Integrative crystallography and molecular dynamics reveal isoform‐specific ligand interaction networks in C4 and nonC4 plant NADP‐malic enzyme

SUMMARY

NADP‐dependent malic enzyme (NADP‐ME) has been repeatedly co‐opted into distinct metabolic roles across plants, most prominently as the decarboxylase of NADP‐ME‐type C4 photosynthesis. In maize, the plastidic C4‐ and nonC4‐NADP‐ME isoforms are closely related in sequence yet display strikingly different catalytic properties, suggesting that small changes in ligand recognition can re‐tune reaction chemistry. However, mechanistic interpretation has been hampered by the scarcity of plant NADP‐ME structures captured in catalytically informative, ligand‐bound states. Here, we integrate X‐ray crystallography with structure‐guided docking and atomistic molecular dynamics (MD) to resolve ligand‐site interaction networks across reaction states. We determined a 2.55 Å structure of maize plastidic nonC4‐NADP‐ME bound to NADP⁺, pyruvate, and Mg²⁺, revealing a conserved NADP‐ME fold with localized active‐site flexibility. Comparison with maize C4‐NADP‐ME uncovers isoform‐specific rewiring of NADP⁺ and pyruvate contacts, with the nonC4 enzyme forming a denser product–cofactor interaction network. To access substrate‐bound states, we reconstructed malate‐NADP⁺‐Mg²⁺ complexes by docking followed by MD, identifying distinct malate‐Mg²⁺ coordination geometries and alternative NADP⁺ positioning between isoforms. Together, these structures and simulations provide a network‐level framework for plastidic NADP‐ME functional diversification and generate testable hypotheses for how ligand coordination drives isoform‐specific catalysis.

Significance Statement

This work provides mechanistic insight into isoform‐specific ligand recognition by plant NADP‐malic enzymes through the complementary use of experimental and in silico structural approaches applied to distinct active‐site states. We integrate X‐ray crystallography, structure‐guided docking, and molecular dynamics to define ligand interaction networks, revealing how subtle structural rewiring underlies functional diversification of closely related C4 and nonC4 isoforms.

INTRODUCTION

Even subtle alterations in active‐site geometry, oligomeric interfaces, or flexible loop regions can reshape the catalytic energy landscape of enzymes, tuning substrate binding, cofactor preference, turnover rate, and regulatory behavior (Tokuriki & Tawfik, 2009; Yang et al., 2020). Such effects are especially pronounced in enzyme families that have diversified through duplication and subfunctionalization, where high overall sequence conservation can mask functionally important local rewiring of interaction networks and dynamics (Conant & Wolfe, 2008; Innan & Kondrashov, 2010).

NADP‐dependent malic enzyme (NADP‐ME) provides a powerful system to explore how modest structural changes support metabolic specialization in plants. NADP‐ME catalyzes the oxidative decarboxylation of malate to pyruvate and CO₂ with reduction of NADP⁺, a reaction that has been repeatedly recruited into distinct physiological contexts (Gowik & Westhoff, 2011; Sage, 2004). Plant NADP‐ME isoforms differ in subcellular localization, kinetic properties, and regulation, supporting functions that range from the C4 carbon‐concentrating mechanism to plastidial “housekeeping” metabolism (Alvarez & Maurino, 2023; Maier et al., 2011; Maurino et al., 2009).

This functional diversification is well illustrated by the plastidic NADP‐ME pair from Zea mays. The photosynthetic C4‐NADP‐ME and the plastidic housekeeping isoform (nonC4‐NADP‐ME) share high sequence identity yet exhibit pronounced differences in catalytic performance and regulation, consistent with adaptation to contrasting metabolic environments (Alvarez et al., 2019; Bovdilova et al., 2019; Saigo et al., 2004). The extent of divergence despite strong sequence conservation suggests that isoform specificity may arise from relatively subtle structural adaptations, potentially involving ligand coordination, conformational transitions, and/or altered interface‐mediated dynamics, rather than wholesale reorganization of the fold.

A major obstacle to a structure‐based interpretation of these differences is the scarcity of plant NADP‐ME structures that capture catalytically relevant ligand‐bound states. In particular, substrate‐bound complexes are difficult to obtain because malate is challenging to trap in crystallo, limiting direct insight into malate coordination, metal‐binding geometry, and their coupling to NADP⁺ positioning. As a consequence, it remains difficult to directly compare substrate and product interaction networks across plant NADP‐ME isoforms using experimental structures alone.

Here, we integrate X‐ray crystallography with structure‐guided modeling to resolve ligand‐site interaction networks across reaction states in maize plastidic NADP‐ME isoforms. We solved a product‐bound crystal structure of maize nonC4‐NADP‐ME and compared product‐bound interaction networks to the maize C4 isoform and available plant references. To access substrate‐relevant states, we reconstruct malate‐bound complexes using docking followed by atomistic molecular dynamics simulations, enabling comparative analysis of malate‐metal coordination and NADP⁺ positioning. Together, these analyses provide a network‐level framework for understanding how conserved catalytic scaffolds support isoform‐specific functional divergence in plant NADP‐MEs.

