Systems Biology with High-Throughput Sequencing Reveals Genetic Mechanisms Underlying the Metabolic Syndrome in the Lyon Hypertensive Rat

Abstract

Background

The metabolic syndrome (MetS) is a collection of co-occurring complex disorders including obesity, hypertension, dyslipidemia, and insulin resistance. The Lyon Hypertensive (LH) and Lyon Normotensive (LN) rats are models of MetS sensitivity and resistance, respectively. To identify genetic determinants and mechanisms underlying MetS, an F2 intercross between LH and LN was comprehensively studied.

Methods and Results

Multi-dimensional data were obtained including genotypes of 1536 SNPs, 23 physiological traits and more than 150 billion nucleotides of RNA-seq reads from the livers of F2 intercross offspring and parental rats. Phenotypic and expression QTL were mapped. Application of systems biology methods identified 17 candidate MetS genes. Several putative causal cis-eQTL were identified corresponding with pQTL loci. We found an eQTL hotspot on rat chromosome 17 that is causally associated with multiple MetS-related traits, and found RGD1562963, a gene regulated in cis by this eQTL hotspot, as the most likely eQTL driver gene directly affected by genetic variation between LH and LN rats.

Conclusions

Our study sheds light on the intricate pathogenesis of MetS and demonstrates that systems biology with high-throughput sequencing is a powerful method to study the etiology of complex genetic diseases.

Background

The metabolic syndrome (MetS), characterized by hypertension, central obesity, dyslipidemia, and insulin resistance, causes increased mortality due to cardiovascular and renal disease.^{1, 2} The etiology of the metabolic syndrome is complex, and is a collection of multifactorial traits each involving environmental and genetic interactions. While twin studies, familial segregation, and inter-correlation analyses have all supported the existence of strong genetic influences on the metabolic syndrome,^3–5 the majority of the genetic determinants of the metabolic syndrome remain to be identified.

Identification of the genetic contribution to complex disease is aided by integrated studies involving genetic variation, transcriptional regulation, and animal models. The Lyon Hypertensive (LH) rat was selectively bred for high blood pressure;⁶ however it has several features common to human MetS including high body weight, cholesterol, and triglycerides, increased insulin and insulin/glucose ratio, and high blood pressure.⁷ The Lyon normotensive (LN) control strain, concurrently bred for normal blood pressure from the same Sprague Dawley (SD) colony, is genetically similar to the LH; however this strain is lean, has normal plasma lipids, and is normotensive. Consequently, the Lyon Hypertensive and Normotensive strains are really to be considered as the Lyon MetS sensitive and Lyon MetS resistant rat strains, respectively, and represent a simplified multigenic model for better understanding the pathological links between MetS and its associated risk for cardiovascular disease.

Previous studies identified quantitative trait loci (QTL) related to MetS phenotypes in the LH rat, including blood pressure, body weight, plasma lipids, glucose, and insulin, among others.⁸ While mapping QTL provide valuable information, integrated systems genetics approaches can complement positional cloning approaches to identify novel genes causing complex traits (reviewed in ⁹). In this study we used a systems genetics approach to identify genes involved in traits defining MetS in the LH rat. In a single F2 intercross between the LH and LN rat strains, we determined MetS phenotypes as well as liver mRNA levels by RNA-seq. Combined mapping of phenotypic QTL (pQTL) and expression QTL (eQTL) followed by co-expression network analyses led to strong candidate genes underlying the metabolic dysfunction in the LH rat.

Materials and methods

Expanded methods are available in the Data Supplement, as indicated below.

Animals

LH/MRrrcAek, LN/MRrrcAek, and their F1 and F2 progeny were bred and maintained in an approved animal facility on a 12-hour light-dark cycle at the University of Iowa. The rats were provided chow (Teklad 7913) and water ad libidum unless otherwise noted as part of the experimental protocol. LH males were bred to LN females to produce F1 offspring, which were brother-sister mated to generate 169 F2 male progeny to be used in this study. All animal protocols were approved by the IACUC at the University of Iowa.

SNP genotyping

DNA was isolated from tail or spleen samples from each F2 offspring and parental rats (DNEasy Blood and Tissue kit, Valencia, CA). SNP genotyping was determined using a custom 1536 Illumina SNP chip, enriched with 453 SNPs selected to tag all haplotypes differing between the LH and LN genomes (Table I and Methods in the Data Supplement).¹⁰ Genotyping was performed according to manufacturer specifications at GeneSeek (©Neogen Corporation). Only SNP calls that were polymorphic between LH and LN strains, and with GenCall scores exceeding 0.4 were accepted for analysis.

