Trans-interface coupling in the tetrameric enzyme underlies allosteric control in Escherichia coli biosynthetic threonine deaminase Authors: Samaneh Khodi, Abeeb Abiodun Yekeen, Haiyan Liu, Quan Chen Published: 2026 ABSTRACT AbstractAllosteric regulation in biosynthetic L-threonine deaminase (BTD) has long been attributed to effector-induced conformational changes. Here, we identify a trans-interface coupling mechanism that regulates the active site architecture of EcIlvA, the BTD homolog of Escherichia coli. Starting from a low-activity, regulatory-domain-truncated variant (EcIlvA1–335, termed CDL) that retains only the catalytic domain and the native linker region, we combined computational screening with experimental selection and directed evolution to identify a variant (designated HSG) bearing three mutations in the linker region (H322L, S328L, G334L). These mutations restore enzymatic activity to approximately two-thirds of the full-length EcIlvA level. Size-exclusion chromatography experiments suggested that the truncated CDL variant is dimeric in solution, whereas the HSG mutations restore the tetrameric form of the full length native protein. High-resolution crystal structures of CDL and HSG reveal two distinct functional states. They demonstrate that substitutions at the monomer–monomer interfaces within the dimer-of-dimers tetrameric assembly trigger inter-subunit rotation, propagating conformational changes across subunits and remodeling interactions at distal dimer–dimer interfaces. This allosteric cascade stabilizes a tetrameric assembly that hosts an open, catalytically competent active site. Molecular dynamics simulations further reveal that tetramerization in this form enhances global structural rigidity while retaining essential flexibility in the substrate-binding loops. Collectively, our findings establish that trans-interface-coupling within the dimer-of-dimers architecture of the tetrameric enzyme acts as a key allosteric mechanism in EcIlvA, providing direct structural and dynamic evidence for quaternary structure–driven allostery in a classical metabolic enzyme.Keywords: Threonine deaminase, Allosteric regulation, Quaternary structure, Conformational coupling, MD simulation Introduction Allosteric effects are often essential for regulating protein functions in signaling or metabolic pathways (Konieczny et al., 2023). The allosteric regulation of an enzyme involves the binding of an effector molecule at a site other than the enzyme's active site, inducing conformational changes that lead to altered enzyme activity (McCullagh et al., 2024; Thirumalai et al., 2019). In multi-subunit allosteric enzymes, such conformational changes may involve coupled intra and inter-subunit rearrangements (Zandany et al., 2008). In fact, cross-subunit communications are implied in the two well-known general models explaining the cooperative kinetics of multi-subunit enzymes, namely, the KNF (Koshland-Némethy-Filmer) model assuming sequential structural changes of different subunits (Koshland et al., 1966) and the MWC (Monod-Wyman-Changeux) model assuming simultaneous conformation transitions of all subunits (Monod et al., 1965). Despite such general kinetic models, the actual structural re-organizations that occur during the allosteric transitions of many enzymes have not been concretely resolved, leaving the underlying biophysical mechanisms elusive. To unravel such mechanisms, it is important to capture and analyze the different conformational states of various allosteric enzymes (Motlagh et al., 2014), and to reveal the key molecular features and interactions that determine the conformational dynamics. A classical example of multi-subunit allosteric enzymes is the biosynthetic L-threonine deaminase (BTD) (Gallagher et al., 1998; Gonzales-Vigil et al., 2011; Jia et al., 2024), which catalyzes the deamination of L-threonine to produce α-ketobutyrate (AKB) and ammonia, the first reaction in the isoleucine biosynthesis pathway (Amorim Franco and Blanchard, 2017). Taken EcIlvA, the BTD of Escherichia coli, as an example, has been well-known that its allosteric inhibition by isoleucine and its activation by valine play an important role in the feedback regulation of the biosynthesis of amino acid residues with branched side chains. The underlying structural basis and mechanisms of these inhibition effects have been of great interest since the early studies of allosteric proteins (Eisenstein, 1995; Isogai et al., 2022; Shulman et al., 2008). However, not all the conformational states involved in the allosteric regulation process have been clearly determined, and the allosteric mechanisms of EcIlvA and related BTDs remain elusive in terms of atomic-level structures and conformational dynamics. Previous studies about the allosteric behaviors of EcIlvA and related BTDs have provided substantial insights into enzyme kinetics (Chen et al., 2013; Isogai et al., 2022). It was found that effectors mainly modulate the apparent substrate affinity of the enzyme without significantly altering the maximum reaction rate (Chipman and Shaanan, 2001). Additionally, the cooperative association of ligands to both the active and the effector sites could be fitted to an expanded two-state concerted allosteric MWC model (Chipman and Shaanan, 2001). Moreover, the dependencies of the enzyme activity on effector concentrations and structure-based ligand docking suggested the presence of two isoleucine binding sites of different affinity and one valine binding site. It was suggested that the two isoleucine sites were sequentially occupied with increasing isoleucine concentration, while the occupation of the valine site could prevent isoleucine binding (Chen et al., 2013). To date, several crystal structures of BTDs have been reported. Two structures of the full-length EcIlvA (PDB IDs: 1TDJ and 9D2Q), each in complex with the cofactor pyridoxal-5′-phosphate (PLP), reveal a homotetrameric assembly organized as a “dimer-of-dimers” (Fig. 1) (Gallagher et al., 1998; Yun et al., 2025), with each protomer comprising an N-terminal catalytic domain and a C-terminal regulatory domain connected by a helical linker (α13). The monomer–monomer interface within each dimer is extensive, involving contributions from both the catalytic and regulatory domains as well as the linker α13. In contrast, the dimer–dimer interface within the tetramer is more localized, engaging only a region of the catalytic domain from each subunit. Yun et al. also reported dimeric crystal structures of both wild-type and Phe352Ala mutant forms of the EcIlvA regulatory domain, each bound to the allosteric inhibitor isoleucine. Comparison of these inhibitor-bound regulatory domain structures with the full-length, uninhibited structure suggests that isoleucine binding initiates conformational changes at Phe352, which is located directly within the inhibitor-binding site. These changes are propagated to distal surface loops of the regulatory domain, which may further extend across the protomer–protomer interface to affect active-site loops and the α13 linker helix of the neighboring protomer. However, because the available inhibited structures lack both the catalytic domain and the α13 linker as key elements required to fully delineate the signal transmission pathway, the precise cross-subunit allosteric mechanism remains unclear. The crystal structure (PDB ID: 3IAU) of a regulatory domain–truncated variant of TD2 (pTD2), the BTD from Solanum lycopersicum (tomato), has been reported (Gonzales-Vigil et al., 2011). This structure comprises the PLP-bound catalytic domain and the α13 linker, which shares 46% sequence identity with the corresponding regions of EcIlvA. The backbone atoms of this truncated catalytic domain structure superimpose onto those of EcIlvA (PDB ID: 1TDJ) with a Cα root-mean-square deviation (RMSD) of 0.78 Å (282 matched atoms). More recently, the crystal structure of full-length IlvA1 from Pseudomonas aeruginosa, which shares 53% sequence identity with EcIlvA, was reported (PDB ID: 8Y1J) (Jia et al., 2024). This structure captures the catalytic domain bound to the reaction product α-ketobutyrate (AKB) and exhibits a Cα RMSD of 0.91 Å (397 matched atoms) relative to the monomer of EcIlvA (1TDJ). Tertiary and Quaternary structures of EcIlvA. (A) Domain organization of full-length EcIlvA showing the sequence ranges of catalytic domain, regulatory domain, and the intervening linker helix. (B) Overall tetrameric structure of full-length EcIlvA (PDB ID: 1TDJ), with an enlarged view highlighting one subunit. (C) Structure of the catalytic domains and linker helices retained in the CDL variant, with an enlarged view of an individual linker helix, indicating the three mutated residues from the CDL to the HSG variant. The above structures provide starting points for understanding the allosteric mechanism of EcIlvA and related BTDs. Firstly, the catalytic domain of these proteins belongs to the type II fold of the β family PLP-dependent enzymes (Yu et al., 2013), which comprises two distinct subdomains (referred to as N1 and N2, respectively) with the PLP tightly embedded between the two subdomains (Gallagher et al., 1998). Studies on other β-family PLP-dependent enzymes suggested the presence of a movable motif of the catalytic domain, whose displacement may facilitate the conformational transition between an uninhibited or active open state (R-state) and an inhibited or inactive closed state (T-state) (Giménez-Mascarell et al., 2018). Secondly, the regulatory domains do not directly contact the catalytic domains of either the same or any different subunits in full-length structures reported so far. This feature of the tertiary and quaternary structure implies that the linker region and the