Difference between restoring and predicting 3D structures of the loops in G-protein–coupled receptors by molecular modeling

Goldfeld et al. (1) presented modeling results on restoring 3D structures of the intra- and extracellular loops in bovine rhodopsin (bRh), human A2A adenosine receptor (A2Ar), turkey β1-adrenergic receptor (β1AR), and human β2-adrenergic receptor (β2AR). In all cases, the lowest energy conformers of the loops matched the corresponding X-ray structures with excellent rmsd values. The extracellular loops for the same receptors were modeled in our article previously (2) with the use of a much less sophisticated modeling procedure. Goldfeld et al. (1) emphasized substantially better accuracy of reproducing the X-ray structures of the loops, especially larger loops, in their study compared with ours. Although their restoration of the loops is impressive, we believe it is important to outline differences between the two modeling procedures (1, 2) that were not sufficiently distinguished by Goldfeld et al. (1).

First, the results of Goldfeld et al. (1) were obtained for restoration of each single individual loop with the other loops fixed in their X-ray conformations. Also, their modeling used some additional specific constraints (atom–atom contacts) within and between loops found by X-ray crystallography for bRh, A2Ar, and β2AR (table S1 in ref. 1). The calculations by Goldfeld et al. (1) restored the conformation of a given loop within the framework of the already known X-ray structure. On the contrary, our de novo calculations (2) predicted multiple combinations of conformations for all the extracellular loops without any preliminary knowledge of the X-ray structures of any extracellular elements surrounding the loops (2).

Second, there are disulfide bridges in the extracellular loops of A2Ar, β1AR, and β2AR. Our modeling deduced the possible systems of disulfides from de novo calculations (2). It was not clear in the work of Goldfeld et al. (1) whether the actual disulfide bridges were deduced from calculations or adopted directly from the known X-ray structures (1); if it is the latter case, their modeling used these experimental constraints as well.

Third, our calculations were routinely performed for loops of larger lengths than those in Goldfeld et al. (1) (compare table 2 in ref. 1 vs. table VI in ref. 2). This point is worth mentioning, as the difficulty of loop prediction is known to drastically increase with the length of the loop.

Fourth, our modeling aimed to predict of a set of possible conformations of the highly flexible extracellular loops in G protein-coupled receptors rather than restoring only one single “snapshot” conformation frozen in the available X-ray structures. Crystal structures capture only one view of multiple dynamic modes of the loop. The snapshot loop structure seen depends on specific experimental conditions, and may correspond neither to the lowest energy conformation obtainable by modeling nor to the most biologically relevant conformer. In this regard, the lowest of the rmsd values calculated for the set of predicted low-energy conformations may serve as a more adequate measure for success of predictions made by modeling (2).