Multiscale conformational heterogeneity in the protein-binding domains of staphylococcal protein A: Possible determinant of functional plasticity

SUMMARY

The Staphylococcus aureus virulence factor staphylococcal protein A (SpA) is a major contributor to bacterial evasion of the host immune system, through high-affinity binding to host proteins such as antibodies. SpA includes five small three-helix-bundle domains (E-D-A-B-C) separated by conserved flexible linkers. Prior attempts to crystallize individual domains in the absence of a binding partner have apparently been unsuccessful. There have also been no previous structures of tandem domains. Here we report the high-resolution crystal structures of a single C domain, and of two B domains connected by the conserved linker. Both structures exhibit extensive multiscale conformational heterogeneity, which required novel modeling protocols. Comparison of domain structures shows that helix1 orientation is especially heterogeneous, coordinated with changes in sidechain conformational networks and contacting protein interfaces. This represents the kind of structural plasticity that could enable SpA to bind multiple partners.

INTRODUCTION

The structural plasticity conferred by conformational flexibility has increasingly been recognized as a likely determinant of function. For example, multiscale heterogeneity in the calmodulin central helix most likely helps it in binding >100 protein targets (Wilson and Brunger, 2000), and a concerted motion seen in both NMR and crystal structures of ubiquitin is proposed to underlie its functional plasticity of promiscuous binding to many different proteins with high affinity (Lange et al., 2008). However, flexibility is manifested in a variety of ways, depending both on the protein itself and on how it is observed. Flexibility is apparent in X-ray crystallography as electron-density inconsistent with a single molecular model – either fully separated peaks or anisotropic density shapes showing fluctuation of atom groupings. In this paper we refer to alternative conformations as conformational heterogeneity rather than flexibility because the latter term implies motion on a relevant time scale, which cannot be determined by crystallography. Many phenomena contribute to conformational heterogeneity in crystal structures, from diverse crystal contacts to functionally relevant conformational fluctuations on a wide range of time and size scales.

Like ubiquitin, staphylococcal protein A (SpA) exhibits broad binding specificity with other proteins. This protein allows Staphylococcus aureus to evade the innate and adaptive immune systems, making it a significant challenge to human health. Among virulence factors responsible for S. aureus pathogenicity, SpA is the best studied and arguably the most important. It is a highly abundant 56kDa multi-domain cell-surface polypeptide with two functionally distinct halves (Fig. 1a). The C-terminal half anchors SpA to the extracellular surface of the peptidoglycan cell wall via the LPXTG motif (Schneewind et al., 1992) and is likely disordered due to its low sequence complexity. In contrast, the N-terminal half is a series of five stable protein-binding domains (E-D-AB-C). Recent studies establish that the conserved sequence KADNKF forms a highly flexible linker between all domains except E to D, which uses the longer sequence KADAQQNKF, also likely to be highly flexible (A.H. and T.G.O, unpublished data). The five domains have sequence identities of 74% to 91% (relative to A domain, Fig. S1) and share the same three-helix-bundle topology. The folding of each domain is thermodynamically uncoupled to the others and displays a gradient of increasing stability toward the more C-terminal modules (A.H. and T.G.O, unpublished). In addition, the B domain rapidly unfolds and refolds approximately 70 times per second (Myers and Oas, 2001), and recent studies establish the same property for the other four domains (A.H. and T.G.O, unpublished data). All five domains can bind the F_c and F_ab regions of host antibodies (Jansson et al., 1998), TNFα receptor 1 (Gomez et al., 2004), von Willebrand factor (Hartleib et al., 2000), and the C1qR component of complement (Nguyen et al., 2000).

Figure 1.

Staphylococcal protein A (SpA) and the crystal structures of C and B–B domains. (a) Schematic showing the organization of SpA and its five protein-binding domains. The conserved linker (green) between E and D domains has a three-residue insertion, indicated by a star. (b) The asymmetric unit of C domain (cyan). The three helices are labeled (Hlx1–3) and the conserved linker regions are indicated (arrow) and colored dark blue. (c) The asymmetric unit of B–B domain. The two copies of B–B domain are colored in shades of green or purple. Each domain (Domain1 and Domain2) and the linker in between is depicted as a ribbon drawing, where Domain2 is rotated by 180°. See also Figure S1.

To date, only two crystal structures of SpA domains have been solved, both single domains in antibody complexes: B domain with F_c (PDB 1FC2) (Deisenhofer, 1981) and D domain with F_ab (1DEE) (Graille et al., 2000). These structures show partner interactions with SpA, but they lack comparison with unbound domain structures, and their lower resolution (2.7–2.8Å) does not allow determination of multiple conformations. The B domain/F_c co-crystal structure (1FC2) lacks coordinates for most of helix3, which originally stimulated some interest in the possibility that helix3 unwinds upon F_c binding. However, subsequent NMR-based amide hydrogen exchange and circular dichroism studies suggested that helix3 is indeed formed in the complex and that the lack of density in this region of the 1FC2 data resulted from a crystallization artifact (Gouda et al., 1998; Jendeberg et al., 1996; Torigoe et al., 1990). Another issue discussed by the groups that determined NMR solution structures of B domain and its variants is the interhelical angle between helix1 and helices2 and 3 (Gouda et al., 1992; Tashiro et al., 1997; Zheng et al., 2004). The latter helices are nearly perfectly antiparallel in all structures, but there is significant variation in the angle between helix1 and helix2, depending on the structure. A low helix1–2 angle was observed in the F_c complex (Deisenhofer, 1981), whereas a larger angle was found in the original solution structures (Gouda et al., 1992; Jendeberg et al., 1996), which were not constrained by residual dipolar coupling (RDC) data. These results led the authors to propose that a conformational change in B domain takes place upon F_c binding to give a low helix1–2 angle. This proposal was supported by kinetic B domain/F_c binding studies, indicating that a conformational change might contribute to k_on (Jendeberg et al., 1995). Two subsequent solution structures of Z domain (an engineered B domain variant) and E domain, based in part on RDC constraints, showed a lower helix1–2 angle in the absence of F_c, which called into question this mechanism (Starovasnik et al., 1996; Tashiro et al., 1997). However, the differences in helix1–2 angle between most structures are not significantly larger than the uncertainties in these most recent structures, so the existence and possible role of a coordinated conformational change in helix1, whose residues form the majority of the contacts with F_c, remains unresolved.

