Three recent papers published in The Journal of Immunology by Dr. Salunke and his coworkers at the National Institute of Immunology in New Delhi have attracted our attention (1–3). We regret to inform you that our findings cast serious doubt on the validity of the structural data presented in these publications. A paper published by the same group elsewhere exhibits similar problems (4).
To summarize, the publications in question attempt to address the role of plasticity in molecular recognition by Abs. Specifics vary from publication to publication, but the overall approach includes determination of multiple crystal structures of corresponding Abs in complex with various peptides. Based on the refined structure models, details of molecular recognition are then derived at an atomic level, and conclusions are presented regarding mechanisms of such recognition.
Unfortunately, in each of these publications describing peptide–Ab complexes determined by x-ray crystallography, the very central claim of a peptide actually bound to an Ab is not supported by evidence. The necessary evidence in the form of electron density is absent and the analyses of the experimental data are systematically flawed when examined according to accepted professional and scientific standards. Deposition of experimental data in the Protein Data Bank (PDB) has been mandatory since 2008, and availability of such data allows us (as well as anyone trained in protein crystallography) to verify the claims presented in the publications of Dr. Salunke. We inspected electron density maps for every protein–peptide complex structure associated with these papers, and found that the required primary experimental evidence, positive omit electron density in support of presented claims, is lacking. No electron density exists for the peptides in the purported protein–peptide complexes.
Our goal here is to correct the scientific record and preserve the integrity of the PDB as a valid database (5, 6). It should also be noted that the electron density map calculation required to visualize these problems requires access to the coordinates and structure factors. Consequently, editors or reviewers are not at fault that these problems were overlooked during the editorial review process, as they likely did not have access to those data at the time. The importance of inspection of electron density fit as a primary means of local validation has been repeatedly pointed out (7–9).
Below we describe the accepted standard validation procedure and compare results of its application to test cases selected from the literature and to seven structure models deposited by Dr. Salunke et al. [2XZQ/2Y06/2Y07/2Y36 (1); 4BH7/4BH8 (2); 4H0H (3)].
Validation procedure
We use an established validation procedure to verify the presence (or absence) of peptide electron density. The procedure includes following steps:
Rerefine the deposited structure models against the deposited experimental data. The refinement results described below were obtained using BUSTER-TNT software (10). Using other modern crystallographic refinement programs [e.g., REFMAC5 (11) or phenix.refine (12)] yields the same results and conclusions.
Compare the average B-factor of the peptide molecule to that of the set of atoms in immediate contact with the peptide (we use a 4 Å interatomic distance cutoff to define the molecular neighborhood). It is expected for a genuine protein–peptide complex that these two sets of B-values will be close. Large discrepancy indicates that the peptide molecule is either present at partial occupancy or that its presence is not supported by electron density.
Remove the peptide molecule from the deposited structure model and rerefine the model without peptide—the omit-map procedure (13, 14). The term omit map here refers to the fact that the model component in question, i.e., the peptide in this case, is omitted from the model refinement to reduce model bias in the electron density map. If the peptide molecule in question is present, the shape of the resulting difference electron density will provide corresponding evidence. The standard approach according to modern practice (15, 16) is to inspect the difference electron density omit map contoured at 3 σ. We list two positive control examples and the negative results for the seven PDB entries from the papers in question.
Positive controls
In order to demonstrate the expected behavior, we have applied the same analysis to PDB entries 3FN0 (17) and 3GGW (18): both are genuine and validated Fab/peptide complexes (17, 18). For these two structures, rerefinement of the original model results in similar B-factors for the bound peptides and the neighboring Ab atoms (36.0/37.7 Å2 and 37.6/27.3 Å2 for 3FN0 and 3GGW, respectively). As expected in the case of bound ligands, the degree of combined dynamic and static disorder observed as expressed by the average B-factor in the peptide molecule and the surrounding protein atoms is very similar, given that these atoms are in direct contact and interact via a number of hydrophobic interactions and hydrogen bonds. As ultimate consequence and as the necessary proof positive, the peptide molecules are clearly evident in the omit electron density map (see Fig.1).
Omit electron density maps for the Ab-bound peptides from PDB entries 3FN0 (left panel) and 3GGW (right panel). 2fo-fc map (blue) contoured at 1σ and fo-fc map (green/red) contoured at ± 3 σ are shown. Peptide model is shown (yellow sticks). This figure and Fig. 2 were rendered using PyMOL (http://www.pymol.org).
Omit electron density maps for the Ab-bound peptides from PDB entries 3FN0 (left panel) and 3GGW (right panel). 2fo-fc map (blue) contoured at 1σ and fo-fc map (green/red) contoured at ± 3 σ are shown. Peptide model is shown (yellow sticks). This figure and Fig. 2 were rendered using PyMOL (http://www.pymol.org).
