O-Glycosylation as a Novel Control Mechanism of Peptidoglycan Hydrolase Activity*

Background: A range of peptidoglycan hydrolases (PGHs) contain low complexity domains of unknown function. Results: O-Glycosylation of the low complexity domain of the Lactobacillus plantarum autolysin Acm2 is a major negative modulator of enzymatic activity. Conclusion: O-Glycosylation represents an autoregulatory control mechanism of PGH activity. Significance: This is the first functional evidence that glycosylation controls the activity of a bacterial enzyme. Acm2, the major autolysin of Lactobacillus plantarum, is a tripartite protein. Its catalytic domain is surrounded by an O-glycosylated N-terminal region rich in Ala, Ser, and Thr (AST domain), which is of low complexity and unknown function, and a C-terminal region composed of five SH3b peptidoglycan (PG) binding domains. Here, we investigate the contribution of these two accessory domains and of O-glycosylation to Acm2 functionality. We demonstrate that Acm2 is an N-acetylglucosaminidase and identify the pattern of O-glycosylation (21 mono-N-acetylglucosamines) of its AST domain. The O-glycosylation process is species-specific as Acm2 purified from Lactococcus lactis is not glycosylated. We therefore explored the functional role of O-glycosylation by purifying different truncated versions of Acm2 that were either glycosylated or non-glycosylated. We show that SH3b domains are able to bind PG and are responsible for Acm2 targeting to the septum of dividing cells, whereas the AST domain and its O-glycosylation are not involved in this process. Notably, our data reveal that the lack of O-glycosylation of the AST domain significantly increases Acm2 enzymatic activity, whereas removal of SH3b PG binding domains dramatically reduces this activity. Based on this antagonistic role, we propose a model in which access of the Acm2 catalytic domain to its substrate may be hindered by the AST domain where O-glycosylation changes its conformation and/or mediates interdomain interactions. To the best of our knowledge, this is the first time that O-glycosylation is shown to control the activity of a bacterial enzyme.

Growth and division are two essential steps in the cell cycle of bacteria. To be able to achieve such processes, bacteria use specific enzymes called peptidoglycan hydrolases (PGHs) 6 that are required to remodel their protective cell wall, which contains peptidoglycan (PG) as a major component. This polymer is composed of glycan strands made of the repeating disaccharide N-acetylmuramic acid-(␤-1,4)-N-acetylglucosamine (MurNAc-GlcNAc) connected by peptidic stems that are linked to MurNAc (1,2). Depending on their specificities, PGHs cleave different PG bonds. Three types of enzymes, termed N-acetylglucosaminidases, N-acetylmuramidases, and lytic transglycosylases, cleave ␤-1,4 bonds of PG strands. In addition to these enzymes that target the glycan chains, the N-acetylmuramoyl-L-alanine amidases cleave the amide bond between the lactic acid side chain of MurNAc and L-Ala of the stem peptide, whereas carboxy-and endopeptidases cleave the peptidic stem or interpeptide bridge (3).
Recently, we investigated the functional role of the 12 predicted PGHs of Lactobacillus plantarum, a commensal species in the gastrointestinal tract of humans, by systematic gene deletion (4). This study revealed that a restricted set of four PGHs (Acm2, Lys2, LytA, and LytH) are key actors for morphogenesis, autolysis, and/or peptidoglycan composition (4). Among these PGHs, Acm2 was shown to be the major autolysin of L. plantarum required for the last step of cell separation during the septation process (4,5). In addition, we demonstrated previously that Acm2 activity is inhibited by GlcNAc O-acetylation of glycan chains and altered by the absence of D-alanylation of teichoic acids, which results in cell perforations at the septum (1,5). The Acm2 protein (Lp_2645) is a putative N-acetylglucosaminidase characterized by a modular structure composed of three domains. The catalytic domain that belongs to the endo-␤-N-acetylglucosaminidase family (CAZy family GH73) (6) is associated with an N-terminal domain rich in Ala, Ser, and Thr (AST domain) and with a C-terminal domain composed of five SH3b (SH3_5) PG binding domains (PFAM08460) (4,7,8). The functional role(s) of the association of such domains has not been investigated so far in any PGH. Moreover, the role of AST and SH3b domains remains poorly understood except for the importance of SH3b domains of PGHs for PG binding (3, 9 -11). Concerning the accessory AST domain of Acm2, other PGHs contain similar low complexity domains composed of repetitive sequences of Ala, Ser, Thr, or Pro residues such as AcmB of Lactococcus lactis, AtlA of Enterococcus faecalis, and p75/Msp1 from Lactobacillus casei or Lactobacillus rhamnosus (12)(13)(14)(15). The function of these repetitive sequences is currently unknown, but they are hypothesized to be involved in cell wall binding or to serve as a cell wall spacer allowing exposure of the catalytic domain at the cellular surface (3,14). Recently, two studies have demonstrated that the low complexity domains of Acm2 and p75/Msp1 are O-glycosylated. The AST domain of Acm2 was shown to contain at least 15 modified sites with N-acetylhexosamine (HexNAc), most likely N-acetylglucosamine based on its recognition by the wheat germ agglutinin, whereas the exact number of O-glycosylated residues and nature of hexoses (possibly mannose) for p75/Msp1 was not investigated in detail (7,16). O-Glycosylation of proteins in eukaryotes has been studied extensively and occurs on a broad range of proteins to confer protection against proteases, to target proteins to their proper cellular localization, to modulate protein-protein interactions, and to increase thermodynamic protein stability or solubility (17). By contrast, little is known about the function of prokaryotic O-glycosylation. Previous studies have shown that this modification mainly occurs on extracellular proteins such as subunits of S-layer, adhesins, flagella, and type IV pili where it is required for either proper bacterial adhesion, motility, or virulence (18 -21). It was also shown that some enzymes involved in the degradation of polysaccharides (e.g. xylanases) or PG as reported above for Acm2 and Msp1/p75 were O-glycosylated, but no function could be assigned except for Msp1/p75 for which O-glycosylation confers a protection against proteases (12,16,(22)(23)(24).
The aim of this study was to investigate the contribution of the different accessory domains and O-glycosylation of Acm2 to its functionality. For this purpose, truncated variants of Acm2 either deleted of its N-terminal AST domain or its C-terminal 5ϫ SH3b PG binding domain were purified in a glycosylated or non-glycosylated form. Our results revealed that these two accessory domains play distinct roles: the SH3b-containing domain is essential for the in vivo functionality of Acm2, PGbinding ability, and enzymatic activity, whereas the N-terminal AST domain and its glycosylation negatively regulate Acm2 activity. Notably, glycosylation of the AST domain not only contributes to its proteolysis resistance but is a major negative effector of Acm2 enzymatic activity, revealing a novel control mechanism of PG-degrading enzymes.

