Linker‐mediated domain separation enhances cold adaptation in cellulases

Abstract

Cold‐adapted cellulases have the potential to reduce energy demands in industrial processes by enabling efficient cellulose saccharification at lower temperatures. However, the structural basis of cold adaptation in bidomain cellulases remains poorly understood. Our prior studies on bidomain amylases proposed a “linker spacer effect”—a greater magnitude of domain separation is associated with improved cold adaptation. Here, we investigated this effect in the bidomain cellulase Cel5G and its linker‐modified variants through molecular dynamics simulations. Domain separation, quantified by domain separation index (DSI), positively correlates with catalytic turnover number at 10°C, meaning that Cel5G variants with a greater DSI demonstrates a higher cold activity. Structural analyses show that disulfide‐bonded loops are pivotal to maintain a high population of extended conformations that involve a greater DSI and reduced interdomain hydrogen bonding (H‐bonding). Besides serving as a spacer, linkers in bidomain enzymes can also modulate the active site by altering the protein's conformational dynamics, fine‐tuning the frequency of H‐bond interactions to influence catalytic residues' capability of binding or reacting to the substrate. Overall, this study enhances our structural and dynamics‐based understanding of cold adaptation of bidomain enzymes, guiding the development of new strategies for engineering cold‐adapted enzymes.

1. INTRODUCTION

Cellulases catalyze the hydrolysis of β‐1,4‐glycosidic bonds in cellulose and have a vast array of applications in industries such as paper and pulp manufacturing, food and juice production, textile processing, and biorefining (Bhardwaj et al. 2021; Hamid et al. ²⁰¹⁵; Juturu and Wu ²⁰¹⁴). The cellulase market is projected to reach $2.45 billion by 2026, accounting for 20% of the total global enzyme market (Ranjan et al. 2023; Yoon et al. ²⁰¹⁴). One of the most significant uses of cellulases is degrading cellulose into fermentable monosaccharides, which can be converted into bioethanol (Vasić et al. 2021). However, saccharification usually requires >50°C as the reaction temperature condition due to feeble activities of cellulases at lower temperatures (Khalili Ghadikolaei et al. 2018). This requirement leads to substantial energy consumption and operational costs in the biofuel industry. Creating cellulases and their mutants capable of catalyzing saccharification at lower temperatures could minimize industrial heating needs, representing a promising solution for both economic efficiency and environmental sustainability (Jurich et al. 2025).

Cold‐adapted cellulases have been isolated from bacteria, earthworms, and fungi (Sonan et al. 2007; Ueda et al. ²⁰¹⁴; Xiao et al. ²⁰⁰²). For example, a cellulase sourced from Eisenia fetida retains 38% activity at 10°C (Ueda et al. 2014), and another from Trichoderma reesei maintains 25% activity at 20°C (relative to their optimal temperature activities at 40°C and 50°C, respectively) (Xiao et al. 2002). In both cases, the observed relative activities exceed the expected 12.5% remaining after a 30°C drop—based on the typical assumption of activity halving every 10°C, derived from Arrhenius equation—demonstrating these cellulases' ability to compensate the loss of activity caused by thermal effects. Cellulases typically feature a bidomain construct, consisting of three structural components: a catalytic domain (CD) for catalysis, a carbohydrate‐binding module (CBM) for substrate recruitment, and a peptide linker connecting both domains (Violot et al. 2005). One of the most extensively characterized cold‐adapted cellulases is Cel5G from Pseudoalteromonas haloplanktis (Garsoux et al. 2004; Sonan et al. ²⁰⁰⁷; Violot et al. ²⁰⁰⁵). Cel5G exhibits maximal catalytic activity at 40°C and maintains over 50% activity at 25°C, demonstrating a 10‐fold higher relative activity compared to its mesophilic homolog Cel5A from Erwinia chrysanthemi (Sonan et al. 2007). This cold adaptation behavior was also observed to depend on the linker length, where shortening Cel5G linker length by two‐thirds leads to a 33% decrease in k _cat at 10°C (Sonan et al. 2007).

