Scientists led by Adam Godzik at the Burnham Institute have combined structural genomics and computational modeling to construct the first complete model, including 3D protein structures, of the metabolic network of an organism, in this case that of one of the first discovered hyper-thermophilic bacteria, Thermotoga maritima. The study appeared in Science in September.
Godzik, who directs the Burnham's bioinformatics and systems biology programs, said that this is analogous to sequencing an entire genome and can be applied to various organisms, including humans. "It was done for many individual proteins and pathways, but as far as I can tell, it's the first example of a functional systems biology network which has this whole layer added," he says. The model points to a small number of protein shapes, or folds, providing a new understanding of the evolution of protein functional networks.
Using a bottom-up approach, Godzik's team first mined the literature for all known biochemical reactions and substrates to find evidence for more than 50 percent of the metabolic reactions. Then, combining experimental, genomic, and computational modeling data, they were able to reconstruct the missing enzymes' structures and functions.
The map consists of 478 metabolic genes, 503 unique metabolites, and 562 intracellular and 83 extracellular metabolic reactions. Of the 478 proteins in the network, 120 were determined experimentally while 358 were solved through modeling. The team validated the reconstruction by showing that they could reproduce T. maritima's ability to grow on many different carbohydrates and produce several known metabolic by-products. However, Godzik says, the model doesn't incorporate interactions between proteins, only interactions between proteins and small molecules.
Next, fold analysis of similar, connected, and unrelated proteins revealed that there were only 182 distinct folds among the 478 proteins, and that enzymes that catalyze similar reactions have a higher probability of having similar folds. This suggests that nature does indeed use slightly tweaked, existing shapes to perform new functions.
"From [an] immediate perspective, it allowed us to look at the distribution of faults on the network," Godzik says. "So we could, for instance, look at relations between folds and functions on the entire network. Similar structures are often found on the same pathway, so we hypothesized that a lot of pathways could have developed by internal duplication in the pathway. And this was noticed on many pathways and that was never analyzed on a functional network covering [the] entire metabolism."
The map is not perfect, however, and will require further work to complete. "On the level of very detailed structures, we know that some of them are wrong," Godzik says. Out of the few hundred models his team built, only about 100 were accurate and some are "very, very crude." While it's still a work in progress, Godzik believes they've hit an important milestone in functional and comparative analysis. "We have at least one model, however crude, for every protein in the network," he says. "[But] we are aware that significant parts of these models are not useful for detailed analysis of interactions yet, so this is something that will be happening in the future."