Evolutionary coupling saturation mutagenesis: Coevolution-guided identification of distant sites influencing Bacillus naganoensis pullulanase activity

Abstract

Pullulanases are well-known debranching enzymes hydrolyzing α−1,6-glycosidic linkages. To date, engineering of pullulanase is mainly focused on catalytic pocket or domain tailoring based on structure/sequence information. Saturation mutagenesis-involved directed evolution is, however, limited by the low number of mutational sites compatible with combinatorial libraries of feasible size. Using Bacillus naganoensis pullulanase as a target protein, here we introduce the “evolutionary coupling saturation mutagenesis” (ECSM) approach: residue pair co-variances are calculated to identify residues for saturation mutagenesis, focusing directed evolution on residue pairs playing important roles in natural evolution. Evolutionary coupling (EC) analysis identified seven residue pairs as evolutionary mutational hotspots. Subsequent saturation mutagenesis yielded variants with enhanced catalytic activity. The functional pairs apparently represent distant sites affecting enzyme activity.

Introduction

Pullulanase (EC 3.2.1.41), generally with a large molecular weight and a multi-domain structure, is an important member of the glycoside hydrolase family [1]. Pullulanase specifically catalyzes the hydrolysis of α−1,6-glycosidic linkages of unmodified substrates and is widely used in medicine, materials, organometallic chemistry, and the saccharification industry [2]. However, the catalytic efficiency of pullulanase needs to be further improved for industrial application. To date, engineering of pullulanase is mainly focused on the catalytic pocket or domain tailoring based on structure/sequence information [3–8].

Directed evolution, a pioneering technology for engineering enzymes, represents an important approach to create enzymes with desired properties such as enhanced catalytic activities [9, 10]. To this end, random mutagenesis-based protein engineering methods, including error-prone polymerase chain reaction and DNA shuffling, have been widely used [9, 11–13]. This approach requires the construction and subsequent screening of large libraries containing mostly negative variants. To narrow the search in the mutation library and reduce screening efforts, structure information-guided saturation mutagenesis methods, such as triple code saturation mutagenesis [14], have been explored to evolve stereoselective and regioselective enzymes with optimized catalytic profiles [15]. In structure-based saturation mutagenesis methods, the amino acid residues selected for substitution are mostly confined to specific spatial regions in the protein such as an active center or substrate-binding domain, rather than potential sites outside of these functional regions, potentially involved in distal effects or residue-residue networks that can modulate enzyme functions [16–22]. This highlights the need for new approaches which expand the selection of mutational hotspots for saturation mutagenesis to the entire protein molecule, while keeping the mutation libraries focused to a comparative small number of residues.

Proteins from diverse organisms can be identified as homologues by sequence comparison, because evolutionary constraints limit substitutions at particular positions. The evolutionary record encoded in multiple sequence alignments of homologous proteins also provides information on residue pair co-variance. A long-standing goal of bioinformatics has been to utilize such sequence co-variation to provide information about residue pair functional correlations and/or contacts [23–32]. A key challenge in this analysis is created by transitive correlations, or relay effects; i.e., to distinguish A-B covariation due to A->B interactions from A-C covariation due to relayed A->B->C interactions. Recently, methods have been developed that can distinguish direct evolutionary couplings from such transitive correlations, allowing reliable analysis of direct evolutionary residue-residue couplings [33–38]. Such evolutionary couplings (ECs) can provide accurate information about residue pair contacts [33–35]. Generally, the highest scoring ECs are between residues that indeed contact one another in the 3D structure [33–35, 37, 39–47].

Global modeling approaches, which treat correlated pairs of residues as dependent on each other, generate high coupling scores only for residue pairs that are likely to be direct correlations, thereby minimizing transitive sequence covariance effects. Mean-field direct coupling analysis (DI), based purely on sequence information, computes residue couplings in a maximum entropy model [33–35, 38, 48]. The resulting couplings can be used to generate distance restraints for modeling 3D structures of proteins, RNA, and complexes [34, 35]. More recently, an alternative pseudo-likelihood maximization direct coupling analysis (PLM) has been described [36, 37]. PLM was shown to outperform DI on a set of protein-domain families with a large number of sequences [36]. Both the DI and PLM methods are implemented in the EVfold (mailto:http://www.EVfold.org/) contact prediction server [34], and the PLM method is implemented in the GREMLIN (mailto:http://gremlin.bakerlab.org/) contact prediction server [37].