RESULTS AND DISCUSSION

Capturing ligand‐site interaction networks in plastidic NADP‐ME isoforms

We set out to compare active‐site architectures (ligand‐site interaction networks) of the maize plastidic C4‐ and nonC4‐NADP‐ME isoforms. This comparison is hampered by the limited availability of fully liganded plant NADP‐ME structures, particularly for catalytically competent substrate‐bound states, which constrains direct inference of substrate/product recognition and isoform‐specific adaptations. To address this gap, we combined (i) a newly determined product‐bound crystal structure of the maize nonC4‐NADP‐ME with (ii) structure‐guided docking and atomistic MD simulations to reconstruct substrate‐bound states that are difficult to capture crystallographically (Figure S1).

Product‐bound crystal structure of nonC4‐NADP‐ME reveals conserved fold with enhanced local flexibility

A product‐bound plant reference structure is available for the tetrameric photosynthetic NADP‐ME from Sorghum bicolor in complex with pyruvate and NADP⁺ (PDB ID: 6C7N; Alvarez et al., 2019), providing an experimentally supported framework for product recognition in a C4‐NADP‐ME isoform. Building on this work, we co‐crystallized nonC4‐NADP‐ME in the presence of NADP⁺, pyruvate, and Mg²⁺ and solved the resulting structure (CSnonC4) by molecular replacement to 2.55 Å resolution (Table S1). The asymmetric unit contains two protomers (A and B; Figure 1a), while the tetramer is generated by crystallographic symmetry in space group P2₁22₁, as reported for other MEs (Chang & Tong, 2003; Tao et al., 2003). Protomer A is well resolved except for the first 33 N‐terminal residues, whereas protomer B displays weaker density and elevated B‐factors, with several disordered segments (Met1‐Thr32, Ala316‐Ala325, Ala334‐Ala337, Ile364‐Ala389, and Glu474‐Lys478) (Figure 1b), consistent with enhanced local flexibility. Despite this, both protomers superimpose closely (Cα RMSD 0.34 Å over 507 residues), and subsequent analyses therefore focus on protomer A unless stated otherwise. The CSnonC4 structure preserves the overall arrangement of secondary structural elements observed in maize C4‐NADP‐ME (CSC4; PDB ID: 5OU5; Figure S2), with 3D superposition yielding RMSDs of 0.36 and 0.50 Å for protomer and dimeric comparisons, respectively. When aligned with the human ME (PDB ID: 1PJ2, used for malate positioning in MD simulations), CSnonC4 shows an average RMSD of 1.14 Å over 504 Cα atoms, with ~51% sequence identity, reflecting a strong structural correspondence despite moderate sequence conservation.

Figure 1.

Crystal structure of nonC4‐NADP‐ME.

(a) Dimeric crystal structure of CSnonC4 (PDB ID: 9XYW). The 2Fo‐Fc electron‐density map (blue mesh), contoured at 0.8σ, is shown around protomer B. NADP⁺ and pyruvate are shown as sticks and Mg²⁺ as a sphere.

(b) Superposition of protomers A (pink) and B (green). Regions of protomer B with elevated B‐factors not modeled in the final structure are highlighted in blue; the start and end residues of each unmodeled segment are indicated.

Overall, CSnonC4 preserves the canonical NADP‐ME fold and aligns closely with CSC4, supporting a conserved global architecture across plastidic isoforms (Figure S2).

Product‐bound active‐site interaction networks differ between C4 and nonC4 isoforms

Inspection of CSnonC4 revealed well‐defined electron density for one NADP⁺ and one pyruvate in the active site. To enable a direct isoform comparison in the product‐bound state, we generated a product‐bound C4‐NADP‐ME model by docking NADP⁺ and pyruvate into the maize apo C4‐NADP‐ME structure (CSC4, Alvarez et al., 2019) (Figure S1A), guided by the strong conservation between maize and sorghum C4 enzymes, and evaluated ligand‐protein interaction patterns using LIGPLOT (Figures 2a,b and 3a,b).

Figure 2.

Comparative active‐site analysis of pyruvate and malate binding in C4, nonC4, and human NAD(P)‐ME.

