Ultrafine Mapping of SNPs From Mouse Strains C57BL/6J, DBA/2J, and C57BLKS/J for Loci Contributing to Diabetes and Atherosclerosis Susceptibility

DNA for mouse strains B6 (JAX strain 000664), BKS (JAX strain 000662), and DBA (JAX strain 000671) were purchased from The Jackson Laboratory (Bar Harbor, ME). DNA from BKS-db (JAX strain 000642) was isolated from tissues using standard phenol/chloroform extraction procedures.

A total of 15,300 single nucleotide polymorphisms (SNPs), covering the genome at an average spacing of ∼167,000 bp (autosomes) and 330,000 bp (X chromosome), were analyzed. SNP analysis was carried out at Illumina (San Diego, CA), using their BeadArray technology (9,10). This involves allele-specific extension from the genomic DNA template followed by PCR amplification of the extended allele. The genotype is determined by the ratio of fluorescent signals from the two allele-specific extension products.

To determine whether SNPs that are informative between B6 and DBA are randomly distributed among all SNPs tested, Runs tests were carried out using the Dataplot program (11) provided by the National Institute of Standards and Technology.

We previously described (12) an F2 cross between B6 and DBA mice, in which liver mRNAs were analyzed for expression of 23,574 genes (13). Linkage maps were constructed, and quantitative trait locus (QTL) analysis was performed using MapMaker QTL (14,15) and QTL Cartographer (16). Logarithm of odds scores were calculated at 2-cM intervals throughout the genome for each of the 23,574 genes. Using these data, eQTLs controlling expression of these genes were located. Significant (logarithm of odds >4.3) eQTLs were identified for 2,123 genes. Those genes, whose expression levels mapped to within 20 Mb of the location of the gene itself, were identified as cis-acting eQTLs.

SNP genotyping was carried out on DNA from B6, DBA, BKS, and BKS-db. Of 15,300 SNPs assayed, 12,473 gave genotype data with sufficient quality across a large strain set to be included in the dataset. Based on the rate of heterozygosity, we estimate an error rate of ∼0.08%. After removing all SNPs that did not have an identified position in build 32 of the public genome database as well as all those that gave no data or heterozygous genotypes for any of the four strains discussed here, 10,517 SNPs remained. These markers were spaced at an average of 2 × 10⁵ bp on the autosomes and at 5 × 10⁵ bp on the X chromosome. The distributions of total SNPs and informative SNPs for each chromosome are shown in Fig. 2 of the online appendix (available at http://diabetes.diabetesjounrals.org). The SNPs covered the entire length of each chromosome at high density, except for regions within 2–4 Mb of the centromere. Beyond these centromeric regions, there were two notable gaps in SNP coverage on the autosomes. The last 10 Mb of chromosome 7 contained only four SNPs, and, on proximal chromosome X, there was a gap of 5.3 Mb between 22 and 28 Mb. With these exceptions, the maximal interval between SNPs was 2.2 Mb on the autosomes and 2.8 Mb on chromosome X.

For the analysis presented here, we selected the subset of 4,710 SNPs that are polymorphic between the B6 and DBA. The distribution of such informative SNPs is significantly nonuniform, with obvious long stretches largely devoid of informative SNPs. For example, on chromosome 1, as shown in Fig. 1 and 2 of the online appendix, there is a large gap (8–10 Mb) in informative SNPs from 140 to 150 Mb. Such regions have been previously noted by others (17) and have been hypothesized to represent blocks of DNA inherited from a common ancestral strain. For instance, on chromosome 10, only 10 of 197 SNPs between 21 and 84 Mb were informative. This same region was noted by Wiltshire et al. (17). Strikingly, on the X chromosome, only 12% of the markers were polymorphic between B6 and DBA, suggesting a large portion of this chromosome is derived from a common ancestor.

