Atomic resolution experimental phase information reveals extensive disorder and bound 2-methyl-2,4-pentanediol in Ca²⁺-calmodulin

Experimental phase information at 1.0 Å resolution was used to validate a more complete multiple-conformation model of disorder in Ca²⁺-calmodulin and identified bound 2-methyl-2,4-pentanediol (MPD) in multiple regions of the protein. This bound MPD is likely to contribute to some of the previously noted peculiar features of the Ca²⁺-calmodulin crystal structure.

Abstract

Calmodulin (CaM) is the primary calcium signaling protein in eukaryotes and has been extensively studied using various biophysical techniques. Prior crystal structures have noted the presence of ambiguous electron density in both hydrophobic binding pockets of Ca²⁺-CaM, but no assignment of these features has been made. In addition, Ca²⁺-CaM samples many conformational substates in the crystal and accurately modeling the full range of this functionally important disorder is challenging. In order to characterize these features in a minimally biased manner, a 1.0 Å resolution single-wavelength anomalous diffraction data set was measured for selenomethionine-substituted Ca²⁺-CaM. Density-modified electron-density maps enabled the accurate assignment of Ca²⁺-CaM main-chain and side-chain disorder. These experimental maps also substantiate complex disorder models that were automatically built using low-contour features of model-phased electron density. Furthermore, experimental electron-density maps reveal that 2-methyl-2,4-pentanediol (MPD) is present in the C-terminal domain, mediates a lattice contact between N-terminal domains and may occupy the N-terminal binding pocket. The majority of the crystal structures of target-free Ca²⁺-CaM have been derived from crystals grown using MPD as a precipitant, and thus MPD is likely to be bound in functionally critical regions of Ca²⁺-CaM in most of these structures. The adventitious binding of MPD helps to explain differences between the Ca²⁺-CaM crystal and solution structures and is likely to favor more open conformations of the EF-hands in the crystal.

1. Introduction  

Calmodulin (CaM) is a highly conserved eukaryotic protein that binds to Ca²⁺ and regulates the activities of hundreds of downstream binding partners. CaM contains two similar domains, each of which comprises two EF-hand Ca²⁺-binding motifs (Babu et al., 1985 ▸). These two domains are separated by a flexible linker region that allows the large interdomain mobility needed for CaM to interact with its structurally diverse targets (Ikura, Clore et al., 1992 ▸; Meador et al., 1992 ▸). Upon binding Ca²⁺, the EF-hand helices in each of the two domains of CaM sample more open conformations and expose clusters of predominantly hydrophobic residues with an unusually high surface density of phenylalanine and methionine (Kuboniwa et al., 1995 ▸; O’Neil & Degrado, 1990 ▸). These flexible hydrophobic pockets are then poised to interact with protein targets in a variety of interaction modes that have been structurally characterized over the past two decades (Franco-Echevarría et al., 2014 ▸; Lau et al., 2012 ▸; Sarhan et al., 2012 ▸; Fallon et al., 2009 ▸; Meador et al., 1992 ▸, 1993 ▸; Yamauchi et al., 2003 ▸).

Because of its physiological importance, small size (148 amino acids), multidomain architecture and extensive conformational dynamics, CaM has been widely studied using a variety of structural and biophysical techniques. Both solution NMR spectroscopy and X-ray crystallography have demonstrated that CaM is highly flexible, particularly in the central linker region comprising residues 70–90 (Ikura, Barbato et al., 1992 ▸; Meador et al., 1993 ▸). Despite general agreement about its structure, NMR and X-ray crystallo­graphy have provided differing views of several important details of Ca²⁺-CaM. For example, the central linker region is helical in triclinic crystal structures of Ca²⁺-CaM (Babu et al., 1988 ▸; Chattopadhyaya et al., 1992 ▸), but NMR spectroscopy and crystal structures in other space groups have shown that the linker is nonhelical from residues 78 to 81 (Fallon & Quiocho, 2003 ▸; Ikura, Barbato et al., 1992 ▸). In addition, an NMR spectroscopic study using residual dipolar coupling information showed that both domains of Ca²⁺-CaM are more closed in solution than in most crystal structures, particularly in the N-terminal domain (Chou et al., 2001 ▸). Fundamentally, these differences provide strong evidence for the highly dynamic character of Ca²⁺-CaM; however, a more specific explanation for these disparities is desirable. Of particular interest is why the EF-hands in the triclinic crystal structure of Ca²⁺-CaM are more open than they are in solution. While crystal lattice artefacts are often invoked to explain the continuous central helix region observed in triclinic Ca²⁺-CaM, the more open EF-hands in each domain are not as easily rationalized using crystal-packing arguments.

