Product binding varies dramatically between processive and nonprocessive cellulase enzymes

Background: Experimentally measured product inhibition constants of cellobiose for cellulases vary significantly. Results: Cellodextrin bound as substrate to cellulases increases product binding to processive cellulases, whereas it does not affect nonprocessive cellulases. Conclusions: The increased binding affinity correlates with hydrogen bonds between the substrate and cellobiose in processive cellulase tunnels. Significance: The results offer an interpretation for the discrepancy in measured inhibition constants. Abstract Cellulases hydrolyze β -1,4 glycosidic linkages in cellulose, which are among the most prevalent and stable bonds in Nature. Cellulases comprise many glycoside hydrolase (GH) families, and exist as processive or nonprocessive enzymes. Product inhibition negatively impacts cellulase action, but experimental measurements of product binding constants vary significantly, and there is little consensus on the importance of this phenomenon. To provide molecular-level insights into cellulase product inhibition, we examine the impact of product binding on processive and nonprocessive cellulases by calculating the binding free energy of cellobiose to the product sites of catalytic domains (CDs) of processive and nonprocessive enzymes from GH Families 6 and 7. The results suggest that cellobiose binds to processive cellulases much more strongly than nonprocessive cellulases. We also predict that the presence of a cellodextrin bound in the reactant site of the CD, which is present during enzymatic catalysis, has no effect on product binding in nonprocessive cellulases, whereas it significantly increases product binding to processive cellulases. This difference in product binding correlates with hydrogen bonding between the substrate-side ligand and the cellobiose product in processive cellulase tunnels and the additional stabilization from the longer tunnel-forming loops. The hydrogen bonds between the substrate- and product-side ligands are disrupted by water in nonprocessive cellulase clefts, and the lack of long tunnel-forming loops results in lower affinity of the product ligand. These findings provide new insights into the large discrepancies reported for binding constants for cellulases, and suggest that product inhibition will vary significantly based on the amount of productive binding for processive cellulases on cellulose.

product binding on processive and nonprocessive cellulases by calculating the binding free energy of cellobiose to the product sites of catalytic domains (CDs) of processive and nonprocessive enzymes from GH Families 6 and 7. The results suggest that cellobiose binds to processive cellulases much more strongly than nonprocessive cellulases. We also predict that the presence of a cellodextrin bound in the reactant site of the CD, which is present during enzymatic catalysis, has no effect on product binding in nonprocessive cellulases, whereas it significantly increases product binding to processive cellulases. This difference in product binding correlates with hydrogen bonding between the substrate-side ligand and the cellobiose product in processive cellulase tunnels and the additional stabilization from the longer tunnelforming loops. The hydrogen bonds between the substrate-and product-side ligands are disrupted by water in nonprocessive cellulase clefts, and the lack of long tunnel-forming loops results in lower affinity of the product ligand. These findings provide new insights into the large discrepancies reported for binding constants for cellulases, and suggest that product inhibition will vary significantly based on the amount of productive binding for processive cellulases on cellulose.
Bacteria and fungi secrete enzyme cocktails consisting of processive and nonprocessive enzymes, which act synergistically to digest cellulose (1,2). Nonprocessive cellulases (endoglucanases) cleave cellulose chains randomly, creating free ends. Processive cellulases (cellobiohydrolases or CBHs) acquire the cellulose chains at free ends and processively cleave cellobiose units from either end. Cellobiose is further hydrolyzed to glucose in solution by β-Dglucosidases. Inhibition of cellulases by cellobiose constitutes a major obstacle for achieving high product yields in the enzymatic hydrolysis of cellulose (3,4), and thus significantly impacts the efficiency of biomass conversion (5,6). Many studies have been designed to quantify cellulase product inhibition (7-13), but measured inhibition constants vary by orders of magnitude depending on factors such as temperature, pH, and substrate (3).
Molecular simulation can offer insights into cellulase action as a complement to experiments (14,15). Recently we used molecular simulations to identify new functions of cellulase sub-domains (16)(17)(18)(19) and suggest a new microkinetic model for processive enzyme action (20,21). In a related previous study, we calculated the binding free energy of cellobiose and glucose to the product site of the catalytic tunnel of a GH Family 7 (GH7) CBH from the fungus Trichoderma reesei (Hypocrea jecorina) using steered molecular dynamics (SMD) (22). We identified targets for protein engineering to decrease product binding (22). However, two outstanding questions regarding product binding are how cellobiose binding to the enzymes affects processive cellulases relative to nonprocessive cellulases, and how product inhibition varies with productive binding of both types of cellulases. In this study, we use SMD to examine the difference in cellobiose binding to the product site in processive and nonprocessive cellulases with and without the enzymes being catalytically engaged.
We have chosen to examine cellobiose binding in processive and nonprocessive enzymes from the GH7 and GH6 families. These families are among the two most important cellulase families produced by fungi as they provide significant and synergistic hydrolytic potential for cellulose deconstruction (23). Specifically, we study the GH7 CBH Cel7A, the GH6 CBH Cel6A, and GH7 endoglucanase Cel7B from T. reesei, which are the three most abundant enzymes in the native T. reesei secretome (24). We also examine the GH6 endoglucanase Cel6B from the fungus Humicola insolens. Cel7A processively hydrolyzes the cellulose chain from the reducing end via a retaining mechanism to produce β-cellobiose, whereas Cel6A preferentially acts from the nonreducing end of the chain via an inverting mechanism to produce α-cellobiose (25)(26)(27). Cel7B and Cel6B are corresponding GH7 and GH6 non-processive cellulases, respectively. We note that H. insolens Cel6B has experimentally been shown to be an endoglucanase because it hydrolyzes amorphous and soluble substrates such as carboxymethyl cellulose (CMC) and phosphoric-acid-swollen cellulose efficiently, whereas it cannot digest crystalline cellulose (28,29).