Conformational heterogeneity is traditionally evaluated at two scales: sidechain-rotamer conformations and tertiary-structure rearrangements, such as hinge motions and secondary-structure reorientations. It includes local backbone differences such as peptide flips or backrubs that accompany sidechain rotamer sampling (Davis et al., 2006), small translational shifts of a few residues, and also sidechain changes between distinct rotamer states, which involve large movements and significant energy barriers. Large-scale heterogeneity involves concerted positional differences of many residues, including changes in helix-helix positioning (Zheng et al., 2004) or β-sheet twist and rare cases of major refolding (Skehel and Wiley, 2000), as well as the long-recognized interdomain hinge motions. Any of these types of conformational dynamics can be integral to enzyme function, allosteric regulation, induced-fit binding and functional plasticity, by enabling the alternative structures required for biological function.

To investigate the molecular basis for SpA flexibility and the connection between local and global conformational heterogeneity more generally, we have determined several X-ray crystal structures of SpA C domain and B–B (two B domains connected by the conserved linker), all in the absence of any partner protein. C domain was solved to 0.9-Å resolution at cryogenic temperature (Fig. 1b). The structure shows many backbone and sidechain alternative conformations. Because previous work suggested that cryogenic freezing limits protein conformational heterogeneity (Fraser et al., 2011; Juers and Matthews, 2004), we also determined a 1.4-Å structure of C domain at room temperature. The B–B construct, determined at 1.5-Å resolution (Fig. 1c), is an informative mimic of the nearly identical A–B or B–C pairs. Because the linker was fully visible, the B–B structure reveals effects of the special case of linker proximity on conformational heterogeneity. This work establishes that SpA protein-binding domains exhibit extensive structural plasticity that presumably helps enable a small domain to accommodate multiple binding partners, and sheds light on the diverse nature of that plasticity.

RESULTS

Overview of C domain and B–B structures

Diffraction data for P2₁ C-domain crystals were collected at cryogenic and room temperatures and solved by single-wavelength anomalous dispersion (Table 1). Both structures contain one molecule in the crystallographic asymmetric unit and include all 58 residues of the SpA C domain (Fig. 1b), plus a Zn²⁺ ion that binds the chain ends. Both also share the same three-helix-bundle topology seen in previous SpA-domain structures (Figure 1b). Including N- and C-caps (Richardson and Richardson, 1988), the residue ranges are helix1: 6–19, helix2: 23–37, helix3: 40–56.

Table 1.

Data collection and refinement statistics

	C domain (cryogenic)	C domain (room temp)	B-B
Data collection
Space group	P21	P21	P65
Cell dimensions
a, b, c (Å)	27.3, 38.4, 28.6	27.8, 38.6, 29.0	44.4, 44.4, 214.8
α, β, γ(°)	90.0, 117.4, 90.0	90.0, 118.2, 90.0	90.0, 90.0, 120.0
	Peak	Peak
Wavelength	0.8	1.0	1.0
Resolution (Å)	50.0–0.9 (0.92–0.9)a	50.0–1.42 (1.44–1.42)a	50.0–1.49 (1.52–1.49)a
Rsym	0.07 (0.20)	0.07 (0.33)	0.10 (0.62)
I / σI	33.7 (4.36)	29.7 (2.54)	28.0 (2.2)
Completeness (%)	97.5 (77.6)	97.0 (73.6)	99.9 (99.6)
Redundancy	6.1 (2.7)	5.6 (2.9)	6.8 (4.4)
FOM	0.84	0.75
Refinement
Resolution (Å)	25.39–0.90	25.5–1.42	37.9–1.49
No. reflections	37938	9889	39026
Rwork / Rfree	11.0 / 12.7	11.2 / 14.1	14.2 / 18.5
No. atoms
Protein	1646	1524	6240
Zinc	1	1
Acetate	3	3
Water	137	49	409
B-factors
Protein	8.4	17.4	19.8
Zinc	5.7	13.2	–
Acetate	7.4	19.4	–
Water	19.7	36.0	32.5
R.m.s. deviations
Bond lengths (Å)	0.019	0.016	0.003
Bond angles (°)	1.802	1.424	0.712
Model Validationb
Ramachandran outliers (%)	0	0	0
Ramachandran favored (%)	98.2	99.0	99.0
Rotamer outliers (%)	0	1.1	0.5
C-beta outliers (%)	0	0	0
Clashscore	0.00	0.00	0.00
Overall score	0.50	0.52	0.50

^aData were collected from a single crystal. Values in parentheses are for the highest-resolution shell.

^bFrom the MolProbity-style validation in Phenix.

Diffraction data for P6₅ B–B crystals were collected at cryogenic temperature and solved by molecular replacement using cryogenic C domain as the search model (Table 1). There are 5 sequence differences between B and C domains, plus an F13W substitution in each domain of the B–B construct (Fig. S1) to allow fluorescence detection (see Methods). Crystals of B–B contain two molecules in the asymmetric unit (Fig. 1c). To limit the confusion associated with having two domains, two chains and alternative conformations, we labeled the chains X and Y, the alternative conformations a – d, and the residues as 1–58 for domain1 and 101–158 for domain2. The overall conformations of chains X and Y of B–B match closely, but domain1 and domain2 differ significantly, as does the detailed heterogeneity. We define the interdomain linker in B–B as the residues between helix3 of domain1 and helix1 of domain2 (i.e., 58 and 101–105). In each chain, this linker is in an extended conformation making substantial contact with domain2 across the same fairly hydrophobic helix1–2 face that binds to F_c. Because of this linker conformation, the two domains are translated and flipped in orientation relative to one another, forming a hook-like structure overall (Fig. 1c.)

