Regulation of the Xylan-degrading Apparatus of Cellvibrio japonicus by a Novel Two-component System*

The microbial degradation of lignocellulose biomass is not only an important biological process but is of increasing industrial significance in the bioenergy sector. The mechanism by which the plant cell wall, an insoluble composite structure, activates the extensive repertoire of microbial hydrolytic enzymes required to catalyze its degradation is poorly understood. Here we have used a transposon mutagenesis strategy to identify a genetic locus, consisting of two genes that modulate the expression of xylan side chain-degrading enzymes in the saprophytic bacterium Cellvibrio japonicus. Significantly, the locus encodes a two-component signaling system, designated AbfS (sensor histidine kinase) and AbfR (response regulator). The AbfR/S two-component system is required to activate the expression of the suite of enzymes that remove the numerous side chains from xylan, but not the xylanases that hydrolyze the β1,4-linked xylose polymeric backbone of this polysaccharide. Studies on the recombinant sensor domain of AbfS (AbfSSD) showed that it bound to decorated xylans and arabinoxylo-oligosaccharides, but not to undecorated xylo-oligosaccharides or other plant structural polysaccharides/oligosaccharides. The crystal structure of AbfSSD was determined to a resolution of 2.6Å. The overall fold of AbfSSD is that of a classical Per Arndt Sim domain with a central antiparallel four-stranded β-sheet flanked by α-helices. Our data expand the number of molecules known to bind to the sensor domain of two-component histidine kinases to include complex carbohydrates. The biological rationale for a regulatory system that induces enzymes that remove the side chains of xylan, but not the hydrolases that cleave the backbone of the polysaccharide, is discussed.

The microbial degradation of lignocellulose biomass is not only an important biological process but is of increasing industrial significance in the bioenergy sector. The mechanism by which the plant cell wall, an insoluble composite structure, activates the extensive repertoire of microbial hydrolytic enzymes required to catalyze its degradation is poorly understood. Here we have used a transposon mutagenesis strategy to identify a genetic locus, consisting of two genes that modulate the expression of xylan side chain-degrading enzymes in the saprophytic bacterium Cellvibrio japonicus. Significantly, the locus encodes a two-component signaling system, designated AbfS (sensor histidine kinase) and AbfR (response regulator). The AbfR/S twocomponent system is required to activate the expression of the suite of enzymes that remove the numerous side chains from xylan, but not the xylanases that hydrolyze the ␤1,4-linked xylose polymeric backbone of this polysaccharide. Studies on the recombinant sensor domain of AbfS (AbfS SD ) showed that it bound to decorated xylans and arabinoxylo-oligosaccharides, but not to undecorated xylo-oligosaccharides or other plant structural polysaccharides/oligosaccharides. The crystal structure of AbfS SD was determined to a resolution of 2.6 Å . The overall fold of AbfS SD is that of a classical Per Arndt Sim domain with a central antiparallel four-stranded ␤-sheet flanked by ␣-helices. Our data expand the number of molecules known to bind to the sensor domain of two-component histidine kinases to include complex carbohydrates. The biological rationale for a regulatory system that induces enzymes that remove the side chains of xylan, but not the hydrolases that cleave the backbone of the polysaccharide, is discussed.
The microbial hydrolysis of the plant cell wall is a pivotal biological process that is integral to the carbon cycle and is thus essential for the maintenance of terrestrial life. This process is also of growing industrial significance, particularly within the bioenergy sector where lignocellulosic biomass is a valuable and abundant substrate for the production of bioethanol and other liquid fuels (1,2). The most abundant hemicellulose polysaccharide in the majority of terrestrial plant cell walls is xylan (3). This polymer consists of a backbone of ␤1,4-linked xylose residues that can be decorated at O-2 and/or O-3 with arabinofuranose or acetyl groups, whereas 4-methyl-D-glucuronic acid is linked exclusively ␣1,2 to the xylan backbone (3). A further elaboration of xylan structure is provided by the esterification of some arabinofuranose side chains, at O-5, with ferulic acid (Fig. 1) that in turn may include a component of the aromatic heteropolymer lignin (3).
