Novel loci for hyperglycemia identified by QTL mapping of longitudinal phenotypes and congenic analysis

We previously reported that four hyperglycemia loci are located on three chromosomes in the Nagoya-Shibata-Yasuda (NSY) mouse model, commonly used to study type 2 diabetes. However, we did not search for hyperglycemia loci across all chromosomes. In this study, we performed quantitative trait loci (QTLs) mapping of longitudinal phenotypes from crosses between NSY (hyperglycemic) and C3H (normoglycemic) mice. We identified four new QTLs for hyperglycemia, namely Nidd5nsy, Nidd6nsy, Nidd1c3h, and Nidd2c3h, on Chromosome 1, 4, 10, and 13, respectively. These QTLs were associated with hyperglycemia in young mice and had attenuated effects in older mice. Nidd5nsy and Nidd6nsy were hyperglycemic with NSY alleles, and Nidd1c3h and Nidd2c3h were hyperglycemic with C3H alleles. We further bred Nidd5nsy congenic mice and demonstrated that Nidd5nsy has a strong effect on hyperglycemia when young, accompanied by insulin resistance and visceral fat accumulation. These results showed that the effects of individual QTLs strengthened or weakened with age, and that the sum of the effects of QTLs captured the age-related deterioration of glucose tolerance in individuals. Our results support the importance of longitudinal phenotypes in the genetic analysis of polygenic traits and have implications for the genetic basis and pathogenesis of type 2 diabetes in humans.

www.nature.com/scientificreports/ at 48 weeks of age, explaining approximately half of the genetic variance 10 . Subsequent studies using consomic mice possessing diabetogenic Chr11 or Chr14 from NSY mice on the genetic background of control C3H mice clearly demonstrated that the genes responsible for hyperglycemia were located on Chr11 and Chr14. However, double consomic mice possessing both Chr11 and Chr14 from NSY mice did not exhibit the diabetic phenotypes of NSY mice 11,12 , indicating that additional genes are responsible for the remaining genetic variance. In our initial study, we identified the loci Nidd1nsy-Nidd4nsy using a two-step approach 10 . Namely, a subpopulation of F2 mice (93 out of 307 mice) exhibiting extreme phenotypes in glucose tolerance were subjected to initial genome screening with a full panel of markers, and three significant chromosomes (Chr11, Chr14, and Chr6) were further analyzed in the whole population of F2 mice. However, this approach prevented the analysis of the remaining chromosomes in detail. Therefore, the aim of this study was to identify new QTLs, besides those on Chr11, Chr14, and Chr6, by analyzing the QTL mapping data of the whole F2 population (307 mice) with sufficient genetic markers and longitudinally obtained phenotypes. Furthermore, we aimed to produce a congenic strain possessing the target chromosomal region containing the newly identified loci and analyze the glucose tolerance and other metabolic traits of these mice longitudinally.

Results
Novel hyperglycemia loci detected by QTL analysis. QTL mapping of the F2 progeny (n = 307) using the R/qtl package identified four novel loci for diabetes on Chr1, Chr4, Chr10, and Chr13, in addition to reconfirming the previously identified Nidd1nsy-Nidd4nsy (Tables 1, 2, Fig. 1, and Supplementary Table S1). In contrast to the previously identified Nidd1nsy on Chr11, which was linked to hyperglycemia at all ages, the effects of the newly identified loci were age-dependent, with significant linkages observed at younger ages, but not older ages (Table 1, Fig. 1, and Supplementary Table S1). The NSY-derived alleles for loci on Chr1 and Chr4 were positively linked to hyperglycemia. Meanwhile, for loci on Chr10 and Chr13, the C3H-derived alleles were positively linked to hyperglycemia. Therefore, loci on Chr1 and Chr4 were designated Nidd5nsy and Nidd6nsy for non-insulin-dependent diabetes mellitus loci 5 and 6 in NSY mice, respectively. However, the loci on Chr10 and Chr13 were designated Nidd1c3h and Nidd2c3h for non-insulin-dependent diabetes mellitus locus 1 and 2 in C3H mice, because they caused hyperglycemia in C3H mice and the NSY allele had a protective effect against hyperglycemia.
Confirmation of newly identified Nidd5nsy on Chr1 by congenic strain analysis. To directly demonstrate the existence of Nidd5nsy on Chr1 and localize and characterize the gene, we constructed a congenic strain possessing the NSY-derived Nidd5nsy region of Chr1 on the control C3H background. The constructed congenic mice, C3H.NSY-Nidd5nsy, carried NSY-derived alleles for Chr1 from D1Mit305 to D1Mit269, including the region of interest ( Supplementary Fig. S1).
