Chronic kidney disease (CKD) is a growing medical problem and a significant risk factor for the development of end-stage renal disease, cardiovascular disease, and cardiovascular mortality. The genetic basis of CKD is recognized, but knowledge of the specific genes that contribute to the onset and progression of kidney disease is limited, mainly because of the difficulty and expense of identifying genes underlying CKD in humans. Results from genetic studies of CKD in rodents often correspond to findings in humans; therefore, we used quantitative trait locus (QTL) analysis to detect genomic regions affecting albuminuria in a cross between C57BL/6J and DBA/2J mice, strains resistant and susceptible to CKD, respectively. We identified several independent and interacting loci affecting albuminuria, including one QTL on mouse chromosome (Chr) 2 that is concordant with QTL influencing urinary albumin excretion on rat Chr 3 and diabetic nephropathy on human Chr 20p. Because this QTL was identified in multiple mouse crosses, as well as in rats and in humans, we used comparative genomics, haplotype analysis, and expression profiling to narrow the initial QTL interval from 386 genes to 10 genes with known coding sequence polymorphisms or expression differences between the strains. These results support the continued use of multiple cross-mapping and cross-species comparisons to further our understanding of the genetic basis of kidney disease.
- quantitative trait loci
chronic kidney disease (CKD) represents a significant medical and economic burden in the United States, affecting roughly 25 million people. The number of patients presently undergoing treatment for end-stage renal disease is nearly 500,000, but this number is projected to soar to 2.2 million patients by 2030 because of the increase in individuals with diabetes and minority populations who demonstrate high risk for kidney disease (49). CKD is also a risk factor for cardiovascular disease, including myocardial infarction, atherosclerosis, stroke, and hypertension (24, 33, 40, 41). Mortality from myocardial infarction, stroke, and coronary artery disease increases with elevated urinary albumin concentrations (30, 32, 38), and the link between CKD and cardiovascular disease is supported by numerous additional studies (7, 9, 15, 18, 22, 23, 29). Therefore, understanding the causes of kidney disease onset and progression is important to improve diagnosis and treatment of both renal and cardiovascular diseases.
The genetic basis of CKD is well established [for review, see Bowden (1) and Korstanje and DiPetrillo (27)], and identifying genes that influence its onset and progression will likely improve our understanding of the disease. Identifying genes underlying CKD in humans is difficult and expensive (1); however, genetic analyses of CKD in rats and mice are easier and more cost-effective than human studies and are relevant to human CKD because the gene content and linear organization of genes along chromosomes (Chrs) in rats and mice parallel the gene content and linear organization of human genes (34, 35). This genomic conservation between species allows comparison of quantitative trait locus (QTL) regions across species. For example, after the discovery of an albuminuria QTL on rat Chr 1 (3, 44), several groups tested the homologous locus on human Chr 10q23 and found linkage to creatinine clearance (20) and end-stage renal disease (14). The finding of QTL in corresponding genomic regions across species is termed concordance, and we recently documented the concordance between several kidney disease loci in rats and humans (27). Thus animal models are effective for dissecting the genetic basis of CKD, with reasonable expectation that the findings from animal models will be relevant to CKD in humans.
The National Kidney Foundation defines CKD according to the level of kidney function and the presence or absence of kidney damage, regardless of the type of kidney disease (33). The principal marker of kidney damage is elevated urinary albumin excretion (31, 33), and microalbuminuria, even in the presence of normal kidney function, is diagnostic for CKD (33). We have shown that DBA/2J (D2) mice exhibit progressive albuminuria and rising blood urea nitrogen concentrations with age, characteristic signs of CKD (16). In the present study, we performed a QTL analysis of male progeny from an intercross between C57BL/6J (B6) and D2 mice to identify genomic regions underlying CKD in mice, using urinary albumin concentration as a marker of CKD.