inter-subunit interfaces play crucial roles in allosteric regulation (Gallagher et al., 1998). In fact, the linker is spatially close to both the proposed effector binding site in the regulatory domain and the active site in the catalytic domain of the same monomer, and constitutes the core monomer–monomer interface in each dimeric module of the overall tetramer, with two linkers from two different monomers forming a coiled coil structure (Gallagher et al., 1998; Yu et al., 2013). The inherent flexibility of the full-length protein has hindered the determination of intact effector-bound structures, leaving the mechanism by which the regulatory domains efficiently transmit signals to the catalytic sites unsolved (Jia et al., 2024; Yun et al., 2025). To establish a concrete mechanism of the allosteric transitions in EcIlvA and related BTDs, it is highly desirable to determine high-resolution structures of potential transient intermediate states (Hu et al., 2023). Aiming at this, here we first examined the catalytic activity of several differentially truncated variants of EcIlvA. We found that while removing both the regulatory domain and the linker from the enzyme abolished catalysis, removing only the regulatory domain but retaining the linker (the resulting variant named as CDL, for catalytic domain and linker) caused low but detectable catalytic activity, which was further supported by cell-growth assays. The markedly reduced activity suggested that, in the absence of the regulatory domain, the catalytic domain and linker module predominantly occupies poorly active conformational states. Therefore, we decided to look for mutants of the CDL that could compensate for this effect by combining computational mutation screening and experimental selection. We obtained a CDL variant harboring three-point mutations in the linker region that restored high enzymatic activity. This variant is designated HSG to indicate that the mutated residues correspond to a histidine, serine, and glycine in the wild-type enzyme. We determined the structures of both CDL and HSG using X-ray crystallography. Based on structural analyses and associated enzymatic activity data, we propose a structure-based allosteric model for EcIlvA in which monomer–monomer packing at a dimer interface is coupled with the formation of the stable symmetric dimer–dimer interfaces of a tetramer organized as a dimer-of-dimers. Notably, the dimer–dimer interfaces are formed not between the linker regions but between the catalytic domains. The concerted formation of two distant dimer–dimer interfaces in the tetramer contributes to maintaining the active site in a catalytically competent conformation. To further investigate the role of linker–linker interactions in promoting the dimer-to-tetramer transition and how tetramerization influences conformational dynamics to enhance catalysis, we performed a series of molecular dynamics (MD) simulations on the dimeric and tetrameric forms of both CDL and HSG in solution. Our study reveals key structural determinants underlying the allosteric regulation of EcIlvA and uncovers how these elements enable cross-scale dynamic coupling between the protein's oligomeric state and local conformation of the active site. Results Removing the regulatory domain from EcIlvA leads to the loss or reduction of catalytic activity To examine the role of the linker and regulatory domain in EcIlvA catalytic activity, we generated two truncated variants: the isolated N-terminal-catalytic-domain-only variant (CD, residues 1–320) and the catalytic-domain-with-linker variant (CDL, residues 1–335). We then assessed their enzymatic activity both in vivo and in vitro, as well as their oligomerization state. The results are compared with those of the full-length EcIlvA. In in vitro enzyme activity assays, the full-length protein exhibited a high specific activity of 10,608.17 U/μmol monomer, CDL showed a significantly reduced activity of 80.93 U/μmol monomer, and CD showed barely detectable activity (1.70 U/μmol monomer). Moreover, CDL showed substantial activity decreases over storage time at 4 °C: from 80.93 U/μmol monomer to 60.42 U/μmol monomer after 24 h and further to 27.47 U/μmol monomer after 72 h (Figs. 2A, B). These results were consistent with the in vivo complementation assays, in which the ilvA gene and its variants were supplied on arabinose-inducible pBAD plasmids and expressed in the ilvA-knockout E. coli K-12 strain JW3745–2. When grown on threonine-containing mineral salt medium (MSM) agar with varied inducer concentrations, cells expressing full-length EcIlvA thrived under both high (0.2%) and low (0.1%) inducer concentrations (Unless noted otherwise, concentrations are given as mass/vol%, i.e., grams of solute per 100 mL of final solution), cells expressing CDL grew only with 0.2% arabinose, and cells expressing CD showed no growth at either 0.2% or 0.1% inducer concentration (Fig. S1). This suggests that the CDL variant exhibits lower catalytic activity compared to the full-length EcIlvA enzyme. Enzymatic activity and size exclusion chromatography (SEC) analysis of EcIlvA and variants. (A) Specific activities of freshly purified proteins. Error bars indicate the standard deviation (SD) from three independent experiments. (B) Time-dependent changes in enzymatic activity of EcIlvA, CDL and HSG during storage at 4 °C. (C) Size exclusion chromatography (SEC) profiles of the purified proteins. Protein concentrations are 20 mg/mL (EcIlvA), 25 mg/mL (CD), 27 mg/mL (CDL), and 20 mg/mL (HSG). Size-exclusion chromatography (SEC) (Fig. 2C) revealed that EcIlvA predominantly exists as a tetramer (224.7 kDa), whereas CDL elutes as a mixture of dimer (71.47 kDa, major peak) and tetramer (minor peak). In contrast, CD migrates solely as a monomer (33.97 kDa). These results indicate that the linker region—present in CDL but absent in CD—contributes to the formation of higher-order assemblies and is critical for maintaining partial tetramerization of the protein. This observation correlates with the residual enzymatic activity retained by CDL but not by CD: assuming that the catalytically active conformation of the catalytic domain is stabilized by tetramer formation, the lost or reduced enzymatic activity of CD or CDL compared to full-length EcIlvA can be attributed to their lost or diminished propensity to maintain productive oligomeric states. The HSG mutant restores activity and forms tetramer as the full-length EcIlvA We used the structure-based sequence design program ABACUS (Xiong et al., 2017) to scan for single mutations in the catalytic domain and linker that potentially stabilize the full-length tetramer structure (PDB ID: 1TDJ). We identified 38 single mutants (Table S1), each mutant was experimentally constructed and assessed with the cell growth assay. Seventeen out of the 38 mutants survived in the presence of 0.2% arabinose induction, while only the S328L and the G334L mutants grew on 0.1% arabinose (Fig. S2 A). Furthermore, in the cell growth assay performed under very low arabinose induction (0.02%), the S328L-G334L double mutant (referred to as CDL-SG) exhibited improved growth compared with CDL, though still poorer than the full-length EcIlvA (Fig. S2B). The specific activity of purified CDL-SG was 1186.73 U/μmol monomer, which decreased to 737.5 U/μmol monomer after 24 h of storage at 4 °C and 368.3 U/μmol monomer after 72 h (Figs. 2A, S2C). To improve from CDL-SG, we conducted three rounds of directed evolution via error-prone PCR (ePCR) to introduce random mutations into CDL-SG. The libraries were first screened on MSM agar plates containing 0.002% arabinose, to ensure that only mutants with significantly increased activity could thrive. Surviving colonies were further analyzed by spot titer testing on MSM agar plates with an arabinose concentration of 0.02%, identifying the most active mutants by examining their growth patterns across different dilution levels. Among the variants, only one mutant from the third round of evolution showed growth similar to that of EcIlvA in spot titer assays (Fig. S2D). This mutant, named HSG, comprises three mutations, H322L, S328L, and G334L, all located within the linker (residues 321–335) (Fig. 1C). We note that another mutation of His322, H322E, has been predicted by ABACUS (Table S1); however, this single mutant supports growth only at 0.2% arabinose and fails at 0.1% (Fig. S2A), and therefore does not improve cell growth relative to CDL (Fig. S2A). Purified HSG exhibited a specific activity of 6810.50 U/μmol monomer, reaching approximately 64.2% of full-length EcIlvA (10,608.17 U/μmol monomer), and retained consistent enzymatic activity after 72 h of storage at 4 °C (Figs. 2A, B). SEC analysis further suggests that the HSG variant predominantly adopts a tetrameric state (Fig. 2C), mirroring the oligomeric state of full-length EcIlvA. Notably, the tetrameric architecture and high catalytic activity of HSG resemble those reported for pTD2, that still retain high enzyme activity (Gonzales-Vigil et al., 2011). This implies a conserved functional link between a properly formed tetrameric state and the enzymatic activity in the regulatory domain-depleted state of EcIlvA. Trans-interface coupling in EcIlvA revealed by structural analysis We solved the structures of CDL and HSG by X-ray crystallography (see statistics of data collection and refinement in Table S2). In both structures, the PLP cofactor is covalently bound to the active-site lysine residue. In the refined structures the adduct is represented as the non-standard amino acid residue LLP (N6-(pyridoxal-5′-phosphate)-l-lysine) in the PDB coordinate files (Figs. S3A, B). In the CDL structure, the asymmetric unit contains two subunits that form an