Genetic linkage of hyperglycemia and dyslipidemia in an intercross between BALB/cJ and SM/J Apoe-deficient mouse strains

Individuals with dyslipidemia often develop type 2 diabetes, and diabetic patients often have dyslipidemia. It remains to be determined whether there are genetic connections between the 2 disorders. A female F2 cohort, generated from BALB/cJ (BALB) and SM/J (SM) Apoe-deficient (Apoe−/−) strains, was started on a Western diet at 6 weeks of age and maintained on the diet for 12 weeks. Fasting plasma glucose and lipid levels were measured before and after 12 weeks of Western diet. 144 genetic markers across the entire genome were used for quantitative trait locus (QTL) analysis. One significant QTL on chromosome 9, named Bglu17 [26.4 cM, logarithm of odds ratio (LOD): 5.4], and 3 suggestive QTLs were identified for fasting glucose levels. The suggestive QTL near the proximal end of chromosome 9 (2.4 cM, LOD: 3.12) was replicated at both time points and named Bglu16. Bglu17 coincided with a significant QTL for HDL (high-density lipoprotein) and a suggestive QTL for non-HDL cholesterol levels. Plasma glucose levels were inversely correlated with HDL but positively correlated with non-HDL cholesterol levels in F2 mice on either chow or Western diet. A significant correlation between fasting glucose and triglyceride levels was also observed on the Western diet. Haplotype analysis revealed that “lipid genes” Sik3, Apoa1, and Apoc3 were probable candidates for Bglu17. We have identified multiple QTLs for fasting glucose and lipid levels. The colocalization of QTLs for both phenotypes and the sharing of potential candidate genes demonstrate genetic connections between dyslipidemia and type 2 diabetes.


Background
Individuals with dyslipidemia have an increased risk of developing type 2 diabetes (T2D), and diabetic patients often have dyslipidemia, which includes elevations in plasma triglyceride and low-density lipoprotein (LDL) cholesterol levels and reductions in high-density lipoprotein (HDL) cholesterol levels [1]. Part of the increased diabetic risk associated with dyslipidemia is due to genetic variations that influence both lipoprotein homeostasis and the development of T2D. Indeed, a few rare gene mutations result in both dyslipidemia and T2D, which include ABCA1 [2], LIPE [3], LPL [4], and LRP6 [5]. Genome-wide association studies (GWAS) have identified >150 loci associated to variation in plasma lipids [6,7] and >70 loci associated with T2D, fasting plasma glucose, glycated hemoglobin (HbA1c), or insulin resistance [8][9][10]. Nearly a dozen of the loci detected are associated with both lipid and T2D-related traits at the genome-wide significance level, including GCKR, FADS1, IRS1, KLF14, and HFE (http://www.genome.gov/GWAStudies/). Surprisingly, half of them have shown opposite allelic effect on dyslipidemia and glucose levels [11], and this is in contrary to the positive correlations observed at the clinical level. Furthermore, it is challenging to establish causality between genetic variants and complex traits in humans due to small gene effects, complex genetic structure, and environmental influences.
A complementary approach to finding genetic components in human disease is to use animal models. Apolipoprotein E-deficient (Apoe −/− ) mice are a commonly used mouse model of dyslipidemia, with elevations in non-HDL cholesterol levels and reductions in HDL levels, even when fed a low fat chow diet [12,13]. High fat diet feeding aggravates dyslipidemia. Moreover, these mice develop all phases of atherosclerotic lesions seen in humans [14] and are extensively used for atherosclerosis research [15][16][17][18]. We have found that Apoe −/− mice with certain genetic backgrounds develop significant hyperglycemia and T2D when fed a Western-type diet but become resistant with some other genetic backgrounds [16,19,20]. BALB/cJ (BALB) and SM/J (SM) Apoe −/− mice exhibit differences in dyslipidemia and T2D-related phenotypes [16]. The objective of the present study was to explore potential genetic connections between dyslipidemia and T2D through quantitative trait locus (QTL) analysis of a female cohort derived from an intercross between BALB-Apoe −/− and SM-Apoe −/− mice.

