Exposing Hidden Alternative Backbone Conformations in X-ray Crystallography Using qFit

Abstract

Proteins must move between different conformations of their native ensemble to perform their functions. Crystal structures obtained from high-resolution X-ray diffraction data reflect this heterogeneity as a spatial and temporal conformational average. Although movement between natively populated alternative conformations can be critical for characterizing molecular mechanisms, it is challenging to identify these conformations within electron density maps. Alternative side chain conformations are generally well separated into distinct rotameric conformations, but alternative backbone conformations can overlap at several atomic positions. Our model building program qFit uses mixed integer quadratic programming (MIQP) to evaluate an extremely large number of combinations of sidechain conformers and backbone fragments to locally explain the electron density. Here, we describe two major modeling enhancements to qFit: peptide flips and alternative glycine conformations. We find that peptide flips fall into four stereotypical clusters and are enriched in glycine residues at the n+1 position. The potential for insights uncovered by new peptide flips and glycine conformations is exemplified by HIV protease, where different inhibitors are associated with peptide flips in the "flap" regions adjacent to the inhibitor binding site. Our results paint a picture of peptide flips as conformational switches, often enabled by glycine flexibility, that result in dramatic local rearrangements. Our results furthermore demonstrate the power of large-scale computational analysis to provide new insights into conformational heterogeneity. Overall, improved modeling of backbone heterogeneity with high-resolution X-ray data will connect dynamics to the structure-function relationship and help drive new design strategies for inhibitors of biomedically important systems.

Fig 1. Flowchart of the qFit 2.0 algorithm.

qFit can operate on each residue in the protein (orange boxes) in parallel (1 ≤ n ≤ N indices are for residues in the protein). Anisotropic refinement gives a thermal ellipsoid for the Cβ (orange model), and refinement with occupancies set to 0 gives an omit map (purple model). These inputs are combined, backbone translations and peptide flips are sampled (blue models), each backbone is decorated with sidechain rotamers, and an MIQP is used to select 1–4 conformations for the residue. Residues with consecutive multiple backbone conformations, called fragments (yellow boxes), are then subjected to a second MIQP to trace compatible alternative backbone conformations across residues. Residues and fragments are combined into an intermediate model. Finally, a Monte Carlo procedure is used to adjust alternative conformation labels (“altloc” identifiers) to minimize steric overlaps, and the final model is refined.

Fig 2. Geometry and distribution of peptide flips in training set.

(A,B) Reference primary conformation peptide (black) and four cluster centroids for secondary peptide conformations (colors), from the side (A) or “top-down” (B). (C) Members from the training set segregate into two ~180° rotated clusters with different translations in the peptide plan (blue vs. red). View from roughly the same angle as (A). (D) Other members from the training set segregate into +120° and -120° rotated clusters (green vs. brown). View from roughly the same angle as (B).

Fig 3. True vs. false positives with synthetic data.

Peptide flip true positives = percent of peptide flips in the actual synthetic model that are present in the qFit 2.0 model. Peptide flip false positives = percent of residues with a peptide flip in the qFit 2.0 model that are not in the actual synthetic model. Rotamer false positives = percent of sidechain rotamers (as defined by MolProbity [12, 34]) in the qFit 2.0 model that are not in the actual synthetic model. True positives in green; false positives in red. Peptide flips in solid lines; rotamers in dotted line. Data is averaged over all four synthetic datasets (corresponding to the four peptide flip geometry clusters in Fig 2) and all three mainchain amplitudes are considered; see Methods.

Fig 4. Multiconformer modeling with qFit results in similar or better crystallographic R-factors.

R_work and R_free are plotted vs. PDB ID sorted from high to low resolution. X’s indicate rerefined original structures and filled circles indicate qFit 2.0 models; both are after refinement with water picking.

Fig 5. qFit 2.0 successfully identifies known peptide flips.

(A-C) Val539-Gly540 in the Kelch domain of human KLHL7 at 1.63 Å (PDB ID 3ii7). 2mFo-DFc electron density is contoured at 1.2 σ (cyan) and 2.5 σ (blue); mFo-DFc electron density is contoured at +3.0 σ (green) and -3.0 σ (red). (A) The deposited model includes alternative conformations for this peptide, which are well justified by the electron density. (B) qFit 1.0 starting from single-conformer input fails to find the second conformation, resulting in peaks in the difference density map (arrow). (C) qFit 2.0 finds both conformations, resulting in the disappearance of the difference peaks. (D-E) Asn42-Gly43 in carbohydrate binding domain 36 at 0.8 Å resolution (PDB ID 1w0n). The Asn42 sidechain (left, darker green/purple) points up out of the image so is visually truncated. In (E-F), 2mFo-DFc electron density is contoured at 1.5 σ (cyan) and 2.5 σ (blue); mFo-DFc electron density is contoured at +3.0 σ (green) and -3.0 σ (red). (D) The deposited structure includes alternative conformations (green and purple) related by a peptide flip, but re-converges too early at the Gly43 backbone N atom, resulting in >4 σ bond length (red and blue fans) and bond angle (red and blue springs) outliers [34]. (E) qFit 1.0 fails to identify the flip, leaving significant difference density map features. (F) qFit 2.0 identifies the flip at Asn43 and also correctly splits Gly43 into separate conformations, thereby flattening the difference map relative to qFit 1.0 and eliminating the covalent geometry errors in the original structure.

