Response to Comment on Three X-ray Crystal Structure Papers

Additional strong support for the existence of the bound ligands can be seen in the structural analyses of germline Ab BBE6.12H3 in ligand-free state and in the states bound to four different peptide ligands (1). Abs often have induced fit on binding of the Ag, reflected in terms of structural changes in the Ab, particularly in elbow angle and conformations of the CDRs (11–14). It is evident that the Ab BBE6.12H3 shows significant changes in the conformations of CDRs when bound to different ligands. Also, elbow angles of the ligand-bound Abs are significantly different from the elbow angle of the unliganded Ab (Table II). This is particularly common in the crystal structures involving conformational flexibility of ligand/receptor.

We disagree also with other statements made by Pozharski et al. It is to be noted that PDB ensures the quality of deposited data by implementing various validation mechanisms. Likewise, it is incorrect to state that the editors or reviewers have no access to the data during the review process. The authors’ obligation to make available the primary data at any stage of the review process is well recognized.

Detailed responses to the specific queries raised by Pozharski et al. along with the methodological clarifications are as follows:

In all instances, while determining the structures, we used the refined native Ab models for molecular replacement. All of the structures were refined using the Crystallography and NMR system (CNS 1.2) at the time of deposition (15). All of the processes were now repeated using CNS 1.3 (15) and PHENIX 1.9 (16), and the results are very similar.

Ligand fit in a crystal structure is judged on the basis of many interlinked factors. In the present case, while modeling the peptide ligands onto the electron density map, three kinds of maps were routinely used: simulated annealing 2Fo–Fc composite omit maps, delete-refine Fo–Fc maps, and Fo–Fc maps (17–19). Change in R_free and appearance of negative or positive peaks in Fo–Fc maps were constantly monitored to ensure correct modeling of the ligand and to avoid any wrong fit and overinterpretation. Excessive dependence on a single factor can be detrimental to unbiased evaluation of the model fit. It should be noted that whereas the peptide ligands were 12 residues long, we modeled only those residues that satisfied all of the above criteria. The refinement of B-factors, B-individual against B-group, was guided by general considerations of the ratio of model parameters to the number of observations (20). Finally, quality of data must be viewed in the context of the specific study design and functional implications.

The main concern is the absence of positive omit electron density as the primary experimental evidence. Whereas Pozharski et al. have visualized positive omit electron density for ligands at a contour level of 3.0 σ, we would like to emphasize that while modeling a molecule, an experienced crystallographer evaluates electron density going back and forth at different contour levels rather than sticking to a specified σ cutoff, to distinguish signal from noise. There are no hard-and-fast rules for setting contour levels when building models in delete-refine Fo–Fc electron density maps or the other electron density maps; the crystallographers are expected to critically evaluate results during the refinement process to ensure signal-to-noise distinction is adequately handled. This is evident from a large number of published structures; only a few are cited in this letter (4, 5, 21–25).

We do not wish to comment on the standards provided by Pozharski et al. because we are not questioning the structures in which ligands are defined at 3.0 σ cutoff. However, we argue that many structures that are crystallographically sound and physiologically interesting have ligands visualized at contour levels less than 3 σ. The statement by Pozharski et al. that “The standard approach according to modern practice is to inspect the difference electron density omit map contoured at 3.0 σ” is not universally valid. We provide examples as positive controls from among the many biologically relevant protein–ligand complexes visualized at a lower σ cutoff, even as low as 1.5 σ (4, 5, 21–25). Indeed, ligand building in delete-refine Fo–Fc maps at contour levels less than 3.0 σ is routinely practiced. A comparison of delete-refine Fo–Fc maps in some of the PDB depositions from the above-mentioned positive controls, contoured at 3.0 σ and at 1.7 σ, is shown in Fig. 1. It is evident that the signal can be distinguished from noise at the lower σ cutoff.

Positive delete-refine Fo–Fc electron density is evident for the peptide models built in each of the maps when observed at σ cutoff range between 2.5 and 1.7. The images of delete-refine Fo–Fc maps in Fig. 2 show relevant positive densities to indicate presence of the peptide in each case. For generating these figures, calculations were done with CNS 1.3 (15). The figures were generated in Coot (26). We would like to re-emphasize that in addition to delete-refine Fo–Fc maps, simulated annealing 2Fo–Fc composite omit maps were also used while modeling the peptides (17–19). We have recalculated simulated annealing 2Fo–Fc composite omit maps after deleting the peptides, and densities for the ligands and electron density for peptides can be visualized at the contour levels between 1.0 σ and 0.7 σ. Therefore, positive omit electron density irrefutably exists in all of the seven structures if visualized at σ cutoffs as mentioned in Fig. 2, which are acceptable as a routine practice.

