Quantitative Trait Loci Affecting Atherosclerosis at the Aortic Root Identified in an Intercross between DBA2J and 129S6 Apolipoprotein E-Null Mice

Apolipoprotein E-null mice on a DBA/2J genetic background (DBA-apoE) are highly susceptible to atherosclerosis in the aortic root area compared with those on a 129S6 background (129-apoE). To explore atherosclerosis-responsible genetic regions, we performed a quantitative trait locus (QTL) analysis using 172 male and 137 female F2 derived from an intercross between DBA-apoE and 129-apoE mice. A genome-wide scan identified two significant QTL for the size of lesions at the root: one is Ath44 on Chromosome (Chr) 1 at 158 Mb, and the other Ath45 on Chr 2 at 162 Mb. Ath44 co-localizes with but appears to be independent of a previously reported QTL, Ath1, while Ath45 is a novel QTL. DBA alleles of both Ath44 and Ath45 confer atherosclerosis-susceptibility. In addition, a QTL on Chr 14 at 73 Mb was found significant only in males, and 129 allele conferring susceptibility. Further analysis detected female-specific interactions between a second QTL on Chr 1 at 73 Mb and a QTL on Chr 3 at 21 Mb, and between Chr 7 at 84 Mb and Chr 12 at 77 Mb. These loci for the root atherosclerosis were independent of QTLs for plasma total cholesterol and QTLs for triglycerides, but a QTL for HDL (Chr 1 at 126 Mb) overlapped with the Ath44. Notably, haplotype analysis among 129S6, DBA/2J and C57BL/6 genomes and their gene expression data narrowed the candidate regions for Ath44 and Ath45 to less than 5 Mb intervals where multiple genome wide associations with cardiovascular phenotypes have also been reported in humans. SNPs in or near Fmo3, Sele and Selp for Ath44, and Lbp and Pkig for Ath45 were suggested for further investigation as potential candidates underlying the atherosclerosis susceptibility.


Introduction
Atherosclerosis is a major cause of myocardial infarction and stroke. Risk factors for atherosclerosis include smoking, obesity, diabetes, hypertension and high plasma cholesterol level, suggesting that formation and progression of atherosclerotic plaques are influenced by complex interactions between multiple genetic and environmental factors. Recent advances in genome-wide association studies (GWAS) of human populations have led to the identification of a number of loci associated with atherosclerosis [1]. However, except for those such as LDLR and apoE that affect plasma cholesterol levels, roles of other genetic factors that influence early pathogenesis of atherosclerosis are not fully understood [2].
Inbred strains of mice show a huge variance in atherosclerotic susceptibility; in apoE-null mice on the different genetic backgrounds, for example, the lesion size in the aortic root areas increases in the order of DBA/2J.C57BL/6J .129. AKR/J. BALB/cJ.C3H/HeJ [3,4]. By utilizing the strain-specific variations in atherosclerosis susceptibility, quantitative trait locus (QTL) analyses on different sets of intercross have identified more than 40 chromosomal loci that affect atherosclerotic plaque sizes [3,5,6,7,8]. Genetic studies in mice primarily focus on the very early stages of atherosclerosis, such as plaque initiation and growth at the aortic root, which complement GWAS where loci are primarily associated with advanced atherosclerosis or clinical vascular events. Thus, although most of the regions described in the mouse QTL mapping studies do not contain genes identified in human GWAS, several strong candidate genes have been proposed. These include Tnfsf4 on Chr 1 [9], Rcn2 on Chr 9 [10], Tnfaip3 on Chr 10 [11], Raet1e on Chr 10 [12], and Adam17 on Chr 12 [13]. Studies have also demonstrated that the bioinformatics approaches such as combined-cross analysis and haplotype analysis help reduce the number of the candidate genes [14]. Most QTL studies in mice have utilized the C57BL/6J strain due to their high susceptibility to atherosclerosis. However, systematic QTL analyses where strains other than C57BL/6J are utilized will be highly informative and can reveal new QTL and do provide an important data source for bioinformatics analyses. Furthermore, mouse studies compensate for human studies by controlling environmental factors and focusing on earlier stages of the progression of atherosclerosis. Recent expansion in genome information available for inbred strains of mice also provides a highly promising approach accelerating this process through genome comparison.
Previously, we have shown that apolipoprotein E-null mice on a C57BL/6J and a 129S6 genetic background (B6-apoE and 129-apoE) develop atherosclerosis differently in the aortic arch and the aortic root area: in the arches, 129-apoE mice form larger plaques than B6-apoE mice, whereas plaques in the roots are smaller in 129-apoE than in B6-apoE [4]. QTL analysis on the F2 generations of 129-apoE and B6-apoE cross identified significant QTL for the arch lesion on Chr 1 and for the root lesion on Chr 9, demonstrating that different genetic factors are likely contributing to the strain-and location-specific differences of plaque development [15]. In the present study, we carried out QTL analysis in an additional intercross between apolipoprotein E-null mice on a DBA/2J and those on a 129S6 genetic background, and identified two significant QTL for the root lesion; Ath44 on Chr 1 which overlaps with but appears to be independent from the previously identified Ath1 [16], and a novel locus Ath45 on Chr 2. We narrowed the candidate regions of each QTL using bioinformatics information.