Previously, we determined an atomic resolution (1.0 Å) X-ray crystal structure of Ca²⁺-CaM from Paramecium tetraurelia (PDB entry 1exr; Wilson & Brunger, 2000 ▸). The high resolution of this data set allowed the detailed characterization of side-chain and backbone atomic disorder on multiple length scales, providing evidence for widespread spatially correlated disorder in the protein. However, several features in those electron-density maps remained ambiguous, including the prominent difference electron density located in both hydrophobic binding pockets (Wilson & Brunger, 2000 ▸). These features were mentioned in the original publication but were left unmodeled owing to uncertainty in their proper assignment. Similar electron-density features have been described in other triclinic Ca²⁺-CaM structures (Chattopadhyaya et al., 1992 ▸), but no adequate explanation for them has been provided. Because these unexplained electron-density features are located in functionally critical regions of Ca²⁺-CaM, it is important to convincingly assign them to a molecular species.

To examine these ambiguous electron-density features with minimal structural model phase bias, we collected a 1.0 Å resolution single-wavelength anomalous diffraction (SAD) data set from Ca²⁺-SeMet CaM. Combined with modern density-modification procedures, the experimental phases produced an electron-density map of comparable quality to the 2mF _o − DF _c electron density calculated from the final refined model. The experimentally phased electron-density map significantly reduced the ambiguity in the assignment of correlated main-chain and side-chain disorder in Ca²⁺-CaM. In addition, this experimental phase information confirms that 2-methyl-2,4-pentanediol (MPD) molecules are located in the C-terminal binding pocket and mediate a crystal-packing contact between adjacent N-terminal domains in the lattice. Electron density in the N-terminal binding pocket is highly suggestive of another bound MPD. MPD has been used at ∼50%(v/v) concentration as a precipitant in all prior triclinic Ca²⁺-CaM structures; therefore, its binding to these regions of the protein offers a potentially general explanation for the more open EF-hands of the triclinic Ca²⁺-CaM crystal structure.

2. Methods  

2.1. Protein expression, purification and crystallization  

P. tetraurelia CaM was expressed using the previously described pKK233-3 expression construct (Wilson & Brunger, 2000 ▸). Selenomethionine (SeMet) was incorporated using the Escherichia coli methionine-auxotroph strain B834 grown in M9 defined medium containing 1% glucose, 50 mg l⁻¹ of each amino acid except methionine and cysteine, 7.5 mg l⁻¹ FeSO₄ and 50 mg l⁻¹ l-SeMet. Protein expression and purification were performed as described previously (Wilson & Brunger, 2000 ▸).

The purified and lyophilized SeMet CaM protein was resuspended at 15 mg ml⁻¹ in 5 mM CaCl₂, 50 mM sodium cacodylate pH 5.0 and crystallized using the hanging-drop vapor-equilibration method. A drop consisting of 4 µl 15 mg ml⁻¹ SeMet CaM in buffer was mixed with 1 µl 2-methyl-2,4 pentanediol (MPD). The drop was equilibrated against a 1 ml reservoir consisting of 50 mM sodium cacodylate pH 5.0, 50% MPD at 4°C. Crystals of the native protein were used for initial microseeding of SeMet CaM crystals using a cat whisker. Subsequent microseeding was performed using SeMet CaM crystals to ensure a uniform population of SeMet-substituted protein.

Ca²⁺-SeMet CaM crystals grew in 2–5 d and reached dimensions of 0.7 × 0.4 × 0.1 mm. The crystals were blade-shaped and were mounted in nylon loops such that the longest crystal axis was not collinear with the pin. The 50% concentration of MPD in the mother liquor is sufficient to cryoprotect the crystals, which were removed from the drop and immersed into liquid propane held at its melting point in a liquid-nitrogen bath.

2.2. Data collection and processing  

SAD data were collected on beamline 9-1 at the Stanford Synchrotron Radiation Laboratory (SSRL) from a single crystal maintained at 100 K. The crystal was illuminated with 12 670 eV incident X-rays (0.9785 Å) and data were measured using a MAR345 image-plate detector with 150 µm pixel size. The X-ray energy is ∼10 eV higher than the peak of the measured Se X-ray fluorescence in similar crystals of Ca²⁺-SeMet CaM. Data were collected in inverse-beam geometry with 20° wedges in both high-resolution and low-resolution passes to avoid overloaded reflections. In order to collect data in the blind cone, a second high-resolution pass was measured from the crystal after reorientation. Diffraction data were indexed, integrated and scaled in HKL-2000 (Otwinowski & Minor, 1997 ▸). Merging statistics, anomalous signal and resolution limit were independently analysed using the unmerged, integrated data with AIMLESS (Evans & Murshudov, 2013 ▸) as implemented in the CCP4 suite (Winn et al., 2011 ▸). Final data statistics are provided in Table 1 ▸.

Table 1. Data collection and processing.

Values in parentheses are for the outer shell.