Phenotyping

Beginning at three weeks of age, parental and F2 rats entered a 12-week phenotyping protocol (Table II and Methods in the Data Supplement). Briefly, body weight, blood pressure (by telemetry), and plasma measures of lipids, glucose and leptin were determined. Ear and tail snips were taken for DNA isolation, and tissues were harvested at the end of the protocol. For each phenotype, differences between the F2 offspring and parental strains were determined using ANOVA. Phenotypic (p)QTL were mapped to the genome using R/qtl and the SNP genotypes determined in the F2 rats.¹¹ Multiple testing error was corrected for by permutation testing (see Methods in the Data Supplement).

RNA-seq

Total RNA from liver tissue obtained after completing the 12 week phenotyping protocol from LH (N = 6), LN (N = 6), and F2 (N = 36) rats was sequenced on an Illumina HiSeq 2000 to produce approximately 30 million 51-bp paired-end reads per sample (Table III in the Data Supplement). The sequenced F2 rats were selected to be enriched for the extremes of the phenotype distributions. Sequences were aligned to the rat genome using Tophat (Figure I in the Data Supplement),¹² and for mapped reads, FPKM were determined using Cufflinks.¹³ Differentially expressed genes (DEGs) between LH and LN were determined using edgeR (version 3.0.0)¹⁴ and corrected for multiple testing using an FDR < 0.05. eQTL mapping was performed in the genotyped F2 offspring using FPKM data as normalized gene expression and the R/qtl and R/eqtl packages, with permutation testing to correct for multiple testing error.¹⁵ Gene Ontology enrichment analyses were performed using DAVID.¹⁶ Detailed methods can be found in the Data Supplement.

Causality tests

To infer causal relationship between gene expression and phenotypes (or another gene expression), we used the single.marker.analysis function from the Network Edge Orienting (NEO) package¹⁷ implemented in R (see Methods in the Data Supplement). In this study, the eQTL peak or mQTL (module QTL) SNPs were used as anchor SNPs. Causality tests were conducted only when traits show correlation with expression value (Pearson R ≥ 0.3). Causality was indicated when the RMSEA (Root Mean Square Error of Approximation) value was < 0.05, which indicate good model fit, and the causality score was > 0.3,^18–20 representing 2 times higher probability of causal model than any other four models. For causality tests with a large number of genes, we focused on global patterns rather than specific genes to avoid the pitfalls of multiple testing that may inflate the false positive rate.

Network construction and analyses

Unsigned weighted gene co-expression network based on the gene expression of 36 F2 rats was constructed using WGCNA package in R,²¹ including 10,088 genes with mean FPKMs >1 across the cohort. Network modules were determined with at least 30 genes and minimum height for merging modules at 0.25 to obtain moderately large modules (see Figure II and Methods in the Data Supplement). Gene ontology enrichment analysis of each module was conducted using DAVID.¹⁶ Module eigengenes were summarized by the first principal component of the module expression profiles. To determine mQTLs, clusters of genes that were regulated by the same eQTL hotspot were selected and tested for enrichment in particular modules using Fisher exact tests.

qPCR validation

Quantitative RT-PCR (qPCR) was performed on liver RNA from all available F2 rats (N=144) and parental controls for the cis-eQTL genes RGD1562963, Aqp11, Prcp, and Mapk9 (see Methods in the Data Supplement). Gene expression (2^−ΔΔCt) differences between groups were determined by ANOVA followed by post-hoc tests for pairwise significance compared to the LH parental control. Correlations between gene expression and phenotypes were determined using Spearman correlation, with nominal P values of < 0.05.

Genome sequencing and analysis

Genome sequencing and identification of variants was performed as previously described.²² Sequence data is available at the EBI Sequence Read Archive under accession number ERP002160. Sequence variants are available at the Rat Genome Database (http://rgd.mcw.edu).The Alibaba2²³ program was used to predict transcription factor binding sites using binding sites from TRANSFAC database (http://www.biobase-international.com/product/transcription-factor-binding-sites).

Data access

RNA-seq raw data and gene expression values of each sample have been deposited in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE50027.

Results

pQTL (Phenotypic Quantitative Trait Locus) mapping

To identify loci and genes contributing to the phenotypes defining MetS, we phenotyped 169 male offspring from an F2 intercross between LH and LN rats, as well as male parental LH (N = 13) and LN rats (N = 20), for traits including blood pressure, plasma lipid, glucose, and leptin levels, and body weight and growth (Table IV and V in the Data Supplement). For all blood pressure, body weight and growth measures, and plasma leptin levels the F2 cohort had an intermediate phenotype, significantly differing from both LH and LN strains (P < 0.01). At 18 weeks of age plasma triglyceride levels in the F2 rats were more similar to LN rats, both with significantly lower levels than LH rats, whereas plasma cholesterol levels at 12 and 16 weeks of age were similar or higher in the F2s than LH.