Atherosclerotic Plaque Size and Plasma Lipids in the Parental, F1, and F2 Mice
Atherosclerotic plaque sizes in the aortic root area and plasma lipids were measured in parental 129-apoE and DBA-apoE, F1, and F2 mice (Table 1). Plaque sizes in the aortic root of male mice were approximately 9 times larger in parental DBA-apoE than in 129-apoE mice. Similarly, root plaque sizes in female mice were approximately 14 times larger in DBA-apoE than in 129-apoE mice. The striking differences in the root plaque sizes between DBA-apoE and 129-apoE mice are consistent with a previous report [3]. Neither DBA-apoE nor 129-apoE mice exhibited statistically significant sex-dependent differences. In the F1 and F2 mice, the plaque size distributions were intermediate between those in the parental strains. Since root lesion sizes and plasma lipids were not normally distributed, log-transformed values were used for QTL analysis (Table S1, Figure 1 and Figure S1).
DBA-apoE mice have lower plasma HDL-cholesterol and higher triglyceride levels than 129-apoE mice. Plasma total cholesterol levels were lower in DBA-apoE than in 129-apoE mice in males only (Table 1). While increased plasma lipid levels are potentially atherogenic, there is only a very small association between the lipid levels and lesion sizes in the F2 population ( Figure S2). Approximately 4% of the variation in the lesion size can be attributed to the total cholesterol (R 2 = 0.04, p,0.01) in males, while 5% of the variation in the lesion size can be attributed to HDL cholesterol (R 2 = 0.05, p,0.01) in females.

QTL Analysis for the Plaque Size in the Aortic Root
Genome-wide scans for QTL that influence atherosclerotic plaque sizes in the aortic root and plasma lipid levels were performed using 170 informative SNP genotypes of the 309 F2 mice (172 males and 137 females). We first performed single-locus scans for root lesions using sex as an additive or an interactive covariate ( Figure 2A). Significant LOD (Logarithm of the odds ratio) scores were 3.6 and 4.7 in the sex-additive (black line) and sex-interactive (red line) analyses, respectively. The detected QTL, LOD scores, 95% confidence intervals (CIs) and the variance of phenotype explained are summarized in Table 2. In the sexinteractive model, two highly significant QTL were detected on Chr 1 (LOD = 8.4, 158 Mb) and on Chr 2 (LOD = 9.4, 162 Mb). In addition, three suggestive QTL on Chr 7 (LOD = 3.5, 78 Mb), Chr 14 (LOD = 3.8, 56 Mb) and Chr 18 (LOD = 3.3, 53 Mb) were obtained ( Figure 2A and Table 2). The LOD score difference between the sex-additive model and the sex-interactive model (LODi) indicates, however, that Chr 7 (LODi = 2.5) and Chr 14 (LODi = 2.0) loci interact with sex ( Figure 2B). The LODi analysis also revealed an additional sex-specific locus on Chr 1 (LODi = 1.9, 75 Mb, Figure 2B). To examine a potential presence of additional QTL on Chr 1 and Chr 2, we calculated LOD score differences between the one-QTL and two-QTL models (DLOD) for each chromosome in the sex-interactive model. The LOD score differences showed not significant but non-negligible evidence for a second QTL at 75 Mb on Chr 1 (DLOD = 1.6). There was no observable evidence of a second QTL on Chr 2 (DLOD = 0.7). Together, our analysis revealed that two major QTL significantly affect the plaque size at the aortic root, one QTL on Chr 1 at 158 Mb and the other on Chr 2 at 162 Mb. Average size in the F2 mice homozygous for the DBA allele of the QTL on Chr 1 at 158 Mb was 2.5 times larger than those in the mice homozygous for the 129 allele of this locus, showing 129 dominance over DBA allele ( Figure 3A). We name this QTL Ath44. Although it overlaps with previously identified Ath1 [17], the identity is uncertain as we describe below. Similarly, average plaque size at the aortic root of F2 mice homozygous for the DBA allele of the QTL on Chr 2 at 162 Mb was 2.5 times larger than the plaque size of mice homozygous for the 129 allele of this locus. At this locus DBA and 129 show additive effects. This QTL also appears to be distinct from any of QTL previously mapped on mouse Chr 2: Athla1 [18], Ath28 [6], Ath39 [19] and Ath41 [8]. We name this novel QTL locus Ath45. As shown in Figure 4, Ath45 corresponds to human chromosome 20, where a locus 20q11 -13 has been linked with the incidence of acute coronary syndromes [20]. However, comparison between the mouse and human QTL did not reduce the interval of Ath45, since human 20q11 -13 locus spans almost 35 Mb (estimated interval: 28 -63 Mb) and include the homologous regions for Ath41 and Ath28.
We next performed QTL analysis on data from each gender to confirm the sex-specific QTL. Significant LOD scores were 3.5 and 3.6 in male-only and female-only analyses, respectively. The male-only scan detected two significant QTL on Chr 2 at 172 Mb (LOD = 5.9) and on Chr 14 at 73 Mb (LOD = 3.9), and one suggestive QTL on Chr 1 (LOD = 3.5, 171 Mb) ( Table 2 and Figure 2C, blue line). In the female-only scan, two significant loci, Chr 1 at 139 Mb (LOD = 4.8) and Chr 2 at 139 Mb (LOD = 3.9), and one suggestive locus on Chr 7 (LOD = 3.2, 78 Mb) were identified (Table 2 and Figure 2C, red line). DLOD on the female Chr 1 showed clear evidence for a second QTL on Chr 1 at around 73 Mb (DLOD = 3.0, arrow in Figure 2C), which was not present in the male Chr 1. The LOD curves indicate that multiple QTL may exist on Chr 2 and Chr 14 in males, and on Chr 2 in females; however, DLOD analysis for these chromosomes did not support this observation (DLOD = 0.8, 0.7, and 1.3, respectively). Thus, these sex-specific analyses: 1) identify a QTL on Chr 14 that is male-specific, 2) identify female-specific QTL on Chr 7 and Chr 1 and 3) are consistent with the results of sex-combined scan.
To better characterize the underlying genetic effect of the QTL reported above and to allow for possible epistatic interactions, we next performed genome-wide multiple QTL mapping allowing for a pair-wise scan. These results and the calculated proportion of the phenotype explained by regression analysis are shown in Table 3. In the sex-combined analysis, the best-fit model accounts for 27.9% of variation for the root lesion sizes. Chr 1 and 2 loci account for 8.2% and 8.8% of the total variance, respectively, and sex accounts for 1.9% of the total variance. The multiple-QTL model showed no evidence for significant interactions between the major effect QTL on Chr 1 and 2. In the male-only analysis, Chr 1, 2 and 14 loci account for 30.7% of the total variance and we detect no evidence of epistatic interactions. This contrasts with the results of our female-only analysis. In females, the multiple-QTL model revealed loci that were not identified by the single-locus genome scans, including QTL on Chr 3 at 21 Mb, Chr 12 at 77 Mb and Chr 17 at 67 Mb. The multiple-QTL model also revealed epistatic interactions in females between the second QTL on Chr 1 at 73 Mb and Chr 3 at 21 Mb, and between Chr 7 at 84 Mb and Chr 12 at 77 Mb. Investigation of the allelic effects of these loci identify that mice with the 129 allele at the Chr 1 locus have reduced plaque size when the animal has at least one 129 allele at the Chr 3 locus ( Figure 3B). Similar results were found on Chr 7, where mice with a DBA allele at the Chr 7 locus tends to suppress the plaque size when the animal has 129 allele at Chr 12 ( Figure 3B). The major effect loci (Chr 1 and 2) and sex specific and epistatic loci (Chr 1, 3, 12 and 17) jointly account for approximately 63.5% of the variance in females. The higher rate of phenotype variance explained was presumably due to more QTL mapped in females.

