A segment of cold shock protein directs the folding of a combinatorial protein

Abstract

It has been suggested that protein domains evolved by the non-homologous recombination of building blocks of subdomain size. In earlier work we attempted to recapitulate domain evolution in vitro. We took a polypeptide segment comprising three β-strands in the monomeric, five-stranded β-barrel cold shock protein (CspA) of Escherichia coli as a building block. This segment corresponds to a complete exon in homologous eukaryotic proteins and includes residues that nucleate folding in CspA. We recombined this segment at random with fragments of natural proteins and succeeded in generating a range of folded chimaeric proteins. We now present the crystal structure of one such combinatorial protein, 1b11, a 103-residue polypeptide that includes segments from CspA and the S1 domain of the 30S ribosomal subunit of E. coli. The structure reveals a segment-swapped, six-stranded β-barrel of unique architecture that assembles to a tetramer. Surprisingly, the CspA segment retains its structural identity in 1b11, recapitulating its original fold and deforming the structure of the S1 segment as necessary to complete a barrel. Our work provides structural evidence that (i) random shuffling of nonhomologous polypeptide segments can lead to folded proteins and unique architectures, (ii) many structural features of the segments are retained, and (iii) some segments can act as templates around which the rest of the protein folds.

It is thought that individual protein domains have arisen by assembly and/or exchange of small gene segments, for example, by exon shuffling or by DNA recombination (1, 2). In the case of exon shuffling, it has been proposed that the encoded segments comprise entire structural elements (3) and also that such segments may be capable of folding autonomously (4, 5). Indeed, the formation of local structure would be expected to facilitate overall folding by restricting the conformational space to be sampled. Local structure has been identified experimentally in denatured proteins (6), for example, helices in apomyoglobin and ribonuclease (7, 8). Local structure has also been identified on the folding pathway of proteins; for example, hydrophobic clusters in barnase (9) form a site for the consolidation of several β-strands and a nucleation point for folding (10). However, it has been more difficult to illuminate experimentally the role of local structure in the evolution of protein domains.

In earlier work, we investigated the ability of a polypeptide segment to serve as a building block for the creation of folded domains (11). We took a segment from a five-stranded, antiparallel β-barrel, the monomeric major cold shock protein (CspA) of Escherichia coli (12, 13). This protein belongs to an ancient family that is present in both prokaryotes and eukaryotes (14) and binds to both RNA and single-stranded DNA (15). Its N-terminal 36 residues correspond structurally to the first three strands and include a hairpin turn and an aromatic cluster, which unfold more slowly than the rest of the domain and are thought to act as a nucleation point for folding (16, 17). Furthermore, this segment has high sequence similarity with a complete exon present in a number of homologous eukaryotic proteins (Fig. 1A) (18). We took this segment and fused it to polypeptide segments from other natural proteins encoded by randomly fragmented genomic E. coli DNA. We displayed the resulting repertoire of chimaeric polypeptides on the surface of filamentous bacteriophage and selected by proteolysis a number of stably folded proteins (as characterized by cooperative unfolding and the distribution of chemical shifts in NMR). To further understand the role of this segment in domain creation, we attempted to determine crystal structures of the chimaeric proteins. Here we describe the structure of one of these proteins, 1b11, in which the C-terminal region incorporates 34 residues from the E. coli 30S ribosomal protein S1 (Fig. 1B) (19). Both CspA (Fig. 2A) and S1 (Fig. 2B) have the same fold but low sequence identity: <30% identity by using clustalw (20).

Fig. 1.

Sequences. (A) Multiple alignment of CspA template and exons from eukaryotic proteins. Sequences are named according to organism of origin, and Ensembl gene IDs (www.ensembl.org) for each eukaryotic sequence are provided here: 1, CspA from E. coli; 2, ENSANGG00000008966 from Anopheles gambiae; 3, CG17334 from Drosophila melanogaster; 4, ENSG00000153048 from Homo sapiens; 5, ENSRNOG00000002610 from Rattus norvegicus; 6, SINFRUG00000134199 from Fugu rubripes; 7, ENSANGG00000019305 from Anopheles gambiae; 8, CG9705 from Drosophila melanogaster. The N-terminal part of the Drosophila (no. 3) exon was truncated to show only the matching region. Positions with 100% identity are shaded. (B) Secondary structure (arrows) in parent proteins and 1b11; CspA template and S1 domain segment are in blue and red, respectively. Other residues (N-terminal His-tag, linker, and C terminus of 1b11) are not colored. Of the 103 residues of 1b11, residues 16-101 were identified in the final structure.

