disequilibrium (LD) among chosen SNPs (r² > 0.8).

White blood cells were used for genomic DNA extraction with phenol–chloroform (Yin et al., 2011). All obtained DNA samples were numbered and maintained at −20°C until further studies were carried out. Next-generation sequencing technology (NGS) was used to analyze the genotypes of the six selected SNPs at the Center for Human Genetics Research, Shanghai Genesky Bio-Tech Co. Ltd., Shanghai, China (Onda et al., 2018). Supplementary Table 1 depicts all relevant primer sequences.

The following normal values were implemented in our research: serum ApoA1 (1.20–1.60 g/l), TG (0.56–1.70 mmol/l), LDL-C (2.70–3.10 mmol/l), TC (3.10–5.17 mmol/l), HDL-C (1.16–1.42 mmol/l), ApoA1/ApoB ratio (1.00–2.50), and ApoB (0.80–1.05 g/l). HLP was diagnosed when TC levels were more than 5.17 mmol/l and/or TG levels were more than 1.70 mmol/l (Guo et al., 2016). Diabetes (Alberti and Zimmet, 1998), body weights (obese, normal, or overweight) (Zhou, 2002), and hypertension (Chalmers et al., 1999; Whitworth, 2003) were diagnosed based on previously established criteria.

The SPSS (version 22.0, IBM Corp., Armonk, NY, United States) was used for all data analyses. The information of measurement parameters is depicted in terms of mean ± SD with the exception of TG levels, which is depicted in terms of median and interquartile ranges. Intergenotype variances between the six SNPs as well as between the proportion of alcohol consumers and smokers were evaluated using the chi-square test. Several common characteristics were tested by independent-sample t-test or Kruskal–Wallis test between both ethnicities. Haploview (Broad Institute of MIT and Harvard, Cambridge, MA, United States; version 4.2) allowed for assessment of the pairwise LD (measured by r²), Hardy–Weinberg equilibrium (HWE), and the frequencies of haplotypes or gene–gene interaction haplotypes in two ethnic groups. HLP diagnosis and its correlation to genotypes/haplotypes were assessed using unconditional logistic regression analysis. Analysis of covariance (ANCOVA) was used to assess the association between genotypes/haplotypes and serum lipid profiles, and P < 0.008 or P < 0.0045 (equivalent to P < 0.05 after adjusting for six SNPs or 11 haplotypes in independent tests by Bonferroni correction) was determined to be possessing significant statistical significance. Related parameters including blood pressure, smoking, blood glucose, sex, BMI, age, and alcohol consumption as covariables were adjusted for the correlation analysis. The generalized multifactor dimensionality reduction (GMDR) was used to determine the ideal combination among gene–gene/environment exposures, haplotype–haplotype/environment exposures, and SNP–SNP/environment exposures (Lou, 2015). GMDR reduces high-dimensional genetic data to a single dimension by exploring interaction models through cross-validation and using maximum likelihood estimates to calculate the score-based statistics of each participant. A good model would ideally possess a cross-validation constancy (CVC) value close to 10, which allows for identification of the best model among all likely candidates. Balanced accuracy testing scores between 0.50 (model prediction results are no better compared to chance) and 1.00 (perfect model prediction results) are indicative of the extent of the case-control status. Permutated P-values for these models were obtained by performing 1,000 permutations. The GMDR analyses included all haplotypes, SNPs, and numerous environmental modifiers, such as smoking, drinking, hypertension, and BMI. The analyses were further adjusted for study population, gender, and age. Statistical significance was ascertained with a P-value of <0.05. Multivariable linear regression analyses implementing stepwise modeling were carried out to discern associations between genotypes, alleles, and various modifiable environmental components with blood lipid profile in the Han and Maonan ethnic groups. Likewise, statistical significance was deemed to be achieved if P < 0.05. An interactive heat map with various serum lipid profile parameters was constructed using the R software (version 3.3.0) (Zhao et al., 2014).