Overview of SpA conformational heterogeneity

The high-resolution data show that SpA domains have much more conformational heterogeneity than observed in the previous crystal structures, or even than other crystal structures generally. Discrete backbone alternative conformations were identified for 55% of residues (31 of the 58 total) in room-temperature C domain, 62% in cryogenic C domain, 61% in B–B chain X, and 74% in B–B chain Y. Doing justice to such complexity required non-traditional approaches to achieve consistency within and between the alternative models, as described in the Methods section. The extensive alternative-conformation modeling contributed to achieving the outstanding validation statistics reported in Table 1.

In addition to sidechain shifts riding on the more global backbone shifts, all three structures have extensive individual sidechain heterogeneity. Figure 2a shows examples for two surface charged sidechains and an interior aliphatic sidechain, with electron density to support as many as four distinct rotamers. To quantify conformational heterogeneity among all six domains, we computed the maximum distance between equivalent atoms of alternative conformations within each residue along the sequence, separately for backbone atoms and for sidechain atoms. The three-dimensional distribution of this heterogeneity within single domains is shown for cryogenic C in Figure 2b as the local width in a putty-sausage diagram. The greatest sidechain heterogeneity occurs at sidechains pointing away from the interior, not unexpectedly. This surface heterogeneity occurs in sidechains without strong intermolecular interactions, but an exception is the Tyr114 sidechains in B–B chains X and Y, which interact with each other across molecules in the asymmetric unit.

Figure 2.

Residue-level conformational heterogeneity. (a) Examples of heterogeneity in C domain. Lys42 and Glu24 sidechains are solvent-exposed, and Ile34 is in the interior. The maximum distance between equivalent sidechain atoms (see text) is depicted for each residue. (b) Putty-sausage diagram of C domain where the relative diameter represents the maximum distance between equivalent atoms of alternative conformations (if present) within each residue. The color coding reflects the type of heterogeneity. (c) Comparison of residue-level conformational heterogeneity of single domains in C domain and B–B structures. B–B chain and domain designations are indicated as a suffix to the protein name. For example, “X1” indicates domain1 of chain X. See also Figures S3 and S4.

Figure 2c and Figures S3 and S4 use different representations to compare the distribution of conformational heterogeneity between the six different domains in these structures. The overall pattern is very similar in the C domains and the first domains of B–B. An interesting exception is the lower heterogeneity of C domain at the beginning of helix3, close to the main cluster of sequence differences between C and B. The most conspicuous contrast, however, is between domain2 of both B–B chains and that of the other four domains. In particular, helix1 is significantly less heterogeneous in domain2, probably due to interaction between the interdomain linker and helix1 of domain2 (discussed below). This perturbation by the linker appears to induce greater changes in conformational heterogeneity than data-collection temperature, sequence differences between B and C domains, or the fact that C domain is 0.5 ± 0.15 kcal/mol more stable than B domain (A.H. and T.G.O, unpublished data).

Conformational coordination of sidechain alternative conformations

Many residue-level conformational changes involve rotamer shifts of neighboring sidechains (Fig. 3a–c), where only some local combinations of conformations are possible without prohibitive steric clashes. Since electron density only shows the sum of all conformations, other information is needed to correctly analyze the conformational clustering, including all-atom sterics and the logic of assigning alternative conformation labels and occupancies.

Figure 3.

Concerted conformational heterogeneity in B–B. (a,b,c) Populations of Tyr114-X, Gln110-Y, and Tyr114-Y for the alt-a (a, 60%), mixed alt-a/alt-b (b, 8%), and alt-b (c, 32%) conformations. (d,e,f) Trp13 in Domain1 (d) and Trp113 in Domain2 (e) constrain the rotamers of nearby residues, while the mixed conformations of the two domains when superimposed (f) are incompatible with one another due to significant overlap of the van der Waals spheres, as indicated by red spikes.

The two domains within each B–B chain have nearly the same backbone conformation, but differ in backbone heterogeneity (see above) and also show differences in sidechain conformation and heterogeneity, some of which occur in coordinated networks. An extreme example is Tyr114 (domain2), which adopts two distinct rotamer conformations, while Tyr14 (domain1) adopts only one rotamer. Trp13 points toward both Tyr14 and Leu17, which must therefore both point away from Trp13 (Fig. 3d), but Trp113 points away from Tyr114 and Leu117, which lets them adopt alternative conformations (Fig. 3d). The Trp13 conformation in domain1 is incompatible with the conformations of Tyr114 and Leu117 in domain2 because that combination would produce serious steric clashes. The position of Trp13 therefore constrains the conformations of surrounding residues, while Trp113 is in turn limited by the presence of the linker bound to domain2. Based on refined occupancies of Gln110 in chain Y and Tyr114 in both chains, we can estimate the populations of three sets of rotamers, depicted in Fig. 3a–c. The most populated set (Fig. 3a) has a population of ~60%, and the second set (Fig. 3c) has a population of ~30%. The least populated set (Fig. 3b) has a much lower population of ~8%, presumably because of a somewhat unfavorable contact between the Tyr rings, which may move apart by an amount too small to be distinguished within the summed density.