approximately C2-symmetric homodimer with an inter-subunit interface (monomer–monomer interface) along a non-crystallographic two-fold axis (the r axis, Fig. S3A). Structural alignment of the monomers revealed a Cα RMSD of 0.40 Å (270 matched atoms), with deviations primarily localized in two regions: (i) N2 subdomain (β3–β5, α5–α6) spanning from Gly71 to Thr152, and (ii) the linker (α13) (Fig. S4A). In chain 2, residues 234–242 (corresponding to the Ω-loop in the active site, see below) lacked discernible electron density, suggesting high conformational flexibility of this region (Fig. S4A). The asymmetric unit of HSG structure corresponds to a tetrameric assembly similar to the full-length wild type protein (Fig. S3B). The organization of the four subunits can be viewed as a dimer-of-dimers, which has an approximate D2 symmetry with three mutually perpendicular non-crystallographic two-fold axes (labeled as p, q, and r, respectively). In this arrangement, each monomer interacts with two neighboring monomers through two distinctive interfaces: the larger interface, referred to as the monomer–monomer interface as it corresponds to the same interface in the CDL dimer structure, is formed around the non-crystallographic r axis; the smaller interface, referred to as the dimer–dimer interface as it is absent from the dimer configuration but is present only in the dimer-of-dimers configuration, is formed around the non-crystallographic p axis and distant from the monomer–monomer interface. Notably, the linker region, which contains all the three mutations of HSG relative to CDL, is involved only in the monomer–monomer interface. Thus, these mutations cannot directly stabilize the dimer–dimer interface to promote tetramer formation. Additionally, the four monomers within a single asymmetric unit of the HSG structure exhibit greater structural uniformity compared to the two monomers in the CDL asymmetric unit, as evidenced by significantly lower cross-subunit Cα RMSD values (0.11–0.16 Å for HSG with 240–287 matched atoms, versus 0.40 Å for CDL with 270 matched atoms) (Fig. S4B). We applied PISA assembly analysis (Krissinel and Henrick, 2007) to the crystal structures (Table S3). For HSG, PISA supported tetrameric assembly in one asymmetric unit (A₄/ABCD) as the most stable biological unit. In this dimer-of-dimers assembly, the two parts of the dimer–dimer interface (namely, the AD interface and the BC interface) within the tetramer are highly similar to each other. At each part, the Q175 residues from opposite dimers (related by the non-crystallographic rotations around the p axis) form proper hydrogen bonds (Fig. 3A), as previously described for EcIlvA (Gallagher et al., 1998). For CDL, a tetramer-assembly comprising two asymmetric units related by a crystallographic symmetry operation (A₄/A₂B₂) was generated besides the single-asymmetric-unit dimer (A₂/AB). However, the two parts of the dimer–dimer interface (namely the AA interface and the BB interface) are qualitatively different: the two symmetry-related Q175 residues form strong hydrogen bonds only at the AA interface, while the two Q175 residues at the BB interface are not mutually hydrogen bonded (Fig. 3B). Structural analysis of CDL and HSG assemblies reveals distinct oligomeric architectures and interface features. (A) Center: the HSG dimer-of-dimers assembly in one asymmetric-unit. Three non-crystallographic axes are indicated. Left and right: close-up panels showing the dimer–dimer interface. The Q175-Q175 pairs of both the A-D and B—C protomer pairs form hydrogen bonds. (B) Center: the CDL dimer-of-dimers assembly comprising two asymmetric units. Three axes, two crystallographic (p and q) and one non-crystallographic (r), are indicated. Left: the Q175–Q175 pair of the A-A protomer pair form hydrogen bonds. Right: the Q175–Q175 pair of the B—B protomer pair do not form hydrogen bonds. Hydrogen bond donor-acceptor distances (Å) are indicated. Q175 is shown in stick and each chain in a separate color; CDL subunits generated from a symmetry operation are shown in grey and labeled “Symmetry A" or “Symmetry B". (C) Relative rotations between the catalytic domains or linkers of the two monomers around the non-crystallographic r axis in CDL and HSG. Rotation axes were derived with the draw_rotation_axis.py script and visualized in PyMOL. To establish which assembly predominates in solution, SEC was performed on CDL and HSG at varied protein concentrations. At a low concentration of 5 mg/mL, CDL eluted as a single major peak consistent with a dimeric form (71.47 kDa) (Fig. S5A). As the concentration was increased to 27 and 50 mg/mL, the dimer remained dominant, with a minor peak consistent with a tetramer assembly (142.94 kDa), indicating concentration-dependent tetramer formation, together with a broad peak attributable to higher-order aggregations (Fig. S5B). In contrast, HSG was eluted predominantly as a form consistent with a tetramer (143.16 kDa) at all examined concentrations (Figs. S5A, B). We acknowledge that while SEC analysis reveals distinct oligomeric states across variants, inferences of oligomeric state distributions based solely on SEC data, particularly obtained here with the Superdex 75 column, is inherently limited. First, the column's resolution is suboptimal across the broad molecular weight range encompassing all relevant species. Second, deviations between observed elution volumes and those predicted from globular molecular weight standards (proteins used as standards are given in Methods) likely arise from differences in hydrodynamic properties (e.g., non-spherical shape and variable compactness) among species. Nevertheless, consistent, variant-specific elution profiles are clearly resolved, and the relative ordering of elution peaks aligns well with predictions from structural models, in strong support of the inferred oligomeric distributions. To identify the key structural determinants underlying the transition between dimeric and tetrameric states, we compared the structures of various oligomeric forms of EcIlvA, CDL, and HSG. The dimeric CDL structure aligns with the corresponding dimer within the full-length EcIlvA tetramer (PDB ID: 1TDJ) with a Cα RMSD of 2.26 Å (610 matched atoms), substantially higher than the monomer-to-monomer alignments between individual CDL subunits and their counterparts in 1TDJ, which are 0.49 Å (240 matched atoms) and 0.52 Å (237 matched atoms), respectively (Fig. S6A). This discrepancy suggests significant conformational differences at the dimer level despite high monomer similarity. In contrast to the dimeric CDL, the tetrameric HSG structure closely resembles the full-length EcIlvA tetramer (1TDJ), with an overall Cα RMSD of only 0.48 Å (1140 matched atoms). When individual HSG subunits are aligned separately to the monomers in 1TDJ, the Cα RMSDs range from 0.41 to 0.48 Å (271–274 matched atoms), nearly identical to the global alignment—indicating high structural consistency across all subunits (Fig. S6B) (The monomer-versus-monomer Cα RMSD between HSG and 9D2Q, the more recent higher-resolution structure of full-length EcIlvA, is 0.26–0.36 Å for 236–273 matched atoms). Additionally, when the dimeric CDL is superimposed onto a corresponding dimer within the HSG tetramer, the dimer-level Cα RMSD is 1.18 Å (563 matched atoms), again markedly higher than the subunit-level RMSDs of 0.36 Å (239 matched atoms) and 0.37 Å (256 matched atoms) for different subunits, respectively (Fig. S6C). The consistently higher dimer-to-dimer RMSDs compared to monomer-to-monomer values—when aligning dimeric CDL with dimers in either the EcIlvA or HSG tetramers—suggest substantial inter-subunit rearrangements within the dimer. To characterize the nature of these structural rearrangements, we analyzed the inter-subunit rotational angles around the r-axis (non-crystallographic axis for CDL and HSG, crystallographic axis for 1TDJ) at the monomer–monomer interface, using the draw_rotation_axis.py script (Schrodinger, 2015). First, we assessed the rotation of the linker region. In the EcIlvA structure, this angle is nearly ideal at 180°, resulting in a planar, in-plane arrangement of the corresponding vectors relative to the axis of rotation (Fig. S7A). In contrast, the CDL structure exhibits a reduced angle of 175°, introducing a significant out-of-plane kink in the rotated vectors. In the HSG structure, the linker rotational angles are 177.9° (A–B dimer) and 179.2° (C—D dimer), respectively (Fig. 3C), indicating that the near-planar configuration is largely preserved. A similar trend is observed for the relative rotations of the catalytic domains: their rotational angles around the r-axis deviate from planarity in CDL but remain close to 180° in both EcIlvA and HSG (Figs. 3C, S7B). These rotational differences directly influence the spatial positioning of key residues at the two cross-subunit parts of the dimer–dimer interface in the tetrameric state. In EcIlvA and HSG, the (near-) ideal planar rotation in each dimer causes the two cross-subunit interface structures to be (almost) identical. At both interfaces, the above mentioned pair of Q175 residues related by the p-axis (crystallographic in 1TDJ and non-crystallographic in HSG) are positioned to form strong inter-subunit hydrogen bonds simultaneously (Fig. 3A). In CDL, the distortion from planarity of the inter-monomer rotation within each dimer is propagated toward the two cross-subunit parts of the dimer–dimer interface (derivable from crystal packing but likely unrealized in solution). This results in divergence of the two cross-subunit interfaces. Only at one interface, the sidechains of the two Q175 residues are positioned to form strong hydrogen bonds. At the other interface, these hydrogen-bonds