Even outside these large obvious regions of low informativeness, there is significant clustering of informative and uninformative markers as judged by tests based on the binomial distribution (11). Thus, there is a significant overrepresentation of clusters containing greater than three uninterrupted informative or uninformative markers between B6 and DBA. This is consistent with retention in modern inbred strains of discernable blocks, both large and small, of ancestral genomes. While the presence of large gaps in the set of informative SNPs lowers the resolution with which we may map the blocks of introgressed DBA DNA in BKS, it is also less likely that there is genetic variation in these regions contributing to differences in the BKS phenotype.

Application of these informative SNPs to identify regions of BKS derived from DBA is shown in Figs. 1 and Table 1. The middle of Fig. 1 shows the locations for all SNPs on chromosome 1 that are informative between B6 and DBA. Horizontal lines on the right side show regions of BKS that carry the DBA alleles for these informative SNPs. Thus, between 33.2 and 70.8 Mb on chromosome 1, all 70 SNPs showed a genotype consistent with this region of BKS deriving from DBA. By contrast, all informative SNPs between 70 Mb and the telomere at ∼200 Mb carried the B6 allele. Table 1 shows the location of all regions in BKS carrying uninterrupted blocks of DBA alleles. These large DBA-derived regions are present on all chromosomes except 2 and 15 and in total represent ∼20% of the genome. Interestingly, the largest DBA-derived block appears on the X chromosome. This chromosome appears to be >80% from DBA.

As previously shown, the BKS strain carries some genomic regions derived from strains other than B6 or DBA (2,8). We have termed these regions as “non-B6–non-DBA.” The fine-scale SNP genotyping described here allows us to closely map the boundaries of these regions. For example in Fig. 1, on proximal chromosome 1 between 3.5 and 22.5 Mb, the SNP genotype rapidly alternated between B6 and DBA alleles. This could derive from a high degree of recombination between B6 and DBA alleles but more likely results from introgression of a DNA segment derived from an unrelated strain. Consistent with this, we observed a high frequency of SNPs interspersed in the same interval that, while not informative between B6 and DBA, nevertheless shows the alternate allele for the SNP in question. These SNPs are labeled “Unique in BKS” on the left side of Fig. 1. The distributions of DBA-like SNPs and BKS-unique SNPs for each chromosome are shown in Fig. 1 of the online appendix.

There are a total of 144 BKS-unique SNPs in the BKS genome. Throughout, the BKS-unique SNPs are most frequently associated with regions where the informative SNPs rapidly alternate between B6 and DBA alleles. In some cases, such as on proximal chromosome 9, the BKS-unique SNPs are concentrated in regions with a low density of SNPs informative between B6 and DBA. Both of these circumstances are consistent with introgression of genomic fragments from an unrelated strain. We have termed these general genomic regions as non-B6–non-DBA to indicate their unknown genetic origin. We use the term “BKS-unique” for individual SNPs whose origin is clearly from a strain other than B6 or DBA.

In no case do we see significant runs of BKS-unique SNPs that are interspersed with purely B6-like or DBA-like regions and well separated (>2–3 Mb) from a B6-DBA transition. The non-B6–non-DBA regions are summarized in Table 2. In total, these regions constitute 8–9% of the BKS genome.

In an attempt to identify the origin of the non-B6–non-DBA regions, we compared alleles for BKS-unique SNPs among a panel of DNAs from 56 other inbred strains (Cervino et al., unpublished data). There was no strong match for this set of SNP alleles other than with BKS-db (Fig. 3 in the online appendix). Of the strains tested, the best match was with 129P3J and other strains of the 129 series. For these strains, 98 of 144 BKS-unique SNPs match BKS. However, while strain 129 matches this specific subset of SNPs in some regions, the overall match between 129 and BKS in the same regions is not high. Similarly, there is no strong match in this set of strains across the complete set of SNP alleles from the non-B6–non-DBA regions listed in Table 2 (data not shown). Petkov et al. (8) suggested that a mouse strain related to BTBR may have contributed these DNA regions, but we observe a close match to BTBR only on part of chromosome 4 (data not shown). Thus, the origin of the non-B6–non-DBA regions in BKS is undetermined. A match may be found in a larger panel of inbred strains, or the contaminant may derive from an outbred or wild strain.

