Main

Determining the molecular basis of natural phenotypic variation, including interindividual susceptibility to common diseases, is a central challenge of postgenome genetics. The availability of genome sequences and genome-scale technologies has enabled new strategies for identifying genes underlying complex phenotypes and has greatly accelerated progress in this field.

The high heritability of variation in gene expression1 has suggested that identification of the genetic determinants of gene expression may give insights into the molecular basis of complex traits. One justification for studying the genetics of gene expression is that transcript abundance may act as an intermediate phenotype between genomic DNA sequence variation and more complex whole-body phenotypes. Mapping of gene expression levels as quantitative trait loci (called eQTLs) has been undertaken in yeast2,3 and more recently in mammals4,5 and has shown that the approach is feasible. As a tool for studying disease phenotypes, however, aside from a single study in F2 mice4, the approach has not been extensively tested.

For the past 50 years, the rat has been a leading model for the study of common, complex human diseases6,7. The availability of the rat genome sequence8 has made feasible studies of gene expression at the level of the genome alongside well-characterized rat phenotypes. The spontaneously hypertensive rat (SHR) is a widely studied model of human hypertension and also has many features of the metabolic syndrome9,10,11,12,13. In the early 1980s, the SHR strain was crossed with the normotensive Brown Norway (BN) strain to generate the BXH/HXB panel of recombinant inbred (RI) strains14,15. Rodent RI panels are powerful and permanent resources for genetic mapping that offer the opportunity to accumulate genetic and physiological data over time16. A further advantage of RI strains, as with chromosome substitution strains17,18, is the ability to study genetically identical biological replicates, which increases trait heritability by reducing environmental variance19.

Here we applied integrated gene expression profiling and linkage analysis to the regulation of gene expression in fat and kidney tissue in the BXH/HXB panel of rat RI strains. We found that these RI strains are a suitable genetic system for large-scale identification of positional candidates and regulatory pathways for previously mapped physiological QTLs (called pQTLs). By comparative mapping, we compiled a data set of candidate genes for investigating the molecular basis of human hypertension. By identifying hundreds of robustly mapped cis- and trans-acting eQTLs in a model system with large numbers of existing pQTLs, we generated a unique and accessible resource to test the hypothesis that genetic variation in gene expression has a key role in the molecular evolution of complex physiological and pathophysiological phenotypes.

Results

Linkage analysis of expression profiles in RI strains

We analyzed genome-wide expression data of 15,923 transcripts collected from fat and kidney of 30 RI strains and the SHR and BN progenitor strains. To assess variability within RI strains and to generate a robust set of data, we analyzed gene expression in fat and kidney from four independent rats from each RI strain and from four or five rats from each progenitor strain. Including parental progenitor strains, we carried out 259 individual array hybridizations.

We first compared gene expression profiles generated from fat and kidney tissue from the parental strains (SHR and BN). Of the 15,923 probe sets present on the array, 2,046 and 1,553 detected differential expression (P < 0.05) in fat and kidney, respectively, between SHR and BN parental strains. We next carried out genome-wide linkage analysis for the expression data generated in the RI strains. Because genetic regulation of gene expression may be detected for genes that are not differentially expressed between the parental strains, we carried out the linkage analysis for expression profiles from all 15,923 probe sets without filtering.

We generated likelihood ratio statistic (LRS) values and established empirical genome-wide significance by permutation testing. At genome-wide significance of P = 0.05, we detected 3,520 and 4,530 linkages in fat and kidney, respectively. To account for multiple testing of 15,923 expression phenotypes, we evaluated the false discovery rate (FDR) at several levels of significance. Although the FDR is relatively high at P = 0.05, the expected number of true linkages is 2,644 in fat and 2,917 in kidney (Supplementary Table 1 online). At a more stringent level of P = 10−3, the FDR is ∼4% in both tissues, corresponding to 509 and 761 expected true positive linkages in fat and kidney, respectively. An independent test for linkage between marker and transcript using the Wilcoxon-Mann-Whitney test showed that 65–68% of these linkages at P = 10−3 are common to the two analyses (Supplementary Table 2 online).