Heterogeneity in helix1 orientation

With the structures of unbound C domain and B–B solved, we now have six structures of SpA domains with which to analyze and compare conformational heterogeneity, as well as the various single-domain structures already in the PDB. Montelione and coworkers (Zheng et al., 2004) previously reported wide disagreement over the tilt angle of helix1. Our structures demonstrate that these differences represent actual differences in helix1 orientation between structures, rather than errors in one or more of the NMR structures. Our results confirm that helix2 is the least variable and helix1 the most, with more change parallel than perpendicular to the helix2–3 plane. To allow quantitative comparison of helix1 orientations among current SpA-domain structures, we transformed the helix axis vectors for each domain into the same coordinate system with helix2 as the principal reference (see Methods). The superimposed helix axes and the relative helix1 orientations for 10 different SpA-domain structures are shown in Figure 4a and Table S1, respectively. Our crystal structures differ by up to 7.5° from one another (Fig. 4b), 10° from 1Q2N (Zheng et al., 2004), and 13° from 1DEE (Graille et al., 2000).

Figure 4.

Global conformational heterogeneity of Hlx1 in SpA protein binding domains. (a) Helix-axis orientations of sixteen different SpA domains in a coordinate system where the x-axis is parallel to Hlx2, and Hlx3 lies nearly in the x-y plane. The cryogenic C domain (C) and 1Q2N Z domain (Z) structures are labeled. (b) Superposition of both C domain structures with the four individual domains of our B–B structure. (c) Superposition of Z domain (1Q2N, variant of B domain) and cryogenic C domain using Hlx2 and Hlx3 for the superposition. (d) Peel-away of interior of the Hlx1-Hlx2/3 interface, shown from both sides. Interfacial sidechains are highlighted with colored spheres. The key core residue at the interface is Ile16, which forms the pivot point for variation in the angle of Hlx1 (see text). (e) The helix differences between 1Q2N and C domain are the consequence of significant rearrangements in the interhelical knob packing for all three helices. Residues and spheres are colored as in panel b, which represent structure affiliation. The pivot, marked in the figure, consists of Ile16.

While there exist some sequence differences between the different domains, these nonconserved residues are not at the helical interfaces, so differences in interhelical angles cannot be attributed directly to local sequence differences. Also, nearly the full range is seen among B–B domains. We hypothesize that subtle differences in packing due to residue-level conformational differences in and out of the direct interface are responsible for helix1 rearrangements in single-domain SpA structures. To test this, we compared the interhelical packing in C domain with the canonical B-domain NMR structure, 1Q2N (Zheng et al., 2004) (Fig. 4c). Helix1 differs most toward its C-terminal end, with a pivot point around Ile16. In analogy to Crick’s “knobs and holes” (Crick, 1953), this important “knob” sidechain nestles into a deep cavity between helix2 and helix3 and makes six sidechain contacts across the interface (Fig. 4d). Ile16 adopts the same conformation in both structures, and four of those six interface contacts (to Phe30, Ile31, Leu45, and Ala48) are identical. This is consistent with previous work (Braisted and Wells, 1996) which suggested that Ile16 stabilizes the packing of helix1 to helix2. The interaction of Leu45 with Ala12 is also the same in both structures. Figure 4d shows the key knob residues on the N- and C-terminal ends of helix1 that intercalate with “hole” residues across the interface, thereby holding the three-helix-bundle together. Most assume unique conformations in each structure, leading to differences in knob locations (Fig. 4d) when viewed down the helical axis. For example, both Leu19 and Asp52 sidechains shift significantly to maintain their contact, while Leu22 changes rotamer from mt (minus, trans) to tp (trans, plus) in most 1Q2N models to maintain contact with Ile19 in the new helix1 orientation.

Effect of the B–B interdomain linker on conformational heterogeneity

In the context of the B–B structure (see Fig. 1c), both interdomain linkers form a rather extensive and well-packed interface with helices1 and 2 of domain2 (including Phe105 and Trp113), on the same surface region that binds F_c in the 1FC2 complex structure. Recent NMR studies have established that 6 residues (K₅₈A₁₀₁D₁₀₂N₁₀₃K₁₀₄F₁₀₅) within the linker sequence are highly flexible in solution (A.H. and T.G.O, unpublished). For this reason, we know that no single domain-domain arrangement, including this one, can predominate in solution. Given the linker flexibility, this interaction must be quite weak, yet it seems the most likely cause of the substantial reduction in conformational heterogeneity for helix1 of domain2 in both B–B chains (Fig. 2c). This conformational sensitivity of helix1 to contacts with other parts of the molecule suggests that its orientation can be readily frozen out when it is part of an interface.

Relationship of conformational heterogeneity to diversity of binding partners

The large set of SpA-domain crystal structures now available, especially the new ones at high resolution, allows analysis of the relationship between intradomain conformational heterogeneity and the potential variety of binding partners. To do this, we have focused on comparing the helix conformation and interface packing of the 1FC2 B domain/F_c complex (Deisenhofer, 1981) with the variety of conformations and interfaces seen in the other crystal structures.

Table S1 shows that the helix1–2 interhelical angle is largest for the two C domain and B–B chain Y domain2 structures, and very small for the B domain/F_c structure (1FC2). Therefore, for maximum contrast, we docked the 1FC2 complex onto C domain by superimposing the 1FC2 B domain onto the cryogenic C structure. We then examined the non-cognate interface between C domain and the docked F_c. The well-fit interface of the cognate interaction (Fig. 5a) is replaced with a combination of huge steric clashes and non-interacting gaps (Fig. 5b). This interaction actually occurs (since C domain is known to bind F_c (Jansson et al., 1998)) and must therefore be capable of relaxing to fit. An attempt to see whether purely sidechain shifts in the C domain could alleviate the clashes was surprisingly successful, but necessitated coordinated rotamer changes for an entire network of large sidechains and failed at adding positive interactions. The majority of sidechain atoms consistently shifted toward the surface of the complex, leaving gaps in the packing.