The results from ipGTT and body weights of congenic C3H.NSY-Nidd5nsy and C3H mice are shown in Fig. 2. At 24 weeks of age, glucose levels were significantly higher in C3H.NSY-Nidd5nsy mice than in C3H mice at all time points post glucose challenge (P < 0.01) (Fig. 2b). gAUC values were also significantly higher in C3H. NSY-Nidd5nsy mice than in C3H mice (P < 0.001) (Fig. 2d). At 12 weeks of age, glucose levels at 120 min post glucose challenge were significantly higher in C3H.NSY-Nidd5nsy mice than in C3H mice (P < 0.05); however, no significant difference was observed at 36 weeks of age, indicating that Nidd5nsy had an age-dependent effect on hyperglycemia. There were no significant differences in body weight between C3H.NSY-Nidd5nsy and C3H mice at any age ( Fig. 2e).
At 24 weeks of age, the most significant differences were observed in the glucose tolerance test, so the diabetesrelated phenotype of C3H.NSY-Nidd5nsy mice was compared to that of C3H mice at 26-28 weeks of age (Table 3 and Supplementary Fig. S2). Insulin secretion (at 26 weeks of age), assessed by the insulinogenic index after glucose challenge, was similar for C3H.NSY-Nidd5nsy and C3H mice. HOMA-IR, an index of insulin resistance, was significantly higher for C3H.NSY-Nidd5nsy mice than for C3H mice (P < 0.05). Additionally, in a separate group from the ipGTT considered for dissection (at 28 weeks of age), C3H.NSY-Nidd5nsy mice exhibited significantly higher body weights (P < 0.01), BMI (P < 0.05), and visceral fat weights (P < 0.05) than C3H mice, suggesting that increases in body fat and insulin resistance were the main causes of hyperglycemia induced by Nidd5nsy. The development of hyperglycemia in the face of insulin resistance with normal insulin secretion suggests the failure of insulin secretion to compensate for insulin resistance. which was mapped to the region near D4Mit219 (95% CI: 31-122 Mbp), was strongly linked to glucose levels at 60 min post glucose challenge (MLS: 3.46, P < 0.05) and gAUC (MLS: 3.40, P < 0.05) at 24 weeks of age (Table 1 and Fig. 1). Nidd6nsy explained 5.2% of the phenotypic variance in glucose levels at 60 min post glucose challenge (Table 2). F2 mice with NSY-derived alleles at D4Mit219 had significantly higher gAUC values than those without these alleles (P < 0.01) (Supplementary Table S1). The Nidd6nsy locus was not linked to glucose levels at 12, 36, or 48 weeks of age. No markers tested on Chr4 were linked to body weight at any age nor epididymal fat pad weight collected at 52 weeks of age. The Nidd1c3h locus on Chr10, which was mapped to the region near D10Mit230 (95% CI: 56-88 Mbp), was strongly linked to blood glucose levels at 0, 30, 60, and 90 min post glucose challenge at 12 weeks of age (MLS: 3.89, 5.17, 3.25, 3.43 and P: < 0.05, < 0.01, < 0.05, < 0.05, respectively) ( Table 1). Consequently, gAUC was strongly linked to this locus (MLS: 4.46, P < 0.01) ( Table 1 and Fig. 1). F2 mice with NSY-derived alleles at D10Mit230 had significantly lower, not higher, gAUC values (P < 0.01) (Supplementary Table S1). NSY-derived alleles at  www.nature.com/scientificreports/ D10Mit230 were not linked to glucose levels at 24, 36, or 48 weeks of age. Additionally, NSY-derived alleles at D10Mit230 were not linked to body weight at any age nor epididymal fat pad weight collected at 52 weeks of age. The Nidd2c3h locus on Chr13, which was mapped to the region near D13Mit134 (95% CI: 11-39 Mbp), was only linked to blood glucose levels at fasting glucose at 12 weeks of age (MLS: 4.41, P < 0.01) ( Table 1). F2 mice with NSY-derived alleles at D13Mit134 had a significantly lower glucose value at 0 min (P < 0.01) ( Table 2).