Saprophytic prokaryotes, such as Cellvibrio japonicus, which utilize the plant cell wall as a major carbon and energy source, synthesize an extensive repertoire of extracellular and membrane-associated hydrolytic enzymes that degrade all the major plant structural polysaccharides, including xylan (for review see Refs. 4,5). Consistent with the complex structure of xylans, the polysaccharide is degraded by a combination of endo-␤1,4-xylanases (6,7), henceforth referred to as xylanases, and a suite of enzymes that remove the side chains that include ␣-glucuronidases (8), arabinofuranosidases (9,10), and xylan esterases (11), which remove glucuronic acid (only from the nonreducing end of xylose polymers), arabinofuranose, and acetyl residues, respectively, from the xylan backbone, whereas the ferulic acidarabinofuranose ester linkage is hydrolyzed by ferulic acid esterases (12). Indeed, the recent completion of the genome sequence of C. japonicus revealed ϳ150 genes encoding complex carbohydrate-modifying enzymes, primarily glycoside hydrolases and carbohydrate esterases, the majority of which attack the plant cell wall (13).
Current understanding of the mechanism by which the plant cell wall, a highly complex and insoluble recalcitrant macromolecular structure, modulates the expression of the plant cell wall-degrading machinery in microorganisms is at best fragmentary. Although the complete xylan-degrading system is expressed when C. japonicus is presented with decorated xylans (14), it is unknown whether there is a common inducer for all the enzymes required for complete saccharification of the polysaccharide, or whether different molecules (derived from decorated xylans) act as inducing triggers for specific subsets of these glycoside hydrolases and esterases. Indeed, Emami et al. (14) have shown that the expression of the bacterium xylanases is controlled by a complex regulatory system. Thus, although xylan induces the transcription of the five C. japonicus xylanase genes, there are clear differences in the temporal expression of these enzymes. It has been proposed that low level constitutive expression of specific glycosyl hydrolases releases mono-or oligosaccharides (from plant cell walls) that enter the microbial cytoplasm and, by binding directly to regulatory proteins, influence their activity (15). Support for this hypothesis is provided by studies on ascomycete plant cell wall-degrading fungi; in these organisms xylose binds to and modulates the activity of the transcription factor XlnR, which activates xylanase gene expression (16). Similarly, the disaccharide sophorose appears to induce fungal cellulase activity by interacting with the transcription factors ACEI and ACEII (17,18), whereas in Geobacillus stearothermophilus the activity of the repressor protein UxuR (which controls transcription of a glucuronoxylan metabolizing system) is modulated by 4-methyl-O-glucuronic acid-xylotriose (19). By contrast, it has emerged recently that extracellular signals can activate bacterial glycoside hydrolases by interacting with the extracytoplasmic domains of hybrid two-component system proteins (20,21), in which the sensor kinase and response regulator are contained within a single polypeptide. The nature of the signals that activate these sensor proteins, however, remains unclear.
Here we have deployed a transposon mutagenesis strategy to identify regulatory proteins of C. japonicus that control the expression of the arabinofuranosidase CjAbf51A, an important and highly expressed component of the xylan-degrading apparatus of the bacterium (9,22). We show that a canonical twocomponent system, comprising AbfS and AbfR, activates the expression of the repertoire of cell-associated enzymes, including CjAbf51A, that remove the decorations from xylan, but does not control the synthesis of the xylanases that attack the ␤1,4-linked xylose polymeric backbone. The ligands that activate AbfS were shown to be arabinoxylo-oligosaccharides. The crystal structure of the sensor domain of AbfS displays a typical PAS-fold (Per Arndt Sim) fold, a large family of proteins with sensory and signaling functions that can often be difficult to identify by sequence homology alone. The structure of the sensor domain of AbfS enabled the likely binding site for the activating ligand to be identified. The biological significance for a regulatory system that controls the expression of enzymes that remove the side chains but not the xylanases that hydrolyze the backbone of xylans is discussed.

MATERIALS AND METHODS
Bacterial Strains, Plasmids, and Culture Conditions-The Escherichia coli strains used in this study were CH545 harboring pJB4JI (14) and BL21 (DE3):pLysS, B834 (DE3) (Novagen). All E. coli strains containing recombinant plasmids were cultured in Luria-Bertani broth (LB) supplemented with 50 g/ml ampicillin or kanamycin, as appropriate, at 37°C unless otherwise stated. To grow C. japonicus, a rifampicin-resistant mutant of the bacterium was cultured in either LB or minimal medium, which consists of M9 salts and the appropriate carbon source, either monosaccharide or polysaccharide, at a final concentration of 0.2%. The Cellvibrio was grown either in liquid culture at 30°C with high aeration (200 rpm) for up to 36 h in medium comprising Ͻ10% (v/v) of the incubator vessel (conical flask) or on solid media at 28°C for up to 2 weeks.
