Structural and mechanistic basis for the heterotetrameric benzaldehyde synthase from petunia

Abstract

Benzaldehyde is a widespread volatile compound produced by plants. Its final biosynthetic step is catalyzed by benzaldehyde synthase (BS), a peroxisomal enzyme composed of α and β subunits, both belonging to the short-chain dehydrogenase/reductase (SDR) family. Here, we report the crystal structure of Petunia hybrida BS, which reveals an α₂β₂ heterotetrameric arrangement. Structural and biochemical analyses show that the α subunits contain the canonical catalytic site, whereas the β subunits have lost catalytic activity but are essential for heterotetramer assembly. Notably, the C terminus of the β subunit extends into the diagonally positioned α subunit, contributing to the formation of the composite benzoyl-CoA substrate-binding pocket. Site-directed mutagenesis and subunit-mixing experiments support noncooperative, additive contributions of protomers within the heterotetramer. This work establishes BS as a rare heterotetrameric plant SDR and demonstrates how subunit specialization and intersubunit arrangement enable function, providing principles for understanding and engineering multimeric enzyme complexes.

Heterotetrameric benzaldehyde synthase forms a composite α-β active site for benzaldehyde production.

INTRODUCTION

Benzaldehyde is one of the most widely distributed and ancient volatile compounds, produced not only by more than half of plant families but also by insects, non-insect arthropods, and microbes (1, 2). It mediates diverse ecological interactions across kingdoms, functioning in arthropods both as a multifunctional pheromone and a defense compound and in plants as a pollinator attractant and antifungal agent (1). Beyond ecological roles, benzaldehyde is a key determinant of fruit aroma and, because of its distinct pleasant almond-like odor, represents the second most important compound in the flavor and fragrance industry after vanillin (2). Recent work reveals that benzaldehyde also suppresses proliferation and metastasis of therapy-resistant pancreatic cancer cells, highlighting its emerging biomedical potential (3).

In plants, benzaldehyde biosynthesis begins from phenylalanine, which is first converted to trans-cinnamic acid by phenylalanine ammonia lyase (4). Subsequent reactions involve activation by cinnamate–coenzyme A (CoA) ligase, side-chain shortening by cinnamoyl-CoA hydratase/dehydrogenase and 3-ketoacyl thiolase yielding benzoyl-CoA, which is then reduced to benzaldehyde (4). The long-standing mystery of the final step was recently resolved by the discovery of benzaldehyde synthase (BS), a peroxisomal enzyme composed of α and β subunits that together catalyzes the conversion of benzoyl-CoA and NADPH [reduced form of nicotinamide adenine dinucleotide phosphate (NADP⁺)] into benzaldehyde, free CoA, and NADP⁺ (Fig. 1A) (4).

Fig. 1. Reaction, size, and structural overview of PhBS.

(A) Substrates and products of the PhBS-catalyzed reaction. (B) Size exclusion chromatography profile of PhBS, shown relative to protein standards (Bio-Rad Laboratories, catalog no. 1511901). mAU, milli-absorbance unit. (C) Structure of the PhBS heterotetramer, comprising two α subunits (blue and light blue) and two β subunits (orange and light orange), with NADP⁺ shown in yellow. (D) Conserved Asp and Arg residues forming the adenine-binding motif of the NADP⁺ pocket. (E) Catalytic tetrad (Asn-Ser-Tyr-Lys) in PhBSα′. (F) Multiple sequence alignment of α subunit orthologs of BS from P. hybrida, S. lycopersicum, A. thaliana, and P. dulcis. The full alignment is provided in fig. S3, with numbering based on PhBSα. Figure 1B was prepared using Igor Pro, version 5.0.3.0 (WaveMetrics, Lake Oswego, OR, USA).

Both α and β subunits of BS are members of the short-chain dehydrogenase/reductase (SDR) superfamily (4), which is among the oldest and most widespread enzyme families in nature, present across all domains of life (5–8). More than 47,000 SDRs have been identified to date, and they play important roles in primary and specialized metabolism, as well as in redox sensing pathways (9). Some members of this family share as low as 20% pairwise sequence identity, indicating early divergence (5, 7), and they occur as monomers, dimers, or tetramers (10). Despite their sequence diversity, all known SDRs share a common α/β folding pattern in which six to seven β sheets are sandwiched between α helices on each side (11, 12), forming a Rossmann fold that enables binding to NAD(H) or NADP(H) (12), the cofactors required for activity (10). Another hallmark of SDRs is the presence of a catalytic tetrad (Asn, Ser, Tyr, and Lys). However, some SDRs lack one or more of these residues (13, 14), suggesting potential for neofunctionalization.

Plants contain 49 SDR families (8), which are involved in the biosynthesis of important hormones such as abscisic acid (15) and indole-3-acetic acid (16), the production of alkaloids (17), steroids (18), and fatty acids (8), as well as detoxification of persistent organic pollutants (19) and disease resistance (20), among other functions (21). In the phenylpropanoid pathway, structural and biochemical characterization of cinnamoyl-CoA reductase (CCR), which reduces various hydroxycinnamoyl-CoA thioesters to their corresponding aldehydes (21, 22), showed that the reaction involves hydride transfer from NAD(P)H and formation of a tetrahedral intermediate at the carbonyl carbon of the CoA thioester substrate, with the resulting oxyanion stabilized by residues of the catalytic tetrad (23). BS is mechanistically related to CCR, as both enzymes catalyze NADPH-dependent reductions of CoA thioesters, but in contrast to CCR, BS exhibits strict substrate specificity, converting benzoyl-CoA into benzaldehyde. Unlike all previously described SDRs in plants, BS requires the assembly of two distinct, individually inactive subunits (α and β) to form an active enzyme complex. To date, the role of each BS subunit in catalysis, the precise conformation of each monomer within the complex, as well as its catalytic mechanism remain unknown.