Historically, the db mutation arose on the BKS genetic background, and most early phenotypic data for the diabetes mutation were derived from this strain. The original BKS-db strain is no longer available. The BKS-db breeding stock currently available from The Jackson Laboratory is heterozygous for the db mutation (because homozygotes are infertile), and as a convenience for breeding, the heterozygous db mutation at 46.7 cM is maintained in opposition to a coat color mutation in the closely linked misty (m) locus at 46.1 cM (Fig. 2). Moreover, the mutant misty gene derives from DBA and was first transferred to B6 (18) and subsequently bred into the BKS-db line (E. Leiter, personal communication). Thus, the current heterozygous breeding stock for BKS-db represents a second-generation congenic with the potential for contamination with genomic regions originating in both B6 and DBA. In congenic strains, the introgressed segment always carries additional sequence beyond markers used to select the congenic segment. In addition, because congenic construction begins with outcrossing between donor and recipient strains, additional regions of the donor genome beyond that of the intended congenic segment frequently become introgressed in the final congenic strain. We used the high-density SNP genotype data to address these issues in the homozygous BKS-db genome.

The leptin receptor gene occurs at ∼100.6 Mb on chromosome 4 in a region where there are a number of non-B6–non-DBA SNPs in the BKS genome (Fig. 2). In fact, for BKS chromosome 4, a large block from ∼85 to ∼130 Mb shows the characteristic SNP pattern of non-B6–non-DBA DNA. This suggests that the original db mutation occurred on a region of the genome derived from a strain other than B6 or DBA. The SNPs proximal to and in the immediate vicinity of the leptin receptor gene are identical between BKS and BKS-db. However, distal to the leptin receptor gene, from ∼102 to ∼110 Mb, the original BKS sequences have been replaced with DNA carrying DBA alleles (indicated by the bar in Fig. 2). This block of DBA alleles in BKS-db is consistent with residual DBA contamination carried along with the misty congenic region derived from DBA. Sequence differences in this congenic contamination may impact the phenotype driven by the db mutation. Distal to this block, only two SNPs, at 111.0 and 111.6 Mb on chromosome 4, distinguish BKS and BKS-db. SNP analysis also reveals a short block of five SNPs on chromosome 15 from 93.9 to 95.8 Mb that were DBA-like in BKS but have become B6-like in BKS-db (Fig. 3). Thus, both possible contaminants from the misty congenic are now fixed in the BKS-db genome. Strikingly, no other contaminating regions associated with construction of the BKS-db congenic were detected by the fine-scale SNP analysis, although additional segments may have been transferred in either regions where the SNPs are not informative between B6 and DBA or smaller insertions between informative SNPs.

In the presence of the leptin receptor mutation, BKS exhibits a progression into full type 2 diabetes accompanied by extensive atherosclerosis. Strikingly, the diabetic and atherosclerotic phenotypes in BKS-db are significantly more severe than in B6-db with, for instance, an ∼100-fold increase in lesion area. While rates of lesion progression and underlying differences in plaque histology and stability have not been widely explored, the basic effect is assumed to result from susceptibility loci conferred by the introgressed segments of DBA genome (4,19). To map the underlying loci, Mu et al. (4) performed a genetic cross between B6-db and BKS and then analyzed female F2 offspring carrying the homozygous leptin receptor mutation. They identified a significant QTL for atherosclerosis on chromosome 12 and suggestive QTLs for plasma glucose on chromosomes 8 and 17 and for plasma HDL cholesterol on chromosomes 5 and 17.

The atherosclerosis locus has been further localized to a subcentimorgan interval on proximal chromosome 12 (19), spanning 12.4–13.2 Mb. Comparison of this locus with the SNP genotype data presented here for BKS (Table 1) shows that it is contained within a short block of DBA-derived genomic sequence that extends from 10.4 to 19.5 Mb.