Transposon Mutagenesis-The transposon Tn10 was randomly inserted into the genome of C. japonicus as described previously (22). To screen 10 4 colonies from the transposon library, plates containing transconjugant colonies were picked and cultured for 4 days at 30°C on LB-agar supplemented with wheat arabinoxylan at which point large colonies were visible. The plates were then overlaid with 3.5 ml of 50 mM sodium phosphate buffer, pH 7.0, containing 4 mM 4-methylumbellerfyl ␣-L-arabinofuranoside (MUA) 6 which, when cleaved, produces the fluorescent compound methylumbelliferone. After 15 min, the plates were inspected over a UV transilluminator, and cells that did not fluoresce (or fluoresced more brightly than the wild type bacterium) were selected for further study. As CjAbf51A is the only C. japonicus enzyme that can hydrolyze MUA (22), the mutants that displayed altered MUAase expression exhibit a change in the expression level of the arabinofuranosidase. To identify the genes interrupted by the Tn10 insertions, chromosomal DNA from selected transconju-  gants were digested with NcoI (cuts once within the transposon outside the kanamycin resistance gene) and cloned into pMTL22p. Transformants containing Tn10 insertions were identified by their kanamycin-resistant phenotype. Using the sequencing of the NcoI C. japonicus insertions, which extends from the transposon into the flanking DNA, to interrogate the sequence of the C. japonicus genome led to the identification of the gene disrupted by Tn10.
Expression and Purification of AbfS SD and AbfR-The region of the AbfS gene encoding AbfS SD (amino acids 42-165 of AbfS) was amplified by PCR using Taq polymerase, the primers 5Ј-CTCCATATGGATATAACGCCGGAAGG-CAACTTCC-3Ј and 5Ј-CTCCTCGAGATTTTCGAGAAACA-TTTCCCAAATGATATCCCG-3Ј, and C. japonicus genomic DNA as the template. The amplified DNA was digested with NdeI and XhoI and cloned into similarly restricted pET22b to generate pJH1. AbfS SD was produced as a soluble, His 6 -tagged fusion protein in E. coli BL21 (DE3) harboring pJH1 by culturing the bacterium at 30°C to mid-exponential phase (A 600 nm 0.6), and recombinant protein expression was then induced by the addition of isopropyl ␤-D-thiogalactopyranoside to a final concentration of 1 mM and incubation overnight at 16°C. The sensor protein was purified to electrophoretic homogeneity by immobilized metal anion chromatography and size exclusion chromatography using Talon TM and Superdex-75, respectively, using methodology described previously (24). AbfS SD was eluted from the Talon column with 50 mM Tris-HCl buffer, pH 8.0, containing 30 mM NaCl and 100 mM imidazole. Selenomethionyl-substituted AbfS SD was overproduced in the E. coli methionine auxotroph B834 (DE3) and purified by the same procedure as for native AbfS SD except that all buffers contained 10 mM 2-mercaptoethanol.
To express the DNA binding domain of AbfR in E. coli, the corresponding region of the gene was amplified by PCR from chromosomal C. japonicus DNA using Taq polymerase and the following primers: 5Ј-AACATATGCTCAAAATTCTCCTC-ATTGATGATG-3Ј and 5Ј-AAGGATCCTTACTGCTCGTC-ATTCAGGGTCAACATG-3Ј. The amplified DNA was cloned into NdeI-BamHI-restricted pET16b site to generate pTN2, which encodes AbfR containing an N-terminal His 6 tag. AbfR was expressed in E. coli BL21 (DE3) harboring pLysS and pTN2 employing the same growth conditions used to produce Abf SD . The response regulator was purified by immobilized metal affinity chromatography, as described for Abf SD , and then by anion exchange chromatography using a Bio-Rad Q column. Unbound protein was removed from the column in 10 mM Tris-HCl buffer (Buffer A), pH 8.0, and AbfR was eluted with a 0 -500 mM NaCl gradient in Buffer A.
