The chitinolytic machinery of Serratia marcescens is one of the best known enzyme systems for the conversion of insoluble polysaccharides. This machinery includes four chitin-active enzymes: ChiC, an endo-acting non-processive chitinase; ChiA and ChiB, two processive chitinases moving along chitin chains in opposite directions; and CBP21, a surface-active CBM33-type lytic polysaccharide monooxygenase that introduces chain breaks by oxidative cleavage. Furthermore, an N-acetylhexosaminidase or chitobiase converts the oligomeric products from the other enzymes to monomeric N-acetylglucosamine. Here we discuss the catalytic mechanisms of these enzymes as well as the structural basis of each enzyme's specific role in the chitin degradation process. We also discuss how knowledge of this enzyme system may be extrapolated to other enzyme systems for conversion of insoluble polysaccharides, in particular conversion of cellulose by cellulases and GH61-type lytic polysaccharide monooxygenases.
The genus Serratia is a member of the Enterobacteriaceae order that currently has 15 cultured type strains registered in the Ribosomal Database Project database . The bacterium is known for its ability to synthesize the red pigment prodigiosin (2-methyl-3-amyl-methoxyprodigiosene [2, 3]) and for being frequently encountered in nosocomial infections [4-7]. Other capabilities have made this bacterium one of the most rigorously studied organisms related to conversion of recalcitrant polysaccharides, namely its simple but yet highly efficient enzymatic machinery for degradation of chitin, a linear insoluble polymer of β-1,4-linked N-acetylglucosamine (GlcNAc).
In 1969, a groundbreaking study by Jaime Monreal and Elwyn T. Reese concluded that Serratia marcescens was the most efficient chitin degrader amongst 100 tested microorganisms . The aim of this work was to locate a source for ‘pure’ chitinase activity which would then allow detailed studies of these enzymes as ‘lytic agents’. Indeed, these findings laid down the cornerstone for a model system employed by many to study enzymatic chitin degradation. A similar study conducted by Carroad and Tom 9 years later , which aimed at finding bacteria efficient in bioconversion of shellfish wastes, led to the same conclusions as the study by Monreal and Reese. Following these studies, the next important contribution to the study of S. marcescens chitinases came when Roberts and Cabib described a simple method for purifying the enzymes by a chitin-affinity strategy . This method was based on a method for chitinase purification published in Methods in Enzymology in 1966  and involved binding the enzymes to an insoluble but easily degradable chitinous substrate. After binding, the mixture was incubated at 30 °C until the chitin was fully degraded. By removing the reaction products by dialysis, the chitinases were left in their pure form. Some years later Fuchs et al. combined the Roberts and Cabib purification method with a size exclusion chromatography step in an attempt to purify the individual chitinolytic components of S. marcescens, and concluded that the bacterium secretes five individual proteins involved in chitin conversion . These proteins would later be named ChiA, ChiB, ChiC1, ChiC2 and CBP21. In 1986, Fuchs et al. and Jones et al. reported the first cloning of S. marcescens chitinases, namely ChiA and ChiB [12, 13]. ChiA and ChiB were cloned by various groups later [14-16], followed by CBP21 [17, 18] and ChiC1/ChiC2 [19-21]. In the meantime, another component of the system, an N-acetylhexosaminidase which is commonly referred to as chitobiase, was discovered and cloned [22, 23]. A more convenient way of releasing the various chitin-active enzymes from chitin was discovered by Watanabe and co-workers who managed to separate chitobiase, CBP21 and the chitinases by applying a pH gradient (pH 6.0 to pH ~ 3.6) .
According to the CAZy sequence-based classification system , all S. marcescens chitinases belong to glycoside hydrolase family 18 (GH18). The catalytic domain of the chitobiase belongs to family GH20, whereas CBP21 has been classified as a CBM33, i.e. a carbohydrate binding module (CBM) in family 33. Following the cloning and purification of the chitinases, CBP21 and chitobiase, X-ray crystallographic structures of these enzymes started to emerge. ChiA was the first Serratia chitinase to have its structure solved, representing the first ever structure of a bacterial or fungal GH18 chitinase . Subsequently, the structures of chitobiase , ChiB  and CBP21  were solved whereas the structure of the catalytic domain of ChiC was solved in 2012 .
Further insight into the S. marcescens chitinolytic machinery was obtained by combining this structural information with a variety of biochemical analyses, including studies of enzyme kinetics and chitin-degrading potential, studies with inhibitors such as the natural product allosamidin  and a variety of mechanistic studies supported by site-directed mutagenesis work. Our current knowledge of the system results from these studies and is reviewed below. Highlights include (a) the discovery of a substrate-assisted catalytic mechanism to cleave the β-1,4 bonds in chitin and chito-oligosaccharides , (b) novel insights into the mechanism of processivity obtained using chitosan, a soluble variant of chitin [31-33], and (c) the discovery of the enzymatic activity of CBP21, which was the key to revealing a novel enzyme family currently referred to as lytic polysaccharide monooxygenases (LPMOs) [34, 35].
The GH18 chitinases ChiA, ChiB and ChiC
Structures and modular organization
The GH18 catalytic domains of ChiA, ChiB and ChiC have a (β/α)8 TIM-barrel fold with crucial catalytic residues (see below) being located on β-strand number 4. All three chitinases are multimodular (Fig. 1, Table 1), having an N-terminal chitin binding module with a fibronectin III (FnIII) like fold (ChiA), or a C-terminal CBM5 chitin binding module (ChiB) or a C-terminal FnIII module coupled to a downstream CBM12 chitin binding module (ChiC). Full-length ChiC, also referred to as ChiC1, tends to be cleaved by endogenous proteases to yield ChiC2, a form observed in the early studies on secreted chitinolytic proteins and comprising the catalytic domain only [19-21] (this is observed both in S. marcescens and when recombinantly expressing full-length ChiC in Escherichia coli). Note that the available crystal structures for ChiA and ChiB comprise the complete two-domain proteins, whereas for ChiC only the crystal structure of the catalytic domain has been determined. However, there exists structural information for individual FnIII and CBM12 modules from work on chitinase A1 from Bacillus circulans (Fig. 1) [36, 37].