QTL Analysis for the Plasma Lipids
Plasma levels of lipids and lipoproteins are well-established risk factors for atherosclerosis. In the current cross the QTLs for atherosclerotic plaque size are independent of those for the plasma lipid levels except for Ath44, which overlaps with a QTL for plasma HDL-cholesterol on Chr 1 at 126 Mb (CI: 102 -174 Mb, LOD = 6.0) ( Figure S3 and Table S2). This locus may affect lesion development through a lipid-related pathway as the 129 allele of the HDL QTL is associated with higher plasma HDL and the 129 allele of Ath44 is protective for atherosclerosis. Two significant QTLs affecting total plasma cholesterol levels were identified by single scan analysis with sex as an interactive covariate on Chr 9 at 37 Mb (LOD = 8.6) and on Chr 11 at 83 Mb (LOD = 6.4) ( Figure  S3 and Table S2). Alleles of 129 at both loci conferred increased levels of total cholesterol. In contrast, at the QTL on Chr 6 at 88 Mb (LOD = 9.6), which affects plasma triglyceride levels in sexcombined and male-only analyses, DBA allele was associated with increased triglyceride levels. In addition, a locus on Chr 9 at 67 Mb was identified as a QTL for triglyceride levels in females (LOD = 5.5). None of these significant loci for plasma total cholesterol/triglyceride levels overlapped with the QTLs for the atherosclerosis.
Candidate Intervals of the Ath44 on Distal Chr 1 The Ath44 localized within confidence range between 152.9 -168.5 Mb (peak at 158.3 Mb) on Chr 1 significantly affects the lesion size at the aortic root and this region harbors more than 100 genes. To refine the candidate interval, we performed haplotype analyses in the region, followed by investigation of gene functions and gene expression levels, and further comparison with overlapping QTL.
Survey of nucleotide sequence differences in the Ath44 regions of Chr 1 between DBA/2J and 129S5, which is the closest relative to 129S6 substrain (see methods section) [21], indicates that DNA sequence of DBA differs from that of 129 extensively. Since no QTL affecting the root plaque size was present in the same region of the B6-apoE x 129-apoE cross, the simplest inference is that the candidate gene for atherosclerosis is likely to be within the blocks where 129 differs from that in DBA but shares the same haplotype with B6. Major blocks of haplotype sharing between 129S6 and C57BL/6 are 152 -153 Mb, 155 -157 Mb, and 160 -168 Mb, together containing approximately 84 genes with DBA-specific SNPs that could influence at the levels of transcription/translation ( Figure 5). For 50 genes with nonsynonymous SNPs, we studied each amino acid substitution for type of changes, position in the protein, homology with other vertebrate proteins, potential side chain modifications, and protein domain context. For example, flavin containing monooxygenase 3, coded by Fmo3, contains five amino acid substitutions unique to DBA. While four are benign changes, the substitution to Asn in DBA from highly conserved and negatively charged Asp at amino acid residue at position 76 in a NAD binding domain is predicted to be damaging when analyzed with Sorting Intolerant from Tolerant (SIFT) program with a score of 0.01 (Table S3). Similarly, there are seven amino acid substitutions between DBA and 129 E-selectin coded by Sele, and Ser201Phe in the first Sushi (CCP/SCR) domain was also predicted to be possibly damaging (SIFT score 0.04). We next looked for the genes relevant to vascular development or to atherosclerosis through knockout phenotypes and human GWAS association data. The 152.9 -168.5 Mb interval is equivalent to human Chr1q22-5. The 152 -153 Mb region contains Hmcn1 (hemicentin 1). Hemicentin 1 is implicated in agerelated macular degeneration in humans and markers within and near HMCN1 have been reported to be strongly associated with VLDL levels (-log 10 P = 13.9) and aortic atherosclerosis (8.9) [22].
The 155 -157 Mb interval is syntenic with Chr1q25, where an intron SNP marker in RGS8 is associated with cardiovascular disease (8.9) and SNPs near and within the CACNA1E gene are associated with HDL cholesterol levels (9.5), echocardiography (5.2) and blood pressures (4.9). The 160 -168 Mb interval contains genes including inflammatory genes (Tnfsf4, Tnfsf18, Fasl) at 163Mb, antioxidant genes (Fmo1 -3) at 165 Mb, and selectins (Sele and Selp) at 166 Mb. Differences in these inflammation related genes could influence early atherogenesis. Notable GWAS associations are TNFSF18 with HDL (9.9), DNM3 with platelet   .7), SELP with P-selectin level (60.0), and ATP1b1 with electrocardiography (15.0). By far the majority of GWAS data indicate that the common genetic variations affecting atherosclerosis-related vascular diseases alter gene expression levels rather than protein function. We therefore performed microarray analysis of aortic tissues from wild type mice to identify differentially expressed genes among the three parental inbred strains. Among the genes located in the 152.9 -168.5 Mb interval, the expression of Fmo1, Fmo2, Fmo3 and Sele was lower in DBA aorta than in B6 or 129 aortas, while the expression of Npl and ATP1b1 was higher (Table 4 and Figure S4). We also estimated gene expression levels in peritoneal macrophages from parental strains by exploratory microarray. In macrophages from DBA mice, the expression of Npl, Tnfsf4, and ATP1b1 was higher than in macrophages from 129 mice (Table 4 and Figure S5).
Since SNPs within the given haplotype blocks of interest could affect the expression of the genes at far distance away from them, we next screened aortic expression QTL (eQTL) data from the Hybrid Mouse Diversity Panel (HMDP) [23] for associations (Table S4). SNPs were selected according to the criteria: 1) located within and near the QTL interval, 2) associated with genes at P, 1.00E-06, and 3) having nucleotide sharing pattern of DBA ? 129 = B6. Majority of SNPs associated with the expression of genes peaked nearby to them (cis) confirming strain-specific differences in their regulations, although not all resulted in overt changes in aortic or macrophage expression in our experiments. Some SNPs affect the genes at .10 Mb away on the same chromosome, including Sox13, Ifi202, Ephx1 and Agfg1. Among these SNPs, rs31529590 at 158.1 Mb is strongly associated with the expression of Ifi202b at 175.9 Mb. On the other hand, there are multiple other cis-acting SNPs for Ifi202b, where nucleotides are shared in 129 and DBA but different in B6. This raises a possibility that two different regions regulate this gene, one unique in DBA and the other shared in 129 and DBA. Indeed, expression levels of Ifi202b are highest in DBA and lowest in B6. Additionally, the analyses revealed that SNPs within Ath44 interval appear to be affecting 16 genes on different chromosomes. Many of the trans-eQTLs peak at rs32041876 located within the Dnm3 gene where multiple noncoding short transcripts including mirRNAs have been detected. However, none of these associate genes showed DBA-specific expression in the aorta.
Collectively, we prioritized candidate genes underling Ath44 on Chr 1 by analyses of haplotype blocks, surveys of gene functions and expressions, and an analysis of an eQTL panel; 24 genes (Table 4) and additional 2 genes with eQTL (Table S4) were selected to be most likely responsible for plaques in the aortic root area.