453 SNPs were used to tag all haplotypes differing between LH and LN across the genomes of the two strains^{10, 24} (Table I in the Data Supplement). pQTLs of 23 phenotypes were calculated based on the F2 cohort and analyzed using R/qtl¹¹. Seventeen pQTLs exceeding the 5% genome-wide threshold for significance (defined by permutation testing) were identified for plasma cholesterol, plasma leptin, body weight, growth curve, heart rate, and blood pressure on chromosomes 1, 2, 10, 15, and 17 (Figure 1 and Table VI in the Data Supplement). Overlapping QTLs for plasma cholesterol and body weight were identified on RNO10, while overlapping QTLs for plasma leptin and blood pressure were identified on RNO17. All other QTLs were non-overlapping.

Figure 1.

Lod plots of representative pQTL. Descriptions of phenotypes are in Table V in the Data Supplement. (a) Adj BW = length adjusted body weight at 16 weeks (wks) old, (b) Chol = plasma cholesterol (mM) at 18 wks old, (c) SBP = systolic blood pressure (mmHg) at 14 wks old, (d) Leptin = plasma leptin (ng/ml) at 18 wks old. Blue line denotes 5% genome-wide significance level; red line denotes 1% genome-wide significance level.

eQTL (Expression Quantitative Trait Locus) mapping and parental rat DEG (Differentially Expressed Gene) analyses

Liver mRNA from parental (N=6) and F2 (N=36 with extremes of the phenotypes) rats were sequenced to produce ~30 million paired-end reads (51 bp) per sample. FPKM (fragments per kilobase of transcript per million fragments mapped)¹³ was used as a normalized quantitative measure of gene expression and eQTL mapping was performed using the R/eqtl package.¹⁵ We identified 1264 suggestive eQTLs (Lod > 3.82) and 276 significant eQTLs (Lod > 4.84). The chromosomal locations of the regulated genes were plotted against the chromosomal locations of their associated SNPs (Figure 2) to visualize both cis-eQTLs (diagonal) and trans-eQTL hotspots (vertical). For example, we identified a single SNP (ENSRNOSNP962219), on RNO17 as an eQTL hotspot. This SNP is associated with 66 trans-eQTLs and one cis-eQTL at 29.9 Mb on RNO17 that exceed 5% genome-wide significance. The cis-regulated gene, RGD1562963, will be discussed in detail below. The genes sharing this eQTL hotspot have significant GO (gene ontology) enrichment in genes involved in mitochondrial function including oxidative phosphorylation (FDR = 0.017), and comprise 22.4% of the 67 genes. Other trans-eQTL hotspots were identified on chromosomes 7, 8, and 14.

Figure 2.

Genome-genome plot of liver eQTL. The location of genes (y-axis) and their peak eQTL SNPs (x-axis) are shown accordingly. Red dots denote eQTL above 1% genome-wide significance; blue dots denote eQTL above 5% genome-wide significance. The red triangle denotes the location of SNP ENSRNOSNP962219 as an example of eQTL hotspot on RNO17.

In liver from LH and LN rats, 610 DEGs were identified with a false discovery rate (FDR) of 5% (Figure III in the Data Supplement). The numbers of genes upregulated and downregulated in the LH rats were similar (322 and 288, respectively). Genes upregulated in the LH rat were significantly enriched for immune response and inflammatory processes while downregulated genes were enriched for fatty acid metabolism (Figure III in the Data Supplement).

Integrate pQTLs, cis-eQTLs and DEGs to identify candidate causal genes

Integration of the liver gene expression with SNPs and phenotypes was used to discover candidate genes responsible for the etiology of MetS. Several criteria were set as filters to select the most promising eQTL-gene pairs causing MetS in the LH rat. First, the genes must be differentially expressed between LH and LN strains. Second, they must have at least one eQTL where the peak SNP co-localizes within the pQTL confidence interval; third, the average F2 FPKM must be > 1, to avoid potential noise in RNA-seq of genes with very low expression. Finally, to address whether the regulated genes are likely to cause the phenotypic outcomes, the Network Edge Orienting (NEO) package¹⁷ was used to infer the causal relationships between genes meeting the above criteria (N=25) and all measured phenotypes, using the eQTL peak SNPs as anchors. Causality tests (LEO.NB.SingleMarker) were calculated between each of the genes and any correlated phenotypes (R ≥ 0.3). Genes causally affecting at least one phenotype were ranked by their maximum scores, using a previously suggested causality score threshold of 0.3,^18–20 equivalent to a 2-fold higher probability compared to any other model (for example, the phenotype modifies the gene’s expression). This approach identified 17 genes (66 SNP-gene-phenotype trios) (Table 1 and Table VII in the Data Supplement). Of these 17 genes, 6 are cis-regulated and fall within pQTLs on RNO 1, 2, 10, and 17, while the remaining 11 are regulated in trans. One gene (Aqp11) has a cis-eQTL that shares the same peak SNP with pQTLs, and the average number of phenotypes causally affected by these 17 genes was 3.3.

Table 1.

Candidate genes identified for MetS traits identified by integrated mapping and network analyses.

