Structural Plasticity within 3-Hydroxy-3-Methylglutaryl Synthases Catalyzing the First Step of β-Branching in Polyketide Biosynthesis Underpins a Dynamic Mechanism of Substrate Accommodation

Abstract

Understanding how enzymes have been repurposed by evolution to carry out new functions is a key goal of mechanistic enzymology. In this study we aimed to identify the adaptations required to allow the 3-hydroxy-3-methylglutaryl (HMG)-CoA synthase (HMGCS) enzymes of primary isoprenoid assembly to function in specialized polyketide biosynthetic pathways, where they initiate β-branching. This role notably necessitates that the HMG synthases (HMGSs) act on substrates tethered to noncatalytic acyl carrier protein (ACP) domains instead of coenzyme A, and accommodation of substantially larger chains within the active sites. Here, we show using a combination of X-ray crystallography and small-angle X-ray scattering, that a model HMGS from the virginiamycin system exhibits markedly increased flexibility relative to its characterized HMGCS counterparts. This mobility encompasses multiple secondary structural elements that define the dimensions and chemical nature of the active site, as well the catalytic residues themselves. This result was unexpected given the well-ordered character of the HMGS within the context of an HMGS/ACP complex, but analysis by synchrotron radiation circular dichroism demonstrates that this interaction leads to increased HMGS folding. This flexible to more rigid transition is notably not accounted for by AlphaFold2, which yielded a structural model incompatible with binding of the native substrates. Taken together, these results illustrate the continued necessity of an integrative structural biology approach combining crystallographic and solution-phase data for elucidating the mechanisms underlying enzyme remodeling, information which can inform strategies to replicate such evolution effectively in the laboratory.

Figure 1

Comparison between assembly of dimethylallyl pyrophosphate, a primary precursor of isoprenoids, and the β-methylation reaction series of polyketide biosynthesis, both of which involve an HMG(C)S homologue.^, (A) The isoprenoid pathway begins with synthesis of acetoacetyl-CoA by acetoacetyl-CoA thiolase (AACT) from two equivalents of acetyl-CoA. The acetoacetyl-CoA is then condensed with a third unit of acetate by a HMG-CoA synthase (HMGCS) to yield (S)-HMG-CoA, followed by reduction to (R)-mevalonic acid by HMG-CoA reductase (HMGR). Finally, the combined action of mevalonate-5-kinase (MK), phosphomevalonate kinase (PMK), mevalonate diphosphate decarboxylase (MDD) and isopentyl pyrophosphate isomerase (IPI), results in dimethylallyl pyrophosphate. (B) β-Methylation is initiated by decarboxylation of malonate to acetate catalyzed by a condensation-incompetent stand-alone ketosynthase (KS⁰). This reaction takes place with the substrate tethered to an ACP. The ACP then becomes the donor (ACP_D) of the acetate nucleophile which is used by the HMGS to attack the β-keto group of a polyketide chain attached to an acceptor ACP (ACP_A). The resulting HMG-like intermediate is successively dehydrated and decarboxylated by two enoyl-CoA homologues to afford the β-methyl product (the blue dot indicates the origin of the methyl group). The inset shows the specific substrates recognized by the virginiamycin M, Cur and bacillaene HMGS homologues.

Figure 2

Crystal structure of the VirC^4A mutant (Cys114Ala/Gln334Ala/Arg335Ala/Arg338Ala) (PDB ID: 8S81). (A) Side view of the homodimer structure. The two VirC^4A monomers are illustrated in cartoon representation, with the two polypeptide chains colored in white (monomer A) and pale green (monomer B), respectively. For clarity, the location of the active site is indicated by a black ellipse in only one monomer. The disordered residues of the β8–α7 loops are represented by dashed red lines, and the β6–β7 loop which forms the other side of the active site is colored in purple. The residues corresponding to helices α2 (which are disordered in monomer A (dashes)) are colored in marine blue. The loop α5–β6 containing the catalytic residue is colored in yellow. (B) Zoom into the active site of monomer A showing the 2Fo–Fc map contoured at 1σ around the disordered β8–α7 loop and helix α2 regions. Residues Asp190–Val203 of the β8–α7 loop (in red) are missing from the 2Fo–Fc map contoured at 1σ. To illustrate this region, the residues directly upstream and downstream the loop are represented in red sticks. In monomer A, the disordered helix α2 (residues Asp30–Asn36) is represented with blue dashed lines, with the 2Fo–Fc map contoured at 1σ in the region of Arg27–Ala40 (residues represented as blue sticks). The catalytic residue is shown as a yellow ball. (C) Zoom into the active site showing the ordered β6–β7 loop (in purple sticks) with the 2Fo-Fc map contoured at 1σ around the residues Val140–Met165. For clarity, residues upstream and downstream of the β8–α7 loop and helix α2 are indicated in cartoon representation.

Figure 3

Active site of the VirC^4A. The active site is located at the interface between the two monomers. The mutated catalytic Cys114Ala is shown in stick form (yellow). The entrance to the active site is located near the C-terminal end of helix α2. Monomer A is shown in white surface representation, while the surface of monomer B is in green, with the regions of helix α2, and the β6–β7 and β8–α7 loops shown in blue, purple and red, respectively. The disordered β8–α7 loop (residues 190–203, red sticks) renders the active site accessible to solvent. The residues of the β6–β7 loop (Val145–Ala161) that form one side of the active site are shown in stick representation, and are well-ordered in the crystal structure. The β6–β7 loop is less conserved in term of sequence within the protein family, and therefore is likely to participate in substrate recognition. The loop is stabilized along its length via a network of hydrogen bonds (dashed lines) that includes water molecules (red spheres), and together with helix α1 and α14, contributes to structuring the active site. In addition, a cation–π interaction between Tyr153 and Arg33 helps to maintain the structure of helix α2 of monomer B, which likely contributes to the interface with both the ACP and the Ppant phosphate moiety of the acyl-ACP substrate.