So far, these algorithms, predicting residue-residue contacts from sequence covariation data, have been used mostly for protein and nucleic acid 3D structure prediction [34, 35, 37, 39–47], or for protein structure determination in combination with sparse experimental data [49], providing insights into protein structure-function relationships. In this regard, the strength of the inferred couplings, especially the top-scoring residue couplings, has been effectively used to predict co-evolving residue pairs and inferred residue-residue contacts, which have been applied in modeling 3D structures with remarkable accuracy. These co-varying residues include interactions in the protein structure which have been probed in the course of evolution.

An alternative application is to exploit ECs for the information they provide on the naturally-occurring co-variation of residues that has occurred in course of protein evolution, which could potentially reveal functional diversity that has been selected for by different evolutionary pressures.

Based on these properties of ECs, we hypothesized that coevolution analysis of protein sequence and prediction of evolutionary couplings could be effective for identifying productive contacts across the entire enzyme molecule for protein engineering, and to create a focused library [50] providing positive variants with desired properties, even for large proteins without known 3D structures. Here, using the industrially-important 926-residue pullulanase from Bacillus naganoensis as a target protein, we describe a new approach called “evolutionary coupling saturation mutagenesis” (ECSM), to engineer the enzyme guided by coevolution analysis. Although approaches using sequence co-variance to target sites for mutagenesis has been previously described [22, 28, 30], the recent advances in bioinformatic analysis outlined above provide more reliable identification of co-evolution due to direct residue contacts, minimizing co-variance due to transitive correlations, and enhancing the effectiveness of this strategy. Moreover, in order to validate the current view [51, 52] that co-evolution of two residues most often relies on a first mutation at one residue site, reducing fitness which followed, by a second mutation at a contacting site, resulting in increasing fitness, we performed saturation mutagenesis for the residues forming an evolutionary coupled pair in a sequential manner.

Results and Discussion

The ECSM method is schematically described in Fig. 1a. For target protein, the residue pairs for saturation mutagenesis are determined using improved methods for coevolution analysis [43, 53–55]. Then, one can construct libraries of saturation mutagenesis for selected EC pairs and screen them to select mutants showing good enzymatic performance. Based on this ECSM method, the target protein can be efficiently engineered towards a desired purpose.

Figure 1.

Evolutionary coupling saturation mutagenesis (ECSM) strategy (a) using pullulanase from Bacillus naganoensis (b) as a target protein. (a) Schematic representation of ECSM strategy. (b) Domain structure of pullulanase from B. naganoensis: CBM, carbohydrate-binding module; X, domain of unknown function.

To verify the feasibility and efficiency of the proposed ECSM method, pullulanase from Bacillus naganoensis (BnPUL), an enzyme with a large molecular weight (~101 kDa) and a multi-domain structure (Fig. 1b), was selected as a model system [6]. BnPUL specifically catalyzes the hydrolysis of α−1,6-glucosidic linkages of unmodified substrate and has been widely applied in drug formulation, processing of polymeric materials, organometallic chemistry, and industrial saccharification (i.e., carbohydrate degradation to soluble sugars) [56]. Previous efforts to engineer a homologous pullulanase through semi-rational design increased catalytic efficiency by 2 times [8]. Since BnPUL is a highly active naturally occurring enzyme, it represents a very challenging target for further improvement.

Prediction of evolutionary couplings and selection of mutation sites

The EVcouplings server (mailto:http://evfold.org) calculates ECs using a maximum entropy model [33, 57], constrained by the statistics of the multiple sequence alignment (MSA), to infer residue pair couplings [33, 34]. The EVcouplings server uses two alternative scoring methods to identify evolutionary coupled residues, the PLM and DI methods, which use different mathematical algorithms to infer the parameters of the maximum entropy model describing the couplings [34, 36]. For very large sequence alignments, PLM becomes asymptotically equivalent to full maximum-likelihood inference, whereas DI remains intrinsically approximate [34, 36]. Therefore, PLM is supposed to consistently outperform DI, as assessed across a number of large protein domain families [34, 36]. Thus, in this work, the EVcouplings-PLM method was used to generate ECs from sequences spanning individual-and sequential pairs of domains within the protein. In addition, EVcomplex (https://evcomplex.hms.harvard.edu), which is designed to calculate ECs between different domain families, was used to calculate additional inter-domain ECs within BnPUL.