Gene Name	eQTL Chr	eQTL LOD	Maker name of peak	Type	Peak position	LH mean FPKM	LN mean FPKM	DEG FDR	Gene Chr	Gene start	Gene end	# Anchor*	# Traits†	Causality score‡
Aqp11	1	7.15	ENSRNOSNP70864	cis	154627781	7.0	12.3	1.4E-04	1	154973798	154983989	2	2	0.68
Pigl	10	7.13	ENSRNOSNP2796578	cis	49338801	1.5	3.9	1.1E-09	10	48629286	48689381	1	1	0.7
Aqp11	1	7.11	ENSRNOSNP45308	trans	136552144	7.0	12.3	1.4E-04	1	154973798	154983989	2	2	0.68
RGD1562963	17	6.79	ENSRNOSNP962219	cis	29903973	1.7	1.1	4.4E-02	17	29733271	29746769	1	1	0.311
Iah1	17	6.34	ENSRNOSNP962219	trans	29903973	79.5	50.2	2.0E-06	6	41875020	41882235	1	4	1.1
Neurl4	10	6.04	ENSRNOSNP2796827	trans	75516973	3.8	7.6	1.6E-08	10	56745822	56757435	2	9	1.14
Pex11b	2	5.29	ENSRNOSNP2786744	cis	191277899	7.8	13.8	2.6E-04	2	191437355	191446244	1	1	0.705
Prcp	1	5.26	ENSRNOSNP2784200	cis	151110344	5.6	3.1	8.7E-09	1	149711289	149763868	2	1	0.588
LOC303140	10	5.07	ENSRNOSNP2796873	trans	80398271	0.4	1.4	7.5E-12	10	39407540	39417112	1	1	1.02
Mapk9	10	5.05	ENSRNOSNP2796424	cis	36183842	15.1	27.2	1.9E-06	10	35344672	35384319	2	9	1.48
Neurl4	10	4.98	ENSRNOSNP2796437	trans	37954559	3.8	7.6	1.6E-08	10	56745822	56757435	2	9	1.14
Phldb3	1	4.93	ENSRNOSNP2783493	trans	44878370	2.8	1.8	1.5E-04	1	79893158	79910727	1	1	0.35
Aspa	10	4.74	ENSRNOSNP2796902	trans	83698648	0.7	2.0	2.4E-09	10	60178512	60199209	1	2	0.352
Popdc2	2	4.59	ENSRNOSNP2785628	trans	51496838	2.2	0.7	1.2E-15	11	64154287	64170282	1	3	0.479
Prmt6	2	4.48	ENSRNOSNP2786609	trans	175912348	1.1	2.0	1.5E-03	2	205909919	205915148	1	1	0.837
Rtn4ip1	10	4.40	ENSRNOSNP313093	trans	47678382	8.0	2.4	5.1E-26	20	47818499	47857741	1	1	0.686
Mapk9	10	4.31	ENSRNOSNP2796840	trans	77352099	15.1	27.2	1.9E-06	10	35344672	35384319	2	9	1.48
Prpsap2	10	4.05	ENSRNOSNP2796516	cis	46015612	5.3	3.8	8.2E-03	10	47890410	47925430	1	1	0.886
Ivns1abp	17	3.96	ENSRNOSNP962219	trans	29903973	24.5	37.2	1.7E-02	13	66223323	66235533	1	6	0.956
Supt4h1	17	3.93	ENSRNOSNP962219	trans	29903973	25.2	18.7	2.4E-02	10	76030806	76036988	1	7	1.11

^*Total number of anchors for causality test of the gene.

^†Number of unique phenotypes that show causal relationship.

^‡The maximum causality score to phenotypes. The eQTL hotspot in RNO17 (ENSRNOSNP962219) is highlighted in bold.

Cis-eQTLs are strong candidates for causing pQTLs, when the cis-regulated genes fall within the pQTL intervals. To further investigate causal cis-eQTL, we performed qPCR for four of the six cis-regulated genes (Aqp11, Prcp, Mapk9, and RGD1562963) in the entire cohort of F2 rats completing the study (N = 144). Using the nearby SNP as a marker for the genotype for each gene, all four genes were found to have genotype-specific expression in the F2 rats (Figure 3a–d), with allelic expression similar to that found in liver from the parental LH and LN rats.

Figure 3.

qPCR validation of cis-eQTLs and their correlation to MetS traits. (a) Aqp11 in F2 rats expression grouped by ENSRNOSNP70864 genotype on RNO1. (b) Prcp expression in F2 rats grouped by ENSRNOSNP2784200 genotype on RNO1. (c) Mapk9 expression in F2 rats grouped by ENSRNOSNP2796424 genotype on RNO10. (d) RGD1562963 expression in F2 rats grouped by ENSRNOSNP962219 genotype on RNO17. (e) Gene expression-phenotype correlations for cis-regulated genes in the F2 rat cohort (N=144). All expression is relative to a standard SD whole-rat universal reference sample.* P < 0.05.