In this study, we investigated the structural and mechanistic basis of BS from Petunia hybrida (PhBS) using a combinatory approach that included x-ray crystallography, biochemical characterization, and computational modeling. We revealed an α₂β₂ heterotetrameric architecture of the functional enzyme, uncovered the catalytic properties of the α subunit, and showed that the β subunit, although catalytically inactive, plays a critical role in stabilizing the enzyme complex and contributing to benzoyl-CoA binding. These findings elucidate a composite active site formed at the interface of two BS subunits and highlight an unexpected quaternary structural organization within the plant SDR family.

RESULTS

PhBS has a heterotetrameric α₂β₂ arrangement

Previously, it was shown that active PhBS requires both the α (PhBSα) and β (PhBSβ) subunits to convert benzoyl-CoA and NADPH into benzaldehyde, NADP⁺, and free CoA (Fig. 1A) (4). To determine the structure of PhBS, both subunits were coexpressed in Escherichia coli using a plasmid designed for dual protein expression, and the resulting PhBSα-PhBSβ complex was purified (see Materials and Methods). Analytical size exclusion chromatography showed that the purified PhBS elutes primarily at a peak corresponding to an apparent molecular weight of 86 kDa (Fig. 1B). Given that each subunit is ∼30 kDa, this elution position is intermediate between that expected for a dimer and a tetramer. Previously reported oligomeric states of SDRs vary widely, including monomers, dimers, and tetramers (22, 24, 25). Therefore, we surmised that either PhBS forms a compact tetramer with a smaller hydrodynamic radius than a typical globular protein of comparable mass (26), or PhBS exists in a rapid equilibrium between dimeric and tetrameric states, resulting in an intermediate elution profile. A minor peak of PhBS with an apparent molecular weight of 237 kDa was also observed, which is consistent with that of an octamer assembly (Fig. 1B).

To gain insights into the structure and oligomeric state of BS, we cocrystallized PhBS with its products (free CoA, benzaldehyde, and NADP⁺) and determined the structure at 2.5-Å resolution. The asymmetric unit contains eight protein chains corresponding to two essentially identical heterotetramers [as indicated by root mean square deviation (RMSD) = 0.13 Å] (fig. S1A), one of which was selected for detailed analysis. The heterotetramer consists of two PhBSα subunits and two PhBSβ subunits in a compact α₂β₂ arrangement (27–29), designated PhBSα, PhBSα′, PhBSβ, and PhBSβ′, respectively, and displays C2 symmetry across the vertical axis (Fig. 1C) (30). Notably, the C-terminal region of the β subunits extends into the potential substrate-binding site of the diagonally positioned α subunits (Fig. 1C and fig. S2). Structural alignments yielded RMSD values of 0.14 Å for α-α′, 0.13 Å for β-β′, and 0.78 to 0.83 Å for the various comparisons between α or α′ and β or β′. Each subunit is composed of β sheets sandwiched between α helices, forming the canonical Rossmann fold characteristic of SDRs (Fig. 1C and fig. S1B) (31). Clear electron density corresponding to NADP⁺ was observed exclusively in the canonical nucleotide-binding pocket of the α subunits (Fig. 1C and fig. S1C), whereas no density was detected for the other two products.

The PhBSα subunit harbors the active catalytic site

Structural analysis of the NADP⁺-binding pocket of PhBSα revealed two highly conserved amino acid residues found in SDRs, Arg⁵⁰ and Asp⁸¹, which are known to coordinate the adenine exocyclic N6 nitrogen of NADP⁺ primarily through hydrogen bonding and ionic interactions (Fig. 1D) (21, 32). In PhBSα, Asp⁸¹ is the first residue of the highly conserved Asp-X-X-Asp motif found in other SDRs (29), but the second aspartate is replaced by alanine, resulting in Asp⁸¹-X-X-Ala⁸⁴ (Fig. 1D). We identified the conserved catalytic tetrad in PhBSα (Asn¹³⁴-Ser¹⁶³-Tyr¹⁷⁸-Lys¹⁸²) (Fig. 1E), along with the canonical cofactor-binding motif Thr²⁴-Gly²⁵-X-X-X-Gly²⁹-X-Gly³¹ (fig. S3), the features that place the PhBSα subunit within the “classical” SDR subfamily (10, 14, 32, 33). In addition to PhBS, BS orthologs from Solanum lycopersicum, Arabidopsis thaliana, and Prunus dulcis have been previously heterologously produced in E. coli and were shown to have BS activity (4). A multiple sequence alignment of α subunits from these species revealed a strict conservation of the catalytic tetrad as well as Arg and Asp residues involved in NADPH binding (Fig. 1F and fig. S3). To probe the functional importance of the catalytic tetrad in PhBSα, we generated S163A and Y178F mutants and assessed their activities in the presence of the wild-type (WT) PhBSβ. BS activity was completely abolished in both cases (Table 1), confirming that the functional catalytic site resides in the α subunit and is essential for the overall BS activity. These findings further demonstrate that the α subunit of BS harbors a functional catalytic center characteristic of a canonical SDR.

Table 1. Effects of α and β subunit mutations on PhBS enzymatic activity.

Data are presented as means ± SD (n = 3 technical replicates). The specific activity of WT PhBS was set as 100% and was 27.89 ± 1.08 nanokatal (nkat)·mg protein⁻¹. ND, not detected.

α Subunit	β Subunit	Relative specific activity (%)
WT	WT	100 ± 3.87
S163A	WT	ND
Y178F	WT	ND
WT	S147A	41.33 ± 0.96
WT	Y162F	60.21 ± 0.60
WT	R221Q/K227T/D240G	ND
WT	R221Q	0.97 ± 0.01