Bacterial Strains, Plasmids, and Growth Conditions
The bacterial strains and plasmids used in the present study are listed in Table 1. Plasmids were constructed in L. lactis subsp. cremoris NZ3900. L. plantarum and L. lactis were grown in MRS broth (Difco) and in M17 broth (BD Biosciences) containing 0.5% glucose (M17-glucose) at 30°C, respectively. When appropriate, 10 g/ml chloramphenicol was added to the media. Solid agar plates were prepared by adding 2% (w/v) agar to the medium.

DNA Techniques and Electrotransformation
General molecular biology techniques were performed according to the instructions given by Sambrook et al. (25). Electrocompetent L. plantarum and L. lactis cells were prepared as described previously (4,26). PCR were performed with Phusion high fidelity DNA polymerase (Finnzymes, Espoo, Finland) in GeneAmp PCR system 2400 (Applied Biosystems, Foster City, CA). The primers used in this study were purchased from Eurogentec (Seraing, Belgium) and are listed in supplemental Table S1.

Construction of the Complementation Vectors
The expression vectors harboring the different truncated forms of acm2 were constructed from pGITR0010, which contains the complete acm2 ORF under the control of nisA promoter (P nisA ), allowing induction of its transcription by the addition of nisin (27). A reverse PCR strategy was applied on plasmid pGITR0010 using primers pairs EcoRI_Delta_AST_ UP/EcoRI_Delta_AST_DO and EcoRI_Delta_5_SH3_UP/ EcoRI_Delta_5_SH3_DO to remove the AST domain (Acm2 ⌬AST; deletion between amino acids 50 and 243) or the five SH3b domains (Acm2 ⌬SH3; deletion between amino acids 428 and 785; see Fig. 3A) of Acm2, respectively (supplemental Table  S1). Subsequently, the resulting amplicons were digested with EcoRI, self-ligated, and transformed to L. lactis NZ3900 by electroporation. The sequences of the two resulting plasmids, pGITR0012 (Acm2 ⌬AST) and pGITR0013 (Acm2 ⌬SH3), were verified, and they were transformed to L. plantarum TR0010 (Acm2 Ϫ ) by electroporation and subsequently used for complementation studies.

Construction of Expression Vectors for Acm2 Purification
For purification purposes, Acm2 and its derivatives devoid of their putative signal peptide were double tagged by an N-terminal His 6 and C-terminal Strep-tag. First, a DNA fragment encoding the mature part of Acm2 (from amino acids 33 to 785) was amplified by PCR from plasmid pGITR0010 using primers N_6XHis_Acm2_NcoI and C_Strep_Acm2_XbaI. The forward and reverse primers contained NcoI and XbaI restriction sites as well as His 6 -and Strep-tag-encoding sequences at their 5Ј-ends, respectively (supplemental Table S1). The resulting amplicon was digested with NcoI and XbaI and cloned into similarly digested pNZ8048, yielding the expression plasmid pGITR0014. Second, to construct different truncated variants of the N-terminal His 6 -and C-terminal Strep-tag-tagged version of Acm2 (named rAcm2), a reverse PCR strategy was set up using plasmid pGITR0014 as template and primer pairs 6XHis_Strep_Delta_AST_UP and 6XHis_Strep_Delta_AST_DO, 6XHis_Strep_Delta_5_SH3_UP and 6XHis_Strep_Delta_5_ SH3_DO, and 6XHis_Strep_AST_UP and 6XHis_Strep_ AST_DO to construct truncated versions of Acm2 lacking the AST domain (rAcm2 ⌬AST), the five SH3b domains (rAcm2 ⌬SH3), and all domains except the AST domain (rAST), respectively (supplemental Table S1). Subsequently, the resulting amplicons were self-ligated and transformed to L. lactis NZ3900 by electroporation. The integrity of the four plasmids was verified by DNA sequencing. The resulting expression vectors, pGITR0014 (rAcm2), pGITR0015 (rAcm2 ⌬AST), pGITR0016 (rAcm2 ⌬SH3), and pGITR0017 (rAST), were extracted from L. lactis NZ3900 and transformed into L. plantarum TR0010 (Acm2 Ϫ ) by electroporation and subsequently used for purification assays.

Purification of rAcm2 and Its Truncated Variants
The various tagged proteins were overexpressed in L. lactis NZ3900 and L. plantarum TR0010 using the nisin-inducible controlled expression system as reported previously (27)(28)(29). Expression was induced (A 600 ϭ 0.3) with nisin A (Sigma-Aldrich) at a concentration of 2.0 and 20 ng/ml for L. lactis and L. plantarum, respectively. Bacteria were grown to midexponential phase (A 600 ϭ 0.8) and harvested by centrifugation at 5,000 ϫ g for 10 min at 4°C. The pellet was washed once with 50 mM potassium phosphate buffer, pH 8.0 and resuspended at an A 600 of 80 in 50 mM potassium phosphate buffer, pH 8.0. Cells were lysed with glass beads using a Precellys cell disrupter (Bertin Technologies) at 5,000 rpm for 5 ϫ 30 s and centrifuged at 20,000 ϫ g for 20 min at 4°C. The His 6 -and Strep-tag-tagged proteins were purified successively by affinity chromatography on Ni 2ϩ with the ProBond TM Purification System (Invitrogen) and with Strep-Tactin Superflow 1-ml columns (IBA) according to the instructions of the manufacturers. The purity of the tagged proteins was checked by SDS-PAGE. The pure proteins were concentrated by ultrafiltration with an Amicon Ultra-0.5 (Ultracel-30 membrane, 30-kilodalton (kDa) cutoff) (Millipore). Protein concentrations were measured using the following three independent methods: Bio-Rad Protein Assay based on the method of Bradford (56), absorbance assay at 280 nm, and finally NanoOrange Protein Quantitation kit (Invitrogen). The pure proteins were stored at 4°C.