Fig. 2.

Structure of 1b11. (A and B) Secondary structure cartoons of CspA (PDB ID code 1MJC) with template residues 1-36 in dark blue (A) and the S1 domain from E. coli polynucleotide phosphorylase (PDB ID code 1SRO) with residues 5-45 (as found in 1b11) in dark red (B). (C and D) 1b11 tetramer (C) and dimer (D) shown with a smoothed backbone tube; chains A and A′ are gray, and B and B′ are green. (E) 1b11 barrel with strand assignments; strands β1-β4 (residues 16-72) of chain A and β5-β6 (residues 73-101) of chain B are colored as in C and D. Strand β6 is split into β6a and β6b to indicate its change of direction. (F) 1b11 barrel showing origin of segments, with strand assignments of parent proteins marked and colored as in A and B.

Materials and Methods

Sample Expression and Purification. 1b11 protein was expressed as described in ref. 11 with an N-terminal hexahistidine tag. After sonication of bacteria prepared from 4 liters of culture (OD₆₀₀ of 0.6 at induction), the soluble fraction was purified by chromatography on a 0.5-ml Superflow NiNTA matrix (Qiagen, Valencia, CA) equilibrated with IMAC buffer (50 mM sodium phosphate/500 ml of NaCl, pH 8). After successive washes with IMAC (5 ml), IMAC/10% (vol/vol) glycerol (5 ml), IMAC/0.4 M NaCl (5 ml), and IMAC/10 mM imidazole (5 ml), the 1b11 protein was eluted with 0.2 M imidazole (2 ml). After dialysis, the 1b11 fractions were applied to a 1-ml anion exchange column (HiTrap Q Sepharose, Amersham Biosciences) in 20 mM Tris·HCl (pH 8) buffer at 4°C and eluted with a linear gradient of NaCl (10-400 mM). The protein was desalted and concentrated to 20 mg/ml by using an Ultrafree concentrator (Millipore) and corresponded to a single major band on SDS/PAGE. The molecular mass as determined by surface-enhanced laser desorption/ionization affinity MS (11,010 Da) was consistent with that calculated (10,997 Da).

Crystallization. 1b11 crystals were grown at 17°C by vapor diffusion of hanging drops of 1 μl of purified 1b11 at a concentration of 10 mg/ml mixed with 1 μl of reservoir solution [0.2 M NaCl/0.05 M Mes, pH 5.6/2.5 M (NH₄)₂SO₄] and suspended over 750 μl of reservoir solution in wells. Propagation by hair seeding was necessary to obtain crystals of reasonable quality. The crystals belong to space group I422, with cell parameters of a = b = 101.9 Å and c = 75.5 Å. There are two molecules per asymmetric unit, with a solvent content of 44% in the unit cell.

Data Collection, Structure Determination, and Refinement. The crystal structure of 1b11 was solved by the single-wavelength anomalous diffraction (SAD) method (Table 1). A gadolinium (Gd) derivative was obtained by soaking a native crystal in mother liquor together with 100 mM of the Gd complex Gd-DO3A (21) for 26 h. For cryoprotection, single crystals were transferred to a drop containing mother liquor and 15% glycerol, in which the crystal was soaked for 1 min before flash-freezing in a liquid nitrogen stream. The native data were collected on beamline ID14 [European Synchrotron Radiation Facility (ESRF)]; the SAD data were collected on beamline ID29 (ESRF) at a wavelength of 1.711 Å, corresponding to the maximum of the fluorescence near the gadolinium L_III edge. The native data were integrated with mosflm (22); the derivative data were integrated with xds (23).

Table 1. Data processing and SAD phasing statistics.

	Native	Gd derivative
Space group	I422	I422
Cell parameters	a = b = 101.9 Å	a = b = 101.9 Å
	c = 75.5 Å	c = 75.5 Å
	α = β = γ = 90°	α = β = γ = 90°
Data collection
Resolution range, Å	70.71–1.90 (2.0–1.9)	38.9–2.90 (3.1–2.9)
Observations	75,247	51,235
Unique reflections	15,820	4,591
Average I/σ (I)	19.3 (3.3)	8.4 (4.4)
Multiplicity	4.8 (3.5)	11.1 (11.2)
Completeness, %	99.0 (99.0)	99.9 (100.0)
Rsym, %	7.8	6.5
Phasing statistics
No. of sites	n.a.	3
Rano, %	n.a.	10.4 (12.6)
FOM (after mlphare)	n.a.	0.528
Anomalous Rcull	n.a.	0.51

Values in parentheses are for the highest-resolution shell. FOM, figure of merit; n.a., not applicable.