Ligand‐protein interactions involving pyruvate (a–c) were analyzed using CSC4, CSnonC4, and the human NAD(P)‐ME structure PDB ID: 1PJ3, whereas those involving malate (d–f) were analyzed using malate‐bound conformations derived from docking and molecular dynamics simulations performed on CSC4, HMnonC4, and the human‐ME structure PDB ID: 1PJ2. Residues are labeled using isoform‐specific numbering. In the LIGPLOT diagrams, hydrogen bonds are shown in green, hydrophobic interactions in violet, and non‐bonded hydrophobic contacts as red dashed semi‐spheres. Conserved residues relative to C4‐NADP‐ME are marked with a green square; residues analogous to human ME are indicated by a red star; and residues matching the nonC4‐NADP‐ME counterpart are denoted by an orange circle. These comparisons reveal that C4 and nonC4‐NADP‐MEs accommodate both substrate and product through divergent ligand interaction networks. This rewiring of active‐site contacts provides a structural basis for their distinct catalytic properties.

Figure 3.

Comparative analysis of the NADP⁺‐binding site in C4, nonC4, and human NAD(P)‐ME in pyruvate‐ and malate‐bound states.

NADP⁺ interaction patterns are shown for pyruvate‐bound structures/models using CSC4, CSnonC4, and the Human NAD(P)‐ME structure PDB ID: 1PJ3 (a–c); whereas those involving malate (d–f) were analyzed using malate‐bound conformations derived from docking and molecular dynamics simulations performed on CSC4, HMnonC4, and the human‐ME structure PDB ID: 1PJ2. Residues are labeled using isoform‐specific numbering. In the LIGPLOT diagrams, hydrogen bonds are shown in green, hydrophobic interactions in violet, and non‐bonded hydrophobic contacts as red dashed semi‐spheres. Residues conserved relative to the C4 isoform are marked with a green square; residues analogous to human ME are indicated with a red star; and residues corresponding to the nonC4 isoform are denoted by an orange circle. Red underlined residues in CSnonC4 and the human‐ME denote positionally conserved residues with different identities between isoforms. These analyses reveal isoform‐specific NADP⁺ coordination patterns across reaction states. Differences in cofactor positioning and stabilization suggest alternative mechanisms of ligand recognition that may contribute to functional specialization.

This analysis identifies clear isoform‐specific differences in product recognition. In CSC4, the pyruvate C3 carbonyl group forms hydrogen bonds with Lys255 and Tyr184 (Figure 2a), whereas in CSnonC4, this group interacts with Glu335 (Figure 2b). In CSnonC4, the pyruvate carboxylate group additionally establishes a salt bridge with Arg245 and a hydrogen‐bond interaction with Asn500 (Figure 2b). Differences in the composition of the active‐site pocket were also observed. In CSC4, only two residues, Thr185 and Leu239, contribute hydrophobic contacts (Figure 2a), while in CSnonC4 five residues shape this region: Asn247, Lys263, Asp336, Asp359, and Asn545 (Figure 2b).

In both isoforms, NADP⁺ helps stabilize pyruvate in the active‐site conformation; however, the NADP⁺ interaction patterns also differ (Figure 3a,b). In CSC4, interactions with NADP⁺ are primarily mediated by the 2′‐phosphate group, which forms hydrogen‐bonding and electrostatic contacts with Arg237 and Asn331 (Figure 3a). In CSnonC4, the interaction between the NADP⁺ phosphate group and the conserved arginine residue Arg245, the structural analogue of Arg237 in CSC4, is maintained; additionally, this phosphate establishes further contacts with Ala395 and Gly393 (Figure 3b). Additional interactions are observed in the nonC4 isoform, where the nicotinamide moiety of NADP⁺ forms two hydrogen bonds, one with the side chain and the other with the amide group of Asn500 (Figure 3b). Furthermore, the adenine ribose phosphate establishes three hydrogen bonds with Ser427 and one with Lys443, while the ribose hydroxyl group interacts with the backbone carbonyl oxygen of Ala392 (Figure 3b). In addition, the NADP⁺ pocket is stabilized in CSC4 by 10 hydrophobic contacts involving Leu239, Gly240, Asp351, Thr355, Ser464, Val466, Asn492, Asn536, Asn537, and pyruvate (Figure 3a), all of which are conserved in CSnonC4 (Figure 3b), which additionally displays nine more hydrophobic contacts involving: Asn339, Gly391, Glu394, Asp426, Ser449, Asp359, Thr471, Leu498, and Gly525 (Figure 3b), indicating subtle differences in NADP⁺ accommodation between both isoforms.

Comparative evaluation of homology‐ and AlphaFold‐derived nonC4‐NADP‐ME models selects a structural framework for downstream analyses

To compare substrate interaction networks in C4‐ and nonC4‐NADP‐ME, we first built a structural model of the maize nonC4 isoform. CSC4 was selected as the primary template due to its high sequence identity. The resulting dimeric homology model (HMnonC4) superposed closely with a CSC4 dimer, yielding a Cα RMSD of 0.4 Å across 1176 residues (Figure 4a), supporting its use for comparative analyses. We then reconstructed malate‐bound active sites in both CSC4 and HMnonC4 by docking malate and performing atomistic MD simulations in the presence of malate, NADP⁺, and Mg²⁺. MD simulations therefore provide a systematic framework to identify stable binding modes and to quantify dynamic interaction patterns within the catalytic site.