Determination of Acm2 Hydrolytic Specificity by Peptidoglycan Digestion
PG was extracted from L. plantarum EB004 as described previously (1). PG (2 mg, dry weight) was then incubated with pure rAcm2 (80 g) at 37°C for 24 h at pH 6.5. Insoluble material was removed by centrifugation. Half of the supernatant containing soluble muropeptides was kept for direct analysis, and the other half was further digested with mutanolysin as described previously (30). The samples were then boiled for 3 min to stop reactions. The soluble muropeptides were reduced with sodium borohydride, separated by reverse phase high performance liquid chromatography (HPLC), and analyzed by matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) MS as reported previously (30). They were identified according to the previously published reference chromatogram (1).

Acm2 Activity Assay Using L. plantarum Dead Cells
NZ7100 (WT) or EB003 (OatB Ϫ ) was used as substrate. L. plantarum strains were grown in MRS medium to midexpo-nential phase (A 600 ϭ 0.8). Cells were harvested by centrifugation at 5,000 ϫ g for 10 min at 4°C, washed twice with deionized H 2 O, resuspended at an A 600 of 200 in deionized H 2 O, autoclaved at 120°C for 20 min, and stored at 4°C. Autoclaved cells were diluted in 50 mM Tris-HCl, pH 7.0 to an A 600 of 0.8. rAcm2 and its variants were each added at a concentration of 0.3 nM in a final volume of 300 l, and the turbidity of the cell suspension was monitored by measuring the A 600 every 10 min with a Varioskan Flash multimode reader (Thermo Fisher Scientific). Acm2 activity was expressed as the percent decrease in A 600 . One arbitrary enzymatic unit was defined as 1% A 600 decrease per hour at a protein concentration of 1 M.

Acm2 Activity Assay Using Micrococcus lysodeikticus Cell Walls
The fluorescence assay was performed using The EnzChek Lysozyme Assay kit (Molecular Probes) according to the manufacturer's manual. Briefly, the rAcm2 activity (0.3 nM) was measured on M. lysodeikticus cell walls that were labeled with fluorescein to such a degree that fluorescence is quenched. The PG-hydrolyzing activity of Acm2 relieved the quenching, resulting in an increase in fluorescence proportional to its activity. Fluorescence was measured with a Varioskan Flash multimode reader (Thermo Fisher Scientific) (excitation/emission maxima, ϳ490/525 nm).

Zymogram Analysis
SDS-PAGE was performed with 8% (w/v) polyacrylamide separating gels. Renaturing SDS-PAGE was performed as described previously (1). The polyacrylamide gels contained L. plantarum EB003 autoclaved dead cells resuspended at A 600 of 0.8 as enzyme substrates. Whole cell extracts or purified proteins were used as samples. Whole cell extracts were prepared as described previously (1). Samples were boiled in denaturing sample buffer and centrifuged for 1 min at 20,000 ϫ g prior to loading. After sample migration, gels were washed for 30 min in deionized H 2 O and incubated for 16 h at room temperature in 50 mM Tris-HCl, pH 6.8, 1 mM DTT containing 0.1% (v/v) Triton X-100. Subsequently, the gels were washed in deionized H 2 O followed by staining with 0.1% methylene blue in 0.01% (w/v) KOH and destaining in deionized H 2 O.

Triton X-100-induced Autolysis in Buffer Solution
L. plantarum strains were grown in MRS medium to midexponential phase (A 600 ϭ 0.8). Cells were harvested by centrifugation at 5,000 ϫ g for 10 min at 4°C; washed once with 50 mM potassium phosphate buffer, pH 7.0; and resuspended at an A 600 of 1.0 in 50 mM potassium phosphate buffer, pH 7.0 supplemented with 0.05% Triton X-100 (1,31). Cell suspensions were transferred into 96-well sterile microplates with a transparent bottom (Greiner, Alphen a/d Rjin, The Netherlands) and incubated at 30°C. Autolysis was monitored by measuring the A 600 of the cell suspensions every 20 min for 5 h with a Varioskan Flash multimode reader (Thermo Fisher Scientific). The extent of autolysis was expressed as the percent decrease in A 600 .

Proteolysis Resistance Experiments
Glycosylated and non-glycosylated AST domains (rAST Lp and rAST Lc , respectively) were incubated for 15 min at 25°C with trypsin at various final concentrations (no trypsin and 1, 10, and 100 g) in a 50 mM Tris-HCl, pH 8 buffer. The reactions were stopped by adding NuPAGE lithium dodecyl sulfate sample buffer (4ϫ; Invitrogen) and boiling at 95°C for 10 min. The reaction products were analyzed by SDS-PAGE (8% (w/v) acrylamide).

Mass Spectrometry Analysis
The total mass of rAST Lp and rAST Lc was measured by MALDI-TOF. 1 l of each protein sample was mixed with 1 l of sinapic acid (4 mg/ml in 70% acetonitrile (ACN), 0.1% trifluoroacetic acid (TFA)). 1 l of the mixture was spotted on a MALDI target plate. The spotted plated was analyzed in linear mode on an Applied Biosystems 4800 MALDI-TOF/TOF Analyzer using a 200-Hz solid state laser operating at 355 nm. MS spectra of the intact proteins were obtained using a laser intensity of 6,000 and 4,000 laser shots per spot in the m/z range of 10,000 -100,000.