Three Gd positions were determined by using the program shake and bake (24). The resulting coordinates were directly used for phasing the SAD data with the CCP4 program mlphare (25) and gave an initial overall figure of merit (FOM) of 0.53. An established procedure that describes more precisely the heavy-atom structure and allows the estimation of phases of the centric reflections during the SAD phasing process was used. Phases were improved by histogram matching and solvent flattening with dm (26). Phases were extended to 1.9 Å by using the native-crystal data set, after which the FOM rose to 0.71.

An initial model was built by using arp/warp (27). Manual building was then carried out by using the program o (28) and alternated with several cycles of refinement by using the program refmac (29). The refinement converged to an R factor of 0.19 and R_free of 0.24 (8.3% of data). The model has excellent geometry, with 94% of residues in core regions of the Ramachandran plot and none in disallowed regions. The final model contains all atoms for residues 16-101 of 1b11, except for residue Arg-58, in which the side chain was not visible. The initial 16 residues contain the 6-His tag, which could not be traced and was not modeled. Representations of the structure were prepared with pymol (30).

Results

Quaternary Structure. 1b11 is predominantly a tetramer in solution (<10% dimer), as is evident from size-exclusion chromatography. It also crystallizes as a tetramer comprising chains referred to as A, A′, B, and B′ (Fig. 2C). These are organized as two homodimers in which the A-B dimer and A′-B′ dimer are identical and related by crystallographic symmetry. Within each dimer (the asymmetric unit) the two molecules A and B (and A′ and B′) are almost identical, are related by noncrystallographic symmetry, and exchange their C termini in a manner previously referred to as domain swapping (31) or, more appropriately in this case, segment swapping (32) (Fig. 2D). Each domain therefore has contributions from two chains. The dimer and tetramer interfaces in 1b11 bury several hydrophobic aromatic residues that are exposed in the monomeric parent proteins CspA and S1. These residues, which originally formed the nucleic acid-binding site in both parents (33, 34), now mediate oligomerization. This finding is consistent with the poor RNA binding of 1b11 (K_d > 1 μM; data not shown).

Domain Architecture. Each 1b11 domain has an overall barrel topology of six antiparallel β-strands (β1-β6). Strands β1-β4 and β5-β6 are contributed respectively by chains A and B (Fig. 2E); strands β1-β3 and β4-β6 are derived respectively from the CspA and S1 domains (Fig. 2F). The barrel is flattened, being elliptical rather than circular in cross section, with one face formed by strands β1, β2, β3, β5, and β6 and the other formed by strands β1, β4, and β6. Strands β1 and β6 are contained in both faces of the barrel, with the necessary change of direction facilitated by Gly-21 and a bulge (Lys-24) in β1 and a bulge (Leu-93) in β6. All six strands in the barrel are linked by main-chain hydrogen bonds to form a closed structure, and the barrel is capped by residue Trp-98 from strand β6.

Structure of Segments. The N-terminal segment of 1b11 (residues 15-50) corresponds to the N-terminal 36 residues of CspA. This segment adopts the same antiparallel, three-stranded conformation as in CspA, with those residues in strand β1 that are responsible for shaping the barrel serving the same purpose in both proteins. The backbone hydrogen bonds within the CspA segment itself are identical in CspA and 1b11 (Fig. 3), with the exception of a further hydrogen bond between Gly-31 and Phe-48 in 1b11. Indeed, the equivalent Cα atoms (1b11 residues 18-50 and CspA residues 3-36) can be superimposed with a rms deviation of 1.4 Å, with divergence mainly at the turns (Fig. 4A).

Fig. 3.

Pattern of main-chain hydrogen bonds of β-strands of CspA and 1b11. Interstrand backbone hydrogen bonds are shown as dashed lines, and residues are shown as circles; small circles indicate residues pointing into the barrel. Hydrogen bonding from β1-β3 of CspA (dark blue) to β4-β5 of CspA (light blue) has been recapitulated in 1b11 by hydrogen-bonding to S1-derived residues in β4 and β6 (gray). Residues in 1b11 that are colored white have no structural equivalent in CspA.

Fig. 4.