With reference to the second allegation, i.e., discrepancies observed in B-factors, it is to be noted that all of our structures except one are at low or intermediate resolutions. The refinement of individual B-factors is fine at very high resolution, but the similar treatment of B-factors at lower resolution can lead to aberrations (20). We had therefore reported the structures after B-group refinement for all except 4H0H (Table I, serial numbers [Sl. Nos.] 15–21).

It has been emphasized, particularly at lower resolution, that a model with fewer restraints for B-factor refinement may be better (20). The B-factor calculations presented by Pozharski et al. are likely to be presented after refining the individual B-factors for all the structures; we were able to reproduce similar values to those obtained by Pozharski et al. by following their procedure. Thus, there are no inconsistencies in the data reported in the papers (1–3). Differences observed in B-factors calculated by Pozharski et al. and by us are primarily due to adoption of different refinement strategies. Indeed, depending on the resolution and the context, crystallographers usually adopt strategies to ensure avoidance of model bias. Although newer strategies are being evolved constantly for each of such circumstances (20), there has not yet been a universalization of norms.

Concerning the difference in the B-factors of peptide and neighboring atoms of macromolecules, we would like to emphasize that the presence of such a difference cannot automatically lead to a conclusion that the ligand is absent. There are several reasons for the B-factors of the ligand to be higher than those of the protein. One reason is that the occupancy of the ligand could be slightly lower, in which case the peptide electron density would be detectable at lower contour levels and the B-factor values would be somewhat higher. However, as long as positive electron density in the unbiased electron density maps is observable, a model could be built. Another reason is that if the model does not have rigid geometry and the contacts with the protein are limited, the B-factors would be higher. More rigid ligands are expected to have better-defined electron density and much lower B-factors in comparison with intrinsically flexible ligands. All of the above are very real circumstances particularly involving proteins associated with interesting physiological processes. Table I (Sl. Nos. 1–14) lists many other examples from the literature to support our argument (4–10). Most of the ligands cited in Table I are either peptides or nucleic acids and exhibit a higher degree of freedom in backbone conformation. In particular, peptides show an even greater degree of freedom contributed by variations in side chain conformations. This degree of freedom is considerably reduced when ligand binds protein with higher affinity and therefore interacts more strongly (owing to multiple sites of interactions). It is important to highlight that the three PDB depositions by Pozharski and colleagues, 3LK0, 3LK1, and 3KL5 (6, 9), show significantly higher B-factors for ligands compared with those of the protein (Table I, Sl. Nos. 12–14).

In our 2013 paper (3), we had two structures of Ab 2D10, bound to two different ligands: a dodecapeptide (4H0H) and a comparatively more rigid monosaccharide (4H0I), with identical binding affinities to the Ab 2D10. Clearly, these two ligands have varying degrees of flexibility and the effect is clearly evident in B-factors. The peptide–2D10 Ab complex (3), which was determined at 2.0 Å resolution, showed somewhat higher B-factors at occupancy of 1.0 for the peptide. Considering that slightly higher B-factors may be due to lower occupancy, we refined the structure by setting an occupancy of 0.8. Indeed, refining with lower occupancy did bring the B-factors down without changing the delete-refine Fo–Fc electron density and the model fit. In other words, the structural information remained unchanged.

Finally, we would like to reinforce that when dealing with biological systems, several variables are critical, and therefore all cases should not be uniformly subjected to arbitrarily restrictive criteria. Such criteria could result in limited applicability and an immense loss of genuine structural data of biological relevance. This would certainly be detrimental to constructive progress in science. More and more molecular biologists are employing crystallographic experimentation to capture complex and dynamic biological processes. Considering the constraints that the macromolecular crystallographic experimentations impose, one has to work at an edge and carefully distinguish between signal and noise. Analyses using multiple parameters ought to be used, as is done by many experienced macromolecular crystallographers, some of whom are mentioned above.