Despite the kinetic characterization, the structural understanding of the origin of cold adaptation of cellulases is limited. The crystal structure is available only for the CD of Cel5G, while the full‐length structure of Cel5G has not been determined by X‐ray crystallography due to the high flexibility of its linker region, nor by NMR because of its large size (Violot et al. 2005). This is typical of most bidomain enzymes, such as amylases (Zhang et al. 2020). To our knowledge, only one full‐length cellulase structure has been determined, which is a trimodular halophilic cellulase with no linker, deposited in the PDB (ID: 5E09), published by Zhang et al. (2015). As such, existing mechanistic discussions about the origin of cold adaptation in bidomain enzymes have focused solely on the amino acid profile (e.g., a reduced proline and increased glycine composition; Garsoux et al. ²⁰⁰⁴; Öten et al. ²⁰²³; Violot et al. ²⁰⁰⁵). Although small‐angle X‐ray scattering experiments suggest that full‐length Cel5G adopts an extended conformation (Violot et al. 2005), the conformational dynamics of the full‐length bidomain enzymes and their relevance to cold adaptation remain poorly understood.

Leveraging high‐throughput molecular dynamics (Jurich et al. 2025; Shao et al. ²⁰²²; Shao et al. ²⁰²³; Shao et al. ²⁰²⁵), our prior studies investigated the molecular basis of cold adaptation in bidomain amylases, which are similar in construct to cellulases but function to decompose starch, rather than cellulose (Ding et al. 2025). Compared to mesophilic amylases, cold‐adapted amylases exhibit a greater magnitude of interdomain separation between CD and CBM, which can be quantified by domain separation index (DSI). Through in silico screening of linkers that result in a greater DSI, we identified a helical linker that transforms a mesophilic amylase from Pseudomonas saccharophila into a cold‐adapted variant, achieving a 3‐fold increase in relative activity at 0°C compared to the wild type. However, it remains unknown whether this “linker spacer effect” applies to cellulases and other bidomain enzymes.

In this study, we investigated the structure–dynamics–function relationships underlying cold‐adapted bidomain cellulases through molecular dynamics (MD) simulations, focusing on how linker‐mediated domain separation influences cold adaptation in Ce5G and its variants. Using Cel5G and three linker‐modified variants as model enzymes, we found that domain separation index is positively correlated with experimentally measured k _cat at 10°C, where cold‐adapted bidomain variants show a greater magnitude of domain separation.

While this trend reflects a distinct metric—k _cat at low temperature rather than relative activity—it reinforces the previously noted “linker spacer effect” seen in cold‐adapted amylases, where increased domain separation was linked to enhanced cold adaptation (Ding et al. 2025). Conformational analyses reveal that cold‐adapted variants leverage their linkers, designed with varying linker lengths and disulfide‐bonded cyclic motifs, to promote domain separation, which leads to a higher population of open conformational states and increases active site flexibility within the CD through linker‐mediated structural dynamics. This study informs design principles for engineering bidomain cellulases with improved cold adaptation, offering potential advancements in industrial cellulose degradation and sustainable biofuel production.

2. RESULTS AND DISCUSSION

2.1. Cold‐adapted cellulases exhibit greater magnitude of domain separation

To investigate the molecular basis underlying cold adaptation in bidomain cellulases, we conducted 5 × 200 ns MD simulations on Cel5G and three of its variants with a modified linker, whose kinetic parameters have been experimentally characterized. They share an identical set of CD and CBM but differ in their linker composition and topology. The Cel5G linker, with a total length of 109 residues, contains three circular loops, each comprising 13 residues, cyclized through disulfide bonds between C314‐C328, C345‐C359, and C382‐C396. In the Δ1L variant, 31 residues (D312–V342) are truncated from Cel5G, including the first disulfide‐bonded loop (C314–C328). In the Δ2L variant, 68 residues (D312–I379) are truncated, encompassing two disulfide‐bonded loops (C314–C328 and C345–C359). Δ2L‐CA preserves the linker length of Δ2L but incorporates C382A and C396A mutations to break the one remaining disulfide‐bonded loop (Figure 1a and Table S1, Supporting Information). The bidomain amylase structures were predicted by AlphaFold2, which demonstrates success in predicting the structures of bidomain enzymes, such as Aspergillus fumigatus EgtA (Doyle et al. 2022) and Chlamydomonas reinhardtii GPD2 (Cruz‐Powell et al. ²⁰²⁴).

FIGURE 1.

Global flexibility in Cel5G and its linker‐modified variants. (a) Schematic representation illustrating the construction of Δ1L, Δ2L, and Δ2L‐CA. (b) RMSD for Δ1L, Δ2L, and Δ2L‐CA at 10°C. RMSD was calculated from all heavy atoms using the most representative structure from hierarchical clustering as a reference. Outliers are not depicted in the box plot.

We first assessed the global flexibility of each enzyme by calculating its RMSD using the most representative structure from hierarchical clustering as the reference (Figure 1b and Table S3). Larger RMSD values are associated with enhanced global flexibility, which has been observed in a cold‐adapted endonuclease compared to its mesophilic counterpart (Benrezkallah et al. 2020). The average RMSD values for Cel5G and its variants differ substantially: Cel5G (18.99 ± 6.28 Å), Δ1L (12.24 ± 4.50 Å), Δ2L (6.34 ± 3.59 Å), and Δ2L‐CA (2.98 ± 1.20 Å) (Table S3). The reduction in global flexibility from Cel5G to Δ2L‐CA parallels the trend in cold activity among the variants, represented by k _cat values at 10°C—12 ± 0.01 s⁻¹ for Cel5G, 10 ± 0.02 s⁻¹ for Δ1L, 9 ± 0.01 s⁻¹ for Δ2L, and 8 ± 0.02 s⁻¹ for Δ2L‐CA.

Since all variants share an identical CD and CBM, the observed differences in RMSD are attributed to changes in the linker‐mediated relative motions between these two domains. To quantify the magnitude of relative motions between CD and CBM, we calculated domain separation index (DSI) by subtracting the R _g sum of both domains from the distance between their geometric centers (Figure 2a). The DSI values for Cel5G, Δ1L, Δ2L, and Δ2L‐CA are 41.3 ± 8.9 Å, 29.4 ± 9.2 Å, 8.0 ± 3.4 Å, and 5.9 ± 1.7 Å, respectively (Table S4), which aligns with the RMSD values. The trend in DSI values among Cel5G variants remains consistent at 50°C (Table S5). Similar to our prior studies on amylases (Ding et al. 2025), we observed that cellulase variants with a greater DSI value demonstrate stronger cold adaptation. The experimental k _cat values show a strong positive correlation with the DSI values with a Pearson correlation coefficient of 0.96 and Spearman correlation coefficient of 1.00 (Figure 2b). Considering the potential limitation of AlphaFold2 in modeling disordered proteins, we evaluated whether 5 × 200 ns simulations provide sufficient sampling. Specifically, we analyzed DSI values for each variant, averaged across five replicates, at 25 ns intervals (Figure S1). All systems demonstrate relatively stable average DSI values over time; for instance, Cel5G experiences a 1.4% change (41.9–41.3 Å) in DSI from 175 to 200 ns. These results support the adequacy of the simulation timescale for sampling large‐scale domain fluctuations and the relative ranking of interdomain separation in Cel5G and its variants. To illustrate relative domain motions, we show the time evolution and distribution of DSI at 10°C from representative 200 ns trajectories (Figure 2c). Taking Cel5G and Δ2L‐CA for comparison, which exhibit the largest and smallest mean DSI values, respectively, Cel5G involves a broad range of DSI values, fluctuating between 32 and 52 Å with an average of 42 Å, indicating the population of extended conformational states (Figure 2d). In contrast, Δ2L‐CA displays a narrow DSI range of 3–9 Å and a low average of 5 Å, demonstrating the population of compact conformational states (Figure 2d). Similar to the cold‐adapted amylase sdA (Ding et al. 2025), the enhanced interdomain separation in Cel5G prevents the two domains from interfering each other and improves CBM's accessibility to recruit polysaccharide substrate in solution (Ding et al. 2021; Ding et al. ²⁰²²). With improved substrate accessibility, Cel5G maintains high activity in cold environments with limited substrate and water mobility. While this trend reflects a distinct metric—k _cat at low temperature rather than relative activity—it complements our previously proposed “linker spacer effect” in cold‐adapted amylases, where increased domain separation was linked to enhanced cold adaptation (Ding et al. 2025).

FIGURE 2.

Comparative analysis of DSI and conformations in Cel5G and its linker‐modified variants (Δ1L, Δ2L, and Δ2L‐CA) at 10°C. (a) Schematic for the definition of domain separation index (DSI). (b) Correlation between DSI and k _cat values at 10°C. (c) Time evolution and distribution of DSI in a single 200 ns trajectory for in Cel5G and its linker‐modified variants at 10°C. (d) Structural overlay of typical snapshots for Cel5G (blue), Δ1L (green), Δ2L (orange), and Δ2L‐CA (red) at 10°C. Colored portions of Cel5G correspond to regions that are deleted in Δ1L (green) and Δ2L/Δ2L‐CA (orange).