PhBSβ subunit serves as a structural component required for BS function

Structural alignment of PhBSα and PhBSβ revealed high overall structural similarity (RMSD = 0.84 Å) (Fig. 2A) despite only ∼30% sequence identity (4). The PhBSβ harbors the first three amino acid residues of the catalytic tetrad, the Asn-Ser-Tyr, while the last, Lys¹⁸² in PhBSα, which completes the tetrad, is substituted by Leu¹⁶⁶ (Fig. 2A). As a key residue of the catalytic tetrad, this Lys residue has been shown to participate in a proton relay that activates substrates for catalysis, binds the ribityl backbone of NADPH, and modulates the pK_a (where K_a is the acid dissociation constant) of the catalytic tyrosine (21, 34, 35). This unusual substitution to Leu suggests that PhBSβ has likely lost its catalytic activity. Moreover, the highly conserved Arg⁵⁰ residue that coordinates the exocyclic N6 nitrogen of the NADP⁺ adenine moiety in PhBSα and in other SDRs like PhCCR1 (21) is also substituted with an Asn in PhBSβ (Fig. 2B), further supporting that PhBSβ is catalytically inactive. A multiple sequence alignment of four orthologous previously characterized β subunits shows strict conservation of the catalytic tetrad’s Asn residue, while the serine residue is conserved in all except AtBSβ, which carries a threonine substitution (fig. S4) (4). The catalytic tetrad’s tyrosine is replaced by asparagine in almond PdBSβ, but in the others, it remains conserved as tyrosine. Leu¹⁶⁶ in PhBSβ corresponds to alanine in tomato SlBSβ and to serine in both almond PdBSβ and Arabidopsis AtBSβ (fig. S4). All analyzed β subunits have an aspartate at the position corresponding to Asp⁸¹ in PhBSα, but only PdBSβ has a positively charged residue, Lys³⁸, which corresponds to Arg⁵⁰ in PhBSα (Fig. 2B and fig. S4).

Fig. 2. Structural comparison of PhBSα′ and PhBSβ′ and analysis of dimer interface contacts.

(A) Structural alignment between PhBSα′ and PhBSβ′, highlighting residue pairs corresponding to catalytic tetrad in PhBSα and corresponding residues in PhBSβ. (B) Adenine binding residues in PhBSα′ and corresponding residues in PhBSβ′. (C) Dimer interface residues in PhBS tetramer. (D) Dimer interface residues involved in π-cation and electrostatic interactions. (E) Potential salt-bridge interaction at dimer interface. PhBS and NADP⁺ color scheme is the same as in Fig. 1.

To assess the role of PhBSβ in the BS enzyme complex, we performed site-directed mutagenesis to examine the contribution of catalytic tetrad residues retained in PhBSβ to enzyme activity. The PhBSβ Y162F and S147A mutants were generated, and when each was analyzed in the presence of WT PhBSα, the BS activity was reduced to 41 and 60% of WT levels, respectively (Table 1). This contrasts with the complete loss of activity observed upon introduction of the equivalent mutations into PhBSα (Table 1), further supporting that PhBSα is the critical catalytic subunit of the BS enzyme complex, whereas PhBSβ likely plays a supportive role. It is possible that the Y162F and S147A mutations in PhBSβ may compromise overall protein folding, heterotetramer formation, and/or cooperativity of subunits, thereby impairing enzyme function. We therefore used far–ultraviolet (UV) circular dichroism (CD) to measure the average secondary structure of PhBSα:PhBSβ (Y162F) and PhBSα:PhBSβ (S147A) at 1:1 ratio (36–38). The spectra show two minima near 208 and 222 nm, characteristic of α helices (38), identical to the WT PhBS spectrum (fig. S5), indicating that activity loss in the mutants is not due to large changes in secondary structure. To determine whether substitutions of the catalytic residues in PhBSβ hinder its interactions with PhBSα, pull-down experiments were performed using untagged WT PhBSα and maltose-binding protein (MBP)-tagged PhBSβ variants. The Y162F and S147A mutations in PhBSβ did not alter its ability to interact with PhBSα, as similar amounts of PhBSα were recovered with mutant and WT forms of PhBSβ (fig. S6). Together, these results suggest that the reduced enzymatic activity is not due to disrupted overall folding or protein-protein interaction. Instead, PhBSβ may exert an allosteric effect on BS activity through substrate or product binding, and the introduced mutations likely interfere with this allosteric regulation.

After establishing that PhBSβ is catalytically inactive, we examined the interface between PhBSα and PhBSβ subunits to investigate the role of each subunit in heterotetramer assembly (Fig. 2C). One key group of residues includes Glu²⁵⁹ and Tyr²⁶⁰ on PhBSα′ in proximity to Arg²²¹, Lys²²⁷, and Asp²²⁸ on PhBSβ′ (Fig. 2D). The same interface residues are also observed between PhBSα and PhBSβ. Electrostatic interactions likely contribute to subunit binding, and Arg²²¹ on PhBSβ′ and Tyr²⁶⁰ on PhBSα′ probably engage in a π-cation interaction, a strong noncovalent force (39, 40). Another potentially relevant contact occurs between Arg²⁵¹ on PhBSα′ and Asp²⁴⁰ on PhBSβ′ (Fig. 2E). To probe the role of these residues at the α-β subunit interaction, we generated the R221Q/K227T/D240G triple mutant of PhBSβ. When assayed with WT PhBSα, this triple mutant completely abolished BS activity (Table 1). Despite retaining the same secondary structure as mixtures of PhBSα and WT PhBSβ (fig. S7), this triple mutant of the β subunit was unable to form a complex with PhBSα (fig. S6). Together, these results support a role for these surface residues in PhBSβ in promoting heterotetramer formation and corroborate the α₂β₂ quaternary arrangement revealed by crystallography as the active form of PhBS. Notably, the single R221Q mutation alone in PhBSβ impaired the α-β subunit complex formation (fig. S6) and reduced BS activity by 100-fold without altering the secondary structure (Table 1 and fig. S7), consistent with the notion that Arg²²¹ participates in π-cation and electrostatic interactions critical for complex stability.