B6 and D2 mice were obtained from The Jackson Laboratory (Bar Harbor, ME) and mated to obtain reciprocal F1 populations (B6 fathers for one F1 population and D2 fathers for the other F1 population). These reciprocal F1 progeny were phenotyped to investigate the role of imprinting and maternal effects on CKD. A separate group of F1 mice produced by mating B6 females to D2 males [(B6×D2)F1] were intercrossed to produce the 340 male F2 mice used for QTL analysis. All mice were maintained on a 12:12-h light-dark cycle, housed in pressurized, individually ventilated cages containing pine shaving bedding and topped with a polyester filter, and allowed access to acidified water and food (18% protein rodent diet, product no. 2018; Harlan Teklad, Madison, WI) ad libitum.
All animal protocols were approved by the Animal Care and Use Committee at The Jackson Laboratory.
Spot urine samples were collected each morning from 10-wk-old male mice by restraining the mice above a 1.5-ml microcentrifuge tube. Samples were collected for four mornings, pooled for each individual mouse, and refrigerated until analyses. Urinary albumin and creatinine concentrations were measured in each sample with a Synchron CX5 clinical chemistry analyzer (Beckman Coulter, Brea, CA). The final urinary albumin concentration was calculated by linear regression from a standard curve generated with the use of mouse albumin standards purchased from Kamiya Biomedical (Seattle, WA). Our group (16) has performed extensive validation of this method to ensure that urine constituents from animals with kidney damage and decreased renal function do not interfere with the detection method.
Darvasi et al. (8) showed that marker spacing of 10–20 cM in a genome scan provides power to detect and resolve QTL similar to that of an infinite number of markers. Therefore, genomic DNA isolated from the tail of each mouse was genotyped with 91 single nucleotide polymorphism (SNP) markers spaced throughout the genome to provide an average marker spacing of ∼17 cM. We expect that this marker density provides sufficient power to detect QTL affecting albuminuria in this population. Genotyping was performed by the Allele Typing Service at The Jackson Laboratory in conjunction with KBiosciences (Herts, UK) using markers chosen from a panel of SNP markers designed to facilitate genotyping in inbred mouse strains (36).
QTL affecting urinary albumin concentration (with and without urinary creatinine concentration as an additive covariate) and urinary albumin-to-creatinine ratio (ACR) were identified by both parametric and nonparametric genome-wide scans (42, 47). Main effect QTL associated with these phenotypes were detected by computing a logarithm of the odds ratio (LOD) score at 2 cM steps over the genome, and these LOD scores were compared with significance thresholds (genome-wide adjusted P = 0.10, suggestive; P = 0.05, significant) computed by permutation analysis (5). QTL confidence intervals were calculated based on Bayesian probability. In addition, interacting QTL affecting each phenotype were identified through a simultaneous search for pairs analysis (42).
Gene expression levels among the kidney tissue samples from D2 and B6 mice were assayed with Illumina mouse v.1.0 BeadChips (Illumina, San Diego, CA). Probe intensity levels for each bead type were generated by the Illumina Bead Studio software. Intensity histograms, box plots, scatter plots, and MA plots were generated with the Bioconductor R/BeadArray package (http://www.bioconductor.org) to detect anomalies in the array data. After these preprocessing diagnostics were performed, expression values for each probe were computed by log transforming the intensities and then quantile normalizing the transformed values to equalize their distribution across all arrays. All statistical tests for detecting differentially expressed probes between the D2 and B6 tissue samples were performed with a modified t statistic incorporating shrinkage estimates of variance components from within the R/MAANOVA package (6, 54). Statistical significance levels for each test were calculated by permutation analysis and were adjusted for multiple testing using the false discovery rate method, q-value (45). Differentially expressed probes were declared at an false discovery rate threshold of 0.05 for each comparison. Probes were annotated to genes using information provided by Illumina and to chromosome locations by mapping each probe to the National Center for Biotechnology Information Build 36 genome using BLAT (25).
Phenotype values for groups are presented as means ± SE. Urinary albumin and creatinine concentrations, urinary ACR, wet kidney weights, and wet kidney weight-to-body weight ratios between parental strains, F1, and/or F2 offspring were compared by ANOVA followed by Bonferroni's multiple comparison posttest. P < 0.05 was considered significant.
Albuminuria in B6, D2, F1, and F2 mice.