Diffraction source	Beamline 9-1, SSRL
Data-collection date	7/4/2000
Wavelength (Å)	0.9785
Temperature (K)	100
Detector	MAR345
Space group	P1
a, b, c (Å)	25.04, 29.41, 52.79
α, β, γ (°)	89.45, 86.29, 82.39
Mosaicity (°)	0.45
Resolution range (Å)	25–1.00 (1.02–1.00)
Total No. of reflections	419257
No. of unique reflections	75551
Completeness (%)	94.0 (84.7)
Anomalous completeness (%)	92.6 (78.8)
Multiplicity	5.5 (3.8)
Anomalous multiplicity	2.8 (2.0)
〈I/σ(I)〉†	11.5 (1.6)
CC1/2 ‡	0.997 (0.294)
R meas (I +, I − merged)	0.121 (2.19)
R meas (I +, I − separate)	0.099 (2.19)
Overall B factor from Wilson plot (Å2)	8.96

^†〈I/σ(I)〉 decreases to 2.0 at 1.04 Å resolution.

^‡CC_1/2 (Karplus & Diederichs, 2012 ▸) was used to determine the high-resolution cutoff.

2.3. SAD phasing and density modification  

Anomalously scattering atoms were located using SHELXC/D (Sheldrick, 2008 ▸), compared with the anomalous difference Fourier maps calculated using model phases from PDB entry 1exr and moved to a common origin to facilitate comparison with the previously solved structure. These 12 heavy-atom sites were input into Phaser for SAD phasing against the full data set extending to 1.0 Å resolution (Read & McCoy, 2011 ▸). The overall figure of merit for the final SAD phases was 0.44.

Density modification was performed using default settings in RESOLVE as implemented in PHENIX (Terwilliger, 2004 ▸), SOLOMON (Abrahams & Leslie, 1996 ▸), Pirate (Cowtan, 2000 ▸), DM (Cowtan et al., 2001 ▸) and Parrot (Cowtan, 2010 ▸) in the CCP4 suite (Winn et al., 2011 ▸). Importantly, we used Hendrickson–Lattman coefficients for the anomalous heavy-atom contribution only (HA_anom) for density modification wherever possible, as these consistently produced superior density-modified maps. Map quality was judged initially by visual inspection and quantified by calculating the correlation coefficient with the final refined 2mF o − DF _c map using PHENIX (Adams et al., 2011 ▸). Of the density-modification programs tested, Parrot (Zhang et al., 1997 ▸; Cowtan, 2010 ▸) delivered the best maps, which could be further improved slightly by using PDB entry 4zgf (Joint Center for Structural Genomics, unpublished work) as a reference data set for histogram matching rather than the default PDB entry 1tqw (Kuntz et al., 2005 ▸).

2.4. Model refinement and validation  

An initial model derived from PDB entry 1exr with all amino-acid side chains in a single conformation was used to build the Ca²⁺-SeMet CaM model. 1exr was used instead of an automatically built model because autobuilding struggled with the weak density of the central helix region, resulting in fragmented models with the two domains separated. Adjustments to the model, including the introduction of ordered solvent molecules and alternate amino-acid conformations, were made in Coot (Emsley & Cowtan, 2004 ▸) with reference to density-modified experimentally phased 2mF _o − DF _c and mF _o − DF _c maps. Cycles of model building and adjustment were alternated with conjugate-gradient least-squares refinement in SHELX against an intensity-based residual target function with geometric (DFIX, DANG, FLAT), ADP (DELU, SIMU) and occupancy (PART) restraints (Sheldrick, 2008 ▸). Friedel-merged intensities were used. Anisotropic ADPs were introduced into the model after convergence of the isotropic ADP refinement, and riding H atoms were included except for the O^γ atom of Ser, the OH atom of Tyr and the N^δ1 atom of His, owing to ambiguities in their idealized position or residue ionization state. In the final cycles of refinement of the converged model, the test set of reflections that were previously sequestered for the calculation of the R _free value (Brünger, 1992 ▸) were included in the refinement. Several close contacts between partially occupied water molecules and disordered protein atoms were identified during model building. These clashes occur between atoms whose occupancies sum to less than unity and are therefore examples of correlated disorder. In order to prevent these regions being incorrectly flagged as van der Waals contact violations, we assigned altloc IDs to the disordered water molecules in the final models.

In order to facilitate direct comparison between the SHELXL-refined model and the PHENIX-refined qFit2.0 models (see below), we used the program phenix.model_vs_data to calculate the R and R _free values for the SHELXL model using the bulk-solvent and anisotropic scaling routines implemented in PHENIX (Afonine et al., 2010 ▸). These values are R _free = 15.6%, R = 13.4% and should be compared with the values reported in Table 2 ▸. The model was validated using Coot (Emsley & Cowtan, 2004 ▸) and MolProbity (Chen et al., 2010 ▸) and the anisotropic ADPs were analysed using PARVATI (Zucker et al., 2010 ▸). Refined model statistics are shown in Table 2 ▸. Structural figures were produced with POVScript+ (Fenn et al., 2003 ▸).

Table 2. Structure solution and refinement.

n.a., not applicable; this quantity is not calculated by the refinement program used.

