Technical Comments

Comment on “Crystal structures of translocator protein (TSPO) and mutant mimic of a human polymorphism”

See allHide authors and affiliations

Science  30 Oct 2015:
Vol. 350, Issue 6260, pp. 519
DOI: 10.1126/science.aab1432

Abstract

Li et al. (Reports, 30 January, p. 555) reported on a crystal structure for a translocator protein (TSPO) from Rhodobacter sphaeroides in which some of the electron density is modeled as a porphyrin. The analysis of the x-ray data discussed here suggests that this assignment is incorrect.

Translocator proteins are implicated in human diseases caused by malfunction of the systems that transport porphyrins, cholesterol, and proteins into mitochrondria (1). In addition, Bacillus cereus translocator protein (TSPO) and other bacterial TSPOs catalyze a light-dependent photooxidation of protoporphyrin IX (PpIX) (2, 3). Thus, a crystal structure of TSPO with a porphyrin bound is likely to shed light on functional properties of this class of proteins. The structure reported by Li et al. describes a TPSO that appears to have a copurified porphyrin bound, which is identified as PpIX in the corresponding Protein Data Bank (PDB) file (PDB accession 4UC1) (4). If this assignment is correct, this structure raises a new set of questions. Because the biochemical data provided by Li et al. (4) show that their purified Rhodobacter sphaeroides TSPO (RsTSPO) protein still binds externally supplied PpIX tightly, does RsTSPO have multiple porphyrin binding sites, and if so, how is the observed site related to the function of the TSPO proteins? Before making the considerable effort it may take to answer these biochemical questions, it might be wise to examine the reliability of the claim that this structure includes PpIX or some other closely related compound.

Close examination of the 4UC1 structure raises several issues, some of which are evident in the figures included in the paper reporting on this structure (4). The PpIX structure shown in figure 4 of that paper appears to be bent instead of planar, and many of its ring substituents do not appear to fit into the “feature-enhanced map” included in their paper. It is also relevant that the PDB validation report for 4UC1 flagged the PpIX and nearly all the headless monooleins that structure is reported to include as outliers both in electron density fitting and in geometry (4). The ligand library density-fitting (LLDF) score is 11.24 for the PpIX, and it is as high as 30.71 for the monooleins (LLDFs greater than 2.0 are considered a concern). In addition, the geometry outlier Z score for the PpIX is 3.53 for bond-length violations, involving 50% of all its atoms, and it is 2.63 for bond-angle violations, involving 40% all of its atoms (Z scores greater than 2.0 are a cause for concern). Thus, it appeared that an independent analysis of this structure that takes advantage of the data deposited for 4UC1 might be warranted.

Fig. 1 Structure re-refinement for 4UC1.

(A) Fractional Rwork (magenta open circles) and Rfree (red filled circles) values after structure re-refinement, using all the available data deposited in the PDB at a resolution of 1.65 Å as a function of reciprocal resolution (1/Å). The reported Rwork (black open diamonds) and Rfree (blue filled diamonds) values for the 4UC1 structure are included for comparison at the truncated 1.80-Å resolution marked by the vertical line. (B) Final weighted 2FobsFcalc map contoured at 1.0 σ is overlaid with a (PpIX) model (thin salmon sticks) and a PEG molecule (thick gold/yellow sticks). Currently, the density is fitted with 9 PEG units. It can also fit 10 units after converting the centrally bound water molecule into a part of the added unit. The PpIX molecule does not fit the density at all. (C) The presence of two Trp side chains (W38/W39, magenta) may have defined the unusually curled conformation of the PEG. The orientation of this figure is very close to figure 4A of (4).

That analysis raised some questions about the structure factors deposited for this structure in the PDB that can only be addressed if the original raw data were publically available, which they are not. The protein used to obtain the 4UC1 structure contained Se-Met, and the data were collected at the Se peak wavelength. The presence of Se-Met in the protein can be demonstrated by refining the structure assuming that only S-Met is present. Large positive residual peaks appear on the S atoms of S-Met residues in the corresponding difference maps, as expected, and an estimated Se-Met/S-Met substitution ratio is about 76% to 24%. For a typical Se-Met data set that extends to such high resolution, one expects that (i) anomalous difference Fourier maps should include positive peaks that are much larger than the noise for all the Se atoms; (ii) the ratio of the anomalous signals to measurement errors should always be greater than 1—i.e., the differences between the Friedel mates, F(hkl) and Embedded Image, of Bragg reflections should be larger than their estimated errors; and (iii) an outstanding experimental map should be obtained using Se anomalous signals (5). Unfortunately, none of these expectations is satisfied by the 4UC1 data set. In fact, the reported, average anomalous signal ratio of the 4UC1 data set was smaller than 1, suggesting that the measurement errors might have been systematically overestimated. If this is so, it would also have an adverse effect on the resolution reported for the 4UC1 data set. A better, and possibly higher, resolution structure would probably emerge if the data were properly reprocessed.

Although the deposited 4UC1 data set extends to a resolution of 1.65 Å, none of the data between 1.80 and 1.65 Å appear to have been used to obtain the 4UC1 structure or to compute its R-factor statistics (4). Exclusion of the weak-intensity high-resolution data, which are invariably poorly explained by structural models, is bound to improve the R-factor statistics of a structure, even though it compromises the utility of R factors as indicators of structure quality and may make modeling errors hard to detect (6). In fact, when all the 4UC1 data were used for structure refinement and the model corrected by hand, R factors improved in all resolution ranges (Fig. 1A). Even for the data between 1.80-Å and 1.65-Å resolution, R factors are still below the 42% limit of Evans and Murshudov (7). Furthermore, it was found that electron densities for the putative PpIX site and several headless monoolein sites were better modeled by other moieties such as low-molecular weight polyethylene glycol (PEG) molecules, the presence of which might be explained by the fact that the crystals in question were grown in the presence of 32% PEG 400 (Fig. 1, B and C). These reassignments of electron density are supported by LLDF values and Z-score values that are all below those deemed acceptable.

Of course, there is more than one way to process diffraction data. One may trade radiation-damaged data with redundancy, anomalous signals with resolution, the precision with the accuracy of the data, and so on. Given this fact, discussion is currently under way about whether deposition of raw data should be required for publication of crystal structures so that they can be reprocessed later by others when questions like those addressed above arise in the future (8).

References and Notes

  1. Acknowledgments: This work is in part supported by National Institutes of Health Project Grant P01 GM022778 and funding from the Steitz Center for Structural Biology, Gwangju Institute of Science and Technology, Republic of Korea. The author thanks P. Moore for discussion and suggestions of this comment.
View Abstract

Navigate This Article