Table 1. Domain structure of the S. marcescens chitinolytic machinery. The protein structure database (Protein Data Bank, PDB) entries are those for the wild-type apo-enzymes. For each enzyme, the top to bottom order of the listed domains reflects N-terminal to C-terminal in the protein. References for the classification systems: CAZy , Pfam , CATH . N/A indicates that the domains are not classified
2.70.50 Coagulation factor XIII; Chain A, domain 1
The roles of the additional CBMs and FnIII modules have been addressed in several studies, but remain partly unresolved. The substrate binding properties of CBMs in general are well documented  but exactly how and to what extent CBMs contribute to the efficiency of substrate conversion is not completely clear. Among the more ‘advanced’ potential functions of CBMs are roles in correctly positioning the catalytic domain, contributions to processive action, and perhaps even local decrystallization of the substrate (e.g. [39, 40]). The 40–60 residue CBM5 and CBM12 modules often found in chitinases are distantly related and appear as one family in Pfam (PF02839; www.pfam.org). Both are characterized by the presence of conserved exposed tryptophans that interact with the substrate (see below). Several studies have confirmed that the presence of these domains increases substrate affinity as well as the efficiency of chitin hydrolysis, especially for the more crystalline chitin forms [41-44].
The FnIII module of ChiA (Pfam name ‘chitinase A N-terminal domain’) contains exposed aromatic residues and mutational studies have clearly shown that these residues contribute to substrate binding and the efficiency of substrate hydrolysis . This domain shares barely detectable sequence identity with FnIII modules that are found in ChiC and in several other chitinases such as ChiA1 from B. circulans [37, 42] and that form a separate Pfam family (Fig. 1, Table 1). Watanabe and co-workers have carried out both functional and structural studies of the latter type of FnIII domain, leading to the conclusion that these domains are not directly involved in chitin binding. Nevertheless they are important for enzyme activity, probably by affecting the overall enzyme structure and the spatial localization of the catalytic domain and the CBM12 domain . In contrast to the N-terminal domain of ChiA, the structurally characterized FnIII domain of ChiA1 from B. circulans does not have exposed aromatic residues on its surface . Model-building studies (not shown) indicated that the FnIII domain of ChiC also lacks surface-exposed aromatic residues.
Apart from the actual bond cleavage, catalysis by these chitinases involves gaining access to a single chain of chitin, substrate binding and positioning and, after cleavage, substrate displacement, as discussed in detail further below. The enzymes may cleave randomly in the chain (endo-action) or they may primarily bind to chain ends (exo-action). Each of these two modes of initial binding may or may not be combined with processive action, the latter meaning that each productive enzyme–substrate association leads to a series of consecutive cleavages, releasing dimeric products from the polymer substrate.
The first GH18 chitinase structures were those of ChiA from S. marcescens  and of a plant single domain endochitinase called hevamine [46, 47]. These structures were puzzling in the sense that the putative catalytic centers of the enzymes did not show the typical arrangement of two carboxylic acids normally found in GHs. One of these residues acts as a catalytic acid that protonates the leaving group, whereas the other acts as a nucleophile in retaining GHs or as a water-activating base in inverting enzymes (see [48, 49] for further details). After having shown that hevamine is a retaining enzyme, Terwisscha van Scheltinga and colleagues proposed a solution to this puzzle on the basis of the crystal structure of hevamine in complex with allosamidin . The pseudotrisaccharide allosamidin  is a highly specific inhibitor of GH18 enzymes and contains an allosamizoline group. The structure of the hevamine–allosamidin complex showed that the allosamizoline group binds in the −1 subsite, leading Terwisscha van Scheltinga and colleagues to propose that it mimics a reaction intermediate in which a positive charge at C1 is stabilized intramolecularly by the carbonyl O atom of the N-acetyl group at C2 . Thus, the nucleophile ‘missing’ from the structures is provided by the substrate itself.
This substrate-assisted catalytic mechanism has since been studied in much detail and there is now ample evidence supporting and refining the original ideas. Furthermore, it has been shown that a similar substrate-assisted mechanism is used by other GHs acting on GlcNAc glycosides, including the GH20 chitobiase, one of the other members of the chitinolytic enzyme system  (see later), and the O-GlcNAc hydrolases of family GH84 . Today, there is an abundance of evidence from crystallographic [51, 52] and biochemical studies [53-55] supporting the idea that the anchimeric assistance from the acetamido group on the −1 sugar leads to formation of a covalent oxazolinium ion intermediate. As shown in Fig. 2, the formation of this intermediate requires distortion of the −1 sugar towards a boat conformation, which entails the formation of an even less favorable half-chair type transition state conformation along the way towards the boat form [56-58]. The distortion of the −1 sugar has been confirmed by various crystallographic studies [50, 59-61].
Some of the most detailed work aimed at unraveling the catalytic mechanism of the GH18 chitinases was done on ChiB from S. marcescens. The GH18 chitinases contain a diagnostic DXDXE motif ending with the catalytic acid (Asp140–Glu144 in ChiB; Figs 2 and 3). Another highly conserved residue is a serine that is part of a less well known but equally diagnostic SXGG motif (Ser93 in ChiB). Additional conserved residues include Tyr214 and Asp/Asn215 (Figs 2 and 3). Using crystallography, site-directed mutagenesis and computational analysis, roles for each of these residues have been proposed as follows (for work on ChiB, see [59, 62, 63]; for some key studies on other GH18 chitinases, see [61, 64, 65]).
Glu144, the catalytic acid/base: In the first phase of the reaction, Glu144 facilitates leaving group departure by donating a proton. In the enzyme–substrate complex Glu144 has a drastically elevated pKa due to the combination of substrate binding and the presence of Asp215 (GH18 enzymes with an Asn in this position have an acidic pH optimum for activity whereas the optimum for ChiB lies around 6–7). The −1 net negative charge on the Asp140–Asp142 pair may also contribute to the elevated pKa. In the second phase of the reaction, after leaving group departure, Glu144 acts as a base. Its pKa is now lower due to departure of the leaving group, which exposes the carboxylate to solvent and reduces the effect of Asp215 . Furthermore, the rotation of protonated Asp142 towards Glu144 will also contribute to reducing the pKa of the latter.