The C terminus of PhBSβ contributes to active site formation

Our cocrystallization approach yielded the PhBS holoenzyme structure with NADP⁺ bound, but no occupancies were observed for free CoA or benzaldehyde. Attempts to soak the crystals with benzoyl-CoA, benzaldehyde, or NADPH also did not produce structures containing these ligands. To overcome this limitation, we used Boltz-1, a recently developed open-source AI model that predicts three-dimensional structures of biomolecular complexes with accuracy comparable to AlphaFold3 (41). The docking simulation was performed with two PhBSα chains, two PhBSβ chains, two NADPH molecules, and two benzoyl-CoA molecules as the inputs. The Boltz-1–predicted heterotetramer closely resembles our crystal structure (RMSD = 0.41 Å; fig. S8). In the Boltz-1–predicted model, NADPH binds to PhBSα subunits in a nearly identical orientation to NADP⁺ in our structure (fig. S9). In addition, the thioester moiety of benzoyl-CoA is positioned near the hydride-donating C4 carbon of the NADPH nicotinamide ring and in proximity to the catalytic residues Ser¹⁶³ and Tyr¹⁷⁸ (42), consistent with a productive substrate orientation for catalysis (Fig. 3A).

Fig. 3. Modeling of PhBS bound to substrates and the proposed catalytic mechanism.

(A) Boltz-1 model of PhBS bound to NADPH (shown in yellow) and benzoyl-CoA (shown in cyan). (B) Surface view of the Boltz-1 model highlighting the composite benzoyl-CoA binding pocket formed by residues of diagonally positioned PhBSα′ and PhBSβ subunits. (C) Surface view of the PhBS crystal structure showing the empty CoA binding groove (dotted oval), contributed by residues of diagonally positioned PhBSα′ and PhBSβ subunits, in an open conformation relative to the Boltz-1 model. (D) Proposed catalytic mechanism of BS, mediated by catalytic residues in PhBSα. Dashed lines indicate hydrogen bonds. PhBS color scheme is the same as in Fig. 1.

The CoA tail of benzoyl-CoA extends toward the C terminus of the PhBSβ subunit positioned diagonally across from the PhBSα′ subunit, with its negatively charged diphosphate group located near three positively charged arginine residues, namely, Arg²⁵⁹, Arg²⁶¹, and Arg²⁶³, on PhBSβ (Fig. 3, A and B). Therefore, we hypothesized that these arginine residues coordinate binding of the CoA diphosphate moiety, thereby promoting specific binding and proper positioning of the benzoyl-CoA substrate. Further evaluation of our solved holoenzyme crystal structure revealed a groove formed in part by the PhBSα′ active site and the C terminus of the diagonally positioned PhBSβ subunit (Fig. 3C). A side-by-side comparison of the Boltz-1 model and the crystal structure shows a pocket in PhBSα′ into which the PhBSβ C terminus could extend (Fig. 3, B and C). Our crystal structure likely represents an open conformation, whereas the Boltz-1 model captures a closed conformation upon substrate binding. Together, these data reveal the role of the β subunit C terminus as an integral component of the benzoyl-CoA binding site in the diagonally positioned α subunit. To test whether these arginine residues on PhBSβ are required for BS activity, we generated triple mutants (R259Q/R261Q/R263Q and R259M/R261M/R263M) as well as a PhBSβ mutant with a C-terminal truncation (Δ256–end). All three PhBSβ mutants were inactive when assayed in the presence of WT PhBSα (Table 2), indicating that the C terminus of PhBSβ is essential for BS activity, likely through its contribution to the benzoyl-CoA binding site formation. To evaluate the contribution of each arginine residue, we generated single PhBSβ mutants (R259M, R261M, and R263M) and examined their impact on BS activity and benzoyl-CoA binding affinities. All three mutants displayed low to residual BS activity, when assayed with WT PhBSα. R261M and R263M, which substitute the two arginine residues closest to benzoyl-CoA, had the greatest impact on activity (Table 2), as their binding affinities for benzoyl-CoA were too weak to be determined (fig. S10). In contrast, the R259M mutant retained ∼16% of WT BS activity (Table 2), as it was still capable of binding benzoyl-CoA with a reduced affinity as measured by microscale thermophoresis (MST) [dissociation constant (K_d) = 1.8 ± 0.7 mM versus 0.9 ± 0.4 mM for WT PhBSβ; fig. S10]. Although R259 is the most distal of the three arginine residues relative to the diphosphate moiety of benzoyl-CoA, it may still contribute to substrate binding and catalysis by enhancing the local positive electrostatic environment, and this residue is conserved among petunia, Arabidopsis, almond, and tomato β subunits (fig. S4). Together, our results indicate that the arginine-rich C terminus of PhBSβ is critical for enzyme activity, with each arginine residue making a substantial, but distinct, contribution.

Table 2. Role of the C terminus of PhBSβ in overall BS activity.

Data are presented as means ± SD (n = 3 technical replicates). The specific activity of WT PhBS was set as 100% and was 31.12 ± 0.44 nkat·mg protein⁻¹.

α Subunit	β Subunit	Relative specific activity (%)
WT	R259Q/R261Q/R263Q	ND
WT	R259M/R261M/R263M	ND
WT	ΔC-term*	ND
WT	R259M	15.93 ± 0.09
WT	R261M	0.93 ± 0.01
WT	R263M	3.29 ± 0.07

^*Amino acids from position 256 to the end were truncated.

Noncooperative assembly of the PhBS heterotetramer

The heterotetrameric architecture of BS prompted us to further investigate the role of PhBSβ in enzyme structure and function. We restored the canonical catalytic tetrad in PhBSβ by generating the L166K mutant (Fig. 2A). Unexpectedly, when the L166K mutant of PhBSβ was mixed with WT PhBSα at a 1:1 ratio, BS activity decreased ∼100-fold (Table 3). The reciprocal mutation in PhBSα (K182L) with WT PhBSβ abolished the activity completely, as expected (Table 3). Similarly, no activity was detected in the mutant combinations PhBSα (K182L):PhBSβ (L166K) and PhBSα (S163A):PhBSβ (L166K), both of which contain inactive α subunits and β subunits with restored canonical catalytic tetrad (Table 3). Further pull-down experiments revealed that L166K mutation in PhBSβ disrupted the assembly of the heteromeric enzyme complex (fig. S6). Overall, these results show that the presence of the canonical catalytic tetrad in the β subunit does not confer BS activity but instead impairs it by disrupting heterotetramer assembly.

Table 3. Noncooperative activity of the PhBS heterotetramer revealed by site-directed mutagenesis subunit mixing.