Figure 5.

Coordination of backbone and sidechain conformational changes with binding of F_c. (a,b) All-atom contacts at binding interfaces. (a) Well-fit, cognate interface of F_c with SpA B domain (low-angle helix 1–2 orientation), in PDB file 1FC2; (b) Poorly fit, non-cognate interface of F_c with superimposed high-helix-angle conformation of C domain from 4NPD. Green and blue dots show good van der Waals contact or H-bonds, and clusters of red spikes show steric clashes. (c,d) Correlation of changes in helix-helix angle with concerted rotamer changes in a network of 7 sidechains (bold colors) for (c) two low helix-angle structures (B–B domain 1 for both chains), similar to the B domain in 1FC2, and for (d) two high helix-angle structures (C domain and B–B chain Y Domain2). Switching only sidechain network conformations on the superimposed, high-angle, non-cognate C model from panel b can eliminate most clashes but neither recreates good van der Waals packing nor any H-bonds. However, the high-angle domain structures each form an extensive, well-packed contact with another SpA-domain partner in the crystal. See also Table S1.

Measurement of interfacial solvent-accessible surface area (Chothia, 1976) is a useful way to correlate structure to binding affinity for evolved, well-packed interfaces, but ignores gaps of up to 2.8Å. As a more sensitive measure for this modeled and possibly ill-packed interface, we used all-atom contact analysis¹⁸, which counts only atom-atom contacts within 0.5Å of ideal van der Waals distance. The cognate interface in 1FC2 has 142Ų total all-atom contact, 110Ų of favorable H-bond plus van der Waals, and one clash (see Table S1 for details). The initial non-cognate interface has only 54Ų favorable and 12 clashes; the rebuilt non-cognate interface has only one clash but still just 68Ų of favorable contact. For comparison, the internal interface between helix1 and 2–3 is 153Ų total (147Ų favorable), the B–B linker contacts domain2 across 106Ų, and a sample of tight biological dimer and inhibitor interfaces ranged from 100 to 460Ų. Crystal contacts typically have fairly little direct contact – a sample ranged from 18 to 40Ų. Of note for SpA, however, C domain makes crystal contact across 121Ų plus a Zn site, and the B–B crystals alternative contacts of 144 and 258Ų; presumably this helps attain high resolution.

The coordinated network of changes in sidechain conformation found in the non-cognate modeling is recapitulated elsewhere in our crystal structures, and it gives a broader context for the constraints on possible rotamer pairings seen in Figure 3 when assigning consistent alternative-conformation models in the B–B structure. Of our six domain structures, the high interhelix angle and sidechain arrangement of C domain is also seen in the second domains of B–B, while the low interhelix angle and sidechain arrangement of 1FC2 B domain is seen in the first domains of B–B. Notably, all six domains make extensive crystal contacts across this helix1–2 surface, in three different arrangements. As shown in Figure 5c–d, seven sidechains make concerted rotamer changes: Ile31 on helix2 and Gln10, Asn11, Phe/Trp13, Tyr14, Leu17, and His18 on helix1. Some of the coordination between adjacent sidechain rotamers is constrained but not all: the more exposed Asn11 and His18 adopt multiple alternative conformations, and are not sequence conserved. In summary, there is substantial coordination between backbone and sidechain conformational heterogeneity in SpA domains, and also a probable relationship between that combined heterogeneity and the ability to form contacts with a variety of protein partners.

DISCUSSION

Interpreting conformational heterogeneity in crystallographic data

Representing protein structures as collections of possible models consistent with the data is common practice for structures determined by NMR. Although these are often called ensembles, there is usually no direct evidence that they represent the true ensemble of structures present in solution. Recent efforts to incorporate dynamic NMR information into NMR model refinement have produced more realistic representations of true ensembles (Lindorff-Larsen et al., 2005). Ensemble building is a relatively new pursuit in X-ray crystallography, and three current methodological approaches are promising but still in their infancies. The first approach repeats crystallographic model building and refinement from multiple randomly seeded starts (DePristo et al., 2004; Terwilliger et al., 2007), which is good at identifying uncertainty within the major conformation but seldom captures large differences or minor populations. The second approach is molecular dynamics-based ensemble refinement, for instance as currently under development in the PHENIX crystallographic suite (Adams et al., 2010b; Burnley et al., 2012). When we tried it on our structures, individual results failed validation and only some of the sidechain alternative conformations evident in electron density were captured. A third approach is the qFit program (van den Bedem et al., 2009), which uses low-level electron density to identify potential alternative conformations for the next atom but has not yet been extended to model complete alternative sidechain conformations or to correlate neighboring changes. Automated methods for modeling and coupling conformational networks, like that used in qFit and CONTACT (van den Bedem et al., 2013), respectively, speed up the tedious process of modeling sidechain heterogeneity and very accurately couple sidechain alternative conformations. However, the subtle anisotropy observed in our electron density, particularly that of the backbone, is difficult for the current automated methods to fit. Therefore we used manual placement of sidechain and backbone atoms to identify all alternative conformations.

Because of these difficulties, our initial steps toward building a complete, consistent conformational ensemble from the crystallographic data were limited to careful identification of alternative conformations one residue at a time. To accomplish this, we extended traditional modeling of discrete alternative conformations based on anisotropy or separate peaks in the electron density (see Methods). Briefly, we kept isotropic B-factors until all visible backbone and sidechain alternative conformations had been built. These conformations were assigned to specific, internally consistent models of alternative conformations using a combination of MolProbity conformational and all-atom clash analysis, plus PHENIX refinement of occupancies.