For the catalytic domain alone (473–787), the polypeptide segment spanning the catalytic and CBM48 domains (315–787), and the polypeptide segment spanning the catalytic domain and the C-terminal segment (473–926), the EVcouplings server identified 47,650 (151 sequences/residue), 29,241 (61 sequences/residue), and 51,574 (113 sequences/residue) non-redundant sequences, respectively. For the combination of the CBM48 and catalytic domains without the linker sequence connecting them, the EVcomplex server identified 21,488 non-redundant sequences. The large number of sequences per residue used for these analyses promises to generate accurate evolutionary coupling information.

Tables 1–2 show residue pairs with scores larger than the significance threshold of 0.8 suggested by Hopf et al. [41], which served as candidates for coupled hotspots used for subsequent saturation mutagenesis. The identified pairs of ECs are summarized in Table 1 (calculated by EVcouplings) and Table 2 (calculated by EVcomplex). Seven residue pairs were selected from the top third of coupled sites, i.e., D614/H539, E530/T520, D541/D473, E777/T730, and K631/Q597 pairs with EVcouplings scores above 1.0, and V328/I565 and Y392/Y571 pairs with inter-domain EVcomplex scores above 1.0. Pairs within ten residues of a higher scoring pair were excluded. All of the pairs selected from the EVcouplings-PLM analysis also gave high-ranking EC scores using the GREMLIN [37, 43] server (Table S1). Table 3 (third column) provides the special distance in the 3D structure between co-varying residues for each seven pairs selected for saturation mutagenesis. As is usually observed for evolutionary coupled residues [45], six out of the seven residue pairs are in close spatial proximity (≤ 6 Å; Table 3). For three of those pairs, saturation mutagenesis yielded enzyme mutants with improved catalytic activity (Table 3; Fig. 2). Significantly, the one residue pair (Y392/Y571) separated by ~15 Å, corresponding to residues which are not in direct contact, did not yield any mutants with improved activity (Table 3).

Table 1.

Evolutionary couplings generated from analysis with EVcouplings (PLM score > 0.8).

Catalytic domain			CBM48 and catalytic domain				Catalytic domain and C-terminal segment
Residue 1	Residue 2	PLMa score	Residue 1	Residue 2	PLM score	Residue position	Residue 1	Residue 2	PLM score	Residue position
D614c	H539	1.4898	D541b	D473	1.267	Catalytic domain	D614	H539	1.3426	Catalytic domain
E610	K532	1.2174	D541	K476	0.98858	Catalytic domain	E530	T520	1.2768	Catalytic domain
E777	T730	1.0485	E530	T520	0.98639	Catalytic domain	E610	K532	1.0537	Catalytic domain
V545	L538	0.99965	E533	L455	0.98564	Catalytic domain/linker	K631	Q597	1.0308	Catalytic domain
K631	Q597	0.87149	E530	L455	0.90629	Catalytic domain/linker	Y814	Q761	0.94657	C-terminal segment/Catalytic domain
			K631	Q597	0.88181	Catalytic domain	V545	L538	0.93468	Catalytic domain
			K451	D443	0.85737	Linker	E777	T730	0.90179	Catalytic domain
			A364	M354	0.84801	CBM48	Y611	K532	0.90062	Catalytic domain
			E610	K532	0.83898	Catalytic domain	T861	D855	0.85811	C-terminal segment
			L538	V483	0.83182	Catalytic domain	Y851	T766	0.85629	C-terminal segment/Catalytic domain

^aPLM means pseudo-likelihood maximization.

^bResidue pairs in bold resulted in enzymes with improved catalytic activity (Fig. 3).

^cResidue pairs in italic were selected for saturation mutagenesis.

Table 2.

Evolutionary couplings generated from analysis using EVcomplex (score > 0.8).

Residue 1	Residue 2	PLMa score	EVcomplex score
V328bc	I565	0.624	3.140
V379	I565	0.438	2.204
Y392	Y571	0.373	1.880
T332	E568	0.303	1.526
W329	M595	0.280	1.408
T332	V566	0.274	1.380
V336	I565	0.257	1.292
R386	Q558	0.255	1.282
T380	E568	0.215	1.084
Y392	D602	0.205	1.032
A388	I565	0.203	1.024
T332	Y569	0.173	0.870
S384	Q558	0.172	0.866
W329	Y569	0.160	0.807

^aPLM means pseudo-likelihood maximization.

