Evolution of Fold-Switching in a Metamorphic Protein

Abstract

Metamorphic proteins switch between different folds, defying the protein folding paradigm. It is unclear how fold-switching arises during evolution. Using ancestral reconstruction and NMR, we studied the evolution of the metamorphic protein XCL1, which has two distinct folds with different functions, making it an unusual member of the chemokine family that generally adopts one conserved fold. XCL1 evolved from an ancestor with the chemokine fold. Evolution of a novel dimer interface, changes in structural constraints and molecular strain, and alteration of intramolecular protein contacts drove the evolution of metamorphosis. Then, XCL1 likely evolved to preferentially populate the non-canonical fold before reaching its modern-day near-equal population of folds. These discoveries illuminate how one sequence evolves to encode multiple structures, revealing principles for protein design and engineering.

One Sentence Summary:

The human protein XCL1 evolved to switch between two completely different folds with different functions.

Metamorphic proteins defy the protein folding paradigm, in which each amino acid sequence adopts one defined fold (monomorphic). They switch reversibly between entirely distinct folds, often with different functions, on the timescale of seconds (1). Metamorphic proteins undergo large-scale structural changes (e.g. alterations in secondary structure and hydrogen bonding networks), whereas allosteric proteins exhibit smaller-scale conformational dynamics (1, 2). While only ~6 metamorphic proteins have been well characterized (2–4), estimates suggest that metamorphic proteins may comprise up to 4% of proteins in the Protein Data Bank (PDB) (5). The emergence of new protein folds on evolutionary timescales has been examined (6–9), but it remains unclear how metamorphic folding evolves in a single protein.

Among metamorphic proteins, the human chemokine XCL1 (lymphotactin) undergoes one of the most significant structural switches, involving complete rearrangement of hydrogen bonding networks (Fig.1A–B) (10). XCL1 belongs to a family of 46 human chemokines: small, secreted proteins that direct immune cell migration. While non-XCL1 chemokines adopt a monomorphic, conserved α+β fold (chemokine fold) (11), XCL1 reversibly interconverts between the chemokine fold and a dimeric, all-β fold (alternate fold) with no structural similarity to other known proteins (Fig.1A) (10). Non-XCL1 chemokines execute two essential functions using one structure, but XCL1 divides these two functions between its two folds: the chemokine structure activates XCL1’s cognate G-protein coupled receptor (GPCR), while the alternate structure binds glycosaminoglycans (GAGs) (10). Additionally, the alternate structure is directly antimicrobial, like a subset of chemokines (12). In a family of proteins that adopt a single fold, how did XCL1 evolve to become metamorphic?

Fig. 1. Evolutionary history of XCL1.

(A) Structures and secondary structure diagrams for XCL1’s chemokine fold (red) and alternate fold (gold). (B) Cartoon comparing hydrogen bonding networks in each fold’s β-strands. Black lines between dots represent pairs of hydrogen bonds between amino acids. Grey shading highlights the same set of residues. Interconversion between XCL1’s chemokine fold and alternate fold requires the β2 strand to rotate 180° and shift by one residue relative to the β1 and β3 strands, establishing an entirely new hydrogen bonding pattern. (C) Ensemble of the top 20 NMR structures for Anc.0 (PDB ID 7JH1), colored by secondary structure. Disulfide bonds are shown as dark grey sticks. (D) Simplified phylogenetic tree showing XCL1’s evolutionary history, beginning with the last shared ancestor (Anc.0) of XCL1 and another chemokine (CCL20). MYA, Million Years Ago. Nodes represent reconstructed ancestral sequences. For each ancestral sequence, % identity to extant human XCL1 is shown in parentheses.

To better understand the evolution of metamorphosis in XCL1, we inferred its phylogenetic tree using ancestral reconstruction (Fig.1D, Fig.S1). This technique utilizes multiple sequence alignments of modern-day proteins from different species to infer amino acid sequences of shared ancestors (13) and has been used to investigate various evolutionary questions (14–19). We created a multiple alignment for 457 chemokine sequences from 30 vertebrate species incorporating 14 chemokine structures (20), from which we inferred the phylogeny of XCL1 and related chemokines using maximum likelihood methods (i.e. the inferred ancestral sequences have the highest probability of producing the modern-day sequences (13, 21, 22)). Phylogenetic analysis indicates that XCL1 is most closely related to another, monomorphic chemokine, CCL20 (Fig.1D, Fig.S1) (12). We resurrected (i.e. expressed and purified) the last shared ancestor of XCL1 and CCL20 (Anc.0). Unlike the temperature- and salt-dependent structural equilibrium of XCL1 (23), the Anc.0 HSQC spectrum is unchanged from 10–50°C +/− NaCl (Fig. S2). We solved the Anc.0 NMR structure: it adopts the canonical chemokine fold (Fig.1C, Table S1). Despite sharing only 40% sequence identity with extant, human XCL1, Anc.0’s structure is highly similar to human XCL1’s chemokine fold (RMSD = 1.45Å) (Fig. 1A, C). Together, these data suggest that a metamorphic protein evolved from a monomorphic ancestor.