Origin of high SpA-domain conformational heterogeneity

Increasingly more structures in the PDB break 1Å resolution, where alternative conformations are most visible. Even among these ultra-high-resolution structures, however, most have about 10% backbone heterogeneity; a few (e.g., 1EXR, 2I16, 1YK4, 3NIR, 2VB1) include alternative conformations that stretch for more than two residues and have between 25% and 36% backbone alternative conformations. We have found only the PDB ID 1M40 β-lactamase (Minasov et al., 2002), fit with 63% backbone alternative conformations, in or near the indeed unusual 55–74% range seen in our SpA structures. The backbone alternative conformations within a single SpA-domain crystal structure differ by only 1Å or less, but their pattern is very similar to the larger changes between structures. Prevalence of sidechain alternative conformations in SpA is large but hard to compare quantitatively with other structures because many ride on backbone shifts.

There are several features of SpA that may lead to this unusual conformational heterogeneity. For example, SpA domains are small, single-chain three-helix bundles, each held together by the Ile16 pivot residue (Fig. 4d). The lack of an extensive hydrophobic core might facilitate conformational heterogeneity by allowing adjustment of interhelical angles with less constraint from interior van der Waals interactions.

Another consideration is that multiple backbone conformations may be less rare in other proteins than suggested by their occurrence in crystal structures, because all types of partial disorder are more prevalent at low resolution but can be well characterized only at high resolution. An unusual feature of SpA domains may be that, despite having high conformational heterogeneity, they can still form high-resolution crystals. All our crystals have extensive contacts that form a continuous zigzag of domains along a crystallographic screw axis – a heterologous contact in P2₁ for C domain, and an alternation of domain1 and domain2 pseudo-2-folds in P6₅ for B–B, each of which uses the helix1–2 face. The overall domain orientations must repeat in exactly 180° for P2₁ and in exactly 60° for P6₅, which these domains achieve using varied interhelical angles and sidechain positions. Therefore we speculate that these well-ordered crystals are enabled by the same plasticity used to bind different biological targets.

Another unusual property of SpA domains is that they unfold and refold very rapidly, in less than 100 ms (Myers and Oas, 2001) (A.H. and T.G.O, unpublished data). Although kinetic and thermodynamic experiments establish that the folding reaction is two-state, the rapid kinetics implies a relatively low activation barrier between the folded and unfolded ensembles. This characteristic of the overall energy landscape may be correlated with the complex native-state landscape suggested by the high conformational heterogeneity we observe. In any case, it seems likely that evolution has converged for functional reasons on a sequence that not only includes a flexible linker but also provides sidechain networks that can support alternative helix packing (Fig. 4) and alternative binding interfaces (Fig. 5).

Possible functional roles of intradomain conformational heterogeneity

From the structure of B domain bound to F_c (PDB ID 1FC2), Deisenhofer identified Tyr14 and Gln10 as residues of B domain involved in binding by forming H-bonds to backbone oxygens of F_c (Deisenhofer, 1981). Gln10 is highly heterogeneous and Tyr114 shows two distinct conformations in our structures (Fig. S3), as part of a network of sidechain differences coordinated with helix1 orientation (see Results). We superimposed the mainchain atoms of residues 110–118 from B–B domain2 on the same residues in 1FC2. Only Tyr114 alt-a (Fig. 6a) t90 is capable of forming the H-bonds for the F_c interface while avoiding clashes, and only when on a low-angle helix1 conformation. Tyr114 alt-b is not compatible with binding because it sterically clashes with the F_c molecule.

Figure 6.

Conformational compatibility of B–B with antibody binding. (a) Tyr114 alt-a (blue) adopts the same conformation as seen in the F_c/B domain complex (1FC2, magenta), but Tyr114 alt-b (cyan) is incompatible. The Tyr alt-a sidechain is posed for interaction with Leu432 from F_c. (b) The Asp36 conformation in Domain1 is able to bind Arg2519 of F_ab (as it does in the structure of F_ab in complex with D domain, 1DEE), but the Asp136 conformation in Domain2 is not possible due to a steric clash with the F_ab molecule. (c,d) Functional relevance of interdomain orientation. (c) The crystallographic interdomain conformation of B–B is incompatible with F_c binding because when Domain1 is posed as in 1FC2, the second domain clashes with the F_c molecule. In order to bind to F_c, B–B must adopt another interdomain conformation. (d) The interdomain orientation of B–B is compatible with binding two F_ab molecules, when superimposed with 1DEE.

Similarly, Silverman and coworkers implicated Asp36 as a residue on D domain that binds antibody fragment F_ab via H-bonding to Arg2519 (Graille et al., 2000). In our set of structures Asp36 and Asp136 are quite heterogeneous, using different rotamers to form H-bonds with different partners. To see which of these conformations are compatible with binding F_ab, we superimposed both domain1 and domain2 of B–B onto 1DEE (Figure 6b). Asp36 in domain1 shows even slightly better H-bonds to the Arg than in 1DEE, while Asp136 in domain2 clashes seriously with the F_ab.

Another intriguing sidechain-level feature of the current structures is a possible role for Phe5 in binding interactions that involve the helix1–2 surface. The Phe5 sidechain flips out to allow contacts with a second molecule in the 1FC2, 1DEE, and 1H0T complexes. Given the rather low order parameter of its NH in solution (A.H. and T.G.O, unpublished), Phe5 must spend considerable time disordered, along with the rest of the linker. However, it is tucked down over one end of the helix1–2 surface in all six of our SpA-domain structures and in the NMR structures 1SS1, 1BDD, 2SPZ, and 1Q2N (where NOEs were observed between the Phe ring and domain surface residues (Doreleijers et al., 2005)). The helix1–2 surface, the most hydrophobic face on SpA domains, forms three quite different, well-packed crystal interfaces in our structures, and is likely to play an important role in binding other biological partners of SpA besides F_c.