Results
The system setup is detailed in Supporting Information. The methods closely follow our previous work (22) wherein we define a reaction coordinate for pulling the cellobiose ligand from the enzyme (Figure 1). Here we report the reversible thermodynamic work required to remove cellobiose from the product side of the CD tunnels, i.e. the computed work is the negative cellobiose binding free energy. The loops forming the binding cleft in Cel6B ( Figure 1D) are open in the PDB structure (1DYS). Comparison with other GH6 structures indicates that the loops may close to form a tunnel when Cel6B is catalytically engaged (30)(31)(32), i.e. when a cellodextrin chain is bound at the active site prior to hydrolysis. However, preliminary results of the simulations starting with a closed structure (as detailed in SI), with the substrate as a cellotetraose (reactant side) and a cellobiose (product side), show the loops open at 10 to 40 ns. Whether or not a cellohexaose substrate can maintain the loops closed before the hydrolysis reaction occurs is beyond the scope of this study. Thus, SMD simulations on Cel6B were conducted starting with the opened PDB structures. Figure 2 shows the potential of mean force (PMF) of cellobiose expulsion from the CD of Cel7A, Cel7B, Cel6A, and Cel6B. We note that the PMF of cellobiose expulsion from Cel7A with a cellodextrin at the reactant site has been published previously (22). The work profiles for the SMD simulations are shown in Figure S2 in Supporting Information. The free energy changes and association constants (K) are summarized in Table  1. The product inhibition constant K I in this work is defined as the equilibrium dissociation constant for reaction E + P → EP. Here E represents the enzyme and P is the product, cellobiose. The product inhibition constant is defined as where K is the equilibrium association constant.
Here we note that we only consider product binding as one form of inhibition. Other contributions that will also affect product inhibition are not taken into account in this study.
Overall, our results demonstrate that the binding free energy of cellobiose to the product binding site of both CBHs is significantly larger than that of either of the endoglucanases within the same GH family, suggesting generally that cellobiose inhibits CBHs more strongly than endoglucanases within the same GH family. These results are qualitatively consistent with experimental studies suggesting cellulose hydrolysis by endoglucanases from the same GH families is less affected by product inhibition than CBHs (9,33). We stress that the quantities in Table 1 are only free energies of cellobiose binding to the product site, such that primarily the relative free energies should be considered. Product inhibition also depends on the relation between the concentrations of product and substrate relative to the rate constants for their association to, and dissociation from, the catalytic center of the enzyme. Additionally, experimental measurements of inhibition constants include other factors such as cellobiose binding to the unfilled reactant sites of cellulases and ligand binding in alternate conformations. Thus, the quantities computed here are not directly equivalent to apparent inhibition constants.
Our results also show a cellodextrin chain bound at the reactant site of the CD has a substantially different effect on the binding free energy of cellobiose to CBHs and endoglucanases. Whereas the bound cellodextrin ligand at the reactant site has no impact on cellobiose binding to the endoglucanases, it increases the binding free energy of cellobiose to CBHs dramatically. Van Tilbeurgh et. al. demonstrated that the binding constant for β-D-cellobiose to Cel6A was increased by an order of magnitude in the presence of 0.33 M glucose (34). They hypothesize that the glucose binding in one subsite can induce a conformational change in the adjacent binding subsites, which subsequently increases the binding affinity for cellobiose. Our observations lend support to this hypothesis in that if glucose occupies the +1 site of Cel6A, this may enable the formation of a stable hydrogen bond with the cellobiose bound at the -2 and -1 sites, or vice versa. The change in cellobiose binding affinity for Cel6A is 8.5 kcal/mol and 3.5 kcal/mol for Cel7A, which results in the calculated association constant differing by several orders of magnitude, as shown in Table 1. This result suggests a difference in the product binding when the CBHs are productively bound on cellulose relative to not productively bound.
One potential mechanism for the increased binding in CBHs is that both the tunnel loops and the reactant side ligand in the CBHs stabilize cellobiose more than the open cleft and putatively more flexible reactant side ligand in the endoglucanases. We tested this hypothesis by comparing the cellobiose interactions with water, the enzyme (shown in Table S2), and the reactant side ligand in processive and nonprocessive enzymes. Figure 3 shows the solvation of the ligands in all four enzymes. In Cel7A and Cel6A, only a few water molecules contact the ligand. From Table 2, hydrogen bonds exist between the cellobiose at the product site and the cellodextrin ligand in the reactant site, which may contribute to the increased binding affinity of cellobiose in processive enzymes. In contrast, the binding clefts in Cel7B and Cel6B are more accessible to water, such that water can diffuse into the cleft and solvate both ligands, thus disrupting the hydrogen bonds between the substrate and product ligands. These structural observations are consistent with the proposed mechanism of greater stabilization in CBHs by increased interaction with protein tunnel walls and the reactant ligand.
We also examined the interaction of water with the cellobiose ligand. Table 3 shows that the cellobiose ligand in the nonprocessive enzymes (Cel7B and Cel6B) exhibits increased interaction energy with water compared to the processive enzymes, due to the geometry of the binding site (clefts vs. tunnels) in the two classes of enzymes.
Because cellobiose is a mixture of roughly 1/3 αand 2/3 β-anomers at equilibrium, experimentally measured product inhibition is a combined effect of both anomers. To investigate the binding affinity of the other anomer rather than the native products (α-anomer to GH6 cellulases and βanomer to GH7 cellulases), we also calculated the relative binding free energy of β-anomer to Cel6A with a cellodextrin chain bound at the reactant site using thermodynamic integration (TI). The calculated ΔΔG is -0.4±0.2 kcal/mol, indicating no significant difference between binding affinities of the two anomers. We thus assume the anomer effect will be insignificant in Cel7A, since the anomeric carbon faces the solvent, and in the endoglucanases where the ligand is more highly solvated.
Finally, we note that our results shown in Figure 2 suggest stronger product binding in Cel6A than Cel7A. Claeyssens et al. conducted binding experiments with Cel7A and Cel6A using the soluble substrates cellobiose and cellotriose (35). They found that cellobiose binds much stronger to Cel7A compared to Cel6A. However, the number of binding sites in these two cellulases is different. There are at least 9 binding sites (from -7 to +2) in Cel7A tunnel and only 6 binding sites (from +4 to -2) in the Cel6A tunnel. To our knowledge, this experimental approach cannot distinguish the specific binding sites of cellobiose occupied in the tunnel. On the other hand, our simulations only focus on the binding of cellobiose at the product site, i.e. the +1 and +2 sites in Cel7A and -2 and -1 sites in Cel6A. Thus, we note that the binding free energies of cellobiose to cellulases calculated here should not be directly compared with experimental binding studies between cellulases with significantly different numbers of binding sites.