Non-XCL1 chemokines have two conserved disulfide bonds, one of which is incompatible with the alternate fold of XCL1. Restoring the missing disulfide bond to XCL1 makes it monomorphic, locked into the chemokine fold (24). Anc.0 has both chemokine disulfide bonds, so we sought to identify the interval where one disulfide was likely lost, because this may have imparted metamorphism. Maximum likelihood methods (13, 21, 22) identified several nodes along the evolutionary trajectory from Anc.0 to extant XCL1. By locating the oldest node whose sequence lacks the cysteines needed for one of the disulfides, we identified the interval where one disulfide was lost (Fig.1D), culminating in an ancestor called Anc.2 (Fig.1D). Anc.2 adopts a single fold at all temperature and salt conditions tested by NMR (10–50°C, +/− NaCl), indicating that disulfide loss alone did not enable metamorphism (Fig.1D, Fig.S3).

Given that loss of a conserved disulfide was necessary but insufficient to impart metamorphism to XCL1, what evolutionary changes resulted in the emergence of metamorphosis? We resurrected the remaining inferred ancestral sequences (Anc.3, Anc.4) and found by NMR that they both interconvert reversibly between two distinct folds (Fig.2A–C, Fig. S4–7) despite sharing only 60 and 68% sequence identity with human XCL1 respectively. This suggests that a wide range of sequences can encode metamorphism, because here metamorphism is encoded by sequences that differ at up to 40% of positions. ZZ-exchange experiments reveal that Anc.3 (k_ex=0.97 s⁻¹) and Anc.4 (k_ex=2.6 s⁻¹) exchange on the timescale of seconds, similar to human XCL1 (k_ex=0.90 s⁻¹ (25)), consistent with XCL1 evolving to remain metamorphic over hundreds of millions of years, balancing occupancy of its two structures by keeping exchange rates in a narrow range.

Fig. 2. Evolutionary progression of metamorphic folding in XCL1.

(A) 1D traces (from HSQC experiments performed at 37°C, 150 mM NaCl, 350 μM protein concentration) for a glycine peak (Gly44 in XCL1) that is diagnostic of XCL1’s metamorphic conformation, shown as spheres on Anc.0’s structure at left. Because Gly44 occupies distinct chemical environments in XCL1’s different native structures, there are two Gly44 peaks with chemical shifts of ~8.0 ¹H ppm (chemokine fold, red) and ~7.4 ¹H ppm (alternate fold, gold). (B) Fractional abundance of the chemokine and alternate folds were calculated as an average of relative HSQC peak volumes for two reporter residues (Gly44 and the indole NH of Trp55 in XCL1). All spectra were collected at 37°C with 150 mM NaCl and 350 μM protein concentration. (C) ZZ-exchange dynamic analysis (27) of Anc.3 at 40 °C and Anc.4 at 25 °C, both with 0 mM NaCl, using the glycine reporter. Cross peaks indicate the presence of structural interconversion. Conditions for each ancestor were chosen to maximize the presence of both folds in order to enhance detection of cross peaks. (D) Simplified phylogenetic tree indicating the back-and-forth evolutionary trajectory of XCL1 metamorphosis. Pie charts represent fractional occupancy of the chemokine fold (red) and the alternate fold (gold).

Because extant human XCL1 occupies its two folds in equal proportion, we sought to determine if Anc.3 and Anc.4 do also. HSQC peak intensities can quantify the fractional population of the different folds of XCL1, which is determined by the concentration-independent equilibrium constant for fold-switching. HSQC experiments show that under near-physiologic conditions (37°C, 150 mM NaCl) at identical protein concentrations, 92±3.7% of the Anc.3 population occupies the chemokine fold (Fig. 2A–B, D, Table S2). However, only 9.3±2.0% of the Anc.4 population occupies the chemokine fold (Fig. 2A–B, D, Table S2). To confirm the robustness of these results to statistical uncertainty in ancestrally reconstructed sequences, we expressed and purified “Alt.” ancestral proteins, where we replaced every residue with the next most likely residue if the next most likely residue was predicted with >20% probability, creating a “worst case” alternate at all positions ancestor, as is the standard in the field (26) (Fig. S8). Consistent with our results for the maximum likelihood ancestors, Alt.Anc.0 and Alt.Anc.2 are monomorphic, whereas Alt.Anc.3 and Alt.Anc.4 are metamorphic (Fig. S8), and Alt.Anc.4 populates the chemokine fold 11±1.4%. This suggests that XCL1 likely evolved from a single-fold ancestor (Anc.2), to a metamorphic ancestor that prefers the chemokine fold (Anc.3), to a metamorphic ancestor that prefers the alternate fold (Anc.4), to an extant metamorph that equally populates two different folds (Fig. 2B, D).