Model	Ca2+-SeMet CaM	Ca2+-SeMet CaM, qFit2.0	Ca2+-CaM, qFit2.0
PDB code	5e1k	5e1n	5e1p
Refinement program	SHELX-14	PHENIX 1.9	PHENIX 1.9
Resolution range (Å)	25–1.00	25–1.00	25–1.00
Completeness (%)	94.0	94.0	97.4
No. of reflections, working set	71779	71737	69345
No. of reflections, test set	3772	3772	7778
R cryst	0.1436	0.1337	0.1249
R free	0.1649	0.1495	0.1424
R all	0.1440	0.1345	0.1267
No. of non-H atoms
Protein	1636	2942	2607
Ion	5	5	5
Ligand	16	16	16
Water	176	178	180
Total	1833	3141	2808
R.m.s. deviations
Bonds (Å)	0.013	0.013	0.010
Angles (°)	n.a.	1.489	1.317
Angle (1–3) distances (Å)	0.0294	n.a.	n.a.
Average B factors (Å2)
Protein	17.5	14.1	13.9
Ion	11.4	10.8	10.3
Ligand	28.2	29.8	21.0
Water	32.3	26.8	25.5
ADP anisotropy†
Protein	0.388	0.432	0.464
Ion	0.414	0.384	0.368
Ligand	0.358	0.578	0.714
Water	0.335	0.479	0.594
Ramachandran plot
Most favored (%)	99.6	99.4	98.7
Allowed (%)	0.4	0.6	0.9

^†ADP anisotropy is defined as the ratio of the smallest to the largest eigenvalue of the anisotropic ADP tensor.

Automated detection and modeling of alternative side-chain and backbone conformations was performed using qFit2.0 (Keedy et al., 2015 ▸). Structure-factor amplitudes and refined single-conformer models were used to calculate 2mF _o − DF _c and mF _o − DF _c maps. For the Ca²⁺-CaM qFit2.0 model (PDB entry 5e1p), the diffraction data were those deposited with the original PDB code 1exr. Electron density was sampled using standard rotamer libraries for all residues that had anisotropic electron density at side-chain or backbone atoms indicative of a potentially disordered residue. One to four conformations were chosen for each amino acid and the resulting models were refined in PHENIX with anisotropic ADPs and riding H atoms against an Friedel-merged amplitude-based maximum-likelihood target with geometric and ADP restraints (Afonine et al., 2012 ▸). The ADP and geometric weights were optimized and the occupancies of alternative conformations were refined but constrained to sum to unity. Model statistics are shown in Table 2 ▸.

Anisotropic ADP validation was performed using PARVATI (Zucker et al., 2010 ▸) and ANISOANL (Winn, 2001 ▸). The anisotropies of the qFit models refined using PHENIX are systematically higher (indicative of more isotropic character of the ADPs) than those refined against the same data in SHELX (Table 2 ▸). This disparity reflects differences in the type and application of anisotropic ADP restraints in these two programs.

3. Results  

3.1. SAD phasing and density modification  

Experimental phases were obtained using single-wavelength anomalous diffraction (SAD) near the Se K edge of Ca²⁺-SeMet CaM crystals. CaM has eight methionines (excluding the N-terminal Met), resulting in the incorporation of eight anomalous scatterers in 148 amino acids. In addition, the five bound Ca²⁺ ions also have a weak anomalous signal (0.58 e⁻) at this wavelength. The absence of point-group symmetry in these crystals poses particular challenges for anomalous data collection, including balancing the requirements for reasonable data completeness and multiplicity with the need to minimize radiation damage to the sample. Consequently, a SAD data set was collected because a two-wavelength MAD data collection would inflict unacceptable radiation damage on the sample. Furthermore, the completeness is lower than desired in the highest resolution bin (85%; Table 1 ▸) owing to complications arising from the blind cone in space group P1 at high resolution. Although the sample was reoriented in order to capture some of the blind-cone data, the range of this reorientation was limited by the beamline geometry.

The SeMet SAD data set contains significant anomalous signal to approximately 1.33 Å resolution based on the anomalous correlation coefficient between random halves of the unmerged data (CC_anom = 0.3) as calculated by AIMLESS (Evans & Murshudov, 2013 ▸; Fig. 1 ▸ a). Some anomalous signal may extend as far as 1.23 Å resolution if a less conservative cutoff of CC_anom = 0.15 is used. As additional measures of anomalous data quality, the mean anomalous signal-to-noise [〈ΔI/σ(ΔI)〉, where ΔI is the Bijvoet difference I ⁺ − I ⁻] and anomalous measurability (Zwart, 2005 ▸) also indicate that anomalous signal is present to 1.3 Å resolution (Fig. 1 ▸ b) using cutoffs of 1.2 and 0.05, respectively. Despite challenges arising from the low symmetry of this triclinic crystal, the anomalous data have an overall completeness of 93% and a multiplicity of 2.8 (Table 1 ▸). These values are likely to represent upper limits attainable from this sample, as more extensive exposure of similar crystals to X-rays resulted in an obvious loss of high-resolution reflections. SAD phasing performed using Phaser (Read & McCoy, 2011 ▸) gives a final figure of merit of 0.44, which indicates good starting SAD phase-angle estimates.

Figure 1.