As shown in Table 1, there were several differences in serum lipid levels and general parameters between the Han and Maonan ethnic groups. We noticed that the Maonan population possessed higher levels of LDL-C, diastolic blood pressure, ApoB, systolic blood pressure, TC, TG, weight, alcohol consumption, pulse pressure, and BMI in contrast to the Han group (P < 0.05–0.001). However, serum ApoA1, HDL-C levels, and the ApoA1/ApoB ratio were decreased in the Maonan ethnicity than those in the Han ethnic group (P < 0.05–0.001). Gender ratio, age distribution, height, glucose, the proportion of smokers, and waist circumference were similarly distributed between both ethnic groups.

As shown in Table 1, further subgroup analyses found that there were significant differences in blood lipid levels and general parameters between the normal and hyperlipidemia populations in both ethnic groups. We noticed that weight, systolic blood pressure, TC, waist circumference, ApoB, the proportion of smokers, BMI, alcohol consumption, TG, diastolic blood pressure, pulse pressure, and LDL-C were raised in patients with HLP in contrast to healthy participants in both ethnic groups. Conversely, patients with HLP had lower levels of ApoA1, HDL-C, and ApoA1/ApoB ratio in contrast to healthy individuals; these findings were similar irrespective of ethnic groups. Interestingly, there was no any stark difference in height, gender, age distribution and glucose levels between HLP patients and normal subjects in both the Han and Maonan ethnic groups.

Figure 1 depicts the location of six detected SNPs of SYTL3–SLC22A3 cluster in a closely genomic area of chromosome 6. All six SNPs were distributed in a manner complying with HWE across both the Maonan and Han groups (P > 0.05 for all). As shown in Table 2, several SNPs possess allelic and/or genotypic frequencies that were significantly varied between the two ethnic groups (rs6455600, rs2129209, and rs446809); between the normal and HLP populations in the Han (rs6455600, rs2129209, rs446809, and rs539298) and Maonan groups (rs6455600, rs446809, and rs539298; P < 0.05–0.001, respectively). These differences may partly account for the differences in blood lipid levels and the risk of hyperlipidemia between the Han and Maonan populations.

As shown in Table 2, the dominant model of rs446809 SNP increased HLP-related morbidity risk in both ethnic groups, whereas the dominant model of rs539298 SNP reduced HLP-associated morbidity in both the Han and Maonan groups. The dominant models of rs6455600 and rs2129209 SNPs decreased morbidity of HLP in the Han ethnicity, whereas the dominant model of rs6455600 SNP raised HLP-related morbidity in the Maonan ethnicity.

As shown in Figure 2, several SNPs were associated with TC (rs6455600, rs2129209, and rs539298), TG (rs446809), and LDL-C (rs446809) levels in the Han and with TC (rs6455600 and rs539298), TG (rs446809), and LDL-C (rs446809) in the Maonan ethnic groups (P < 0.008–0.001 and P < 0.008, respectively, had statistical significance after Bonferroni correction).

Figure 3 suggests the presence of significant pair wise LD among selected loci across both ethnic groups. Figure 4 indicates the associations of 11 haplotypes with blood lipid levels. Several haplotypes were correlated with TC (SYLT3 A-C-A-A; SLC22A3 A-A and C-A; and SYTL3–SLC22A3 A-C-A-A-A-A and A-C-A-A-A-G), LDL-C (SLC22A3 A-A, C-A, and C-G and SYTL3–SLC22A3 A-C-A-A-A-A), and TG (SLC22A3 C-G and SYTL3–SLC22A3 A-C-A-A-C-G and G-T-C-A-C-A) in the Han and with TC (SYLT3 A-C-A-A; SLC22A3 A-A, A-G, and C-G; and SYTL3–SLC22A3 A-C-A-A-A-A, A-C-A-A-C-G, and G-T-C-A-A-A), LDL-C (SLC22A3 A-A, A-G, and C-G and SYTL3–SLC22A3 A-C-A-A-C-G), and TG (SLC22A3 C-G and SYTL3–SLC22A3 G-T-C-A-A-A) in the Maonan ethnic groups (P < 0.0045–0.001 and P < 0.0045, respectively, had statistical significance after Bonferroni correction).

