A

(Isolated TF's not shown)

1 10 100 Numher ot Regulators

All promoters * N(k) = 2140 k TF-'IN" • N(k) = 30 k'i T

Yeast transcription regulation network

DMA/RNAyPralfiiil Biûeysithesis Environmental RffspçKiw Mataboliam Call Cyclo

DsratopmentBt Pioççsses

Figure 8. Transcription factor network of Saccharomyces cerevisiae. A) Diagram of the connections between 106 TF's. This graph is implied by the data reported in Lee et al.36 The connections were inferred from the data using the thresholding criteria used in the original paper. Only the connected graph is shown. The characterization of theTFs into categories (color coding) is according to the paper. B) Power law fit of both the "core" network amongTFs (described in the text and diagramed in A) for both "in" (round, pink-center points, pink line) and "out" (square points, blue line) networks. This is compared on the same graph to the overall promoter distribution (shown in A). C) Conceptual diagram showing how the "core" network and the "effector" network are related. Acolor version of this figure is available online at http://www.Eurekah.com.

not yet been fully elucidated for E. coli there is enough information to make a preliminary comparison. Although the data for this comparison is not obtained by the same methods, there is reasonable evidence for the overall structure of the regulatory network based on the integration of wide variety of approaches. Babu and Teichman have made such an integration in their presentation of the transcription network37 based on comparative genomics of the transcription factors in E. coli and a detailed survey of the literature reporting regulatory relationships.38 When we analyze this inferred network using the same statistical approach as for the yeast system we find (albeit with less statistical strength because of the smaller number of genes) that the picture is remarkably similar. We have represented the data in Figure 9A.

The transcription factor regulatory network comparison between yeast and E. coli then suggests an overall structure that has a central core network, consisting of transcription factors regulating each other's expression, and an external, "effector gene" network, regulated by the outputs from this core (Fig. 9B). The central cores are both more highly connected, as indicated by their lower exponent in the power-law fit, than the effector networks. What is more surprising than this, however, is that the exponents are very similar for the two organisms even though one is a prokaryote and the other a eukaryote.

This similarity is surprising, since yeast is much more complex in a variety of ways than E. coli. This raises the question of how an organism as complex as yeast is (relative to E. coli in any case) can have the same overall topological structure of its core regulatory network. The increased complexity may arise from several sources outside of the TF network. First, there are more transcription factors in yeast, and genes are regulated individually, rather than in operons, as in E. coli. Second, it is becoming clear, but is not yet fully understood, that there are a number of transcription cofactors (proteins that regulate gene transcription by binding to transcription factors, but not to DNA) that regulate genes by binding to multiple, bound TF's. This can lead to a large increase in regulatory complexity in yeast, and these kinds factors are not found (or are exceptional) in E. coli. Third, it is likely, judging from the number of RNA-binding proteins, that expression is regulated at the level of the RNA (post-transcriptional regulation) much more in yeast than in E. coli. Clearly the TF network comparison is only part of the story, yet it is significant that the core TF networks are very similar in structure.

Conclusions and Summary

We have discussed two substantially different methods and views of transcription networks, one based on correlative, "influence" analysis using time series analysis, and another based on direct TF binding analysis. While they are different and complementary in many ways, they are connected through the underlying mechanism of global gene regulation and control. Both methods provide insights into the structure of gene expression networks as well as powerful frameworks for data mining. All genes are regulated directly by TF's binding cis-regulatory regions. Thus, when a nontranscription factor gene is seen to "influence" another by time series analysis, we know that it does so through a "hidden" set of interactions that involve TFs, perhaps through regulated chemical modifications of TFs or transcription of TF genes. Most TF's are too weakly expressed, relative to the "effector genes" for the mRNA levels to be followed by array experiments. Thus the "central computer " we discussed as the core network plays the role, in some sense, as a set of "hidden variables" for the effector genes that are followed in a time series analysis.

Clearly a variety of methods must be used to elucidate fully the structure and function of gene expression networks, even in single celled organisms as "simple" as yeast and E. coli. The next stage of analysis of these networks should bring the details of the regulatory interactions, and a full picture of the network to the point where the dynamics, stable states and state transitions of the networks can be predicted and compared with experiment.

0 0

Post a comment