Ethics statement
All procedures were in accordance with current National Institutes of Health guidelines (https://grants.nih.gov/ grants/olaw/Guide-for-the-Care-and-use-of-laboratoryanimals.pdf ) and approved by the institutional Animal Care and Use Committee (protocol #: 3109). Blood was drawn from the retro-orbital plexus of overnight fasted mice with the animals under isoflurane anesthesia.

Animals, experimental design and procedures
BALB and SM Apoe −/− mice were created using the classic congenic breeding strategy, as described [16]. BALB-Apoe −/− mice were crossed with SM-Apoe −/− mice to generate F 1 s, which were intercrossed by brother-sister mating to generate a female F 2 cohort. Mice were weaned at 3 weeks of age onto a rodent chow diet. At 6 weeks of age, female F 2 mice were started on a Western diet containing 21 % fat, 34.1 % sucrose, 0.15 % cholesterol, and 19.5 % casein by weight (Harlan Laboratories, TD 88137) and maintained on the diet for 12 weeks. Mice were bled twice: once before initiation of the Western diet and once at the end of the 12-week feeding period.
Overnight fasted mice were bled into tubes containing 8 μL of 0.5 mol/L ethylenediaminetetraacetic acid. Plasma was prepared and stored at −80°C before use.

Housing and husbandry
Breeding pairs were housed in a cage of 1 adult male and 2 females, and litters were weaned at 3 weeks of age onto a rodent chow diet in a cage of 5 or less. At 6 weeks of age, F 2 mice were switched onto the Western diet and maintained on the diet for 12 weeks. All mice were housed under a 12-h light/dark cycle at an ambient temperature of 23°C and allowed free access to water and drinking food. Mice were fasted overnight before blood samples were collected.

Measurements of plasma glucose and lipid levels
Plasma glucose was measured with a Sigma glucose (HK) assay kit, as reported with modification to a longer incubation time [21]. Briefly, 6 μl of plasma samples were incubated with 150 μl of assay reagent in a 96-well plate for 30 min at 30°C. The absorbance at 340 nm was read on a Molecular Devices (Menlo Park, CA) plate reader. The measurements of total cholesterol, HDL cholesterol, and triglyceride were performed as reported previously [13]. Non-HDL cholesterol was calculated as the difference between total and HDL cholesterol.

Genotyping
Genomic DNA was isolated from the tails of mice by using the phenol/chloroform extraction and ethanol precipitation method. The Illumina LD linkage panel consisting of 377 SNP loci was used to genotype the F 2 cohort. Microsatellite markers were typed for chromosome 8 where SNP markers were uninformative in distinguishing the parental origin of alleles. DNA samples from the two parental strains and their F 1 s served as controls. Uninformative SNPs were excluded from QTL analysis. SNP markers were also filtered based on the expected pattern in the control samples, and F 2 mice were filtered based on 95 % call rates in genotype calls. After filtration, 228 F 2 s and 144 markers were included in genome-wide QTL analysis.

Statistical analysis
QTL analysis was performed using J/qtl and Map Manager QTX software as previously reported [19,22,23]. One thousand permutations of trait values were run to define the genome-wide LOD (logarithm of odds) score threshold needed for significant or suggestive linkage of each trait. Loci that exceeded the 95th percentile of the permutation distribution were defined as significant (P < 0.05) and those exceeding the 37th percentile were suggestive (P < 0.63).

Prioritization of positional candidate genes
The Sanger SNP database (http://www.sanger.ac.uk/ sanger/Mouse_SnpViewer/rel-1410) was used to prioritize candidate genes for overlapping QTLs affecting plasma glucose and HDL cholesterol levels on chromosome (Chr) 9, which were mapped in two or more crosses derived from different parental strains for either phenotype. We converted the original mapping positions in cM for the confidence interval to physical positions in Mb and then examined SNPs within the confidence interval. Probable candidate genes were defined as those with one or more SNPs in coding or upstream promoter regions that were shared by the parental strains carrying the "high" allele but were different from the parental strains carrying the "low" allele at a QTL, as previously reported [24].