To validate differential expression detected by microarray and to determine whether interstrain differences in gene expression could be due to DNA sequence variants affecting probe binding, we measured mRNA abundance in parental and RI strains by quantitative RT-PCR and sequenced the target region of a subset of cis-regulated transcripts. We found significant sequence differences that could affect probe binding for only 1 of the 15 transcripts that we examined: this transcript, in the major histocompatibility complex (MHC), showed significant sequence variation between SHR and BN that accounted for the apparent differential expression detected on the microarray (data not shown). We observed strong concordance between the microarray and RT-PCR data for differential expression between the parental strains (Supplementary Fig. 1 online). In addition, we confirmed eQTL linkages for nine transcripts showing strong genome-wide linkage (P < 10−3; Supplementary Table 3 online), including those showing small relative changes between genotypic classes.

Characterization of cis- and trans-acting eQTLs

Linkages of individual expression phenotypes to multiple tightly linked markers give rise to artificially inflated numbers of eQTLs. We removed this redundancy using a custom algorithm (Supplementary Fig. 2 online), delineating a data set of 2,118 and 2,490 nonredundant eQTLs at genome-wide significance level of P < 0.05 in fat and kidney, respectively (Table 1). We examined which of these eQTLs were regulated in cis or in trans by defining a cis-acting eQTL as having a linkage peak within 10 Mbp of the physical location of the probe set (Supplementary Fig. 3 online). The proportion of eQTLs regulated in cis or in trans varied substantially in accordance with the genome-wide significance of the eQTL linkage. At a genome-wide threshold of P < 0.05, 60–65% of the eQTLs were regulated in trans in both tissues. At a higher significance level (P ≤ 10−4), however, 85–100% of eQTLs were regulated in cis (Table 1). These results may reflect the large gene effects of some cis-acting eQTLs and the probable oligo- or polygenic trans-acting influences on gene expression.

Table 1 Number of eQTLs detected in fat and kidney data sets at different genome-wide thresholds of significance
Full size table

A small proportion (∼15%) of the eQTLs detected independently in kidney and fat were common to both tissues. These 311 eQTLs (Table 1) can be considered replicated linkages and probably reflect common regulatory mechanisms that are shared between the two tissues. A very high proportion of these shared linkages are cis-acting (70% at P < 0.05). This suggests that the preponderance of trans-acting eQTLs observed in the distinct fat and kidney data sets belong to tissue-specific networks for control of gene regulation.

Some of the most significantly linked cis-acting eQTLs detected in the RI strains also showed marked differential expression in the parental strains (Fig. 1 and Table 2). Although a number of these eQTLs were shared between kidney and fat (e.g., Cd36 and Ilf3), others showed tissue-specific differences in regulation of gene expression (Sah, also called Sa, and the transcribed sequence 1388437_at). Many of these cis-acting eQTLs showed similar distributions of gene expression in the parental strains and in the RI strains when separated according to marker genotype at the linkage peak (Fig. 1). This is suggestive of a monogenic, or near monogenic, cis-acting regulation of gene expression for these genes and for others with linkage at similar levels of significance (Table 2). Nevertheless, 675 of all linked transcripts showed linkage to two or more loci, emphasizing the generally complex nature of regulation of gene expression.

Figure 1: Expression values of parental and RI strains for 12 transcripts in kidney and fat.
figure 1

The first column in each plot shows expression levels for all 30 RI strains. The second and third columns show expression levels for the replicates from each parental strain (SHR and BN, respectively). The fourth and fifth columns show expression values for RI strains by SHR and BN marker genotype, respectively. RMA-normalized expression values are shown on the y axis. Detailed properties for all genes depicted are given in Table 2. These genes, all with cis-acting eQTLs at genome-wide significance level of P ≤ 5 × 10−4 in either kidney or fat, were selected according to the following criteria: (i) four genes previously reported to show cis-acting differential expression or association with an SHR phenotype (a–d); (ii) four genes with highly significant cis-acting eQTLs in both tissues (e–h); (iii) two genes with highly significant eQTLs and with biological relevance in kidney tissue (i–j); (iv) two genes with highly significant eQTLs in fat tissue (k–l). Gene names and linkage statistics are given in Table 2.

Full size image
Table 2 Linkage statistics and biological properties of cis-regulated genes
Full size table

SNP detection and SNP frequency in eQTL genes

To identify DNA sequence variants that may represent candidate quantitative trait nucleotides underlying eQTLs, we generated sequence data for seven of the most statistically significant cis-acting eQTL genes (Pik3c3, Myh9, Kif1c, Aox1, Ascl3, Dgat2 and Gnpat). Although variations in all parts of the gene can affect transcript levels, we focused on variations in the promoter or cDNA that might directly influence transcriptional activity or mRNA stability.

We identified sequence variants in the exons or upstream regulatory regions in six of the seven genes. As a first step towards assessing the functional importance of these sequence variants, we determined the allele status for all sequence variants in three additional inbred strains by resequencing and compared the results to the reference rat genome sequence. Only one of these genes, Pik3c3, had a distribution of allelic variants (one promoter SNP and one silent substitution in exon 19) that was common to all of the studied hypertensive strains (SHR/Ola, SHR/Mdc and SHRSP/Mdc) and distinct from that of the normotensive strains (BN.Lx/Cub, BN/SsNHsd/Mcwi and WKY/Mdc). Further sequencing of Pik3c3 from genomic DNA detected 19 SNPs distributed throughout the gene on two distinct haplotypes (Supplementary Fig. 4 online).

Pik3c3 showed allelic segregation with hypertension across strains, is one of the most statistically significant (P < 10−6) cis-acting eQTL genes in our data set and resides within the blood pressure QTL BP46 (pQTL ID 107) and in a chromosome 18 SHR congenic strain (SHR.BN-D18Rat32/D18Rat12; pQTL ID 105; Supplementary Table 4 online). Pik3c3 was upregulated by a factor of 1.4, as shown by quantitative RT-PCR, in kidneys from the SHR strain compared with those from both the BN and the chromosome 18 congenic strains (both P < 0.05), confirming cis-regulated control of Pik3c3 expression in the SHR strain. Kidneys transplanted from congenic SHR.BN-D18Rat32/D18Rat12 rats into SHR rats led to a highly significant drop in blood pressure (by 12 mmHg) compared with kidneys transplanted from SHR rats into other SHR rats (data not shown). These data indicate that the congenic kidney is sufficient to mediate blood pressure changes and support the idea that Pik3c3 is a good candidate for involvement with hypertension in the SHR strain.

The finding of SNPs in six of the seven cis-acting eQTL genes that we studied led us to investigate the SNP frequency at the level of the genome compared with the frequency in eQTL genes. We first inspected the number of SNPs identified in a previously published data set of rat cDNAs20. For the 8,986 genes on the RAE230A array with Ensembl IDs, SNPs were detected between the stroke-prone SHR (SHRSP) and the BN reference sequences in 2,092 genes (23.3%), which we considered the observed SNP rate across the genome in this SNP data set. We then determined the number of genes in this data set that contained SNPs in which we detected cis- and trans-acting eQTLs and found significant enrichment for SNPs in the cis-regulated eQTL genes compared with either the trans-regulated eQTL genes or the observed rate across the genome (Table 3).

Table 3 Percentage of transcripts with sequence variants
Full size table

Relationship of eQTLs to pQTLs

We addressed how the eQTL data can be used to identify candidate genes underlying SHR phenotypes, many of which have been analyzed by QTL mapping in experimental crosses over the past 15 years. Cis-acting eQTLs are good candidates for these pQTLs because they show strain-specific differences in gene expression that are under the control of DNA sequence variants in or close to the gene itself.

