Genome wide association study identifies a functional SIDT2 variant associated with HDL-C levels and premature coronary artery disease.

Abstract

Objective:

Low HDL-C is the most frequent dyslipidemia in Mexicans, but few studies have examined the underlying genetic basis. Our purpose was to identify genetic variants associated with HDL-C levels and cardiovascular risk in the Mexican population.

Approach and Results:

A GWAS for HDL-C levels in 2,335 Mexicans, identified four loci associated with genome-wide significance: CETP, ABCA1, LIPC and SIDT2. The SIDT2 missense Val636Ile variant was associated with HDL-C levels and was replicated in 3 independent cohorts (P=5.9×10⁻¹⁸ in the conjoint analysis). The SIDT2/Val636Ile variant is more frequent in Native American and derived populations than in other ethnic groups. This variant was also associated with increased ApoA1 and glycerophospholipid serum levels, decreased LDL-C and ApoB levels, and a lower risk of premature CAD. Because SIDT2 was previously identified as a protein involved in sterol transport, we tested whether the SIDT2/Ile636 protein affected this function using an in vitro site-directed mutagenesis approach. The SIDT2/Ile636 protein showed increased uptake of the cholesterol analog dehydroergosterol, suggesting this variant affects function. Finally, liver transcriptome data from humans and the Hybrid Mouse Diversity Panel (HMDP) are consistent with the involvement of SIDT2 in lipid and lipoprotein metabolism.

Conclusions:

This is the first genome wide association study for HDL-C levels seeking associations with coronary artery disease in the Mexican population. Our findings provide new insight into the genetic architecture of HDL-C and highlight SIDT2 as a new player in cholesterol and lipoprotein metabolism in humans.

INTRODUCTION

Observational epidemiologic studies have reported that low plasma high density lipoprotein cholesterol (HDL-C) concentrations are an independent risk factor for cardiovascular disease.^{1, 2} Heritability of HDL-C serum levels has been estimated as high as 70% in various populations, including Mexican-Americans.^{3, 4} Genome wide association studies (GWAS) have successfully identified more than 150 loci associated with lipid levels mainly in European populations,^5–8 while relatively few GWAS have been performed in Mexicans.^9–12 The main HDL-C associated loci identified in Europeans are also associated with HDL-C levels in Mexicans, although novel loci have been reported in the latter group.¹⁰ Notably, a functional variant apparently private to the Americas (ABCA1 Arg230Cys) was found to be associated with lower HDL-C levels in Mexicans.^{13, 14} Although low HDL-C levels are a well-established cardiovascular risk factor, Mendelian randomization studies have shown that most genetic variants associated with this trait are not associated with cardiovascular risk, suggesting that this relationship is not necessarily causal.^15–17 In this regard, it has been postulated that pleiotropic effects of the genetic variants or HDL-C particle functionality rather than HDL-C plasma concentrations may affect cardiovascular risk.^18–21

Low HDL-C levels are highly prevalent in the Mexican population.²² This population group has been underrepresented in GWAS, and few lipid-associated variants have been tested for coronary artery disease (CAD) risk in Mexicans.²³ Therefore, we performed a GWAS for HDL-C levels in Mexican individuals, using a multiethnic array that includes rare and common genetic variants for Hispanic populations. We identified four loci associated with HDL-C levels with genome-wide significance, including a missense variant in the SIDT2 gene (Val636Ile, rs17120425). We then sought to replicate these associations in two independent cohorts: the Genetics of Atherosclerotic Disease (GEA) and the Morbid Obesity Surgery (MOBES) studies, and to test their possible effect on CAD risk. Lastly, we explored the effect of the SIDT2 gene, a member of a novel family of cholesterol transporters,²⁴ on lipid metabolism using existing liver transcriptome data from the Hybrid Mouse Diversity Panel, as well as the effect of the SIDT2 Val636Ile variant on uptake of the cholesterol analog dehydroergosterol (DHE) in vitro using a site-directed mutagenesis approach.

MATERIALS AND METHODS

The data that support the findings of this study have not been deposited in a public repository because this is inconsistent with consent documents. The data are also part of other studies in progress, but are available from the corresponding author upon reasonable request.

Study populations

A detailed description of all cohorts is provided in the Data Supplement. The Obesity Research Study for Mexican Children (ORSMEC) and Mexican Adult cohorts were used for the discovery phase. The Genetics of Atherosclerotic Disease (GEA) cohort, the Mexican Obesity Surgery (MOBES) cohort and a group of Native American individuals were used for the replication phase.

Ethics Statement

This study was conducted according to the principles expressed in the Declaration of Helsinki and was approved by the Ethics Committees of participant institutions. All adult participants provided written informed consent prior to inclusion in the study. For children, parents or guardians of each child signed the informed consent and children assented to participate.

GWAS and quality control

Genomic DNA was isolated from peripheral white blood cells using standard methods. A total of 2,335 children and adults included in the discovery phase were genotyped using the Multi-Ethnic Genotyping Array (MEGA, Illumina Inc., San Diego, CA), USA), which included >1600k SNPs. This array includes both common and rare variants in Latin American, African, European and Asian populations. Standard quality control (QC) measures were as previously described (Data supplement).²⁵ Identity-by-descent (IBD) was estimated using Plink v1.07.²⁶ A total of 624,242 SNPs remained after QC measures.

Selection of SNPs for the replication analyses

Ten SNPs at 7 loci found to be significantly associated with HDL-C levels in the discovery phase (P<1×10⁻⁵) were selected for replication in two independent cohorts (GEA controls and MOBES). These were the lead SNP of 4 loci (rs711975 in STAG1, rs6866550 in LOC107986449, rs983309 in LOC157273, and rs1077834 in LIPC), and 6 independent SNPs at 3 loci (rs4149310 and rs9282541 in ABCA1; rs948690 near BUD13 and rs17120425 in SIDT2, both near the APOA5 cluster; and rs12448528 and rs11508026 in CETP). SNPs were defined as independent when LD with other variants within the locus was low (r² and D’ <0.2). Genotypes of selected SNPs were obtained from microarray data, or genotyped using KASP (LGC, U.S. http://www.lgcgroup.com) or Taqman assays (ABI Prism 7900HT Sequence Detection System, Applied Biosystems). The rs17120425 variant was also genotyped in 302 Native Mexicans using Taqman probes. Call rates exceeded 95% and no discordant genotypes were found in 10% of duplicate samples. No SNPs deviated from Hardy–Weinberg equilibrium in any group (P>0.05).