Data are presented as means ± SD (n = 3 technical replicates). The specific activity of WT PhBS was set as 100% and was 28.02 ± 0.45 nkat·mg protein⁻¹.

α Subunit	β Subunit	Relative specific activity (%)
WT	L166K	1.01 ± 0.01
K182L	WT	ND
K182L	L166K	ND
S163A	L166K	ND
WT	WT/L166K*	73.38 ± 1.58
WT/S163A*	WT	48.82 ± 0.75
WT/Y178F*	WT	53.83 ± 1.25

^*WT and mutant subunits were mixed at a 1:1 molar ratio.

To investigate subunit-subunit cooperativity within the complex, PhBSβ (L166K) was mixed with WT PhBSβ and WT PhBSα at a 0.5:0.5:1 ratio, and the resulting mixture exhibited ∼75% of WT activity (Table 3). In parallel, mixtures containing either PhBSα (WT):PhBSα (S163A):PhBSβ (WT) or PhBSα (WT):PhBSα (Y178F):PhBSβ (WT), both at 0.5:0.5:1 ratios, showed ∼50% activity, indicating a lack of α-α cooperativity (Table 3) (43). Together, these results suggest that BS activity scales with the number and identity of functional subunits in the α₂β₂ complex. The α subunits appear to contribute independently to catalysis, providing no evidence for α-α cooperativity. For the β subunit, the ∼75% activity observed in the WT/L166K mixing experiment is consistent with a noncooperative, additive contribution, although this interpretation should be considered in light of the impaired assembly of the L166K variant. Thus, the mixing experiments support a noncooperative, additive model for BS function while also indicating that impaired assembly of PhBSβ (L166K) contributes to the reduced activity of complexes containing this variant.

DISCUSSION

Specialized metabolism generates a vast array of metabolites that are not required for immediate survival but provide adaptive advantages in changing environments (44, 45). To meet these demands, plants have evolved highly specialized enzymes that catalyze noncanonical chemistries, often through the mechanism of gene duplication followed by neofunctionalization (46). Recent examples include neofunctionalized α-carbonic anhydrases in Lycopodium alkaloid biosynthesis and a neofunctionalized BAHD acyltransferase that enables a pericyclic step in salicylic acid formation (47, 48). In this work, we show that BS, the enzyme catalyzing the final step of plant benzaldehyde biosynthesis, assembles as a rare α₂β₂ heterotetramer composed of two α and two β subunits, representing a metabolic innovation achieved through multimeric enzyme complex formation. Several plant SDRs, including SAD-C from Pisum sativum (49, 50), TS2 from Zea mays (51), MUR1 from A. thaliana (52), 3-hydroxy-3-methylglutaryl-CoA reductase also from A. thaliana (24), secoisolariciresinol dehydrogenase from Podophyllum peltatum (53), and β-keto acyl carrier protein reductase from Brassica napus (54), form tetramers, but all are homotetrameric. Unlike these homotetrameric SDRs, BS functions as an α2β2 heterotetramer composed of distinct catalytic and noncatalytic subunits. Although its oligomeric architecture is distinct, the BS α subunit retains the canonical SDR catalytic tetrad (Asn-Ser-Tyr-Lys) and catalytic mechanism, as confirmed by site-directed mutagenesis. Structural alignment further shows that PhBSβ adopts an overall fold highly similar to that of PhBSα despite sharing only ∼30% sequence identity. However, substitutions at conserved SDR residues in PhBSβ indicate that it has lost canonical catalytic function. In addition, mutating key α-β interface residues identified from the heterotetramer structure abolished BS activity, demonstrating that PhBSβ plays a supportive, noncatalytic role in the BS heterotetramer and is required for proper complex assembly, which is essential for enzymatic function.

Since our crystallization yielded only NADP⁺-bound PhBS, we used Boltz-1 (41) to model the binding mode of benzoyl-CoA and NADPH. The Boltz-1 prediction closely matched our crystal structure, reproducing the α₂β₂ heterotetrameric quaternary structure and positioning both NADPH molecules in the same locations and conformations as the NADP⁺ observed in PhBSα. In this model, benzoyl-CoA adopts a catalytically productive conformation relative to the catalytic tetrad and NADPH in PhBSα. The model also shows that the C terminus of PhBSβ extends into the active site of PhBSα′, together forming the benzoyl-CoA binding site at the interface of diagonally positioned α and β subunits. Mutation of the C-terminal arginine residues in PhBSβ that coordinate the CoA phosphates abolished BS activity, confirming the role of the β subunit C terminus in forming the benzoyl-CoA-binding site and, in turn, underscoring the importance of the α₂β₂ heterotetrameric arrangement for catalysis. Notably, the C termini of other SDRs are often variable and structurally flexible and have been reported both to contribute to substrate recognition and to undergo conformational changes upon substrate binding (10, 55–58). This flexibility in the progenitor β subunit was likely co-opted during evolution to enable formation of the BS heterotetramer.

In an effort to understand why BS evolved a heterotetrameric architecture, we restored the missing catalytic lysine residue in PhBSβ by introducing the L166K substitution. Contrary to expectation, L166K reduced BS activity by ∼100-fold (Table 3) and also impaired α-β complex formation, indicating that reconstitution of the canonical catalytic tetrad in the β subunit is not sufficient to confer activity and instead perturbs productive assembly. Notably, however, mixing PhBSβ (L166K) with WT PhBSβ in equimolar amounts restored ∼75% of WT activity (Table 3), a value higher than expected if only fully WT α₂β₂ complexes were active. This result suggests that, unlike the interface mutant R221Q, which disrupts complex formation directly, L166K does not abolish assembly completely but more likely weakens or destabilizes the heterotetramer in a manner that can be partially compensated by the presence of WT β subunits. Analysis of intersubunit cooperativity in catalysis further showed that equimolar mixtures of WT PhBSα with either PhBSα (S163A) or PhBSα (Y178F) yielded ∼50% activity (Table 3), consistent with independent contributions of the two α active sites and linear scaling with the number of functional α protomers (43). Recent work on SDRs has shown that intersubunit interactions stabilize active sites and that generating equivalent active sites in nontetrameric enzymes requires longer polypeptide chains, suggesting that the tetrameric arrangement may reduce the biosynthetic cost of enzyme assembly and remain evolutionarily advantageous even when its active sites operate independently (59).