To illustrate the overlap between individual cis-acting eQTLs and previously mapped SHR pQTLs, we plotted the locations of the most stringently mapped (P ≤ 10−4) cis-acting eQTLs against the chromosomal locations of known SHR pQTLs (Fig. 2). This generated a data set of genes that have cis-acting sequence variants that mediate interstrain differences in gene transcription or mRNA stability (Fig. 2; details of the pQTLs and lists of the cis-acting eQTLs with P ≤ 10−4 (FDR ≤ 1% in both tissues) are given in Supplementary Tables 4 and 5 online). The sequence variants may also underlie (patho)physiological phenotypes mapped as pQTLs in crosses between SHR, BN and other strains (Fig. 2). These cis-regulated eQTL genes merit testing in genetic and functional assays as positional candidates for colocalizing SHR pQTLs and potentially for related human QTLs. Other cis-acting eQTLs may be considered 'orphan' eQTLs, candidate genes for pQTLs that have yet to be mapped.

Figure 2: Locations of cis-acting eQTLs and previously mapped SHR pQTLs.
figure 2

Chromosomes are shown in blue. The arrowheads on the left of each chromosome (yellow, fat; red, kidney; green, shared) represent the location of the probe set for each cis-acting eQTL with P ≤ 10−4. Previously identified pQTLs in the SHR strains and in the RI strains are shown on the right of each chromosome. Gray boxes represent pQTLs for which both flanking markers are mapped. White boxes represent pQTLs for which incomplete flanking marker information was available, in which case the flanking marker(s) are estimated to be 10 Mbp from the linkage peak. The numbers on the pQTL bars correspond to pQTL information reported in Supplementary Table 4 online. Some pQTLs in Supplementary Table 4 online do not appear in this diagram owing to a lack of physical mapping information for the genetic markers that define the pQTL. Gene names, relative changes in parental strains and public database references for probe sets are given in Supplementary Table 5 online. Scale in Mbp.

Full size image

Trans-acting eQTLs represent loci that influence expression of genes or transcripts remote from the eQTL itself. Coincidental mapping of trans-acting eQTLs for multiple transcripts to the same chromosomal location, as observed on chromosomes 3 and 17 (Fig. 3), may represent a shared regulatory transcriptional control mechanism by a single gene at the eQTL. The locations of trans-acting eQTLs in relation to known SHR pQTLs (Fig. 3), together with the locations of cis-acting eQTLs (Fig. 2), may point to genes and regulatory pathways underlying individual SHR pQTLs.

Figure 3: Location of trans-acting eQTLs and previously mapped SHR pQTLs.
figure 3

pQTLs are shown as in Figure 2. The arrowheads on the left of each chromosome (colors as in Fig. 2) represent the locations of the markers at the linkage peak for each trans-regulated eQTL with P ≤ 10−2. This threshold of significance was chosen because trans-acting eQTLs are expected to have smaller effects than cis-acting eQTLs4. Gene names, relative changes in parental strains and public database references for probe sets are given in Supplementary Table 8 online. Scale in Mbp.

Full size image

Comparative analysis of blood pressure QTLs

To explore the applicability of the detected fat and kidney cis-acting eQTLs to human hypertension, we formed a data set of 255 cis-acting eQTL genes with FDR ≤ 5% that were contained within rat pQTLs for blood pressure and left ventricle and cardiac mass previously mapped in the SHR strain (Supplementary Table 4 online). We used the Ensembl Ensmart database to map Affymetrix probe set identifiers of each cis-acting eQTL to rat Ensembl genes and then to identify the putative human orthologs. We determined the physical location of each human ortholog on the human genome sequence and compared these locations with the locations of previously mapped human blood pressure QTLs (Supplementary Table 6 online). This yielded a set of 73 human orthologs of cis-regulated rat eQTL genes that are contained within human blood pressure QTLs and are candidate genes for human hypertension (Table 4).