^bResidue pairs in bold resulted in enzymes with improved catalytic activity (Fig. 3).

^cResidue pairs in italic were selected for saturation mutagenesis.

Table 3.

Summary of results from saturation mutagenesis of strongest EC-coupled site pairs.

Method	EC pair	Inter- residue distanceb (Å)	Distance from active sitec (Å)	Number of active mutants at 1st site	Residues at 1st in active single mutantsd	Number of double mutants increasing activity vs. WTe	Mutations in double mutants with increased activity	kcat ratio vs. WT	kcat /Km ratio vs. WT	1st site inactive mutants screened at 2nd sitef	Percentage of double mutants restoring activityg
EVC-PLMa	D541/D473	6.0	26	8	A, I, L, V, Y, E, M, R	3	D541I/D473E D541I/D473I D541I/D473Q	2.3 1.7 1.5	2.9 2.6 2.9	N/A	N/A
EVC-PLM	K631/Q597	2.9	16	10	A, V, I, L, S, T, E, P, R, H	2	K631V/Q597K K631V/Q597S	1.8 2.3	4.0 5.6	N/A	N/A
EVC-PLM	D614/H539	2.8	17	7	A, I, L, V, E, T, S	0				N, W, M, G, Q	6–11%
EVC-PLM	E530/T520	2.6	24	N/A	N/A	0				N/A	N/A
EVC-PLM	E777/T730	2.7	8	8	D, Q, Y, N, A, S, G, V	0				N/A	N/A
EVcomplex	V328/I565	4.0	25	8	A, G, I, L, M, T, W, D	2	V328L/I565L V328L/I565V	2.4 2.4	3.0 4.2	H, S, Q, E, Y	7–16%
EVcomplex	Y392/Y571	15	16	7	A, V, S, F, W, L, I	0				N/A	N/A

^aEVC means EVcouplings and PLM means pseudo-likelihood maximization.

^bMinimum distance between non-hydrogen atoms of residue pair.

^cMinimum distance between C-α atoms in residue pair and catalytic triad (D619, E648, D733).

^dActivity per unit culture volume: 46–113% of wild type.

^eActivity per unit culture volume: D541/D473: 127–175%; K631/Q597: 152–177%; V328/I565: 154–160%.

^fActivity per unit culture volume: 9–11% of wild type.

^gActivity per unit culture volume: 43–65% of wild type.

Figure 2.

The spatial location in the homology model of BnPUL of the three residue pairs (K631/Q597, V328/I565, D541/D473) identified as mutational hotspots based on EC analysis resulting in mutant enzymes with improved catalytic activity. (a, c) Backbone representation with the X25 and X45 domains shown in teal, the CBM48 domain shown in blue, and the catalytic domain and C-terminal segment shown in gray. The residues in the catalytic triad in the active site of the enzyme (D619, E648, and D733) are shown in purple space-filling representation. The three residue pairs yielding improved catalytic activity are also shown in space-filling representation, with K631/Q597 in green, V328/I565 in yellow, and D541/D473 in red. The view in panel c is rotated by 180˚ from that in panel a. (b, d) Surface representation of BnPUL in the same orientations and with the same color-coding as in panels a and c. These images were all produced using the program PyMOL.

Construction and screening of ECSM pullulanase libraries

Enzyme engineering was performed using a system in which full-length pullulanase from B. naganoensis fused to an N-terminal PelB secretion signal is expressed under a T7 promoter from plasmid pET-28a-PelB in E. coli strain BL21(DE3). We used the NNK degenerate primers encoding 32 distinct codons, while saturating all 20 amino acids. When one amino acid residue is targeted by NNK, 96 colonies (~3 times 32) are necessary to cover 95% of the variants [58–60].