Together, our results demonstrate that the PhBS active site is formed by diagonally paired subunits (PhBSα-PhBSβ′ and PhBSα′-PhBSβ) (Fig. 1C). The catalytic tetrad resides on PhBSα, whereas PhBSβ contributes its C-terminal residues to form the benzoyl-CoA binding pocket, underscoring the unique heterotetrameric architecture of BS adapted to its catalytic function. On the basis of results presented here and prior literature describing the general SDR catalytic mechanism (10, 32, 56, 60), we propose the catalytic mechanism of BS (Fig. 3D). Catalysis occurs in the α subunit, which houses the canonical SDR catalytic tetrad and the NADPH-binding site. Tyr¹⁷⁸, Lys¹⁸², and Asn¹³⁴ help coordinate the ribose of the nicotinamide moiety of NADPH (32), positioning the cofactor for hydride transfer. Benzoyl-CoA is positioned by the composite pocket at the α-β interface, so that its thioester carbonyl lies adjacent to the C4 of the nicotinamide moiety of NADPH. Hydride transfer from NADPH to this carbonyl generates a tetrahedral thiohemiacetal, with the resulting oxyanion stabilized by Ser¹⁶³ and Tyr¹⁷⁸ (10, 22). The ε-amino group of Lys¹⁸² lowers the pK_a of the p-hydroxyl of Tyr¹⁷⁸, aligning with its role to stabilize the oxyanion (56). Collapse of the tetrahedral intermediate cleaves the C─S bond, releasing CoA as the leaving group, while benzaldehyde and NADP⁺ form. Product release resets the active site for subsequent turnovers.

The fact that both subunits adopt Rossmann folds and can self-oligomerize in multiple ways (4) raises the question of why neither subunit alone can catalyze benzaldehyde biosynthesis. The Boltz-1–predicted homotetramer of PhBSα shows that NADPH and benzoyl-CoA are positioned productively for catalysis and that the C-termini of the diagonal subunits are associated with the active sites. However, only 3 of the last 12 residues, Thr²⁷³, Ser²⁸¹, and Ser²⁸², are polar, whereas the remaining residues are nonpolar and none are positively charged, in contrast to the three arginines present in the PhBSβ C terminus (fig. S11, A and B). In contrast, the Boltz-1–predicted homotetramer of PhBSβ places the substrates far from the Rossmann fold motif, indicating a nonproductive assembly (fig. S11C). Because the PhBS active site resides in the α subunit, we compared PhBSα′ from the substrate-bound Boltz-1 model with NADP⁺-bound CCR1 from Sorghum bicolor [Protein Data Bank (PDB) 5TQM], which functions as a monomer (22), and found that flexible loops containing Arg²³⁸, Arg²⁹⁸, and Arg³⁰⁶ in SbCCR1 are positioned close to the CoA phosphates, whereas these residues are absent in PhBSα′ (fig. S12). Using PhBSα as a query in a PDB structural search identified enoyl–acyl carrier protein reductase (InhA) from Mycobacterium tuberculosis as the most similar reductase, with a predicted score of 0.87. InhA is an SDR involved in the biosynthesis of mycobacterial cell wall components and exists as a homotetramer in solution (61), but unlike PhBS, its active site pocket does not directly involve the C termini of diagonal subunits (fig. S13, A and B). By contrast, an α₂β₂ heterotetrameric SDR, (S,R)-2-(α-hydroxybenzyl)succinyl-CoA dehydrogenase (BbsCD) from Thauera aromatica (30), contains two active subunits and two supportive subunits, and its active site groove connects with the C termini of the inactive diagonal subunits (fig. S13, C and D), in a manner similar to that of PhBS.

Overall, this work provides the structural and mechanistic characterization of a heterotetrameric plant SDR, yet several questions remain. For instance, when α and β subunits from different species were mixed, activities varied widely (4), suggesting that either the β subunits differentially modulate α subunit catalysis or that cross-species heterotetramers assemble inefficiently. Pairing AtBSα with PdBSβ yielded the highest BS activity, nearly tripling the activity of PhBS (4). These observations, together with our structural insights, open opportunities to use computational modeling platforms such as AlphaFold3 and Boltz-1 to rationally engineer more efficient BS variants or other multimeric SDRs, enabling improved enzymatic production of benzaldehyde and other valuable compounds for industrial biosynthetic applications.

MATERIALS AND METHODS

Heterologous expression and purification of the PhBS complex

DNA fragments corresponding to the open reading frames of PhBSα (OK095279.1) and PhBSβ (OK095280.1) were both cloned into the two multiple cloning sites of the pACYCDuet-1 plasmid, which is designed for coexpression of two proteins in E. coli. In one site, we cloned PhBSα followed by a Gly-Ser (GS) motif, a tobacco etch virus (TEV) cleavage site, and a 6×His-tag. In the other site, we cloned MBP, followed by (GS)₄, a TEV cleavage site, a GS motif, and PhBSβ. This construct was used to transform E. coli BL21 (DE3)–competent cells, and colonies were selected on LB/agar plates containing chloramphenicol (35 μg/ml). A single colony was used to inoculate 10 ml of LB with chloramphenicol (35 μg/ml) and incubated overnight at 37°C with shaking at 220 rpm. The overnight culture was then used to inoculate 1 liter of Terrific Broth (TB) containing chloramphenicol (35 μg/ml). Cultures were grown at 37°C with shaking at 220 rpm to an optical density at 600 nm (OD₆₀₀) of ∼0.4 to 0.6, transferred to a precooled shaking incubator at 18°C, induced with 1 mM isopropyl-β-d-1-thiogalactopyranoside (IPTG) after 30 min, and incubated overnight.