Anomalous signal in the Ca²⁺-SeMet CaM Se SAD data set. In (a), the correlation coefficient for anomalous differences over random halves of the data (CC_anom; orange circles) and the correlation coefficient for intensities over random halves of the data (CC_1/2; blue squares) are shown as a function of resolution. The dashed line shows the CC_anom cutoff of 0.3 (1.33 Å resolution) and the dotted line shows a less conservative cutoff of 0.15 (1.23 Å resolution). In (b), the extent of anomalous signal is plotted as the ratio of the anomalous difference signal to noise (red circles) and anomalous measurability (blue squares). All three estimates suggest that there is measurable anomalous signal to ∼1.3 Å resolution in this data set.

As SAD phase probability distributions are bimodal, phase improvement by density modification was necessary to obtain optimal electron-density maps, although the SAD-phased maps were clearly interpretable without modification. Density modification was performed with default settings as implemented in several current software packages (see §2). While all produced excellent maps, the best-performing program was Parrot (Zhang et al., 1997 ▸; Cowtan, 2010 ▸), which produced an electron-density map (Fig. 2 ▸) that was comparable to others in well ordered regions but showed clearer electron density for many disordered or partially occupied residues. We suspected that the superior performance of Parrot may be owing in part to its default use of the high-resolution (1.2 Å) reference structure 1tqw for histogram matching. This atomic resolution reference structure should more closely resemble the expected features of the electron density of Ca²⁺-SeMet CaM at 1.0 Å resolution than would the default settings in other programs. In order to optimally match reference and target electron density, we selected PDB entry 4zgf as a new reference for histogram matching. The 1.0 Å resolution 4zgf structure (Joint Center for Structural Genomics, unpublished work) contains both selenomethionine and Ca²⁺ and thus closely matched the atomic composition and data-set resolution of Ca²⁺-SeMet CaM. The resulting Parrot density-modified map had a correlation coefficient (CC) of 0.88 with the final 2mF _o − DF _c electron density, a CC of 0.77 with the final refined Ca²⁺-SeMet CaM model and was slightly better by visual inspection than the map obtained using PDB entry 1tqw as a reference. Overall, the Parrot 2mF _o − DF _c map CC was 0.05–0.1 higher than those obtained using maps calculated with other density-modification programs. The sensitivity of the final electron-density map to the particular density-modification algorithm employed indicates that this Ca²⁺-SeMet CaM SAD data set could be useful for the testing and improvement of density-modification approaches at atomic resolution.

Figure 2.

A view of the density-modified SAD electron density around the second Ca²⁺ site. Parrot-modified, SAD-phased experimental electron density is shown at 1.0σ (orange) and 4.0σ (magenta) for an EF-hand Ca²⁺-binding site (site II). The map contoured at 4.0σ shows separation of atomic electron density for Glu67 and other residues.

3.2. Modeling conformational disorder and comparison with 1exr  

As expected, the final refined Ca²⁺-SeMet CaM model is nearly identical to the previously refined 1.0 Å resolution crystal structure (PDB entry 1exr; Wilson & Brunger, 2000 ▸), with a C^α r.m.s.d. of 0.1 Å. The protein adopts the same extended, dumbbell-shaped conformation as seen in all other triclinic Ca²⁺-CaM structures (Fig. 3 ▸). As in PDB entry 1exr, there is a fifth Ca²⁺ ion that mediates a lattice contact between Glu47 and Asp58 of a neighboring molecule.

Figure 3.

Location of disordered residues and bound MPD in Ca²⁺-SeMet CaM. The locations of modeled disordered residues (yellow spheres) and of bound MPD are shown on a ribbon diagram of the SHELXL-refined model of Ca²⁺-SeMet CaM (blue). N- and C-termini are labeled.

Ca²⁺-CaM is one of the premier examples of physiologically relevant protein conformational dynamics. Consequently, crystalline Ca²⁺-CaM is more extensively disordered than most proteins that diffract to atomic resolution. We previously modeled 37 disordered residues in the structural model of Ca²⁺-CaM (PDB entry 1exr) using conservative criteria that excluded many ambiguously disordered residues (Wilson & Brunger, 2000 ▸). For the Ca²⁺-SeMet CaM model, we relaxed these criteria and manually built a total of 57 residues into alternate conformations, which represents 39% of the protein (Fig. 3 ▸). The larger number of alternate conformations in Ca²⁺-SeMet CaM is solely a consequence of this less conservative approach to identifying disordered residues, as the final 2mF _o − DF _c electron-density maps for this structure and the prior 1.0 Å resolution Ca²⁺-CaM structure are quite similar, with a CC of 0.94. In addition, the disordered residues common to both 1exr and Ca²⁺-SeMet CaM are similarly modeled, despite the fact that the disorder model in 1exr was not used as a reference or a starting point for the Ca²⁺-SeMet CaM model. Moreover, the degree of overall atomic disorder between the two models is also similar, with an average ADP for all protein atoms of 17.4 Ų and an average protein ADP anisotropy of 0.388 for Ca²⁺-SeMet CaM, which is comparable to the average protein ADP of 16.4 Ų and protein ADP anisotropy of 0.353 previously reported for 1exr. The consistencies between 1exr and our new model suggest functional conformational disorder, rather then idiosyncrasies resulting from cryocooling (Keedy et al., 2014 ▸). Therefore, the increased number of modeled disordered residues in Ca²⁺-SeMet CaM represents an improved model of conformational disorder compared with the conservative approach taken with 1exr.