The aligned sequences of Anc.2 and Anc.3 differ in 26 positions distributed throughout sequence and structure (Fig. S1C). To uncover how these sequence changes introduced metamorphosis, we examined these positions based on three criteria: sequence comparison of positions at the alternate structure’s dimer interface, sequence comparison of positions likely to impact overall structural flexibility, and analysis of residue-residue contacts in the chemokine structure (Fig.3A). Another sequence-based criterion that has been used to predict protein metamorphosis is inaccurate secondary structure prediction (5, 28); however, each ancestral sequence, and WT XCL1, are predicted to adopt the secondary structure profile of the chemokine fold (Fig. S9).

Fig. 3. Metamorphic folding is enabled by the combined presence of changes in dimer interface, structural flexibility, and residue contact networks.

(A) Schematic of the hypothesis that metamorphosis evolved over time via a subset of key sequence changes from Anc.2 towards Anc.3. (B) Dimer interface positions that are polar in Anc.2 and hydrophobic in Anc.3. (C) Positions likely to be important for structural flexibility, and creation of strain in the chemokine fold. (D) Positions selected for replacement in Anc.2 using contact network analysis (Cα, red spheres). Anc.0 disulfides shown in black. (E-G) HSQC spectra for Anc.2 (E), Anc.2m (F) and Anc.2m I38T (G) (50°C, 20 mM NaPO₄ (pH 6.0)). Boxes: conformational reporter residues. (H) ZZ-exchange analysis for Anc.2m (50°C, 20 mM NaPO₄ (pH 6.0)) using reporter residue Gly41. Top: cross peaks indicate interconversion. Bottom: curves fit to peak intensities vs. time to calculate kinetic parameters. k_forward (k_for), interconversion from the alternate fold towards the chemokine fold; k_reverse (k_rev), interconversion in the reverse direction.

Several metamorphic proteins form protein-protein interfaces likely to stabilize one of their two structures, including RfaH (29), Selecase (30), and XCL1 (23). For XCL1, the alternate structure forms a stable dimer interface that is likely important for metamorphic folding, so we performed sequence comparison of Anc.2 and Anc.3 at XCL1 dimer interface positions. We found that three interface positions switch from polar or charged in Anc.2 to aliphatic side chains in Anc.3 (Y11I, T38I, and R43L) (Fig. 3B), making contacts with the rest of the apolar dimer interface more favorable. Sequence comparison also identified two changes between Anc.2 and Anc.3 that likely increase structural flexibility (P15R) or create strain (a deletion in the β1–β 2 loop at position 29) in the chemokine fold (Fig. 3C).

Networks of noncovalent residue-residue contacts are key indicators of protein conformation and stability (31, 32) that mediate essential biological phenomena (33–35). Because metamorphosis likely relies on a nuanced balance of conformational stability across multiple protein folds, we suspected that changes in noncovalent contact networks might contribute to the evolution of protein metamorphosis. We compared contact networks in human XCL1 versus Anc.0 and identified seven positions that make > 3 contacts in Anc.0 and < 3 contacts in XCL1, and whose identities differ between Anc.2 and Anc.3. We focused on these sequence substitutions because they decreased connectivity in Anc.0 (potentially permissive for interconversion) but minimized changes in connectivity in XCL1 (preserving the chemokine fold) (31) (Fig. 3D, S10), suggesting that mutations that permit rewiring of these contacts could enable metamorphic folding. These positions largely cluster around Anc.0’s second disulfide bond (Fig. 3D, S10C), “gluing” its structure in place. After the disulfide was lost, this remaining structural “glue” “dissolved” as metamorphosis evolved. Position 27, which points into the “glue” region, is a glutamine in Anc.0, Anc.2, and Anc.3. But in Anc.4, it is a bulky lysine, which is likely to sterically destabilize the chemokine fold, perhaps partly accounting for the fact that Anc.4 prefers the alternate fold. In human XCL1, this position is a threonine, perhaps facilitating XCL1’s shift to a 50/50 structural equilibrium.