Aqp11 and Prcp both fall within QTL on RNO1 for measures of body weight (Figure 1a and Table VI in the Data Supplement). Both RNA-seq and qPCR determined liver Aqp11 expression is downregulated in the LH compared to the LN allele (Table 1; Figure 3a). Furthermore, expression of Aqp11 is negatively correlated with a co-localized pQTL trait – Adj BW 16 wk (nominal P < 0.05; Figure 3e). Conversely, Prcp expression is significantly upregulated in LH liver (Table 1; Figure 3b). Prcp expression is also significantly correlated with all traits having co-localized pQTL (Adj BW 12 wk; Adj BW 16 wk; Growth AUC; Figure 3e). These data suggest that genetic variation causing concurrent downregulation of Aqp11 and upregulation of Prcp cause obesity in the LH rat. Of note, both genes have been reported as MetS-related genes through mouse knock out studies (Table VII in the Data Supplement).^25–27

Network construction and module QTL (mQTL) analyses

While the integrated eQTL/pQTL approach identified regulated genes with gene expression and sequence variation between the LH and LN strains, as well as functional relevance to the mapped traits, the phenotypic variation due to single genes in a multifactorial disease like MetS may be limited. As such, the expression data generated from RNA-seq can also be used to unveil the mechanisms of etiology of this disease at the network level. A weighted gene co-expression network was constructed based on the gene expression measures of the 36 F2 individuals, consisting of 10,088 genes with average FPKM > 1. Fifteen modules were obtained from the network analysis. Figure 4a is a global view of the network. Together the top 5 modules contain 7781 genes, over 77% of all genes included in the dataset, and have strong gene ontology enrichment (Figure 4b). Other modules, such as the cyan and pink modules, have strong gene ontology enrichment in immune regulation and inflammatory response, similar to that of the enrichment in the upregulated genes in the LH compared to LN rats (Figure III in the Data Supplement).

Figure 4.

Co-expression network studies of F2 liver. (a) Multidimensional scaling based on distance matrix of the expression of all genes in the co-expression network. Genes are colored according to their module color. (b) Gene ontology enrichments for the five largest modules using all genes used in network as background. (c) Module QTLs (mQTLs). Each bar represents the number of genes (Lod > 3) regulated by the eQTL hotspot (x-axis label) by their corresponding module colors. Only mQTLs with at least 50 regulated genes are plotted. Red arrow denotes mQTL co-localizing with a pQTL.

Module eigengenes (ME) were calculated to represent the gene expression of each module, and correlations between multiple module eigengenes and multiple MetS-related traits were observed (Figure IV in the Data Supplement). For example, the grey module eigengene has strong correlation with leptin, glucose, white adipose and multiple blood pressure related traits. Module QTLs (mQTLs) were also determined, defined as an eQTL-peak SNP of multiple genes that show significant module enrichment based on the Fisher exact test (see Methods in the Data Supplement). In this analysis, we relaxed the eQTL Lod score threshold to Lod > 3 in order to increase the power of mQTL detection. As shown in Figure 4c, genes sharing the same eQTL peak SNP are enriched in a limited number of modules.

Of the mQTLs, ENSRNOSNP962219 is located in a pQTL region for blood pressure and plasma leptin on RNO17 (Figure 1) and has the largest number of genes regulated by this eQTL hotspot. Other mQTLs, including the strong trans-eQTL hotspots on RNO 7, 8, and 14, are not located in pQTL regions and thus were not studied further. ENSRNOSNP962219 is an mQTL of both the turquoise (P = 3.7×10⁻⁷, Fisher exact test) and blue modules (P = 2.0×10⁻⁴², Fisher exact tests) (Figure 4c). These two modules are the top two largest modules and represent distinct functional categories. The turquoise module is enriched for genes involved in transcriptional regulation; the blue module is enriched for genes associated with translation, ribosomes, mitochondria and oxidative phosphorylation (Figure 4b). As shown in Figure 5a, these two modules also show significant correlations with multiple phenotypes, including body weight, white adipose tissue (WAT), blood pressure, and plasma triglycerides. To test if these two modules have causal relationship with phenotypes, we used ENSRNOSNP962219 as the anchor SNP to calculate the causality scores of the module eigengenes to the correlated phenotypes. As shown in Figure 5b, the blue module shows a causal relationship to white adipose, while the turquoise module shows a causal relationship to multiple phenotypes including body weight, white adipose and blood pressure. Therefore these two modules are predicted to causally affect most of the correlated phenotypes.

Figure 5.