We investigated the genetic basis of CKD using an intercross between B6 and D2 mice. Male D2 mice exhibited higher urinary albumin concentrations than B6 male mice (Table 1) and were more susceptible to kidney disease than female D2 mice, since neither B6 nor D2 female mice had detectable urinary albumin concentrations (data not shown). Because renal histology was not notably different between male D2 and B6 mice, we focused on urinary albumin concentrations and ACR as phenotypes of interest and generated reciprocal F1 progeny to test for maternal or imprinting effects on these phenotypes. F1 progeny generated in either direction displayed significantly lower urinary albumin concentrations and ACR than D2 mice but not B6 mice (Table 1), suggesting that maternal and imprinting effects do not alter albuminuria in these mice.
QTL underlying albuminuria in the (B6×D2)F2 population.
Urinary albumin concentrations in the F2 population exhibited a highly skewed distribution (Fig. 1A), which are most appropriately analyzed by nonparametric QTL analysis (2, 28). We performed both nonparametric and parametric (normal model) genome scans, with permutation-based significance thresholds in both analyses to ensure that type I errors were controlled. We first analyzed log urinary albumin concentrations using a nonparametric genome scan (Fig. 1B, summarized in Table 2) and detected a significant, main effect QTL on Chr 2 (Albq5), as well as suggestive QTL on Chr 4 (Albq6) and Chr 5 (Albq7). A normal model genome scan of log urinary albumin concentrations also detected main effect QTL at similar locations (data not shown), demonstrating that the use of a normal model with permutation-based significance thresholds did not result in spurious QTL detection. The nonparametric analysis is limited to simple genome scans, which precluded us from performing multiple regression analysis, determining population variance explained, or fitting multiple QTL models for this nonnormal trait.
Urinary albumin concentrations are affected by urine volume; to address whether differences in urine volume altered the results, we performed parametric and nonparametric QTL analyses for the ratio of urinary albumin to creatinine concentrations, as well as a parametric analysis for urinary albumin concentration with urinary creatinine concentration as an additive covariate. We identified Albq5 as significant QTL in all three analyses and Albq7 as significant in the parametric analyses, but Albq6 on Chr 4 failed to meet the suggestive threshold (data not shown).
We localized Albq5 to distal Chr 2 (peak 62 cM; Fig. 1C); however, the shape of the LOD curve and the large confidence interval suggested that multiple QTL may be present on Chr 2. D2 mice contributed a recessive albuminuria susceptibility allele at this locus because urinary albumin concentrations were not different between B6 homozygous and heterozygous F2 mice, but levels for both were significantly lower than urinary albumin concentrations in D2 homozygous F2 mice (Fig. 1D). Fine mapping localized Albq7 to proximal Chr 5 (peak 25 cM; Fig. 1E) between 9 and 45 cM, where D2 mice contributed an additive albuminuria susceptibility allele; heterozygous F2 mice displayed significantly greater urinary albumin concentrations than B6 homozygous F2 mice but significantly lower urinary albumin concentrations than D2 homozygous F2 mice (Fig. 1F).
We genotyped each of the 340 F2 progeny with markers spanning the genome, so we performed a normal model, pair-wise genome scan to identify epistatic interactions affecting albuminuria. We found a significant interaction between a locus on Chr 8 (Albq8, peak 32 cM) and Albq6. D2 mice contributed an additive high-albuminuria allele at Albq6 in the main scan (data not shown), where mice heterozygous at Albq6 had higher urinary albumin concentrations than B6 homozygous mice. D2 homozygosity at Albq8 significantly increased urinary albumin concentrations in those mice heterozygous for Albq6 (Fig. 2A). D2 mice also contributed a high albuminuria allele at Albq7 in the main scan (Fig. 1F). We found that D2 homozygosity at Albq7 was significantly modified by the presence of homozygous B6 alleles at Albq9 (Chr 14, 22 cM; Fig. 2B). The main effect and interacting QTL affecting albuminuria are summarized in Table 2.
Evidence for multiple QTL on Chr 2.