Trait value distributions
Fasting plasma glucose and lipid levels of F 2 mice were measured before and after 12-weeks of Western diet. Values of fasting plasma glucose, non-HDL cholesterol and triglyceride levels of F 2 mice on both chow and Western diets and of HDL cholesterol level on the chow diet were normally or approximately normally Fig. 1 The distributions of trait values for fasting plasma glucose, HDL, non-HDL cholesterol and triglyceride of 228 female F 2 mice derived from an intercross between BALB-Apoe −/− and SM-Apoe −/− mice. Fasting blood was collected once before initiation of the Western diet (left panel) and once after 12 weeks on the Western diet (right panel). Graphs were created using a plotting function of J/qtl software distributed ( Fig. 1). Values of square root-transformed HDL cholesterol levels on the Western diet showed a normal distribution. These data were then analyzed to search for QTLs affecting the traits. Loci with a genomewide suggestive or significant P value are presented in Table 1.

Fasting glucose levels
A genome-wide scan for main effect QTL revealed a suggestive QTL near the proximal end of Chr9 for fasting glucose when mice were fed the chow diet (2.37 cM, LOD: 2.21) (Fig. 2 [21]. The suggestive QTL on distal Chr5 (101.24 cM, LOD 3.198) was novel. The BALB allele conferred an increased glucose level for both of the Chr9 QTLs while the SM allele conferred increased glucose levels for the 2 Chr5 QTLs ( Table 2).

Fasting lipid levels
Genome-wide scans for main effect QTLs showed that HDL, non-HDL cholesterol, and triglyceride levels were each controlled by multiple QTLs (Figs. 3, 4 and 5; Table 1). For HDL, 3 significant QTLs, located on Chr1, Chr7 and Chr9, and 1 suggestive QTL on Chr10, were identified. All 3 significant QTLs for HDL were detected when mice were fed either chow or Western diet, while the suggestive QTL on Chr10 was found when mice were on the chow diet. The significant QTL on Chr1 replicated Hdlq5, which had been mapped in numerous crosses [25]. The Chr7 QTL replicated Hdlcl1, initially mapped in (PERA/EiJ x B6-Ldlr)) x B6-Ldlr backcross [26]. The Chr9 QTL replicated Hdlq17, previously mapped in B6 x 129S1/SvImJ F 2 mice [27]. The suggestive QTL on Chr10 overlapped with Hdlq26 mapped in a SM/J x NZB/BlNJ intercross [28]. For all 4 HDL QTLs, F 2 mice homozygous for the SS allele had higher HDL levels than those homozygous for the BB allele (Table 2). For non-HDL cholesterol levels, 6 suggestive QTLs were detected when F 2 mice were fed the chow diet, and 2 significant and 2 suggestive QTLs were detected on the Western diet (Fig. 4). The 2 significant QTLs on Chr2 and Chr16 and the suggestive QTL on Chr11 were novel. The former 2 QTLs were named Nhdlq15 and Nhdlq16, respectively. Nhdlq15 peaked at 31.8 cM on Chr2 and affected non-HDL levels in a dominant mode from the BB allele while Nhdlq16 peaked at 46.66 cM on Chr16 and affected non-HDL levels in a dominant mode from the SS allele. The rest replicated previously identified ones in other mouse crosses: The Chr1 QTL peaked at 66.95 cM, overlapping with Chol7 mapped in an intercross of 129S1/SvImJ and CAST/Ei mice [29]. The Chr5 QTL overlapped with Hdlq34 mapped in PERA/EiJ × I/ LnJ and PERA/EiJ × DBA/2 J intercrosses [30]. The Chr6 QTL overlapped with Pnhdlc1, initially mapped in a B6 x CASA/Rk intercross and then replicated in B6 x C3H Apoe −/− F 2 mice [31,32]. The Chr8 QTL replicated Nhdlq1, initially mapped in B6 x 129S1/SvImJ F 2 mice [33]. The Chr9 QTL replicated Nhdlq11, initially mapped  in B6 x C3H Apoe −/− F 2 mice [32]. The Chr12 QTL peaked at 44.14 cM, overlapping with Nhdlq12 mapped in a B6 x C3H Apoe −/− F 2 intercross [32]. For triglyceride levels, 3 suggestive QTLs, located on Chr1, 2, and 5, respectively, were identified (Fig. 5). The Chr1 QTL peaked at 97 cM, 17 cM distal the Apoa2 gene (80 cM). The Chr2 QTL replicated Tgq11, mapped in an intercross between DBA/1J and DBA/2J [34]. The Chr5 QTL was novel.