2.2. Structural basis underlying linker spacer effect

To investigate the effect of linker length and linker topology on domain separation, we constructed cysteine‐to‐alanine variants of Cel5G and Δ1L to eliminate their disulfide‐bonded loops: Cel5G‐CA (C314A, C328A, C345A, C359A, C382A, C396A) and Δ1L‐CA (C314A, C328A, C351A, C365A) (Table S1). Together with the previously described Δ2L‐CA, these constructs allow direct comparisons between linker architectures with and without disulfide‐bonded loops at three different linker lengths: Cel5G vs. Cel5G‐CA (109 aa), Δ1L vs. Δ1L‐CA (78 aa), and Δ2L vs. Δ2L‐CA (41 aa) (Figure 3a). In this set, Cel5G, Δ1L, and Δ2L contain 3, 2, and 1 disulfide loops, respectively, while their CA counterparts contain only linear peptide linkers.

FIGURE 3.

Association of domain separation and H‐bond formation at the CD‐CBM interface. (a) Effects of linker length and disulfide‐bond loop presence on domain separation. (b) Frequency distribution of the number of interface H‐bonds in Cel5G, Δ1L, Δ2L, and Δ2L‐CA. (c) Effects of DSI on the frequency and number of H‐bonds formed at the CD‐CBM interface. (d) Frequency of interface H‐bond formation in Cel5G and its linker‐truncated variants (Δ1L, Δ2L, and Δ2L‐CA) at 10°C.

We conducted 5 × 200 ns MD simulations for Cel5G‐CA and Δ1L‐CA and compared the average DSI values for all six Cel5G variants (Cel5G, Cel5G‐CA, Δ1L, Δ1L‐CA, Δ2L, and Δ2L‐CA) (Figure 3a and Table S4). Despite having the same linker length, Cel5G‐CA's average DSI is lower than Cel5G's by 32.9 Å. Similarly, Δ1L‐CA's average DSI is 14.3 Å lower than that of Δ1L, and Δ2L‐CA's average DSI is 2.1 Å lower than that of Δ2L. These results indicate that variants with disulfide‐bonded loops as a “spacer” exhibit significantly greater DSI than their linear‐peptide counterparts (Figure 3a). In comparing Cel5G, ∆1L, and ∆2L, the presence of additional loops amplifies the “linker spacer effect” of the linker. Contrary to the common belief that a longer linker better separates domains, the results reveal no correlation between linker length and domain separation. For example, Cel5G‐CA, which has the longest linker (109 residues) shows a lower average DSI than Δ1L‐CA (Cel5G‐CA: 8.4 ± 1.7 Å vs. Δ1L‐CA: 15.0 ± 6.8 Å). Cel5G‐CA has a similar DSI to Δ2L‐CA (6.1 ± 1.7 Å), even though the latter's linker is 68 residues shorter. Previous research on Cel5G suggests that its enhanced cold adaptation was primarily due to its long linker (Sonan et al. 2007). However, our findings demonstrate that the presence of disulfide‐bond loops within the linker, rather than the linker length, plays a dominant role in spacing CD and CBM to induce cold adaptation. Notably, this disulfide‐bonded spacer represents an additional structural feature, beyond the helical linker motif identified in cold‐adapted amylases in our previous study (Ding et al. 2025), that contributes to enhanced domain separation. Both studies suggest a new approach for engineering bidomain enzymes: instead of modifying the active site loops, engineering the linker—a structural component that is presumably indirect to the catalytic function—can enhance cold adaptation by domain separation.