Figure 4.

Structural comparison of homology‐ and AF‐modeled nonC4‐NADP‐ME.

(a) Superposition of the nonC4‐NADP‐ME homology model (HMnonC4) with the C4‐NADP‐ME crystal structure (CSC4; PDB ID: 5OU5). Protomers A and B are shown in cartoon and ribbon representations, respectively, to emphasize local differences in Cα positions. Blue arrows mark regions of pronounced deviation that may influence isoform‐specific dynamics and/or substrate accessibility.

(b) Superposition of the AlphaFold3‐derived nonC4‐NADP‐ME model (AFnonC4) with CSC4, highlighting overall fold conservation and localized conformational differences relative to the template structure.

Secondary‐structure prediction using ESPript 3.2 revealed a high degree of structural conservation between the plastidic NADP‐ME isoforms (Figure S2). Minor differences were observed in helical elements, notably in residues 194–199 following α‐helix 7: In the C4 isoform, this segment arranges as a 3₁₀ helix (η8; i → i + 3), whereas in nonC4‐NADP‐ME, it forms a canonical α‐helix (α8; i → i + 4). This subtle variation alters the local helical packing, while β‐sheet elements remain fully conserved.

To complement the homology model, we generated an AlphaFold3‐derived structure (AFnonC4; Figure 4b), which confirmed the overall fold of the homology model and aligned closely with CSC4 (RMSD of 1.0 Å over 1176 residues of all Cα atoms; Figure 4b). This model confirmed global structural conservation and revealed increased local conformational variability relative to the template.

To investigate the structural basis of RMSD deviations between the modeled nonC4‐NADP‐ME structures, we performed a comparative analysis of dimer interfaces and active‐site conformations using PISA and LIGPLOT, respectively. For both HMnonC4 and AFnonC4, the interface between protomers A and B was examined with a focus on hydrogen bonds (Figure 5a and Table S2), salt bridges (Figure 5b and Table S3), and residues contributing to interface stabilization or destabilization (Figure 5c,d and Table S4). Both models share nine hydrogen bonds at the dimer interface (Table S2 and Figure 5a). AFnonC4 displays nine unique hydrogen bonds involving the residue pairs R131‐R218, Y155‐R208, F207‐K147, G212‐Y106, and N151‐S205. Similarly, HMnonC4 exhibits nine unique hydrogen bonds, including Q160‐Q154, Q160‐Q156, K147‐G204, Y106‐P210, N151‐R208, and R208‐R171, which involve residues absent in AFnonC4. Notably, specific atoms within residues R208, K147, and Y106 form hydrogen bonds in both structures while participating in distinct interaction networks in each model. Salt‐bridge analysis identified 11 interactions in HMnonC4 and 12 in AFnonC4, with seven conserved between both structures (Figure 5b and Table S3). HMnonC4 exhibits four unique salt bridges, three involving R171‐D167 interactions and one involving R110‐E170 (Figure 5b and Table S3). By contrast, AFnonC4 contains five unique salt bridges, including three involving D219‐R131 and one each between E122‐R218 and R110‐D111 (Figure 5b and Table S3). In addition to conserved interactions, residue R110 engages different salt‐bridge partners in each model, interacting with E170 in HMnonC4 and with D111 in AFnonC4. Analysis of interface stabilization revealed a total of six stabilizing residues in both AFnonC4 and HMnonC4, with five residues being fully conserved: P135, P136, P210, Y214, and F297. Model‐specific contributions include A137 in AFnonC4 and R131 in HMnonC4 (Figure 5c and Table S4). Among destabilizing residues, only R110 is conserved, while N160 and E170 are unique to AFnonC4, and D219 is exclusive to HMnonC4 (Figure 5d and Table S4).

Figure 5.

PISA‐based comparison of the protomer A‐B dimer interface in the HMnonC4 and AFnonC4‐NADP‐ME models.

Venn diagrams summarize the overlap in (a) hydrogen bonds, (b) salt bridges, (c) stabilizing residues, and (d) destabilizing residues identified by PISA (see also Tables S2–S4). Residues are color‐coded by model: HMnonC4 (orange) and AFnonC4 (violet). Boxed labels highlight residues participating in the indicated interactions. “x2” and “x3” denote residues involved in two or three equivalent interactions, respectively.

Together, these analyses show that the homology‐ and AF‐derived models display distinct patterns of interfacial interactions. In this setting where a closely related, high‐resolution crystal structure is available (CSC4) and sequence identity is high (>82%), the homology model provides the closest approximation to experimentally observed conformations, consistent with its lowest global Cα RMSD.