Confirmation of previously identified hyperglycemia loci: Nidd1nsy-Nidd4nsy. Nidd1nsy-Nidd4nsy, which were identified in our previous study using only 68 markers with the MAPMAKER/QTL program 10 , were reconfirmed in this study using more markers with the R/qtl package with the same data set as before. Although the Nidd1nsy-Nidd4nsy loci detected in our previous study may not be exactly the same as the present ones, the peaks of the previously detected QTLs are located within the 95% CI of the QTLs detected in the present study. Nidd1nsy on Chr11 in the region near D11Mit242 (95% CI: 39-51 Mbp) was linked to glucose levels at 12, 24, 36, and 48 weeks of age (Table 1 and Fig. 1), similar to the results of our previous study 10 . Nidd1nsy was the locus most strongly linked to hyperglycemia at 24, 36, and 48 weeks of age, and the third most strongly linked locus at 12 weeks of age. Nidd1nsy was not linked to body weight at any age nor epididymal fat pad weight at 52 weeks of age. Nidd2nsy was reconfirmed on Chr14 (Table 1 and Fig. 1). The LOD peak was located near D14Mit59 (95% CI: 21-89 Mbp) for hyperglycemia and near D14Mit160 (95% CI: 22-80 Mbp) for epididymal fat pad weight. Nidd2nsy was not linked to body weight at any age. Nidd3nsy was reconfirmed on Chr6 (Table 1 and Fig. 1). The LOD peak for glucose levels was located near D6Mit135 (95% CI: 124-138 Mbp), while that for epididymal fad pad weight was located near D6Mit54 (95% CI: 94-127 Mbp). Nidd3nsy was strongly linked to epididymal fat pad weight at 52 weeks of age (MLS: 6.93, P < 0.01) ( Table 1). Nidd3nsy was not linked to body weight at any age. Nidd4nsy was identified in the centromeric region of Chr11 (95% CI: 12-52 Mbp) on the same chromosome but at a different position from Nidd1nsy. Nidd4nsy was strongly linked to glucose levels at 12 weeks of age (Table 1 and Fig. 1) and was not linked to body weight at any age.

Discussion
This study identified four novel hyperglycemia QTLs, including Nidd5nsy on Chr1, Nidd6nsy on Chr4, Nid-d1c3h on Chr10, and Nidd2c3h on Chr13. For Nidd5nsy and Nidd6nsy, NSY-derived alleles predisposed mice to developing hyperglycemia, while for Nidd1c3h and Nidd2c3h, the C3H-derived allele predisposed mice to hyperglycemia. Further, these QTLs were linked to hyperglycemia at younger ages, namely, transient hyperglycemia. The previously identified hyperglycemia loci Nidd1nsy-Nidd4nsy on Chr6, Chr11, and Chr14 were also Table 2. Summary of hyperglycemia QTLs in NSY with C3H mice. * 1 95% confidence interval (CI), corresponding to a 1.8-LOD support interval. * 2 Physical positions were estimated from marker positions. * 3 Mode of inheritance of the NSY allele was determined using the degree of dominance, which is the ratio of the dominance effect to the additive effect (see "Materials and methods" section). * 4 Number of genes in the 95% CI was estimated using the Mouse Genome Database (http:// www. infor matics. jax. org). Values were compared using one-way ANOVA with post hoc test (Bonferroni): *P < 0.05, **P < 0.01 vs CC; † P < 0.05, † † P < 0.01 vs NC. www.nature.com/scientificreports/ www.nature.com/scientificreports/ reconfirmed in this study. The effect of Nidd5nsy, the QTL most strongly linked to hyperglycemia among the newly identified loci, was clearly demonstrated in the congenic study. Moreover, the congenic study clarified that Nidd5nsy harbors a locus that causes hyperglycemia, insulin resistance, insufficient insulin secretion to compensate for insulin resistance, and visceral obesity. The effects of four novel QTLs were age-dependent, with significant linkages detected in young ages which diminished at older ages. Generally, diabetes persists or worsens with age unless therapeutic intervention with diet and exercise is provided after onset. Therefore, a single QTL was expected to cause a similar phenomenon, i.e., persistence or worsening with age. However, the longitudinal phenotypes in this study demonstrated that the effects of some QTLs strengthened while others weakened with age, suggesting that summation of the QTL effects resulted in age-related deterioration of glucose tolerance in the individual.
