SYSTEMS BIOLOGY:

DYNAMICS OF CELLULAR LEVEL FUNCTION AND REGULATION DERIVED FROM MURINE EXPRESSION ARRAY DATA

Abstract

A major open question of systems biology is how genetic and molecular components interact to create phenotypes at the cellular level. Although much recent effort has been dedicated to inferring effective regulatory influences within small networks of genes, the power of microarray bioinformatics has yet to be used to determine functional influences at the cellular level. In all cases of data-driven parameter estimation, the number of model parameters estimable from a set of data is strictly limited by the size of that set. Rather than infer parameters describing the detailed interactions of just a few genes, we chose a larger-scale investigation so that the cumulative effects of all gene interactions could be analyzed to identify the dynamics of cellular-level function. By aggregating genes into large groups with related behaviors (megamodules), we were able to determine the effective aggregate regulatory influences among 12 major gene groups in murine B lymphocytes over a variety of time steps. Intriguing observations about the behavior of cells at this high level of abstraction include: (i) a medium-term critical global transcriptional dependence on ATP-generating genes in the mitochondria, (ii) a longer-term dependence on glycolytic genes, (iii) the dual role of chromatin-reorganizing genes in transcriptional activation and repression, (iv) homeostasis-favoring influences, (v) the indication that, as a group, G protein-mediated signals are not concentration-dependent in their influence on target gene expression, and (vi) short-term-activating/long-term-repressing behavior of the cell-cycle system that reflects its oscillatory behavior.

Supplemental material

The following figure represents the actual biological influences within a cell as obtained from analysis of gene expression data. (Click here for larger image.)

• Supporting Table 1
• Supporting Table 2
• Supporting Table 3
• Supporting Text
• Supporting Figure 5
  Fig. 5. Convergence in the calculation of the transition matrix. Each point is the correlation between the final presented 1.5-h transition matrix and the corresponding matrix derived from a random set of n of 33 data vectors. That the trend goes to 1 indicates that the algorithm is finding a stable solution despite noise in the data. It also appears that n > 26 is sufficient to give very similar results.
• Supporting Figure 6
  Fig. 6. Predictions using the transition matrix. For the 1.5-h transition, we compared the predicted log fold change in expression level with that observed, for 5 of 33 representative ligand experiments. Predicted levels (gray) and actual observations (white) closely coincide.

Back to Systems Biology page

Back to Recent Publications page

Back to NECSI Research page