Folding and packing of segments in 1b11 and parent proteins. (A and B) Superposition of Cα backbones of the CspA segment in CspA (blue, residues 1-36) and 1b11 (gray, residues 18-50) (A) and the S1 segment in S1 (red, residues 6-40) and 1b11 (gray, residues 64-98) (B). In B, 1b11 strands β5-β6a could be superimposed to S1 strands β2-β3 (albeit poorly); 1b11 strands β4 and β6b have been spatially rearranged such that no superposition to S1 was possible. (C) CspA (blue) and 1b11 (gray) barrels, with key packing residues in the C-terminal half of the barrel core shown as spheres. (D) Cartoon representation of residues 2-141 of the Tm0160 protein from Thermotoga maritima (PDB ID code 1VJL) (www.jcsg.org). The N-terminal three-stranded sheet similar in configuration to the N-terminal segment in CspA is shaded orange. The rest of the structure (light gray) comprises another similar three-stranded sheet, as well as various helices, but does not form an OB-type barrel.

The C-terminal segment of the 1b11 barrel (residues 64-98) has been recruited from the S1 30S ribosomal protein (residues 364-398). The structure of this domain can be inferred by homology with another S1 domain from the E. coli polynucleotide phosphorylase (Fig. 2B) (34), which has the same OB (oligonucleotide/oligosaccharide-binding) fold as CspA. 1b11 broadly retains the secondary structure of the S1 segment; residues in strands β4-β6 of 1b11 correspond to those in strands β1-β3 of S1. However, the detailed β-strand assignments differ; in particular, strand β6 in 1b11 comprises all residues of the equivalent β3 strand in S1, as well as the first seven residues of the following loop. There are also major topological differences. If 1b11 β5 and β6a are superimposed on the β2-β3 hairpin of S1 (albeit poorly), the remaining corresponding residues of 1b11 and S1 do not superimpose at all because of a major spatial rearrangement (Fig. 4B).

Discussion

A Unique Architecture. OB-fold proteins are typically five-stranded, monomeric β-barrels characterized by a bulge and two small residues that allow strands to curve and shape the barrel (14). 1b11 retains these determinants; it derives two (Gly-21 and the β1 bulge) from the CspA segment and one (Gly-64) from the recruited S1 segment. However, 1b11 is not typical, because it forms a six-stranded barrel, assembles to a tetramer, and has two strands involved in segment swapping. These properties may reflect the fact that 1b11 was created in a single step of recombination, creating exposed hydrophobic surfaces that are most readily sequestered by oligomerization and domain/segment swapping (35). Oligomerization provides a powerful way of stabilizing proteins and may also provide further advantages compared with monomers, including avidity effects and clefts for binding or catalytic sites. Alternatively, over the course of natural evolution, such oligomers may resolve to monomers as mutation renders these surfaces more hydrophilic.

Complete OB folds typically comprise five strands, and there are no examples of six-stranded OB barrels. There are, however, examples of segment-swapped OB-fold proteins. The molybdate/tungstate binding protein (MOP) is a trimer of segment-swapped dimers comprising five-stranded barrels in which the C-terminal strand is swapped (36). BiMOP (37) arises from a duplication of MOP but is a monomer consisting of a tandem repeat of two segment-swapped MOP barrels. However, 1b11 resembles none of these examples and is therefore a unique variant of this fold. Recently, a protein was created with unique architecture (and topology) by de novo design (38); our work suggests that unique architectures also may be created by recombining segments of natural proteins.

Structural Templating. It is surprising that the structure of the CspA segment is highly retained and that an OB-type barrel is created around it. This finding suggests that the CspA segment may have acted as a template by directing the structure of the recruited segment to complete a barrel. Not only does the CspA segment recruit additional β-strands as necessary from the nonhomologous sequence of S1, but it rearranges the S1 segment. The rearrangement involves more than a simple strand-for-strand replacement; only one of the strands (β4 in 1b11) recruited to replace the two missing CspA strands corresponds to a strand of the S1 structure (β1 in S1); the other (β6b in 1b11) corresponds to a loop region of S1 (C-terminal to β3 in S1) that is transformed into a strand (Fig. 1B). These strands complete the barrel, and the two superfluous strands from S1 (β2 and β3 in S1) are accommodated as an extra hairpin (β5 and β6a in 1b11) at the periphery of the barrel, where they least disrupt the overall fold (Fig. 2E). The overall barrel architecture is thereby maintained; furthermore, the S1 segment recapitulates the hydrogen-bond pattern of the C-terminal strands of CspA despite different primary sequences (Fig. 3). The extra hairpin of 1b11 forms additional hydrogen bonds that close the barrel by juxta-position of antiparallel strands (rather than parallel strands as in CspA).

