SEARCH

SEARCH BY CITATION

Mastering Your Domains

  1. Top of page
  2. Mastering Your Domains
  3. Finding a Family Resemblance

Despite sophisticated algorithms for structural analysis, the final prediction of protein domain content is often remarkably subjective—you know one when you see it. Unfortunately, the reality is more complicated, and different strategies can disagree on how many domains a protein contains and where the boundaries fall, largely due to the absence of a consistent thermodynamic definition of ‘protein domain’.

Porter and Rose attempt to bring clarity to this situation with a formal definition that yields predictions that are generally corroborated by experimental data. Their method entails determining the m-value for a given protein segment, measuring the extent to which that polypeptide is denatured in a given concentration of urea. If the m-value for a protein segment in isolation is roughly equivalent to its m-value within the context of the whole protein—as quantified by a ‘qualifying ratio’ (QR)—that segment can be considered a true, independent domain. 1

The authors developed an algorithm called ‘structure-energy equivalence of domains’ (SEED), which predicts domains by calculating QR values across entire protein sequences. Importantly, SEED's domain boundary determination process emphasizes the energetic contributions of solvent-exposure of backbone surface area, which the authors cite as the dominant energy term in folding, rather than side-chain interactions. SEED accurately mapped nine proteins for which the domain structure had previously been experimentally obtained; importantly, the experimental and SEED results predicted more domains than the leading domain-prediction algorithms, CATH and SCOP. They also saw divergence between results obtained with SEED and CATH for an additional set of 71 proteins, with experimental data generally supporting SEED's predictions when available. These findings generally suggest that many proteins may contain more and smaller individual domains than previously believed.— Michael Eisenstein

Porter, L.L. & Rose, G.D. Proc. Natl. Acad. Sci., USA, Published online 25 May 2012, DOI: 10.1073/pnas.1202604109.

Finding a Family Resemblance

  1. Top of page
  2. Mastering Your Domains
  3. Finding a Family Resemblance

As the portals between the intracellular and extracellular environment, membrane proteins play a pivotal role in cellular physiology. It is therefore unsurprising that nearly 50% of our current pharmacopoeia is directed at channels, receptors and other molecules studding the cell surface. Unfortunately, structural characterization remains a chore—membrane proteins are notoriously hard to crystallize, and most prediction algorithms have exhibited limited effectiveness.

By taking an evolutionary perspective on membrane protein structure, Hopf et al. have now achieved important progress on this front. Their EVFold_membrane algorithm assembles alignments of protein homologues from multiple species, which it then analyzes to identify evolutionarily conserved pairs of interacting residues. This analysis attempts to uncover the most probable set of relationships across the entire protein sequence, eliminating biases that emerge in more ‘local’ approaches for determining probable co-evolution. The models are further filtered to eliminate those that are incompatible with membrane constraints or secondary structure predictions. 2

The researchers selected 11 proteins containing transmembrane helical domains for which no 3D structural information was available, and used EVfold_membrane to predict all-atom structures for each. In many cases, the resulting structures closely resembled proteins that are virtually unrelated from a sequence perspective but which perform a similar function. The algorithm also performed well in a benchmarking study using proteins of known structure, generating models that accurately reflected existing crystal structures in 22 out of 25 cases. The authors note that the resulting structures are likely to contain inaccuracies when examined at higher resolution, but should nevertheless prove useful as a tool for structural genomics efforts or even drug discovery.— Michael Eisenstein

Hopf, T.A. et al. Cell, 149, 1–15 (2012).