Cells were harvested by centrifugation and resuspended in buffer A [50 mM tris (pH 8.0), 100 mM NaCl, 10% (v/v) glycerol, 5 mM β-mercaptoethanol, and 1 mM phenylmethylsulfonyl fluoride], followed by sonication. The supernatant was clarified by centrifugation, filtered through a 0.2-μm syringe filter, and stirred with ∼12 ml of preequilibrated 50% amylose resin slurry [New England Biolabs (NEB)] at 4°C for ∼40 min using a magnetic stir bar. The mixture was then loaded onto a gravity-flow column, washed with ∼10 column volumes of buffer A, and eluted with buffer A supplemented with 10 mM d-maltose. The eluent was applied to a preequilibrated QIAGEN nickel-nitrilotriacetic acid (Ni-NTA) gravity-flow column, washed with ∼10 column volumes of buffer A followed by buffer A with 50 mM imidazole, and eluted with ∼15 ml of buffer A containing 250 mM imidazole. To remove the affinity tags (MBP on PhBSβ and 6×His on PhBSα), 1 mg of His-tagged TEV protease was added, and the sample was dialyzed at 4°C for 16 hours against ∼5 liters of buffer B [50 mM tris (pH 8.0), 100 mM NaCl, and 10% (v/v) glycerol]. Following dialysis, the protein solution was passed three times each through Ni-NTA resin and amylose resin to remove uncleaved protein and His-tagged TEV. Recombinant proteins were further purified by gel filtration on an ÄKTA Pure FPLC system (GE Healthcare Life Sciences) using a HiLoad 16/600 Superdex 200 pg column. The main peak was collected, analyzed by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and concentrated to ∼10 mg/ml using Amicon Ultra-15 centrifugal filters (Millipore) with a 30-kDa cutoff. Approximately 1 ml of concentrated protein was centrifuged in a 1.5-ml Eppendorf tube for 20 min at maximum speed in a tabletop centrifuge at 4°C to pellet any aggregates. The resulting supernatant was transferred to a fresh tube, the concentration was remeasured, and the protein was aliquoted into ∼100-μl portions, flash frozen in liquid nitrogen, and stored at −80°C until further use.

Mutagenesis, heterologous expression, and purification of recombinant PhBS subunits

Site-directed mutagenesis of PhBSα and PhBSβ subunits was performed using the respective WT coding regions cloned into the pMAL-c5X expression vector (NEB), which encodes an N-terminal MBP tag. The R221Q/K227T/D240G mutation and single-amino acid substitutions, except for single-arginine mutations at the C terminus of the β subunit, were introduced by amplifying the entire plasmid with primers containing the desired mutations (table S1) and recombining the products using the ClonExpress II One Step Cloning Kit (Vazyme). For constructs that include a single arginine or three simultaneous arginine substitutions, as well as C-terminal deletion, fragments harboring the intended mutations were amplified using mutation-specific primers (table S1) and cloned into the BamHI site of pMAL-c5X using the same kit. All plasmid sequences were verified by Sanger sequencing and transformed into E. coli Rosetta2 (DE3) pLysS-competent cells (EMD Millipore). Single colonies were cultured overnight in 2 ml of LB medium containing carbenicillin (100 μg/ml) and chloramphenicol (34 μg/ml) at 37°C. Overnight cultures were diluted 100-fold into fresh LB medium with the same antibiotics and grown at 37°C until an OD₆₀₀ of 0.4 to 0.6 was reached. Cultures were cooled to room temperature, induced with 0.1 mM IPTG, and incubated overnight at 18°C with shaking. Cells were harvested by centrifugation at 4°C and lysed using BugBuster Protein Extraction Reagent (Merck KGaA). Lysates were clarified by centrifugation at 10,000g for 30 min at 4°C, and supernatants were filtered through a 0.45-μm syringe filter before purification. Recombinant proteins were purified using amylose resin (NEB) according to the manufacturer’s instructions.

X-ray crystallography

Purified PhBS complex protein at 10 mg/ml in buffer B was incubated with free CoA, benzaldehyde, and NADP⁺ (each at 300 μM) for 0.5 hours at 30°C. The enzyme solution was mixed 1:1 with precipitants from Molecular Dimensions sparse-matrix screens (Molecular Dimensions, Sheffield, UK). Sitting drop vapor diffusion trays were set up using a Formulatrix NT-8 Drop Setter (Formulatrix, Bedford, MA). After 10 days, clusters of thin, rectangular PhBS crystals appeared in the condition 0.1 M sodium chloride, 0.1 M sodium formate, 0.1 M bis-tris propane (pH 8.5), and 25% (v/v) PEG (polyethylene glycol) Smear Medium at 18°C. Individual crystals were transferred to fresh precipitant solution supplemented with 15% glycerol before cryo-freezing. Diffraction data were collected at the Cornell High Energy Synchrotron Source (CHESS), HPBio MX & FlexX beamline (Ithaca, NY). The best crystal diffracted to 2.5-Å resolution. Data were processed automatically with DIALS (62) version 3.14 and scaled using AIMLESS (63) version 0.7 from the CCP4 software suite (64) version 9.0. The crystal belonged to the P1 space group with cell dimensions a = 68 Å, b = 73.7 Å, c = 134.1 Å, α = 90.6°, β = 103.7°, and γ = 115°. Complete data collection statistics are provided in table S2. The PhBS structure was solved by molecular replacement using PHASER (65) version 2.8. Initial search models were generated with AlphaFold3 from the PhBSα and PhBSβ sequences, with terminal loops pruned before use. Two α₂β₂ heterotetramers were located in the asymmetric unit. Following rigid-body refinement and positional refinement with phenix.refine (66) from PHENIX (67) suite version 1.21, missing terminus loops were built using COOT (68) version 0.9. Simulated annealing was included in the early refinement cycles to reduce model bias introduced by AlphaFold3. After initial refinement, clear difference Fourier density was observed for the active site ligand, and NADP⁺ was modeled into this density (fig. S1C). Several rounds of refinement and model building were carried out, and water molecules were added during the final refinement stages.