In order to efficiently identify mutants with largely decreased enzyme activity, we used an effective high-throughput screening method, the red-pullulan plate assay (Fig. 3). For every single mutant, we picked 96 colonies after transformation and spotted them at 45 positions on plates. After mutation at the first site, we selected only variants showing enzyme activity in the assay for saturation mutagenesis at the second site. For the 7 pairs, a total of 66 inactive single mutants (Table 3) were discarded prior to the generation of 1,056 double-mutants. Out of those, 154 double mutants showing at least wild-type activity in the plate assay were overexpressed and the enzyme activity per unit volume (which reflects both enzyme activity and overexpression yield) was assayed in the supernatant obtained after cell lysis. This resulted in the selection of seven double mutants (Table 3) corresponding to three residue pairs from the EC co-variance analysis (Fig. S2; residue pairs are in bold in Tables 1 and 2), which were then purified in order to accurately measure their Michaelis-Menten parameters (Fig. 4; Table S2). The fact that no mutants with improved catalytic activity could be identified for four of the residue pairs, indicates that the covariation of these pairs does not provide guidance for enhancing catalytic performance.

Figure 3.

Flowchart of the two-step saturation mutagenesis performed for two of the seven residue pairs (Table 3). The red plate pullulanase assay is performed after saturation mutagenesis of the first site. Left branch: Compensatory 2nd site mutations identified for 1st site mutations with greatly reduced activity (< ~10% of WT). Right branch: 2nd site mutations for 1st site mutations retaining ~WT activity with significantly increased enzyme activity (Table 3).

Figure 4.

Michaelis-Menten parameters of the seven double mutants (a) and the eight quadruple mutants (b) derived from the evolutionary coupling saturation mutagenesis libraries. The values were measured in triplicate and estimates of the standard deviations are indicated as error bars.

To explore the potential recovery of enzyme activity for reduced activity single mutants by introducing compensatory mutation in the second position, we selected five such single mutants for each of the two pairs D614/H539 and V328/I565 (Table 3; rightmost section) and employed saturation mutagenesis for the second position. Intriguingly, we identified at least one compensatory mutation bringing the activity back to 43–65% of the wild-type (Fig. 3; Table 3) in each case. This finding reveals that interdependent evolutionary plasticity at the coupled sites is required for co-evolution of residue pairs with strong ECs [51, 52].

Catalytic efficiency of engineered enzymes

The seven double mutants derived from our ECSM libraries exhibit 1.5–2.4 folds increases of k_cat and 2.6–5.6 folds increases of k_cat/K_m relative to the WT enzyme (Fig. 4a; Table S2), with K631V/Q597S being the best performing double mutant. Such increases in catalytic performance are impressive given the fact that WT BnPUL is a highly active enzyme originating from an organism which evolved in a carbohydrate rich environment [61]. Moreover, the seven double mutants with enhanced catalytic efficiency exhibit no detectable differences in terms of temperature and pH dependence of their activities (Fig. S3). For comparison, we performed saturation mutagenesis for two residue pairs with low PLM score (V361/K327, PLM score = 0.41021; S721/E428, PLM score = 0.084614) and found that the resulting mutants did not significantly affect enzymatic activity (Table S3).

In order to evaluate potential synergy arising from combining the double mutants and to allow further directed evolution, the five best performing ones (Fig. S2; Table S2) were selected to create a total of eight quadruple mutants (Fig. S4; Table S4). Apparently, construction of quadruple mutants did not bring significant improvement of activity for most of the double mutants (Fig. S4). Concerning the catalytic performance of the mutants, on the other hand, the eight quadruple mutants exhibit 2.5–3.0 folds increases of k_cat and 3.8–6.3 folds increases of k_cat/K_m relative to the WT enzyme (Fig. 4b; Table S4). As shown in Table S2 and Table S4, both double and quadruple mutations lower the K_m values of the enzyme obviously, while the quadruple mutants exhibit similar K_m values as those of the double mutants, and thus the increases of k_cat/K_m after combination of positive double mutants mostly due to the increases of k_cat values. Of the mutants, the best performing quadruple mutant K631V/Q597K/D541I/D473E shows a 3.0-fold increases of k_cat and 6.3-fold increases of k_cat/K_m relative to the WT enzyme. As we observed for the double mutants, all quadruple mutants exhibit WT BnPUL temperature and pH dependence of their activities (Fig. S5). In addition, by measuring the half-lives of the variants at 55 °C, the mutations at K631/Q597 and/or V328/I565 did not change much of the stability of the enzyme, while the mutations at D541/D473 gave negative effects on the stability of the enzyme, suggesting that the residue pair D541/D473 might be the functional sites relating to the stability of the pullulanase (Table S5; Table S6).