As illustrated in Tables 3, 4, the dominant haplotype and G × G interaction haplotype were the SYTL3 A-C-A-A (>55% of the samples) and SYTL3–SLC22A3 A-C-A-A-C-A (>29% of the samples). The haplotypes of the SYTL3 A-C-A-A; SLC22A3 A-A and A-G; and SYTL3–SLC22A3 A-C-A-A-A-A raised HLP-linked morbidity in both the Maonan and Han ethnicities, whereas the haplotypes of the SLC22A3 C-G and SYTL3–SLC22A3 A-C-A-A-C-G decreased HLP-related morbidity in both ethnic groups. In the Han group, the haplotypes of the SLC22A3 C-A and SYTL3–SLC22A3 A-C-A-A-C-A and G-T-C-A-C-A had a protective effect, while the haplotype of SYTL3–SLC22A3 A-C-A-A-A-G increased HLP-related morbidity. In the Maonan group, the haplotypes of SYTL3–SLC22A3 G-T-C-A-A-A and G-T-C-A-C-A raised HLP-related morbidity (P < 0.05–0.001).

As presented in Table 5, several significant three-loci models comprising of SNP–SNP (rs9364496, rs6455600, and rs446809, possessing a testing accuracy of 56.71% and a CVC of 9 of 10), SNP–environment (rs446809 and rs539298 and BMI > 24 kg/m², possessing a testing accuracy of 70.88% and a CVC of 10 of 10), haplotype–haplotype (SYTL3 A-C-A-A and SLC22A3 A-G and C-G, possessing a testing accuracy of 60.65% and a CVC of 10 of 10), haplotype–environment (SLC22A3 A-G and C-G and BMI > 24 kg/m², possessing a testing accuracy of 70.42% and a CVC of 10 of 10), gene–gene (SYTL3–SLC22A3 A-C-A-A-A-A, A-C-A-A-A-G, and G-T-C-A-C-A, possessing a testing accuracy of 58.86% and a CVC of 10 of 10), and gene–environment (SYTL3–SLC22A3 A-C-A-A-A-G and G-T-C-A-C-A and BMI > 24 kg/m², possessing a testing accuracy of 58.45% and a CVC of 10 of 10) interactions were detected in the Han ethnicity (P < 0.001, respectively). Additionally, several different significant three-loci models comprising of SNP–SNP (rs6455600, rs446809, and rs539298, possessing a testing accuracy of 64.63% and a CVC of 10 of 10), SNP–environment (rs446809 and rs539298 and BMI > 24 kg/m², possessing a testing accuracy of 70.83% and a CVC of 10 of 10), haplotype–haplotype (SYTL3 A-C-A-A and SLC22A3 A-A and A-G, possessing a testing accuracy of 58.87% and a CVC of 10 of 10) haplotype–environment (SLC22A3 A-A and A-G and BMI > 24 kg/m², possessing a testing accuracy of 58.85% and a CVC of 10 of 10), gene–gene (SYTL3–SLC22A3 A-C-A-A-C-A, A-C-A-A-C-G, and A-T-C-G-C-A, possessing a testing accuracy of 57.18% and a CVC of 10 of 10), and gene–environment (SYTL3–SLC22A3 A-C-A-A-C-A and A-C-A-A-C-G and BMI > 24 kg/m², possessing a testing accuracy of 57.88% and a CVC of 8 of 10) interactions were also observed in the Maonan ethnicity (P < 0.0041–0.001, respectively).

Figure 5 shows an entropy-based interaction dendrogram constructed based on MDR, which reveals the presence of a strong powerful redundancy interactions between the rs6455600 and rs446809 (SNP–SNP), rs539398 and BMI > 24 kg/m² (SNP–environment), SLC22A3 A-G and C-G (haplotype–haplotype), SLC22A3 C-G and BMI > 24 kg/m² (haplotype–environment), SYTL3–SLC22A3 A-C-A-A-A-G and G-T-C-A-C-A (gene–gene), and SYTL3–SLC22A3 G-T-C-A-C-A and BMI > 24 kg/m² (gene–gene–environment) in the Han group. At the same time, there were strong powerful redundancy interactions between the rs446809 and rs539298 (SNP–SNP), rs539398 and BMI > 24 kg/m² (SNP–environment), SYTL3–SLC22A3 A-C-A-A-C-G and A-T-C-G-C-A (gene–gene), and SYTL3–SLC22A3 A-C-A-A-C-G and BMI > 24 kg/m² (gene–gene–environment) in the Maonan group. In addition, there were other strong powerful synergy interactions between SLC22A3 A-A and A-G (haplotype–haplotype) and SLC22A3 A-G and BMI > 24 kg/m² (haplotype–environment) in the Maonan group.