Candidate Intervals of Ath45 on Chr 2
The DBA allele of Ath45 at 162.0 Mb (CI: 156.5 -165.1 Mb) is associated with larger aortic root plaques significantly in the DBA-apoE x 129-apoE cross. Our previous B6-apoE x 129-apoE cross showed a suggestive QTL at the same location (peak: 154 Mb, CI: 112 -164 Mb, LOD: 3.8) where B6 allele confers atherosclerosis susceptibility over 129 allele [15], suggesting that the locus is likely shared in DBA and B6. Consequently, regions where 129 haplotypes differ from those which are shared by DBA and B6 were further considered. The haplotype-sharing pattern of 129 ? DBA = B6 occurs in 42 genes ( Figure 5). The features of these genes analyzed using the same strategies described above are summarized in Table 5.
Although the syntenic region in humans has not been associated with atherosclerosis, the region (157.8 -158.3 Mb) includes the functional candidate Lbp, which encodes lipopolysaccharide binding protein. By interacting with CD14, Lbp plays key roles in innate immune signaling pathways, including toll-like receptor signaling pathway. Lbp protein of 129 contains Gly25Cys, and Tyr284His substitutions compared to the proteins in B6 and DBA. Gly-25 is a conserved residue at the signal peptide cleavage site and the change to Cys-25 could alter processing of the matured protein. Additionally, Tyr-284 is in the conserved lipid binding  Table 4. Haplotype analysis of Ath44 on Chr 1.   domain and its change to His-284 could alter its function. Each change is predicted to be deleterious with SIFT score of 0.01and 0.03, respectively (Table S3). Lbp also showed reduced expression in 129 macrophages compared to DBA and B6 (Table 5 and Figure S5), although further confirmations are required.
In the 160 -164 Mb interval, an Asn22Ile substitution in the Ift52 gene coding for an intraflagellar transport protein essential for the formation and maintenance of flagella and cilia is potentially damaging with a SIFT score of 0.03 (Table S3). Asn-22 is a conserved residue within an ABC-type uncharacterized transport system domain, which is found in various eukaryotic and prokaryotic intra-flagellar transport proteins. Pkig, coding for protein kinase inhibitor gamma, showed lower aortic expression in 129 than in DBA. Pkig protein of 129 also has two amino acid substitutions compared to Pkig proteins in B6 and DBA, Gly52Ala and Ser70Asn. Both Gly-52 and Ser-70 are highly conserved during evolution. Gly52Ala substitution is possibly damaging (SIFT score 0.08), while Ser70Asn might be benign (Table S3). Aortic expression of Ada, which encodes adenosine deaminase, is expressed more than two times higher in 129 than in DBA and B6. ADA plays an important role in purine metabolism and adenosine homeostasis, thereby modulates signaling of extracellular adenosine and T-cell activation. While ADA deficiency causes autosomal recessive SCID (severely combined immunodeficiency) due to toxic effects of substrate accumulation, a lower activity of the enzyme appears to be associated with reduced incidence of CAD in humans [24] and may be beneficial in controlling intimal formation as SMC-derived adenosine can inhibit cell proliferation in culture [25]. Higher Ada gene expression in the aorta of 129 mice (Table 5 and Figure S6), however, cannot simply account for their reduced susceptibility to atherosclerosis. Additionally, although no nonsynonymous SNPs or gene expression difference were observed, human GWAS has indicated strong associations of a missense mutation and an intron variant in HNF4a with plasma HDL levels (-log 10 P = 15.0) and type 2 diabetes (-log 10 P = 9.5), respectively. Additionally, decreased angiogenesis was observed in mice lacking syndecan 4 coded by Sdc4 [26].
Finally, eQTL data of HMDP panels provided 6 additional genes in the Ath45 interval, expression of which is associated with SNPs at P,1.00E-06, having nucleotide sharing pattern of 129 ? B6 = DBA (Table S5). Of these 6 genes, Plac9 on chromosome 14 and Exosc1 on chromosome 19 were affected in trans. Microarray data showed Dynlrb1 and Plac9 have lower expression in 129 compared to DBA and B6, although their association with atherosclerosis has not been reported. Together, we reduced the candidate interval of Ath45 on Chr 2 according to haplotype blocks, gene functions and gene expression levels, suggesting 16 genes and additional 2 genes with eQTL as the putative candidates for the root atherosclerosis.