Coincident QTLs for fasting glucose and lipids
LOD score plots for Chr9 showed that the QTL for fasting glucose (Bglu17) coincided precisely with the QTLs for HDL (Hdlq17) and non-HDL (Nhdlq11) in the  confidence interval (Fig. 6). F 2 mice homozygous for the BB allele exhibited elevated levels of fasting glucose and non-HDL but decreased levels of HDL, compared to those homozygous for the SS allele (Table 2). These QTLs affected their respective trait values in an additive manner.

Correlations between plasma glucose and lipid levels
The correlations of fasting glucose levels with plasma levels of HDL, non-HDL cholesterol, or triglyceride were analyzed with the F 2 population (Fig. 7).
Hdlq17 on Chr9 has been mapped in multiple crosses, including B6 x 129, B6 x CAST/EiJ, B6-Apoe −/− x C3H-Apoe −/− , and B6-Apoe −/− x BALB-Apoe −/− crosses Fig. 4 Genome-wide scans to search for loci influencing non-HDL cholesterol levels of female F 2 mice fed a chow (a) or Western diet (b). Two significant loci on chromosomes 2 and 16 were identified to affect non-HDL cholesterol levels of mice fed the Western diet [24,27,31,32,[36][37][38]. We conducted haplotype analyses using Sanger SNP database to prioritize positional candidate genes for both QTLs. Prioritized candidate genes for Hdlq17 are shown in Additional file 1: Table S1, and candidate genes for Bglu17 are shown in Additional file 2: Table S2. Most candidates for Hdlq17 are also candidate genes for Bglu17. These candidates contain one or more non-synonymous SNPs in the coding regions or SNPs in the upstream regulatory region that are shared by the high allele strains but are different from the low allele strains at the QTL. All candidate genes were further examined for associations with relevant human diseases using the NIH GWAS database (http://www.genome.gov/GWAStudies/). Sik3, Apoa1, and Apoc3 have been shown to be associated with variations in total, HDL, LDL-cholesterol or triglyceride levels [6,7,39], and Cadm1 with obesity-related traits [40].