Omit electron density maps calculated using experimental data from (1–3). 2fo-fc maps (blue) contoured at 1σ and fo-fc maps (green/red) contoured at ± 3 σ are shown. Panels correspond to the following PDB entries: 2XZQ/2Y06 (first row), 2Y07/2Y36 (second row), 4BH7/4BH8 (third row), 4H0H (bottom row). Deposited peptide model is shown (yellow sticks).
Omit electron density maps calculated using experimental data from (1–3). 2fo-fc maps (blue) contoured at 1σ and fo-fc maps (green/red) contoured at ± 3 σ are shown. Panels correspond to the following PDB entries: 2XZQ/2Y06 (first row), 2Y07/2Y36 (second row), 4BH7/4BH8 (third row), 4H0H (bottom row). Deposited peptide model is shown (yellow sticks).
These results serve to establish expected positive proof for the presence of the peptides when comparing them to the structure models in question.
Negative results
2XZQ/2Y06/2Y07/2Y36 (1), 4BH7/4BH8 (2), 4H0H (3): average B-factors for the peptides and surrounding atoms are shown in Table 1. The discrepancies between the deposited models and the results of rerefinement are discussed below. In all seven cases, upon rerefinement of the original models, B-factors of peptide ligands significantly exceed that of their immediate molecular neighborhood. These significant discrepancies provide clear indication of a lack of scattering contribution originating from the purported peptide molecules.
PDB ID . | Deposited Model . | Rerefined Model . | ||
---|---|---|---|---|
Bpeptide . | Bprotein . | Bpeptide . | Bprotein . | |
2XZQ | 64.9 | 24.9 | 102.8 | 29.1 |
2Y06 | 71.7 | 65.4 | 145.6 | 87.7 |
2Y07 | 37.6 | 41.4 | 111.4 | 61.1 |
2Y36 | 66.4 | 39.3 | 139.0 | 39.8 |
4BH7 | 64.5 | 26.8 | 132.3 | 30.8 |
4BH8 | 83.1 | 31.4 | 98.7 | 36.9 |
4H0H | 91.4 | 27.7 | 105.5 | 31.9 |
PDB ID . | Deposited Model . | Rerefined Model . | ||
---|---|---|---|---|
Bpeptide . | Bprotein . | Bpeptide . | Bprotein . | |
2XZQ | 64.9 | 24.9 | 102.8 | 29.1 |
2Y06 | 71.7 | 65.4 | 145.6 | 87.7 |
2Y07 | 37.6 | 41.4 | 111.4 | 61.1 |
2Y36 | 66.4 | 39.3 | 139.0 | 39.8 |
4BH7 | 64.5 | 26.8 | 132.3 | 30.8 |
4BH8 | 83.1 | 31.4 | 98.7 | 36.9 |
4H0H | 91.4 | 27.7 | 105.5 | 31.9 |
Values are shown for both original models as deposited in the PDB and upon rerefinement in BUSTER-TNT (10).
In agreement with these findings, the difference omit maps in all seven cases show no evidence of peptides when contoured at the 3 σ level (see Fig. 2).
Inconsistencies between refinement results and tables in publications
Table I shows the comparison of relative B-factors for the peptides in the structures discussed above as they appear after a cycle of rerefinement to the same values as reported in the corresponding models deposited in the PDB. The tabulated values for Bpeptide are unexpectedly and inexplicably lower than the actually refined Bpeptide values, while the Bprotein values are much closer. The average B-factor of a structure model may vary when it is refined with different programs and/or a different set of parameters, but it has never been observed that a different refinement would produce much higher B-factors for specific groups of atoms such as the purportedly bound peptide relative to the protein to which it is bound. In one case (4H0H), the occupancy of the peptide was set to 0.8, which can only be justified if that brings B-factors of peptide and Ab into agreement (not the case here despite the reduced occupancy). We assume that Dr. Salunke and his coauthors are aware that the overall B-factors of neighboring components in a crystal structure cannot differ drastically.
Summary
High relative B-factors of the purported peptides serve as a first indication of problems with the peptide models. The absence of the peptides is firmly supported by the clear absence of positive omit difference density. Tabulated values for mean peptide B-factors in the deposited PDB files are unexplainably lower and inconsistent with those obtained by standard crystallographic refinement practice and cannot be reconciled with the demonstrated absence of electron density. The genesis of these structure models is unknown to us, but in our opinion, it is abundantly clear that they are erroneous and do not support the conclusions of the corresponding papers.
Footnotes
Abbreviation used in this article:
- PDB
Protein Data Bank.