Large-scale analysis of the human and mouse transcriptomes

Abstract

High-throughput gene expression profiling has become an important tool for investigating transcriptional activity in a variety of biological samples. To date, the vast majority of these experiments have focused on specific biological processes and perturbations. Here, we have generated and analyzed gene expression from a set of samples spanning a broad range of biological conditions. Specifically, we profiled gene expression from 91 human and mouse samples across a diverse array of tissues, organs, and cell lines. Because these samples predominantly come from the normal physiological state in the human and mouse, this dataset represents a preliminary, but substantial, description of the normal mammalian transcriptome. We have used this dataset to illustrate methods of mining these data, and to reveal insights into molecular and physiological gene function, mechanisms of transcriptional regulation, disease etiology, and comparative genomics. Finally, to allow the scientific community to use this resource, we have built a free and publicly accessible website (http://expression.gnf.org) that integrates data visualization and curation of current gene annotations.

Figure 1

Expression of tissue-specific genes. Genes with tissue-specific expression patterns were identified for all tissues in the human (A) and mouse (B) datasets. “Tissue-specific” was defined as expressed with AD greater than 200 in one tissue and less than 100 AD in all other tissues. Tissues were sorted by the number of tissue-specific genes found. The five tissues in human and mouse with the most tissue-specific genes are labeled. Replicate samples from one tissue were averaged, and genes and tissues were clustered by using cluster and visualized by using treeview (25). Red, up-regulated; green, down-regulated; black, median expression. Tissue labels: a = testis, b = pancreas, c = liver, d = placenta, e = thymus, f = mammary gland, g = thyroid, and h = salivary gland).

Figure 2

Differential expression of GPCRs and kinases. Pfam was used to identify GPCRs (PF00001, PF00002, and PF00003) and kinases (PF00069, PF00433, PF00454, and PF00625) from the genes interrogated in the gene expression atlas. Data were filtered to remove genes that were not expressed in the atlas (max AD < 200) and not differentially expressed (ANOVA P > 0.05), and the remaining genes were visualized as described previously. The gene identities for these Pfam families, as well as for all Pfam families, can be viewed on the web site (http://expression.gnf.org).

Figure 3

Identification of pituitary-specific response elements. The gene expression atlas was used to identify pituitary-enriched genes (Left). Genomic sequence up to 5 kb upstream of the translational start methionine was searched for conserved motifs. On the Right is a potential regulatory element identified in the upstream genomic sequence of the genes in this cluster. This element is similar to a previously described Pit1 binding site from the growth hormone 2 structural gene.

Figure 4

Potential markers for prostate cancer were identified by comparing gene expression in normal tissues with normal and tumor prostate samples. Fifty candidate makers are visualized here, and the top eight gene identities are shown.

Figure 5

Comparison of gene expression for mouse/human ortholog pairs. Putative ortholog pairs between mouse and human genes were identified by LocusLink symbol. (A) Gene expression patterns across 16 tissues for these 799 gene pairs were compared. The distribution of correlation coefficients is plotted. (B) The 427 gene pairs with correlation coefficients greater than 0.6 were sorted by tissue of maximum expression and visualized as described previously. (C) One hundred twenty-eight gene pairs have negative correlation in their gene expression pattern. The expression pattern for collagen XV is shown here. Mouse collagen XV is highly expressed in the uterus, whereas human collagen XV shows highest expression in the placenta.