Mutagenesis of Abf SD -Amino acid substitutions were introduced into AbfS SD using PCR-based methodology deploying the QuikChange mutagenesis protocol (Stratagene).
DNA Binding Assay-Electrophoretic mobility shift analysis (EMSA) used infra-red dye-labeled probe corresponding to the promoter of abf51A (gene encoding CjAbf51A) extending from nucleotides Ϫ13 to Ϫ218 (ϩ1 is the 1st nucleotide of the structural gene). The probe was generated by PCR using C. japonicus as the template DNA and primers, listed in supplemental material, labeled at the 5Ј end with IRDye 700 (IRD; synthesized by MWG Biotec). The corresponding unlabeled probe was generated in the same way but using primers without the infra-red label. Double-stranded oligonucleotide competitors were produced from complementary single-stranded oligonucleotide pairs (supplemental material) by annealing at 96°C for 10 min in 10 mM Tris-HCl buffer, pH 7.5, containing 50 mM NaCl and 1 mM EDTA. AbfR (0.1-1 g), IRD-labeled probe (5-7.5 ng), and where appropriate, double-stranded oligonucleotide competitors (200-fold excess) or other competitors at a 100-fold excess were incubated for 30 min at 4°C in 10 mM Tris-HCl buffer, pH 7.5, containing 50 mM KCl and 1 mM dithiothreitol. The mixture was then loaded onto a 5% (w/v) polyacrylamide-TBE gel and electrophoresed for 90 -120 min at 4°C in 0.5ϫ TBE buffer. Images were captured using an Odyssey Infrared Imaging System (LI-COR Biosciences).
Enzyme Assays-To assess the capacity of the isolated transposon mutants to synthesize plant cell wall hydrolases, the selected strains were cultured on minimal media in which the sole carbon sources were wheat arabinoxylan, hydroxyethylcellulose, and carob galactomannan, respectively, at a concentration of 0.2%. At late exponential phase/early stationary phase, the culture supernatant and the harvested cells were assayed for acetyl and ferulate esterase (11), xylanase (25), arabinofuranosidase (22), ␣-glucuronidase (26), mannanase (27), and endoglucanase (28) activities.
Ligand Binding Assays-The binding of AbfS SD to a variety of soluble polysaccharide ligands was assessed by affinity gel electrophoresis and fluorescence spectrometry. Affinity gel electrophoresis was based on the method of Ref. 29. Potential polysaccharide ligands were added to the native gel mixture immediately prior to polymerization at a concentration of 0.1%. In the fluorescence spectrometry method, the 2.5-ml reaction consisted of 50 mM sodium phosphate buffer, pH 7.0, containing 1.0 M AbfS SD and potential ligands at a concentration ranging from 0 to 1.5 mM for oligosaccharides and 0 -1 mg ml Ϫ1 for polysaccharides. After thorough mixing the protein was excited at 280 nm, and the emission spectrum was scanned from 290 to 450 nm. The fluorescence data were fitted to a single site model using nonlinear regression to determine the dissociation constant (K D ).
Sequence Comparison Analysis-Proteins were screened for sequence similarity within the Pfam data base (Wellcome Trust Sanger Institute) using the HMMER2 suit of programs. The output for the probability that two proteins, or sequence motifs, are related are given as Hidden Markov Model (HMM) scores (bits) and expectation values. An HMM score is related to the log base 2 of the probability of the alignment to the model divided by the probability of the sequence matching because of the "randomness of amino acid sequence." Thus a score of 100 bits means it is 2 100 -fold more likely that the sequence is a match to the model rather than because of a random event. The expectation or e score is the number of different alignments with scores equivalent to or better than a score that is expected to occur by chance. The lower the E value, the more significant the score.
Crystallization of AbfS SD -Crystallization conditions for native AbfS SD were obtained by sparse-matrix screening at 20°C using the hanging-drop vapor diffusion technique. The best native crystals grew from AbfS SD solutions at a concentration of 21.5 mg/ml mixed with 0.9 M KH 2 PO 4 , and although this solution was buffered with 0.1 M Tris-HCl, pH 8.0, the actual pH of the crystallization mother liquor was ϳ6.0. Although selenomethionyl-AbfS at 18.7 mg/ml also crystallized from phosphate-containing solutions, isomorphous crystals that were subsequently used for data collection were actually obtained from 1.8 M (NH 4 ) 2 SO 4 , 0.1 M NaCl, 0.1 M sodium cacodylate, pH 6.5. All crystals were prepared for data collection by adding glucose to 30% saturation to the crystallization mother liquor. Crystals were flash-cooled by plunge-freezing into liquid nitrogen before being mounted for x-ray diffraction data collection.