are disrupted (Fig. 3B). This likely causes diminished tetramerization in solution. To confirm the role of Q175 in stabilizing the dimer–dimer interfaces, we constructed the HSG-Q175A (HSG-A) and HSG-Q175S (HSG—S) mutants, and assessed their effects on cell growth, protein oligomerization and enzyme activity. Both mutants failed to support cell growth on MSM media containing 0.1% arabinose (Fig. S8A). SEC analysis revealed distinct elution profiles consistent with different oligomeric assembly: HSG-A exhibited one peak suggesting aggregated proteins possibly due to partial unfolding or misfolding, and another peak comparable to the dimer peak of CDL (Fig. S8B), whereas HSG-S showed a predominant peak consistent with a dimeric form (Fig. S8C). Specific activity measurements showed that HSG-S and HSG-A retained only ∼10.0% and ∼ 3.2% of original HSG activity, respectively (680.17 U/μmol monomer for HSG-S and 217.67 U/μmol monomer for HSG-A, versus 6810.5 U/μmol monomer for HSG; Fig. S8D). These data reveal that residue Q175, which evidently stabilizes the dimer–dimer interface based on the structure, is of key importance for catalytic competence despite being distant from the active site. Interestingly, despite their severely impaired tetramerization as evidenced by SEC, both HSG-S and HSG-A exhibited significantly higher specific activity than CDL (80.93 U/μmol monomer), suggesting residual catalytic capacity. This activity may have stemmed from functional tetramers present in solution but undetectable by SEC because of reduced stability (i.e., the mutations can cause the tetrameric form to be only transiently stable). That said, we acknowledge that we cannot definitively exclude the possibility that the dimeric species retain some low-level catalytic activity. Nevertheless, we expect any such dimeric activity to be significantly lower than that of the active tetrameric form. To gain further insights into why removing the regulatory domain and subsequent mutations of HSG change the organization at the monomer–monomer interface, we conducted a structural analysis of the monomer–monomer interfaces via Biovia Discovery Studio Visualizer (Biovia, 2025). In EcIlvA, hydrogen bonds and non-bonded interactions, notably involving His322, Ser328, Arg325, Phe321, Val327, and, Cys331, stabilize the interface, whereas CDL lacks detectable hydrogen bonds at the interface and exhibits minimal hydrophobic contacts. In HSG, His322Leu and Ser328Leu mutations reduced polar interactions but introduced compensatory non-polar interaction involving Leu328, Phe321, Leu324 and Arg325 (Table S4, S5) (Fig. S9 A-C). Among the key residues analyzed, His322 and Phe321/Phe278 are of relatively large, rigid sidechains, and could potentially facilitate through-space conformational coupling. His322 indeed plays a pivotal role in the linker-linker interface interactions in EcIlvA (Table S4, S5), and its mutation to Leu restored tetramerization in HSG (Fig. 2C). Phe321 and Phe278 are spatially proximate to His322 (Figs. S9A, B) and are conserved across species (Wang et al., 2018). We constructed the HSG-F321A and HSG-F278A mutants, and found them to grow only under maximal induction, while the HSG-FF double mutant (F321A-F278A) failed to grow at 0.2% arabinose (Fig. S8E). SEC analysis revealed that HSG-FF produced two peaks (Fig. S8F), corresponding to a dimeric form and a partially unfolded or misfolded state. The activity measured for the dimer fraction was 75.74 U/μmol monomer, corresponding to ∼1% of the activity of HSG (Fig. S8D). The strong effects of these mutations to HSG on tetramerization and enzyme activity provide additional evidence to support that disturbances at the monomer–monomer interface can be propagated to the distant dimer–dimer interface as well as the active site. The catalytic domains in the crystal structures of CDL and HSG are in distinctive states To resolve structural variations within subunit, we performed structural superpositions of EcIlvA, CDL, and HSG, focusing on one subunit (subunit A) from a single asymmetric unit in each structure. The CDL-EcIlvA superposition revealed pronounced divergences in the N2 subdomain (β3–β5, α5–α6), α3a, α10, the Ω-loop (residues 234 to 244), and the linker (Fig. 4A). In contrast, HSG-EcIlvA superposition showed only minor deviations at a few flexible loops (indicated by arrows) and linker, with preserved α3a and α10 conformations (Fig. 4B). The CDL-HSG comparisons mirror the differences between the CDL and EcIlvA structure, including differences at the N2 subdomain, the Ω-loop, and the linker, revealing maximal displacements in α6 of the N2 subdomain and the Ω-loop (Fig. 4C). Subunit-level structural comparisons between EcIlvA, CDL and HSG. (A) Structural superposition of subunit A from CDL and EcIlvA. A rotated view highlights the alignment of the N2 subdomain (comprising β-strands 3–5 and helices 5–6), which exhibits a notable overall movement. (B) Superposition of subunit A from HSG and EcIlvA. Arrows indicate regions with minor structural deviations. (C) Overlay of subunit A from CDL and HSG. Rotated views emphasize movements of the N2 subdomain and the Ω-loop. (D) Top views of the active site cavity mapped onto the protein surface. The flexible N2 subdomain and Ω-loop are shown as green and pink cartoons, respectively. In EcIlvA and HSG, the substrate-binding cavity (yellow) remains solvent-exposed, and the pyridoxal phosphate (PLP) cofactor (red sticks) is buried beneath the Ω-loop. In contrast, CDL adopts a closed conformation: the N2 subdomain and Ω-loop move toward each other, sealing the cleft leading to the active site. This rearrangement exposes the PLP to solvent. (E) Active site cavities depicted as red envelopes in subunit A of each structure. Cavity volumes are similar in EcIlvA (854 Å3) and HSG (840 Å3), but significantly reduced in CDL (599 Å3). Previously reported active site residues are shown in sticks. Close-up views highlight altered positions Ser86 and His90 in CDL (orange) relative to EcIlvA (blue), compared to the conserved geometry observed in HSG (green). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) To investigate the possible implications on enzyme activity of the within-subunit structural variations, we carried out comparative analyses of exposed surfaces, which revealed a dynamic cleft between the N1 and N2 subdomains and connected to the PLP-binding cavity. The geometry of this cleft is modulated by the movements of the N2 subdomain (specifically α6) and the Ω-loop (Fig. 4D). Given the absence of substrate in the EcIlvA-PLP structure, we attempted to infer the substrate-binding pocket using available complex structures from the database. In the complex structure of AKB and BTD from P. aeruginosa PAO1 (PDB ID: 8Y1J), Ser75 and His79, which align structurally with Ser86 and His90 in EcIlvA, are likely residues bordering the binding site cavity (Jia et al., 2024). Additionally, Yu and colleagues' structural alignment of EcIlvA with a specific C-terminal domain (PDB ID: 2GN2) from Salmonella typhimurium (encoded by tdcB), identified Pro156, Phe157, Val162, and Gln166 as critical, conserved substrate-binding residues (Yu et al., 2013). In the tetrameric structures of both EcIlvA and HSG, the above active site residues—along with Ser86 and His90—bordered an open tunnel for substrate binding (highlighted in yellow in Fig. 4D) within the dynamic cleft mentioned above. In this configuration, the Ω-loop is positioned near the PLP cofactor. In contrast, the dimeric CDL structure lacks the characteristic of an open cleft between the N1 and N2 subdomains. Instead, the movements of the N2-subdomain and the Ω-loop relative to the rest of the catalytic domain leads to the closing of the cleft, occluding the substrate-binding site and exposing the PLP pocket (Fig. 4D). To quantify the variations in structural features of interest, we used PyMOL plugin PyVOL (Smith et al., 2019) to estimate cavity sizes and compared the results for different structures (Fig. 4E). The active site cavity is of large volumes in EcIlvA (∼850 Å3) and HSG (∼840–1000 Å3), while in the CDL structure the cavity is of much-reduced volumes (∼600 Å3) (Fig. 4E), in consistence with the closure of the cleft for substrate-binding pocket. Visually, the CDL structure also undergoes substantial rearrangements of key substrate-binding residues Ser86 and His90 compared to EcIlvA and HSG (Fig. 4E). Taken together, the catalytic domain in the tetrameric HSG structure harbors an open, accessible substrate binding cavity and a well-packed PLP binding pocket, likely representing a catalytically active state. In contrast, the crystal structure of CDL lacks an open cavity and well-organized Ω-loop, likely representing a catalytically inhibited state. Moreover, key interactions at one half of the dimer–dimer interface are disrupted in the CDL crystal structure. To examine if the inhibited state in the allosteric regulation of EcIlvA is also associated with weakened interactions for tetramerization, we performed SEC on the EcIlvA protein after overnight dialysis with isoleucine. The retention time was consistent with the size of the dimeric state (Fig. S8G). The post-dialysis protein also showed a loss of activity in converting threonine to AKB (142.77 U/μmol monomer), to approximately 1.3% of untreated EcIlvA (Fig. S8D). This suggests that allosteric inhibition of EcIlvA by isoleucine is associated with destabilization of the tetrameric form of the protein. Molecular dynamics simulations reveal coupling between tetramerization and conformational stability To explore protein dynamics across different oligomeric and conformational states, we performed MD simulations of dimeric CDL (DCDL), tetrameric CDL (TCDL), dimeric