Although the linker spacer effect is observed in cellulase Cel5G and its variants, the structural basis explaining how the linker mediates domain separation remains elusive. We hypothesize that Cel5G variants with smaller DSI are more likely to form interdomain interactions, which restrict the relative motion between domains and consequently reduce the global flexibility of these variants. To investigate the interdomain interactions of Cel5G, Δ1L, Δ2L, and Δ2L‐CA, we analyzed the H‐bonding interactions at the interface between CD and CBM, including: (1) the H‐bond frequency, defined as the percentage of snapshots showing at least one formed H‐bond (donor‐acceptor distance cutoff: ≤3 Å; donor–hydrogen–acceptor angle cutoff: ≥135°) out of a total of 10,000 snapshots across five replicas (Figure 3b), and (2) the distribution of H‐bond numbers within these snapshots (Figure 3c). Consistent with our hypothesis, Cel5G and Δ1L, the two variants with relatively large DSI values, exhibited <5% interface H‐bond frequencies (Figure 3b). In contrast, over 65% of snapshots in both Δ2L and Δ2L‐CA contain at least one interface H‐bond (Figure 3b). Regarding the H‐bond number distribution, all variants predominantly form single interface H‐bonds (Figure 3c and Table S5), with the frequency being 2.6%, 26.7%, and 34.4% for Δ1L, Δ2L, and Δ2L‐CA, respectively. Consistent with this trend, the frequency of forming two interface H‐bonds is 1.7%, 23.3%, and 24.4% for Δ1L, Δ2L, and Δ2L‐CA, respectively. As a minor population, Δ2L and Δ2L‐CA exhibit similar frequencies of forming three (13.8% vs. 8.8%) and four (4.1% vs. 2.0%) interface H‐bonds (Table S6). By comparison, Cel5G variants with smaller DSI exhibit a higher tendency of forming more H‐bonds at the CD‐CBM interface, thereby stabilizing the interdomain interactions and favoring compact conformational states that impair enzymatic activities at lower temperatures.

Maintaining a high DSI is critical in mitigating interdomain interactions and maintaining cold adaptation for bidomain enzymes. To investigate the threshold of minimal DSI required to prevent interface H‐bond formation in Cel5G variants, we performed statistical analysis of both the frequency and number of interface H‐bonds across Cel5G, Δ1L, Δ2L, and Δ2L‐CA, using 40,000 snapshots (2000 snapshots × 5 replicas × 4 variants) (Figure 3d). The interdomain H‐bond frequency reaches its peak (i.e., 89.5%) when the DSI is between 2 and 4 Å, involving approximately 2.2 H‐bond pairs (Figure 3d). When DSI exceeds 4 Å, both the frequency and number of interface H‐bonds decrease with increasing DSI. Interface H‐bonds rarely form (less than 1.2%) when DSI exceeds 14 Å, and none forms when DSI exceeds 20 Å (Figure 3d). For Cel5G, 95% of the snapshots exhibit a DSI greater than 20 Å (average DSI: 41.3 Å; Figure S2), resulting in the absence of interface H‐bonds (Table S5). For Δ1L (average DSI of 29.4 Å), 12.8% of the snapshots show a DSI less than 20 Å (Figure S2), resulting in interface H‐bonds in 5% of snapshots (Table S5). These findings suggest that when designing cold‐adapted bidomain enzymes, both the mean DSI value and its distribution should be considered. Although this statistical analysis is based on Cel5G variants, this DSI threshold (20 Å) can serve as a general reference for bidomain enzymes as DSI is independent of the size of each functional domain (Figure 2a). Altogether, these results reveal new potential avenues for enhancing cold adaptation through rationally tuning linker topology. In addition to disulfide‐bonded loops, alternative structural motifs may also promote interdomain separation. For example, helical linker elements, which have previously been associated with increased domain separation in cold‐adapted amylases (Ding et al. 2025), could be explored in cellulases as well. Similarly, incorporating some elements of rigid synthetic linkers—such as polyproline linkers (Arora et al. 2002)—may offer novel avenues for optimizing the cold activity of bidomain enzymes. Through such targeted linker engineering, domain mobility and low temperature catalytic performance could possibly be fine‐tuned, expanding the scope of cold‐adapted enzymes for industrial applications.