To further narrow the list of candidate genes, we have attempted to identify genes that are differentially regulated between B6 and DBA and whose expression is regulated in cis. We previously identified cis-acting eQTLs in an F2 intercross between strains DBA and B6 (12). The genes listed in Table 3 were selected on the following criteria: 1) Logarithm of odds score for the eQTLs was greater than 4.3, 2) physical map location of the gene is within 20 Mb of the eQTL peak, and 3) physical map location of the gene is within a DBA block introgressed in BKS. The r² value indicates the percentage of expression variation for the gene attributable to the eQTL at this locus. (Online appendix Table 3 is a complete version of Table 3, which only shows chromosome 12.) That these genes are regulated in cis strongly suggests that local sequence variation within or near the gene affects expression differences observed among animals in the B6 × DBA cross.

Thus, the genes listed in Table 3 are likely candidates for any phenotypic differences introduced by the DBA blocks within BKS. This greatly focuses the attempt to identify candidate genes. For instance, there are two DBA-like regions on chromosome 12 in BKS covering a total of ∼55 Mb. The National Center for Biotechnology Information MapViewer identifies ∼430 genes in these combined regions. For the same regions, we identify only 18 cis-acting eQTLs meeting our criteria. To further focus on those genes specifically related to diabetes susceptibility, we measured correlation of each gene’s expression with diabetes-related traits in the same animals. These traits included plasma levels of insulin and lipids (triglycerides, free fatty acids, and total, unesterified, and HDL cholesterol), measures of obesity (body weight, total fat pad weight, and weights of individual fat depots), and atherosclerosis (fatty streak lesions in the aortic arch). Genes whose expression showed significant correlation (P < 0.05) with one of these traits are shown in bold in Table 3, and the specific correlated phenotype is noted.

These cis-regulated genes will be useful in identifying candidates for genes influencing the BKS phenotypes. For instance, genetic crosses between B6 and BKS were analyzed by Coleman and Hummel (7) revealing non-Mendelian inheritance of diabetes susceptibility and suggesting that multiple loci contribute to the phenotype (7). Coleman and colleagues (20,21) identified variations in cytosolic malic enzyme activity between BKS and B6 and speculated that they may play a role in diabetes susceptibility. Malic enzyme catalyzes the NADP⁺-dependent oxidative decarboxylation of malate to pyruvate and CO₂, producing the NADPH required for de novo fatty acid synthesis. Coleman and Kuzava’s (20) studies showed that the B6 strain carries the high-activity allele of the malic enzyme regulator (Mod1r), whereas BKS carries the low allele and that the Mod1r locus segregates in part with diabetes susceptibility. Coleman and collegues (20,21) mapped the Mod1r locus to proximal chromosome 12. Examination of the cis-regulated genes on proximal chromosome 12 (Table 3) shows a number of possible candidates for this susceptibility, most of which are uncharacterized genes. One particularly interesting candidate in this region is lipin, a gene known to modulate adiposity, insulin resistance, and atherosclerosis (22,23). The cis regulation of lipin, as detected in expression array data the B6 × DBA cross, is consistent with the fivefold differential expression of the mRNA that we observe between B6 and BKS by quantitative PCR (Fig. 4). Moreover, the lipodystrophy observed in mice carrying a natural mutation in lipin and the increased adiposity seen in muscle- and adipose-specific transgenics (24) strongly suggest that variation in mRNA levels has a corresponding impact on lipin activity. While lipin expression was not correlated with diabetes-related traits in the B6 × DBA cross, it is quite possible that the observed cis-regulated variation in lipin expression only becomes relevant in the context of obesity driven by leptin receptor deficiency or that the insulin resistance/diabetes phenotype results from interaction of lipin expression differences with other loci within the non-B6–non-DBA regions of BKS.