Molecular docking and dynamics reveal isoform‐specific active‐site interactions with the substrates

To compare substrate interaction networks in C4 and nonC4‐NADP‐ME, we reconstructed malate‐bound active sites using docking followed by atomistic MD simulations with malate, NADP⁺, and Mg²⁺ (Figure S1B). MD simulations revealed distinct substrate coordination modes in the active sites of the two isoforms, as assessed by LIGPLOT analysis (Figures 2 and 3). In C4‐NADP‐ME, the dominant conformational cluster shows malate stabilized by a bidentate salt bridge involving Arg237 and Lys255, together with hydrogen‐bond interactions with Tyr184 and Thr185, and direct coordination to Mg²⁺ (Figure 2d). By contrast, the malate in the nonC4 isoform adopts a more compact and closed binding mode (Figure 2e). Here, malate coordination involves Asp426 and a bidentate interaction with Mg²⁺, giving rise to a non‐canonical metal‐binding geometry (Figure 2e). This alternative malate coordination mode differs from that observed in the C4 isoform but represents a chemically viable geometry, as supported by similar coordination patterns reported in PubChem (Figure 2e). Moreover, the Mg‐O‐C angle of the malate carboxylate group remains centered around ~90° throughout the simulation, indicating a stable coordination geometry between the carboxylate oxygen and the Mg²⁺ ion. The observed fluctuations are consistent with a dynamic coordination environment influenced by solvent and neighboring residues.

In the nonC4 isoform, the malate‐binding pocket is additionally stabilized by non‐hydrophobic interactions involving residues conserved with the C4 isoform (Tyr192, Leu247, and Asp359), a residue conserved with the human NAD(P)‐ME (Glu335), and the nonC4‐specific residue Asp358 (Figure 2e). In this isoform, Tyr192 participates in hydrophobic contacts that contribute to pocket stabilization. By contrast, and consistent with previous mechanistic work implicating this residue in the catalytic mechanism of human NAD(P)‐ME (Tao et al., 2003), the homologous Tyr184 residue in the C4 isoform forms a hydrogen bond with malate (Figure 2d).

Besides, distinct NADP‐binding patterns further differentiate the two isoforms. In the C4 isoform, NADP interacts with Arg237, Asn331, Gly383, Gly385, Glu386, Ala387, Ser419, Lys435, Asn492, and Ser535 and establishes contacts with malate (Figure 3d). Conversely, in the nonC4 isoform, NADP engages Arg245, Thr363, Gly391, Ala392, Glu394, Asn500, and Ser543 (Figure 3e), highlighting isoform‐specific cofactor recognition and stabilization mechanisms.

From substrate to product: Conserved catalytic scaffolds but rewired ligand interaction networks

To place the maize plastidic isoforms into a mechanistic framework, we compared their ligand‐site interaction networks with those observed in human NAD(P)‐ME, for which malate‐ and pyruvate‐bound reference structures are available (Chang & Tong, 2003; Tao et al., 2003). Pyruvate‐binding interactions were assessed relative to the pyruvate‐bound human structure (PDB ID: 1PJ3), whereas malate‐binding interactions were evaluated against the malate‐bound human structure (PDB ID: 1PJ2).

In the maize C4 enzyme (CSC4), both malate‐ and pyruvate‐bound states engage the conserved Tyr/Lys pair (Tyr184 and Lys255) (Figure 2a,d), mirroring the interactions formed by the corresponding residues in human NAD(P)‐ME (Figure 2c,f). These residues have been proposed to constitute the catalytic pair in the human enzyme (Chang & Tong, 2003; Tao et al., 2003), and their conservation supports a shared catalytic scaffold in which the Tyr/Lys pair remains central to ligand positioning across reaction states. By contrast, the nonC4 isoform adopts an alternative substrate‐binding arrangement. In the malate‐bound ensemble, malate does not directly contact residues equivalent to the proposed Tyr/Lys catalytic pair, despite the overall conservation of the active‐site cavity. (Figure 2d–f). Instead, malate is stabilized within a pocket shaped by conserved positions including Tyr192, Leu247, Asp359, and Glu335 (the latter shared only with the human enzyme; Figure 2e), consistent with an isoform‐specific substrate accommodation strategy.