Several questions arose during our study. First, QTL mapping indicated that Nidd5nsy on Chr1 was most strongly linked to hyperglycemia at 12 weeks of age, whereas congenic mice exhibited only mild hyperglycemia at 12 weeks of age and significantly increased blood glucose levels at 24 weeks of age. In addition, an increase in body weight, which was not evident in QTL mapping, was detected in the congenic analysis. A possible reason for this discrepancy is that the effect of Nidd5nsy may depend on interactions with other genes of the NSY donor line, which were lost during the construction of the congenic line. In other words, Nidd5nsy is affected by gene-gene interactions with age-dependent and -independent effects, resulting in overall phenotypes of type 2 diabetes in NSY mice that vary with age. Polygenic disorders, such as type 2 diabetes and obesity, are affected by multiple gene-gene and gene-environment interactions [13][14][15] . Thus, it is necessary to consider that genes with age-dependent effects may influence these interactions. Second, the increase in visceral fat mass was not evident in QTL mapping, but was detected in the congenic analysis. One of the possible reasons for this discrepancy may be the difference in phenotyping age: 52 weeks in QTL mapping and 28 weeks in the congenic study. Another possibility is that the congenic mice were affected by Nidd2nsy on Chr14 and Nidd3nsy on Chr6 from C3Hderived alleles, which increased visceral fat 16,17 .
This study mapped a new QTL for hyperglycemia, Nidd5nsy on Chr1. The human syntenic region corresponding to the peak LOD score is Chr1q24.2-32.1. This region contains loci for type 2 diabetes that have been reported in several studies listed in the GWAS catalog (https:// www. ebi. ac. uk/ gwas/ home) (Supplementary Table S2) [18][19][20][21][22][23][24] . Although there are still many genes in this region, and identifying susceptibility genes or disease-causing variants is a formidable challenge, these syntenic regions in mice and humans provide important information about common pathways and regulatory genes in both species.
In mouse studies, several QTLs for blood glucose were reported in the region identified in this study. A QTL on distal Chr1, Nob3, has been identified in genome-wide scans of diabetic NZO with non-diabetic C57BL/6J 25 . On Chr4, some QTLs have been identified in NZO with SJL mice (Nidd/SJL) 26,27 , in NZO with several mouse lines 28 , and in NZO with DBA mice (Nidd/DBA) 29 . On Chr10, a QTL has been identified in SM/J and A/J recombinant inbred strains (t2dm1sa) 30 . On Chr13, no QTL has previously been identified in the region of interest. The data in the present study together with previous reports point to the importance of these chromosomes in conferring susceptibility to hyperglycemia.
Nidd1c3h and Nidd2c3h are distinct from other QTLs in that the NSY-derived alleles from diabetes-prone mice are protective, while the C3H-derived alleles from normoglycemic mice confer susceptibility to type 2 diabetes. Since type 2 diabetes is a polygenic disease and each locus was identified in crosses of two strains by the relative strength of the locus on each phenotype, it is possible that some genes may induce opposite effects. For example, previous studies have reported contradictory effects of Nidd2nsy and Nidd3nsy on visceral fat weight; alleles from disease-prone NSY mice decreased visceral fat pad weight, whereas alleles from disease-resistant C3H mice increased visceral fat pad weight 16,17 . Such loci effects were also previously reported in NOD mice, a www.nature.com/scientificreports/ well-known animal model of another polygenic disease, type 1 diabetes [31][32][33] . Loci inducing the opposite effects may arise as revertants, mutations that reverse pathological processes in disease models. The genes and gene products of such loci provide important information for establishing effective methods of disease prevention and intervention. This study had some limitations. First, the current NSY-derived Nidd5nsy congenic region is approximately 85 Mb, which contains more than 500 genes, making it difficult to examine candidate genes at this time. Furthermore, functional studies suggest that hyperglycemia in Nidd5nsy congenic mice is due to both insulin resistance and a compensatory defect in insulin secretion in response to insulin resistance, suggesting that two or more genes may be involved rather than only one. To identify susceptibility genes, more regionally restricted congenic analysis is necessary. Second, in this study, we performed QTL analysis across all chromosomes and found several new QTL, each of which accounted for at most 10% of the phenotypic variance, and collectively could not explain the entire phenotypic trait. This suggests that gene-gene and gene-environment interactions may be involved and need further investigation. Third, we used the R/qtl package in the current study 34 , which differed from that used in our previous studies 10,17 , because the previous program could not be used. Despite differences in programs, the results for Nidd1nsy-Nidd4nsy were almost identical for both studies, indicating the accuracy of the methods used and validating the new QTLs obtained in this study.