Turquoise and blue modules have causal relationship with phenotypes. Description of phenotypes are in Table V in the Data Supplement. (a) Heatmap of correlation coefficients between phenotypes and eigengenes of turquoise and blue modules. The correlation coefficient and nominal P value (in bracket) are shown in each cell. P values ≤ 0.01 correspond to FDR < 0.1 based on Benjamini-Hochberg FDR correction. (b) Causality scores of blue and turquoise eigengenes with their correlated phenotypes. Colors represent module colors. Vertical dashed line denotes the cutoff to infer causal relationships. (c) Causality scores of the cluster of genes sharing the ENSRNOSNP962219 eQTL hotspot with their correlated phenotypes. For each phenotype, the causality scores of all eligible genes in one specific module to this phenotype are shown as a box. Genes in turquoise and blue modules analyzed separately and denoted by color. The number above each box denotes the number of genes eligible for causality tests (see Methods in the Data Supplement), and only boxes with at least 20 eligible genes are shown. The horizontal dotted line denotes the cutoff to infer causal relationships.

To further study the role that mQTL ENSRNOSNP962219 plays on gene expression regulation and MetS phenotypes, we focused on the 293 genes sharing the eQTL peak SNP ENSRNOSNP962219 at a Lod > 3. Of these 293 genes, 279 (95.2%) fell into either the turquoise (118 genes) or blue modules (161 genes) (Figure 4c). For each of these 279 genes, we conducted causality tests to determine whether the gene’s expression causally affects phenotypes that are correlated with this specific gene. We found 136 (84.5%) of the 161 genes in blue module causally affect white adipose, and more than 80% of the 118 genes in turquoise causally affect adjusted body weight and diastolic blood pressure. A number of genes show causal relationship to other phenotypes (Figure 5c). These results are consistent with the causality tests using eigengenes, suggesting these two modules are responsible for many MetS phenotypes in the LH rat. Furthermore, it suggests the turquoise network has pleiotropic effects on common MetS phenotypes.

RGD1562963 is a potential driver gene of the gene cluster sharing a common eQTL hotspot within pQTL for MetS traits

As described above, ENSRNOSNP962219 defines a trans-eQTL hotspot, suggesting genes in the hotspot are affected, directly or indirectly, by a common driver gene. In principle, the driver gene would be regulated in cis (i.e. a cis-eQTL) at ENSRNOSNP962219 and have causal relationship to trans-regulated genes sharing the same eQTL peak. To find the driver gene, we selected the 10 genes in the 95% eQTL hotspot confidence interval (RNO17:1-39 Mb) with an eQTL Lod > 3 (relaxed criteria). Only one cis-eQTL gene, RGD1562963, exceeded the Lod 4.8 threshold for significance (Lod 6.8) (Figure 6a). Two other cis-eQTL genes had suggestive linkage, Sfxn1 and Prelid1 (both with Lod 3.93). Furthermore, only RGD1562963 is differentially expressed between the parental LH and LN strains, and shows allele specific regulation in the F2 cohort (Table 1; Figure 3c).

Figure 6.

Determination of the driver gene for the ENSRNOSNP962219 eQTL from ten putative cis genes. (a) Lod scores of putative cis-eQTL in the trans-eQTL interval. (b) Haplotypes and genes located within the eQTL confidence interval. Orange box denotes the eQTL region, and vertical red dashed line denotes the eQTL peak. The haplotype blocks were generated through whole genome sequencing of LH and LN strains.¹⁰ (c) The number of SNVs (including SNPs and indels) for each candidate driver gene, including the 5 kb upstream genomic region of each gene. The genes are ordered according to their genomic locations. (d) Number of regulated trans genes in this gene cluster by each of the candidate driver genes based on causality tests. Causality score cutoff of 1 are used to infer the causal relationship.

If a gene is cis-regulated, it also must have sequence variation regulating its expression. Genome sequencing of both the LH and LN strains identified approximately 643,000 SNVs and 327,000 indels between the strains.²² These variants cluster into haplotypes derived from different ancestral chromosomes.¹⁰ In the 95% confidence interval of the eQTL hotspot on RNO17, only three small haplotypes differ between LH and LN; one of these haplotypes (29.7 – 30.3 Mb) contains RGD1562963 (Figure 6b). There are 86 sequence variants between LH and LN rats in the genomic interval encoding RGD1562963 and the 5 kb upstream. In comparison, Sfxn1 has only one intronic indel and Prelid1 has no SNVs between the strains (Figure 6c). Three non-synonymous variants (V36I, R135H, and C139Y) were identified in RGD1562963, although functional variant prediction tools (Polyphen and SIFT^{28, 29}) predict these variants to be benign. We also identified 3 variants within relatively conserved regions shown as conservation score peaks on UCSC conservation track based on comparisons of 9 vertebrates, and 19 variants within putative transcription factor binding sites for PHO2, Oct-1 and GATA-1 (Alibaba2²³) (Table VIII and Figure V in the Data Supplement). One variant (SNV1) in the predicted promoter of RGD1562963 fell into both categories and is a strong candidate for the functional variant regulating RGD1562963 expression.