It appears that the CspA segment has selected a sequence capable of recreating a stable fold and positioned the residues to make contacts compatible with the fold. OB folds have a highly conserved Gly at the start of strand β4, which facilitates the turn of the protein chain across the barrel; the S1 segment supplies a Gly from strand β1 of S1 (Gly-64 in 1b11) for this purpose. Furthermore, in the hydrophobic core of the 1b11 barrel, S1-derived core residues of 1b11 (Val-67, Ile-71, and Ile-96) are recruited to occupy very similar locations to core residues in CspA (Val-51, Ile-55, and Val-67, respectively) (Fig. 4C). However, the core packing is not completely recreated; for example, Gly-69 in 1b11 replaces Phe-53 in CspA, leading to the reorientation of the Met-19 (Met-5 in CspA) side chain into the barrel interior. The CspA segment has thereby acted as a template and imposed its preferred conformation by selecting a segment able to satisfy secondary structure requirements, hydrogen-bond patterns, fold determinants, and core barrel packing.

The ability of the CspA segment to act as a template is consistent with the observation that its β2-β3 hairpin folds first and appears to serve as a nucleation point for folding of CspA (17). This early folding is likely to be mediated by hydrophobic clusters, which, in CspA, are present both at the protein surface and within the barrel. Multiple sequence alignments of OB-fold proteins show that core residues corresponding to CspA Val-9, Ile-21, Val-51, and Val-67 are highly conserved and constitute a further predicted folding nucleus (39). Segments with nucleation points may be particularly well suited to serve as templates for the creation of new domains because they can fold autonomously. Because the S1 segment also includes a structurally homologous β2-β3 hairpin turn, it may also constitute a nucleation point. Indeed, this hairpin retains much of its structure in 1b11 despite the overall rearrangement of the S1 segment to fit the CspA template, which raises the possibility that in the early stages of folding there are two main pathways: one initiated by the autonomous folding of the CspA segment (and collapse of the S1 segment to fit) and the other by the S1 segment (and collapse of the CspA segment to fit). However, the CspA-like structure of the CspA segment “dominates” over S1 in 1b11. There are presumably several reasons for this, including the intrinsic stabilities of each isolated segment, the extent to which the other is able to provide missing fold or packing determinants, and the contribution of each monomer to the stability of the oligomer. Here, the CspA fragment dictates the fold of the barrel, and the S1 segment provides missing determinants and also mediates the oligomerization of 1b11.

Although the three strands of CspA are retained in 1b11 as a sheet and reconstitute an OB-type barrel, they would presumably be capable of being incorporated as a sheet into other architectures. For example, the natural protein RNA polymerase subunit RPB8 is an eight-stranded barrel composed of a repeat of two incomplete, OB-like, four-stranded sheets (40). The Tm1060 protein from Thermotoga Maritima (Protein Data Bank ID code 1VJL) (Fig. 4D) contains two OB-like, three-stranded sheets (A. Murzin, personal communication), each sheet having a curved first strand incorporating a bulge and a small residue. However, it does not form a barrel at all; an entirely different fold is created by the helices and linking elements comprising small stretches of β-structure against which the sheets pack.

The Building Blocks of Evolution. Our results show that the fusion of nonhomologous sequence segments can lead to the generation of folded domains, which may be promoted by the ability of the segments to fold autonomously. We would expect that autonomous folding would be promoted through hydrophobic clusters and also with an increased length of the polypeptide; short peptides do not generally fold autonomously (41), and their structures are dictated by protein environment. For example, peptides of 5-11 residues can adopt alternative secondary structures in different protein contexts (42-45). By contrast, longer peptides can make sufficient internal interactions to fold autonomously; theoretical calculations suggest a critical size of ≈38 residues (46), which is significantly smaller than the size of a typical protein domain (100-250 residues) (47) but consistent with the average length of exons (35-40 residues) (48, 49).

A further aspect of the work is the ability of one segment to act as template for the folding of the rest of the polypeptide, which may have evolutionary implications. If folded domains arise more frequently from segments able to fold autonomously, then evolution should favor the enrichment of such segments and the folds built around them. Indeed, the power law distribution for folds (50) is entirely consistent with competition between templates, with more successful templates giving rise to the dominant folds with large numbers of members (51). Presumably, there is also a requirement for segments that are deformable and those that act as linkers; it is likely that many sequences will be capable of fulfilling this requirement (and these would not direct specific folds). Irrespective of the role of subdomain segments in the evolution of protein architectures, the discovery of segments with strong templating abilities may provide further insights for the artificial evolution and design of engineered proteins (52-54).