Protein Structure and Function. Application of Bioinformatics Methods - John Rigden 2014

Prediction of Membrane Protein Structure
Prediction of Spatial Structure

As in the case of Globular Proteins, two main approaches are used to predict the three-dimensional Structure of TM proteins: ab initio modeling and Homology modeling, which are described in Chapters 1 and 3 of this book.

Homology modeling, also known as comparative modeling, relies on using known close structures as templates to build a spatial model of the protein under study. The method is based on the observation that Cell/13.html">Protein Structure is significantly more conserved than its Amino Acid Sequence. Consequently, even in the presence of pronounced sequence divergence, proteins may share common structural properties—in particular, a similar fold—provided their sequences show detectable similarity (typically over 30% sequence identity). Obtaining high-resolution crystallographic structures is fraught with numerous Technical Challenges, especially when it comes to TM proteins. Therefore, homology modeling emerges as a highly promising approach. The structural models generated this way can provide invaluable insights for formulating hypotheses about protein function and can help guide the direction of future experimental work. The homology modeling process itself can be broadly divided into four main stages: Template Selection, sequence alignment between the target protein and the template, model construction, and Model Evaluation. Necessary iterations can be performed at each stage to progressively refine the quality of the final model (Sanchez and Sali 1997; Marti-Renom et al. 2000). Table 4.5 lists several commonly used homology modeling programs.

Class="center">Table 4.5. Some of the most frequently used homology modeling programs (adapted from Wallner and Elofsson 2005)

Among the tools listed in Table 4.5, only SWISS-MODEL (Peitsch 1996), which features a 7TM/GPCR interface, was specifically designed for TM proteins. Consequently, special attention must be paid to the potential presence of polar side chains protruding into the Hydrophobic core of the membrane in homology models. Certain modeling tools, such as SCWRL (Canutescu et al. 2003), do account for the specific side-chain Structural Features of transmembrane proteins. While these tools do not completely solve the problem, incorporating them into Model Building significantly improves the quality of extramembrane regions. To date, there remains a shortage of specialized tools dedicated exclusively to TM protein modeling. Nevertheless, recent studies demonstrate that bioinformatics Methods currently applied to soluble proteins—ranging from Hydrophobicity profiling to Secondary structure prediction and homology modeling—can be just as successfully utilized for TM proteins (Forest et al. 2006). Indeed, key application areas for TM protein modeling include the identification and validation of drug targets, as well as the characterization and optimization of Ligand binding. Homology-based drug design approaches have thus far been applied to investigate A wide variety of Kinases, including the epidermal growth factor receptor Tyrosine kinase (Ghosh et al. 2001), Bruton's tyrosine kinase (Mahajan et al. 1999), and Janus kinase 3 (Sudbeck et al. 1999).

Ab initio (or de novo) modeling involves generating a Three-Dimensional Protein Structure in the absence of any prior structural information about the target or its homologues. Research in this area generally focuses on three main pillars: generating low-resolution partially overlapping protein structures, developing accurate energy-scoring Functions, and implementing efficient conformational sampling methods. Although the majority of these methods were originally developed for globular proteins, significant efforts have also been made to predict the Structure of Transmembrane proteins.

ROSETTA (Rohl et al. 2004) is an ab initio modeling server designed to identify the minimum-energy structure for a given amino acid sequence through potential energy function analysis. The prediction pipeline relies on continuous feedback between individual methodological components, allowing for ongoing optimization of both potential Energy Functions and calculation algorithms. A modified version of ROSETTA (Barth et al. 2007) incorporates atomic-level potential functions describing protein-membrane interactions, explicitly accounts for lipid-membrane Protein Interactions, and treats Hydrogen Bonds implicitly. The results indicate that this model successfully captures the key physical forces governing the folding and stability of Membrane Proteins. The method is capable of predicting the structures of small TM proteins (fewer than 150 residues) with a resolution better than 2.5 Å, an accuracy comparable to predictions for small Water-soluble Protein domains. The membrane-adapted ROSETTA approach, combined with homology modeling and domain assembly techniques, has been successfully used to model the structures of the Kv1.2 and KvAP potassium channels. The resulting models showed a high degree of structural similarity to their respective crystallographic counterparts. Modeling both the open and closed states of these channels helped elucidate the mechanisms underlying voltage-gated channel opening and closing, demonstrating that the gating mechanism is driven by conformational changes within the protein structure. These modeling insights led to testable hypotheses for subsequent experimental validation (Yarov-Yarovoy et al. 2006).

FRAGFOLD (Jones 2007) is a protein tertiary structure prediction method based on the assembly of supersecondary structure fragments using a simulated annealing algorithm. The core concept is to dramatically restrict the conformational search space by pre-selecting fragments from high-resolution protein structure libraries. In the FILM method (Pellegrini-Calace et al. 2003), a Membrane Potential term is added to the standard FRAGFOLD energy functions (which account for pairwise interactions, solvation, covalent bonds, and hydrogen bonds). This membrane potential was derived from a statistical analysis of a dataset comprising 640 transmembrane helices with experimentally determined topologies, extracted from 133 proteins found in the SWISS-PROT database. Benchmarking tests on known structures of small proteins demonstrate that the method achieves an acceptable level of accuracy in predicting both helix topology and overall protein conformation.