Hence, the ECSM approach compares very favorably with site-directed mutagenesis [7, 8, 62, 63] and regional truncation [3, 5, 64] methods, which increased the catalytic efficiency only about two folds (Table S7).

Spatial location of mutational hotspots

Fig. 2 shows the spatial location in the homology model of BnPUL of the six residues (none of which is highly conserved; Fig. S6) forming the three residue pairs (K631/Q597, V328/I565, D541/D473) identified as mutational hotspots based on EC analysis resulting in mutant enzymes with improved catalytic activity (Fig. 4). For all three pairs, residues forming the pair are in close spatial proximity (< 6 Å; Table 3), as is usually observed for evolutionary coupled residues [45]. Interestingly, however, all six residues are located more than 14 Å away from each of the three residues forming the catalytic triad of BnPUL. Hence, although the four residues (K631, Q597, D541, D473) identified using EVcouplings are largely buried and located on the catalytic domain opposite to the catalytic site, it might be that they are of importance for substrate, intermediate or product binding. Alternatively [19, 20], these residues may affect [16–18, 65–67] the structure and/or dynamics of the active site to improve catalytic function. In contrast, the two residues (V328, I565) identified using EVcomplex are largely buried in the catalytic-CBM48 domain interface. Hence, it is very likely that those residues enhance catalytic performance through distal effects.

Molecular dynamic simulations of pre-reaction state

To further understand the role of identified residues and the corresponding mutations on catalytic performance of the enzyme, molecular dynamic simulations were performed to analyze the structural characteristics of pre-reaction state. According to the previous QM/MM calculation, the rate-determining step of α-amylase was believed to be the nucleophilic attack from the key residue of aspartic acid to the C1 atom [68]. Fig. 5a shows accessibility of the two carboxyl oxygen atoms of Asp619 toward C1 anomeric carbon of the targeted α−1,6-glucosidic bond for the pre-reaction state, in which the carboxylate attacks on the sp3 carbon and breaks down the C-O bond at the rate-determining step and afterwards the anomeric carbon would be neutralized by a water molecule under the catalysis of the general acid/base of Glu648 and Asp733. It is highly essential for the carboxylate oxygen, anomeric carbon, and the leaving glycosyl group (OR) to orient in a good linear alignment in the rate-determining transition state. Surprisingly it was computationally observed that the Asp619 OD1 atom in the highest-efficient mutant (K631V/Q597K/D541I/D473E) points to the C1 atom statistically and dynamically, much better than those in the WT enzyme (Fig. 5b). Thus, it is conclusive that the distal mutations can significantly promote the reactive population of pre-reaction state and global effect might be very important in enzyme engineering [69–71].

Figure 5.

Interaction between the active residue Asp619 and the targeted α−1,6-glucosidic bond in the pre-reaction state (a) and conformation maps of the highest-efficient mutant and the WT enzyme in the pre-reaction state simulations (b).

Conclusions

Using ECSM, pullulanase mutants were created with significantly enhanced catalytic activity when compared with WT BnPUL (k_cat and k_cat/K_m were increased by of ~3 and ~6, respectively). These gains of catalytic activity are remarkable for a highly evolved naturally occurring enzyme, and they compare very favorably with previously engineered variants [3, 5, 7, 8, 62–64]. Since high thermostability and high enzyme activity at comparatively low pH are key for industrial applications of pullulanases [62], it is important to observe that the mutant enzymes exhibit virtually the same dependence of activity on temperature and pH as the WT, making these mutants suitable for industrial applications. Importantly, none of the six residues (i.e. three pairs) identified from calculations of ECs is located in, or close to the active site. This finding indicates that mutations made at these residue pairs propagate through the enzyme structure to affect enzyme activity.

We showed that the calculation of evolutionary couplings is effective for identifying residue pairs for protein engineering by saturation mutagenesis. Even for large proteins without known 3D structures, the ECSM approach allows one to focus on a small number of top scoring residue pairs in order to generate small saturation mutagenesis libraries [50]. Importantly, coevolution analysis enables the identification of (peripheral) distant sites influencing enzyme activity, i.e., information which cannot be derived with structure-based engineering protocols focusing on the active site and its vicinity. Finally, our study showed that residues with strong ECs are very tolerant to mutations, and that a second mutation following a mutation which results in largely reduced activity can restore activity to nearly WT levels. Overall, our findings support the current view [50, 51] that co-evolution of two residues most often relies on a first mutation resulting in unchanged or reduced fitness, which is then followed by a second mutation resulting in increased fitness.