Characterization of the Glycosylation Pattern by Mass Spectrometry
Protein Precipitation and Digestion-10 l of each protein sample was precipitated with the methanol/chloroform/water method described by Wessel and Flügge (32). Precipitated proteins were solubilized in 20 l of 100 mM triethylammonium bicarbonate buffer (Sigma-Aldrich), pH 8.0 by sonication for 5 min in a bath sonicator (Bioruptor, Diagenode). The proteins were digested with trypsin (Promega) for 16 h at 37°C. After digestion, samples were vacuum-dried (SpeedVac SC200, Savant).
Mass Spectrometry Analysis-The spotted plates were analyzed in reflector mode on an Applied Biosystems 4800 MALDI-TOF/TOF Analyzer using a 200-Hz solid state laser operating at 355 nm. MS spectra were obtained using a laser intensity of 3,600 and 2,000 laser shots per spot in the m/z range of 800 -4,000, whereas MS-MS spectra were obtained by automatic selection of the 20 most intense precursor ions per spot using a laser intensity of 4,000 and 2,000 laser shots per precursor. Collision-induced dissociation was performed with an energy of 1 kV with air gas at a pressure of 1 ϫ 10 6 torrs. Data were collected using Applied Biosystems 4000 Series Explorer TM software.
LC/MS-MS data were processed using Applied Biosystems GPS Explorer TM 3.6 software. For peptide identification, a local database containing the Acm2 sequence was used. Tolerance was set to 200 ppm on the precursors and 0.3 Da on the fragments. One trypsin miscleavage was authorized. For modifications, methionine oxidation and HexNAc glycosylation (203.08 Da) on Ser, Thr, and Asn were selected. HexNAc-modified peptides were checked by manual de novo sequencing on the MS-MS fragmentation spectra.

Fluorescence Microscopy
Conjugation reactions between Alexa Fluor 514 Carboxylic Acid, Succinimidyl Ester, Mixed Isomers (1-mg unit size; Invitrogen) and purified Acm2 or its derivatives were performed according to the manufacturer's instructions. L. plantarum cells were grown to mid exponential phase (A 600 ϭ 0.8). 2 ml of cells were pelleted; washed once with 50 mM phosphate buffer, pH 5.5; resuspended in 1 ml of the same buffer; and finally incubated for 30 min at room temperature with the dye-conjugated proteins. Cells were then pelleted; washed once with 50 mM phosphate buffer, pH 5.5; and resuspended in the same buffer. 5 l of cells was adsorbed on a polylysine-precoated slide for 30 min at room temperature and observed using an Axio Observer Z1 inverted microscope. Micrograph analyses were performed with AxioVision 4.8 software (Carl Zeiss).

Preparation of Peptidoglycan Model Surfaces and Functionalized Atomic Force Microscopy (AFM) Tips
PG from Bacillus subtilis (Sigma) was covalently immobilized onto gold surfaces. Silicon wafers coated by thermal evaporation with a thin layer of chromium (around 5 nm) followed by a thin layer of gold (around 30 nm) were rinsed in ethanol, cleaned for 15 min by UV-ozone treatment, rinsed in ethanol, and dried with N 2 . The clean surfaces were immersed overnight in a 1 mM thiol solution of 90% 11-mercapto-1-undecanol (Sigma) and 10% 16-mercaptohexadecanoic acid (Sigma), rinsed in ethanol, briefly sonicated in ethanol solution, and dried with N 2 . The supports were then immersed for 30 min into a solution of 10 g/liter N-hydroxysuccinimide and 25 g/liter N-(3-dimethylaminopropyl)-NЈ-ethylcarbodiimide hydrochloride (Sigma), rinsed with Milli-Q water, incubated for 1 h with 50 g/ml PG solution, rinsed with buffer, and immediately used. AFM gold-coated cantilever (OMCL-TR400PB-1, AtomicForce, Olympus) were rinsed with ethanol, cleaned for 15 min with UV-ozone, rinsed with ethanol, dried with a gentle flow of N 2 , and immersed overnight in a 0.1 mM solution of 90% HS-C 11 -(ethylene glycol) 3 -OH thiols (ProChimia) and 10% HS-C 11 -(ethylene glycol) 3 -nitrilotriacetic acid thiols (ProChimia). AFM tips were then rinsed with ethanol, dried with N 2 , and immersed in a 40 mM aqueous solution of NiSO 4 , pH 7.2 for 30 min. Tips were then incubated in a 200-l droplet of a 100 g/ml His 6 -tagged protein solution for 1 h, rinsed, and stored in buffer (PBS implemented with 150 mM NaCl to avoid protein precipitation) until use.

AFM Measurements
AFM images and force curves were obtained using a Nanoscope V multimode AFM (Bruker Corp., Santa Barbara, CA) at room temperature. After rinsing in buffer, the model surfaces were mounted on an AFM magnetic steel puck using double sided tape and transferred to the AFM liquid cell while avoiding dewetting. All force curves were measured with a maximum applied force of 250 piconewtons using a constant approach and retraction speed of 1,000 nm/s using cantilevers with a nominal spring constant of ϳ0.02 newton/m (AtomicForce, Olympus). Measurements were taken at three different spots on the model surfaces (n Ͼ 200). Prior to each experiment, the presence and uniformity of the PG layer was confirmed through a scratching test where the layer was first imaged in liquid medium under large force (Ͼ10 nanonewtons) on a 1 ϫ 1-m 2 area followed by recording of an image on a larger area (5 ϫ 5 m 2 ) at lower force. The resulting image allowed determination of the PG layer thickness (ϳ2 nm), which was consistent with earlier data obtained for the same system (33).

