The publication of the complete DNA sequence of human chromosome 21 two years ago1 provided the basis for identifying every one of the genes on this chromosome. On pages 582 and 586 of this issue, Reymond and colleagues2 and Gitton and co-workers3 extend this tremendous accomplishment by analysing the expression of the mouse equivalents of these human genes in many different mouse tissues, and at several different stages of development. Chromosome 21 was chosen for these large-scale, functional-genomics analyses because of its importance in human development — Down's syndrome occurs when a person inherits three copies of chromosome 21 instead of the normal two.
Affecting one in 700 live births, Down's syndrome has a significant impact on afflicted people, their families and society. The disorder invariably results in cognitive impairment and characteristic variations in digits and in the ridge formations on the hands and feet. Triplication of chromosome 21 is also a risk factor for congenital heart disease and Hirschsprung's disease, a colon disorder, as well as many other developmental abnormalities. But it remains unknown how inheriting three copies of chromosome 21 produces this broad spectrum of problems. Finding out when and where each chromosome 21 gene is expressed during development is a crucial step towards understanding the syndrome.
Of course, it's difficult to analyse gene expression during human development. But studies of mice are a valuable alternative, not least because one then has access to every tissue at all stages of development; moreover, many human genes have equivalents (orthologues) in mice. Reymond et al.2 and Gitton et al.3 started by identifying orthologues of most of the confirmed and 'predicted' human chromosome 21 genes. Reymond et al. then used three experimental approaches to detect their expression in mice, looking at 12 adult tissues and 6 developmental stages. This extensive survey forms the basis of the group's chromosome 21 'gene atlas'.
All three approaches used by Reymond et al. are based on detecting messenger RNA (because the first step in gene expression is the production of an mRNA copy of the gene). For 'in situ hybridization', a label is incorporated into a DNA or RNA molecule that is complementary to a given mRNA. The complementary molecule (cDNA or cRNA) is then incubated with a whole mouse embryo (for 'whole-mount in situ hybridization') or with a tissue section, and binds to the target mRNA. Developing the label produces a coloured product, revealing where the mRNA was present (Fig. 1). In an embryo, resolution is at the level of a tissue or organ. Tissue-section (histological) analysis provides higher resolution, to the level of cells, but with decreased sensitivity.
Figure 1: Analysing gene expression in the heart by in situ hybridization.
a, b, Whole-mount in situ hybridization, showing expression of the mouse Sh3bgr gene, an orthologue of the human SH3BGR gene, at two different stages of embryonic development. c, A tissue section of the mouse embryo, showing the expression of the Sh3bgr gene in developing heart. E, embryonic day of development. (a, b, provided by V. Marigo; c, provided by G. Eichele.)
High resolution image and legend (29K)The third tactic used by Reymond et al. was to isolate mRNA from a tissue or organ; next, the mRNA was 'reverse transcribed' to produce a cDNA molecule, many copies of which were then generated by the polymerase chain reaction. This amplification technique, RT-PCR for short, is extremely sensitive, allowing the detection of even a few copies of an mRNA. Meanwhile, Gitton et al. used these same techniques to provide an independent look at one early developmental stage. They also extend Reymond et al.'s results by analysing the rapidly developing postnatal brain.
Several broad trends were evident in both data sets. For instance, during early embryonic development, many of the mouse orthologues of human chromosome 21 genes are expressed broadly, throughout many tissues; this pattern becomes more restricted as development proceeds. Reymond et al. also show that, in adult mice, the number of different genes expressed in a given tissue varies greatly: using RT-PCR, for example, they detect the expression of a surprising 85% of chromosome 21 genes in adult brain, but only 21% in muscle. Both groups also pinpoint specific genes that might be relevant to Down's syndrome, by virtue of their expression in the brain, limbs, heart or gut — structures commonly affected by triplication of chromosome 21.
Gitton et al. also looked further at the kinds of genes that are expressed broadly or in restricted patterns, by considering the mouse genes for which there are related sequences (paralogues) in a single-celled organism (yeast), in nematode worms and in fruitflies. They found that most of the mouse genes that had paralogues in all three lower organisms were expressed broadly or ubiquitously in the mid-gestation embryo. Only a few showed an expression pattern that was restricted to specific tissues, suggesting that these genes might have general 'housekeeping' functions that are required in all cells. By contrast, many of the genes with paralogues in the two multicellular organisms but not in yeast were expressed in tissue-restricted patterns. As a group, these regionally patterned genes were involved in cell–cell communication and signal transduction (functions that are less elaborate in yeast).
In addition, Gitton et al. looked for functional clustering — the occurrence of groups of genes that are coordinately expressed in the same tissues; such genes might be involved in common processes. They used a computational method (described in ref. 4) to search for these functional gene clusters. To do this, the authors examined large existing gene-expression databases called expressed sequence tag (EST) libraries. ESTs are short nucleotide sequences, hundreds or thousands of which are obtained from randomly picked clones in a cDNA library; each cDNA (and therefore each EST) library represents the mRNAs expressed in a particular tissue at a specific point in time. Here, Gitton et al. identified 190 relevant EST libraries, in which the chromosome 21 genes were represented by 4,186 ESTs (on average, there were 22 ESTs per gene). Using a combination of correlation and matrix analyses, the authors identified eight functional clusters of genes that were expressed to a similar degree in different libraries, suggesting that the clustered genes are regulated together.
This approach is attractive in that it uses an objective statistical assessment to find far more complex patterns in large data sets than can be observed by eye. But there are caveats. In theory, the procedure is minimally influenced by the size of the EST library, yet most of the eight gene clusters were seen only in the largest libraries. And, surprisingly, no clusters were found from the brain, despite the extensive representation of brain ESTs and the observation2, 3 that many genes are coordinately expressed in restricted brain regions. This does not negate the value of the correlations, but does suggest that the method may be influenced by other, not immediately obvious factors.
Genome-sequence databases and EST libraries, coupled with appropriate search engines, have transformed the process of discovering genes. The development of centralized gene-expression databases and computational tools such as those described here is crucial for the more complex, multi-dimensional problem of tracking the expression of these genes through space and time. Several refinements will, however, be needed to allow such databases to capture all the information about gene expression. Users need a way of assessing the quality of the data, which is inevitably uneven in such a large resource and becomes more so if multiple investigators contribute data to a central compilation. For example, Reymond et al. found that 98–99% of the analysed genes showed detectable expression by whole-mount in situ hybridization after 9.5 days of embryonic development. But, using the same procedure, Gitton et al. found a signal for only 70% of genes at this stage. Standard controls, such as reference gene sets or relative statistics (for instance, the 98% versus 70% quoted here), will be needed to evaluate a given result.
It will also be important to improve on the static, two-dimensional images of whole embryos. In the laboratory, stained embryos must be rotated to allow them to be viewed from all angles. Software analogous to that used to present three-dimensional molecular structures, combined with three-dimensional imaging by optical sectioning or other approaches, may allow comparable manipulation of these images. On the computational side, the correlation approach to gene expression is but an early step along the road to obtaining methods that will relate patterns of gene expression to developmental events.
Nonetheless, gene-expression databases such as these2, 3 are already valuable for analysing where and when genes come into play during development, and so will provide clues to what those genes do. Ultimately, these spatiotemporal patterns must be linked to structural and physiological changes that result from — and which themselves further alter — gene expression. The resulting knowledge of the roles of individual genes in specific developmental pathways must then be combined with an understanding of how development differs from normal in genetic disorders, before we can comprehend and, eventually, ameliorate these syndromes.
