Hypothalamic transcriptomes of 99 mouse strains reveal trans eQTL hotspots, splicing QTLs and novel non-coding genes

Abstract

Previous studies had shown that the integration of genome wide expression profiles, in metabolic tissues, with genetic and phenotypic variance, provided valuable insight into the underlying molecular mechanisms. We used RNA-Seq to characterize hypothalamic transcriptome in 99 inbred strains of mice from the Hybrid Mouse Diversity Panel (HMDP), a reference resource population for cardiovascular and metabolic traits. We report numerous novel transcripts supported by proteomic analyses, as well as novel non coding RNAs. High resolution genetic mapping of transcript levels in HMDP, reveals both local and trans expression Quantitative Trait Loci (eQTLs) demonstrating 2 trans eQTL 'hotspots' associated with expression of hundreds of genes. We also report thousands of alternative splicing events regulated by genetic variants. Finally, comparison with about 150 metabolic and cardiovascular traits revealed many highly significant associations. Our data provide a rich resource for understanding the many physiologic functions mediated by the hypothalamus and their genetic regulation.

Keywords: RNA-Seq; eQTL; evolutionary biology; genomics; high fat diet; hypothalamus; mouse; neuroscience; obesity; trans-eQTL.

Figure 1.. Genomic properties of novel genes are similar to known non coding genes.

Novel genes and isoforms are defined by Cufflinks class code 'u' and 'j' respectively. Distributions of transcript length (A), and maximal hypothetical peptide length (B) of novel genes (yellow), new isoforms (purple), known non coding transcripts (dashed line) and known coding transcripts (solid line). Transcriptional complexity (number of transcripts per locus, (C) and splicing complexity (number of exons per transcript, (D) of novel genes, novel isoforms, known coding and know non coding transcripts. DOI: http://dx.doi.org/10.7554/eLife.15614.004

Figure 2.. Genetic regulation of expression in the mouse hypothalamus.

(A) Number of genes affected by trans (blue), local (yellow) or both (striped) variants as a function of statistical threshold. (B) Gene level expression quantitative trait loci (eQTL, top) but not transcript specific (isoQTL, bottom) show trans eQTL hotspots. Density shows the number of interactions at lower statistical thresholds (1e–6), red shows interactions passing 1E-12 threshold. Yellow indicates cis acting variants. (C). Genetic regulation occurs on every level, but gene level regulation is more prevalent than transcript specific cases. Supplementary figure shows correlations between allele expression in F1 and local eQTLs identified in HMDP hypothalamus. DOI: http://dx.doi.org/10.7554/eLife.15614.006

Figure 2—figure supplement 1.. Allele specific expression in whole brain correlation to local eQTL.

DOI: http://dx.doi.org/10.7554/eLife.15614.007

Figure 3.. Alternative splicing in the mouse hypothalamus.

Alternative splicing events were classified (see Materials and methods) into 7 types: alternative 3’ splice (A3, blue), alternative 5’ splice (A5, purple), alternative first exon (AF, orange), alternative last exon (AL, brown), mutually exclusive exons (MX, black), retained intron (RI, green), and skipped exon (SE, dark red). All events were quantified in each sample for percent spliced in (PSI). (A) Number of alternative splicing events of each type (solid color), and number of genes affected by these events (light color). (B) Example of partial exon skipping in Colq gene. DBA shows the complete inclusion of the exon (therefore PSI = 1), while in C57BL/6 there is partial exon skipping (PSI = 0.78). (C) Number of alternative splicing events with and without local QTL signal (solid and light color respectively). (D) Alternative splice QTLs are mapping to the same chromosome, for all types of events, indicating that most of genetic regulation is by local (and likely cis acting) variants. (F) Distance between most significant SNP for each event and gene start. The largest effect is typically within 1 Mb of the gene. (G) An example of mapping of mutually exclusive exon event in Nnat gene mapping to SNP rs32019082. (E) Distribution of all PSI values of each event type in all samples. DOI: http://dx.doi.org/10.7554/eLife.15614.008

Figure 4.. eQTL mapping suggests trans eQTL hotspots in the hypothalamus regulate expression of hundreds of genes.

(A) Mouse genome was broken into 100 kb bins. The plot presents genome wide counts of genes which expression is associated with SNPs in that region, in trans. (B) Zoom of trans eQTL locus on chromosome 15. Peak SNP (associated with most genes in trans, rs31703733) is shown in red, color of other SNPs indicates r² to rs31703733. Lower track shows the 10 genes which expression is associated with rs31703733 locally. C,D,E pertain to the 10 genes associated locally with rs31703733 and therefore potentially mediate the trans effects. (C) Summary table about each gene. (D) Heatmap showing correlation of expression between genes associated with rs31703733 locally, and genes associated with rs31703733 in trans. Color indicates Pearson correlation coefficient. (E) Example correlation between potential regulator (RApgef3) and trait (HDL levels). DOI: http://dx.doi.org/10.7554/eLife.15614.010

Figure 5.. Expression of long non coding RNAs in the hypothalamus is phenotypically relevant.

(A) Expression of heatmap known non coding RNAs and novel isoforms of these genes n the HMDP. Six lncRNAs (top cluster, Meg3, Gm26924, Snhg4, Miat, 6330403K07Rik, and Malat1) are highly expressed in almost all samples. (B) Novel isoforms of lncRNAs are expressed at a similar level of known ones. (C) Long non coding RNAs are associated with multiple phenotypes in the HMDP. (D) An example of association between a non coding RNA C330006A16Rik and average food intake. DOI: http://dx.doi.org/10.7554/eLife.15614.011

Figure 6.. RNA editing is prevalent at the mouse hypothalamus at low levels.

(A) A total of 8462 editing sites were identified in the HMDP, with A to G accounting for >70% of the modifications. (B) Number of sites identified in each strain (color coding as in A). (C) Editing level at 90 sites, that were detected in at least 70% of the samples, were mapped. Heatmap shows variation in editing in these sites among the strains. (D) An example of an edited site in Ociad1 gene, and its genome wide mapping result. DOI: http://dx.doi.org/10.7554/eLife.15614.012

Figure 7.. Groups of genes are associated with multiple related phenotypes in HMDP, although not necessarily enriched for GO ontology or specific pathways.

(A) Fraction of co-shared genes among the 500 genes most associated with a phenotype. (B) Enrichment analysis of the top 500 genes associated with each of the 150 phenotypes results in a small number of significant associations. DOI: http://dx.doi.org/10.7554/eLife.15614.013

Figure 8.. RNA-Seq analysis framework.

General workflow used for analysis of RNA-Seq data in this study. Initial demultiplexed samples (fastq files) were aligned to the mouse genome with STAR, merged in one file per strain, and transcripts assembeled with cufflings. The resulting assembly files (one from each strain) were merged with GENECODE M2 annotation into unified assembly. The abundance of each transcript in the unified assembly was estimated in sample specific alignment files. DOI: http://dx.doi.org/10.7554/eLife.15614.015