Fine mapping of the 11q23 locus

For fine mapping of the 11q23 locus, we increased the density of imputed SNPs using the IMPUTE2 algorithm, and phased genotypes from the 1000 Genomes Project and from NAT trios as separate reference panels. Unphased genotypes from the discovery cohorts were imputed with default parameters for effective population size, Markov chain Monte Carlo iterations and hidden Markov Models. Only genotypes imputed with a probability > 0.9 were included in the analysis. Because of special interest in rs11216230 (SIK3 gene), which lies within the 11q23 locus and was previously associated with HDL-C in Mexicans,¹⁰ only subjects in whom this SNP was successfully imputed were included in the analysis (1162 children and 981 adults).

To identify distinct association signals arising from multiple causal variants in the same region, we screened for independent index SNPs in the chr11q23 region using conditional and join multiple-SNP (COJO) in GCTA.^{27, 28} The GCTA model made use of summary statistics from the random-effects meta-analysis in the discovery cohorts, and genotype data from these cohorts as a reference for LD across the region. We confirmed the index variant identified by GCTA (rs17120425) through conditional regression analysis using this SNP as covariate. We performed an additional conditioned analysis using the TG level lead SNP within the 11q23 region (rs964184), as covariate. Moreover, to identify the credible set of variants with a 99% probability of containing a causal variant at the SIDT2 locus, we used Bayesian fine-mapping estimating the approximate Bayes factors of association.²⁹ Afterwards, variants in this locus such that their cumulative posterior probability was greater than or equal to 0.99 were selected by using unscaled variance. Unconditioned statistics from our meta-analysis were used for Bayesian fine-mapping.

Mendelian Randomization

In order to test the causal effect of HDL-C levels on CAD we performed Mendelian Randomization (MR) analyses by using HDL-C associated SNPs as an instrument. In MR analyses, genetic variants act as proxies for HDL-C levels in a manner independent of confounders.³⁰ We used the inverse-variance weighted (IVW) method³¹ which assumes that all genetic variants satisfy the instrumental variable assumptions (including zero pleiotropy). We also performed MR-Egger regression,³² which allows each variant to exhibit pleiotropy. SNPs found to be associated with HDL-C (n=10) were included in the MR analyses. Both methods were performed with the aid of the Mendelian Randomization R package.³³

HEK293T cell cultures and wildtype and SIDT2/Ile636 transfection

Human embryonic kidney (HEK293T) cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA). Cells were grown on 35 mm culture dishes using Dulbecco’s modified Eagles medium (DMEM) (Invitrogen, Carlsbad, CA) supplemented with 10% of heat inactivated fetal bovine serum (Wisent, premium quality, Canada), penicillin-streptomycin and glutamine (Life Technologies) in an incubator with humidity control at 37 °C and 5% CO₂.

Human SIDT2 was cloned from human cDNA CGI-40 (AF151999.1) obtained from the Riken Consortium (Japan). The product was cloned in pEGFP-N1 (Clontech, Mountain View, CA. USA). The SIDT2/Ile636 variant was produced using the QuikChange® Site-Directed Mutagenesis Kit (Stratagene, Santa Clara, CA. USA) following manufacturer’s instructions. The wildtype SIDT2/Val636 variant will be referred to as SIDT2, and the isoleucine variant will be referred as Ile636/SIDT2. The primers used to change valine for isoleucine were the following: FORWARD 5´- GCGTTCTGGATCATTTTCTCCATCATT-3´ and REVERSE 5´-CGCAAGACCTAGTAAAAGAGGTAGTAA-3´. Constructs were sequenced prior to use.

The plasmids containing SIDT2-GFP and Ile636/SIDT2-GFP were transfected into HEK293T cells grown on 35 mm Petri dishes with a glass bottom (MatTek, Ashland, MA, USA). For transfection we used a mixture of 1 μg of DNA and 6 μl CaCl₂ reaching 60 μl of final volume with distilled H₂O. The mixture was added by dropping to HeBS buffer (50mM HEPES, 280mM NaCl, 1.5mM Na₂HPO₄, pH 7.05) and the final mixture was incubated for 30 minutes prior to replacing with DMEM. The cells were incubated overnight with the transfection mixture, which was then replaced with fresh medium to perform the assays 24 hours later.

Fluorescent cholesterol analog dehydroergosterol (DHE) uptake experiments

The naturally occurring blue fluorescent cholesterol analog DHE was purchased from Sigma (Saint Louis, MO). HEK293T cells expressing either SIDT2-GFP or SIDT2/Ile636-GFP were incubated with 5mM DHE solution, adding 300 mM methyl-β-cyclodextrin (MβCD). The solution was carefully resuspended and diluted in PBS to obtain a DHE/MβCD ratio of 1:8 (mol/mol) and 100 μL of this solution were added to the HEK293T cells. Cells were monitored using the scanning confocal microscope (FV1000, Olympus Japan). Focal plane was positioned at the middle of the cells. Confocal images were acquired every 15 seconds with no averaging to reduce photobleaching. Excitation was at 300 nm using a solid-state laser and emission was collected at 535 nm.

Liver transcriptome analysis in the Hybrid Mouse Diversity Panel (HMDP) and humans

HMDP:

The HMDP is a collection of approximately 100 well-characterized inbred strains of mice that can be used to analyze the genetic and environmental factors underlying complex traits such as dyslipidemia, obesity, diabetes, atherosclerosis and fatty liver disease. We analyzed the liver transcriptome of strains from this panel carrying the human cholesteryl ester transfer protein (CETP) and the human ApoE3-Leiden transgenes. At the age of about 8 weeks, these mice were placed on a “Western style” synthetic high fat diet supplemented with 1% cholesterol.³⁴ After 16 weeks on this diet, plasma lipid profiles were measured by colorimetric analysis as previously described³⁵ and animals were euthanized for the collection of liver tissue. Total RNA was isolated from the left lobe using the Qiagen (Valencia, CA) RNeasy kit (cat# 74104), as described.³⁶ Genome wide expression profiles were determined by hybridization to Affymetrix HT-MG_430 PM microarrays. Microarray data were filtered as previously described.³⁷ The ComBat method from the SVA Bioconductor package was used to remove known batch effects.³⁸ All animal work was conducted according to relevant national and international guidelines and was approved by the UCLA Institutional Animal Care and Use Committee (IACUC).