2.3. Reduced interdomain interactions destabilize active site hydrogen bonds

Cold‐adapted single‐domain enzymes often have a reduced number of H‐bonds at active sites to maintain conformational flexibility and high activity at low temperatures (Liu et al. 2023). To investigate whether this mechanism applies to bidomain enzymes, we analyzed the H‐bond formation frequency at E222, an active‐site nucleophile (Violot et al. 2005), in Cel5G, Δ1L, Δ2L, and Δ2L‐CA (Figure 4a). Analysis of 10,000 snapshots (2000 × 5) from 200‐ns MD simulations of the four variants shows that E222 forms H‐bonds with five residues in CD: R57, H100, N134, F195, and Y196 (Figure 4b). Among all these H‐bonds, the most significant changes across the four variants were observed in the interaction between E222 and N134 (Figure 4a). In Cel5G, the most cold‐adapted variant, this H‐bond forms at a frequency of merely 7%, meaning that both residues bear high flexibility. In Δ1L and Δ2L, this H‐bond occurs more frequently, at 40% and 34%, respectively. Furthermore, in Δ2L‐CA, the variant with the lowest cold activity, the H‐bond forms at a very high frequency of 81%, indicating strong stability of this interaction (Figure 4a). We also analyzed Cel5G‐CD, which contains only the CD of Cel5G without the linker and CBM. Interestingly, the H‐bond formation frequency between E222 and N134 in Cel5G‐CD (78%) is similar to that in Δ2L‐CA (81%) (Figure 4a). Both Δ2L‐CA and Cel5G‐CD show low k _cat values (8 ± 0.02 and 6 ± 0.02 s⁻¹, respectively) at 10°C, compared to other Cel5G variants (Sonan et al. 2007). These values support an inverse relationship between the frequency of the active site H‐bond between E222 and N134 and catalytic activity at low temperature. E222 serves as a key catalytic base, facilitating cellulose hydrolysis by activating a water molecule, which then attacks the glycosidic bond in cellulose to initiate cleavage (Figure 4b) (Violot et al. 2005). Thus, we hypothesize that the increased participation of E222 in residue‐residue H‐bonding may stabilize the charge of the carboxylate, lowering its pK_a and diminishing its ability to act as an effective specific base.

FIGURE 4.

Active site H‐bond stability. (a) Frequency of individual H‐bonds involving the active site E222 in Cel5G, Δ1L, Δ2L, and Δ2L‐CA. (b) Schematic representation of residues involved in H‐bond formation at the active site E222 in Δ2L‐CA, and the contact between cellobiose and E222 mediated by H‐bonds with a water molecule.

Comparing the average pK_a values of residue E222 across five trajectories, the pK_a of Cel5G is 0.33 units higher than that of Δ2L‐CA (Figure 5a). A larger pK_a is beneficial for the catalysis of hydrolysis reactions, since a drop in the pKa of a residue can reduce its ability to deprotonate water (Geronimo et al. 2018). In Cel5G, E222 also forms an average of 1.96 hydrogen bonds with surrounding water molecules, whereas it forms only 1.79 in Δ2L‐CA (Figure 5b). After examining 40,000 snapshots (2000 snapshots × 5 replicas × 4 variants Cel5G, Δ1L, Δ2L, and Δ2L‐CA), we found that the presence of a hydrogen bond between E222 and N134 reduces the pK_a of E222 by 0.31 units relative to snapshots lacking this interaction (Figure 5c). The greater frequency of H‐bonding between E222 and N134 in Δ2L‐CA may be responsible for the observed reduction in the ability of E222 to directly interact with its surrounding water molecules, resulting in decreased activity. By minimizing E222 H‐bonding interactions, Cel5G retains both the appropriate pK_a and the capacity to engage effectively with water, enhancing its cold activity. These results show that besides serving as a spacer, linkers in bidomain enzymes can also modulate the active site by fine‐tuning the frequency of H‐bond interactions to influence catalytic residues' capability of binding or reacting to the substrate. This resembles allosteric regulation in enzymes, even though the active‐site changes arise from linker repositioning rather than from effector binding.

FIGURE 5.

Effects of linker region, active‐site H‐bonding, and DSI on the pK_a of E222. Error bars represent standard error. (a) pK_a of E222 in Cel5G and Δ2L‐CA. (b) Average number of H‐bonds between E222 and water in Cel5G and Δ2L‐CA. (c) pK_a of E222 in frames with versus without an H‐bond between E222 and other residues.