A similar divergence is observed for pyruvate binding. In the nonC4 isoform, pyruvate occupies an active‐site pocket that does not directly engage the residues proposed to form the catalytic pair, yet it is framed by positions that are partly conserved across isoforms. For example, Lys263 in nonC4 aligns with Lys255 in the C4 isoform and Lys183 in the human enzyme, and Arg245 corresponds to human Arg165, indicating retention of key structural elements within the active‐site environment (Figure 2a–c, Chang & Tong, 2003; Tao et al., 2003). Despite this conservation, the interaction patterns differ markedly between isoforms. In nonC4, the pyruvate carboxylate forms hydrogen bonds with Arg245 and Asn500, residues conserved with the human enzyme but not used in the same way in C4‐NADP‐ME, while the carbonyl group interacts with Glu335 (Figure 2b). By contrast, in C4‐NADP‐ME the carbonyl group forms hydrogen bonds with Tyr184 and Lys255 (Figure 2a), with the homolog lysine residue (Lys183) also participating in the corresponding interaction in the human enzyme (Figure 2c). Consistent with these differences, the hydrophobic pocket surrounding pyruvate is more extensive in nonC4‐NADP‐ME, being formed by Asn247, Lys263, Asp336, Asp359, and Asn545; four of these residues are conserved with the human enzyme (Figure 2b,c). Notably, Lys263 appears to undergo a functional shift within this pocket: rather than serving primarily as a structural component, its equivalent residue forms direct hydrogen bonds to pyruvate in the C4 and human enzymes (Figure 2a–c). Together, these observations reveal isoform‐specific ligand interaction networks that may contribute to the distinct kinetic properties of C4‐ and nonC4‐NADP‐MEs (Alvarez et al., 2019; Böhm et al., 2026), while remaining consistent with the conserved overall fold and catalytic architecture described for ME more broadly (Chang & Tong, 2003; Tao et al., 2003).

NADP ⁺ positioning and retention across reaction states correlates with isoform‐specific turnover

Comparison of the cofactor‐binding site in the malate‐bound state reveals pronounced isoform‐specific differences in NADP⁺ positioning. In the C4 isoform, the catalytic C4 atom of the nicotinamide ring is oriented toward and makes direct contact with malate, consistent with a geometry conducive to hydride transfer (Figure 3d). By contrast, in both the nonC4 and human enzymes, the nicotinamide ring does not contact malate directly but instead interacts with a conserved Asn residue shared by these two enzymes (Figure 3d–f).

Additional divergence is evident at the adenine‐associated phosphate of NADP⁺. In the C4 isoform, this phosphate forms two interactions with Ser419 and Lys435, residues that are not conserved in the nonC4 isoform (Figure 3d,e). Conversely, in the nonC4 isoform, the corresponding phosphate engages Glu394 and Gln440 (Figure 3d,e). Despite these differences, nonC4‐NADP‐ME retains most residues involved in NADP⁺ coordination in C4‐NADP‐ME, with the notable exception of Ala392, which is instead conserved in the human enzyme (Figure 3d–f).

These structural differences become particularly evident when comparing malate‐ versus pyruvate‐bound states. The C4 isoform shows a pronounced loss of NADP⁺ contacts upon product formation, whereas in the nonC4 and human isoforms, most NADP⁺ interactions are retained in the pyruvate‐bound state, yielding a denser product–cofactor interaction network. This contrast is consistent with the established kinetic hierarchy (higher k _cat for the C4 isoform relative to nonC4 and human isoforms) (Alvarez et al., 2019; Böhm et al., 2026; Hsieh et al., 2006) and suggests a simple structural interpretation: Stronger cofactor/product coordination in nonC4 and human enzymes could favor product retention and slow pyruvate release, whereas weaker coordination in the C4 enzyme could facilitate product dissociation and higher turnover.

Conclusions and outlook: Opportunities and limitations of combining modeling and crystallography to resolve NADP‐ME function

By integrating X‐ray crystallography with structure‐based modeling (homology modeling, docking, and atomistic MD), we mapped ligand‐site interaction networks in plastidic NADP‐ME across product‐ and substrate‐relevant states. Across isoforms and ligand states, the active site retains a conserved structural scaffold, even as ligand binding is accompanied by local rearrangements. This combination of conservation and state‐dependent reorganization supports a model in which isoform specificity emerges primarily from rewiring of interaction networks within a largely preserved catalytic architecture (Figures 2 and 3). Collectively, our data define structural features that likely contribute to functional diversification between C4‐ and nonC4‐NADP‐MEs and underscore the value of combining in silico and experimental approaches to connect sequence/structure variation with catalytic behavior (Alvarez et al., 2019; Böhm et al., 2026).

At the same time, the present work highlights clear priorities for future studies. Substrate‐bound plant reference structures remain a major bottleneck, and reconstructed malate‐bound states necessarily depend on assumptions about protonation, metal coordination, and ligand placement that are difficult to fully validate in the absence of experimental trapping. Progress will therefore benefit from complementary strategies that directly constrain substrate‐state geometry, including crystallographic trapping with non‐reactive analogs, Cryo‐EM approaches, systematic mutagenesis guided by the interaction networks identified here, and kinetics under controlled Mg²⁺/pH regimes. Together, such efforts should refine substrate coordination models, test the proposed network‐based determinants of turnover and product release, and ultimately enable a more quantitative, state‐resolved structure–function framework for plant NADP‐ME evolution.