AbfS SD Structure Determination-The crystals of AbfS SD belong to the primitive orthorhombic space group P2 1 2 1 2 with unit cell dimensions a ϭ 107.8 Å, b ϭ 80.0 Å, c ϭ 102.0 Å. Additional crystal statistics are indicated in Table 1. For phasing purposes, diffraction data from crystals of native and selenomethionyl-substituted AbfS SD were collected, each to a maximum resolution of 3.0 Å resolution, on beamlines BM14 and ID14-EH3 at the European Synchrotron Radiation Facility in Grenoble, France, respectively. The diffraction data were processed and scaled using MOSFLM (30) and SCALA (31).
The structure of the sensor domain was solved by the single wavelength isomorphous replacement anomalous scattering (SIRAS) technique. Unmerged but scaled data from the selenomethionine dataset (Selenomet, Table 1) were input into SHELXD (32) and the atomic positions of 18 selenium sites were found. These sites were input into SHARP (33) with the native (native 1, Table 1) and selenomethionine (Se-Met, Table  1) datasets for the calculation of SIRAS phases. The self-rotation function indicated the presence of three 2-fold noncrystallographic symmetry (NCS) axes, and a noncrystallographic screw axis was identified by weak 00l reflections with l 3n. The initial electron density map was improved by density mod-ification in SHARP (33). A region of two helices was fitted in real space (FFFEAR (34)) to the density-modified SIRAS map to identify all six NCS-related molecules. A molecular envelope for one molecule of AbfS SD was obtained from a skeleton trace in O (35). This molecular envelope was used with the NCS operators to calculate solvent-flattened, 6-fold NCS-averaged maps in density modification (36) to 3.0 Å. The electron density map was significantly improved by these phase-refinement procedures with the density modification figure of merit increasing from 0.57 to 0.76. The improved map permitted the manual building in O of one AbfS SD molecule, and this model was copied to the other sites by the NCS operators. Additional electron density was present for chains B and E, corresponding to the C-terminal ␣-helix. This model was then refined against 2.6 Å resolution data collected at a subsequent date from a native crystal (native 2, Table 1). 5% of the reflections were omitted from the refinement and used for cross-validation. Refinement used the maximum likelihood target in REFMAC (37) with TLS (translation libration and screw) groups and tight (for main chain atoms) and medium (for side chain atoms) NCS restraints between the chains, interspersed with manual correction in COOT (38). Water molecules were placed automatically with REFMAC/wARP (39) and verified manually. Analysis of the model with MolProbity (40) showed just two Ramachandran plot outliers in the structure. The final Abf SD model was refined against the native 2 data to R cryst /R free values of 0.204/0.250, respectively. The PDB accession code for Abf SD is 2VA0.

Isolating Regulatory Mutants
To interrogate the global regulatory systems that control the plant cell wall-degrading apparatus of C. japonicus, a 10 5 -colony transposon library was constructed using Tn10. A subset of

Xylan Sensor
the library, comprising 10 4 clones, was screened using a soft agarose overlay containing MUA (arabinofuranosidases cleave MUA to release the fluorescent product 4-methylumbelliferone; see under "Materials and Methods") for aberrant expression of the arabinofuranosidase CjAbf51A, a key component of the enzyme consortium that hydrolyzes decorated xylans (9,22). Southern hybridization of 20 colonies showed that the transposon had integrated randomly in the Cellvibrio genome (data not shown). Tn10 is therefore a more useful genetic tool than Tn5 (used previously to mutate C. japonicus (22)), as there are several Tn5 integration hot spots in the genome of the bacterium. This screening strategy generated five colonies that failed to express significant levels of CjAbf51A on media that induce synthesis of the enzyme in wild type C. japonicus. Chromosomal DNAs flanking the transposon were cloned into pMTL22 and sequenced to identify the genes interrupted by Tn10 insertion.