Asp142: This residue plays several interconnected roles. First, its interaction with the acetamido group of the −1 sugar fixes this group in a distorted conformation that promotes nucleophilic attack on the anomeric carbon. Second, through the same interaction, Asp142 stabilizes the developing positive charge on what is to become the oxazolinium ion intermediate. Finally, as alluded to above, the mobility of Asp142 may play a role in the necessary ‘cycling’ of the pKa of Glu144.
Asp140: This partly buried residue seems crucial for keeping Asp142 protonated. Mutation of Asp140 to Asn or Ala drastically reduces activity and yields an acidic shift in the pH optimum for activity.
Ser93 and Tyr10: These residues interact with Asp140, relatively deep ‘down’ in the β-barrel (Fig. 3). Structural data show that rotation of protonated Asp142 towards Glu144 leads to adjustments of Tyr10 and Ser93 that have two consequences. First, the adjusted side chains of Tyr10 and Ser93 partly fill the space left by Asp142. Second, and probably more importantly, due to these adjustments both residues now act as donors in strong hydrogen bonds with Asp140, thus compensating in part for the negative charge of the latter.
Tyr214: As shown in Fig. 2, the phenolic hydroxyl of the tyrosine side chain stabilizes the acetamido group in the distorted conformation that promotes nucleophilic attack on the anomeric carbon. Several of the mutational studies cited above have shown that mutation of Tyr214 (to Phe or Ala) has a drastic effect on catalytic efficiency.
Asp215: This residue is involved in binding the −1 sugar in its distorted conformation (Fig. 2). Furthermore, the negative charge on this residue helps to raise the pKa of the catalytic acid, especially in the enzyme–substrate complex where the glycosidic oxygen is placed between the carboxyl groups of Asp215 and Glu144. In line with these roles, mutation of Asp215 to Asn (which can still hydrogen bond to the substrate) only has a small effect on maximum activity but leads to an acidic shift of the pH optimum for activity. Mutation of Asp215 to Ala is highly deleterious for activity .
Importantly, the experimental data for ChiB show that catalysis involves many residues beyond the key residues Asp142 and Glu144. In fact a larger part of the TIM-barrel core may be involved in catalysis, as suggested by the fact that there is another highly conserved aspartate even further down (Asp137) in what in fact is a DXXDXDXE motif (note that the D137N mutant of ChiB could not be produced due to protein folding/stability issues). Also, GH18 enzymes tend to have positively charged residues in the ‘bottom’ of the TIM-barrel (i.e. near Asp137) that could affect active site electrostatics, e.g. through an effect on the pKa of Asp140. The GH18 chitinases provide one of the better studied GH types as far as the complexity and extent of the catalytic site are concerned. Clearly, such complexity is not unique to GH18 enzymes, since fine-tuning and, in retaining GHs, cycling of pKa values is a common enzyme feature. In-depth studies of a family 11 retaining xylanase from B. circulans provide another example of such complexity [66-68].
Exo- versus endo-action, processivity and directionality in the GH18 chitinases
Whereas the three S. marcescens chitinases employ the same catalytic machineries, the architectures of their active sites suggest that they have different roles in chitin degradation. As shown in Fig. 4, ChiA and ChiB have deep substrate binding clefts lined with aromatic residues, suggesting that these enzymes are processive exo-acting enzymes [69-71]. One key feature of the ‘deep-cleft’ chitinases is the presence of the so-called α + β insertion domain that forms one ‘wall’ of the substrate binding cleft . This 60–80 residue domain is inserted between strand 7 and helix 7 of the TIM-barrel in only part of the GH18 enzymes. The domain may contain extra loops that participate in substrate binding, as is the case in ChiB (with a 79-residue α + β domain) but less so in ChiA (with a 61-residue α + β domain). ChiC lacks the α + β domain and has a much shallower substrate binding cleft lacking most of the aromatic residues seen in ChiA and ChiB (Fig. 4), suggesting that ChiC may be a non-processive endo-acting enzyme.
Structural alignment of ChiA and ChiB reveals several striking features. First, the carbohydrate binding domains are situated in opposite directions (Figs 1 and 4) suggesting that the two enzymes degrade the polymer from different ends, ChiA from the reducing end and ChiB from the non-reducing end. This suggestion, originally derived from comparing the two structures [26, 27], was supported by work from Hult et al. who used tilt microdiffraction in combination with specific labeling of the reducing ends to determine the directionality of chitin degradation by ChiA and ChiB . Second, both enzymes have linear paths of surface-exposed aromatic amino acids starting from the chitin binding domains and extending into the catalytic domains. From work discussed below, as well as studies on other enzymes, it is now clear that hydrophobic stacking interactions between aromatic residues and sugar moieties offer a solution to the problem of how processive enzymes manage to remain attached to their substrates while at the same time retaining the ability to slide during the processive mode of action [33, 70, 74-79].
Processivity and its directionality are challenging to measure (see  for an extensive discussion of possibilities and pitfalls). Measuring final dimer/trimer ratios is a classical method for determining the degree of processivity for glycosidases in general [81-83], but this method is prone to errors, primarily because the ratios are strongly affected by specific, not processivity-related, binding preferences of individual enzymes for oligomeric intermediate products . Final dimer/trimer ratios after degradation of β-chitin by the three individual chitinases were determined as 7.3 for ChiA, 12.6 for ChiB and 4.1 for ChiC , which is compatible with the idea that ChiA and ChiB are more processive than ChiC (note that ChiC by all standards seems absolutely non-processive; see later).