Discussion
In the present study, we identified multiple QTLs that significantly affect atherosclerotic plaque size in the root area in an F2 population derived from a cross between DBA-apoE and 129-apoE. We carried out genome analysis of two QTLs, Ath44 and Ath45, that are significant in male and female combined analyses, by combining the current QTL information with information obtained from our previous B6-apoE x 129-apoE cross. Ath44 on Chr 1 overlaps with Ath1, but they are likely to be distinct, while Ath45 on the Chr 2 is a novel locus.
A large number of loci have been mapped in mice for atherosclerotic plaque development, and some of these loci overlap. However, whether these loci are identical or discrete is difficult to determine since experiments carried out in different laboratories have used different designs, such as husbandry, types of diet, use of sensitizing mutations, sex and lineage of cross, which likely affect the quantitative phenotype [27,28]. QTL information from crosses using apoE-deficient mice in six different combinations of seven laboratory inbred backgrounds have been reported including our current work; four of them involves C57BL/6J partnered with C3H/HeJ, BALB/cJ, FVB/NJ or 129S6, and the remaining two involves DBA/2J, partnered with 129S6 or AKR/ J. While the information derived from these multiple crosses should help define the regions responsible for the QTL, there are several notable discrepancies. The best example is Ath1on Chr 1. Ath1 was originally identified to affect diet-induced atherosclerosis in the aortic root of the female mice of recombinant inbred strains between C57BL6/J and C3H/HeJ, and between C57BL6/J and BALB/cJ [17]. Tnfsf4 (Ox40l) at 163Mb has been suggested as the gene underlying this locus [9]. Tnfsf4 encodes a ligand of OX40, which is a member of the tumor necrosis factor (TNF) receptor superfamily. In females, the mRNA level of Tnfsf4 was 3.7 times higher in atherosclerosis-susceptible C57BL/6J mice than in atherosclerosis-resistant C3H/HeJ mice [9], and the Tnfsf4-OX40 signaling is proposed to affect the susceptibility of atherosclerosis by modulating T cell proliferation [9]. Ath1 was subsequently mapped at 158 Mb (CI: 135 -161 Mb) in the analysis of the F2 population from C57BL/6J-apoE x C3H/HeJ-apoE intercross [29], and at 186Mb (CI: 178 -190 Mb) in C57BL/6J-apoE x BALB/cJ-apoE intercross [8]. In both of these crosses, mice were fed a Western-type diet containing high fat/high cholesterol, and Ath1 was not replicated in chow-fed C57BL/6J-apoE x C3H/HeJ-apoE [30] nor in a cross of LDLR 2/2 mice on C57BL/6J and BALB/cByJ backgrounds fed a diet low in cholesterol [19]. Ath1, therefore, appears to be restricted to studies where a cholic acid containing diet was used to exacerbate atherosclerosis. Thus crosses that use genetic susceptibility in the form of apoE 2/2 and LDLR 2/2 , or cholic acid-containing high fat/high cholesterol diets may yield different loci, reflecting distinct susceptibility alleles and underlying mechanisms.
The Ath44 overlaps with Ath1. However, while the B6 allele at the Ath1 locus conferred increased atherosclerosis compared to both the C3H/HeJ and BALB/cJ alleles, the DBA allele of Ath44 was associated with increased atherosclerosis compared to either the B6 or 129 alleles. Although the variations within or near Tnfsf4 could be responsible for the increased atherosclerosis susceptibility of DBA over 129 in our cross, they cannot explain both Ath1 and Ath44 since DBA shares an identical DNA sequence in and near Tnfsf4 with the atherosclerosis-resistant BALB/cJ mice, while 129S6 and C3H/HeJ share the same sequence with the atherosclerosis-susceptible C57BL/6J allele. In addition, Ath44 was detected in both male-only and female-only analysis in our study, whereas the effect of Tnfsf4 variance appears to be femalespecific. Furthermore, our DBA-apoE x 129-apoE F2 mice were on regular chow. It appears likely that multiple loci within 150 -170 Mb are affecting root atherosclerosis, and that one or more of these loci could be interacting with either a diet containing high cholesterol or cholic acid.
Several other groups have performed crosses of apoE-deficient mice on low cholesterol diets and not replicated the Ath1 locus. For example a DBA-apoE x AKR-apoE cross did not detect a significant QTL near Ath1 [6,31], and only one of the two groups of B6-apoE x FVB-apoE cross analyzed detected a QTL distal to the Ath1 locus at 178 Mb [5]. The haplotype sharing pattern of the interval 157.5 -159 Mb, where both FVB/NJ and AKR/J share DNA sequence with DBA2 but differ from 129 and B6, is consistent with these observations. Furthermore, while C3H/HeJ and BALB/c also share haplotypes with DBA/2J not with C57BL/6J in this region, but there is no indication that their alleles cause larger plaques than the C57BL/6J allele. Thus, these crosses of apoE 2/2 mice suggest, although not conclusively, that 157.5 -159 Mb is an unlikely candidate region. Instead, the genetic changes responsible for increased plaque size in the root of DBA mice are more likely to be within the intervals where DBA sequence uniquely differs from others. The region containing a cluster of flavin containing monooxygenase genes, Fmo1, Fmo2 and Fmo3 is particularly interesting because these genes are highly expressed in aortic tissues, their expression is significantly low in DBA compared to 129 and B6, and Fmo3 contains a potentially damaging amino acid substitution. A recent report identified a SNP in DBA mice that affects an enhancer site for farnesoid X receptor (Fxr/Nr1h4) binding [32]. Although this may not be the causal SNP for atherosclerosis, these results suggest a possible mechanism for altered Fmo3 expression [32]. Whether and how reduced function of these Fmos affects increased risk for plaque size at the aortic root in DBA require further investigation, since Rgs8, Glul, Cacna1e, Prrx1, Sele, Selp and Atp1b1 that affect vascular health are also within these regions.