Discussion
There are two distinct consequences of cellobiose affinity at the enzyme product sites, namely product expulsion and product inhibition. Our results have implications for both. The large increase in cellobiose affinity to the CBHs with cellodextrin bound suggests that product expulsion will be slow when the enzymes act processively on cellulose, which suggests that product expulsion may be a rate-limiting factor in the processive cycle. With Cel7A, we note that an 80-fold higher apparent inhibition constant with cellulose as substrate, compared to that on chromogenic disaccharides, indicates weaker binding of cellobiose during processive action, assuming that the extent of productive binding is equal between the two experiments (9). Our results suggest that productive binding is thus a key quantity in inhibition measurements. If experimental measurements conducted at the same level of productive binding exhibit lower inhibition on cellulose than on chromogenic substrates, this lends support to the hypothesis of an active product expulsion mechanism in GH7 CBHs proposed by Ubhayasekera (36).
In this mechanism, the chemical energy released upon hydrolysis can be converted into kinetic energy released upon hydrolysis of the glycosyl-enzyme intermediate, which may be utilized to expel cellobiose from the product site. A similar coupling of product expulsion to catalysis is not evident in the inverting mechanism of Cel6A, so in this case slow cellobiose release may play a significant role.
While product expulsion only depends on the desorption rate for cellobiose, product inhibition is the combined result of association and dissociation rates for different binding modes of both substrate and product, binding alone and together. Figure 4, inspired from reference (9), illustrates a simple putative mechanism for product inhibition of a processive cellulase hydrolyzing a cellulose chain. The calculated differences in free energy due to differences between productively bound and nonproductively bound enzymes shown in Table 1 result in apparent inhibition constants that vary over 3 orders of magnitude. Therefore, productive binding should be accounted for when measuring the apparent inhibition constants for processive cellulases. The large discrepancy of the experimentally measured apparent inhibition constants, that range over several orders of magnitude, may be caused by the ratio of the free enzymes (E) and the catalytically engaged enzymes (ES), which depend on the substrate, the experimental conditions, and the relative amounts and types of endoglucanases and CBHs present in an enzyme cocktail. Furthermore, with Cel7A it has been experimentally demonstrated that the processive action often comes to a halt, and the enzyme remains nonproductively bound on the surface until it is eventually released and free to diffuse to a new point of initial attack (37,38). This led to the proposal that enzyme release may be a major ratelimiting factor for some processive cellulases. Our results, indicating much higher association constants of cellobiose to CBHs product sites with cellodextrin bound in the substrate side of the tunnel, imply that the release of the catalytic domain from a cellulose chain will be slower in the presence of cellobiose because of increased substrate stabilization, thus contributing to apparent product inhibition.