RESULTS
The Major Autolysin Acm2 of L. plantarum Is an N-Acetylglucosaminidase-The Acm2 annotation and our previous conclusions based on analysis of PG extracted from an Acm2-deficient mutant strain suggest that Acm2 possesses an N-acetylglucosaminidase activity (4). To directly confirm this hydrolytic specificity, a His 6 -Acm2-Strep-tag variant (rAcm2) was overproduced intracellularly in L. lactis and L. plantarum using the nisin-inducible controlled expression system followed by successive purification by nickel and D-desthiobiotin affinity chromatography. This two-step purification was required to obtain a homogenous preparation of the full-length Acm2 protein (supplemental Figs. S1 and S2) because Acm2 was shown previously to be prone to proteolysis (4). By using renaturing SDS-PAGE (zymogram) with L. plantarum autoclaved cells as substrate, it appeared that the rAcm2 protein was active regardless of the expression host used (see below and data not shown). Subsequently, the hydrolytic specificity of Acm2 was examined by digestion of purified L. plantarum PG. To simplify the identification of muropeptides resulting from Acm2 activity, PG extracted from L. plantarum EB004 (OatA Ϫ OatB Ϫ ) was used as substrate because it is totally depleted of O-acetylation (both MurNAc and GlcNAc) and is of well established composition (1). This PG was incubated with rAcm2 and then with or without mutanolysin (muramidase), and the resulting soluble muropeptides were separated by reverse phase HPLC (Fig. 1). Soluble muropeptides released by Acm2 digestion in the absence or presence of mutanolysin were clearly different (Fig. 1, profiles I and II), showing that Acm2 does not possess muramidase activity. MALDI-TOF analysis of muropeptides that resulted from Acm2 digestion alone, corresponding to peaks A, B, C, and D, generated molecular ions with m/z values of 891.38, 1813.85, 1812.60, and 1883.89, respectively (Fig. 1, profile I, and Table 2). The fraction of soluble muropeptides obtained by the combined digestion with Acm2 and mutanolysin included the appearance of new peaks that were annotated as A 1 Table 2). Taken together, these data conclusively demonstrate that Acm2 displays an N-acetylglucosaminidase activity.
The AST Domain Is O-Glycosylated When Expressed in L. plantarum but Not in L. Lactis-Acm2 has been reported previously to undergo cytoplasmic O-glycosylation in L. plantarum that is restricted to its AST domain (7). To investigate whether the AST domain is decorated with a specific glycosylation pattern and whether glycosylation is conserved when the protein is expressed in a different bacterial background, the AST domain was purified from L. lactis and L. plantarum (rAST Lc and rAST Lp , respectively). When separated by SDS-PAGE, rAST Lp clearly displayed a higher molecular mass as compared with rAST Lc (Fig. 2A, left panel). The total mass of the two variants was evaluated by MALDI-TOF. Although the measured total mass of rAST Lc corresponds to the calculated mass of the non-glycosylated rAST domain (22,772 Da), the most abundant form of rAST Lp displays a measured total mass of 26,607 Da ( Fig. 2A, right panel). Because it was reported previously that the AST domain contains HexNAc (m/z value of 203.08), probably N-acetylglucosamine, we estimated that the predominant form of rAST Lp contains 19 HexNAc residues. The number of HexNAc residues present on the rAST Lp is probably variable according to the broad basis of the rAST Lp peak ( Fig. 2A, right panel). rAST Lp and rAST Lc were trypsinized, and the resulting peptides were separated by reverse phase HPLC and analyzed by MALDI-TOF/TOF mass spectrometry. The analysis of the resulting spectra gave a sequence coverage of the AST domain higher than 95% when purified either from L. plantarum or L. lactis (Fig. 2B). Seven glycopeptides containing HexNAc were identified by the fragmentation process for rAST Lp (Fig. 2B and supplemental Table  S2), whereas no glycopeptides were found for rAST Lc (illustrated for glycopeptide II in supplemental Fig. S3). The same observation was made with trypsinized full-length rAcm2 where no glycopeptides could be identified with rAcm2 purified from L. lactis (data not shown). A total of 21 mono-Hex-NAc glycosylation sites were identified on 15 Ser and 6 Thr residues, confirming that glycosylation of Acm2 was O-linked (Fig. 2C). However, despite a high sequence coverage that allowed us to identify precisely each glycosylation site, we were unable to define a unique motif surrounding Ser and Thr residues that could determine the specificity for O-glycosylation. Nevertheless, this analysis revealed that glycosylated Ser/Thr residues at 12 of 21 positions are preceded either by threonine or alanine, defining two weakly conserved glycosylation motifs (Fig. 2C, right panel). Importantly, this analysis showed that a non-glycosylated variant of rAST and full-length rAcm2 could be produced in vivo for functional studies when purified from a heterologous expression host like L. lactis.
SH3b Binding Domains Are Required for Appropriate Cell Separation and Septal Localization of Acm2-In a previous study, we showed that an L. plantarum mutant strain depleted of Acm2 exhibited a chaining phenotype (4). This result suggested that Acm2 was involved in the last step of cell separation during the septation process (4,5). To determine which domain is required for Acm2 cell-separating activity and proper localization, in vivo complementation and fluorescent labeling experiments were performed, respectively. For in vivo experiments, the acm2 gene deletion mutant was complemented by plasmids encoding Acm2 (positive control), Acm2 depleted of its five SH3b domains located at the C terminus (Acm2 ⌬SH3), or Acm2 without its N-terminal AST domain (Acm2 ⌬AST) (Fig. 3A). Because acm2 mutant cells display a defect in cell separation, the number of cells per chain exhibited by complemented mutant strains was compared with that of the wild type (Fig. 3B). We observed that complementation with Acm2 and Acm2 ⌬AST was able to restore daughter cell separation similarly to that in the wild type, whereas Acm2 ⌬SH3 has a low ability to restore the wild-type phenotype. Indeed, the mean number of cells per chain for complementation with Acm2 ⌬SH3 is ϳ3 compared with ϳ4 in the control acm2 mutant. These results suggest that SH3b domains are required for in vivo Acm2 activity potentially through a lack of targeting at the  Table 2.