Enzyme activity assay

BS activity assays were performed in 100 μl of reaction mixtures containing 50 mM bis-tris buffer (pH 6.5), 200 μM benzoyl-CoA, and 2 mM NADPH. Freshly purified MBP-tagged PhBSα and PhBSβ subunits were mixed at a 1:1 molar ratio and incubated at 4°C for 24 hours before use in enzyme assays. Reactions were initiated by adding 0.8 μg of the preincubated PhBS complex and incubated at 28°C for 10 min. Reactions were terminated by adding 100 μl of ice-cold methanol, followed by vigorous vortexing and centrifugation at 15,000g for 20 min. Supernatants (5 μl) were analyzed by high-performance liquid chromatography on an Agilent 1260 Infinity II system (Santa Clara, CA, USA) equipped with an Atlantis dC18 column (2.1 mm by 150 mm, 3 μm; Waters). The column was maintained at 30°C with a flow rate of 0.4 ml/min, and separation was achieved using a 10-min linear gradient from 10% acetonitrile in water to 100% acetonitrile. Benzaldehyde was detected at 248 nm with a diode array detector (1260 DAD HS, Agilent) and identified by comparison of retention time and UV spectrum with an authentic standard. Quantification was performed using a standard curve generated from the authentic benzaldehyde standard.

CD spectroscopy

CD spectra were measured using a J-1500 spectropolarimeter (Jasco, Tokyo, Japan) at room temperature. PhBS subunit mixtures were prepared in the same manner as for enzymatic assays and diluted to a final concentration of ∼7 μM with buffer B, and CD spectra were collected using a demountable Hellma 106-QS quartz cuvette with a 0.01-mm optical path length. CD spectra were collected from 250 to 190 nm using a 0.5-nm data pitch, a 200 mdeg/1.0 dOD (difference in optical density) CD scale, a 4-s digital integration time, a 1.00-nm bandwidth, continuous scanning mode, and a scan speed of 50 nm/min. Three scans were averaged for each sample, and the resulting spectra were baseline-corrected by subtracting the corresponding buffer blank. All CD figures were prepared using Igor Pro, version 5.0.3.0 (WaveMetrics, Lake Oswego, OR, USA).

Pull-down assay

To perform pull-down assays between subunits, the PhBSα coding sequence was subcloned into the BamHI site of pMAL-c6T expression vector (NEB), which harbors an N-terminal 6×His-MBP dual tag and a TEV cleavage site, as described above. The recombinant 6×His-MBP-TEV-PhBSα protein was produced and purified as described above and was incubated with TEV protease at a mass ratio of 5:1 (PhBSα:TEV) for 24 hours at 4°C. The mixture was buffer exchanged to pull-down buffer [20 mM tris (pH 7.4) and 150 mM NaCl] using an Econo-Pac 10DG desalting prepacked gravity flow column (Bio-Rad Laboratories) and incubated with Ni-NTA resin for 1 hour at 4°C to capture uncleaved PhBSα, His-tagged TEV, and His-tagged free MBP. WT PhBSβ and its mutants, S147A, Y162F, L166K, R221Q, R221Q/K227T/D240G, as well as free MBP, were produced in E. coli and purified as described above and were buffer exchanged to pull-down buffer using Zeba spin desalting columns (7 kDa molecular weight cut-off, Thermo Fisher Scientific). About 140 μg of each β subunit or 80 μg free MBP was mixed individually with about 60 μg of tag-free PhBSα in 600 μl of pull-down buffer and was rotated for 20 hours at 4°C in the presence of 20 μl of amylose resin slurry (NEB). Before adding the amylose resin, a 5-μl sample was taken from each mixture and mixed with 1 μl of 5× Laemmli buffer as input controls. After incubation, the supernatant was discarded, and the amylose beads were washed twice with 1 ml of pull-down buffer, transferred to fresh centrifuge tubes, and washed one additional time with 0.5 ml of pull-down buffer supplemented with 0.05% n-dodecyl-β-d-maltoside. The pull-down complexes were eluted by boiling the beads with 100 μl of 2× Laemmli buffer. Three microliters each of input or 10 μl each of pull-down samples was analyzed with SDS-PAGE and subsequent Coomassie brilliant blue staining.

Microscale thermophoresis

MST measurements were performed on a Monolith.X instrument (NanoTemper Technologies) using MO.control v2.7.1 software. Before labeling, ligand, buffer, and protein samples were screened for autofluorescence at 670 nm to verify compatibility with the RED-tris-NTA second-generation His-tag labeling dye (NanoTemper Technologies, MO L018). Equimolar mixtures of PhBSα and PhBSβ subunits (2 μM each) were prepared as described above and equilibrated in MBP column buffer [20 mM tris-HCl (pH 7.4), 200 mM NaCl, and 1 mM EDTA] supplemented with 10 mM maltose for 20 hours at 4°C. Protein complexes were labeled by mixing samples 1:1 with 100 nM RED-tris-NTA dye in phosphate-buffered saline with Tween 20 (PBS-T) [137 mM NaCl, 2.7 mM KCl, and 10 mM phosphate buffer (pH 7.4) supplemented with 0.05% (v/v) Tween 20] and incubating for 30 min at 25°C in the dark according to the manufacturer’s instructions.

Labeled protein was mixed 1:1 with a benzoyl-CoA dilution series (4 mM to 31.3 μM in 60% PBS-T) and incubated for at least 30 min at room temperature before loading into standard Monolith capillaries (NanoTemper Technologies, MO K022). MST measurements were acquired at 25°C using 100% excitation power and medium infrared laser power. Fluorescence at 670 nm was recorded throughout the thermophoresis trace, and ΔF_norm values at 2.5 s were exported for analysis in GraphPad Prism v9.3.1.

Quality control assessment of MST traces showed no evidence of hydrophobic sticking, capillary adsorption, or aggregation, and no ligand-induced changes in fluorescence or photobleaching rate were observed. K_d values were determined by fitting ΔF_norm values to a one site-specific binding model, and nonlinear least squares regression was used to estimate K_d and associated parameters.

Supplementary Materials

This PDF file includes:

Figs. S1 to S13

Tables S1 and S2

References