As an objective measure of the degree of conformational disorder in these two proteins, qFit2.0 was used for automated alternate main-chain and side-chain conformation modeling based on 2mF _o − DF _c electron density (Keedy et al., 2015 ▸). qFit2.0 identifies 138 disordered residues in Ca²⁺-CaM (data from PDB entry 1exr) and 137 residues in Ca²⁺-SeMet CaM, indicating that these two proteins are similarly disordered. The exceptionally high percentage of disordered residues identified by qFit2.0 (93%) is consistent with the highly flexible nature of Ca²⁺-CaM, although some overestimation of disorder is also possible, as an earlier implementation of qFit had an estimated false-discovery rate of 5–10% (van den Bedem et al., 2009 ▸).

Given the large number of disordered residues, it is unsurprising that a significant fraction are difficult to model. Experimental phase information has been shown to improve the identification and modeling of conformational disorder in other systems (Burling et al., 1996 ▸; Brodersen et al., 2000 ▸) and shows similar benefit in this system. An example is Arg94, which was identified as disordered in both the manually built and qFit2.0 models. The manual model features two conformations, but three conformations were identified by qFit2.0 based on automated sampling of the model-phased 2mF _o − DF _c electron density. The third qFit2.0 alternate side-chain conformation, a rotamer about the χ₁ angle, is not visible in 2mF _o − DF _c electron density contoured at 0.9–0.7σ and was therefore missed by manual building into the same maps. However, inspection of the density-modified SAD-phased maps at 0.9σ shows electron density supporting the third conformation (Fig. 4 ▸ a). Importantly, qFit2.0 was able to accurately model the disorder at Arg94 using only model phases, but the inclusion of less biased experimental phase information in the density-modified maps makes this case of disorder easier to identify and confidently model. 2mF _o − DF _c electron density supporting the third conformation of Arg94 becomes visible at ∼0.5σ, which is below the contour level that most crystallographers typically use for model construction. The utility of such low-contour electron-density map features in characterizing protein conformational heterogeneity has recently been demonstrated (Lang et al., 2010 ▸, 2014 ▸; Fraser et al., 2009 ▸). Additional examples where experimental phase information supports qFit2.0 modeling of side-chain disorder in low-contour electron-density features are Glu14, Val108 and Glu123 (Figs. 4b ▸, 4c ▸ and 4d ▸).

Figure 4.

Experimental phase information clarifies cases of ambiguous discrete disorder. In all panels, Parrot-modified, SAD-phased electron density at 0.9σ is shown in orange and final 2mF _o − DF _c electron density at 0.9σ is shown in blue. The darker-colored atoms are the manually built disordered model and the residues in lighter grey are those automatically built by qFit2.0. Arg94 (a), Glu14 (b), Val108 (c) and Glu123 (d) all show clearer evidence of discrete disorder in the experimental electron-density map than in the 2mF _o − DF _c map at this contour level.

3.3. Previously unexplained electron-density features are consistent with MPD bound to Ca²⁺-CaM in multiple locations  

Inspection of the density-modified SAD-phased electron density revealed strong evidence for two molecules of MPD: one mediates a lattice contact involving helices I and III of the N-terminal domain and the other is located in the C-terminal hydrophobic binding pocket (Figs. 3 ▸, 5 ▸ a and 5 ▸ b). Both of these areas contained residual mF _o − DF _c electron density in the prior structure of Ca²⁺-CaM (PDB entry 1exr) suggestive of an unidentified bound species. Because this density was weak and thus their assignment was ambiguous, these features were conservatively but incorrectly modeled as water. In this new structure, the density-modified experimental electron density for these MPD molecules is highly consistent with the final 2F _o − DF _c map, enhancing the confidence that these two MPD molecules are correctly assigned and minimally influenced by model bias, which can be especially pernicious when modeling partially occupied or otherwise disordered ligands (Pozharski et al., 2013 ▸).

Figure 5.

MPD binds to Ca²⁺-CaM at multiple locations and overlaps with a critical target-binding site. (a) shows the density-modified, SAD-phased electron density contoured at 0.9σ (orange) and 2mF _o − DF _c electron density at 0.9σ (blue) around the MPD-binding site in the C-terminal hydrophobic binding pocket. Surrounding residues are labeled. (b) shows the MPD bound at the N-terminal lattice contact between residues in helix I (black lines) and helix III (background, light gray lines). In both panels, the experimental electron density unambiguously supports a bound MPD molecule. In (c) mF _o − DF _c electron density at ±3.0σ (green, red) is shown with the density-modified map contoured at 0.9σ (orange) for a potential MPD site in the N-terminal domain. The disordered MPD molecule (gray lines) is broadly consistent with this density feature but was not included in the final model. In (d) the C-terminal lobe of the crystal structure of Ca²⁺-CaM bound to a peptide from smooth muscle myosin light-chain kinase (smMLCK; PDB entry 1cdl) is shown in ribbon representation, with the peptide in green and CaM in yellow. The C-terminal domain of Ca²⁺-SeMet CaM (blue) is superimposed, with MPD rendered as sticks. The MPD site overlaps closely with the position of the indole side chain of Trp800 in the smMLCK peptide.