How do linkers affect active site H‐bond formation from a distance of 55 Å (measured between E222 and P349, the residue in the middle of Cel5G linker sequence)? To address this question, we used the Floyd‐Warshall algorithm (Floyd 1962; Warshall ¹⁹⁶²) to analyze the shortest H‐bond pathways (those involving the minimum number of residues, with all H‐bond frequencies on the path above 2%) from N206 to E222 in Δ2L, and from S203 to E222 in Δ2L‐CA, as N206 and S203 are involved in the most frequent interface H‐bonds (frequencies above 30%; see Figure S3), and compared these pathways with those in Cel5G (Figures 6 and Table S7). In Δ2L, the shortest H‐bond pathway connecting N206 to E222 consists of nine bonds spanning ~57 Å, compared to Cel5G, where the same pathway involves seven bonds over ~52 Å. The shortest pathway from S203 to E222 in Δ2L‐CA covers six bonds across ~42 Å, whereas in Cel5G, it spans eight bonds over ~63 Å. Notably, the Δ2L and Δ2L‐CA pathways include relatively frequent H‐bonds involving Q174, H194, S173, and N134—residues absent in the corresponding Cel5G pathways (Table S7). The presence of the N134‐E222 H‐bond in the shortest pathways for Δ2L and Δ2L‐CA supports the significance of distal interface H‐bonds, such as Q367‐N206 in Δ2L and N369‐S203 in Δ2L‐CA, on active‐site interactions. Thus, linker‐induced changes in interface H‐bonding are propagated to the active site, modulating cold activity. These results emphasize the structural basis of the “linker spacer effect” in mediating a dynamics‐driven tuning mechanism.

FIGURE 6.

Shortest H‐bond pathways in Cel5G, Δ2L, and Δ2L‐CA. (a) Pathways in Cel5G and Δ2L from surface residue N206 to active site E222. (b) Pathways in Cel5G and Δ2L‐CA from surface residue S203 to active site E222. The shortest H‐bond pathways were calculated using Floyd‐Warshall algorithm (Floyd 1962; Warshall ¹⁹⁶²), where a shortest pathway is defined as one involving the minimum number of residues, with all H‐bond frequencies above 2%. The dashed lines represent H‐bonds, with colors ranging from red (high frequency) to blue (low frequency).

3. CONCLUSION

In this study, we investigated the structural and dynamic underpinnings of cold adaptation in bidomain cellulases, using Cel5G and its linker‐modified variants as model enzymes. Our results revealed that the domain separation index positively correlates with catalytic efficiency at low temperatures, highlighting the importance of linker‐mediated domain separation for cold adaptation. Variants with disulfide‐bonded loops in the linker demonstrate enhanced interdomain separation, reduced interdomain hydrogen bonding, and improved substrate accessibility, all of which contribute to higher cold adaptation. Contrary to the conventional belief that longer linkers inherently promote domain separation, we found that the topology of the linker, specifically the presence of disulfide‐bonded loops, plays a dominant role. Additionally, our findings suggest that linkers not only act as spacers but also modulate catalytic activity by altering the conformational landscape of the enzyme, influencing active site flexibility and hydrogen bonding.

This study parallels our observations in bidomain amylases, supporting the “linker spacer effect” as an engineering strategy for cold adaptation in bidomain enzymes. These principles could advance the design of enzymes for industrial applications, particularly in sustainable biofuel production, by reducing the energy costs associated with cellulose saccharification at low temperatures.

4. MATERIALS AND METHODS

4.1. Protein structure and preparation

This study investigated Cel5G and five of its truncated variants: Cel5G‐CA, Δ1L, Δ1L‐CA, Δ2L, and Δ2L‐CA (Table S1). AlphaFold2 on the ColabFold server (version 1.5.3) was used to predict the structures of the enzymes (see sequences in Data S2) (Jumper et al. 2021; Mirdita et al. ²⁰²²). The pLDDT scores are provided in Data S3. We compared the structural similarity of Cel5G's CD crystal structure (PDB ID: 1TVN) (Violot et al. 2005) to the AlphaFold2‐predicted catalytic domains using the align command in PyMOL. The backbone RMSD indicated strong agreement between the AlphaFold2‐predicted and crystallographic CD structures across all variants (Table S8). The AMBER19 tleap tool was used to prepare the predicted structures for MD simulations (Case et al. 2019). The AMBER ff14SB force field was applied (Maier et al. 2015). All missing atoms were added through tleap, and the positions of hydrogens and protonation states of titratable residues in the proteins were determined using the get_protonation() method from EnzyHTP (Shao et al. 2022), supported by PDB2PQR and PROPKA3 (Dolinsky et al. 2007; Olsson et al. ²⁰¹¹; Yang et al. ²⁰²³). Disulfide bonds were identified by PDB2PQR and established with the bond command in tleap, also facilitated by EnzyHTP automation (Table S2).