To identify sequence changes sufficient to introduce metamorphic folding, we mutated positions in Anc.2 (monomorphic) to match Anc.3 (metamorphic) based on the findings described above (Table S3). We evaluated the resulting variants for fold-switching via HSQC experiments. We found that combining the three dimer interface, two flexibility-altering, and seven connectivity-altering substitutions (11/26 possible sequence changes; Anc.2m; Fig. 3A–D, Table S3) conferred the ability to populate the alternate fold (Fig. S11, Fig. 3E–H). Anc.2 variants incorporating all three dimer interface substitutions alone, or the two flexibility-altering changes alone, or the seven connectivity-altering changes alone are not metamorphic and only occupy the chemokine fold (Table S3), suggesting that these three sets of changes had to occur in combination to introduce metamorphic folding. Reverting the least disruptive mutation at the dimer interface (I38T) in Anc.2m disables metamorphosis, causing Anc.2m I38T to adopt exclusively the chemokine fold, indicating that the T38I mutation is necessary but insufficient for metamorphic folding (Fig. 3G, Table S3). This highlights how mutations that stabilize an alternate fold through formation of a protein-protein interface can critically impact metamorphic folding, as is seen in other metamorphic proteins (29). Together, these findings show that concurrent changes in interaction interfaces, structural flexibility, and noncovalent residue-residue contact networks act together to drive the evolution of protein metamorphosis.

Lattman and Rose proposed that a protein’s specific folded structure can be encoded either by amino acids distributed throughout its sequence (distributed control), or discrete sites (centralized control) (36). While both distributed (37) and centralized (38) control of folding have been described, the requirement for numerous, structurally distant mutations of different kinds to enable metamorphosis suggests that distributed control governs XCL1’s structure.

Despite growing interest in metamorphic proteins (2–5), the mechanism by which one sequence encodes two structures remains unclear. Here, ancestral reconstruction illuminates the molecular evolution of protein metamorphosis in XCL1. Metamorphic proteins have been hypothesized to represent evolutionary bridges, or “snapshots” of proteins “caught in the act” of evolving a new fold (8, 39, 40). Were this the case for XCL1, its ancestors would likely have progressed from populating only the chemokine fold to adopting exclusively the alternate fold. Rather, our data suggest that human XCL1 likely evolved from a metamorphic ancestor that may preferentially adopt the chemokine fold to one that may preferentially adopt the alternate fold, before evolving to occupy both structures in approximately equal proportion. This suggests that XCL1 is not evolving from one fold to a new one, but is evolving to remain metamorphic, indicating that metamorphosis may be a molecular phenotype that was selected for in XCL1 rather than a transient feature of an evolutionary intermediate.

Why would metamorphosis be favored? All chemokines activate GPCRs and bind GAGs, and some chemokines, including XCL1, are directly antimicrobial (12, 41–44). Non-XCL1 chemokines carry out these three functions using one fold, whereas XCL1’s chemokine fold activates its cognate GPCR and its alternate fold binds GAGs and is directly antimicrobial (10, 12, 41). As such, metamorphic folding could confer dynamic control over the fractional population, and therefore the activity, of each structure (Fig. 4A), avoiding the need to transcribe and translate multiple genes and degrade or inhibit multiple proteins in order to turn multiple functions on and off, thus allowing for fast switching between functional states mediated by distinct folds. It could also enhance a specific function (e.g. GAG binding), or permit acquisition of a new function (e.g. antimicrobial activity). Specifically, metamorphic folding may enable XCL1 to kill pathogens directly at the site of infection (alternate fold) and stimulate antigen cross-presentation to effector cells (chemokine fold), coordinating humoral and cell-mediated immune responses and allowing for spatiotemporal regulation of multiple functions (Fig. 4A). If fold-switching thus enhances the ability of other metamorphic proteins to carry out their biologic roles, then metamorphic proteins may be more common than previously thought. Why metamorphic folding evolved in XCL1 and not other chemokines, however, remains mysterious.

Fig. 4. Functional advantages and design insights provided by protein metamorphosis.

(A) Metamorphosis enables spatiotemporal control of protein function through modulation of the relative population of two different folded states in a concentration-dependent manner. For example, XCL1 can populate the antimicrobial fold close to a site of infection and the GPCR binding fold further away. (B) The emergence in nature or de novo design of fold-switching proteins may require (top) creation of a protein-protein interface that stabilizes the alternate fold, (middle) incorporation of molecular flexibility and strain, and (bottom) optimization of residue-residue contact networks to avoid thermodynamic or kinetic trapping in a single fold.

Analysis of proteins like XCL1 that reversibly switch between two well-defined folds can inform the de novo design of fold-switching proteins, which remains a challenge. Our work suggests that designed fold-switching proteins may need to incorporate protein-protein interfaces that stabilize one fold, structural constraints and strain to enable interconversion, and residue-residue contact networks that allow stable formation of both folds without trapping the protein in either fold (Fig. 4B). LOCKR, a designed protein switch system currently being developed for therapeutic use, has already been constructed by stabilizing a protein-protein interface, demonstrating the utility of this approach (45, 46). Moreover, the broad span of XCL1 sequence space compatible with metamorphic interconversion implies that design of fold-switching proteins could be within reach, especially with the help of principles uncovered in this study.