HSG (DHSG), and tetrameric HSG (THSG) in explicit water. The initial structures for DCDL and TCDL were derived from the X-ray crystal structures, with the disordered loop in chain 2 modeled based on its conformation in chain 1 of the asymmetric homodimer. For DHSG and THSG, starting structures were taken from the tetrameric HSG crystal structure, using chains 1 and 2 only for the DHSG simulation. Each system was simulated in triplicate, with each independent run extending to 200 ns. Full-length EcIlvA was not simulated, as the mechanism examined here operates through the linker and the catalytic domain, both of which are fully retained in CDL and HSG. Moreover, as discussed above, the internal structures of these domains and their tertiary assemblies are highly conserved between HSG and full-length EcIlvA. We analyzed Cα RMSD relative to the starting structures by aligning individual monomers (monomer-level superposition, 323 matched Cα atom pairs per chain) and entire dimers (dimer-level superposition, 646 matched Cα atom pairs per dimer). RMSD trajectories converged across all three replicates within each system, indicating equilibration over the simulation time (Fig. S10; Table S6). At the monomer level, TCDL and THSG exhibited lower average RMSD values (ranging from 1.62 to 1.88 Å for TCDL and 1.91 to 2.10 Å for THSG) than the dimer simulations, with THSG exhibiting the smallest fluctuations and inter-replica variations among the four systems (Fig. S10C). In contrast, one monomer in the DHSG dimer showed higher average RMSD (2.89 Å) and greater variability (DHSG chain 1 in Fig. S10B), while the other monomer (DHSG chain 2 in Fig. S10B) and both monomers in DCDL (Fig. S10A) yielded average RMSDs of 2.33 to 2.60 Å. At the dimer level, THSG demonstrated average RMSD values of 2.86–2.87 Å (Fig. S10C), higher than TCDL (2.33–2.35 Å; Fig. S10D) but lower than those observed for DHSG (2.98 Å) and DCDL (3.24 Å) (Figs. S10A, B). Again, THSG exhibited the smallest fluctuations and inter-replica variations throughout the simulations, whereas TCDL showed substantially larger replica-to-replica variability despite its lower mean value. Thus, in the MD simulations, the HSG tetrameric state showed the highest structural stability for both intra-subunit conformation and inter-subunit structural organization. To assess structural compactness, we calculated the radius of gyration (Rg), which reflects the spatial distribution of atoms relative to the protein's center of mass (Khan et al., 2021). At the monomer level, Rg profiles were similar across all systems, fluctuating stably around a mean of about 19.0 Å (Fig. S11 and Table S6). The generally small variations of Rg indicate that overall structural integrity was well maintained in all simulations, in agreement with the RMSD analysis. Slight increases in Rg variability were observed in some chains of DCDL and DHSG (Figs. S11A, B) relative to THSG (Fig. S11C), consistent with reduced conformational stability in the dimeric context. The TCDL simulations showed comparable variability of the monomer Rg as the simulations of the dimeric form (Fig. S11D). At the dimer level, a clearer distinction emerged: THSG and DHSG dimers exhibited stable Rg values averaging around 26.7 Å, whereas DCDL and TCDL dimers displayed consistently higher Rg, averaging around 27.3 Å to 27.6 Å (Table S6), despite comparable fluctuation amplitudes (Fig. S11). These differences align with the distinct inter-subunit packing observed in the crystal structures of CDL versus HSG, underscoring how sequence variation influences quaternary organization. Beyond global RMSD and Rg, we examined changes in intra-subunit Cα–Cα distances during simulations relative to initial structures. To facilitate our discussions, we first compared the starting conformations of the different simulations (i.e., CDL for DCDL and TCDL simulations versus HSG for DHSG and THSG simulations) (Fig. 5). The ΔDist map (difference in intra-subunit Cα–Cα distances between CDL and HSG, scale range − 1.3 to +1.3 Å) (Fig. 5A) revealed two prominent structural differences: (i) reduced distance between the Ω-loop and the rest of the subunit in HSG, indicative of tighter active site loop packing in HSG; and (ii) increased separation of part of the N2-subdomain (more specifically, the L3, L4, and L5 loops and α6, or flexible regions of the N2 subdomain) from the rest of the protein, corresponding to an opening of the substrate-binding cavity. Structural superposition in Fig. 5A confirms these features. The ΔDist maps representing changes in intra-subunit Cα–Cα distances from initial to average structures (averaged over three replicates and monomer-aligned models) further reveal system-specific behaviors (Figs. 5B-D). We note that the color scales differ across panels (CDL vs. DCDL: −0.60 to +1.5 Å; HSG vs. DHSG: −1.3 to +5.1 Å; HSG vs. THSG: −0.59 to +2.5 Å). This variation reflects the distinct dynamic ranges of each comparison, therefore, quantitative assessments between panels refer to the corresponding scale bars rather than relying solely on color intensity. The DCDL simulation recapitulated patterns seen in the CDL-to-HSG comparison, with ΔDist values reaching up to +1.5 Å, contraction of the Ω-loop toward the core and opening up of part of the N2 subdomain (Fig. 5B). This suggests that in solution, the CDL structure undergoes partial reorganization resembling the HSG-like state, likely due to destabilization upon release from the crystalline environment. The ΔDist map from the CDL initial structure to those sampled in the TCDL simulations (Fig. S12A, color scale: −4.7 to +2.2 Å) showed similar HSG-like reorganization as seen in the DCDL simulations, but of considerably greater magnitude. Superposition of CDL with the averaged monomer structure in TCDL indicates relatively large local displacements of the linker, the Ω-loop and the peripheral α6, α3a and PPFD elements (Fig. S12B). Comparative analysis of intra-chain Cα–Cα distance matrices and structural superpositions. Heatmaps display differences in (averaged) intra-chain Cα–Cα distance matrices (ΔDist) relative to a reference structure, accompanied by corresponding structural superpositions. For each panel, the color scale was independently normalized to the ΔDist range of that specific comparison to enhance contrast; the numerical scale range is indicated for each panel. Color coding indicates distance deviations: blue (< 0 Å, shortened distances), white (≈ 0 Å, minimal change), and red (> 0 Å, elongated distances). Average structures were derived from equilibrated molecular dynamics simulation trajectories. The N1 and N2 subdomains are labeled along the top of each heatmap, with key structure elements indicated on the right. In the superposed structures, individual chains are colored distinctly; regions exhibiting large conformational changes are annotated. (A) From the CDL crystal structure to the HSG crystal structure. (B) From the CDL crystal structure to the averaged structure of DCDL simulations. (C) From the HSG crystal structure to the averaged structure of DHSG simulations. (D) From the HSG crystal structure to the averaged structure of THSG simulations. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) In contrast to the ΔDist values from CDL to DCDL, most ΔDist values from HSG to DHSG were positive, reaching up to +5.1 Å, indicating overall shortening of intra-subunit distances and increased structural compaction during simulation (Fig. 5C). Superimposed structures in Fig. 5C show this compaction is localized rather than global, consistent with stable Rg. Notably, the Ω-loop moved away from the L11, α11 and linker regions, reflecting instability in the active site conformation. In THSG, however, ΔDist values from HSG were considerably smaller, remaining almost half the magnitude observed in DHSG within a range of −0.59 to +2.5 Å (Fig. 5D), and no significant structural deviations from starting structure were observed in the averaged structures (Fig. 5D), indicating high conformational stability. Thus, the HSG state was better maintained in both the DHSG and THSG simulations than the CDL state in the DCDL and TCDL simulations. Moreover, comparative analysis between THSG and DHSG demonstrates that the intra-subunit conformation of HSG is significantly better preserved in the tetrameric state than in the dimeric state, supporting coupling between tetramer formation and local structural stability. To evaluate dynamic changes in the substrate-binding cavity—located at the interface between the N1 and N2 subdomains—we analyzed Cα–Cα distances between contacting residues across these domains (defined as pairs with Cα–Cα < 8.0 Å). As contact maps were highly consistent across chains within each system, so values averaged over different chains were used for analysis. Fig. S13 shows contact maps for segments with detectable inter-subdomain contacts: the flexible Ω-loop (residues 234–242) and three mobile loops in the N2 subdomain (L3, L4, L5). The CDL structure exhibited 19 cross-subdomain contacts, whereas HSG had only 2 (Fig. S13A), consistent with a more closed cavity in CDL and a more open cleft in HSG. Strikingly, all simulations showed increased contact density compared to their respective starting structures—most prominently in DCDL, where every residue in the Ω-loop (234–242) formed new interactions with the N2 subdomain, resulting in 44 contacts (Fig. S13B). In TCDL, cross-subdomain contacts likewise increased, but far more modestly than in DCDL, yielding 26 contacts (Fig. S13B). In contrast, DHSG and THSG, although showing increased contacts versus the initial structure HSG, maintained sparser contact patterns (Fig. S13B), preserving a more open cavity configuration similar to HSG. To