MATERIALS AND METHODS

Protein structures

NAD(P)‐ME structures were obtained from the Protein Data Bank (PDB): CSnonC4 (PDB ID: 9XYW), CSC4 (PDB ID: 5OU5), sorghum NADP‐ME (PDB ID: 6C7N), and human NAD(P)‐ME in complex with malate (PDB ID: 1PJ2) or pyruvate (PDB ID: 1PJ3). In addition, a structural model of maize nonC4‐NADP‐ME (UniProt: A0A4Y2) was generated using AlphaFold3 (Abramson et al., 2024).

Molecular docking

Molecular docking analyses were carried out using the HADDOCK 2.4 web server (van Zundert et al., 2016), applying its standard protocol for protein–ligand interactions. For each system evaluated, the highest‐ranked structure within the most populated cluster was selected for downstream examination. The resulting complexes were analyzed for interaction profiles using the Protein–Ligand Interaction Profiler (PLIP) (Salentin et al., 2015), and the docking conformations were visualized and explored in 3D using PyMOL version 2.5 (Schrödinger, LLC).

Molecular dynamics simulations

All‐atom MD simulations were performed to investigate the structural differences between C4 and non‐C4 MEs complexed with malate, NADP, and Mg²⁺ ligand topologies were prepared using SwissParam (Bugnon et al., 2023; Zoete et al., 2011). System setup and simulations were carried out using GROMACS 2021 (Bekker et al., 1993), applying the CHARMM03 force field (Brooks et al., 1983) and the TIP3P explicit solvent model (Jorgensen et al., 1983). Each protein–ligand complex was solvated in a triclinic box with 0.15 M NaCl to mimic physiological ionic strength. The systems were energy‐minimized and equilibrated prior to 500‐ns production runs in the NPT ensemble at 300 K and 1 bar.

Trajectory analyses included monitoring structural stability and residue interactions. Representative conformations were extracted based on clustering analysis. Protein–ligand interaction profiles were characterized using LIGPLOT (Wallace et al., 1995), and structural visualizations were conducted with PyMOL version 2.5 (Schrödinger, LLC). This approach enabled a comparative evaluation of malate coordination and active‐site conformational dynamics. Angles describing the Mg²⁺‐malate coordination were monitored during the simulation. In particular, the Mg‐O‐C angle, defined between the magnesium ion, the coordinating carboxylate oxygen, and the corresponding carboxyl carbon of malate, was calculated as a function of time using the gmx angle.

To set up the malate position during simulations, we first surveyed all available NADP‐ME structures in the PDB, identifying 61 entries (Data S1). Sequence identity analysis indicated that only 31 structures exceeded the 50% threshold considered reliable for homology conservation. From this subset, we selected the human‐ME isoform co‐crystallized with malate, NAD⁺, and Mn²⁺ (PDB ID: 1PJ2) as a reference. This structure provides a fully defined ligand‐bound active site and thus a reliable template for substrate positioning in CSC4 (the apo structure of maize C4‐NADP‐ME) and HMnonC4. HMnonC4 was selected for further analyses due to its close structural resemblance to CSC4, compared with the AF‐derived model (Figure 1), allowing us to start MD simulations from all apo‐modeled structures as a common reference point (Figure S1B). Initial malate poses for MD were generated by docking, guided by the malate position of the human‐ME structure (PDB ID: 1PJ2) and the pyruvate position of the Sorghum C4‐ME structure (PDB ID: 6C7N).

NonC4‐NADP‐ME crystallization and crystal harvesting

Protein expression and purification were performed according to previously described protocols (Alvarez et al., 2019; Böhm et al., 2026). Initial crystallization screening on CSnonC4 was performed using a protein solution at 10 mg/mL in weak buffer consisting of 20 mM Tris–HCl pH 8.0, 5 mM MgCl₂, 40 mM pyruvate, and 2 mM NADP. Sitting drops (0.2 μL of protein solution +0.2 μL of commercial precipitant solution) were set up on Jena Bioscience MRC 2‐well crystallization plates (Jena, Germany) with the help of an Art Robbins Gryphon robot (Sunnyvale, CA, USA). Equilibration was done at 4 °C with manual visualization on a Formulatrix Rock Imager 1000 apparatus (Bedford, MA, USA). Crystals appeared with a solution consisting of 8% (w/v) PEG 20000, 8% (w/v) PEG 550 MME, 50 mM Tris–HCl pH 8.5, and 1.2 M sodium formate. Samples were directly harvested from the robotic plates using nylon loops, cryo‐protected in their respective mother liquor added with glycerol and flash cooled in liquid nitrogen for subsequent analysis.