Humans:

Total RNA was extracted from liver biopsies of 144 MOBES participants using Trizol reagent (Invitrogen). Clinical and biochemical characteristics of this subgroup of patients are described in Table SVI in the Data Supplement. RNA sequencing was performed as previously described.³⁹ Briefly, RNA quality was assessed using the Bioanalyzer RNA chip analysis to ensure that the RNA integrity number was >7. Complementary DNA libraries were prepared using the TruSeq RNA Stranded Total RNA Library Preparation kit (Illumina) and sequenced using an Illumina HiSeq2500 instrument, generating approximately 50 million reads/sample. After data quality control, sequencing reads were mapped to the human reference genome using TopHat software v2.0.1⁴⁰ and quantified using Cufflinks software.⁴¹

Statistical Methods

For the discovery phase, genome-wide association with HDL-C was tested independently in 4 groups (normal-weight and obese children and normal-weight and obese adults) under an additive linear mixed model with sex, age and BMI percentile (children) or BMI (adults) as fixed effects, and the genetic relatedness matrix as a random effect. An inverse variance method was used to perform a meta-analysis of the 4 groups. A Z-score was calculated for each SNP in each study, which summarized the magnitude and direction of the effect relative to the minor allele. The Z-score for the meta-analysis was calculated from the weighted sum of the individual study statistics.⁴² Genetic relationship matrices from genome-wide data were considered for the analysis using GCTA software.²⁷ A P-value <5×10⁻⁸ was considered genome-wide significant, suggestive significance was defined as a P-value <1×10⁻⁵. Group heterogeneity in the meta-analysis was evaluated by I² and Cochrane’s Q, using the R package meta. We used publicly available databases such as the GWAS Catalog (https://www.ebi.ac.uk/gwas/home) to annotate associated SNPs. SNPs within a 1Mb range of the SIDT2/Val636Ile variant were included in a locus zoom plot.⁴³ The Mexican Americans from Los Angeles (MXL) 1000 Genomes sample was used to define LD patterns.

For the validation phase, linear regression under additive models was used to test for genetic associations with log-transformed lipid traits (HDL-C, LDL-C, TC and TG levels) in the GEA control, MOBES and Native American cohorts. In the GEA cohort, associations were also tested for log-transformed apoA1 and apoB levels. In the MOBES cohort, associations were also tested for lipid classes. Genetic associations with premature CAD in the GEA cohort were tested using multiple logistic regression under additive models. All tests were adjusted by age, sex and BMI. Associations were tested using the SPSS Statistics package (IBM SPSS Statistics, version 24, Chicago, IL, USA), and statistical significance was considered at P<0.05.

Correlations of normalized SIDT2 liver expression with serum lipid levels and the normalized liver transcriptome were performed using biweight midcorrelation (bicor) coefficient with the R package WGCNA, a robust alternative to Pearson’s correlation coefficient not sensitive to outliers.⁴⁴ Genes significantly correlated with SIDT2 expression in liver (P≤1.0×10⁻⁴) were tested for pathway enrichment analysis using Metascape and ToppGene Suite software.^{45, 46} Enrichment P values <0.05 after FDR correction were considered significant.⁴⁷

RESULTS

Low HDL-C levels were highly prevalent in Mexican children and adults (24.7 and 57.7%, respectively). This trait was significantly more frequent in obese as compared to lean individuals (40.9% vs 10.5% respectively in children, and 78.1% vs 32.7% respectively in adults; P<0.001) (Tables SI and SII in the Data Supplement). To identify loci associated with HDL-C levels, we carried out a GWAS in 2 independent cohorts of Mexican children and adults using a multi-ethnic array. Of note, the same HDL-C associated loci were found in children and adults, regardless of obesity status, and effect sizes were similar and showed the same directionality. There was no significant evidence of heterogeneity (Table SVII), and therefore a fixed-effects meta-analysis was conducted.

In total, we identified 98 variants distributed across 7 chromosomal regions associated with HDL-C levels with genome wide or suggestive significance (P<1.0×10⁻⁵) after adjusting for age, sex, BMI and ancestry (Figure 1 and Table SVII in the Data Supplement). Most of the 98 variants were also associated with other lipid parameters. Four loci (CETP in chromosome 16, ABCA1 gene in chromosome 9, LIPC in chromosome 15, and a region near the APOA5 cluster in chromosome 11) showed genome-wide significance (P<5.0×10⁻⁸) adjusted by age, sex and BMI (Table 1). The same loci reached genome-wide significance when adjusted only by age and sex (Figure SII). Notably, the lead SNP rs17120425 in 11q23 near the APOA5 cluster is a missense variant (Val636Ile) within the SIDT2 gene, which was associated with HDL-C levels for the first time in the present study (P=4.4×10⁻¹²). The sex stratified meta-analysis showed that the effect size of SIDT2/Val636Ile on HDL-C levels was similar in females (B=3.45, P=3.6×10⁻⁷) and males (B=3.22, P=2.1×10⁻⁵) (Table SVIII).

Figure. 1. Manhattan plot for HDL-C levels in the discovery phase.

Plot showing the -Log10 transformed P-value of SNPs for 2,335 Mexican children and adults. The red line indicates the genome-wide significance level (P=5×10⁻⁸). Genes closest to the SNP with lowest P-value at each locus are indicated.

Table 1.

Lead SNPs associated with HDL-C levels in the meta-analysis including Mexican children and adults.

