Originally submitted as a Reader Response on 18th June, 2006

I read with great interest the paper by Yang Zhang et al. in which they report the results obtained with the "threading assembly refinement" (TASSER) methodology applied to model G-protein coupled receptors. This approach represents a remarkable achievement and I must congratulate the authors and also thank them for making their results available.

Since I am interested in the renal vasopressin V2 receptor, I downloaded the compressed "P30518.model.tar.bz2" file and examined the five best models offered. The 3D models, on average, were very close to the model I had obtained through homology modeling of the human vasopressin receptor (P30518) using the latest rhodopsin template 1u19A, with the help of the interactive Project Mode of the Swiss-Model Server, using the DeepView program. Some differences were noted, however, and seem to be due to the approach utilized to construct these models. Indeed, the authors stated that "only the sequence of the given GPCR is passed to TASSER and no other extrinsic knowledge (e.g., active sites and binding regions, experimental restraints, etc) is incorporated into our structure prediction approach." Thus, out of the five models offered, only one allowed for the conserved disulfide bond to be established between the cysteines residues 112 and 192 (see Schulein R et al. FEBS Lett 2000 May 4;473(1):101-106). The distance between c112 and c192 in the other four models was such that no disulfide bond could be established between these two residues. A disulfide bond, between two hemi-cystines, one in the top of TM helix 3 and the other in the extracellular loop, is also present in rhodopsin, and seems to be fairly conserved in class A GPCRs. Perusal of other GPCR models offered by the authors, reveal that some models have this disulfide bond, while others do not. While in the fourth model of the V2 receptor the S-S bond was not spontaneously established like in other GPCR models offered, the distance between c112 and c192 was sufficiently close to allow this bond to occur. Unfortunately, in this particular model, and in contradistinction to the others, the highly conserved NPxxY sequence in the last TM helix is shifted upwards by one full helical turn. While I agree with the authors that the absence of constraints eliminates potential bias and allows for rearrangements of TM helices as compared to rhodopsin that could have occurred with evolution, some compromises may be required. Thus, to my recall, the authors of the ab initio method "MembStruk" took into account the conserved disulfide bridge and other requirements during construction of their models. Although the rearrangements/reorientation of TM helices, not possible with the standard homology approaches, definitively represents a major progress, a compromise may be considered in which a minimum of well-established constraints should be kept, including, for instance, the fact that the loops are not likely to curl up along the TM helices, since this space is most likely occupied by the plasma membrane. I am tempted to select one of their other non-V2 receptor models and use it as a template in the Swiss-Model served to further improve my own V2 receptor model. It would be great if this new methodology could be incorporated into an interactive server like the DeepView/Swiss-Model Server, which would allow investigators to customize the modeling of GPCRs of their choice.

Submitted by: Jacques Durr
E-mail: jadurr@att.net
Occupation: MD
University of South Florida