In three of the five colonies that displayed no arabinofuranosidase activity, Tn10 had inserted into abf51A, the gene encoding the arabinofuranosidase CjAbf51A, thus explaining the lack of this enzyme activity, whereas the other two colonies contain transposon insertions in a gene designated abfS. The 467amino acid protein encoded by abfS contains all the sequence motifs expected of a two-component histidine kinase (Gen-Bank TM accession number ACE83229); toward the C terminus is a histidine phospho-acceptor domain (HMM score ϭ 33.5; expect ϭ 7.9e Ϫ7 ) and a kinase catalytic domain (HMM score ϭ 116.9; expect 6.4e Ϫ32 ). The N-terminal region of AbfS contains two transmembrane-spanning helices that flank an ϳ140amino acid sequence that, by analogy to other two-component histidine kinases, includes the sensory domain that is located in the periplasmic space. Immediately upstream of abfS is abfR (GenBank TM accession number ACE84843; the sequences of AbfS and AbfR are shown in the supplemental material), which encodes a 232-amino acid protein, designated AbfR, whose N-terminal region (ϳ120 amino acids) displays the canonical features of the receiver domain of a response regulator (HMM score ϭ 145.9; expect ϭ 1.2e Ϫ40 ). These features include the conserved trio of aspartates and the Lys-Pro motif, which are required for magnesium-dependent phosphoryl transfer, and the consensus Ser/Thr and Tyr/Phe/His residues of the "aromatic switch," the conformational change that occurs on response regulator activation (41,42). The effector domain of AbfR belongs to the OmpR/PhoB family (expect ϭ 3e Ϫ18 ) of "winged helix" DNA-binding proteins (43), the largest subfamily of response regulator effector domains (44). Thus, abfS and abfR encode a two-component system that most likely regulates the expression of CjAbf51A.

Biochemical Properties of abfS Mutants
To explore the physiological and biochemical influence of the transposon-disrupted abf genes, C. japonicus was cultured on a range of different liquid media, and arabinofuranosidase activity was measured. In the three abf51A mutants no detectable arabinofuranosidase activity (employing the MUAase assay) was detected when the strains were cultured on wheat arabinoxylan confirming that CjAbf51A is the major arabinofuranosidase expressed by C. japonicus. In the abfS mutant, CjAbf51A activity was substantially reduced in comparison with the wild type C. japonicus when cultured on media that induce arabinofuranosidase expression ( Table 2). Although the expression of the cellulases and xylanases produced by C. japonicus was unaffected, the abfS mutant displayed greatly reduced levels of esterase and ␣-glucuronidase activity, enzymes which, together with CjAbf51A, remove the side chains from decorated xylans ( Table 2). To confirm that the biochemical effect of the transposon insertion into abfS was because of inactivation of the kinase sensor, and not through a polar effect on a downstream gene, the wild type sensor was introduced into the abfS mutant. The data (see Table 2) show that the introduction of pTN1 (contains the abfR-abfS locus cloned into the conjugative plasmid pRG960sd (14)) into the abfS mutant restores arabinofuranosidase, esterase, and ␣-glucuronidase activity. These data indicate that the genes encoding the enzymes that remove the xylan decorations are co-regulated by the response regulator cognate to AbfS, which is most likely to be AbfR given their proximity in the genome.

Binding of AbfR to the Promoter of abf51A
To provide further evidence that AbfR targets the promoter of abf51A, the DNA binding domain of the response regulator (AbfR DBD ) was deployed in EMSA experiments using IRDye-labeled abf51Ap (promoter region of abf51A) as the probe. The data show that AbfR DBD causes significant electrophoretic retardation of the probe in the presence of poly(dI-dC) and salmon sperm DNA, but not when nonlabeled abf51Ap was added to the reactions (Fig. 2). These data show that AbfR binds to the promoter region of the CjAbf51A gene and thus is likely to directly modulate transcription of abf51A. To probe further the binding site of AbfR within abf51Ap, EMSAs were carried out with AbfR DBD and a series of overlapping oligodeoxynucleotides that extend across the promoter region. The only oligodeoxynucleotide to prevent probe retardation consisted of nucleotides Ϫ81 to Ϫ120 of the promoter region (Fig. 2) (ϩ1 is the first nucleotide in the structural gene). As the flanking, The activity of cellulase-and xylan-degrading enzymes in wild type and transposon mutants of C. japonicus grown on 0.2% arabinoxylan C. japonicus was cultured, and enzyme activities were determined as described under "Materials and Methods." The activity of the six enzymes, expressed as nanomoles of product/mg of cellular protein, was assayed in cells harvested from cultures grown for 48 h. The plasmid pTN1 contains the abfR-abfS locus cloned into the conjugative plasmid pRG960sd. noncompeting, oligodeoxynucleotides extend from Ϫ65 to Ϫ104 and Ϫ105 to Ϫ144, respectively, the binding site is centered around nucleotides Ϫ104 and Ϫ105.