4.2. Molecular dynamics simulations

MD simulations for Cel5G and its variants were performed using the AMBER19 software package (Case et al. 2019). All hydrogen‐containing bonds were constrained using the SHAKE algorithm (Ryckaert et al. 1977). Each enzyme was solvated in a periodic truncated octahedron box with a 10 Å buffer of TIP3P water, and Na⁺ counterions were added to neutralize the system. The whole solvent box was relaxed by performing 10,000 steps of the steepest descent method followed by 10,000 steps using the conjugate gradient method. After minimization, heating commenced from 0 K to the target temperature of 283.15 K over 36 ps in a constant volume environment, followed by an equilibration phase of 4 ps under NVT conditions (constant number of particles N, volume V, and temperature T) and an additional 1 ns equilibration under NVT. The temperature was regulated by the Langevin thermostat, and restraints were applied on the backbone C_α, C, and N of the amide group (2 kcal mol⁻¹ Å⁻²) from minimization through equilibration (Loncharich et al. 1992). Production runs were subsequently conducted under constant volume and temperature conditions for 210 ns, with snapshots recorded every 100 ps. For each enzyme variant, five independent MD runs were performed, and analyses focused on the last 200 ns of each production run, resulting in a total of 2000 snapshots per run. The simulations adopt a 2‐fs time step.

4.3. Molecular dynamics trajectory analysis

Domain distance, radius of gyration, DSI (Data S3), root‐mean‐square deviation (RMSD; Data S4), and H‐bonds (Data S5) were computed from the MD trajectory data using the cpptraj tool in AMBER (Roe and Cheatham III ²⁰¹³). The mean values for each parameter were calculated based on 5 × 200 ns MD trajectories (including 5 × 2000 snapshots). The categorization for the CD, linker, and CBM was conducted by referring to Sonan et al. (2007) based on sequence alignment (Data S2). Domain distance was determined based on the distance between the geometric centers of each domain, and the radius of gyration was calculated for the entire structure (R _g), C‐terminal domain ($R_{\mathrm{g}}^{\mathrm{C}}$), and N‐terminal domain ($R_{\mathrm{g}}^{\mathrm{N}}$), with all involved heavy atoms considered. DSI was calculated as the difference between the domain distance and the sum of the $R_{\mathrm{g}}^{\mathrm{N}}$ and $R_{\mathrm{g}}^{\mathrm{C}}$ values: DSI = domain distance − ($R_{\mathrm{g}}^{\mathrm{N}}+R_{\mathrm{g}}^{\mathrm{C}}$). RMSD analysis was mass‐weighted (backbone heavy atoms, including Cα and amide group atoms), using the centroid structure of the most populated cluster as a reference. Clustering was performed via cpptraj using average‐linkage clustering with the number of clusters set to 10 (Tables S9–S12). H‐bonds were computed with an angle cutoff of ≥135° and a distance cutoff of ≤3 Å. Interface H‐bonds were defined as H‐bonds having a donor and acceptor from different domains. pK_a values were calculated for each snapshot using PROPKA (version 3.5.0) under default settings. Shortest H‐bond pathways were identified in the network where each residue was a node and H‐bond frequency was the edge weight between nodes using the Floyd–Warshall algorithm in the Python package NetworkX (Floyd 1962; Hagberg et al. ²⁰⁰⁸; Warshall ¹⁹⁶²).

AUTHOR CONTRIBUTIONS

Robbie Ge: Data curation; formal analysis; writing – original draft; writing – review and editing; visualization; investigation. Ning Ding: Conceptualization; methodology; investigation; formal analysis; supervision; visualization; project administration; writing – review and editing. Yaoyukun Jiang: Methodology; software; supervision. Zhongyue J. Yang: Writing – review and editing; conceptualization; investigation; supervision; funding acquisition; project administration.

Supporting information

Data S1. Supplementary Information.

Data S2. Supplementary Information.

Data S3. Data sheet.

Data S4. Data sheet.

Data S5. Data sheet.

Data S6. Data sheet.

Ge R, Ding N, Jiang Y, Yang ZJ. Linker‐mediated domain separation enhances cold adaptation in cellulases. Protein Science. 2025;34(8):e70198. 10.1002/pro.70198

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in Zenodo at https://zenodo.org/records/14206281, reference number 14206281.