Table 4 Comparative analysis of rat and human blood pressure and blood pressure–related QTLs
Full size table

Discussion

Identification of genes that underlie complex (polygenic) traits remains a challenge despite the availability of the genome sequences of humans and other species21,22. We previously combined linkage analysis with microarray-based expression profiling to identify a defective gene, Cd36, that underlies complex disease phenotypes in the SHR strain12,23. Other investigators have successfully used this approach in the study of both monogenic and complex traits24,25,26. But the relative paucity of successful studies using this strategy, and the nature of the Cd36 genomic deletion27 that led to detection of dysregulated Cd36 expression on the microarray, raised questions about the generality of this approach.

Here we applied global gene expression profiling and linkage analysis to study the regulation of gene expression in fat and kidney tissue in the BXH/HXB panel of rat RI strains. We used this panel to map the genetic determinants of gene expression in the SHR strain for 15,923 genes in two of the key tissues in the pathophysiology of the metabolic syndrome. After assessing genome-wide significance, removing redundancy and accounting for multiple testing using FDR, we found more than 1,000 eQTLs in each tissue, of which several hundred were common to both kidney and fat. These eQTLs represent a large source of candidate genes for the scores of pQTLs that have been mapped in the SHR strain.

As previously observed2,3,4,5,28, a powerful feature of this experimental design is the ability to discriminate between cis- and trans-acting influences on gene expression. In keeping with previous results4, we found that, at relatively low levels of significance (P < 0.05), a high proportion of detected eQTLs, ∼65%, result from trans-acting regulators of gene expression. At higher levels of significance (P ≤ 10−4), 85–100% of eQTLs arise from cis-acting regulation and have larger, mainly monogenic effects on gene expression. This suggests that, owing to sequence variation in the gene itself, cis-acting regulation is more easily detected than variation in genes with secondary effects on transcription of other genes and underscores the more complex nature of trans regulation.

Investigation of sequence variation in eQTL genes showed that the SNP frequency was much higher in genes with cis-acting eQTLs than in genes with trans-acting eQTLs or than the rate observed across the genome (Table 3). This observation has a number of possible explanations. First, because the SHR and SHRSP strains are closely related, and both are genetically distant from the BN strain, the increase in polymorphisms detected in cis-regulated genes could reflect identification of causative nucleotide variants that underlie cis-acting control of gene expression. Second, cis-regulated genes may lie in SNP-rich chromosomal regions, and SNPs in these genes could therefore be considered markers of chromosomal regions of phylogenetic diversity between SHR-related strains and the BN strain. If this is the case, a haplotype map of the rat could point to chromosomal regions containing a high density of cis-regulated genes.

Cis-acting eQTLs are of particular interest as positional candidate genes for pQTLs (Fig. 2 and Supplementary Tables 4 and 5 online). Among many other cis-acting eQTLs, our data showed cis-acting control of gene expression for Cd36 and Sah, in which cis-acting control of gene expression or significant intragene sequence variation has been documented23,29. Cd36 is represented twice on the microarray by nonoverlapping probe sets. Our data identified strong cis-acting eQTLs for both probe sets (P ≈ 10−6) in fat and in kidney (Table 2 and Supplementary Table 5 online). One of the probe sets (1386901_at) is derived from sequence in the 3′ untranslated region that is deleted from the SHR genome23,27, and the second (1367689_a_ at) is derived from sequence in the Cd36 coding region and uses oligonucleotides that do not differ in sequence between the SHR and BN strains27. This confirms that we can detect the chromosomal deletion found previously with cDNA microarrays23 and also shows, using probe set 1367689_a_at, that the cis-acting eQTL for Cd36 cannot simply be attributed to strain differences in probe affinity for Cd36 mRNA.

For Sah, our data showed highly significant differential expression in kidney (P = 10−9) between SHR and BN parental strains and strong cis linkage in the RI strains (P = 10−5; Fig. 1c and Table 2 and Supplementary Table 5 online). The genotype-dependent bimodal distribution of Sah expression in the RI strains (Fig. 1c) indicates essentially monogenic cis regulation, as previously reported29. Although Sah is now known not to be a primary determinant of hypertension in the SHR strain30,31, the detection of this cis-acting eQTL in kidney but not in fat demonstrates the ability of this system to identify tissue-specific differences in gene expression, as shown also for many other genes in our data set.