Interestingly, the degree of processivity for family 18 chitinases can be determined by using the water-soluble chitosan as the substrate . Chitosan is obtained by random partial deacetylation of chitin  and it is possible to prepare ‘chitin-like’ chitosans, i.e. chitosans with a high degree of acetylation. Since a correctly positioned N-acetyl group in the −1 subsite is essential for catalysis by GH18 chitinases (Fig. 2), binding of chitosan may be non-productive because the sugar bound in the −1 subsite may lack an N-acetyl group. Formation of a non-productive complex could lead to complete substrate release followed by re-binding. Alternatively, in processive enzymes, the substrate could remain loosely associated with the enzyme and be displaced until a productive complex is formed. Because the repetitive unit in chitin and chitosan is a dimer, the latter processive scenario would imply that every oligomeric product resulting from the same initial enzyme–substrate association, except for the very first, would be even-numbered in length. This is exactly what was observed for ChiA and ChiB, but not for ChiC. When chitosan is the substrate, ChiA and ChiB generate a surplus of even-numbered oligomers while ChiC produces equal amounts of both even- and odd-numbered oligomers [32, 86].
Chitosan offers another method for measuring both processivity and the endo/exo character, which is based on simultaneously monitoring the generation of new reducing groups and the reduction in substrate viscosity . Using this method, it was shown that treatment with ChiC has the same effect on chitosan as acid hydrolysis. For both treatments, the viscosity reduction per generated reducing end is at a maximum, meaning that ChiC is an endo-acting, non-processive enzyme. The same method confirmed that ChiA and ChiB are processive enzymes and also showed that both tend to bind chitiosan in an endo fashion (as opposed to chitin, which seems to be attacked in an exo fashion ).
The contribution of aromatic residues to the processivity of ChiA and ChiB has been studied using these methods for analysis of processivity. Early work [45, 87, 88] had already revealed that aromatic residues are important for interactions with the substrate and that mutation of these residues tends to lead to modest reductions of chitin hydrolyzing activity. These studies were mainly focused on residues remote from the catalytic center and did not address processivity. Studies of ChiA and ChiB have shown that aromatic residues near the catalytic center are crucial in determining both the degree and the directionality of processivity. The first clue came when it was shown that mutation of Trp97 in the +1 subsite of ChiB (Fig. 4) to Ala nearly abolishes processivity . The Trp220Ala mutation in the +2 subsite yielded similar results, i.e. a large reduction in processivity. Mutation of the analogous residues in ChiA, Trp275 and Phe396 respectively (also to Ala; Fig. 4), reduced the degree of processivity only slightly . In ChiA, however, replacement of Trp167 in the −3 subsite by Ala led to the same drastic reduction in the degree of processivity as observed for mutation of Trp97 in the +1 subsite of ChiB. These results show that aromatic residues close to the catalytic center are important for the processive mechanism. Moreover, and of major importance, the results show that it is the aromatic residues in the substrate binding subsites, i.e. the subsites interacting with the polymeric substrate rather than the oligomeric product, that determine processivity. In ChiA, which presumably breaks down chitin from its reducing end, the polymeric part is thought to bind to the – subsites and, indeed, a Trp in the −3 subsite seems crucial for processivity. In ChiB, which presumably breaks down chitin from its non-reducing end, the polymeric part is thought to bind to the + subsites and, indeed, Trp residues in the +1 and +2 subsites seem crucial for processivity.
These studies of ChiA and ChiB revealed an interesting pay-off between processivity and intrinsic enzyme speed. As expected, the non-processive mutants were less effective in degrading crystalline chitin but, quite remarkably, these mutants showed increases in their activities towards chitosan by an order of magnitude. So, while processivity contributes to the degradation of crystalline polysaccharides, probably because once detached single-polymer chains are kept from reassociating with the solid material, the data for the ChiA and ChiB mutants show that this processivity comes at a large cost in terms of enzyme speed. These observations show that the rate-limiting steps in chitin and chitosan hydrolysis are different, and this was confirmed in a follow-up study by Zakariassen et al. (2010) on the kinetics of polymer degradation by ChiA variants . Using Eyring analysis, the authors determined the activation energies for various enzyme–substrate combinations. The results showed that association with the substrate is rate limiting in chitin hydrolysis (reflected in a large entropic term); in this case, the Trp to Ala mutations have an unfavorable effect on the activation enthalpy due to loss of favorable enzyme–substrate interactions. On the other hand the activation energy of chitosan hydrolysis is dominated by an enthalpic term, suggesting that product displacement and release are rate limiting. In this case, the Trp to Ala mutations have a favorable effect on the activation enthalpy, reflecting that fewer interactions need to be broken during product displacement.
The different architectures of the active sites of ChiA and ChiB are also reflected in the thermodynamic signatures of allosamidin binding. Allosamidin binds to subsites −3 to −1, which are ‘substrate’ subsites in ChiA and ‘product’ subsites in ChiB. While allosamidin binds equally strongly to both ChiA and ChiB at pH 6.0 (Kd ~ 0.2 μm), binding to ChiA is driven by enthalpy whereas binding to ChiB is driven by entropy . Mutation of Trp167, which is crucial for processivity in ChiA, to Ala reduced the enthalpy gain of allosamidin binding by 4.4 kcal·mol−1. In another study, addressing the ‘substrate’ subsites of ChiB, it was shown that mutation of Trp97 to Ala leads to a 200-fold increase in Km for (GlcNAc)4 while kcat is increased . This underpins how the ‘stickiness’ of the ‘substrate’ subsites, which is needed for processivity, affects enzyme efficiency.
The difference between non-processive and processive GH18 chitinases has been explored through molecular dynamics simulations of enzyme–substrate complexes . These simulations revealed differences that seem to relate to the more dynamic on–off ligand binding processes associated with non-processive action. In short, non-processive enzymes have more flexible catalytic centers and their bound ligands are more solvated and flexible.
