Tuesday, February 27, 2007

A case of the wrong hand

A perfect example of the danger of blindly trusting the output of computers, as well as the self-correcting nature of science, was provided by the lab of Geoffrey Chang. The work here concerned the crystal structures of several transporters in the cell membranes of bacteria. One, called MsbA, is a member of the class of proteins called ABC transporters because of the part of their structure known as the "ATP-binding cassette" that is involved in providing the energy for transport. MsbA is involved in transporting lipids from one side of the membrane to the other, but it is related to proteins involved in drug resistance, including a protein in cancer cells that enables them to pump out chemotherapy drugs.

Given the difficulty of determining the structures of membrane proteins, Chang's structure of MsbA was seen as a great success. However, major doubt was cast on his results when the structure of another ABC transporter called Sav1866 was determined. Sav1866 is a true multidrug resistance pump, and thus represented a step toward understanding the more pharmacologically relevant members of the ABC transporter family. The problem was, the structure of Sav1866 looked very different from Chang's MsbA structure. In particular, the two halves of the transporter were much more intertwined in Sav1866, such that there was a large twist in the overall structure and that the bundles of helices on each side of the central pore consisted of elements from both halves, rather than a single half as reported for MsbA (see the article with new structure here).

It turns out that the source of the discrepancy is something very simple and totally unintentional. Chang's group used a homemade program to analyze the data, and an error in this program caused two columns of numbers to be swapped. This in turn inverted the coordinate axes (made the structure the wrong hand, as crystallographers call it). This is apparently an easy mistake, but one that is guaranteed to totally ruin a structure. Since the structure was in the low enough resolution range where modeling the atoms into the observed electron density takes some guesswork, the group was still able to build a reasonable model even though the structure was backwards. You can see a figure showing the effect of the reversal in this news article from Science.

Unfortunately, the group used the same data analysis program to analyze another structure, of a bacterial multidrug resistance transporter from a totally different family called EmrE. Once again, they were able to build a satisfactory structure in terms of chemical bonds into their backwards electron density (though I must say I smelled a rat the moment I first saw the structure, due to the fact that the side-chain packing was far worse than anything I'd seen in comparable resolution structures. I cannot comment about MsbA because if I ever saw that one it was too long ago for me to remember what it looked like). All in all, this mistake resulted in the retraction of five scientific papers.

Several researchers have commented on the fact that there were reasons to doubt the proposed sructure of EmrE due to discrepancies with other experimental results. The major lesson here is to double-check all computer output to make sure it is reasonable, both from the standpoint of expected side chain conformations and biochemical data on the protein of interest. While some inferences from biochemical data also can be flawed (i.e. data on residues responsible for substrate binding to enzymes have been shown to sometimes identify residues far from the active site even in good structures), there are cases in which it is worthwhile at least to go over a proposed crystal structure with a fine-toothed comb to make sure everything is reasonable.

No comments: