Quantum chemistry has reached a point where some gas phase properties of small molecules can be computed more accurately than they can be measured. For these systems science can say "problem solved", give itself a light pad on the back, and move on to other things - like how to do the same for molecules twice the size. Or in solution.
I don't believe there is an equivalent success story in biomolecular modeling, but, as I'll argue here, the prediction of the pKa values of ionizable residues in proteins be a close contender!
Jens Erik Nielsen and co-worker recently published a very interesting paper entitled "Remeasuring HEWL pKa values by NMR spectroscopy: Methods, analysis, accuracy, and implications for theoretical pKa calculations" in Proteins.
One of the most interesting implications of the paper, in my opinion, can be found at the bottom of Table III. Here the authors compare Asp, Glu, and His pKa values in HEWL measured using the NMR chemical shifts of backbone nuclei to those obtained from side chain 13C atoms.
The RMSD values are 1.3, 1.2. and 0.6 pH units for 15N, 1HN, and 1Hα, respectively (note that the latter could only be obtained for 7 of the 11 residues). Now compare this to the corresponding values for pKa values predicted with the pKD and PROPKA servers: 0.9 and 0.8.
Thus, if you are not willing to go through the considerable trouble that is assigning side chain chemical shifts for a protein, you are better off predicting the pKa values than measuring them! At least for this protein ... and only if you can't live without knowing all the pKa values ... but still.
Nielsen and co-workers estimate the accuracy of the 13C-derived to be accurate to within 0.1 - 0.2 pH units, so, no, the general problem is not solved. But if more people do as careful a job measuring pKa values as Jens, we may find that we are closer than we think.
2 comments:
Hi Jan
Its correct that "you are better off predicting the pKa values than measuring them" if you measure on amide H or N. But you should _never_ try to estimate pKa values from these chemical shifts. That is one of the points in our paper. The chemical shifts of the amide group are much to sensitive to the overall electrostatic environment in the protein to be useful as probes for the protonation state of the side chain. In some proteins you see some of the larges changes in amide chemical shifts for Ile and Phe! I would not call it a fair attempt to determine the pKa value of an amino acid side chain by following the changes in amide chemical shifts. Therefore, I do not think that comparing such pKa values with predicted pKa values makes sense.
Determining the pKa values of Glu and Asp do not require assignment of the side chains. You can measure the chemical shifts of Cg/Cd of Asp/Glu in an NMR experiment where these shift are correlated with the shifts of the backbone amide H, which you know. Thus the additional assignment you need to do is minimal.
Cheers,
Kaare
"I would not call it a fair attempt to determine the pKa value of an amino acid side chain by following the changes in amide chemical shifts. Therefore, I do not think that comparing such pKa values with predicted pKa values makes sense."
Yes, but many of the experimental pKa values used to benchmark, or (even worse) train, pKa prediction methods were derived from backbone chemical shifts. This gives a completely wrong picture of the accuracy of pKa prediction methods.
"Determining the pKa values of Glu and Asp do not require assignment of the side chains. You can measure the chemical shifts of Cg/Cd of Asp/Glu in an NMR experiment where these shift are correlated with the shifts of the backbone amide H, which you know. Thus the additional assignment you need to do is minimal."
OK, I didn't know that! Even less excuse to report experimental pKa values based backbone chemical shifts. These still appear in the literature.
Thanks for your input, and the very first comment on this blog! One of many I hope.
Post a Comment