Biochemical Properties of the Sensor Domain of AbfS
The sensor domain of AbfS (AbfS SD ), corresponding to residues 42-165 of the full-length protein, was expressed in E. coli and purified to electrophoretic homogeneity. Affinity gel electrophoresis was used to screen the ligand (polysaccharide) specificity of AbfS SD . Quantification of binding was determined by fluorescence spectroscopy. Representative data, displayed in Fig. 3 and the complete data set tabulated in Table 3, show that AbfS SD binds strongly to arabinoxylans from wheat and rye and weakly to glucuronoxylan and oat spelled xylan. No binding of AbfS SD to mannans, ␤-glucans, arabinan, galactan, or arabinogalactan was evident. As AbfS SD is predicted to be located in the periplasm of C. japonicus, it is likely that oligosaccharides (rather than polysaccharides) will comprise the natural ligand for AbfS as they will be translocated across the outer membrane. Thus, the capacity of AbfS SD to bind to oligosaccharides was assessed by fluorescence spectroscopy. Fig. 4 depicts the titration of the sensor domain with AX 2 in which the Xyl-2 (sugars are labeled from the reducing end sugar which is designated Xyl-1) of xylobiose is decorated with an ␣1,3-linked L-arabinofuranose residue. The fluorescence emission spectra are quenched by AX 2 , indicating that the oligosaccharide binds to AbfS SD and increases the polar environment of one or more tryptophans. AbfS SD was also shown to bind to AX 3 (xylotriose decorated with an ␣1,3-linked L-arabinofuranose at Xyl-2), but did not interact with xylotriose, xylobiose, arabinose, cellobiose, or mannohexaose (see Table 3). These data indicate that the target ligands for AbfS SD are xylo-oligosaccharides decorated at O-3 with arabinofuranose moieties. A partial titration curve of AbfS SD with AX 2 and AX 3 revealed an approximate K D for the two oligosaccharides of 700 M (the titration curve could not be completed as the protein started to precipitate after the addition of 5 M eq of the ligand). The reported affinities of the sensor domain of histidine kinases for their target ligands are highly variable. For example, DcuS binds to its C 4 dicarboxylate ligands with K D values between 0.45 and 3 mM (45,46), whereas the dissociation binding of CitA for citrate is in the micromolar range (47,48).

Crystal Structure of AbfS SD
Overall Structure-The structure of AbfS SD was solved using a selenomethionine-containing AbfS SD derivative and refined to convergence at a resolution of 2.6 Å. The final atomic model of AbfS SD contains six protein monomers comprising 120 amino acids, 99 water molecules, and 6 phosphate ions. Each monomer is comprised of a ␣/␤/␣ sandwich (Fig. 5). The central ␤-sheets contains four anti-parallel ␤-strands in the strand order ␤1, ␤4, ␤3, and ␤2 that is flanked by four ␣-helices (three N-terminal of ␤-strand 1 and one C-terminal of ␤-strand 4) on one face and one ␣-helix, extending from the loop connecting   a Oligosaccharide binding was assessed using fluorescence spectroscopy. The potential polysaccharide ligands were initially screened for binding using affinity gel electrophoresis. Fluorescence spectroscopy was used to determine the affinity for those carbohydrates that displayed binding in the affinity gel electrophoresis screen. b NB means no binding detected. c AX 3 , xylotriose (trisaccharide of ␤1,4-linked xylose residues) in which the central xylose is substituted with an ␣1,3-linked arabinofuranose. d AX 2 , xylobiose (disaccharide of ␤1,4-linked xylose residues) in which the nonreducing xylose is substituted with an ␣1,3-linked arabinofuranose.