Chr: position (GRCh37)	Locus	rs ID	Test Allele	Type of variant	Z-Score	P-value
3:136235908	STAG1	rs711975	A	Intronic	4.636	3.6 × 10−6
5:130021677	LOC105379171/ LOC107986449	rs6866550	G	Intergenic	4.489	7.2 × 10−6
8:9177732	LOC157273	rs983309	A	Intronic	−5.056	4.3 × 10−7
9:107589134	ABCA1	rs4149310	A	Intronic	5.734	9.8 × 10−9
9:107620835	ABCA1	rs9282541	A	Missense. Arg230Cys	−8.040	8.9 × 10−16
11:116529468	LINC02702/BUD13	rs948690	A	Intergenic	5.708	1.1 × 10−8
11:117063003	SIDT2	rs17120425	A	Missense. Val636Ile	6.925	4.4 × 10−12
15:58723479	LIPC	rs1077834	A	Upstream	−6.055	1.4 × 10−9
16:56985555	CETP	rs12448528	A	Intergenic	−7.842	4.4 × 10−15
16:56999328	CETP	rs11508026	A	Intronic	9.226	2.8 × 10−20

HDL-C values were log transformed for the analysis. P-values for additive models were adjusted for sex, age, and either BMI or BMI percentile in children (n=2,335). HDL-C, High-density lipoprotein cholesterol; Chr: Chromosome.

Fine mapping of the 11q23 region

After imputation to increase SNP density within this region, a total of 3010 SNPs spanning 1 Mb in chromosome 11q23 were included in the analysis. Figure 2A shows the locus zoom plot for rs17120425 in the 11q23 region. We identified 96 other SNPs within this region associated with HDL-C levels (P<5×10⁻⁸; Table SIX). Five of these SNPs were in LD (r²>0.6) with rs17120425 (3 within the SIK3 gene, 1 in PCSK7 and 1 in PAFAH1B2), while two (rs948690 and rs2156121 near BUD13) were not in LD with the lead SNP (r²<0.2). Of note, rs11216230 within the SIK3 gene, previously associated with HDL-C levels and in high LD with SIDT2/Val636Ile (r²=0.75), was also significantly associated with HDL-C levels (P=1.1×10⁻⁸). The GCTA joint regression analysis of the 3010 genotyped and imputed SNPs in the 11q23 locus identified a single distinct association signal, represented by SIDT2/Val636Ile (P=1.5×10⁻¹¹). In addition, when conditioned on SIDT2/Val636Ile (rs17120425), only rs948690 and rs2156121 near BUD13 remained associated with HDL-C levels (P=1.9×10⁻⁵ and 4.3×10⁻⁵, respectively; Figure 2B). Bayesian fine-mapping identified a not well-resolved 99% credible set within this locus (96 variants comprised the credible set), however the index SNP (SIDT2/Val636Ile) showed the maximum posterior probability (PP_max= 0.302).

Figure 2. Locus zoom plots for associations with HDL-C levels in the 11q23 region, unconditioned and conditioning on rs17120425 (SIDT2/Val636Ile).

Plots show the regional associations before (A) and after (B) conditioning on rs17120425 (SIDT2/Val636Ile), which showed the strongest association with HDL-C levels (P=9.2×10⁻¹²). SNPs are colored based on their linkage disequilibrium (r²) with rs17120425 (purple diamond) using AMR subjects from 1000 Genomes data. -Log10 (P-values) indicate the significance of association for each SNP. Arrows on the horizontal blue lines show the direction of transcription, and rectangles represent exons.

Because this region (11q23) is well known for containing gene variation associated with TG levels, we analyzed whether rs964184, robustly associated with TG levels in previous studies, affected the association of the lead SNP with HDL-C levels. This SNP was strongly associated with higher TG levels (P=1.74×10⁻²²) and significantly associated with lower HDL-C levels (P=0.003) in our study, however conditioning on rs964184 did not modify the association of rs17120425 (SIDT2/Val636Ile) with HDL-C levels (P=7.3×10⁻¹¹, Figure SIII).

Notably, minor allele frequencies of ABCA1/Arg230Cys and SIDT2/Val636Ile variants are highest in populations from the Americas (Table SX). We then genotyped the SIDT2/Val636Ile variant in two independent Native American populations from central Mexico (Totonacs and Nahuas), which was also associated with higher HDL-C levels in these indigenous groups (B=2.81 mg/dL; P=0.027; Table SXI). Moreover, the rs17120425 “A” allele was significantly more frequent in Native Mexicans (15%) than in Mexican Mestizos (10.3%, P=0.001), and local ancestry analysis revealed that this allele was found in a block of Native American origin in 98% of individuals. However, according to EHH analysis, LD extension break down was similar in the ancestral and derived rs17120425 alleles, with no evidence of positive selection (Figure SIV, Methods supplement).

Replication of associations with HDL-C Levels and other lipid traits in independent cohorts

We then sought to replicate associations with HDL-C levels in GEA controls. Nine of the 10 variants associated with HDL-C levels in the discovery phase replicated in GEA controls (P<0.02, Table 2). Altogether, these nine variants explained ~25% of HDL-C level variation (P_{Genetic Risk Score}=1.6×10⁻²¹). Table 2 shows associations of these SNPs with other lipid traits in the GEA cohort. SIDT2//Val636Ile and rs948690 (near BUD13) were both associated with higher HDL-C and lower ApoB levels. SNP rs948690, but not SIDT2/Val636Ile, was also significantly associated with lower TG levels. In the MOBES cohort, rs11508026 and rs12448528 (CETP, P=8.8×10⁻⁵ and 0.013, respectively), rs9282541 (ABCA1, P=4.8 ×10⁻⁴) and SIDT2/Val636Ile (P=0.005) were significantly associated with HDL-C levels (Table 3). The association of SIDT2/Val636Ile with HDL-C levels was highly significant in the conjoint analysis including the discovery phase and replication cohorts (B=3.25, P=5.9×10⁻¹⁸) (Table SVIII).

Table 2.

Association of selected SNPs with lipid traits in the GEA cohort.