Of course, this use of expression array data will only detect candidate genes whose expression is regulated in cis in liver and not genes whose function is altered in the absence of a corresponding shift in expression. Moreover, these expression data are derived from a B6 × DBA cross and are only appropriate for identifying candidates for BKS phenotypic variation in regions of the BKS genome that derive from DBA.

The identification of genetic factors contributing to the diabetes phenotype in strain BKS is an important goal that may reveal new pathways involved in the disease. We now report the characterization of genomic differences between BKS and B6. We also survey cis-acting genetic variations between B6 and DBA at these loci. These results should be useful for the further mapping and, ultimately, positional cloning of genes underlying susceptibility to diabetes- and atherosclerosis-related traits.

The identification of cis-acting genetic variants in liver is an exciting strategy for narrowing the list of candidate genes in the BKS-db phenotype. However, while liver is likely to be a major organ impacting susceptibility to insulin resistance and diabetes in BKS, it is important to pursue the impact of genetic variations in other tissues as likely contributors to these phenotypes. Also, it will be important to verify the involvement of specific genomic regions of BKS using congenic strains. Such congenics will provide an ideal platform for subsequently investigating the effects of specific candidates and pursuing mechanistic and protein expression studies.

Numerous regions of BKS appear to have been derived from strain DBA (Table 1), and a smaller number are of unknown origin (Table 2). Lueders (3) provided the first evidence for DBA-like regions in BKS by studying strain-specific proviral loci. Naggert et al. (2) subsequently used 321 microsatellite markers polymorphic between B6 and DBA to genotype the BKS strain. They observed that several DBA clusters were observed in BKS, but because the markers were widely and unevenly spaced, some clusters were not identified and the remainder poorly defined (2). Recently, Petkov et al. (8) reported an analysis of the BKS genome using a panel of 1,636 SNPs. From this data, they identified 6 of the 19 non-B6–non-DBA blocks that we report here. Furthermore, the finding that DBA-db mice exhibit a diabetes-susceptible phenotype similar to BKS-db suggests that most or all diabetes susceptibility loci in BKS are of DBA origin (25).

The density of our SNP mapping is not sufficient to detect very short non-B6 regions in the BKS strain. For instance, although all SNPs tested on chromosome 2 showed the B6 allele, previous microsatellite mapping has shown a DBA allele for D2Mit1 at 3.80 Mb in BKS (2). This is consistent with the presence of a small fragment of DBA genome (<0.44 Mb) inserted between flanking SNP markers rs422932 at 3.45 Mb and rs3680350 at 3.88 Mb, both of which show the B6 allele in BKS. It is also possible that the D2Mit1 genotype in BKS recently evolved independently of DBA.

More recently, Mu et al. (4) examined the relationship between diabetes and atherosclerosis in the BKS-db strain. They constructed a cross between BKS on the db background and, following the feeding of an atherogenic cholic acid–containing diet, observed striking differences in lesion size, glucose levels, and HDL levels. Interestingly, no relationship was observed between glucose levels and lesion size. These studies (4) led to the identification of a major gene locus for atherosclerosis designated Ath6 located on chromosome 12 in a DBA-like cluster. Interestingly, the peak of this QTL occurred at ∼17 Mb, coincident with the Lipn1 gene described above.

The same study identified suggestive QTLs for plasma glucose levels on chromosomes 8 and 17 and for plasma HDL cholesterol levels on chromosomes 5 and 17. The plasma glucose QTLs were identified near D8Mit195 (83.4 Mb) and D17Mit24 (36.0 Mb), both corresponding to DBA-like regions in BKS. In both cases, the DBA-like region contains a number of strongly cis-regulated genes including likely regulatory factors such as a zinc finger protein on chromosome 8 (Zfp617) and the E2F transcription factor on chromosome 17, in addition to several uncharacterized genes that may underlie the QTLs.

These studies (4,21) did not reveal evidence for sex-linked inheritance of diabetes or atherosclerosis susceptibility, and therefore the extensive DBA-like regions on the X chromosome are unlikely to be involved in the diabetes susceptibility.