Most interestingly, the large total set of SpA domain structures now available, especially these new ones at high resolution, allow study of possible coupling between intradomain conformational heterogeneity and the variety of binding partners. We have shown that: 1) Helix1 orientation differs by 7.5° among our six domain structures, and up to 13° from other crystal and NMR structures; 2) Helix orientation is correlated with a network of sidechain rotamer changes on the helix1–2 face (Fig. 5); and 3) F_c binding to that face is compatible only with the low-helix-angle form of the concerted sidechain/helix rearrangement, while other well-packed contacts incorporate the high-angle version (Table S1). It is extremely difficult to establish that backbone conformational change is essential for binding many partners, but we can definitively say that concerted backbone/sidechain changes are in fact used for that binding.

Multi-domain conformational preferences in SpA

The contact between D domain and F_ab uses a different face of SpA domains than the contact between B domain and F_c. As a result, a single domain could actually bind both F_ab and F_c at the same time, which Graille et al. (2000) demonstrated by building a hypothetical model of the ternary complex. While some sidechain conformations and helix orientations found in C domain and B–B are compatible with binding F_c (as discussed above), the domain-domain relationship observed in each B–B chain would block binding of F_c to either domain due to major steric overlap with the other linked domain (Fig. 6c). This observation confirms the NMR evidence that, in solution, B–B is free to adopt many other domain-domain conformations, thus allowing F_c to bind. Unlike the case with F_c, the global conformation of the B–B chain is compatible with binding up to two F_ab molecules simultaneously on the negatively charged helix2–3 faces, with no steric interference (Fig. 6d). Co-crystals of B–B or other SpA multi-domain constructs with F_c or F_ab would be desirable in order to understand SpA binding.

EXPERIMENTAL PROCEDURES

Plasmid construction

The C-domain gene was PCR cloned from the SpA-N gene. The PCR primers added a 5′ NdeI and a 3′ BamHI site, and were subsequently cloned into the T7 expression plasmid pAED4(Doering, 1992). The B–B gene was synthesized by GENEWIZ, Inc. in a pUC57 cloning vector, and was subsequently cloned into the pAED4 expression vector. It contained the F13W substitution to aid detection of the protein.

Protein expression and purification

Plasmids were transformed into Escherichia coli BL21(DE3) cells using standard transformation procedure. A single colony of transformed bacteria was used to inoculate a 50 ml culture of LB media with 0.1 mg/mL ampicillin. This starter culture was incubated at 37°C until the O.D. reached 0.8–1.0, whereupon it was used to inoculate 1L cultures that were allowed to grow to O.D. 0.8–1.2 at 37°C. IPTG was then added to a final concentration of 1 mM, and the cultures were incubated for an additional 6–8 hours. The cells were harvested by centrifugation and resuspended in 20–30 mL of 50 mM Tris pH 8.8, 1 mM EDTA, and protease inhibitor cocktail (AEBSF, pepstatin, bestatin and E-64). The cells were lysed in a French pressure cell, and insoluble material was centrifuged from the lysate. The lysate was brought to pH 9.0, and micrococcal nuclease was added to digest large DNA fragments for 15 min. The resulting solution was brought to 4 M guanidinium HCl (BioBasic, Inc.) and 20mM TCEP was added. In two successive steps, the solution was dialyzed in a 5% acetic acid buffer, which precipitated many cellular materials, but not expressed proteins. After centrifugation of the insoluble material, the resulting solution was allowed to dialyze overnight into deionized water. The protein was further purified using two types of ion exchange chromatography. First, the protein was loaded onto a strong cation exchanging, SP Sepharose (GE) column in 50 mM acetate buffer at pH 3.6. The column was eluted with a NaCl gradient (typically 100 mM to 500 mM gradient) in a volume of 600–800 mL and collected in 10 mL fractions monitored by a UV detector (Bio-Rad) at 278 nm. The fractions comprising the protein elution peak were checked for purity by SDS-PAGE. The purest fractions were pooled and dialyzed against deionized water. Subsequently, the resulting solution was loaded onto a weak anion exchanging DEAE Sephacel (GE) column in 25 mM Tris (BioBasic, Inc.) at pH 10.0 and eluted with a 800 mL NaCl gradient (typically 0 to 250 mM gradient) into 10 mL fractions monitored by a UV detector at 278 nm. The protein elution peak fractions were checked for purity by SDS-PAGE. The purest fractions were pooled and dialyzed against deionized water. The final solution was lyophilized and stored in a desiccator. Purities of the final protein stocks were confirmed by SDS-PAGE to be >95% pure. The masses of all proteins were confirmed by ESI-MS.

Crystallization, data collection, and initial models

C-domain protein solution was prepared at 20 mg/mL from lyophilized protein and mixed in a 1:1 ratio with crystallization solution (55 mM ZnCl₂, 16.5% PEG 6000, and 0.1 M MES pH 6). Crystals formed by hanging-drop vapor diffusion within five days and were of the space group P4₃32, although not of diffraction quality. Seed stock was prepared by adding approximately 10 of the P4₃32 crystals into a 50 μL aliquot of well solution in a Hampton Seed Bead, and the resulting microseeds were used in a new crystal screen. Diffraction-quality crystals were formed by hanging-drop vapor diffuction within two days by mixing C domain, well solution (30 mM NaSCN, 26% (w/v) PEG 3,350, and 0.5% (v/v) glycerol), and seeds in a ratio of 3:2:1, respectively.

