Book of Abstracts: Albany 2005
Structure Calculations of Helical Membrane Proteins -- A Computational Approach
We explore structure calculation strategies for helical membrane proteins. We seek to take advantage of the structural autonomy of the secondary structure units as postulated by the two-stage model of membrane protein synthesis and folding. Our test models are glycophorin A (a homo-oligomerizing membrane coil) and aquaporin (a channel protein). We investigate rationales for the utilization of solution NMR data such as backbone dihedral angles (DA-s), distance-based restraints (NOE-s) and residual dipolar couplings (RDC-s) in structure calculations of single transmembrane (TM) helices as well as of multi-spanning membrane proteins.
We use and compare two alternative folding schemes: a one-step torsion angle simulated annealing from an extended chain conformation, and a two-step procedure (rigid-body refinement of pre-folded canonical TM helices) inspired by the grid-search methods traditionally used in membrane protein predictions.
We find that if TM helices are correctly identified by NMR or complementary methods such as hydrophobicity searches of their primary sequence, then imposition of backbone dihedral angle restraints can take the place of exhaustive NOE assignments. We also analyze the impact of the deviations from canonicity, commonly found in channel proteins, on single-helix as well as all-protein structure calculations. We therefore suggest that NMR data that can accurately identify TM helices (chemical shift index, RDC-s through the ?dipolar waves?) should be used to select regions of the membrane protein that can be restrained into α-helical formations. DA imposition should take into account protein class-dependent shifts in the Ramachandran maps. Experimental approaches should focus on maximizing the number of inter-helical NOE-s that can be assigned and may require the use of longer-range distance information such as that available from the use of paramagnetic labels. The advantage of our proposed strategy resides in the significant reduction in NOE assignments implied by the elimination of intra-helical distance-based restraints.
On a more technical level, we find that a two-step simulated annealing followed by a short round of restrained molecular dynamics with more realistic sampling of helix-helix packing forces offers a slight advantage over the standard extended chain simulated annealing. The improvement in accuracy is more significant for glycophorin A, a homodimer with nearly canonical TM helices. For a bundle of non-identical and non-canonical TM segments such as the ?hourglass? topology of aquaporin, the overall resolution of both methods is low and scales almost linearly with the number of TM helices. Nevertheless, we suggest that in this case the accuracy of the two-step procedure can be further improved by (i) targeting the helical DA assignments towards average protein class-specific values, and (ii) making use of the ?initial guess? for inter-helical packing provided by the analysis of RDC patterns and NOE ?contacts?.