Chr: position (GRCh37)	Locus	rs ID	Test Allele	HDL-C		TC		LDL-C		TG		ApoA1		ApoB
				B (SE)	P- value	B (SE)	P- value	B (SE)	P- value	B (SE)	P- value	B (SE)	P- value	B (SE)	P- value
3:136235908	STAG1	rs711975	A	1.53 (0.55)	0.006	3.21 (1.63)	0.032	2.99 (1.42)	0.023	−5.37 (4.49)	0.038	2.32 (1.67)	0.136	0.70 (1.27)	0.500
5:130021677	LOC105379171/ LOC107986449	rs6866550	G	1.80 (0.72)	0.013	4.38 (2.15)	0.044	2.84 (1.86)	0.197	−1.08 (5.91)	0.949	2.23 (2.20)	0.286	1.74 (1.66)	0.403
8:9177732	LOC157273	rs983309	A	−2.32 (0.50)	2.0×10−6	−4.30 (1.60)	0.005	−2.69 (1.35)	0.016	4.03 (6.46)	0.076	−4.80 (1.43)	0.001	0.19 (1.19)	0.982
9:107589134	ABCA1	rs4149310	A	1.25 (0.49)	0.007	3.68 (1.55)	0.017	0.83 (1.32)	0.650	14.86 (6.22)	0.011	4.22 (1.38)	0.001	0.94 (1.15)	0.371
9:107620835	ABCA1	rs9282541	A	−3.81 (0.71)	5.5 ×10−−9	−6.49 (2.21)	0.004	−2.54 (1.88)	0.312	−1.53 (8.39)	0.639	−8.81 (2.06)	2.0×10−6	−1.36 (1.68)	0.452
11:116529468	LINC02702/ BUD13	rs948690	A	1.10 (0.52)	0.017	−1.11 (1.56)	0.374	−0.57 (1.35)	0.705	−12.19 (4.27)	0.001	1.94 (1.60)	0.126	−2.93 (1.21)	0.011
11:117063003	SIDT2	rs17120425	A	3.35 (0.75)	4.2×10−5	−3.66 (2.31)	0.157	−6.22 (1.96)	0.004	−4.75 (8.75)	0.477	10.29 (2.14)	3.0×10−6	−3.92 (1.74)	0.038
15:58723479	LIPC	rs1077834	A	−0.59 (0.46)	0.204	−0.15 (1.46)	0.930	1.66 (1.23)	0.090	−5.44 (5.87)	0.025	−3.77 (1.30)	0.009	−0.90 (1.08)	0.281
16:56985555	CETP	rs12448528	A	−3.77 (0.57)	2.9×10−12	−6.02 (1.85)	0.001	−2.96 (1.57)	0.062	3.91 (7.48)	0.340	−5.26 (1.66)	2.1×10−4	−2.09 (1.39)	0.118
16:56999328	CETP	rs11508026	A	2.79 (0.47)	3.8×10−10	2.25 (1.51)	0.130	−0.62 (1.28)	0.830	1.38 (6.06)	0.621	4.08 (1.35)	0.001	−0.49 (1.12)	0.979

P-values were calculated using log transformed HDL-C, TC, LDL-C, TG, ApoA1 and ApoB levels. B values were calculated for the minor alleles and are expressed as mg/dL. P-values for additive models were adjusted for sex, age, and BMI (n=1,592), SNPs rs711975, rs6866550 and rs948690 were genotyped in a subset of 1,036 individuals recruited as controls. Chr: Chromosome; HDL-C, High density lipoprotein cholesterol; TC, Total cholesterol; LDL-C, Low density lipoprotein cholesterol; TG, Triglycerides; ApoA1, Apolipoprotein A1; ApoB, Apolipoprotein B.

Table 3.

Association of selected SNPs with lipid traits in the MOBES cohort.

Chr: position (GRCh37)	Locus	rs ID	Test Allele	HDL-C		TC		LDL-C		TG
				B (SE)	P- value	B (SE)	P- value	B (SE)	P- value	B (SE)	P- value
3:136235908	STAG1	rs711975	A	−0.10 (0.63)	0.903	−3.10 (2.71)	0.275	−4.66 (2.22)	0.036	5.77 (5.13)	0.582
5:130021677	LOC105379171/ LOC107986449	rs6866550	G	−0.06 (0.76)	0.753	−0.62 (3.31)	0.752	−0.45 (2.73)	0.694	−1.15 (6.26)	0.840
8:9177732	LOC157273	rs983309	A	−0.38 (0.65)	0.810	−2.80 (2.81)	0.648	−3.54 (2.30)	0.255	6.25 (5.24)	0.236
9:107589134	ABCA1	rs4149310	A	0.66 (0.60)	0.302	3.70 (2.62)	0.305	3.16 (2.15)	0.162	2.98 (4.88)	0.345
9:107620835	ABCA1	rs9282541	A	−3.77 (0.96)	4.8 ×10−4	−8.78 (4.27)	0.077	−6.38 (3.50)	0.073	−4.26 (7.99)	0.844
11:116529468	LINC02702/BUD13	rs948690	A	−1.22 (0.57)	0.095	−2.13 (2.47)	0.652	−2.09 (2.03)	0.617	6.57 (4.67)	0.142
11:117063003	SIDT2	rs17120425	A	2.92 (0.98)	0.005	7.79 (4.30)	0.072	3.72 (3.54)	0.226	9.15 (8.04)	0.196
15:58723479	LIPC	rs1077834	A	−0.86 (0.57)	0.192	3.13 (2.50)	0.347	4.08 (2.05)	0.143	−2.57 (4.67)	0.264
16:56985555	CETP	rs12448528	A	−1.77 (0.72)	0.013	−1.71 (3.14)	0.705	0.12 (2.58)	0.668	−0.29 (5.86)	0.999
16:56999328	CETP	rs11508026	A	2.16 (0.55)	8.8×10−5	0.11 (2.44)	0.774	−1.14 (1.99)	0.695	−3.99 (4.54)	0.277

P-values were calculated using log transformed HDL-C, TC, LDL-C and TG levels. B values were calculated for the minor alleles and are expressed as mg/dL. P-values for additive models were adjusted for sex, age, and BMI (n=555). Chr: Chromosome; HDL-C, High-density lipoprotein cholesterol; TC, Total cholesterol; LDL-C, Low-density lipoprotein cholesterol; TG, Triglycerides.

The SIDT2 Val636Ile variant is associated with serum glycerophospholipid levels in the MOBES cohort.

We then sought associations of the 4 HDL-C associated SNPs in the MOBES cohort with lipid classes known to be the main components of HDL-C lipoprotein particles (19 classes of cholesterol esters, glycerophospholipids and TG).^{48, 49} SIDT2/Val636Ile was significantly associated with higher glycerophospholipid serum levels (P<0.05) (Figure 3), particularly with total phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylglycerol (PG), phosphatidylinositol (PI) and phosphatidylserine (PS) as adjusted by age, sex, BMI and lipid lowering treatment. In contrast, ABCA1 (rs9282541) and CETP variants (rs12448528 and rs11508026) were not significantly associated with any of these lipid classes.