The study findings indicate that type 2 diabetes in NSY mice is regulated by the summation of novel QTLs identified in this study and previously reported QTLs (Chr6, Chr11, and Chr14), with each QTL exerting different age-related effects. Furthermore, this study demonstrates the importance of longitudinal analysis of phenotypes during the genetic analysis of polygenic traits.

Materials and methods
Animals. NSY mice 6 were obtained from the Branch Hospital of Nagoya University School of Medicine (Nagoya, Japan). C3H/HeNcrj (C3H) mice were purchased from Charles River Laboratories (Kanagawa, Japan). All mice (including NSY, C3H, F1, F2, and congenic mice) were bred at Osaka University Medical School, housed in an air-conditioned room (22-25 °C) with a 12 h light-dark cycle (6:00-18:00 h), and provided tap water and standard rodent chow diet (Oriental Yeast, Tokyo, Japan) ad libitum. Male mice were used in all experiments. The animal study protocol was approved by the Osaka University Graduate School of Medicine Ethics Committee and complied with the ARRIVE guidelines. All experiments were performed in accordance with relevant guidelines and regulations.
Construction of F2 mice and phenotypic characterization. As previously described 10,17 , F1 mice were produced by crossing an NSY male and a C3H female, and F1 mice were intercrossed to produce F2 mice. F2 mice (n = 307) underwent an ipGTT (2 g glucose/kg body weight) after overnight fasting, and blood glucose levels were measured at 0, 30, 60, 90, and 120 min at 12, 24, 36, and 48 weeks of age. Blood glucose levels were measured directly by the glucose oxidase method using Glutest E (Sanwa Kagaku Kenkyusho Co. Ltd., Nagoya, Japan). gAUC was calculated according to the trapezoid rule. F2 mice were euthanized by intraperitoneal administration of pentobarbital sodium (Dainippon, Osaka, Japan) at 52 weeks of age. The epididymal fat pads of F2 mice (n = 221) were dissected and weighed. The phenotypes of F2 mice are shown in Supplementary Fig. S3.
Genotyping of F2 mice and QTL analysis. Genomic DNA, extracted from the liver or tail using the standard phenol-chloroform method, was used as a template for polymerase chain reaction (PCR). Microsatellite marker information was obtained from the Mouse Genome Database (https:// www. infor matics. jax. org). A total of 107 microsatellite markers were used to cover the whole genome (Supplementary Table S3, Supplementary Fig. S4). The markers were amplified by PCR with non-labeled or labeled primers. Non-labeled PCR products were electrophoresed on 9% polyacrylamide gel and visualized by ethidium bromide staining. Labeled PCR products were electrophoresed on 4% denaturing polyacrylamide gel using an ABI Prism 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA) with GeneScan 350 ROX (Applied Biosystems) as an internal lane size standard. The PCR fragments were measured using GeneScan Analysis software, version 3.5 (Applied Biosystems) and genotyped using Genotyper software, version 3.6 (Applied Biosystems).
Normality tests were performed for all phenotypic traits, and for traits that deviated significantly from normality, Box-Cox transformations were used to bring them as close to normal as possible. QTL analysis was performed using these trait values. QTL analysis by interval mapping was performed using the R/qtl package, using multiple imputation approach to complement missing genotype data [34][35][36] . An LOD score was calculated at a 1 cM step within each interval across the linkage map constructed. The 95% CI was obtained by a 1.8-LOD drop method. Threshold LOD values were obtained from multiple imputation approach at α = 0.05 and α = 0.01, by applying 1000 permutations for each trait. The mode of inheritance of the QTL was determined by using the degree of dominance, which is the ratio of the dominance effect to the additive effect, as described by Kenney-Hunt et al. 37 . QTL analysis was performed on 307 mice for glucose and body weight and 221 mice for epididymal fat pads.

Construction of a congenic strain and phenotypic characterization.
A novel congenic strain (C3H.NSY-Nidd5nsy) was constructed with the genomic region identified in QTL analysis using a markerassisted speed-congenic method 38 , as previously described 11,12,16,39,40 . Namely, (NSY × C3H) F1 male mice were backcrossed with female C3H mice. Males that were heterozygous for Nidd5nsy were selectively mated with female C3H mice and male progeny that were heterozygous for Nidd5nsy were used for the next generation. This process was repeated until all markers in the background genome were homozygous for the C3H-derived