X‐ray data collection and processing, structure resolution, and refinement

X‐ray diffraction data were collected on several crystals at the P13 beamline at the DESY—PETRA III storage ring operated by the EMBL (Hamburg, Germany) (Cianci et al., 2017), with a Dectris EIGER X 16 M detector (Baden, Switzerland). Initial data processing was done with XDS (https://xds.mr.mpg.de/) and Aimless from the CCP4 suite (Agirre et al., 2023), with the best sample diffracting to 2.55 Å resolution in the orthorhombic space group P2₁22₁. A total of 5% of the recorded reflections were separated for cross‐validation purposes. The structure was solved by the molecular replacement method as implemented in CCP4 Cloud with Balbes (Long et al., 2008), using the coordinates of the pigeon cytosolic NADP‐ME (PDB ID: 1GQ2) as the starting model (Yang et al., 2002), followed by automatic model building with ModelCraft (Bond & Cowtan, 2022). Next, the model was subjected to several cycles of manual inspection and modification with Coot (Emsley et al., 2010) and refinement with Refmac5 applying non‐crystallographic symmetry restraints (Murshudov et al., 2011). In the last cycles, a total of two Mg²⁺ ions as well as two NADP, two formate, two pyruvate, and 13 water molecules were added to the model. The final coordinates were validated with MolProbity (Chen et al., 2010) and with the validation module of Coot. Detailed statistics on data collection, refinement, validation, and Protein Data Bank (PDB) deposition are presented in Table S1. The final model displays R_work and R_free values of 0.244 and 0.308, respectively, and acceptable stereochemical quality, with 88.0% of residues in favored regions of the Ramachandran plot, 10.9% in allowed regions, and 1.1% outliers.

CONFLICT OF INTEREST

The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.

Supporting information

Data S1. Overview of all ME structures solved to date. The table lists the PDB accession codes and sources, indicates the presence of ligands and substrates, and specifies the method used for structure determination.

Figure S1. Workflow for active‐site interaction analysis in C4 and non‐C4‐NADP‐ME isoforms with pyruvate and malate. (A) Pyruvate‐bound models were generated from the available crystal structures of the maize C4 isoform (CSC4; PDB ID: 5OU5) and the maize non‐C4 isoform (CSnonC4; PDB ID: 9XYW). For CSC4, pyruvate was docked with its initial placement guided by the pyruvate pose observed in sorghum NADP‐ME (PDB ID: 6C7N). (B) Malate‐bound models required different starting structures for each isoform. Because docking malate into CSnonC4 (PDB ID: 9XYW) necessitates removal of the co‐crystallized pyruvate, the HMnonC4 model (0.4 Å RMSD to CSnonC4) was used as the starting structure. For the C4 isoform, malate was docked directly into the apo CSC4 structure (PDB ID: 5OU5). In both isoforms, malate placement was guided by the malate‐bound human NAD(P)‐ME structure (PDB ID: 1PJ2) and by the pyruvate position in the sorghum NADP‐ME reference structure (PDB ID: 6C7N). MD—molecular dynamics.

Figure S2. Secondary‐structure comparison of maize C4 and nonC4‐NADP‐ME isoforms and the human homolog. Secondary‐structure elements were predicted using ESPript 3.2, with CSC4 (A) and HMnonC4 (B) as structural templates. α‐helices, β‐strands, and η (3₁₀) helices are indicated. A yellow star highlights the region immediately following α‐helix 7, where secondary‐structure predictions diverge between the C4 and nonC4 isoforms. Direct ligand‐interacting residues are indicated using the same symbols as in Figures 2 and 3: squares for the maize C4 isoform, circles for the maize nonC4 isoform, and stars for the human malic enzyme. Pyruvate‐interacting residues are shown in violet and malate‐interacting residues in blue.

Table S1. Crystallographic data collection and refinement statistics of nonC4‐NADP‐ME.

Table S2. Hydrogen‐bond interactions between protomers A and B in nonC4‐NADP‐ME AF and HM models. The table lists the interacting residues and atoms identified by PISA, together with the corresponding distances (Å). Conserved interactions are highlighted in gray, whereas differences are shown in bold. Red asterisks mark residue atoms that are conserved between both models but engage with a new, different interaction partner.

Table S3. Salt‐bridge interactions between protomers A and B in nonC4‐NADP‐ME AF and HM models. The table lists the interacting residues and atoms identified by PISA, together with the corresponding distances (Å). Conserved interactions are highlighted in gray, whereas differences are shown in bold.

Table S4. Stabilizing and destabilizing residues between protomers A and B in nonC4‐NADP‐ME AF and HM models. The table lists the residues and atoms involved in stabilizing or destabilizing interactions identified by PISA. Conserved residues are highlighted in gray, whereas differences are shown in bold.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author.