In this study, we investigated seven of the most statistically significant cis-regulated eQTL genes from kidney as positional candidates for involvement in hypertension. Nucleotide variations in one of these genes, Pik3c3, formed a haplotype that associated with hypertension in a small range of hypertensive and normotensive strains (Supplementary Fig. 4 online). Pik3c3 resides in both an SHR blood pressure QTL and an SHR congenic segment that carries a hypertension gene on this region of chromosome 18 (refs. 32,33). Additional physiological and biochemical data (Supplementary Note online) support the idea that Pik3c3 is a candidate for involvement in hypertension in the SHR strain. Existing functional data should similarly encourage investigation of other cis-acting eQTL genes (Table 2 and Supplementary Table 5 online), such as Pik3r1, the regulatory subunit of PI3K, and Ace, which has been extensively investigated as a candidate for involvement in cardiovascular phenotypes34,35.

Eleven of the detected cis-acting eQTLs are located in or close to the MHC on chromosome 20 (Fig. 2). Because we observed cDNA sequence variation between the SHR and BN strains in the single MHC gene that we tested, which accounted for apparent differential expression between strains, several of the apparent cis-acting eQTLs in the MHC probably have a similar basis. Because we found no significant sequence variation in an additional 15 non-MHC genes, however, we do not believe that this accounts for more than a very small proportion of all cis-acting eQTLs.

The detection in this study of known cis-acting regulation of gene expression raises the question of whether this approach is of value in identifying the genes that underlie pQTLs. The SHR defect in Cd36 has been conclusively shown in complementation studies to result in pathophysiological phenotypes36, whereas upregulation in SHR Sah gene expression, detected here and elsewhere, is not now believed to contribute to the hypertensive phenotype30,31. Several other examples now exist in which a causal relationship has definitively been shown between cis-acting eQTLs and functional or physiological phenotypes2,4,26. The available data therefore suggest that the combined expression and linkage approach may be useful for pQTL gene identification, particularly when applied on a genome scale.

The trans-acting eQTLs that we mapped represent transcripts whose abundance is regulated by loci remote from the genomic locus of each of these genes. Yvert et al. analyzed in detail a set of trans-acting eQTLs and defined clusters of significantly coregulated genes with functional effects on yeast biology3. We found several large groups of genes, up to 43 in a single group (Fig. 3), with colocalizing trans-acting eQTLs, suggestive of coregulation by a common gene in the eQTL. Some of these groups of trans-acting eQTLs overlap with pQTLs, suggesting that they may play a part in mediating the development of SHR phenotypes. Functional and in silico analyses of the genes in these groups may advance understanding of the regulatory pathways that underlie these phenotypes.

To show how mining of our data set may be applied to the study of human disease phenotypes, we identified a set of 73 robustly mapped cis-regulated rat eQTL genes with FDR < 5% that lie in SHR blood pressure–related pQTLs and whose human orthologs reside in QTLs for human hypertension (Table 4 and Supplementary Table 6 online). These genes are good candidates for underlying human hypertension, although for many there is little functional information. Some of the genes for which functional information does exist show circumstantial association with hypertension. For example, glutathione S-transferase mu-type 1 (Gstm1) has a strong cis linkage (P < 10−5) in kidney tissue, is differentially expressed by a factor of ∼2.4 fold between the parental strains and lies in a cluster of blood pressure QTLs on rat chromosome 2. Gstm1 has also been implicated previously in hypertension in genetic studies in the SHRSP strain37 and in humans38, suggesting that it may have a role in hypertensive rat strains other than SHR and in humans. Like Gstm1, other genes in this data set play a part in cellular resistance to oxidative stress or exhibit different biological characteristics that make them plausible positional candidates for human hypertension (Supplementary Note online).