Discussion
BALB and SM are two mouse strains that exhibit distinct differences in HDL, non-HDL cholesterol, and type 2 diabetes-related traits when deficient in Apoe [16]. BALB-Apoe −/− mice have higher HDL, lower non-HDL cholesterol, and lower glucose levels than SM-Apoe −/− mice when they are fed a Western diet. To identify the genetic factors responsible for these differences, we performed QTL analysis on a female cohort derived from an intercross between the two Apoe −/− strains. We have identified four loci contributing to fasting glucose levels, four loci contributing to HDL cholesterol levels, nine loci for non-HDL cholesterol levels, and three loci for triglyceride levels. Moreover, we have observed genetic connections between dyslipidemia and type 2 diabetes in that the QTL for fasting glucose is colocalized with the QTLs for HDL and non-HDL cholesterol on chromosome . Three suggestive loci were identified for triglyceride levels 9 and these coincident QTLs share a large fraction of potential candidate genes. We identified a significant QTL on chromosome 9, peaked at 26 cM, which affected fasting plasma glucose levels when mice were fed a chow or Western diet. We named it Bglu17 to represent a novel locus regulating fasting glucose levels in the mouse. This locus is overlapping with a significant QTL (not named) for blood glucose levels on the intraperitoneal glucose tolerance test identified in a BKS-Cg-Leprdb +/+m x DBA/2 intercross and a suggested QTL identified in a B6-Apoe −/− x BALB-Apoe −/− intercross [21,35]. Interestingly, we found that Bglu17 coincided precisely with Hdlq17, a QTL for HDL cholesterol levels, and Nhdlq11, a QTL for non-HDL cholesterol levels. The colocalization of two or more QTLs for different traits suggests that these traits are controlled either by the same gene(s) or closely linked but different individual genes. Hdlq17 has been mapped in multiple crosses derived from inbred mouse strains whose genomes have been resequenced by Sanger, including B6, 129, BALB, C3H/HeJ, and CAST/EiJ [24,27,31,32,[36][37][38]. Nhdlq11 was previously mapped in a NZB/BINJ x SM/JF2 cross and a B6- [32,41]. To determine whether Bglu17 and Hdlq17 share the same underlying candidate genes, we performed haplotype analyses on those crosses that led to the identification of the QTLs. The number of shared genetic variants between Bglu17 and Hdlq17 was surprisingly high. Of them, Sik3, Apoa1, and Apoc3 are located precisely underneath the linkage peak of Bglu17 and Hdlq17, and they are also functional candidate genes of Hdlq17. Indeed, recent GWAS studies have associated these three genes with dyslipidemia or variations in HDL, LDL cholesterol, and triglyceride levels [6,39,42]. The finding in this study strongly suggests that one or more of these "lipid genes" might be the causal gene(s) of Bglu17, contributing to variation in fasting glucose levels. Although it is unknown how they affect glucose homeostasis, one probable effect path is through the influence on plasma lipid levels, which then predispose variation in glucose-related traits. The current Fig. 6 LOD score plots for fasting glucose, HDL, and non-HDL cholesterol of F 2 mice fed the Western diet on chromosome 9. Plots were created with the interval mapping function of Map Manager QTX. The histogram in the plot estimates the confidence interval for a QTL. Two green vertical lines represent genome-wide significance thresholds for suggestive or significant linkage (P = 0.63 and P = 0.05, respectively). Black plots reflect the LOD score calculated at 1-cM intervals, the red plot represents the effect of the BALB allele, and the blue plot represents the effect of the SM allele. If BALB represents the high allele, then the red plot will be to the right of the graph; otherwise, it will be to the left observation on the significant correlations of fasting glucose levels with HDL, non-HDL cholesterol, and triglyceride levels in this cross supports this speculation. Plasma lipid levels, especially non-HDL cholesterol, of the F 2 mice were significantly elevated on the Western diet, so were the fasting glucose levels. When fed the Western diet, Apoe −/− mice display a rapid rise in non-HDL cholesterol levels, often reaching their peak within a couple of weeks (unpublished data), whereas their blood glucose levels rise more slowly and gradually within 12 weeks [43,44]. This difference in onset suggests a causal role for plasma lipids in the rise of blood glucose in the Apoe −/− mouse model. A significant reverse correlation was observed between plasma HDL cholesterol levels and fasting glucose levels in this cross on either chow or Western diet. This result is consistent with the findings of prospective human studies that low HDL levels can predict the future risk of developing T2D and low HDL levels are more prevalent in diabetic patients than in the normal population [45,46]. HDL can increase insulin secretion from βcells, improve insulin sensitivity of the target tissues, and accelerate glucose uptake by muscle via the AMPactivated protein kinase [47]. A significant correlation of non-HDL cholesterol levels with fasting glucose levels was also observed in this cross, and the correlation was extremely high when mice were fed the Western diet. Emerging human studies have also revealed associations of non-HDL cholesterol and ApoB with fasting glucose levels and incident type 2 diabetes [48][49][50]. We previously observed that the elevation of non-HDL cholesterol levels in Apoe −/− mice during the consumption of a Western diet induces a chronic, low-grade inflammation state characterized by rises in circulating cytokines and infiltration of monocytes/macrophages in various organs or tissues [13,17,20,43]. Inflammation in the islets impairs β-cell function [20]. LDL can also directly affect function and survival of β-cells [51]. In addition, high levels of LDL can induce insulin resistance due to its lipotoxicity and effect on endoplasmic reticulum stress [1].
Plasma triglyceride levels were strongly correlated with fasting glucose levels in this cross on the Western diet, although no significant correlation was found when mice were fed the chow diet. Despite the strong correlation, no overlapping QTLs were observed for fasting glucose and triglyceride. The reason for the discrepancy between non-HDL cholesterol and triglyceride in terms of the presence or absence of colocalized QTLs is unclear.
A suggestive QTL for fasting glucose near the proximal end of chromosome 9 (2.37 cM) was detected in this cross, initially on the chow diet and then replicated on the Western diet. The LOD score plot for chromosome 9 has shown 2 distinct peaks, one with a suggestive LOD score at the proximal end and one with a significant LOD score at a more distal region, suggesting the existence of two loci for fasting glucose on the chromosome. The bootstrap test, a statistical method for defining the confidence interval of QTLs using simulation [52], also indicated the existence of two QTLs for the trait on chromosome 9. We named the proximal one Bglu16 to represent a new QTL for fasting glucose in the mouse. Naming a suggestive locus is considered appropriate if it is repeatedly observed [53].
Two suggestive QTLs for fasting glucose on chromosome 5 were identified when mice were fed the Western diet. The proximal one replicated Bglu13, recently mapped in the B6-Apoe −/− x BALB-Apoe −/− cross [21]. One probable candidate gene for this QTL is Hnf1a, which encodes hepatocyte nuclear factor 1α. In humans, Hnf1a mutations are the most common cause of maturity-onset diabetes of the young (MODY) [54]. The suggestive QTL in the distal region was novel.
Most of the QTLs identified for plasma lipids confirm those identified in previous studies, whereas two QTLs for non-HDL are new and named Nhdlq15 and Nhdlq16, respectively. The QTLs on distal chromosome 1 for HDL and triglyceride has been mapped in a number of mouse crosses, and Apoa2 has been identified as the underlying causal gene [55]. However, the QTL (~90 cM) mapped in this study showed that it was more distal to the Apoa2 gene (80 cM), thus suggesting a different underlying causal gene.