Overall, the newly developed ECSM approach, guided by natural sequence co-variance captured in the evolutionary record, promises to become a valuable tool for future enzyme engineering synergizing strongly with established techniques [72].

Materials and Methods

EVcouplings and EVcomplex analysis

Full-length 926-residue pullulanase from B. naganoensis contains five domains identified using the Conserved Domain Database at NCBI: CBM41, residues 5–100 (PF03714); X45, residues 109–163; X25, residues 162–250; CBM48, residues 315–393 (PF02922); and the catalytic domain, residues 473–787 (PF00128). The Jackhmmer algorithm [73] was used to search the Uniprot [74] database to generate multiple sequence alignments as input for EC calculations. Using the EVcouplings server (http://evfold.org) with default parameters, ECs were determined from sequences comprising the catalytic domain (473–787), the catalytic and CBM48 domains (315–787), the catalytic domain and the C-terminal segment (473–926) using EVcouplings-PLM. ECs with PLM scores higher than 1.0 were selected for saturation mutagenesis, excluding pairs within ten residues of a higher scoring pair (Table 1). Next, additional inter-domain ECs between the catalytic and CBM48 domains were identified using the EVcomplex server (https://evcomplex.hms.harvard.edu) with default parameters. Inter-domain ECs with EVcomplex scores higher than 1.0 were selected for saturation mutagenesis, excluding pairs within ten residues of a higher scoring pair (Table 2).

Materials

A QuikChange Lightning Site-Directed Mutagenesis Kit was purchased from Stratagene (La Jolla, CA, USA). Oligonucleotides were synthesized by Talen-bio Technology (Shanghai, China). Plasmid preparation kit I (D6943–02) was purchased from Omega Bio-tek (Norcross, GA, USA). DNA sequencing was conducted by Talen-Bio Scientific (Shanghai, China). All commercial chemicals were purchased from Sigma (Saint Louis, MO, USA), Takara (Otsu, Japan), or Sinopharm Chemical Reagent (Shanghai, China). Pullulan was purchased from Tokyo Chemical Industry (Chuo-ku, Tokyo), and red pullulan was ordered from Megazyme (Wicklow, Ireland).

Primer design and library construction

The full-length enzyme was expressed with a 31-residue N-terminal extension in a pET vector (pET-28a-PelB) using lactose induction of T7 polymerase in E. coli strain BL21 (DE3). The full sequence of the N-terminal extension is MKYLLPTAAAGLLLLAAQPAMAMDIGINSDP. The first 22 residues (underlined) in this sequence encode the signal sequence of the PelB protein from E. carotovora, while the final nine residues arise from the expression vector. The list of primers is shown in Table S8. The primers contained a random sampling of the nucleotides at each of the three positions in the codon for the mutated amino acid. For the double mutant T520/E530, a single primer containing both saturation mutation sites were designed. For the other double mutants, expression systems were generated in two steps. First, primers for saturation mutagenesis at one of the two mutation sites were used with the WT sequence as the template. After digestion of the PCR products with Dpn I to remove the WT template plasmid, 2 μL of the reaction mixture was used to transform E. coli BL21 (DE3). After plating, individual colonies were picked, and based on sequencing results, plasmids were selected representing the full set of 19 amino-acid substitutions at the target site. The single mutants were screened for enzyme activity as described in the next paragraph, and inactive mutants were discarded. Next, each of the plasmids encoding active single mutants were used as a template for saturation mutagenesis at the second mutation site following the same protocol.

Screening

Colonies of cells containing mutant plasmids were picked and spotted at 45 positions on plates containing solid auto-induction Studier [75] medium (recipe given below) with 2% agar, 1% red pullulan, and 50 μg·mL⁻¹ kanamycin, and the plates were incubated overnight at 37 °C (Fig. S1). Colonies forming transparent zones on the screening plates were picked for protein overexpression and used to inoculate 5 mL LB medium in the presence of kanamycin (50 μg·mL⁻¹). After incubation for 10 h at 37 °C with shaking at 200 rpm, 1 mL of the culture was transferred to a 250 mL flask containing 50 mL auto-induction medium (10 g·L⁻¹ α-lactose, 1.0 g·L⁻¹ glucose, 5.0 g·L⁻¹ glycerol, 6.8 g·L⁻¹ KH₂PO₄, 0.25 g·L⁻¹ MgSO₄, 10 g·L⁻¹ tryptone, 5.0 g·L⁻¹ yeast extract, 7.1 g·L⁻¹ Na₂HPO₄, 0.71 g·L⁻¹ Na₂SO₄, and 2.67 g·L⁻¹ NH₄Cl, pH 7.5) supplemented with kanamycin (50 μg·mL⁻¹). After cultivation at 37 °C with shaking at 200 rpm for the first 2–3 h, the culture was incubated at 17 °C with shaking at 200 rpm for another 60 h to express the target protein.