The fine mapping plot of Albq5 showed a broad QTL on Chr 2 (Fig. 1C) that suggested the presence of multiple QTL, so we used several approaches to investigate this hypothesis. First, we reanalyzed Chr 2 with seven additional genotyped markers spanning the QTL interval, and the fine mapping plot distinctly showed two separate QTL with peaks at 34 and 74 cM on Chr 2 (Fig. 3A). Second, we transformed the urinary albumin concentration data to minimize the impact of the nonnormal trait distribution. To do this, we rank ordered the log albumin concentration across all mice and calculated the inverse cumulative normal distribution. We then extended the nonparametric analysis to pair scan and rescanned just Chr 2, which suggested the presence of three interacting QTL at 36, 74, and 134 cM (Fig. 3B). Finally, we fit an additive model for 1, 2, or 3 QTL to determine the statistical evidence for multiple QTL on Chr 2. The 3 QTL additive model was significant compared with 1 QTL models at 36, 74, and 134 cM, as well as 2 QTL models containing different QTL pairs at the three loci (data not shown). Together, these analyses provide solid evidence for three linked loci on Chr 2.
Narrowing Albq5 using bioinformatics tools.
Shike et al. (43) previously identified an albuminuria QTL spanning from 78 to 103 cM on distal Chr 2 in a cross between KK/Ta (KK) and BALB/c (BALB) mice (43). The middle Chr 2 QTL found in our cross, ranging from 68 to 84 cM (Fig. 3), overlaps the Chr 2 QTL identified in the KKxBALB cross. Therefore, we applied several newly developed bioinformatics tools to narrow this QTL (11). The QTL identified by Shike et al. ranges approximately from 137 to 181 Mb on Chr 2 and contains 558 genes, whereas the QTL found in our cross spans from ∼129 to 160 Mb and contains 386 genes (Fig. 4). The region of overlap between these two crosses extends from 137 to 160 Mb and contains 302 genes. A portion of this region of mouse Chr 2, from 137 to 152 Mb, is concordant to a QTL linked to diabetic nephropathy on human Chr 20p (21). Thus comparison of mouse-human QTL homology reduced the number of likely candidate genes to 133.
After narrowing the QTL interval by comparing mouse-human homology, we applied interval-specific haplotype analysis (11) to further narrow the interval. We downloaded a set of 49,575 SNP genotypes spanning the region of QTL overlap in the mouse crosses (137–160 Mb) for BALB, KK, D2, and B6 mice from the Mouse Phenome Database (www.jax.org/phenome), which contains publicly available data from the Perlegen mouse genome resequencing project. Because the genotype information for the KK/Ta and BALB/c substrains was not available, we downloaded genotype data for the KK/HL and BALB/cBy substrains for use in our haplotype analysis. A comparison of BALB/c with BALB/cBy using 1,934 SNPs in the TJL3 SNP dataset in the Mouse Phenome Database revealed that these substrains are 99.84% identical (1,931/1,934 SNPs genome-wide and 141/141 SNPs on Chr 2 are the same between the substrains). We then compared the strain distribution pattern for each SNP across the four strains to identify SNP markers where BALB and B6 share alleles and differ from D2 and KK. Of the 49,575 SNPs tested across the interval, 11,230 (22.7%) fit this pattern. These SNPs fell within 25 annotated genes and 5 Riken clones (Table 3), of which only 6 contained nonsynonymous coding polymorphisms that also fit the appropriate strain distribution pattern. In addition, we identified six genes as differentially expressed in kidneys from B6 and D2 mice. Overall, application of these bioinformatics tools allowed us to narrow our focus from the 386 genes in the initial QTL interval to 10 genes with known coding polymorphisms or renal expression differences between B6 and D2 mice.
Our findings show that urinary albumin concentration is a valid trait for QTL analysis and that differences in urine output, as estimated by urinary creatinine concentration, did not greatly affect the results of our analysis. The nonnormal distribution of albumin concentrations in the F2 population was of concern, but the results from the normal model and nonparametric analyses were very similar. We conclude that albuminuria QTL can be accurately and reliably detected by nonparametric analysis of albumin concentrations without the need to account for creatinine concentrations, by using either ACR or multiple regression adjustment with creatinine as an additive covariate.