As to the Ath45 on Chr 2, the SNPs within and near most of the genes in the 157 -164 Mb interval are unique to 129 and other strains all share SNPs with DBA, which is consistent with the absence of obvious QTLs observed in crosses of apoE-null mice on other strain combinations than those involving 129-apoE. One of the primary candidates is Lbp, in which two amino acid substitutions predict a reduced function of the 129 protein compared to those in DBA and B6. Athero-protective role of the 129 Lbp is supported by the observed association between serum Lbp levels and CAD in humans [33] and welldocumented links among chronic infections, inflammation, and atherosclerosis, as well as roles of LBD, CD14 and TLR in atherosclerosis [34]. Although further evidence is required, Pkig is another primary candidate in the 160 -164 Mb interval, which encodes c-AMP dependent protein kinase inhibitor gamma. While its role in vascular disease is not yet to be determined, Pkig is expressed in many tissues including heart and endothelial cells [35,36]. Our observation of a 30% reduction in the Pkig expression in aortic tissues of 129 mice compared to DBA mice as well as amino acid changes in the highly conserved residues of 129 Pkig protein compared to DBA proteins warrants further examination of their functional relevance. Hnf4a, coding for hepatic nuclear factor 4, is also located in Ath45, although expression difference in parental strains was not detectable and no amino acid changes are present in the protein sequences. As a transcription factor, Hnf4a regulates expression of many genes in atherosclerosisprone aortas of mice [37]. Two QTL for plasma triglyceride, Tgq31 and Tgq32 [38], and two QTL for HDL cholesterol, Hdlq19 [39] and Hdlq70 [38], have been mapped to the overlapping region with Ath45 and Hnf4a has been proposed as an HDL cholesterol associated gene [40]. However, we did not detect any QTL for plasma TG or HDL cholesterol on Chr 2 in the present 1296DBA intercross. Thus, if Hnf4a is contributing to the athero-protection in the aortic root of 129-apoE mice, its role is likely to be independent of the mechanisms that control plasma lipid concentration.
While comparative genome analysis of inbred mouse stains is a powerful approach, there are limitations in our study. One is that the confidence interval estimated by the Bayesian method could be too stringent. For example, Ath45 peak in males extends further to the end of the chromosome and conservative estimation of confidence interval with one-LOD drop suggests that Ath45 could overlap with Ath28 previously found in a DBA/2J and AKR/J cross [6]. If each QTL contains multiple causative genes, the haplotype sharing pattern may be more complex than the assumption we made in our analyses. Another limitation is that the Sanger genome project includes complete genome sequence of 129S1 and 129S5 but 129S6. However, 129S6 strain was derived in 1992 from a cross between 129S5 and 129/SvEv-Gpi1 c , both established in 1987-1988 from a colony of 129S1 (129/SvEv) maintained by M.J. Evans since 1969 [21]. The genome of each inbred laboratory strain of mice is a unique mosaic of old polymorphisms between the parental chromosomes from different subspecies of Mus musculus. Fixation of new variations in the inbred strains during the last 100 years or so since their derivation has been minimum [41]. Thus the probability is small that a new nucleotide change arose in these chromosomal regions of our interest, has been fixed in the 129S6 population, and is functionally relevant. Nevertheless, possible presence of SNPs recently arose in the parental lines of apoE-deficient strains cannot be eliminated completely.
Our current study also revealed the presence of genderspecific QTLs. The QTL on Chr 14 is significant only in males and the 129 allele confers the susceptibility over DBA. It peaks at 73 Mb with a broad confidence interval of 21 -106 Mb, and co-localizes with Ath13 that was previously identified in males of B6-apoE x FVB-apoE cross [5]. Our QTL may well be identical to Ath13 since both were detected in the male-only and the sex-combined analyses but not in the female-only scans. Female-specific epistatic interactions between a second locus on Chr 1 at 73 Mb and Chr 3 at 21 Mb, and between Chr 7 at 84 Mb and Chr 12 at 77 Mb have not been reported and require further investigations. Nevertheless, the QTL on the Chr 1 at 73 Mb may well be the same as Ath30, which was identified by B6-apoE x C3H-apoE cross [29] and was also female-specific. A proteoglycan desmin encoded by Des has been implicated in atherosclerosis in humans [42] and has also been proposed as a candidate gene underlying Ath30 [23]. Similarly, Chr 7 locus co-localizes with QTL, Ath3 [43] and Aorls2 [44] previously identified in wild type female mice fed an atherogenic diet from crosses between B6 x DBA/2, A/J x B6, respectively. It also overlaps with and may well be identical to Ath3, which was identified as a female-specific QTL in a B6-apoE x C3H-apoE cross [29]. We note that the QTL region on Chr 7 contains imprinting domains including one equivalent to Prader-Willi/Angelman syndromes gene cluster on human Chr 15. Epigenetic interactions make comparative genome analysis difficult since the absence of QTL in a certain cross no longer provides simple information. Nevertheless, even though each locus accounts for a small portion of the phenotype variations, accumulations and interactions of such minor factors substantially contribute to the individual susceptibility to atherosclerosis. The interactions between QTLs and the differences between males and females could also provide important keys to identify the underlying genes.
In conclusion, with the use of publicly available mouse genome information and published data of crosses of apoE null mice on various genetic backgrounds, we suggest that the Ath44 and Ath45 are likely to be in 155 -157 Mb or 165.3 -168 Mb intervals on Chr 1 and in 160 -164 Mb on Chr 2. Candidate genes of 25 or so in each locus were further reduced to suggest several genes for future investigation of how these loci influence atherosclerosis susceptibility at the aortic root of mice. These genes include, Fmo3, Sele, and Selp for Ath44, and Lbp and Pkig for Ath45, and multiple SNPs within the equivalent regions in human genome have been shown to associate with cardiovascular phenotypes. While determining the causalities still requires a long process, and the causative genes in mice and humans may differ, searching genetic variations that influence atherosclerotic plaque development with multiple inbred strains of mice would likely lead to the mechanistic studies of known associations as well as to identification of novel genes/pathways involved in the pathogenesis of atherosclerosis. This information in turn should have important implications for prevention and treatment of atherosclerosis.