The data implicate RGD1562963 as being cis-regulated in LH rats. However, to be the driver gene of the trans-eQTL cluster, it must also affect downstream trans-eQTL genes. Therefore, we performed causality testing between RGD1562963 and the 278 trans-eQTL genes that are located on different chromosomes (Lod > 3). RGD1562963 is predicted to causally affect 100 of the 278 trans-eQTL genes (Figure 6d). These results support RGD1562963 as the likely driver gene of the gene cluster sharing the ENSRNOSNP962219 eQTL hotspot. Together the data indicate the genetic dysregulation of RGD1562963 directly affects the turquoise and blue modules through the trans-eQTL genes, resulting in a cascade of dysregulation of the downstream module genes, which in turn affect the MetS traits.

Discussion

The etiology of the Metabolic Syndrome is complex, as it is a collection of several underlying multifactorial traits. To identify novel genes and pathways that underlie the individual components characterizing MetS, as well those that could explain the co-occurrence of these traits, we applied a systems genetics approach in LH and LN rats, inbred strains that show genetic MetS susceptibility and resistance, respectively. Through integrating pQTL and eQTL from a segregating F2 intercross between LH and LN, and parental gene expression differences, followed by co-expression network analysis and tests for causality, we identified 17 genes predicted to be causal for at least a subset of the MetS phenotypes. We identified several cis-regulated genes that fall within pQTL for MetS traits which are strong candidates for the pQTL (Table 1, Figure 3). Furthermore one of these cis-regulated genes (RGD1562963) also falls within a trans-eQTL hotspot on RNO17 and is a strong candidate driver gene of downstream gene regulation (Figure 6). The genes in the trans-eQTL hotspot fall within modules that are causally related to the measured MetS traits, with function related to global transcriptional regulation and mitochondrial oxidative phosphorylation, providing candidate mechanisms underlying a genetic component of MetS in the LH rat model (Figure 5).

The study identified pQTL for MetS traits on RNO1 (body weight measures and plasma cholesterol), 2 (growth, plasma cholesterol, heart rate), 10 (body weight measures and plasma cholesterol), 15 (heart rate), and 17 (blood pressure measures, plasma leptin). Comparing these results with our previous study,⁸ we found concordance of many of the pQTL identified in this study. However, there are also differences. For example, we did not confirm the previously mapped blood pressure QTL on RNO2 and RNO13. Furthermore, while body weight, plasma lipids and blood pressure all mapped to RNO17 in the previous study, only the blood pressure QTL were replicated. Some of the differences could be due to differences in cohort size, with the previous study including over 300 animals. Furthermore, the phenotypes in the previous study were measured at 29–31 weeks of age, compared to 12–18 weeks of age in the present study, which could significantly influence the QTL detection. For instance, it is known that the blood pressure in LH continues to increase between 18–29 weeks of age.^{7, 30} The fact that blood pressure solely mapped to RNO17 at a younger age could suggest that the RNO17 locus initiates hypertension, while loci on RNO2 and RNO13 involve disease progression evident at later ages.

In the pQTL for body weight and growth on RNO1 we identified two cis-eQTL genes of note - Prcp (also known as angiotensinase C) and Aqp11 (aquaporin 11). Prcp encodes a lysosomal prolylcarboxypeptidase, which regulates both the renin-angiotensin and kallikrein-kinin systems.³¹ It has more recently been shown to be a key enzyme involved in the degradation of α-MSH (α-melanocyte-stimulating hormone) to an inactive form unable to inhibit food intake.³² Deletion of Prcp in mice reduces body weight and attenuates the metabolic effects of diet-induced obesity.^{25, 27} It is significantly upregulated in LH compared to LN liver (fold change = 1.8; FDR = 9x10⁻⁹), and its expression is nominally correlated to the QTL traits, including body weight and plasma cholesterol, making it a candidate causal gene for the QTL (Figure 3e).

Another cis-regulated gene in the RNO1 pQTL is Aqp11, a member of the aquaporin major intrinsic protein family that facilitates the transport of water and small neutral solutes across cell membranes. As shown in Figure 3e, Aqp11 expression is downregulated in the LH, correlated with body weight, and causally affects leptin level and bodyweight. LH rats have a variant in the 3′UTR (1g.154973845T>C) as well as in the 5′ upstream region (1g.154985253A>T) of Aqp11 that are in positions with conserved sequence in rat, mouse, human, dog, and opossum. Of note, the 5′ upstream variant is located in a predicted transcription factor binding site for C/EBPalp, which has been reported to bind to the promoter and modulate the expression of the gene encoding leptin, providing a possible causal variant for altered leptin levels.³³ Interestingly, liver specific knockout mice showed increased HDL cholesterol as well as periportal hepatic rough endoplasmic reticulum (RER) vacuolization associated with ER stress.²⁶