For the cryogenic data set, cryo-preservation of the crystal was achieved with the addition of 30% glycerol to the crystallization condition, and the crystal was frozen by direct immersion in liquid nitrogen. All data were collected remotely at the Advanced Photon Source (APS) at Argonne National Laboratory, beamline 22-ID (SER-CAT). Data were processed and scaled with HKL2000 (Otwinowski and Minor, 1997). For the cryogenic C-domain structure, despite an I/σI of greater than 2 at the next higher resolution bin, the data were truncated to 0.9-Å resolution because completeness dropped below 70%. C-domain structures were solved using the single-anomalous dispersion method at the high-remote for zinc at 0.8-Å wavelength (cryogenic structure) and 1.0-Å wavelength (room-temperature structure) using SHELXC/D/E(Sheldrick, 2010) (Table 1). The initial models for C domain were built in COOT(Emsley et al., 2010).

The B–B protein solution was prepared at 20 mg/mL from lyophilized protein and mixed in a 1:1 ratio with crystallization solution (60 mM MES (pH 6), 3.0 M ammonium sulfate, and 2.5% (v/v) glycerol). Diffracting crystals formed by hanging-drop vapor diffusion within two months and were of the space group P6₅. Cryo-preservation of the B–B crystal was achieved by transferring to sodium malonate (3.4 M and pH 6), and the crystal was frozen by direct immersion in liquid nitrogen. All data were collected at the Advanced Photon Source (APS) at Argonne National Laboratory, beamline 22-ID (SER-CAT), at 1.0-Å wavelength. Data were processed and scaled with HKL2000(Otwinowski and Minor, 1997) (Table 1).

The B–B structure was solved by molecular replacement in PHASER (McCoy et al., 2007) using the cryogenic C-domain structure as the search model, which has a 91% sequence similarity to each domain of B–B. The search model was modified to remove any alternative conformations and ten mobile residues at the N terminus. The initial model for the individual domains of B–B was built by AUTOBUILD (Terwilliger et al., 2008) and then rebuilt in COOT (Emsley et al., 2010). The residues in the interdomain linkers were manually built in COOT (Emsley et al., 2010).

Refinement and heterogeneous model building

Refinement of all models was carried out in phenix.refine(Adams et al., 2010a). Initial refinement was performed using two isotropic B factors per residue. Once the model was placed, it was refined using isotropic Bs for all atoms (Figure S2a). After several rounds of iterative refinement and model building, the visible concerted anisotropy was used to fit mainchain alternative conformations in segments (e.g., helix1). Each alternative-conformation segment was rotated and translated in COOT(Emsley et al., 2010) to fit the direction of anisotropy in the carbonyl oxygen 2F_o − F_c density, effectively separating these alternative conformations (Figure S2b). The sidechain for each alternative conformation was then fit down to 0.3–0.5σ levels. Candidates that could not be fit rotamerically or that later refined to zero occupancy were discarded. To avoid bad geometry across peptide bonds, backbone alternative conformations were continued to the flanking Cα atoms if needed, using backrubs (Davis et al., 2006) in a new PHENIX (Adams et al., 2010a) utility written for this purpose.

After all mainchain and sidechain alternative conformations were fit, several rounds of occupancy and anistropic B-factor refinement were performed. All atoms but H were refined as anisotropic for the cryogenic C-domain structure, and all atoms but H and waters were refined as anisotropic for the room-temperature and B–B structures. The 2F_o-F_c and F_o-F_c density at each residue was checked after each round of refinement. Final refinement for all structures was carried out in phenix.refine using automatic weight optimization. The final refined models for the cryogenic structure and room-temperature structure had R_work / R_free values of 11.0%/12.7% and 11.2%/14.1%, respectively, while the final refined model for the B–B structure had R_work / R_free values of 14.2%/18.5%.

Illustrations

Coordinate files for molecules in crystal contact were produced in Chimera (Pettersen et al., 2004). Their all-atom contact dots were produced by the interface feature in MolProbity, counted in Mage, and converted to Ų by dividing by the default contact-dot density of 16/Ų. Figure 5 was made in KiNG. Figure 4a was constructed using Prekin and Mage. All remaining structural illustrations were made in PyMOL(Schrodinger, 2010). Figure 2c was made in Mathematica(Inc., 2010).

Supplementary Material

supplement

Figure S1, Related to Figure 1. Sequence alignment for the five protein-binding domains of SpA. Conserved residues are indicated in red. The F13W substitution is indicated in blue. The linker between E and D domains includes a three-residue insertion.

Figure S2, Related to Experimental Methods. Discrete alternate conformations in C domain. (a,b) Anisotropic 2F_o − F_c density (contoured at 1σ) around Phe5 using (a) only anisotropic B factors and one conformation and (b) two discrete conformations. Excessive F_o − F_c density at 2.5σ remains in panel a but is resolved in panel b.

Figure S3, Related to Figure 2. Conformational heterogeneity in C domain, by residue and type. Number of conformations is shown as bars and maximum distance between identical atoms is shown as points connected by lines. (a) Data collected at 100 K. (b) Data collected at room temperature. The similarity of the two plots suggests that the conformational heterogeneity is conserved over a wide range of temperatures.

Figure S4, Related to Figure 2. Conformational heterogeneity in B–B, by residue and type. Residue numbering scheme is 1 – 58 for domain1 and 101 – 158 for domain2. Number of conformations is shown as bars and maximum distance between identical atoms is shown as points connected by lines. (a) Data for chain X of the ASU. (b) Data for chain Y of the ASU. The pattern and extent of conformational heterogeneity in domain2 is quite different from that of domain1, in both chains.

Table S1, Related to Figure 5. Comparison of interhelical angles and contact surfaces (see Notes that follow).

HIGHLIGHTS.

• Staphylococcal protein A is a highly dynamic protein that binds to many proteins

• Atomic-resolution crystallography detects alternative conformations in most residues

• Helix 1 conformation is tuned by sidechain rotamers around a pivot residue

• Concerted backbone and sidechain changes allow for binding many partners