Conclusions
We characterized the binding affinity of cellobiose to the CD of two processive cellulases (Cel7A and Cel6A) and two nonprocessive cellulases (Cel7B and Cel6B). Our results demonstrate that cellobiose binds to CBHs more strongly than to endoglucanases. Our results also suggest that cellobiose binding can increase by several orders of magnitude when a processive cellulase is productively bound (ES'). These results thus offer an interpretation for the discrepancy in measured inhibition constants. We propose that catalytic engagement has an important effect on apparent inhibition constants for processive cellulases and should be taken into account to characterize product inhibition.

Methods
The CHARMM22 force field (39,40) with the CMAP (41-43) correction were used for the protein, and the C35 carbohydrate force field parameters (44,45) were used for the cellodextrin chain and the glycosylation. The system was generated via the CHARMM-GUI web server (46). Each of the four cellulases with the presence or the absence of the cellodextrin chain at the reactant site (total eight systems) was placed in an equilibrated truncated octahedral box of TIP3P water molecules (44,45). Water molecules that overlapped with the protein and carbohydrate heavy atoms were removed. To produce a neutral system, Na + ions were introduced by transforming the water molecules into ions randomly. Particle Mesh Ewald summation (47) was used for electrostatics with a 6 th order b-spline interpolation, a Gaussian distribution with a width of 0.34 Å, and a mesh size of 96 x 96 x 96. The nonbonded interaction cutoff was 12 Å. The covalent bonds to hydrogen atoms were fixed using SHAKE (48). The system was relaxed in the NPT ensemble at 300 K with a Nosé-Hoover thermostat (49,50) and 1 atm for 100 ps with a step size of 1 fs. CHAMBER (51) was used to convert the CHARMM protein structure file, coordinate file, and associated force field files to an AMBER topology file and coordinate file. The PMEMD engine in AMBER (52) was used to conduct MD simulations. Each system was simulated in the NVT ensemble for 10 ns at 300 K with a step size of 2 fs. Subsequently, 40 starting structures were selected from the last 8 ns of the 10 ns simulation as the initial structures of 40 SMD simulations, which were conducted using the Jarzynski implementation in Amber. During the SMD simulations, the distance between the two dummy atoms, shown in Fig. 1, was gradually increased with a speed of 1 Å/ns in 14 ns. The force constant was 5000 kcal/(mol·Å 2 ).
We used Jarzynski's equality ΔG = -kT ln〈exp(-W/kT)〉 to construct the potential of mean force (PMF) (53)(54)(55). By computing the work to drive the system from the bound state to the unbound state, and averaging the work exponentially, the binding free energy of cellobiose can be extracted. The uncertainty of SMD simulations was measured using the bootstrap method (56

Supporting Information
The details of the atomistic models and TI simulation protocols ( Figure S1; Table S1); the definition of the reaction coordinates in Figure 1; the accumulated work profiles for the 40 SMD simulations.          by guest on July 9, 2020 Table 2. The number of hydrogen bonds and interaction energy (in kcal/mol) between cellobiose and cellodextrin. This analysis is based on 50-ns MD simulations. The cellobiose left the product site at ~30 ns in the two nonprocessive enzymes with a cellodextrin bound in the reactant site, thus the data for the nonprocessive enzymes is all prior to the product leaving. A hydrogen bond cutoff of 3.5 Å and an angle criterion of 60° from linear were used.