Animals
ApoE-deficient mice on a 129S6/SvEvTac genetic background (129-apoE) were generated in our laboratory as previously described [4,15]. ApoE-deficient mice on a DBA2J background (DBA-apoE) were obtained from the Jackson Laboratory (#007067, D2.129P2 (B6)-apoE tm1Unc /J). Male 129-apoE mice were mated to female DBA-apoE mice to generate F1 hybrids, which were intercrossed to generate F2 progeny. Mice were fed with a regular mouse chow (Prolab IsoPro RMH 3000; Agway Inc) and handled in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health under protocols approved by the Institutional Animal Care and Use Committees of the University of North Carolina at Chapel Hill (Protocol Number: 11-028).

Phenotyping
Atherosclerotic plaque sizes were measured as previously described [4,15]. Briefly, mice at 4 months of age were anesthetized with a lethal dose of 2,2,2-tribromoethanol, perfused with 4% paraformaldehyde, and the heart and the aorta were dissected. Atherosclerotic lesions in the roots were evaluated by sectioning the proximal aorta and the heart containing the aortic root. Plaque areas were measured on the captured images using Image J 1.43 software (http://rsb.info.nih.gov/ij/). Plasma levels of total cholesterol and triglycerides were measured using commercial kits from Wako Chemicals USA (Richmond, VA) and Stanbio Laboratory (Boerne, TX), respectively. HDL cholesterol levels were determined after removing apoB containing lipoproteins with magnesium/dextran sulfate [45].

DNA Isolation and Genotyping
Genomic DNA was isolated from the livers and purified using DNeasy Tissue Kit (Qiagen, Hilden, Germany), followed by quantification using PicoGreen dsDNA Assay Kit (Molecular Probes, Eugene OR, USA). Single nucleotide polymorphism (SNP) genotyping was performed using the Illumina BeadArray technology with Mouse Low Density Linkage SNP panels, which is an optimized set of 377 SNPs covering the 19 autosomes and the X chromosome and includes approximately four SNPs per each 27 Mb interval across the entire mouse genome.

QTL Analysis
We performed QTL analyses using R/qtl program (v.1.26-14, http://www.rqtl.org) [46,47]. The main effect QTL were identified by genome-wide single-QTL scans using the 'scanone' function with standard interval mapping. Genotype probabilities between markers were computed with 1 cM intervals and with a genotyping error rate of 0.001. The significance thresholds were determined by 10,000 permutations. QTL were considered significant if LOD scores exceeded 95% (p,0.05) of the permutation distribution; they were considered suggestive if the scores exceed 37% (p,0.63) distribution. 95% confidence intervals (CIs) for QTL positions were calculated by Bayesian credible interval function of the software. To determine sexspecific QTL, main scans were performed with sex as an additive covariate or as an interactive covariate. If the LOD score difference between the two models (LODi) was greater than 2, the sex-QTL interaction was considered to be significant [48]. We also performed QTL analysis in males and females separately. The presence of two QTL on the same chromosome was assessed by comparing two models: a model with one QTL and a model with both QTL. If the LOD score difference (DLOD) was greater than 2, two QTL were considered to be present on the same chromosome [48]. For each QTL, model of inheritance was determined according to allelic effect at the nearest marker of a QTL by performing linear regression using the additive and dominant/recessive models.
Multiple-QTL models were detected by the 'stepwiseqtl' function of R/qtl, which perform a forward/backward stepwise search algorithm to select the best-fitting model. Penalties for the model comparison corresponding to a = 0.1 were calculated using the result of 1,000 permutations with the 'scantwo' function. The proportion of variance explained and LOD score under the model were calculated using the 'fitqtl' function. A multiple imputation was used for the analyses. All the genetic positions (cM) were converted to physical positions (Mb, NCBI Build 37) using the Mouse Map Converter (http://cgd.jax.org/ mousemapconverter/).