Figure 3. Heat map of associations between the SIDT2/Val636Ile variant and lipid classes in the MOBES cohort.

Color intensity reflects the Beta value (red for positive, blue for negative) obtained from linear regression between HDL-C associated SNPs and lipid classes in the MOBES cohort (n=375), adjusted by age, sex and BMI. *P-value <0.05, ** P-value ≤0.001.

Association with premature CAD and Mendelian randomization

We then tested whether HDL-C variants were also associated with premature CAD in the GEA cohort. Five of the ten variants were significantly associated with CAD risk: rs7119175 (STAG1), rs948690 (near BUD13) and SIDT2/Val636Ile were associated with higher HDL-C levels and lower CAD risk, while rs9282541 (ABCA1) and rs1077834/LIPC were associated with lower HDL-C levels and lower CAD risk (Table 4). Mendelian randomization analyses were performed including the 10 variants significantly associated with HDL-C levels. There was no evidence of a causal effect of HDL-C levels on CAD (IVW P=0.585 and MR-Egger P= 0.322; Table SXII).

Table 4.

Association of HDL-C associated SNPs with premature coronary artery disease in the GEA cohort.

Chr: position (GRCh37)	Locus	rs ID	Allele Frequency			OR (95% CI)	P-value
			Test Allele	Controls (n=1,592)	Premature CAD (n=1,095)
3:136235908	STAG1	rs711975	A	35.8	30.6	0.77 (0.67–0.90)	0.001
5:130021677	LOC105379171/ LOC107986449	rs6866550	G	16.7	15.0	0.86 (0.71–1.05)	0.149
8: 9177732	LOC157273	rs983309	A	29.3	30.5	1.06 (0.93–1.21)	0.370
9: 107589134	ABCA1	rs4149310	A	34.4	36.6	1.11 (0.98–1.26)	0.116
9: 107620835	ABCA1	rs9282541	A	9.9	7.1	0.70 (0.56–0.86)	0.001
11:116529468	LINC02702/ BUD13	rs948690	A	51.6	46.7	0.81 (0.70–0.93)	0.003
11: 117063003	SIDT2	rs17120425	A	9.5	8.0	0.80 (0.65–0.99)	0.039
15: 58723479	LIPC	rs1077834	A	42.6	39.1	0.87 (0.77–0.98)	0.027
16: 56985555	CETP	rs12448528	A	20.5	18.1	0.87 (0.75–1.02)	0.087
16: 56999328	CETP	rs11508026	A	36.3	39.0	1.11 (0.99–1.26)	0.087

Associations with premature CAD were adjusted for age, sex and BMI. SNPs rs711975, rs6866550 and rs948690 were genotyped in a subset of 1,036 controls and 625 cases. P-values were calculated by logistic regression. HDL-C, High-density lipoprotein cholesterol; Chr: Chromosome; CAD, coronary artery disease; OR, Odds ratio; CI, confidence interval.

Fluorescent cholesterol analog uptake is enhanced in cells expressing Ile636/SIDT2.

It was previously suggested that the transmembrane conserved cholesterol binding (CRAC) domain of murine Sidt2 associates with the cholesterol analogue dehydroergosterol in HEK293 cells.²⁴ Because the SIDT2/Val636Ile variant is near the transmembrane CRAC domain in the human SIDT2 protein, we evaluated the effect of this variant on DHE uptake in HEK293T cells. Interestingly, DHE uptake was enhanced in cells expressing the Ile636/SIDT2 protein as compared to cells expressing wildtype SIDT2, reaching the highest difference approximately 1.5 minutes after adding DHE to the culture (6.85 ± 0.42 AU vs 9.40 ± 0.73 AU, P<0.01) (Figure 4, Video SI in the Data Supplement).

Figure 4. Uptake of the blue fluorescent cholesterol analog dehydroergosterol (DHE) by cells expressing wildtype and Ile636/SIDT2.

(A) Confocal microscopy images of HEK293T cells expressing the empty vector, wildtype and Ile636/SIDT2, at times 2.5, 3.5 and 5 minutes after adding the fluorescent cholesterol analog DHE to the culture. (B) Plot showing mean fluorescence intensity over time after adding DHE to the culture. Values show mean ± standard deviations from at least 4 independent experiments, each experiment shows mean values from all the cells in the focal plane. The red dots represent cells transfected with the empty vector; the black dots represent cells expressing wildtype SIDT2, and blue dots represent cells expressing Ile636/SIDT2. Addition of DHE is indicated with an arrow. *P-value <0.05; **P-value <0.01.

SIDT2 liver expression correlates with the expression of genes involved in lipid and lipoprotein metabolism

Because unidentified variants in LD with SIDT2/Val636Ile may affect expression of this gene, we tested for differences in hepatic SIDT2 expression levels according to genotype in the MOBES cohort, but the differences were not significant (P=0.486). Moreover, human SIDT2 liver expression showed no significant correlation with lipid traits including HDL-C, TC, or TG levels (Figure 5A). In contrast, in HMDP mice fed with an atherogenic diet, hepatic Sidt2 expression correlated positively with HDL-C levels (r=0.312, P=0.002), and negatively with total cholesterol and TG levels (r=−0.381, P=1.2×10⁻⁴; r=−0.304, P=0.002, respectively) (Figure 5B).

Figure 5. Correlations of liver SIDT2 expression and lipid levels in mice and humans.

Correlations of normalized liver Sidt2 expression with HDL-C, total cholesterol, and TG serum levels in MOBES cohort participants (A) and mice from the HMDP (B). (C) Venn diagram depicting the overlap of genes significantly correlated with SIDT2 liver expression in HMDP mice and MOBES cohort participants (P<1.0×10⁻⁴). (D) Significantly enriched pathways in mice and humans.

To analyze correlations between SIDT2 expression and the liver transcriptome in mice, we used data from the genetically diverse HMDP where environmental factors are controlled. A total of 2,240 genes showed significant correlations with Sidt2 expression, and the most significantly enriched pathway was lipid and lipoprotein metabolism (P_FDR=1.5×10⁻⁷). In human liver tissue SIDT2 expression correlated with the expression of 2,782 genes. Consistent with observations in mice, the lipid and lipoprotein metabolism pathways were among the most significantly enriched. SIDT2 expression correlated with the expression of 227 genes that were shared by both mice and humans, and as expected metabolism of lipids and lipoproteins was the most significantly enriched pathway (Figure 5C–D).