Attie and colleagues (26–30) carried out detailed studies of diabetes susceptibility, using genetic segregation, congenic strains, and expression array analysis, in a cross between B6 and BTBR on an obese (ob) background. They identified loci contributing to adiposity, lipid metabolism, and insulin/glucose levels differing between BTBR (diabetes susceptible) and B6 (diabetes resistant). However, none of these loci, with the exception of an obesity locus on chromosome 13, map within the DBA- or non-B6–non-DBA–derived chromosomal regions of BKS. Thus, although the phenotypic differences observed in the BTBR × B6 and BKS × B6 crosses are similar, separate genes likely determine these traits. Even so, there may be significant overlap in the overall pathways and mechanisms leading to diabetes susceptibility in these two models. In particular, Attie and colleagues have suggested that suppressed hepatic lipogenesis may be central to the relative diabetes susceptibility of BTBR-ob, and our ongoing studies show that BKS-db has strikingly suppressed hepatic lipogenesis relative to B6-db (data not shown).

As described above, in a cross between strains DBA and B6 (12), we used expression array data to identify eQTLs for several thousand genes. As shown in Table 3, about one-third of these eQTLs mapped over the corresponding structural gene and are therefore likely to be cis acting. We propose that this type of data will be very valuable in prioritizing candidate genes underlying the increased diabetes susceptibility observed in BKS-db. However, there are some hypotheses implicit in this proposal that merit validation. First, we assume that mRNA variation measured by array reflects real differences in mRNA levels, as might be determined by more quantitative methodology. The validity of mRNA measurements by array is usually checked only on a small scale by investigators pursuing specific candidate genes. Recently however, systematic assessment of this issue showed 70% correlation between Agilent array data and rigorous quantitative PCR for a panel of 100 mRNAs selected from a wide set of abundance classes (31,32). Moreover, when the same arrays are used in fluor-reversed pairs, as was done for the B6 × DBA cross, the correlation greatly improves (33), in our experience reaching 90%.

A second hypothesis is that mRNA variation mapping to the gene location (i.e., a cis-regulated gene) reflects a local sequence variation that differentially modulates the mRNA level in an allele-specific manner. Such cis regulation is what we would expect for genetic variation underlying a phenotypic QTL. To test this hypothesis, we used a direct test of cis regulation. Among the genes whose expression was predicted to be cis regulated in the B6 × DBA cross, we selected a set of 29 genes containing SNPs within the coding region. We then isolated liver mRNA from (B6 × DBA) F1 animals. Such animals are heterozygous at all loci, and their cells will contain a mixture of B6- and DBA-specific transcripts. Genes that are regulated in trans should have equal transcript levels from each allele. By contrast, genes that are differentially regulated in cis will show a corresponding enrichment in transcripts from one of the two alleles. Taking advantage of the SNPs in these genes, we used direct sequencing of RT-PCR products to determine the relative transcript levels from each allele. Cis regulation was confirmed in 19 of 29 genes (34).

To provide investigators with potential candidates underlying the DBA regions in BKS, we have catalogued the strongest cis-acting variations in Table 3. As discussed above, there is high correlation of this expression array data with more quantitative measures of mRNA levels, and, furthermore, independent cis-trans tests confirm that genes in Table 3 show strong differences in allele-specific transcript levels, strongly suggesting true cis regulation. However, investigators pursuing specific QTLs will need to validate their most promising candidate genes at the level of protein mass or activity. Because of significant posttranscriptional effects for many genes, overall mRNA levels from array data show only 40–75% correlation with the corresponding protein (35,36). Also, it should be noted that the eQTLs were not identified on db background, and thus there are probably differences in gene regulation between BKS-db and B6-db that were not observed in our cross. Nevertheless, most such cis-acting variations are likely to be at least partly independent of the genetic background or the state of obesity and diabetes of the mice.

National Institutes of Health Grants HL28481, HL60030, and HL70526 supported this work.