Both MPD molecules are bound in regions that bias the two domains of Ca²⁺-CaM towards more open EF-hand conformations (Fig. 3 ▸), likely contributing to the discrepancy between the crystal structure and the more closed solution structure of Ca²⁺-CaM (Chou et al., 2001 ▸; Wilson & Brunger, 2000 ▸). For the N-terminal MPD mediating the lattice contact between helices I and III, these intermolecular interactions would not be possible if these helices were shifted into the more closed conformation observed in solution (PDB entry 1j7o; Chou et al., 2001 ▸) owing to steric conflicts between multiple residues. In particular, the 26° shift of helix I observed in the solution structure would place Glu14 into conflict with MPD and also disrupt other lattice contacts. Although it is impossible to formally separate the contributions of crystal lattice contacts and MPD binding to the more open conformation of the crystalline N-terminal lobe, we can conclude that MPD could not reside at this location if the more closed solution structure had been crystallized. A similar steric conflict is observed with MPD bound in the C-terminal hydrophobic pocket, where the shifted helix VIII in the solution structure (PDB entry 1j7p) would place residues Met144 and Met145 into steric conflict with the bound MPD. Therefore, binding of MPD both favors more open conformations of the two domains and facilitates triclinic lattice formation.

MPD has a chiral center at the C4 atom (secondary alcohol position) and racemic MPD was used as a precipitant to grow these crystals. Both molecules of bound MPD were modeled in their 4S configuration, which is consistent with the electron density. We cannot rule out, however, that both enantiomers might be present at both bound MPD sites. The electron density for MPD bound in the C-terminal pocket of Ca²⁺-SeMet CaM could support either enantiomer reasonably well, as the density corresponding to the C5 atom of S-MPD is weak and could also agree with the O4 atom of R-MPD. The environment of this atom is hydrophobic and thus is less consistent with an O atom (in R-MPD) than with a C atom (in S-MPD), leading to the tentative assignment of S-MPD. We note that the C5 atom has the highest ADPs and lowest electron density of any atom in both MPD molecules, consistent with possible averaging over MPD enantiomers.

While the C-terminal hydrophobic pocket contains a clear MPD molecule, there is ambiguous but significant (3–6σ) mF _o − DF _c electron density in the N-terminal pocket that overlaps partially with a feature in the density-modified SAD map (Fig. 5 ▸ c). Although this electron density is difficult to definitely assign and was left unmodeled in the final structure, it is consistent with a disordered MPD molecule (Fig. 5 ▸ c). These two conformations make chemically plausible contacts with surrounding residues and are in a highly hydrophobic local environment that would be expected to interact with less polar components of the mother liquor such as MPD.

3.4. MPD binds in a critical region of the C-terminal pocket and is correlated with side-chain disorder  

The presence of bound MPD in the C-terminal binding pocket is intriguing since Ca²⁺-CaM uses these surfaces to bind its many targets. One of the best-characterized examples is the complex between Ca²⁺-CaM and a peptide derived from smooth muscle myosin light-chain kinase (smMLCK; PDB entry 1cdl; Meador et al., 1992 ▸). Superimposing the structure of Ca²⁺-CaM–smMLCK with Ca²⁺-SeMet CaM shows that the MPD in the C-terminal binding pocket overlaps with Trp800 from smMLCK (Fig. 5 ▸ d). Trp800 penetrates deeply into the hydrophobic binding pockets of Ca²⁺-CaM and is a critical binding determinant for the interaction with smMLCK (Meador et al., 1992 ▸; Bagchi et al., 1992 ▸). Because this site would be partially occluded in the more closed conformations of the EF hands, bound MPD molecules are likely to bias the C-terminal (and possibly N-terminal) domains towards more open, target complex-like conformations in the triclinic crystal structure of Ca²⁺-CaM.

The C-terminal lobe MPD molecule was refined to an occupancy of 0.52, which is similar to the refined occupancy of the nearer of the two alternate conformations of Ile136. This similarity is significant because the Cδ1 atom of the minor second conformation of Ile136 would clash (2.6 Å C–C atom separation) with the CM atom of MPD if both were fully occupied (Fig. 6 ▸). This provides an unequivocal example of side-chain disorder coupled to MPD binding, which imposes a correlation between protein conformational fluctuations and the occupancy of this binding pocket. Although not the only example of spatially correlated disorder in Ca²⁺-CaM, the adventitiously bound MPD clearly illustrates the importance of Ca²⁺-CaM conformational heterogeneity for binding to the hydrophobic pockets.