As shown in Table 6, individuals bearing the rs6455600 CT/TT and rs446809 CA/AA genotypes increased the risk of HLP compared to those with the rs6455600 CC and rs446809 CC genotypes [adjusted odds ratio (OR) = 1.319, 95% confidence interval (CI) = 1.060–1.812, P = 0.008] in the Han group. The carriers of the rs539298 AG/GG genotypes and BMI > 24 kg/m² (adjusted OR = 0.735, 95% CI = 0.482–0.924, P = 0.006), SLC22A3 A-G and C-G (adjusted OR = 0.676, 95% CI = 0.523–0.874, P < 0.001), SLC22A3 C-G and BMI > 24 kg/m² (adjusted OR = 0.658, 95% CI = 0.408–0.861, P = 0.004), SYTL3–SLC22A3 A-C-A-A-A-G and G-T-C-A-C-A (adjusted OR = 0.762, 95% CI = 0.612–0.915, P = 0.032), and SYTL3–SLC22A3 G-T-C-A-C-A and BMI > 24 kg/m² (adjusted OR = 0.649, 95% CI = 0.433–0.816, P = 0.011) reduced the risk HLP. On the other hand, carriers of the rs446809 CA/AA and rs539298 AG/GG (adjusted OR = 1.242, 95% CI = 1.018–1.522, P < 0.001), rs539298 AG/GG and BMI > 24 kg/m² (adjusted OR = 1.263, 95% CI = 1.122–1.657, P < 0.001), SLC22A3 A-A and A-G (adjusted OR = 1.838, 95% CI = 1.609–2.567, P < 0.001), SLC22A3 A-G and BMI > 24 kg/m² (adjusted OR = 2.360, 95% CI = 2.140–2.922, P < 0.001), and SYTL3–SLC22A3 A-C-A-A-C-G and BMI > 24 kg/m² (adjusted OR = 1.649, 95% CI = 1.359–1.901, P < 0.001) raised the risk of HLP in the Maonan group.

The relationship between the genotypes and/or alleles of the six selected SNPs and blood lipid characteristics across both Han and Maonan ethnicities is shown in Supplementary Table 2. We noticed that the genotypes of rs6455600 and rs539298; and the haplotypes of SYTL3 A-C-A-A, SLC22A3 C-A and A-G, and SYTL3–SLC22A3 A-C-A-A-A-A and A-C-A-A-A-G were associated with TC and the genotype of rs446809; and the haplotypes of SLC22A3 C-G, and SYTL3–SLC22A3 A-C-A-A-C-G and G-T-C-A-C-A were related to TG levels in the Han ethnicity. In the meantime, the allele of rs539298; the genotype of rs6455600; and the haplotypes of SYTL3 A-C-A-A, SLC22A3 A-A, A-G and C-G; and SYTL3–SLC22A3 A-C-A-A-A-A and A-C-A-A-C-G were associated with TC and the allele and genotype of rs6455600 and the haplotypes of SLC22A3 A-G and C-G were related to TG in the Maonan ethnicity (P < 0.005–0.001).

The relationship between serum lipid parameters and numerous modifiable variants including waist circumference, gender, alcohol drinking, age, BMI, blood glucose, blood pressure, and cigarette smoking across both the Han and Maonan ethnic groups are shown in Supplementary Table 3 (P < 0.05–0.001).

Figure 6 is a Pearson correlation analysis that proves the relationship between haplotypes; integrative variants; the SYTL3 rs9364496, rs6455600, rs2129209, and rs9456350 and SLC22A3 rs446809 and rs539298 SNPs; and serum lipid profiles. Various modifiable factors comprising of alcohol consumption, cigarette smoking, age, gender, and several known cardiovascular risk factors including BMI, blood glucose, and blood pressure were also linked to blood lipid profiles in both the Maonan and Han ethnic groups.