Peak
Proposed structure  Fig. S4A and data not shown). The results show that fluorescent rAcm2 localized mainly at the septum of the different strains but with a different patterning: dots in late division cells of wild-type and Oat Ϫ strains, dots or large bands for dividing TagO Ϫ cells, and a diffuse patterning for Cps1-4 Ϫ cells with labeling of the septum, poles, and along the lateral cell wall (Fig. 4, supplemental Fig. S4A, and data not shown). Importantly, rAcm2 ⌬AST displays the same labeling pattern as rAcm2 on wild-type cells, whereas rAcm2 ⌬SH3 showed a complete absence of localization (illustrated for wildtype strain in supplemental Fig. S4B). We also repeated these experiments with non-glycosylated rAcm2 and its variants purified from L. Lactis, which generated results that were similar to those obtained with the glycosylated forms obtained from L. plantarum (data not shown). These experiments reveal that Acm2 (glycosylated or not) mainly localizes in the septum area of dividing cells through its SH3b binding domains and independently of the presence of its AST domain. In addition, cell wall polymers have an impact on Acm2 sublocalization with a dominant role for cell surface

polysaccharides (CPSs) compared with wall teichoic acids (WTAs).
Acm2 Binds PG through its SH3b Domains-In view of the importance of the five SH3b domains for in vivo Acm2 activity and its subcellular localization, we directly investigated whether these domains were implicated in the PG-binding ability of Acm2. For this purpose, Acm2-PG interactions were evaluated by single molecule AFM measurements. rAcm2 and rAcm2 ⌬SH3 purified from L. plantarum (rAcm2 Lp and rAcm2 ⌬SH3 Lp ) were linked to a nitrilotriacetic acid-Ni 2ϩ -terminated AFM tip to probe a model surface of PG, which was covalently attached on supports using N-hydroxysuccinimide/N-(3-dimethylaminopropyl)-NЈ-ethylcarbodiimide hydrochloride chemistry (supplemental Fig. S5). Commercial B. subtilis PG was used as a model surface because its basic composition is identical to that of L. plantarum (meso-diaminopimelic acid direct cross-linking). After validating the model surfaces using AFM imaging (see "Experimental Procedures"), single molecule force spectroscopy measurements were performed at three different spots of the surface for rAcm2 Lp or rAcm2 ⌬SH3 Lp . Fig. 5 shows the resulting adhesion force histograms, rupture length histograms, and representatives force curves for rAcm2 Lp (Fig. 5A) and rAcm2 ⌬SH3 Lp (Fig. 5B). The proportion of adhesive events between the PG surface and Acm2 was 29% with the remaining curves being non-adhesive (Fig. 5A, left panel). The rAcm2 Lp adhesive force histogram showed a maximum located at 50 piconewtons with the majority of the curves being located in the 50 -100-piconewton range (Fig. 5A, left panel). These values correlate with previously reported values for receptorligand complexes at the same loading rate (33,34). The rupture distances were located mostly in the 50 -100-nm range, and the force-distance curves could be fitted with the wormlike chain model, which typically describes the behavior of single protein unfolding (Fig. 5A, right panel) (35). However, in contrast to what we observed for rAcm2 Lp , the number of adhesion events for rAcm2 ⌬SH3 Lp appeared significantly lower than that for the full-length protein (9 versus 29%) (Fig. 5B). These data demonstrate that binding of rAcm2 Lp to PG is mostly mediated by its SH3b domains. Nevertheless, the shape of the representative curves and the very short rupture distances (Ͻ50 nm directly after the contact zone) observed with rAcm2 ⌬SH3 Lp did not exclude a contribution of the AST domain to PG binding (Fig.  5B, right panel). To address this question, adhesion of the purified rAST Lp domain was evaluated in a similar experiment, revealing that the number of adhesion events (5%) and the distances of rupture were lower than those observed for rAcm2 ⌬SH3 Lp (Fig. 5C). This result suggests that the AST domain is not involved in the PG-binding capacity of rAcm2 Lp . Subsequently, we investigated whether the absence of O-glycosylation could modify the PG-binding ability of rAcm2. To this end, AFM experiments were repeated with purified rAcm2 Lc and its mutant harboring the empty plasmid pNZ8048 (Acm2 Ϫ ϩ CTL), acm2 mutant overexpressing acm2 (Acm2 Ϫ ϩ Acm2 ϩ ), or acm2 deleted of either its AST domain (Acm2 Ϫ ϩ ⌬AST ϩ ) or its five SH3b domains (Acm2 Ϫ ϩ ⌬SH3 ϩ ) grown in the presence of nisin (20 ng/ml). Mean values Ϯ95% confidence intervals (bars) (number of chains Ͼ500 for each strain) are shown. Significance is based on a log-linear model (distribution of error, Poisson; link function, log) 2 ϭ 2933.94, df ϭ 4, p Ͻ 0.0001. Letters a, b, and c represent groups that are significantly different at the 5% ␣ level. truncated variants. No significant differences of PG-binding forces could be observed among rAcm2 Lc , rAcm2 Lp , and their derivatives (supplemental Fig. S6). As a control, the specificity of the interactions and the proper attachment of the proteins to the gold tip were also confirmed by recording adhesion curves between an AFM tip terminated with nitrilotriacetic acid-Ni 2ϩ without the addition of the protein and the model surface. The resulting curves display typical adhesion for nonspecific interactions such as electrostatic forces where the adhesive events occur directly after contact (supplemental Fig. S7). Taken together, these results demonstrate the importance of SH3b domains, but not the AST domain or its glycosylation, in recognition and binding of Acm2 to PG.