QTL analyses in rodent models are important because they often correspond to human disease loci, as our group (27) recently documented in a review of concordance between rodent and human QTL underlying CKD. We resolved Albq5, detected on mouse Chr 2, into three linked QTL by genotyping additional markers across the interval and by using a nonparametric pair-scan analysis. The proximal QTL (36 cM) is concordant to portions of human Chrs 11p and 2q, which has been linked to renal function in families with type 2 diabetes [2q33.3 (37)] and to early onset of end-stage renal disease in families enriched for nondiabetic nephropathy [2q32.1 (13)]. The distal QTL (134 cM) is concordant with human Chr 20q11.2–13.1. Although this region has not been linked to CKD in humans, it contains the matrix metalloproteinase 9 gene that is associated with end-stage renal disease (19). The middle QTL (74 cM) overlaps an albuminuria QTL detected between KK/Ta and BALB/c mice (43) and is concordant with both the urinary albumin excretion QTL found on rat Chr 3 [Rf-3 (44)] and the human QTL affecting ACR on Chr 20p12 (21). Albq7 on mouse Chr 5 is concordant with an end-stage renal disease QTL on human Chr 4p15 (12), but the homologous region of rat Chr 14 has not been linked to CKD. The findings of concordance between QTL influencing urinary albumin concentration in this mouse cross and CKD QTL detected in rats and humans are consistent with the high degree of concordance previously documented for CKD between rodents and humans.
One way to capitalize on the concordance between kidney disease loci detected in humans and animal models is through comparative genomic analyses. After comparing the regions of overlap between Chr 2 QTL identified in separate mouse crosses, we used comparative genomic analyses to further narrow the QTL interval. Comparison of this QTL region with the rat Chr 3 QTL interval failed to narrow the interval because the whole mouse region was contained within the concordant rat region. However, the mouse interval is concordant to two regions of the human genome, Chr 20p11–13 and Chr 20q11–13. Human Chr 20p12–13 is linked to nephropathy in Pima Indians (21), whereas human Chr 20q has not been linked to kidney disease-related phenotypes. This allows us to focus our search for candidate genes within the homology region that has been linked to nephropathy in both species (human Chr 20p12–13 and mouse Chr 2-137-152 Mb) and substantially narrow the list of candidate genes.
Our use of comparative genomic analyses assumes that the same gene causes kidney disease between the different mouse strains and in humans. There is strong evidence to support this concept; the finding of QTL concordance between mice and humans for numerous complex traits, including kidney disease (27), blood pressure (46), bone mineral density (26), atherosclerosis (50), and plasma lipid concentrations (51, 52), suggests that the same genes underlie QTL that are concordant between mice and humans. More specifically, several different causal QTL genes have been proven in animal models and found to be associated with the corresponding human disease, such as Ox40l for atherosclerosis (53), Hc for liver fibrosis (17), and Ctla4 for diabetes (48). Samuelson et al. (39) recently identified Fbxo10 and Frmpd1 as genes underlying a rat QTL for breast cancer risk, and the homologous human genes were associated with breast cancer. Similarly, Chang et al. (4) used our published interval-specific haplotype analysis of overlapping blood pressure QTL on mouse Chr 1 (10) to identify candidate genes within a concordant human blood pressure QTL. Of the nine candidate genes that they tested, two of the three genes that were associated with blood pressure in human populations were supported by QTL and haplotype analysis in mice. Therefore, we believe that there is compelling evidence to support the use of comparative genomic analyses to effectively narrow QTL intervals that are concordant between rodent models and humans and to identify high-quality candidate genes.
The use of mice in the genetic analysis of CKD has been primarily limited to genetic studies using knockout or transgenic mouse models. We believe that the success of this mouse cross, as well as the earlier cross using diabetic KK/Ta mice, to identify albuminuria QTL that are concordant to human CKD QTL supports the further use of mouse crosses for the genetic dissection of CKD. Additional mouse crosses would facilitate the use of bioinformatics methods for narrowing QTL (11) and would provide a complementary approach to genetic studies of CKD ongoing in rats, which can provide insight into the genetic factors most likely to influence CKD in humans.
This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases Grant DK-69381 (B. Paigen) and National Institute of General Medical Sciences Grant GM-070683 (G. A. Churchill).
We thank the Histology Service at The Jackson Laboratory and Dr. David Feldman of the Novartis Institutes for BioMedical Research (East Hanover, NJ) for assistance with renal histology.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2007 the American Physiological Society