Transglycosylation in GH18 chitinases
Some family 18 chitinases are known to catalyze transglycosylation in addition to hydrolysis. Since transglycosylation is a kinetically controlled reaction, efficient transglycosylation requires an enzyme with an active site architecture that disfavors correct positioning of the hydrolytic water molecule and/or favors binding of incoming carbohydrate molecules, through strong interactions in positive subsites . While wild-type ChiB does not yield detectable levels of transglycosylation products, wild-type ChiA does show a minor activity . To attempt to improve transglycosylation efficiency, specific mutations were introduced in ChiA and ChiB that were likely to disfavor correct positioning of the hydrolytic water molecule and/or favor binding of incoming carbohydrate molecules. Asp313/142 in ChiA and ChiB, respectively, is the second Asp in the conserved DXDXDE motif (its role in the catalysis is described above). Quantum mechanics/molecular mechanics have shown that mutating Asp142 in ChiB to Asn leads to a change in active site electrostatics that could lead to lower hydrolyzability of the oxazolinium ion intermediate or an increased probability of the intermediate being attacked by an incoming sugar, perhaps due to effects on the catalytic water . Furthermore, the Asp142 to Asn mutant of ChiB has a reduced Km for oligomeric substrates which could imply increased affinity for sugar acceptors since this affinity may to some extent be reflected in the Michaelis–Menten constant [62, 93]. Interestingly, both the ChiA-Asp313 to Asn mutant and the ChiB-Asp142 to Asn mutant indeed showed increased transglycosylation activity.
Mutation of Trp97 in the +1 subsite of ChiB and Phe396 in the +2 subsite of ChiA (Fig. 4) to Ala led to reduced transglycosylation activity, probably due to reduced acceptor affinity. Interestingly, mutation of Phe396 to Trp in the already transglycosylating ChiA-Asp313 to Asn mutant, a mutation meant to increase acceptor affinity, led to a clear increase in the transglycosylation efficiency. Similar results have been obtained very recently by Umemoto et al. , using a plant GH18 chitinase. These results show that the aromatic residues that seem to determine the degree and direction of processivity also may co-determine to what extent the enzyme in question may yield transglycosylation products.
Chitobiase, a family GH20 N-acetylhexosaminidase
The 95.8 kDa chitobiase (Uniprot ID Q54468) is the only characterized GH20 from S. marcescens. The GH20 family hosts enzymes that cleave off the non-reducing end sugar of oligosaccharides containing β-N-acetylhexosamines (β-N-acetylhexosaminidase, EC 184.108.40.206, and β-1,6-N-acetylglucosaminidases) or β-6-SO3-N-acetylglucosamine (β-6-SO3-N-acetylglucosaminidase), as well as enzymes acting on lactose (lacto-N-biosidase, EC 220.127.116.11). The functions of GH20 enzymes range from recycling gangliosides in the lysosomes of higher organisms to food scavenging in bacteria and fungi. Apart from chitin conversion, some bacterial GH20 enzymes may also be involved in the hydrolysis of N-glycans and are sometimes classified as virulence factors .
The primary role of the S. marcescens chitobiase is to convert the main product of chitinase action, GlcNAc2, to GlcNAc. It should be noted though that chitobiase and other β-N-acetylhexosaminidases are in general also capable of cleaving off single GlcNAc moieties from the non-reducing end of longer soluble chito-oligosaccharides . Production of chitobiase is induced by both GlcNAc and (GlcNAc)2 and during chitin degradation the enzyme is both secreted into the medium and retained in the periplasm [23, 98]. An S. marcescens mutant lacking the chitobiase grows better on GlcNAc than on (GlcNAc)2 showing that the bacterium depends on chitobiase for optimal utilization of chitin . Chitobiase may also be important to alleviate possible (product) inhibition of the chitinases by (GlcNAc)2, although available data indicate that such product inhibition is not very strong (mM range) .
The structure of chitobiase
The chitobiase gene was first cloned and expressed in 1989 by Kless et al. . The structure of chitobiase was the first of a GH20 enzyme and its complex with the substrate (GlcNAc)2 provided important structural evidence of substrate-assisted catalysis in this enzyme family . The enzyme comprises four domains (Fig. 5, Table 1): domains I, II and IV are organized around a central catalytic domain (III). Domains I, II and III are closely associated, whereas domain IV forms a protrusion. Chitobiase is the largest member of the chitinolytic machinery and its active site pocket and most common substrate (chitobiose) are intriguingly small compared with the vast size of the enzyme. The structure of domain I resembles that of a CBM2 , but there is no experimental evidence supporting a function in binding to an insoluble substrate (binding of chitobiase to chitin is weak, G. Vaaje-Kolstad and S. Lislebø, unpublished observations). Neither domain II (α + β topology) nor domain IV (two β-sheets) bear resemblance to known protein domains with carbohydrate-related activities and the functional roles of these domains are enigmatic.
It is interesting to note that there is large domain variability amongst the known members of the GH20 family. For example, one of the best characterized β-N-acetylhexosaminidases, SpHex, lacks domains I and IV , whereas other GH20 members have different auxiliary modules with unknown functions (for examples see [100, 101]).
Domain III hosts the active site and has an (β/α)8 fold like the catalytic domains of GH18 enzymes. The wall of the entrance to the catalytic site is also defined by domain I and partially by the linker joining domains I and II (Fig. 5). The active site pocket has the shape of a boot and is defined by two subsites (−1 and +1; Fig 5). The −1 subsite hosts the non-reducing sugar and it contains tryptophans, making hydrophobic interactions with the sugar, as well as polar residues forming hydrogen bonds with hydroxyl groups, the 2-acetamido nitrogen and the glycosidic oxygen. Thus, the −1 subsite provides an environment that forces the bound sugar into a distorted conformation that facilitates hydrolysis of the glycosidic bond. The +1 subsite provides few interactions with the bound sugar, which is somewhat surprising considering that the many interactions in the −1 subsite are offset by substrate distortion . Apparently, the net binding energy in the −1 subsite is sufficient for effective and productive substrate binding.
