Predictive Models of Microbial Cell
and Community Functions
Even the simplest microbes command a vast repertoire of complex self-regulating chemical and physical processes. An ultimate goal of the Genomics:GTL program is to develop predictive models of microbial cell and community functions; because of the complexity of microbes, however, the first generation of models will not reach the level of individual biochemical reactions. Instead, they will operate at a level in which cellular pathways are described either qualitatively (as being present or absent) or quantitatively in terms of average concentrations and activity rates derived from experimental data. Despite their lack of chemical detail, these models will provide a powerful tool for integrating and analyzing the very large new biological data sets and, in some cases, predicting cellular behavior under changing conditions.
The prospect for pathway-level modeling is demonstrated by recent research using steady-state models of biological networks in whole cells and kinetic models of individual biochemical pathways. The first approach, metabolic network modeling, combines simplified models with successive constraints to identify an "envelope" of expected cell behaviors under different conditions. Such modeling depends only on the nature rather than the rates of reactants and products of metabolic transformations, and most data for building the model can be derived directly from annotated genomes (see figure below). For example, this type of model could identify which nutrients and metabolic pathways are essential under specific conditions. Metabolic network models eventually will allow scientists to infer phenotypic properties directly from functionally annotated genomes. Models can identify possible metabolic processes, but kinetic information about each pathway is necessary to simulate the cells' dynamic behavior.
In the second approach, models of pathway kinetics require very fast computers and extensive empirical data, including reaction rates and substrate concentrations, to study every step of the biological system to be modeled. Kinetic models have been applied successfully to some very well characterized pathways (see figure below). Since detailed biochemical data generally are not available for pathways, however, comprehensive whole-system models will be possible only after further research has been conducted and computing power has advanced significantly.
The full promise of predictive simulations of microbial function will require a sustained partnership among experimental and computational biologists, mathematicians, and computer scientists. A number of advances are critical in data collection, data management, and modeling methods. Additionally, close collaborations between modelers and biologists are needed to collect complete and consistent experimental data sets for constructing models. The Genomics:GTL program will establish such multidisciplinary partnerships and create data sets and computational methods necessary to reach a predictive understanding of microbial life.
Click to enlarge.
Metabolic maps provide a framework for studying the consequences of genotype changes and the relationships between genotypes and phenotypes. This metabolic network model for Escherichia coli incorporated data on 436 metabolic intermediates undergoing 720 possible enzyme-catalyzed reactions. In this diagram, the circles contain abbreviated names of the metabolic intermediates, and the arrows represent enzymes. The very heavy lines indicate links with high metabolic fluxes. Analyses were correct 90% of the time in predicting the ability of 36 mutants with single-gene deletions to grow on different media. [J.S. Edwards and B. O. Palsson, Proc. Nat. Acad. Sci. 97, 552833 (2000)]
This pathway kinetics model depicts the mechanisms of the "decision circuit" that commits a bacterial virus [lambda] to one of two alternate pathways in its life cycle. The lytic path sets the stage for immediate replication of the virus and destruction of its Escherichia coli host cell, while the lysogenic path selects for the incorporation of viral DNA into the host genome, allowing the virus to remain in a dormant state.
Click to enlarge.
In the diagram, bold horizontal lines indicate stretches of double-stranded DNA, arrows over genes show the transcription direction, and dashed boxes enclose operator sites that comprise a promoter control complex. The core of the decision circuit is the four-promoter , five-gene regulatory network; initiation of pathway actions involve other coupled genes not shown. Many pathogenic organisms use a similar mechanism of concentration-independent probabilistic pathway selection to switch surface features and evade host responses.
In the model above, pathway selection at different virus concentrations, predicted using a kinetic model of the genetic regulatory circuit, is consistent with experimental observations. Developing this model required nearly 40 empirical rate constants and the use of a supercomputer. [A. Arkin, J. Ross, and H. H. McAdams, Genetics 149, 1633-48 (1998)]