DISCUSSION

Low HDL-C levels are the most common dyslipidemia in Mexicans, both in adults and children.^{22, 50} Consistently, low HDL-C levels were highly prevalent in children and adults of the present study. Moreover, the prevalence of low HDL-C levels was significantly higher in obese than in lean individuals, in line with the high prevalence of metabolic syndrome observed in the Mexican population.⁵¹

Here, using a GWAS for HDL-C levels in different cohorts of the Mexican population, we identified four loci genome-wide associated with HDL-C levels at chromosomes 9 (ABCA1), 11 (near the APOA5 cluster), 15 (LIPC) and 16 (CETP). Of note, the effect and direction of the associations were consistent in the 4 study groups (normal weight and obese children and adults). Variants in ABCA1, LIPC and CETP have been robustly associated with HDL-C levels across populations.^{6, 11, 48} Of note, the signal in ABCA1 is the Arg230Cys variant (rs9282541) previously reported as associated with lower HDL-C levels by our group, apparently private to Native American and derived populations.^{13, 14} The signal in chromosome 11 near the APOA5 cluster corresponds to a missense variant within the SIDT2 gene (Val636Ile, rs17120425), associated with higher HDL-C levels. This association was replicated in 3 independent cohorts including Native Americans from Mexico. This is relevant because Native Americans live in rural areas, while the GEA and MOBES cohorts are urban populations, known to have different dietary habits, which may affect HDL-C levels.⁵²

To our knowledge, this is the first time that the SIDT2/Val636Ile variant has been associated with HDL-C levels in a GWAS. Although intronic SIDT2 variants (rs2269399, rs7107152, rs1242229 and rs1784042) have been previously associated with lipid traits and the metabolic syndrome in multi-ethnic cohorts, mainly in Koreans,^{53, 54} none of these variants are in LD with SIDT2/Val636Ile (r²<0.2). It is likely that previous GWAS in Mexicans failed to identify this SNP as associated with HDL-C because SIDT2/Val636Ile was not included in the microarray platforms used in these studies, and the paucity of Native American references for imputation.^9–12 Although the SIDT2/Val636Ile variant is not private to the Americas, it is more frequent in Native Americans (15%) and Hispanics (6%), than other ethnic groups (internationalgenome.org). Local ancestry analyses revealed that the derived “A” allele was of Native American origin in most individuals. However, the extended haplotype homozygosity (EHH) analysis of SIDT2/Val636Ile did not show evidence of recent positive selection.

The 11q23 region contains several lipid-associated genes in close proximity, and fine-mapping showed a cluster of HDL-C associated variants distributed in several genes. It was thus necessary to define whether these signals are independent of SIDT2/Val636Ile. Notably, conditional and joint multiple-SNP analysis of this region identified a single distinct signal associated with HDL-C levels represented by SIDT2/Val636Ile, suggesting there are no other relevant independent signals associated with HDL-C levels within this locus. A previous GWAS in the Mexican population identified two intronic SIK3 gene variants within the 11q23 region, rs139961185 associated with TG levels and rs11216230 with higher HDL-C levels.¹⁰ The latter association was replicated in this study and in an independent Hispanic population.¹² After conditioning the analysis on SIDT2/Val636Ile, the association of rs11216230 with HDL-C levels lost significance (P=0.500), indicating that at least in our study, this variant is not independent of SIDT2/Val636Ile. Moreover, conditioned analysis on rs964184 (APOA5 cluster), the lead SNP for TG levels in several populations,^9–11 also associated with HDL-C levels in the present study, did not modify the association of SIDT2/Val636Ile with HDL-C levels. This suggests that the association of SIDT2/Val636Ile with HDL-C levels is independent of the APOA5 locus. The credible set of variants obtained by Bayesian fine-mapping was not well-resolved, however the index SNP (SIDT2/Val636Ile) showed the maximum posterior probability (PP_max=0.302). Thus, fine mapping results suggest that, although SIDT2/Val636Ile can be considered as potentially casual, there may be other yet unidentified variants in this locus that influence serum HDL-C levels.

In Mendelian randomization analyses, genetic variants act as proxies for HDL-C levels in a manner independent of confounders to analyze the causality of HDL-C levels on coronary artery disease. Our MR analysis is consistent with previous studies suggesting that higher HDL-C levels are not causally protective against coronary heart disease.^15–17 This suggests that the effect of individual variants on CAD risk may be mediated by pleiotropic effects on other cardiovascular risk factors, or on HDL-C composition and functionality.^{18–21, 55} We thus explored whether SIDT2/Val636Ile affects other cardiovascular risk parameters in addition to HDL-C levels. Notably, this variant was also associated with higher ApoA1 levels, and lower LDL-C and ApoB serum levels in the GEA cohort. APOA1 is the major protein component of HDL-C particles, and a Mendelian randomization analysis in Finnish individuals reported that ApoA1 was not associated with risk of CAD.⁵⁶ A multivariable Mendelian randomization study examining serum lipid and apolipoprotein levels reported that only ApoB retained a robust relationship with the risk of CAD,⁵⁷ and recent Mendelian randomization studies suggest that ApoB is the primary lipid determinant of cardiovascular disease risk.^{58, 59} Thus, the association of SIDT2/Val636Ile with decreased cardiovascular risk observed in the GEA study could be mediated by its effect on ApoB levels. However, the association with premature CAD had suggestive significance (P=0.039) and needs to be confirmed in independent cohorts.

HDL lipidome composition has been associated with HDL-C functional properties,^{60, 61} Notably, of the 4 main variants associated with HDL-C levels in the MOBES cohort, only SIDT2/Val636Ile was associated with lipid species, specifically with higher serum concentrations of several glycerophospholipid classes including PE, PG, PC, PI and PS. It has been reported that HDL-C particles enriched in phospholipids can increase HDL-C stability,⁶² while decreased levels of phospholipids in HDL-C were found to impair cholesterol efflux and decrease the cardiovascular protective effects of HDLs.^{49, 61} Particularly, recent studies indicate that phosphatidylserine, a minor component of the monolayer surface of HDL-C, is enriched in small, dense HDL-C particles, which display potent anti-atherosclerotic activities.^{62, 63} Although we measured phospholipids in serum and not directly in HDL-C particles, a limitation of the study, the association of SIDT2/Val636Ile with higher phospholipid levels is consistent with the lower cardiovascular risk conferred by this variant.