SH3b Binding Domains and AST Domain Play Antagonistic Roles in the Control of Acm2 Activity-Previous studies reported that subdomains of modular PGHs could positively or negatively affect enzyme activity (36 -40). As a first investigation of the contribution of SH3b domains and of the AST domain to Acm2 activity, we examined Acm2 activity from the complemented strains that were described above using PGH zymography of total cell extracts and Triton X-100-induced autolysis assays with dead cell suspensions. The zymogram experiments showed that the wild-type strain displayed significant amounts of cell wall-associated PGH activity and that truncation of SH3b domains seemed to severely affect cell wallassociated PGH activity because no activity could be detected after overnight renaturation for the mutant strain complemented with the expression construct for Acm2 ⌬SH3 (Fig.  6A). In contrast, the deletion of the AST domain seemed to enhance cell wall-associated PGH activity because a more intense lytic band was detected. The numerous activity bands that were observed likely result from the proteolytic degrada-FIGURE 5. Interaction forces between the different forms of Acm2 and a model PG surface. Adhesion force histograms (left column) and rupture distance histograms (together with representatives curves; right column) obtained by recorded force-distance curves between the protein-functionalized AFM tip and PG using the strategy depicted in supplemental Fig. S5 are shown. Tips were functionalized with the full-length version of Acm2 purified from L. plantarum (rAcm2 Lp ; A), Acm2 truncated for its SH3b domains (rAcm2 ⌬SH3 Lp ; B), and AST domain of Acm2 (rAST Lp ; C). The wormlike chain fit (red) describing the unfolding of rAcm2 Lp is presented for one representative force-distance curve (A, right column, top curve). pN, piconewtons. tion of Acm2 as reported previously (4). Triton X-100-induced autolysis experiments confirmed these observations because Acm2 ⌬SH3 was unable to restore the autolysis observed with the wild type or the mutant complemented with the full-length Acm2, whereas the expression of Acm2 ⌬AST induced a significantly increased autolysis (Fig. 6B). However, these variations in Acm2 activity of the different truncated variants could result from differences in enzymatic activity and/or cell-binding capacity. In addition, different levels of expression of the Acm2 variants cannot be excluded. To clarify these aspects, the activity of the Acm2 variants purified from L. plantarum (rAcm2 Lp , rAcm2 ⌬SH3 Lp , and rAcm2 ⌬AST Lp ) was compared. As a first validation, the activity was examined by PGH zymography. The results showed that purified rAcm2 ⌬AST Lp and rAcm2 Lp proteins display the anticipated single activity band, whereas the rAcm2 ⌬SH3 Lp activity could only be detected after extended renaturation times (7 days) (Fig. 6C). Because zymography is a semiquantitative activity assay, the enzymatic activity was measured by incubating the same amount of purified pro-teins with a suspension of autoclaved dead cells (OatB Ϫ strain) to monitor the turbidity decrease. Cells of the OatB Ϫ strain that are deprived of GlcNAc O-acetylation were shown previously to be more susceptible than WT cells to digestion by Acm2 (1). To ensure that the same amount of proteins was used, their concentrations in all subsequent assays were validated by three different methods that give consistent results (see "Experimental Procedures"). The data obtained showed that the ⌬SH3 variant does not display any hydrolytic activity in contrast to the ⌬AST variant that exhibited a significantly higher activity (ϳ10-fold) than the full-length rAcm2 Lp protein (Table 3 and Fig. 6D). Taken together, these data indicate that SH3b binding domains of Acm2 are not only required for PG binding but are also necessary for Acm2 hydrolytic activity, whereas the AST domain is not essential for the functional role of Acm2 and actually negatively modulates Acm2 activity.
O-Glycosylation Inhibits the PG Hydrolytic Activity of Acm2-To study whether glycosylation has an impact on the PG hydro- lytic activity of Acm2, we incubated the same concentration of rAcm2 Lp (glycosylated) or rAcm2 Lc (non-glycosylated) with a suspension of autoclaved dead cells of L. plantarum (OatB Ϫ strain) and monitored the turbidity decrease. As internal positive controls, rAcm2 ⌬AST Lp and rAcm2 ⌬AST Lc were used because both are not glycosylated and are expected to display a similar enzymatic activity. Remarkably, the results showed that the PGH activity of rAcm2 Lc was ϳ6-fold higher than that of rAcm2 Lp , revealing that the non-glycosylated form of Acm2 is significantly more active than its glycosylated counterpart, whereas the non-glycosylated positive controls (rAcm2 ⌬AST Lp and rAcm2 ⌬AST Lc ) behaved similarly (Table 3 and Fig. 7A). Interestingly, the PGH activity of rAcm2 ⌬AST Lc was ϳ1.5-fold higher than that of rAcm2 Lc , showing a weak contribution of the AST domain deprived of glycosylation to Acm2 activity (Table 3 and Fig. 7A). Altogether, these data suggest that glycosylation and to a lesser extent the AST domain negatively impact Acm2 activity. To confirm the impact of glycosylation on rAcm2 activity, experiments were repeated with autoclaved dead cells of L. plantarum WT, which is a more resistant substrate than cells from the OatB-deficient strain. Using the same protein concentration as in the previous experiment, rAcm2 Lc showed a significant activity, whereas rAcm2 Lp did not display any detectable activity on this more resistant substrate (supplemental Fig. S8A). In addition, a commercial lysozyme assay that measures hydrolytic activity on fluorescent M. lysodeikticus cell walls was used. Fluorescence intensity was significantly higher at all incubation times for the non-glycosylated rAcm2 Lc compared with rAcm2 Lp (supplemental Fig.  S8B). Taken together, these results demonstrate that O-glycosylation inhibits the PG hydrolytic activity of Acm2.
O-Glycosylation Enhances Stability of the AST Domain against Proteases-To investigate whether glycosylation could confer protection against proteases, the same concentration of glycosylated and non-glycosylated variants of the AST domain (rAST Lp and rAST Lc , respectively) were incubated (15 min) with various concentrations of trypsin ranging from 0 to 100 g/ml. The results revealed that the glycosylated rAST Lp domain was more resistant to degradation relative to its nonglycosylated counterpart at a concentration of 10 g/ml trypsin. Indeed, degradation products are clearly visible for rAST Lc at that concentration but are absent for rAST Lp (Fig. 7B, left  panel, indicated by asterisks). On the other hand, more resistant degradation products are detected for rAST Lp compared with rAST Lc at a concentration of 100 g/ml trypsin (Fig. 7B, right panel, indicated by asterisks). These data suggest that glycosylation partially reduces trypsin digestibility of the AST domain and thus stabilizes the protein by conferring a protection against proteases.