Catalysis in the chitobiase
As alluded to above, the structures and catalytic mechanisms of GH20 β-N-acetylhexosaminidases and GH18 chitinases were unraveled at approximately the same time. As excellently described in a landmark paper by Tews et al. , it turned out that both enzymes use a mechanism that includes anchimeric assistance and formation of an oxazolinium ion intermediate (Figs 2 and 6). Next to the catalytic acid (a glutamate) both enzymes have an aspartate that interacts with the acetamido group of the −1 sugar and positions it for nucleophilic attack on the anomeric carbon. Interestingly, in GH20 enzymes the aspartate and the glutamate are neighboring residues while in GH18 chitinases they form a DXE motif. Viewed in a historical perspective, the first kinetic evidence for neighboring group participation in β-N-acetylhexosaminidases was provided by Yamamoto  who used free energy relationships to analyze catalysis by a protein isolated from Aspergillus oryzae , presumably a GH20 enzyme.
In their original crystallographic study , Tews et al. determined the structure of a chitobiase–(GlcNAc)2 complex that turned out to be a ‘pre-Michaelis complex’ having a distorted sugar in the −1 subsite with the glycosidic oxygen positioned for protonation by a catalytic acid (Glu540; Fig. 5). Furthermore, and most importantly, the structure showed the 2-acetamido group of the −1 sugar being positioned in such a way that its participation as a nucleophile could be envisioned. Further evidence for neighboring group participation was obtained when chitobiase was proved unable to hydrolyze chito-oligosaccharides lacking a 2-acetamido group on the non-reducing sugar . Mutational studies later confirmed Glu540 as the catalytic acid/base and also indicated an important role for Asp539 in substrate binding and positioning of the 2-acetamido group for nucleophilic attack on the anomeric carbon . The role of Asp539 has been confirmed by studies of the corresponding amino acid in other GH20s (e.g. SpHex from Streptomyces plicatus) where the role of this residue in activation of the 2-acetamido group has been convincingly demonstrated [105, 106]. In an excellent study published in 2005, Vocadlo and Withers  studied α-deuterium kinetic isotope effects to show that the reaction intermediate indeed is a covalent dicyclic oxazoline intermediate rather than a stabilized oxocarbenium ion as has sometimes been proposed (e.g. [30, 61, 107]). Notably, accumulated experimental data on the catalytic mechanism of GH20 enzymes [22, 50, 52, 54, 105, 106, 108-110] have been instrumental for our current understanding of the substrate-assisted mechanism used by other enzymes hydrolyzing GlcNAc containing glycosides, including GH18 chitinases.
CBP21, an example of a family CBM33 lytic polysaccharide monooxygenase
During the initial characterization of the S. marcescens chitinolytic proteins, a small (~ 21 kDa) and abundant secreted protein was identified . Although no activity could be specifically related to this protein, it was still reported as ‘chitinolytic’. The gene encoding this protein was later cloned, and the protein was expressed, purified and partially characterized by Suzuki et al. . The latter study showed that the protein had strong affinity for β-chitin, which also inspired the name (chitin binding protein 21 or CBP21; the number 21 reflects the estimated molecular weight in kDa). While Suzuki et al.  could not detect any enzymatic activity for this protein, it was clear that CBP21 was an important part of the chitinolytic machinery as it was secreted in large amounts by S. marcescens during chitin degradation and since its production was co-regulated with chitinase production and induced by (GlcNAc)2 [17, 111].
At the time of its discovery, CBP21 was one of few proteins identified and characterized in the CBM33 family. The first proteins studied from this family were all isolated and cloned from various Streptomyces strains, a major effort carried out by the Schrempf group of Osnabrück University. The first family 33 CBM to be isolated and characterized was CHB1 from Streptomyces olivaceoviridis . CHB1 was shown to bind strongly to α-chitin and was also observed to bind to fungal hyphae. Since CHB1 had no apparent hydrolytic activity, it was proposed to aid the Streptomyces chitinases in chitin degradation by invading and loosening the chitin structure. The presence of several conserved aromatic amino acids led Schnellmann et al.  to propose a chitin binding function for these residues, primarily because it was known at the time that the interaction surfaces of carbohydrate binding enzymes or modules often contained aromatic residues that are crucial for binding (see [38, 113] for reviews). However, it was shown later that most of these aromatic amino acids are located in the hydrophobic core of CBM33s which makes a role in substrate binding highly unlikely . Since 1994, several other family 33 CBMs have been isolated and characterized, showing binding specificities for both α- and β-chitin [114-121]. Prior to 2005, the functional roles of these CBM33s were unknown, but they were generally thought to play a role in chitin degradation as they were found to be co-expressed with chitinases [111, 122, 123].
The year 2005 was a turning point for our understanding of how CBM33s work, and CBP21 played a central role in obtaining this understanding. The first clues were provided by the X-ray crystallographic structure of CBP21 (the first of a CBM33-type protein) that was accompanied by data exploring the contribution of conserved solvent-exposed residues to substrate binding . Remarkably, while most CBMs carry solvent-exposed aromatic amino acids mediating binding to carbohydrates, the structure of CBP21 showed that the conserved binding surface of this protein is dominated by polar residues. A second study on CBP21 published shortly after then showed that CBP21 was able to boost the degradation of chitin by chitinases, presumably by increasing the accessibility of the substrate, although the mechanism was unknown at the time . This mechanism was unraveled in a groundbreaking study in 2010, demonstrating that CBP21 is an enzyme capable of depolymerizing chitin chains through an oxidative mechanism  (the mechanism will be discussed in detail below). Subsequently, similar enzymatic activities have been demonstrated for other members of the CBM33 family, including CBM33s that are active on cellulose . After some initial confusion concerning the naming of these novel enzymes, most workers in the field have now adapted the term lytic polysaccharide monooxygenases (LPMOs). Interestingly, it has been shown that proteins originally classified as glycoside hydrolase family 61 (GH61)  are also LPMOs  (see below).