The SIDT2 protein is found mainly in lysosome membranes,⁶⁴ is a lysosomal nucleic acid transporter, and is expressed in several tissues, including the liver.^65–67 The mammalian SIDT2 protein has high homology to the C. elegans cholesterol uptake protein-1 (CUP-1).⁶⁸ SIDT2 has been identified as a sterol-interacting protein⁶⁹ and more recently as a cholesterol-binding protein.²⁴ Moreover, SIDT2 is predicted to contain two CRAC domains (Cholesterol Recognition/interaction Amino Acid Consensus), found in a broad range of proteins involved in cholesterol transport, metabolism and regulation.^{24, 70} Specifically, the transmembrane CRAC domain from human SIDT1 and mouse SIDT2 appears to bind cholesterol.²⁴ The Val636Ile variant and the CRAC domain are located within the same transmembrane segment, and this variant is 19 amino acids upstream the tyrosine CRAC domain residue, suggested to interact with the cholesterol OH-polar group.²⁴ It is unknown if the Val636Ile variant modifies the interactions of the CRAC domain with cholesterol, thus affecting circulating lipid levels.

In the present study, HEK293T cells expressing the Ile636/SIDT2 protein showed higher cholesterol analog DHE uptake than those expressing the wildtype protein. This increased uptake observed in vitro, may affect circulating levels of cholesterol-rich lipoproteins. Although the mechanism by which this variant increases HDL-C serum levels is unknown, ABCA1-mediated cholesterol efflux and HDL-C formation are primarily dependent on autophagy for cholesterol source.⁷¹ Sidt2^−/− deficient mice show blocked autophagosome maturation, apparently by altering the fusion of autophagosomes with endosomes/lysosomes, the terminal stage of autophagosome maturation, which affects lipid metabolism. ⁷² It has been suggested that autophagy might simultaneously regulate the uptake and use of cholesterol in cells.⁷³ It is thus tempting to speculate that the putative gain of function of the Ile636 SIDT2 protein would enhance autophagosome-lysosome fusion, increasing both cholesterol uptake and autophagy-mediated cholesterol efflux which is primarily ABCA1 dependent, thus elevating HDL-C levels. Further studies are required to establish how this variant affects lipid metabolism in humans.

Sidt2 knockout mice show a wide range of metabolic phenotypes including impaired glucose tolerance likely due to compromised NAADP-involved insulin secretion.^{72, 74–77} Meng et al.,⁷⁷ showed that Sidt2 deficient mice present significantly increased serum total cholesterol, TG and LDL-C levels, and significantly lower HDL-C serum levels as compared to Sidt2^+/+ mice. These findings are consistent with the inverse correlation between hepatic Sidt2 expression and total cholesterol and triglycerides levels, and the direct correlation of Sidt2 expression with HDL-C levels observed in the HMDP. Moreover, Sidt2^−/− mice not only have altered lipid serum levels, but also showed impaired liver function and liver steatosis.^{72, 76, 77} Consistently, Sidt2 expression correlated with increased fat liver content in HMDP mice (data not shown). In contrast, in the MOBES cohort SIDT2 liver expression did not correlate with serum lipid levels or liver steatosis. This discrepancy may be explained by differences in the design of the HMDP and the MOBES cohort. Firstly, HMDP mice had atherosclerotic-promoting transgenes (apoE-Leiden and human CETP) and were fed a “Western” diet. While this provides a unique model to study lipid metabolism, it may also introduce a bias towards certain mechanisms. Moreover, liver biopsies were performed in MOBES participants during bariatric surgery and the patients had previously been submitted to a ketogenic diet. Several studies have reported that diet regulates the hepatic expression of genes involved in lipid metabolism.^{78, 79} Whether dietary components affect SIDT2 expression and its correlation with metabolic traits, requires further study.

Because correlations of the expression of a single gene with metabolic traits may overlook common disease mechanisms among species,⁸⁰ we performed pathway enrichment analyses. In this regard, SIDT2 expression correlations with the liver transcriptome revealed that the most significantly enriched pathways were lipid and lipoprotein metabolism in both species. Altogether, these findings support a role of SIDT2 in lipid metabolism. Further studies are required to better understand the mechanisms by which SIDT2 participates in cholesterol and lipoprotein metabolism at the cellular and systemic levels, and its role in cardiovascular risk.

In conclusion, this is the first study assessing genetic variants contributing to HDL-C levels and coronary artery disease in the Mexican population. Our GWAS revealed for the first time that the SIDT2/Val636Ile variant is associated with increased HDL-C and phospholipid levels and decreased risk of CAD. We also provide evidence that the SIDT2/Val636Ile variant is functional, increasing the uptake of a cholesterol analog in vitro. Our data support a role of SIDT2 in cholesterol and lipid metabolism. The mechanisms by which this protein affects lipid and metabolic parameters in humans require further investigation.

HIGHLIGHTS.

• A GWAS in Mexican individuals identified the SIDT2/Val636Ile variant associated with higher HDL-C levels.

• SIDT2/Val636Ile was associated with higher HDL-C, ApoA1 and glycerophospholipids, and with lower LDL-C and ApoB levels in independent cohorts.

• The SIDT2/Val636Ile variant was also associated with a lower risk of premature CAD.

• Transcriptomic data in mice and humans identified lipid and lipoprotein metabolism pathways as the most significantly correlated with SIDT2 liver expression.

• Cholesterol analog dehydroergosterol uptake was increased in HEK293T cells transfected with Ile636/SIDT2 as compared to wildtype SIDT2.

Abbreviations

ApoA1

Apolipoprotein A1

ApoB

Apolipoprotein B

CAD

Coronary artery disease

GEA

Genetics of Atherosclerotic Disease

GWAS

Genome wide association studies

HDL-C

High density lipoprotein cholesterol

HMDP

Hybrid Mouse Diversity Panel

LDL-C

Low density lipoprotein cholesterol

MOBES

Morbid Obesity Surgery

TC

Total cholesterol

TG

Triglycerides