Previous studies taking the combined expression and linkage approach used simple eukaryotes, transformed cell lines from healthy individuals or segregating populations that are no longer available for additional phenotyping2,3,4,5. Here we applied the same strategy to study, in two different mammalian tissues, regulation of gene expression in the BXH/HXB panel of rat RI strains. RI strains have a number of advantages for this type of study, including the ability to make measurements in multiple genetically identical animals from the same strain to increase trait heritability and the ability to accumulate new phenotypes over time. In addition, the continued breeding of these strains, which are publicly available on a collaborative basis (from M.P. and V.K.), the public availability of the expression data set in this study and of the linkage maps generated in these strains39 give value to these results outside the context of the immediate findings presented here. Alternative study designs in consomic or congenic strains could provide similar long-term resources and should complement the approach presented here16,17,18. In our study, we mapped genetic determinants of gene expression in a single strain combination, SHR × BN. But the coincidental mapping of physiological phenotypes in several crosses23,40, taken together with the data for Pik3c3 showing association between allelic variants and hypertension across strains, suggests that the eQTLs identified in this study will probably be relevant to SHR phenotypes mapped under a variety of environmental conditions or in strain combinations other than SHR × BN.

The value of our results is enhanced by the extremely detailed studies (over a 30-year period) of hypertension, other components of the metabolic syndrome and further unrelated SHR phenotypes9,12,15,23,36,41,42,43. Many of these phenotypes have been subjected to genetic analysis in the BXH/HXB RI panel or are amenable to such analysis. Any phenotype that segregates in the SHR × BN strain combination can potentially be analyzed by this approach, either using the RI strain expression data set from this study or by generating new expression data sets in other tissues or studying additional transcripts in this RI panel.

Methods

Strains and tissue.

We produced a set of RI strains by inbreeding between members of the F2 generation resulting from the cross of the two highly inbred strains: BN (BN.Lx/Cub) and SHR (SHR/Ola)14. We used 30 RI strains (BXH and HXB) at F60. We housed rats in an air-conditioned animal facility and allowed them free access to standard laboratory chow and water. All experiments were done in agreement with the Animal Protection Law of the Czech Republic (311/1997) and were approved by the Ethics Committee of the Institute of Physiology, Czech Academy of Sciences, Prague. We killed rats at 6 weeks of age. We collected tissues from four unfasted males of each RI strain and from four or five rats from each parental strain between 9:00am and 10:00am, froze them in liquid nitrogen and stored them at −80 °C.

Preparation of labeled cRNA and hybridization.

We extracted total RNA from retroperitoneal fat pads12 and from whole kidney from four or five rats of each strain using Trizol reagent (Invitrogen) and purified it using an RNeasy Mini kit (Qiagen) in accordance with the manufacturer's protocol. We synthesized double-stranded cDNA from total RNA without pooling samples, synthesized biotinylated cRNA from cDNA using the MEGAscript T7 kit (Ambion) and nucleotide analogs (Perkin Elmer) in the kidney samples and using Bioarray High Yield RNA Transcript Labelling Kit (Enzo Diagnostics) in fat. We hybridized 15 μg of the fragmented cRNA samples to rat expression Affymetrix RAE 230A GeneChips arrays in accordance with the Affymetrix protocol.

Analysis of expression data.

We computed gene expression summary values for Affymetrix GeneChip data using the robust multichip average (RMA) algorithm44, which uses background adjustment, quantile normalization and summarization. Statistical comparison of expression data in the parental strains was by student's t-test (two-tailed) without correction for multiple testing. To determine relative changes, we back-transformed raw RMA output values to the raw intensity scale (anti-log).

Validation of microarray gene expression data.