Figure 4

SRCD spectra of VirC^4A in the presence of potential ACP partners. (A) Spectra of proteins VirC^4A (in blue), holo-ACP_A (in yellow) and VirC^4A/holo-ACP_A complex (in red). The theoretical summation of the VirC^4A and holo-ACP_A spectra is shown in black. The spectrum of the VirC^4A/holo-ACP_A complex does not superimpose on the theoretical spectrum, revealing a substantial increase in secondary structure upon interaction between the two partners. (B) Spectra of proteins VirC^4A (in blue), holo-ACP_5a (in yellow) and VirC^4A/holo-ACP_5a complex (in red). The theoretical summation of the VirC^4A and holo-ACP_5a spectra is shown in black. The spectrum of the combined VirC^4A and holo-ACP_5a superimposes well on the theoretical summation, revealing no substantial change in secondary structure when VirC^4A is mixed with noninteracting holo-ACP_5a.

Figure 5

Fit between the experimental SAXS data, the VirC^4A crystal structure and the VirC^4A AlphaFold2 model. The experimental SAXS curve for VirC^4A is represented in black, with the Guinier plot inset. The scattering curve calculated from the crystal structure (PDB ID: 8S81) using CRYSOL (in red) agrees well with the experimental SAXS curve (χ² = 1.88), but a discrepancy is nonetheless apparent in the range of 0.12–0.2 Å^–1. The theoretical SAXS curve (in blue) calculated using CRYSOL from an AlphaFold2 model of dimeric VirC^4A, yielded a χ² of 1.42 relative to the experimental data, and showed a major divergence from the curve calculated from the crystal structure in the range of 0.12–0.2 Å^–1.

Figure 6

Reconstruction of the missing β8–α7 loop using the Ensemble Optimisation Method (EOM 3.0),^, and comparison with the VirC^4A crystal structure and AlphaFold2-derived VirC^4A model. (A) Comparison of the theoretical scattering curve (in yellow) calculated from the EOM model that contains the β8–α7 loop and helix α2 which are missing from the VirC^4A crystal structure, with those calculated from the VirC^4A crystal structure (in red) (χ² = 1.88) and the AlphaFold2 model of VirC^4A (in blue) (χ² = 1.42). The fit between the experimental and the reconstructed models using EOM was evidently improved in the range of 0.12–0.2 Å^–1 (light gray shading), as was the overall fit (χ² = 1.11). (B) The three models of open forms of VirC^4A that contain the β8–α7 loop which were calculated using EOM 3.0 are consistent with the SAXS data. All of the models reveal minor differences in the location of the helix α2 (shades of blue/violet). By contrast, the models exhibit major differences in terms of the conformation of the β8–α7 loop (shades of orange/yellow). The three EOM models further show that the β8–α7 loop does not cap the active site, but adopts multiple open conformations. The amplitude of the loop movement varies from one monomer to the other, and is also model-dependent. Our interpretation of these results is that the β8–α7 loop is likely to adopt a continuum of open conformations in solution.

Figure 7

Overall flexibility of VirC^4A as judged using NMA coupled with the SAXS data. (A) Comparison of the SAXS experimental data (in black) with the scattering curves calculated from the SREFLEX open-form models leads to an improved fit (χ² = 1.07–1.09) in the range of 0.12–0.2 Å^–1 (light gray shading), in comparison to that calculated from the crystal structure (in red) (χ² = 1.88) and the AlphaFold2 model (in blue) (χ² = 1.42). (B) Comparison of the closed state modeled by AlphaFold2 (in pale green) with the four open state models (in white) derived from SREFLEX. The closed conformation of the β8–α7 loop observed in the AlphaFold2 model is shown in red, while the open conformations found with SREFLEX are represented in yellow. Comparison of the models also reveals that the subdomain containing helices α1 and α2 (shown in marine blue for the AlphaFold2 model and in violet for the SREFLEX models) is mobile. Overall, NMA combined with SAXS data reveals multiple flexible regions in VirC^4A, implicating them in both accommodation of the substrate in the active site and binding of the ACP partner.

Figure 8

Overall shape and volume of the VirC active site. (A) The cavity in the closed AlphaFold2 model has been delineated using CASTp 3.0. Monomer A is represented in mesh with the active site cavity shown in surface representation (in white), while monomer B is shown in cartoon representation (in pale green) along with its active site cavity (in gold). Helix α2, and the β6–β7 and β8–α7 loops are colored in marine blue, purple and red, respectively. The two residues Arg193 and Glu199 (shown as red sticks) form a lid over the active site in the closed conformation. The volume of the active site in this conformation is approximately 427.8 ų, with a potential interaction surface of 580.2 Ų. The measured 19 Å depth of the cavity would not accommodate binding of either of the two identified acyl-Ppant substrates. (B) To mimic the minimal conformational motion necessary for substrate binding, residues Arg193 and Glu199 (indicated with *) have been mutated in silico to Ala. The resulting active site is relatively linear, with an increased volume of 666.2 ų and approximately twice the surface area (1071.9 Ų). Furthermore, the cavity depth increases to 44 Å, a distance compatible with substrate binding. (C) Detailed view of the enlarged active site. Secondary structure elements surrounding the active site, as well as the β8–α7 loop and helix α2, are colored in pale green, red and marine blue, respectively, and the catalytic Cys114 in green. Cavity measurements are shown using dashed lines.