Figure 6.

Binding of MPD in the C-terminal pocket is correlated with side-chain disorder at Ile136. The density-modified map contoured at 0.9σ (orange) is shown with the C-terminal MPD molecule and its nearby residue Ile136. Ile136 samples two conformations (black and gray) and one (gray, conformer b) sterically conflicts with the bound MPD as shown by the dashed line.

4. Discussion  

The 1.0 Å resolution Ca²⁺-SeMet CaM SAD data set represents one of the higher resolution examples of experimental phase information in the PDB. Because bimodal SAD phase probability distributions require improvement by density modification before optimal maps are obtained, this high-resolution SAD data set provides a potentially useful resource for optimizing the performance of density-modification algorithms at atomic resolution. Our results show that Parrot was the best-performing algorithm tested, although we used default parameters for the tested programs and thus cannot rule out the possibility that other programs may perform better with different parameterization. Although the quality of all density-modified maps was comparable in well ordered regions, Parrot-calculated maps showed stronger and more easily interpreted density for disordered residues, which are of particular interest in this system. Experimental phase information is highly valuable when used to clarify these weak electron-density features, as experimentally phased maps are not subject to phase bias introduced from a structural model and thus can be used to critically evaluate the placement of alternative side-chain conformations and ligands into ambiguous electron density. We note that density modification represents the deliberate introduction of particular forms of phase bias into the map; however, this bias does not favor a particular structural model and is thus less problematic than traditional sources of model phase bias when interpreting difficult regions of electron density.

The value of experimental phase information for characterizing conformational heterogeneity has been recognized for some time (Brodersen et al., 2000 ▸; Burling et al., 1996 ▸). This analysis of Ca²⁺-SeMet CaM confirms and extends a prior characterization of the extensive side-chain disorder present in crystalline Ca²⁺-CaM based on model phases (Wilson & Brunger, 2000 ▸). Although much of the modeled disorder in PDB entry 1exr is supported by this new Ca²⁺-SeMet CaM structure, less stringent criteria for introducing alternative conformations as well as the availability of experimental phase information allows a significant expansion of the disordered model. The manually built Ca²⁺-SeMet CaM model contains ∼39% of the residues in alternate conformations, while the automated disorder-modeling approach implemented in qFit2.0 indicates that 93% of the residues sample multiple conformations. Considered in the light of recent results from diverse proteins (Deis et al., 2014 ▸; Fraser et al., 2009 ▸, 2011 ▸; Keedy et al., 2014 ▸; Lang et al., 2010 ▸, 2014 ▸), there is compelling evidence that crystalline proteins are more extensively disordered than previously thought and that traditional criteria for modeling disorder are probably too conservative. Being free of model bias, experimental phase information could aid in the development of new criteria for the detection and modeling of protein conformational disorder. Because experimental phases are observations and thus increase the observation-to-parameter ratio in refinement, they may be fruitfully used in ensemble-refinement approaches, where they could mitigate concerns about potential overfitting of the data (Burnley et al., 2012 ▸; Burling & Brünger, 1994 ▸), although we did not explore this application here. Moreover, experimental phase information would be a particularly powerful way to characterize conformational heterogeneity when obtained from crystals at room temperature.

These SeMet Ca²⁺-CaM SAD data have resolved a longstanding question about the proper assignment of ambiguous electron-density features in both hydrophobic binding pockets of Ca²⁺-CaM, which we have shown to be MPD. The observation of bound MPD at multiple locations helps explain why the EF-hands of Ca²⁺-CaM are more open in triclinic crystal structures than in solution, as MPD binds in a target-like fashion to at least the C-terminal binding pocket and probably the N-terminal pocket as well. In addition, an MPD molecule mediates a lattice contact that bridges helices I and III of neighboring N-terminal lobes, stabilizing more open conformations of this domain. Ca²⁺-CaM is a difficult protein to crystallize, and nearly all prior structures have used MPD as a precipitant. Even in CaM structures determined in different space groups, electron-density features suggestive of bound MPD are present (Wilson & Brunger, 2003 ▸). Interestingly, one structure of Ca²⁺-CaM that used PEG rather than MPD as a precipitant (PDB entry 1prw) shows the molecule in a dramatically different, partially closed conformation (Fallon & Quiocho, 2003 ▸). Furthermore, a structure of the Ca²⁺-CaM N-terminal domain alone (PDB entry 3ikf) obtained using PEG as a precipitant shows residues from the central linker region bound in the N-terminal pocket of neighboring molecules, similar to the location of bound MPD in the triclinic structure (O’Donnell et al., 2009 ▸). Therefore, we propose that MPD acts as both ligand and precipitant in many crystal structures of Ca²⁺-CaM, and that the presence of MPD as an adventitious CaM ligand needs to be incorporated into future interpretations of this well known structure.

Supplementary Material

PDB reference: Ca²⁺-calmodulin from Paramecium tetraurelia, 5e1p

PDB reference: selenomethionine derivative, 5e1k

PDB reference: 5e1n