We used quantitative real-time PCR (TaqMan) to compare mRNA levels of 16 transcripts in kidneys that show a range of relative changes in expression between the SHR and BN progenitors. To validate the linkage data, we also measured mRNA levels by quantitative real-time PCR for nine cis-regulated transcripts across all RI strains. We reverse-transcribed DNA-free total RNA (2 g) with oligo(dT) primers (Gibco-BRL), Superscript II reverse transcriptase (Gibco-BRL) and dNTP (Boehringer Mannheim) in 40 μl of reaction buffer (Gibco-BRL). We designed primers and probes using Primer Express 1.0 (Applied Biosystems). TaqMan analysis used an Applied Biosystems 7700 system (Perkin Elmer). We normalized expression levels to 18S rRNA expression by using the 2−ΔΔCT method.

Map construction.

We constructed a linkage map of 1,011 autosomal markers for all chromosomes using MAPMAKER/Exp. 3.0 (ref. 45) using two- and four-point linkage analysis and published marker genotypes39,46,47. We corrected the map based on known physical positions of markers and optimized it by 19 iterative steps using multipoint linkage analysis. Map details are given in Supplementary Table 7 online. We retrieved physical map positions of genetic markers from Ensembl or by alignment of available marker sequences to the rat genome using blastn. Markers that could not be mapped using blastn but that were located between physically anchored markers were placed on the physical map by interpolation. Eleven percent of all genetic markers could not be placed on the physical map and were used only for the linkage analysis.

Mapping of eQTLs.

We derived mean expression values from the four replicates for each RI strain and each tissue after application of the Nalimov outlier test at P < 10−3. We carried out genome-wide linkage analysis for each of the 15,923 expression traits and the 1,011 genetic markers using the eQTL Reaper program (K.F. Manly; University of Tennessee Health Science Center, Memphis, Tennessee), which generates an LRS as a measure of the significance of a possible eQTL. eQTL Reaper establishes genome-wide significance by permutation test, estimating an empirical genome-wide probability for observing a given LRS score by chance48. For each probe set, permutations are carried out until an LRS greater than that for the real data is observed or until 1,000,000 permutations have been done, conditional on at least 1,000 permutations being done. The permutation procedures for the eQTL Reaper program correct, for individual expression phenotypes, for multiple testing across genetic markers to give a genome-wide corrected P value.

Calculation of FDR.

The eQTL Reaper genome-wide P value accounts for multiple testing across genetic markers but not for multiple testing across the 15,923 expression measurements. We estimated the number of falsely discovered linkages at a given genome-wide significance level by calculating the q value49, defined as the minimum positive FDR for a fixed significance threshold (Supplementary Table 1 online).

Definition of cis- and trans-acting eQTLs.

Cis-acting regulatory variants are polymorphisms located at or near a gene that influence mRNA levels of that gene. We defined cis-acting eQTLs as eQTLs that map within 10 Mbp of the physical location of the probe set on the genomic sequence (20 Mbp total window size; Supplementary Fig. 3 online). Other eQTLs were defined as acting in trans. We obtained physical locations of probe sets from Affymetrix or Ensembl.

SNP detection in eQTL genes.

To identify DNA sequence variants that could underlie eQTLs, we generated sequence data for the cDNA sequence and 2–5 kb upstream of exon 1 for seven of the most statistically significant cis-regulated eQTL genes from the kidney data set. We carried out direct sequencing from PCR-amplified cDNA or genomic DNA using primers designed by Primer 3.0 from Ensembl annotations. We purified PCR products by shrimp alkaline phosphatase (Promega) and exonuclease I (Promega) treatment and sequenced them directly on an ABI 3730 Sequencer (Applied Biosystems). To determine allele status in three additional inbred strains, we resequenced SNPs in promoters, exons and exon-intron boundaries on genomic DNA from WKY/Mdc, SHR/Mdc and SHRSP/Mdc strains.

URLs.

ArrayExpress is available at http://www.ebi.ac.uk/arrayexpress/.

Accession numbers.

dbSNP: unique SNP identifiers, ss35032354–ss35032394. ArrayExpress: scanned microarray data, E-AFMX-7.

Note: Supplementary information is available on the Nature Genetics website.