Subsequently, the cells were centrifuged and lysed by sonication. The supernatant obtained before and after cell lysis was assayed for pullulanase activity by measuring aldehyde release from a reaction mixture containing 0.2 mL of 2% (w/v) pullulan in 0.1 M sodium acetate buffer (pH 4.5) and 0.2 mL enzyme solution diluted with 0.1 M sodium acetate buffer (pH 4.5), which was incubated at 60 °C for 20 min. Next, the amount of released aldehyde was determined using dinitrosalicylic acid (DNS) and measuring the absorbance at 540 nm spectrophotometrically. One unit of pullulanase activity is defined as the amount of pullulanase that releases 1 μmol of aldehyde per min under these reaction conditions [76]. The reported enzyme activity (Fig. S2 and S4) corresponded to the intracellular (supernatant after cell lysis) activities. All assays were performed in triplicate. Double-mutants exhibiting enzyme activity greater than that of WT pullulanase (cut-off 580 U·mL⁻¹) were selected for protein purification [6] to accurately measure the Michaelis-Menten parameters.

Colonies from D614/H539 and V328/I565 showing no transparent zones on the red pullulanase screening plates were picked for sequencing. Five single-mutants were selected for protein overexpression, enzyme activity per unit culture volume was measured, and second site saturation mutagenesis was performed. For the resulting double-mutants which showed transparent transformation zones on the plate enzyme activity per unit culture volume was measured.

Determination of Michaelis-Menten parameters

K_m and k_cat of purified enzymes were determined based on a previously described method [77] with measurements at the following pullulan concentrations: 0.1, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, 3.0, 4.0, 5.0, 6.0, 8.0, 10.0, 15.0, 20.0 mg·mL^-1. The k_cat and K_m values were obtained by fitting the initial rate data to the Michaelis-Menten equation using nonlinear regression with the GraphPad Prism [78]. Values are shown as means from three replicates with standard deviation.

Molecular dynamic simulations of pre-reaction state

The initial protein structure used in the pre-reaction state analysis was constructed with homology modeling [79], in which a crystal structure (PDB ID: 2WAN) of pullulanase from B. acidopullulyticus was employed as template (sequence identity: 63.67%). The complexation of substrate and the enzyme was referred with the similar triad (D-E-D) in human α-glucosidase maltase-glucoamylase [80]. The enzyme-substrate complexes were further optimized with the CHARMm force field in Discovery Studio software package 3.5. The saccharide substrate was structurally optimized with the semi-empirical method (AM1) and parameterized for the following-up molecular dynamic simulation with the antechamber program in the Gaussian16 [81] and AMBER12 [82] software package. Water molecules was assigned with the TIP3P model, and the general Amber force field (GAFF) and ff03.r1 force field were applied for the classical molecular dynamic simulation. The complexes were placed in a truncated octahedral box of water molecules, extending 10.0 Å along each dimension. A certain number of counterions Cl-were added to neutralized the calculated system. The MD systems were first minimized by the steepest descent minimization of 1000 steps followed the conjugate gradient minimization of 9000 steps, heated up from 0 to 300 K at constant volume in 50 ps, and equilibrated for another 50 ps without any restraints. In the molecular dynamic simulations, the Particle Mesh Ewald (PME) method was employed for long-range electrostatic interactions. Finally, multiple 10 ns of trajectories were collected for the further pre-reaction state analysis [69–71].

Abbreviations

EC

evolutionary coupling

PLM

pseudo-likelihood maximization direct coupling analysis

DI

mean-field direct coupling analysis

SM

saturation mutagenesis

BnPUL

pullulanase from Bacillus naganoensis

MSA

multiple sequence alignment

pullulanase

(EC 3.2.1.41)