The current and previous studies found QTL for multiple MetS traits on RNO17, which were replicated in consomic rat models.^{8, 34} Therefore, we speculate there may be a gene or genes having pleiotropic effects on the MetS spectrum. In this study, we identified a cis-regulated gene on RNO17, RGD1562963, as a putative driver gene regulating multiple downstream genes. The trans-eQTL genes fall within two co-expression modules which are causally related to body weight and blood pressure measures. Therefore we speculate RGD1562963 dysregulation causes a cascade of downstream gene dysregulation that ultimately impacts body weight and blood pressure. However it must be noted that RGD1562963 expression alone was not significantly correlated with these traits; it showed only a nominal correlation to plasma cholesterol levels (Figure 3) which was not significant after correction for multiple testing of 23 traits. One possibility for these results is that RGD1562963 may be a driver gene but unrelated to the LH phenotypes. Another possibility is that the expression of MetS involves multiple pathways and the variation in a single gene, without considering its downstream effectors, may be insufficient to identify a causal relationship. The expression difference in RGD1562963 between the LH and LN genotypes, while consistent across our studies, is relatively modest (Figure 3). There may not be enough variation in RGD1562963 alone to detect significant correlation with the phenotypes, but its dysregulation may induce a cascade of downstream gene dysregulation (ie the genes in the trans-eQTL hotspot), and the conglomeration of these genes show causal relationships to the MetS traits.

RGD1562963 has no functional annotation. Its ortholog in the human (C6Orf52) is also only annotated as an open reading frame with no predicted function; orthologs have also been identified in several other species. The closest sequence homology with RGD1562963 or its orthologs is with TRNAU1AP, coding for an RNA-binding protein in a complex that is required to generate selenoproteins, which are important antioxidants such as Glutathione peroxidases and Thioredoxin reductase.^{35, 36} The RNA binding-domain characterizing TRNAU1AP (RRM domain) also has a role in regulating transcript stability.³⁷ Both functions directly relate to the functional enrichment of the blue (mitochondrial oxidative phosphorylation) and turquoise (transcriptional regulation) modules. However, the sequence similarity between RGD1562963 (and its orthologs) and TRNA1AP do not include the RNA binding domain or any known functional domain or motif, thus predicting its function is premature and will be the focus of future studies.

The expression of RGD1562963 is also strongly associated with the blue co-expression module, including genes involved in mitochondria and increased oxidative phosphorylation. This is evidenced by the mRNA upregulation of several NADH dehydrogenase genes found in the blue co-regulation module. Interestingly, genes involving oxidative phosphorylation have also been found to be upregulated in livers from diabetic obese patients as compared to diabetic lean patients.³⁸ While increased respiration is likely aimed at maintaining homeostasis, chronic activation could eventually lead to opening of the mitochondrial permeability transition pore and apoptosis,³⁹ causing further defects in lipid metabolism in the liver. While the role of the mitochondria is well established in most traits characterizing MetS, further studies are needed to elucidate the mechanism of RGD1562963 and its downstream effects on the metabolic syndrome.

Transcriptome studies only capture a temporal and spatial snapshot of gene expression. Some gene expression variation may be due to a consequence of the MetS state, rather than driving MetS. Furthermore, cell types may differ in the groups being compared. For example, there is significant enrichment of DEGs between LH and LN rats involved in inflammatory response; this may be due to immune cell infiltration into the liver rather than liver gene expression differences. To address these issues, gene expression could be measured prior to significant disease onset. Because the phenotypes are present by 5 weeks of age in the LH,⁴⁰ a temporal study at early ages may better define the time of MetS initiation and provide more homogeneous cell types in the liver.

Systems genetics approaches are predictive and hypothesis generating, but require experimental validation. From the F2 cohort of 169 male rats, a subset of 36 rats was selected for the eQTL studies. While these numbers are in line with studies in RI strains, they are low for many eQTL studies in F2 cohorts. While the rats were selected based upon phenotypic extremes, power may be insufficient to detect some eQTL, leading to possible Type II error. However, combining the eQTL mapping with systems biology approaches have led to both genes and pathways that are strong candidates for functional validation. Additional experimental validation is required to prove the role of RGD1562963 as a driver gene of the trans-eQTL hotspot and its role in MetS in the LH rat. For example, siRNA knockdown of RGD1562963 in hepatocytes or in vivo should alter the expression of genes in the trans-eQTL hotspot and affect liver metabolism. These studies are ongoing. Studies in congenic or knockout strains involving RGD1562963 will be necessary to prove a causal role for this novel gene in the MetS traits in the LH rat.

This study uses a unique rat model of MetS (the LH and LN inbred rat strains) to integrate genetic, transcriptomics, and systems genetics approaches in the identification of candidate genes and mechanistic pathways causing the diverse phenotypes defining MetS. Through our studies we identified cis-regulated genes that are strong causal candidates for determining body weight and plasma lipid levels. One of these cis-regulated genes, RGD1562963, is also a putative master regulator of downstream pathways involving global transcriptional regulation and mitochondrial function. Future studies that validate and determine the role of these genes in the LH rat facilitates not just the translational studies of MetS-susceptibility genes in humans, but also generalized disease mechanisms that offer in-roads to personalized therapies.