Conclusion
We have identified multiple QTLs contributing to dyslipidemia and hyperglycemia in a segregating F 2 population. The finding on the colocalization of QTLs for fasting glucose, HDL and non-HDL cholesterol levels and the sharing of probable candidate genes has demonstrated genetic connections between dyslipidemia and type 2 diabetes. The close correlations of fasting glucose with HDL, non-HDL cholesterol, and triglyceride support the hypothesis that dyslipidemia plays a causal role in the development of type 2 diabetes [1]. The haplotype analysis has prioritized candidates for either chromosome 9 QTL down to a handful of genes. Nevertheless, functional studies need to be performed to prove causality.

Additional files
Additional file 1: Table S1. Haplotype analysis to prioritize candidate genes for HDL QTL Hdlq17 on chromosome 9. Sanger SNP database (http://www.sanger.ac.uk/sanger/Mouse_SnpViewer/rel-1410) was used to prioritize candidate genes for this locus, which were mapped in 3 separate crosses derived from different inbred strains. A high allele strain is the one that displays a larger allelic effect on HDL level at the locus and a low allele strain is the one showing a smaller allelic effect on HDL at the locus. (PDF 86 kb) Additional file 2: Table S2. Haplotype for glucose QTL Bglu17 on chromosome 9. Analysis was performed in the same way as for Hdlq17 illustrated above. (PDF 82 kb)