Getting Closer to the Whole Picture

See allHide authors and affiliations

Science  27 Apr 2007:
Vol. 316, Issue 5824, pp. 550-551
DOI: 10.1126/science.1142502

A major challenge of biology is to unravel the organization and interactions of cellular networks that enable complex processes such as the biochemistry of growth or cell division. The underlying complexity arises from intertwined nonlinear and dynamic interactions among large numbers of cellular constituents, such as genes, proteins, and metabolites. As well, interactions among these components vary in nature (regulatory, structural, and catalytic), effect, and strength. The reductionist approach has successfully identified most of the components and many interactions but, unfortunately, offers no convincing concepts and methods to comprehend how system properties emerge. To understand how and why cells function the way they do, comprehensive and quantitative data on component concentrations are required to quantify component interactions. On page 593 of this issue, Ishii et al. (1) provide unsurpassed complete and quantitative data of components at the various constituent levels in a bacterial cell.

Rather than a reductionist viewpoint (that is, a deterministic genetic view), the pluralism of causes and effects in biological networks is better addressed by observing, through quantitative measures, multiple components simultaneously, and by rigorous data integration with mathematical models (2). Such a systemwide perspective (so-called systems biology) on component interactions is required so that network properties, such as a particular functional state or robustness (3), can be quantitatively understood and rationally manipulated.

The technical challenges of the systems biological approach (4) are mainly along four lines (see the figure): (i) systemwide component identification and quantification (“omics” data) at the level of mRNA, proteins, and small molecular weight metabolites; (ii) experimental identification of physical component interactions, primarily for information processing networks; (iii) computational inference of structure, type, and quantity of component interactions from data; and (iv) rigorous integration of heterogeneous data. The last step is required to achieve a holistic, quantitative, and predictive understanding through mathematical models that enable an iterative cycle between prediction and experiments, the hallmark of systems biology.

A systems roadmap.

The comprehensive component concentrations reported by Ishii et al. provide input data for inferring component interactions using computational methods. The challenge for computational modeling methods yet to be developed is to predict the functional network state from the concentrations and to infer the information processing network that controls the functional state.

The development of experimental methods relating to the first challenge has made tremendous advances in the past decade, but the level of sophistication and the associated costs have led to a situation where primarily single-component data—that is, data solely on genes, proteins, or metabolites—are available. Until the study by Ishii et al., at best two different types of component data were reported for a given experiment, which severely limited progression along the iterative cycle between experiments and theory.

By joining forces among specialized labs, Ishii et al. report systemwide data on three main component layers of cells—transcriptome (mRNA), proteome (protein), and metabolome (metabolites)—with a particular focus on central carbon metabolism of the model bacterium Escherichia coli. Beyond component concentrations, the functional endpoint of gene, protein, and metabolite interactions—the intracellular metabolic fluxes—were quantified from 13C-labeling experiments (5). In a laborious procedure, data on steady-state growth were collected from 24 mutant strains of E. coli in which a different gene that functions in carbon metabolism was removed from each strain. All mutants were grown at the same specific rate (6), thus minimizing indirect effects of the largely different mutant physiology that would otherwise hamper data interpretation from batch cultures.

The highly reproducible results from these genetic perturbation experiments were complemented with steady-state growth data from the wild-type bacterium grown at a range of growth rates in the same culture conditions, with the extreme cases of near starvation and almost unlimited supply of glucose (the limiting nutrient). An interesting observation is the active response of the bacterium's metabolic system to environmentally dictated changes in growth rate. There were global alterations in the expression level of many mRNAs and proteins. By contrast, upon genetic perturbation of the metabolic system, surprisingly few changes were observed at any component level (besides some obvious and inevitable local perturbations such as altered educt and product concentrations due to a deleted reaction). These results indicate that metabolic networks employ fundamentally different strategies to maintain active operation in the face of genetic or environmental perturbations. However, many important questions remain unanswered. Why does this particular distribution of flux in metabolism emerge from the determined component concentrations? What is actively regulated and by which mechanisms? Which mechanisms contribute to the observed robustness?

These questions remain open because the known component interactions have not yet been considered and because the regulatory network that controls metabolism is only partly known and qualitatively understood at best. But Ishii et al. provide, for the first time, a quantitative data set that includes both the constituting components and the functional state of a metabolic network. From these data, we can begin to unravel the conditional and quantitative relevance of regulatory interactions and discover new circuits, thereby addressing the question of how a functional state arises from the components. The call is thus open to integrate this heterogeneous data set into a coherent whole from which testable hypotheses on general principles and network regulation can be derived. Although this particular data set will not be sufficient—for example, it lacks time-resolved dynamic data, and the conditions studied are limited—its unprecedented completeness has the potential to become a cornerstone for computational systems biology.

A number of computational approaches are already available to integrate subsets of the Ishii et al. data. Statistical analyses of metabolic and transcriptional data, for example, can identify key features that are important for respiratory oscillations (7). Alternatively, computational mapping of transcriptome and metabolome data onto a graphic model of component interactions can provide guidance for dissecting control at the level of genetic regulation from regulation of protein activity (8). In contrast to statistical analyses, a method rooted in constraint-based modeling (4) allows combining metabolite concentrations with metabolite flux data using thermodynamic principles to derive hypotheses about active or new regulatory mechanisms (9).

The extensive data set reported by Ishii et al. now opens the way to use existing computational approaches and to develop new ones to extract new biological insights about a fundamental physiological process. From this starting point, model-based design of targeted experiments for further conditions will reduce the unrealistically tedious and expensive collection of large-scale data sets while working toward a truly holistic understanding of cellular behavior. We are one step closer.


Navigate This Article