investigate variations in inter-subunit structural organization during the simulations, we monitored the time evolution of rotation angles of both the linker and the catalytic domain around the previously defined non-crystallographic r axis (Fig. 6). These angles were computed from 309 matched Cα–Cα pairs for the catalytic domains (residues 12–320) and from 14 matched Cα–Cα pairs for the linkers (residues 321–334). The DCDL system exhibited the largest deviations from planarity and the highest degree of angular fluctuations (Fig. 6A). In contrast, THSG displayed the smallest deviations from planarity and minimal variability (Fig. 6C), indicating greater rigidity in its quaternary arrangement. DHSG showed intermediate behavior: the rotational dynamics of its catalytic domain exhibited decreased deviations from planarity and decreased fluctuations compared to DCDL, whereas its linker rotation fluctuated less than in DCDL but remained persistently offset from 180°, adopting a stable yet distorted (non-planar) orientation (Fig. 6B). In the TCDL simulations, the inter-domain rotational angles exhibit variations intermediate between those in the DCDL and the THSG simulations (Fig. 6D). Thus, assembly into a tetramer to some extent damped the large deviations from planarity observed in the dimeric DCDL. Nevertheless, the relative rotations between the monomers, in particular between the linkers, are considerably more mobile in TCDL than in THSG. Time trajectories of the rotational angles between the linkers or the catalytic domains around the r axis at the monomer–monomer interface. Per simulated system, the trajectories from three independent simulations are shown in different colors. (A) The DCDL trajectories. (B) The DHSG trajectories. (C) The THSG trajectories. (D) The TCDL trajectories. Collectively, the MD simulations highlight three key insights. First, at the intra-subunit level, the CDL conformation exhibits lower structural stability than the HSG conformation under simulated solution conditions; moreover, its conformational dynamics during MD simulations partially recapitulate a transition toward an HSG-like state. Second, tetramerization stabilizes the intra-subunit HSG conformation but not the CDL conformation, consistent with the SEC data suggesting that HSG exists predominantly as a tetramer in solution, while CDL adopts a dimeric state. Third, the HSG-specific mutations located in the linker regions restrict the rotation between the linkers from two neighboring monomers to a planar configuration. In the HSG tetramer, this geometric constraint stabilizes the inter-subunit interactions simultaneously at two distal halves of the dimer–dimer interface, resulting in trans-interface coupling. Interactions at distal interfaces modulates the active site conformational dynamics in MD simulations To compare the conformational dynamics of the different simulated systems, we calculated the root-mean-square fluctuations (RMSFs) of Cα atoms, with each chain superimposed independently to compute the fluctuations. For each of the systems DCDL, DHSG and THSG, the RMSF profiles were highly consistent across the three replicate simulations (Figs. S14A–C) (TCDL is not further analyzed because SEC indicates it to be not a dominant state in solution. DHSG, although also not a dominant state in solution, is analyzed to provide controls for DCDL to highlight the mutational effects, and for THSG to highlight the effects of tetramerization). The averaged RMSF profiles from the replicates are presented and compared across systems in Fig. 7A, with regions exhibiting notable differences highlighted on the structural model shown in Fig. 7B. While the overall RMSF profiles are similar among the three systems, distinct differences are evident. Comparisons of structural fluctuations in MD simulations of different systems. (A) Root-mean-square fluctuation (RMSF) profiles of Cα atoms. Residue-wise RMSF values were computed after superposition of individual chains and averaged over three independent MD replicates for each system. Various structure components are labeled and shaded using consistent colors across panels. (B) Structure of the HSG tetramer and a representative monomer, illustrating the spatial organization of key structural components. Labeled regions correspond to those annotated in panel (A), with identical coloring to facilitate direct comparison between dynamic behavior and structural context. When comparing DCDL against DHSG, DCDL shows increased fluctuations in residues 110–120 within the N2 subdomain and around residue 240 (the Ω-loop). These elevated fluctuations likely reflect the transition of the initially closed CDL structure toward a more open, HSG-like conformation during the simulation (see also Fig. 5B). Comparison between DHSG and THSG reveals that several regions (L1, α8, and α11), which are at the dimer–dimer interface in the tetrameric form (THSG) but solvent-exposed in the dimeric DHSG, exhibit lower RMSF in THSG. Interestingly, at several regions forming the active site—including the α3–α6 segment, the PPFD loop, the GGGG loop, and the Ω-loop—THSG displays higher RMSF than DHSG. Thus, despite smaller overall deviations from the initial structure and reduced quaternary structural variability in THSG as shown above, the local flexibility of active-site elements is increased in the tetramer. Notably, this increase in local dynamics is consistent with the differential changes in intra-subunit Cα–Cα distances observed in the simulations: DHSG exhibits increased local compactness (Fig. 5C), whereas THSG largely maintains the initial distances (Fig. 5D). Dynamic coupling at allosteric sites or subunit interfaces is a key mechanism underlying efficient signal transmission in protein complexes (Resende-Lara et al., 2020). To probe the collective motions in the simulated systems, dynamic cross-correlation matrices (DCCMs) were computed (Yu and Dalby, 2020) based on Cα atom fluctuations from the last 50 ns of MD trajectories, after structural superposition as dimers. The resulting correlations are visualized as heatmaps in Fig. S15. Since intra-chain correlation patterns were highly similar across chains within each system (Figs. S15A–D), the diagonal submatrices covering chain 1 were extracted for detailed examination (Figs. 8A–C). Dynamic Cross-Correlation Matrices (DCCMs) of Cα-atoms for Chain 1 in different simulations. A-C, DCCMs were calculated at the dimer level using the last 50 ns of MD trajectories, focusing on Cα atoms. The submatrix corresponding to Chain 1 was extracted and displayed to highlight intra-chain dynamic correlations. Color scale: blue (< 0, anti-correlated motions), white (= 0, uncorrelated), red (> 0, correlated motions); intensity reflects the strength of correlation or anti-correlation. Sequence regions of the N2 and N1 subdomains are labeled along the top axis. Correlated blocks identified in the DCCM maps are annotated along the right margin. (A) The DCDL simulations. (B) The DHSG simulations. (C) The THSG simulations. (D) Averaged tetramer structure from the THSG simulations, with a representative monomer (chain 1) highlighted and shown in an enlarged view. Segments corresponding to different dynamically correlated blocks identified in panel (C) are depicted with different colors. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Visual inspection of the DCCM maps in Figs. 8A to 8C suggests three distinct segments along the polypeptide chain, each corresponding to a diagonal block in the correlation matrix: segment D1, spanning from the N-terminus to α8, including the GGGG loop; Segment D2, encompassing the Ω-loop and its flanking helices; Segment D3, composed of helices α11–α13 that form the monomer–monomer interface. In the THSG system, D1 can be further subdivided into D1_1 (N-terminus to α6) and D1_2 (α7, α8, and their adjacent PPFD and GGGG loops), reflecting more structured internal dynamics within this region. These segments are mapped onto the three-dimensional structure in Fig. 8D. Notably, the DCDL simulations show anti-correlated motions between D2 (Ω-loop) and D3 (monomer–monomer interface). In contrast, both DHSG and THSG systems exhibit enhanced positive correlations between these two regions (indicated by dashed black boxes in Figs. 8A-C). Additionally, in the DCDL model, D3 displays strong positive correlations with helices α4 and α5 in D1—correlations that are markedly reduced in the DHSG and THSG models (pink boxes in Figs. 8A-C). These correlations reflect a possible signal transmission mechanism from the monomer–monomer interface toward the active site. A possible signal transmission mechanism from the dimer–dimer interface toward the active site is reflected by the increased modularity within D1 in THSG, allowing its subdivision into D1_1 and D1_2. This change from DHSG to THSG arises from weakened positive correlations of the PPFD loop and α7 with residues in D1_1, which is likely due to α7 being positioned at the core of the dimer–dimer interface in the tetrameric THSG assembly. Given that both the PPFD loop (upstream of α7) and the GGGG loop (upstream of α8) make significant contributions to the active site architecture, the observed differences in DCCM patterns between DHSG and THSG provide a mechanistic explanation for how tetramerization—by restricting conformational flexibility at the dimer–dimer interface—can propagate dynamic changes to the distal active site. Discussion Although the allosteric regulation of EcIlvA has been extensively characterized biochemically and kinetically over several decades, the exact underlying mechanism has not been fully elucidated because the conformational changes of the catalytic domain in the allosteric process have not yet been resolved at