Microarray Analysis
Aortas were dissected from 3-to 5-month-old male mice. Peritoneal macrophages were isolated from 3-month-old mice of both sexes 4 days after the intraperitoneal injection of 1 ml thioglycollate. Tissues/macrophages from 4-5 mice were pooled and total RNA was extracted using an RNeasy Mini Kit (QIAGEN), according to manufacturer's protocol. RNA quality was assessed by an Agilent Bioanalyzer. Total RNA (250 ng) was used to synthesize fragmented and labeled sense-strand cDNA and hybridize onto Affymetrix arrays. The Affymetrix HT WT User Manual was followed to prepare the samples. Briefly, the WT Expression HT Kit for Robotics (Ambion) was used to generate sense-strand cDNA from total RNA. Following synthesis of sense-strand cDNA, the cDNA was fragmented and labeled with the Affymetrix GeneChip HT Terminal Labeling Kit. The Beckman Coulter Biomek FXP Laboratory Automation Workstation with the Target Express set up was used to prepare the samples with these two kits. Fragmented and labeled cDNA was used to prepare a hybridization cocktail with the Affymetrix GeneTitan Hybridization Wash and Stain Kit for WT Arrays. Hybridization, washing, staining and scanning of the Affymetrix array plate was carried out using the Affymetrix GeneTitan MC Instrument. GeneChip Command Console Software (AGCC) was used for GeneTitan Instrument control. Affymetrix Expression Console Software was used for basic data analysis and quality control. All the samples were processed using a single Mouse Gene 2.1 ST 24-Array Plate (Affymetrix), except for comparison between 129 and DBA macrophages, which was carried out at a different time using Mouse Gene 2.0 ST Array (Affymetrix). The array data was deposited to Gene Expression Omnibus (GEO) Database (accession number GSE53006).

Comparative Genomics and Haplotype Analysis
SNPs and nucleotide sequence comparisons of mouse strains were obtained from publicly available resources (http://www. sanger.ac.uk/resources/mouse/genomes/, http://useast.ensembl. org/Mus_musculus/, and http://www.informatics.jax.org/). San-ger genome project includes whole genome sequence of 129S1 and 129S5 but 129S6. However, 129S6 strain was derived in 1999 from a cross between 129S5 and 129/SvEv-Gpi1 c , two substrains of a 129S1 strain maintained by M.J. Evans (129/SvEv) since 1969 [21]. No nucleotide differences were detected in the relevant regions. Consequently, we used 129S5 sequence to represent 129S6 sequence in our comparisons. The haplotype patterns of genomes in the QTL interval were analyzed using UNC Mouse Phylogeny Viewer (http://msub.csbio.unc.edu/) and Perlegen Mouse SNP Browser (http://mouse.cs.ucla.edu/perlegen/). The likelihood that an amino acid change is detrimental to a protein was examined by Sorting Intolerant From Tolerant (SIFT) program (http://sift.jcvi.org/) [49]. Information of human GWAS data was obtained from NHGRI GWAS catalog (http://www. genome.gov/gwastudies/). Gene expression data in the aorta of mice in the Hybrid Mouse Diversity Panel (HMDP) [21] was obtained from GEO Database (accession number GSE38120). SNPs were selected from the eQTL data of the HMDP according to the criteria: 1) located within and near the QTL interval, 2) associated with genes at P,1.00E-06, and 3) having nucleotide sharing pattern of DBA ? 129 = B6 (Ath44) or 129 ? B6 = DBA (Ath45). Clusters of SNPs linked were represented by one with smallest P value, followed by the shortest distance to the start of the gene when it is on the same chromosome. All the genomic locations are in Mb.

Statistical Analysis
Data were analyzed using JMP software version 8.0 (SAS Institute, Cary, NC). Comparisons of multiple groups in trait values and gene expression levels in the aorta were done by oneway analysis of variance (ANOVA) followed by Tukey-Kramer's HSD test. The distributions of phenotypes in F2 mice were assessed for normality by Shapiro-Wilk test and Kolmogorov test. For gene expressions in macrophages, the genes in Ath44 or Ath45 did not differ by sex. The fold differences were therefore the average of male and female samples. At least two independent pools of samples for each genotype were analyzed and their average was used to estimate relative ratios and level of gene expression for the present study.   Effects of amino acid (AA) substitutions were predicted by SIFT (Sorting Intolerant From Tolerant) program. SIFT scores show the probability that an amino acid change is damaging, ranged from 0 to 1. AA substitutions with SIFT score #0.05 are predicted to be deleterious; substitutions with SIFT score .0.05 to be tolerated.

(DOCX)
Table S4 SNPs within the Ath44 interval associated with gene expressions in the aorta. SNPs within the 152.9 -168.5 Mb interval of Chr 1, having nucleotide sharing pattern of 129 = B6 ? DBA and associated with genes at P,1.00E-06 were selected from the eQTL data of the Hybrid Mouse Diversity Panel (HMDP) [21]. Those in linkage disequilibrium are represented by a single SNP with lowest P-value and nearest to the associated gene represents. Distance is the position of SNP site relative to the start position of the gene. For each SNP, expression levels of the associated genes in the aorta and macrophages are shown as ratios of two strains. Genes that show DBA-specific expression are bolded. a p,0.05, b p,0.01, c p,0.001.

(DOC)
Table S5 SNPs within the Ath45 interval associated with gene expressions in the aorta. SNPs within the 156.5 -165.1 Mb interval of Chr 2, having nucleotide sharing pattern of 129 ? B6 = DBA, and associate with genes at P,1.00E-06 were selected from the eQTL data of the Hybrid Mouse Diversity Panel (HMDP) [21]. A SNP with lowest P-value and nearest to the associated gene represents those in linkage disequilibrium. Distance is the position of SNP site relative to the start position of the gene. For each SNP, expression levels of the associated genes in the aorta and macrophages are shown as ratios of two strains. a p,0.05, b p,0.01, c p,0.001. Genes that show 129specific expression are bolded. (DOC)