The structure of CBP21
CBP21 is a monomeric enzyme with a compact distorted β-sandwich fold/architecture consisting of two β-sheets stabilized by two disulfide bridges and a hydrophobic core consisting of multiple tryptophans (Fig. 7). The CATH database classifies CBP21 as having a ‘coagulation factor XIII’ topology (Table 1). A stretch of 62 residues joining β-strands 1 and 2 contains three helices (one α- and two 310-helices) that constitute most of the surface that hosts conserved residues involved in both substrate binding and the active site. An important feature of the structure is the strictly conserved N-terminal histidine. The imidazole side chain and the N-terminal amino group of this conserved residue constitute a copper binding site in conjunction with another strictly conserved histidine residue (His114 in CBP21; Fig. 7). The binding surface is flat, i.e. without any grooves or pockets that could harbor a carbohydrate moiety. Multiple conserved, mostly polar residues cover the substrate binding surface. A selection of these residues have been experimentally shown to be important for substrate binding  and recently the binding surface has also been mapped by NMR methods  (Fig. 7). While the binding surface is dominated by polar side chains, both mutagenesis and NMR studies have shown that the only solvent exposed aromatic amino acid (Tyr54) is also important for substrate binding.
Catalysis by CBM33-type LPMOs
In the 2010 study unraveling the enzymatic activity of CBP21, it was shown that CBP21 is a metal-dependent enzyme that cleaves chitin chains in the presence of an external electron donor and molecular oxygen . This reaction leads to generation of a regular non-reducing chain end and an oxidized chain end (Fig. 8). The oxidized chain end appears as an aldonic acid in solution, and isotope labeling experiments showed that the two carboxylic oxygens in this product are derived from molecular oxygen and water, respectively (Fig. 8). Most probably, the reaction produces a δ-1,5 lactone as the initial oxidized product, which at physiological pH will be hydrolyzed to an aldonic acid (the equilibrium is driven towards the lactone form at acidic pH; indeed, at lower pH both types of products can be observed ).
Addition of a metal ion chelator (EDTA) or mutation of the metal binding His114 led to inactivation of CBP21, demonstrating the enzyme's metal dependence [34, 124]. Initial studies on both CBP21  and a GH61-type LPMO  led to the idea that these enzymes could function with a wide variety of divalent metal ions. This is remarkable, since one would expect these otherwise cofactor-free redox enzymes to contain a redox-active metal. However, subsequent studies on several LPMOs [127, 129-131] and on CBP21 itself  convincingly showed that the enzymes in fact are copper-dependent (the initial studies most probably overlooked the presence of sufficient ‘background’ amounts of copper in buffers, solvents and the substrates used).
A recent NMR study of CBP21  showed that the enzyme binds Cu+ more strongly than Cu2+ with dissociation constants of ~ 1 nm and 55 nm, respectively. Thus CBP21 is able to protect once reduced copper ions from oxidation (Fig. 8). These observations suggest a sequence of events that is compatible with a catalytic mechanism for LPMOs that has been suggested by Phillips et al.  and Beeson et al. : the copper ion is reduced on the enzyme, which leads to increased affinity for molecular oxygen . Upon oxygen binding, an electron is transferred from Cu+ to O2, generating a superoxo intermediate that may initiate the reaction by abstracting a hydrogen from the substrate (Fig. 8).
Interestingly, when applying the reaction conditions discovered in 2010  the boosting effect of CBP21 on chitinase activity turned out to be much larger than was observed in the 2005 study  that first described this effect. It is believed that CBP21 acts on regions of the substrate (e.g. crystalline surfaces) that are likely to be the least accessible for chitinases. By introducing chain breaks CBP21 generates new attachment points for chitinases. Furthermore, the introduction of negative charge is likely to disrupt crystal packing which would augment the beneficial effect of CBP21 on substrate accessibility. Indeed, under optimal conditions, CBP21 has a tremendous effect on chitin particles, which become highly amorphous (see Fig. 2C in ).
The synergistic action of CBP21 and each of the individual chitinases from S. marcescens  suggests that CBP21 has a fundamentally different role than the GH18 glycoside hydrolases. To illustrate this difference, Vaaje-Kolstad et al.  compared the product spectrum generated by CBP21 and non-processive endo-acting ChiC. Both enzymes are thought to act randomly on chitin chains in an ‘endo’ fashion and are thus expected to generate longer oligosaccharides. If the same chain is cut twice, products that are sufficiently short to be detectable by mass spectrometry will emerge. In this experiment, chitin was incubated with either ChiC or CBP21, each supplemented with a chitin deacetylase that, by deacetylating GlcNAc, increases product solubility. Figure 9 shows that the two enzymes yield fundamentally different product profiles. The products generated by CBP21 show a clear dominance of even-numbered oligomers. Keeping in mind that the repeating unit of a chitin chain is chitobiose, production of an even-numbered product implies that the chitin chain from which this product originates has been attacked twice from the same side. This is compatible with this chitin chain being part of some sort of crystalline packing, where only one face of the chain is accessible. On the other hand, the products released by ChiC represent a continuum of lengths, indicating that ChiC approaches its substrate from ‘any side’ as could be the case if ChiC were to act on the more amorphous regions of the substrate.
Similarities between the S. marcescens chitinolytic machinery and cellulolytic enzyme systems
Figure 10 provides a schematic overview of the chitinolytic machinery of S. marcescens that is based on the data discussed above. Several studies have shown that the enzymes discussed above need to be combined to achieve efficient chitin degradation [99, 124, 134]. Similar enzyme systems exist for cellulose, an important insoluble plant polymer composed of β-1,4-linked glucose . Enzymes degrading chitin or cellulose face similar challenges, and the Serratia enzymes ChiA, ChiB and ChiC and chitobiase have obvious counterparts in processive cellobiohydrolases attacking the cellulose reducing (GH7, GH48) and non-reducing (GH6) ends, endo-cellulases (e.g. GH5, GH7, GH9 and GH12) and beta-glucosidases (GH1, GH3), respectively .