atomic resolution. Here, we combine biochemical assays, computational and experimental mutation screening, and structural biology to address this problem. We reveal that cross-subunit interactions of the tetrameric enzyme contribute substantially to catalytic competence by stabilizing an open, active conformation of the catalytic domain, thereby enabling efficient substrate binding and turnover. First, we found that removal of the regulatory domain (resulting in CDL) substantially alters oligomeric equilibrium in solution, changing the original tetrameric state to a predominantly dimeric state, which features a closed active site. As a result, CDL exhibits severely reduced catalytic activity (0.76% of full-length EcIlvA). We then engineered HSG, a CDL variant with three mutations in the linker region, which restores the tetrameric form in solution and regained 64.2% of wild-type-level activity. We further obtained high-resolution crystal structures capturing two distinct functional states: one (CDL) that is likely inactive (or at least associated with very low activity) and the other (HSG) associated with high activity. These structures provide snapshots of key states in the allosteric transition. Although the dimer-of-dimers tetrameric assembly could be recovered from the crystal structures for both CDL (by using two symmetry-related units in the crystal structure) and HSG (single asymmetric unit in the crystal structure), there are subtle but important differences: in HSG, the two cross-subunit halves of the dimer–dimer interface maintain similar inter-subunit hydrogen bonds, while in CDL the hydrogen bonds between residue Q175 of the two contacting monomers are only maintained at one half of the dimer–dimer interface but are disrupted at the other half. This observation is consistent with the observation that in solution, the dominant aggregation form of CDL is a dimer, with higher-order forms being detected as minor species only at higher concentrations. In contrast, HSG exists as tetramer in solution across all tested concentrations. Our further analyses of the structures suggest a trans-interface coupling mechanism in a dimer-of-dimers tetrameric architecture: in CDL, the two subunits contained in one dimer are rotated 3.8° out of plane around the non-crystallographic local two-fold axis, preventing the simultaneous formation of two stable interfaces that together comprise the full dimer–dimer interface in the tetramer. In contrast, HSG and full-length EcIlvA adopt nearly planar arrangements around the same axis, allowing simultaneous formation of key interactions, especially hydrogen bonds between two Q175 residues of opposite dimers, at the two distal cross-subunit interfaces for dimer–dimer formation (Gallagher et al., 1998). The linker region mutations in HSG relative to CDL are localized at the monomer–monomer interface, far away from the dimer–dimer interface. However, these mutations enhance hydrophobic packing at the monomer–monomer interface and pre-organize the dimers so that residues at the dimer–dimer interface, including the pairs of Q175, are positioned for the productive assembly of the high activity tetrameric form. That a rotation of only ∼4° is sufficient to dismantle the cross-subunit hydrogen bonds suggests that EcIlvA operates close to an energetic threshold at which small perturbations can drive switch-like changes in both quaternary interactions and catalytic activity (Pillai et al., 2024; Selwood and Jaffe, 2012). MD simulations of the dimeric and tetrameric forms of both CDL and HSG corroborate the structural model and add a dynamic dimension to it. They confirm that the assembly into the HSG-type tetramer enhances conformational stability. THSG maintains a rigid quaternary structure and preserves the open substrate-binding cleft, whereas both dimeric forms drift toward greater overall flexibility and active-site collapse. Moreover, THSG exhibits higher global rigidity coexisting with enhanced local dynamics in functional loops that line the active site (e.g., Ω-loop, PPFD, GGGG). This combination of stable overall scaffold and locally mobile structural elements is consistent with the locally frustrated energy landscapes characteristic of allosteric enzymes, in which a stable fold permits the localized motions required for signal propagation (Ferreiro et al., 2011; Frauenfelder et al., 1991; Freiberger et al., 2019). It is interesting to compare the allosteric mechanism revealed here with that suggested in the work of Yun et al. (Yun et al., 2025). There they discussed a possible allosteric pathway involving direct interactions of the regulatory domain of one protomer with the active site loops of the adjacent protomer through the monomer–monomer interface. They also pointed out that conformational changes of the regulatory domain can cause position shifts of the linker helix of the adjacent protomer, suggesting possible roles of the relative positioning of the linker helix in allosteric control. Their latter observation is consistent with our results that the CDL-to-HSG sequence changes, which are localized to the linker region, eventually lead to change of catalytic activity. Beyond the previous findings, we discovered that direct interactions between the regulatory domain and the active-site loops of the adjacent protomer are not absolutely required. Instead, trans-interface coupling, characterized as a relative rotation between the two monomers within the dimer causing a disruption of the hydrogen bonds at the distal dimer–dimer interface, is sufficient to trigger the activity switch. BTDs from different organisms share a highly conserved structural architecture, including the characteristic PLP-dependent catalytic fold at the active site (Eliot and Kirsch, 2004; Gallagher et al., 1998). However, their allosteric mechanisms may differ substantially. For instance, although the regulatory domains of E. coli IlvA (EcIlvA) and Staphylococcus aureus IlvA (SaIlvA) are structurally similar, SaIlvA neither binds nor responds to isoleucine (Yun et al., 2025). We note that the evidence supporting the proposed trans-interface coupling mechanism described in this study is currently limited to EcIlvA. Although it remains unclear to what extent this mechanism applies to other BTDs, we anticipate that it may also operate in other BTDs that feature analogous cross-subunit interfaces and a similar interface-to-catalytic-site communication pathway as EcIlvA. Methods pBAD expression constructs The full-length ilvA gene was amplified from genomic DNA of E. coli DH5α (Trans5α, Cat. No. CD201) and cloned into the arabinose-inducible pBAD/Myc-His-A vector (Invitrogen) using XhoI and HindIII restriction sites, generating pBAD-ilvA. The sequences encoding the catalytic domain alone (CD; residues 1–320) and the catalytic domain together with its C-terminal linker region (CDL; residues 1–335) were subcloned into the same vector to generate pBAD-cd and pBAD-cdl. A stop codon was introduced at the 3′ end of each insert to prevent expression of the C-terminal Myc-His epitope encoded by the vector. All constructs were verified by gene sequencing. In vivo activity assays All pBAD constructs were transformed into E. coli JW3745–2 (CGSC #10733), an ilvA-knockout K-12 strain serving as an isoleucine auxotroph that enables functional complementation assays based on threonine-dependent growth. Mineral salt medium (MSM) containing 0.1 g/L NaCl, 0.2 g/L KCl, 0.5 g/L (NH₄)₂SO₄, 50 mg/L CaCl₂·2H₂O, 0.2 g/L MgSO₄·7H₂O was prepared (Karami et al., 2014). The medium was supplemented with 0.2% (w/v) glucose, 0.4 mM L-threonine, 50 μg/mL kanamycin, 50 μg/mL ampicillin, and arabinose at concentrations ranging from 0.002% to 0.2% (w/v) as inducer. For functional screening, cells carrying the pBAD constructs were inoculated onto these MSM agar plates and incubated at 37 °C to evaluate threonine deaminase activity. JW3745–2 cells carrying the empty pBAD vector served as the negative control, while cells carrying pBAD-ilvA served as the positive control. For spot titer assays, mid-log phase cells expressing EcIlvA and its variants were normalized to OD600 = 1. Serial dilutions of cultures (10−1 to 10−6) were prepared and 1 μL of each dilution was spotted onto MSM plates. Plates were incubated at 37 °C for 18 h, and growth was scored at each dilution. Protein purification and Size exclusion chromatography In order to purify the proteins, the genes were subcloned into the pET-22b(+) vector (Novagen) using the NdeI and XhoI restriction sites, generating constructs encoding C-terminal His₆-tagged proteins. Plasmids were transformed into E. coli Rosetta (DE3) pLysS (Novagen), and cultures were grown in LB medium containing 100 μg/mL ampicillin and 34 μg/mL chloramphenicol at 37 °C until OD₆₀₀ = 0.6. Protein expression was induced with 0.8 mM IPTG, and cultures were incubated for a further 20 h at 16 °C with shaking at 200 rpm. Cells were harvested by centrifugation at 3300 ×g for 15 min at 4 °C and resuspended in binding buffer (50 mM Tris-HCl, 300 mM NaCl, pH 7.8). Cells were lysed by sonication on ice (60% amplitude, 25 kHz, 1 s on / 2 s off, 30 min total), and the lysate was clarified by centrifugation at 32,000 ×g for 30 min at 4 °C. The supernatant was loaded onto a pre-equilibrated Ni Sepharose 6 Fast Flow column (Cytiva, 17531803). The column was washed with 3 column volumes of wash buffer (50 mM Tris-HCl, 300 mM NaCl, 20 mM imidazole, pH 7.8), and bound protein was eluted with the same buffer containing 500 mM imidazole. Eluted protein was dialyzed for 12 h at 4 °C against 25 mM Tris-HCl (pH 7.5), 50 mM NaCl, 0.1 mM EDTA, and 0.02 mM PLP, then concentrated and applied to a Superdex 75 16/600 column (Cytiva) pre-equilibrated with 25 mM Tris-HCl (pH 7.5) and 300 mM NaCl. Chromatography was performed at a flow rate of 1 mL min−1 at 16 °C. Fractions corresponding to the major elution peak were pooled and used for downstream experiments. Theoretical molecular weights were calculated using ProtParam-ExPasy Tool (web.expasy.org/protparam/). The column was calibrated with the Bio-Rad Gel Filtration Standard (Cat. #151–1901), containing thyroglobulin (670 kDa), bovine γ-globulin (158 kDa), chicken ovalbumin (44 kDa), equine myoglobin (17 kDa) and vitamin B12 (1.35 kDa). The apparent molecular weights quoted for them were extrapolated from the calibration curve and used only as an approximate indication of relative size. Enzyme activity assay The activity assay was conducted in a final volume of 1 mL reaction mixture containing 100 mM potassium phosphate buffer (pH 8.0), 1 mM pyridoxal 5′-phosphate (PLP), and 100 μL purified enzyme. To ensure that aggregated proteins were excluded from the analysis, fractions corresponding to size-defined SEC peaks were collected, the protein concentration was determined by A₂₈₀ measurement immediately prior to the assay (Table S7). The reaction was started by adding L-threonine to a final concentration of 40 mM, incubated at 25 °C for 15 min, then stopped with 1 mL of stop solution containing 1% (w/v) semicarbazide and 0.9% (w/v) sodium acetate. AKB formation was quantified by absorbance at 254 nm using a molar extinction coefficient of 540 M−1 cm−1 (Guillouet et al., 1999). Specific activity was calculated as μmol AKB min−1 mg−1 protein and was expressed as molar activity (μmol AKB min−1 μmol monomer−1) to enable comparison across constructs of differing length. All assays were performed in three independent experiments, and values are reported as mean ± SD. ABACUS-based screening for stabilizing mutations Single-site substitutions predicted to stabilize CDL were identified using ABACUS, a computational protein design program that employs a statistical energy (scoring) function to optimize amino acid sequences for a given backbone structure (Xiong et al., 2017). We systematically evaluated the ABACUS energy change induced by every possible single-point mutation across the monomer sequence. As the input structure, we used a homodimeric model derived from the full-length E. coli IlvA crystal structure (PDB ID: 1TDJ), retaining the cofactor pyridoxal-5′-phosphate (PLP) in the catalytic domain. To apply ABACUS (which natively handles single-chain inputs) to this homodimer without modifying its source code, we concatenated the two identical monomer sequences into a single artificial chain of double length, with the sequence repeated twice. Consequently, each single mutation introduced in the original monomer corresponds to an identical, symmetric double mutation in the merged chain. Substitutions yielding a net negative total energy change (i.e., lower ABACUS score) relative to the wild-type residue were predicted to be stabilizing. This comprehensive scan identified 87 residues. Of these, 19 residues are in direct contact with PLP (residues 60, 61, 62, 89, 157, 186–192, 241, 242, 286, 288, 289, 315, and 316) and 68 residues are located at the inter-chain interface (residues 55, 59, 116, 234, 275, 278, 279, 282, 284, 317–322, 324–332, 335, 342, 344, 372, 378–380, 406, 408, 412, 414, 415, 417–419, 421–427, 437, 439, 441, 462–464, 466, 469–476, 502, 505–507, 510–512). A total of 38 substitutions within the catalytic domain or the linker region [residues 10–335] were prioritized for experimental validation. Site-directed mutagenesis Site-directed mutagenesis was carried out by whole-plasmid PCR amplification using pBAD/Myc-His-cdl as the template and complementary primers carrying the desired mutation. Reactions were performed with PrimeSTAR Max DNA Polymerase (Cat. #R045A). The parental methylated template was digested with QuickCut™ Dpn I (Code No: 1609) for 1 h at 37 °C, and the digested product was transformed into competent E. coli JW3745–2 cells. The transformants were initially plated on screening MSM agar plates containing 0.2% (w/v) arabinose. The grown clones were identified and verified through sequencing. The growth of confirmed active clones was subsequently evaluated on MSM plates containing 0.1%, 0.05% and 0.02% (w/v) arabinose. To further explore mutations that could enhance stability, a mutagenesis library was created through error-prone PCR and constructed using MEGAWHOP cloning as previously described (Miyazaki, 2003). The colony survived in the presence of 0.02% arabinose was used as the template. After three rounds of random mutagenesis, colonies that showed the most significant growth on MSM supplemented with 0.02% (w/v) arabinose after 16 h were selected for sequencing. Final candidates were evaluated by spot titer test on MSM supplemented with 0.2% (w/v) arabinose, and the mutants showing growth comparable to the full-length EcIlvA protein were retained for further characterization. Crystallization The collected major peaks eluted from SEC were concentrated to different concentrations at 4 °C using an Amicon Ultra-15 Centrifugal Filter Units (Merck-Millipore). The proteins used for crystallization were purified and concentrated to 9 mg/mL for CDL and 26 mg/mL for HSG. Crystallization screening was conducted using commercial kits from Hampton Research, Molecular Dimensions, and Jena Bioscience collections. The Mosquito® liquid handling robot was employed to perform the hanging-drop vapor diffusion method. HSG crystals appeared after 14 days in a solution containing 1.5 M (NH₄)₂SO₄, 0.1 M Tris at pH 8.5. CDL crystals emerged after 9 days in a solution comprising 0.2 M CaCl₂·2H₂O, 0.1 M HEPES sodium pH 7.5 and 28% (v/v) PEG 400. Crystals were then transferred to a reservoir containing 30% (v/v) glycerol and rapidly cooled in liquid nitrogen for data collection. Structure determination Diffraction data were collected using the PILATUS 6 M detector at the BL18U1 beamline, operating at a wavelength of 0.97915 Å at the Shanghai Synchrotron Radiation Facility (SSRF). The crystal structure of HSG was refined to a resolution of 2.56 Å, while CDL crystal was refined to 2.9 Å. Data processing, including indexing, integration, and scaling, was performed using the XDS package (Brehm et al., 2023; Kabsch, 2010). Further processing, scaling, and merging of the data, with 5% flagged as FreeR, were conducted using the CCP4 suite of data reduction tools (Agirre et al., 2023). Phenix AutoBuild was employed to rebuild the top solutions from the search models (Terwilliger et al., 2008) while RESOLVE was used for density modification (Terwilliger, 2003). Model rebuilding was performed in Coot (Emsley et al., 2010), and the final structures were refined by Phenix (Headd et al., 2012). The statistics for data collection and structural refinement can be found in Table S2. The structural visualization was performed using PyMOL Molecular Graphics System (Schrodinger, 2015). The axes and rotational angles were derived using the draw_rotation_axis.py script in PyMOL. Molecular dynamics simulations MD simulations were conducted using GROMACS 5.1 (Abraham et al., 2015). The topology files for both the protein and ligand were generated using the CHARMM36 all-atom force field (July 2022) (Croitoru et al., 2021). Systems were solvated using the SPC water model within a cubic box of 10 nm, with periodic boundary conditions applied. Neutralization was achieved by adding Cl− and Na+ ions at a concentration of 0.15 mol L−1. To minimize the maximum force on each atom, steepest descent and conjugate gradient algorithms were sequentially applied to achieve a maximum force of less than 1000 kJ mol−1 nm−1. The system temperature was maintained at 300 K using velocity-rescaling with a reference temperature (ref_t) and a coupling time constant (tau_t) of 0.1 ps. Pressure was regulated at 1.0 atm using a Parrinello-Rahman barostat with a coupling constant of 2 ps. Equilibration was executed under a constant volume NVT ensemble for 100 ps, followed by a constant pressure NPT ensemble for an additional 100 ps (Singh et al., 2023). Three independent 200 ns sampling runs were performed for each system. MD trajectory analysis The MD trajectories were analyzed using g_rms, g_gyrate, and g_rmsf (Van Der Spoel et al., 2005) to get root-mean-square deviation (RMSD), radius of gyration (Rg) and root-mean-square fluctuations (RMSF). The rotational angle was derived from the matrix trace using cos θ = (trace(R) − 1)/2, the rotation axis was extracted from the antisymmetric components of R, and the signed angle assigned based on the orientation of this axis relative to a fixed reference direction (Choi, 2006). The cross-correlation of Cα atom displacements was carried out from MD simulation trajectories to identify intra and interdomain correlated motions (Ichiye and Karplus, 1991). The covariance matrices were computed using the g_covar tool in ASCII format and then converted into a normalized cross-correlation matrix using Python (v3.x) with the Matplotlib (v3.x) and NumPy (v.1.12) libraries. Residues 12–334 were selected for each chain as the common structurally equivalent region. All the graphs were generated using OriginPro software (Stevenson, 2011). Trajectories were analyzed using the MDTraj (McGibbon et al., 2015) and MDAnalysis (Naughton et al., 2022) packages via Python scripts. CRediT authorship contribution statement Samaneh Khodi: Writing – original draft, Software, Methodology, Data curation, Conceptualization. Abeeb Abiodun Yekeen: Data curation. Haiyan Liu: Writing – review & editing, Validation, Supervision, Funding acquisition. Quan Chen: Writing – review & editing, Validation, Supervision, Funding acquisition. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.