Last year, data from several yeast DNA microarray experiments were pooled to create a compendium of gene expression profiles in Saccharomyces cerevisiae. Data comparisons on such a massive scale can provide rich insights into gene function and biology. Kim et al. have now pooled data from 30 different laboratories to create a similar compendium that covers ∼93% of the Caenorhabditis elegans genome. By visualizing data in three dimensions to create a gene expression landscape — in which genes that share expression patterns cluster together to form expression mountains — the authors were able to obtain a wealth of details on how genes are co-regulated in the worm.

Motivated by the need to develop high-throughput functional genomics approaches to analyse gene function, Stuart Kim and colleagues combined data from 553 microarray experiments, in which gene expression was compared between wild-type and mutant worms, and between worms grown under different conditions. To identify genes that are co-regulated, they first assembled a gene expression matrix that contains the relative expression level for each gene in each microarray experiment programme. A two-dimensional scatter plot, in which genes with similar expression profiles lie close to each other, was converted into a three-dimensional map, in which the z axis corresponds to gene density in a given area. Each gene was then assigned to a cluster, or expression mountain, and each of these was numbered — the biggest mountain was zero and the smallest was 43.

The expression landscape is visually impressive, but is it biologically meaningful? Overlaps between certain expression mountains and groups of genes that are known to share a common function are significant. In addition, the authors confirmed the validity of this representation, for example by randomizing the data and adding noise.

Once satisfied with their method, Kim et al. went off to explore the genetic landscape they had generated. They found that some mountains group together genes that are expressed in similar tissues, so for example there is a muscle mountain and a germ-line mountain. Other mountains are made up of genes with similar cellular functions — there is a histones mountain and a ribosomal genes mountain. By showing that the mountains were enriched in particular sets of genes, Kim et al. were able to attribute physiological significance for 30 out of 44 of the mountains. Zooming in on particular mountains confirmed and augmented already known genetic links between genes. For example, 89% of previously identified sperm-enriched genes clustered in mountain 4 and genes that encode principal sperm proteins, protein kinases and phosphatases fell into three separate subclusters on mountain 4.

The exploratory trips of Kim and colleagues into this expression landscape have already yielded much information on gene co-regulation, and they have amply shown the superiority of this visual data representation. New gene interactions, unexpected gene co-regulation, assignment of function to new genes and much more await future explorers.