Interestingly, ChiA and ChiB, which act processively in opposite directions, both belong to GH family 18 and use a similar, retaining enzyme mechanism for hydrolysis. Processive cellulases from the same GH family, on the other hand, seem to degrade cellulose in one direction only. For degradation from the reducing end cellulases use both retaining (GH7) and inverting (GH48) mechanisms. However, for degradation from the non-reducing end only the inverting mechanism (GH6) has been observed. Glycoside hydrolysis by a classical (non-substrate-assisted) retaining mechanism as used by cellulases involves formation of a covalent enzyme substrate intermediate where the enzyme is linked to the C1 carbon of the glucose at what will become the newly generated reducing end [69, 136]. In the case of a cellobiohydrolase working from the non-reducing end, this intermediate would contain the product, cellobiose, whereas the remaining polymer, which needs to remain associated with the enzyme during processive action, in fact would need to be displaced in order to create access for the water molecule used to hydrolyze the intermediate. It is thus conceivable that such covalent intermediate binding of cellobiose could interfere with a processive mechanism, which might explain why only the inverting mechanism has been observed for cellobiohydrolases that are thought to act from the non-reducing end. Similar reasoning could in principle apply to ChiB, where the intermediate state would entail formation of an oxazolinium ion of the product, chitobiose. However, in this case it is not the enzyme but the N-acetyl group of the substrate that binds to the C1 carbon of the new reducing end, which, apparently, makes a difference relative to cellulases.
Recently, it has become clear that also CBP21 has counterparts in cellulolytic enzyme systems . So far, cellulose-active LPMOs have been found in fungi, where they are currently classified as GH61 [127, 130, 132, 137, 138], and in bacteria, where they are currently classified as CBM33 . These cellulose-active LPMOs are receiving massive attention because they make enzymatic saccharification of cellulose much more efficient, which is of major importance for the development of second generation biofuels and the bio-economy in general [35, 126]. Indeed, the latest commercial cellulase cocktails such as Cellic CTec2 from Novozymes contain GH61s .
All in all, the past decade has shown that chitin degradation is a useful model system for cellulose degradation. Chitin is easier to work with than cellulose, one reason being the availability of the convenient soluble derivative chitosan. Chitin degradation products are more soluble (or, rather, less insoluble) than cellulose degradation products or can be made more soluble by deacetylation (e.g. Fig. 9). Some of the enzyme systems involved are relatively simple, with less multiplicity (e.g. ), although this certainly is not a general feature (e.g. ). Studies of chitinases have been instrumental in creating deeper understanding of enzyme processivity and the directionality thereof [33, 74, 80, 141] and work on the S. marcescens chitinolytic system [18, 34, 124] has led to the discovery of LPMO activity.
One intriguing issue concerns the pay-off between processivity and intrinsic enzyme speed that has been observed for the GH18 enzymes [33, 74, 80]. Processive cellobiohydrolases are dominant components in natural and commercial cellulose mixtures and are generally considered to be very important in the degradation of the least accessible forms of cellulose. Extrapolating from the chitinase data, it is conceivable that these enzymes suffer from ‘intrinsic slowness’, which in part may be due to the tendency of the enzymes to get stuck on the substrate [142-144]. With current knowledge at hand, one may speculate that more efficient enzyme mixtures may be developed by reducing the amount of processive ‘sticky’ enzymes while at the same time developing strategies that increase accessibility for non-processive cellulases, e.g. by increased use of LPMOs.
The chitinolytic machinery of S. marcescens, depicted in Fig. 10, represents a well-studied, minimal, but seemingly complete enzyme system for degradation of insoluble polysaccharides. Studies of this enzyme system have provided novel insights into how the catalytic centers of GHs are differently shaped for conducting complementary tasks, and have led to the discovery of LPMOs.
Accumulated data for the three chitinases show that the most important difference between ChiA and ChiB on the one hand and ChiC on the other hand relates to the enzymes being processive or not, rather than to the enzymes acting exo or endo, respectively. The studies with chitosan show that ChiA and ChiB can bind in an endo mode  but their deep and rather ‘closed’ substrate binding clefts will normally prevent them from doing so on highly inaccessible substrates. Thus, on a real insoluble substrate such as crystalline chitin, ChiA and ChiB will have a strong preference for binding to chain ends, as is indeed observed . Obviously any further (processive) cleavage resulting from an initial productive enzyme–substrate association event is an ‘exo’ cleavage.
As to processivity, the studies with chitosan lead to one important conclusion that so far has not been discussed very much in the literature: processivity is not directly related to catalysis and cleavage of a glycoside bond, but is rather a result of limited diffusion that is controlled by the substrate binding surface of the enzyme (e.g. specific tryptophans, as seen in ChiA and ChiB). The chitosan data clearly show that the formation of a non-productive enzyme–substrate complex is not necessarily followed by substrate release. Instead, the substrate is displaced while remaining associated with the enzyme; thus an initially non-productive binding mode may eventually be reorganized into a productive binding mode.
The 2005 discovery of the chitinase-boosting activity of CBP21 presented the first clear example of a substrate-disrupting protein, the existence of which had been predicted as early as 1950 during studies on cellulose degradation . At the time (in 2005), CBP21 was thought to be non-catalytic. Already in 1969, when screening for an efficient chitin degrading organism, Reese and co-workers pointed out that the so-called prehydrolytic factor (C1) that they had anticipated to exist in cellulolytic enzyme systems was likely to also be part of chitinolytic enzyme systems . Today, we know that Reese et al. were right and that these ‘prehydrolytic factors’ are LPMOs.
Importantly, almost all fundamental studies of chitinases, and also many studies of cellulases, are carried out using non-natural substrates, i.e. substrates that have been processed in some way. For example, in shrimp shells, chitin is part of a complex heteropolymeric structure rich in proteins and minerals . Chitinous fungal cell walls are also complex heteropolymeric structures  (likewise, plant cell walls are highly complex heteropolymeric structures). In laboratory experiments, one usually uses deproteinized and demineralized chitins as substrates. Thus, most of what we know about chitinolytic enzyme systems today concerns the ability of these systems to convert model substrates. While this by no means disqualifies current knowledge, it should be noted that effective implementation of this knowledge in for example biorefining of chitinous biomass will require additional research. Most probably, when working with real non-processed or hardly processed biomass, additional enzymes such as proteases or additional GHs are needed.
We thank Jerry Ståhlberg for helpful discussions concerning analogies between chitinases and cellulases. Our research on chitinases has been supported by various grants from the Research Council of Norway, most recently by grants 186946 and 196885 (to V.G.H.E.), 209335 and 177542 (to M.S.) and 214138 (to G.V.-K.).