Fig 6. qFit 2.0 finds a hidden peptide flip at room temperature.

(A) Met519-Thr520 in mouse RNA binding protein 39 (RBM39) is modeled with just a single conformation in chain A of the 1.11 Å room-temperature structure (PDB ID 4j5o, pink). There appear to missing unmodeled conformations based on mFo-DFc difference electron density contoured at +3.0 σ (green) and -3.0 σ (red). 2mFo-DFc electron density is shown contoured at 0.9 σ (cyan) and 2.5 σ (dark blue). (B) Although there is diversity for this region in chain B of the asymmetric unit from this structure and in chains A and B from the 0.95 Å cryogenic structure (PDB ID 3s6e, blue), none of the other instances explain the electron density at RT in chain A. There is also no clear evidence for missing alternative conformations in these other instances. (C) In the RT qFit model, the peptide has two alternative conformations (both in red) related by a flip. The new second conformation positions the Met519 sidechain differently (down in this view). Collectively, these changes better explain the local electron density.

Fig 7. qFit 2.0 identifies alternative glycine conformations.

This small loop in the 1.69 Å structure of Hyp–1 protein from St. John’s wort (PDB ID 3ie5) includes several glycines: 49, 50, and 52. (A) The deposited structure (orange) depicts these glycines with single conformations. The qFit 1.0 model (red) does the same, because it cannot sample alternative glycine conformations. (B) The qFit 2.0 model identifies alternative conformations (green/purple) for the entire loop, including all three glycines, based on subtly anisotropic backbone O atoms (arrows). 2mFo-DFc electron density contoured at 1.0 σ (cyan) and 3.0 σ (blue); mFo-DFc electron density contoured at +3.0 σ (green) and -3.0 σ (red).

Fig 8. Extra backbone heterogeneity in qFit 2.0 helps discover new sidechain heterogeneity.

(A) Histogram of difference in maximum Cα displacement across all combinations of alternative conformations between qFit 2.0 and qFit 1.0 for the test set. Vertical dotted line at 0 difference. (B) Maximum Cα displacements for qFit 2.0 vs. 1.0 for residues with a newly discovered sidechain rotamer in the qFit 2.0 model but not in the qFit 1.0 model. Many of these residues fall above the diagonal line, meaning the Cα moves more in the qFit 2.0 model than in the qFit 1.0 model. (C-D) Thr157 in cyclophilin A at room temperature (PDB ID 3k0n). 2mFo-DFc electron density is contoured at 0.5 σ (cyan) and 3.0 σ (blue); mFo-DFc difference electron density is contoured at +3.1 σ (green) and -3.1 σ (red). (C) The deposited structure has alternative rotamers that were correctly manually modeled. (D) qFit 1.0 does not move the backbone and misses the alternative rotamer, as evidenced by a peak of +mFo-DFc density (arrow). (E) qFit 2.0 does move the backbone (note especially the backbone carbonyl displacement), and successfully identifies the alternative rotamer.

Fig 9. Hidden unmodeled peptide flips in the inhibitor-gating “flaps” of HIV–1 protease.

(A) In the 1.39 Å structure of a mutant of HIV–1 protease bound to a novel inhibitor (PDB ID 3qih), the Ile50-Gly51 tight turn interacts with the dimer-related copy of itself, Ile50’-Gly51’ (boxed region). Chain A in orange, chain B in red. The inhibitor (sticks) binds in two overlapping poses immediately adjacent to these flaps. (B) This dimer interface, viewed as if from above in (A), is asymmetric in the deposited structure: both copies of the peptide point downwards in this view. However, positive difference electron density (arrows) suggest unmodeled conformations. (C) qFit 2.0 models this region with coupled asymmetric peptide flips, such that both copies of the peptide point down (~70%, green) or both point up (~30%, purple) in this view. The multiconformer model has diminished difference electron density peaks, suggesting it is a better local fit to the data. Residual difference peaks may reflect unmodeled partial-occupancy waters that are mutually exclusive with the new protein alternative conformations. 2mFo-DFc contoured at 1.2 σ (cyan) and 3.0 σ (blue); mFo-DFc contoured at +3.0 σ (green) and -3.0 σ (red).