DISCUSSION
In this work, we provide new insights in the role of accessory domains in the control of the hydrolytic activity of PGHs. Our results of the study of the N-acetylglucosaminidase Acm2 revealed a unique antagonism between SH3b domains and O-glycosylation of the AST domain in the control of its enzyme activity.
We demonstrated that SH3b domains of the N-acetylglucosaminidase Acm2 are essential for its physiological role in cell separation. In addition, we showed that these domains were fully required for Acm2 septal localization and that they appear to be the only Acm2 domains involved in the cell wall binding process. Interestingly, PG accessibility of wild-type cells for

hydrolytic activity of glycosylated and non-glycosylated rAcm2 variants
The O-glycosylation status refers to rAcm2 variants either purified from L. plantarum (ϩ) or from L. lactis (Ϫ). Data are mean values from triplicates ϮS.D. Significant differences between rAcm2 Lp (glycosylated) and rAcm2 Lc (non-glycosylated) or rAcm2 Lc (non-glycosylated) and rAcm2 ⌬AST Lc (non-glycosylated) are based on t test. ND, not detected; A.U., arbitrary units.  A 600 values between rAcm2 Lp and rAcm2 Lc or rAcm2 ⌬AST as well as between rAcm2 Lc and rAcm2 ⌬AST are all significantly different between 1 and 5 h. Significance is based on Student's t test with a p value of Ͻ0.001. B, proteolysis resistance of glycosylated (rAST Lp ) and non-glycosylated (rAST Lc ) AST domains of Acm2 toward trypsin. Protein samples were incubated with the final concentrations (g/ml) of trypsin indicated at the bottom of the gel. rAST Lc degradation products appearing at 10 g/ml trypsin and more resistant degradation products of rAST Lp at 100 g/ml trypsin are indicated by asterisks.

O-Glycosylation
Acm2 binding is very limited in the septal area, whereas in similar studies with other cell-separating PGHs, labeling of the whole septum area is generally observed (41)(42)(43)(44)(45). We reported previously that GlcNAc O-acetylation of PG inhibits Acm2 activity either by blocking glycan strand cleavage or impairing Acm2 binding (1). However, the absence of O-acetylation has no impact on the Acm2 localization pattern, suggesting that this PG modification is not a major player in the spatial distribution of Acm2, which is dominated by CPS and WTA. Unexpectedly, CPSs are the major hindering components for Acm2 binding in L. plantarum, whereas in other Gram-positive bacteria, this role has been reported for WTA (41,43,(45)(46)(47). In these studies, WTA deficiency was shown to have a global impact on PGH sublocalization, whereas in our experiments, these PG decorations only affected Acm2 binding to the septum area. At the enzymatic level, the depletion of SH3b domains dramatically affected Acm2 activity but without a complete inhibition, suggesting that they are not required as such for catalysis but that SH3b-mediated PG binding is a key step for PG hydrolysis (see model in Fig. 8).
The functional role(s) of low complexity PGH domains and their eventual glycosylation such as found in the AST domain of Acm2 has been poorly investigated so far except for the Msp1/ p75 PGH of L. rhamnosus GG for which O-glycosylation confers protection against proteases (16). To address the importance of glycosylation, glycosylated and non-glycosylated Acm2 variants were purified. The absence of enzymatic or chemical treatment to remove glycosylation was an essential step for keeping proteins as close as possible to their native state. The mass spectrometry analysis of the purified glycosylated Acm2 and the native extracellular enzyme revealed a common set of glycopeptides showing a conservation of the glycosylation pattern (data not shown and Ref. 7). In addition, the fine mapping of 21 glycosylated sites identified both Ser and Thr residues as mono-HexNAc-glycosylated but failed to reveal a unique recognition motif for O-glycosyltransferase(s). Recent studies hypothesized that prokaryotic O-glycosylation takes place on a protein by recognition of a structural motif rather than a specific amino acid sequence (17,48). Nevertheless, simple O-glycosylation motifs could be identified such as that found for the BF2494 glycoprotein of Bacteroides fragilis (49). In the latter, a 3-residue glycosylation site motif (D(S/T)(A/I/L/V/M/T)) was experimentally validated (49). Unfortunately, the three-dimensional structure of the AST domain or orthologous domains is not available for the mapping of glycosylation sites, but secondary structure predictions suggest that the AST domain is mainly unstructured (data not shown), which does not support the hypothesis of the recognition of a structural motif. In support of the hypothesis of a linear glycosylation site motif, we identified two weakly conserved motifs in the AST domain, which may indicate that more than one O-glycosyltransferase is involved. Concerning the specificity of O-glycosyltransferase(s) for the AST domain, the absence of glycosylation on Acm2 purified from L. lactis is striking because this species contains an orthologous enzyme, the N-acetylglucosaminidase AcmB, that also contains a glycosylated low complexity domain rich in Ser and Thr residues (14), 7 implying that the O-glycosylation machinery of L. lactis is highly specific for its endogenous proteins. The identification of the dedicated O-glycosyltransferase(s) will be a major step to clarify the mechanism involved in the selectivity of the glycosylation sites.
A major contribution of this work is that glycosylation of the AST domain is a key process in the negative control of Acm2 enzymatic activity as attested by the relaxed hydrolytic activity of its AST truncated variant and its non-glycosylated state on different PG substrates (Fig. 8). We also showed that O-glycosylation partially protects the AST domain from proteolysis, which could be particularly relevant to maintaining the negative control of the AST domain. Despite the fact that O-glycosylation was hypothesized to be involved in PGH subcellular localization or substrate recognition (3,14), we were not able to show any difference in terms of localization or PG adhesion. An important question is how glycosylation of the AST domain could control the hydrolytic activity of Acm2. Although some extra domains have been shown to act as autoinhibitors of PG hydrolytic activity by obstructing or restricting access to actives sites, they are not of low complexity and do not appeared to be glycosylated (50 -54). The best predicted structural homolog of 7 M.-P. Chapot-Chartier, unpublished data. The removal of the five SH3b domains (Acm2 ⌬SH3) strongly inhibited Acm2 activity due to its inability to bind PG (Inactive status), and the absence of the AST domain (Acm2 ⌬AST) or its O-glycosylation (Acm2 ⌬AST Lc ) resulted in an up-regulation of the Acm2 activity (Relaxed status). In the case of the glycosylated form Acm2 Lp (Controlled status), the access of the Acm2 catalytic domain to its substrate is proposed to be hindered by the glycosylated AST domain by three potential and non-exclusive mechanisms. I, O-glycosylation of the AST domain has a conformational impact that reduces its flexibility and mobility. II, GlcNAc residues of the AST domain directly interact with the catalytic site, resulting in a partial Acm2 inhibition. III, GlcNAc residues of the AST domain are recognized by SH3b domains, resulting in a closed conformation.