␤-strands 1 and 2, on the other face. Despite containing six molecules in the crystallographic asymmetric unit (see supplemental Fig. 1), none of the interfaces bury a sufficiently large surface area (49) to be physiologically relevant (50). Other histidine kinase sensor domains have also shown some plasticity in forming nonphysiological multimers in crystal lattices (51)(52)(53)(54), presumably as a consequence of crystal packing forces that override the weak self-association affinity for the sensor when the isolated domain is removed from its natural dimeric state in the membrane of the cell (55)(56)(57).
Relationship of AbfS SD to Other Structures-AbfS SD displays a typical PAS fold (reviewed in Ref. 58). Arguably the best characterized PAS domain is unrelated to two-component signaling; photoactive yellow protein (PYP), which senses light via a covalently bound para-coumaric acid cofactor (PDB 2PHY (59)). AbfS SD shares this fold with the extracellular domains from other histidine kinases, a selection of which are displayed alongside AbfS SD in Fig. 6. These include, but are not restricted to, the fumarate-receptor DcuS (PDB 1OJG (54)), the citratesensor CitA (PDB 1P0Z (60)) (Fig. 6), the quorum sensor LuxQ (61), and the divalent cation-binding PhoQ from Salmonella typhimurium and E. coli (PDB 1YAX (52) and PDB 2HJE (51), respectively). Cheung et al. (51) have recently described a sub-family of the PAS fold, the PDC (PhoQ-DcuS-CitA) domain, based on strict criteria for the alignment of a small subset of structurally related segments (30 -39 residues). Using the more global approach of secondary structure matching (62), all six chains of AbfS SD align, with similar root mean square deviations (2.6 -3.5 Å over 75-90 matched residues, with Z-and Q-scores of 1.7-2.9 and 0.39 -0.20 with 8 -15% sequence identity) to both PDC proteins, such as CitA (60) and DcuS (54), and non-PDC polypeptides exemplified by PYP and the intracellular PAS-A domain from KinA (53). Indeed the closest structural orthologue to AbfS SD is PhoR. 7 Ligand-binding Site-Several PAS domains have been solved in complex with their target ligand or cofactor, for instance PYP with para-coumaric acid (59), FixL and its close neighbor DosH with heme (63, 64) (Fig. 6), and the sensor domain of the histidine kinases PhoQ with calcium (52) and nickel (51), and CitA with citrate (60). Comparison of the structures of these PAS domains that bind macromolecular target ligands (other than simple divalent cations ion in the case of PhoQ) may therefore provide insight into the mechanism of ligand recognition by AbfS. The ligand-binding site in these PAS domains is found within the central part of the conserved ␤-sheet, where ␤-strands 4 and 5 in particular form the interior region of the site.
As we were unable to obtain crystals of AbfS SD in complex with arabinoxylo-oligosaccharides, the ligand-bound forms of PAS domains were used to guide mutagenesis of AbfS SD to identify the carbohydrate-binding site. Ligand binding was assessed using AX 2 and fluorescence spectroscopy. The majority of the mutations had no effect on ligand binding (Table 4); however, the mutants L107A and D118A displayed a reduced  affinity for the oligosaccharide, whereas the L107A/D118A variant exhibited no binding to AX 2 . Leu-107 in AbfS SD is located in the final ␣-helix, opposite Asp-118 on ␤-strand 3, and the alkyl sections of these side chains are within 4 Å of each other. Superimposition of AbfS SD with the other ligand-bound PAS complexes reveals that Leu-107 is within 6 Å of both the citrate in CitA and the para-coumaric acid in PYP and approaches within 2.5 Å of the iron at the center of heme in FixL and DosH. Therefore, the carbohydrate-binding site in AbfS SD appears to overlap with the ligand-binding sites in other structurally characterized PAS domains, pointing to conservation in structure/function relationships in this important signal transduction domain.
The observed change in fluorescence upon ligand binding is intriguing, as this implies that the polar environment of at least one tryptophan changes on proteincarbohydrate association. There are two tryptophans in AbfS SD , at positions 80 and 159. Trp-159 is situated in the C-terminal ␣-helix that is only partially ordered in the crystal structure. Trp-80 is located on the opposite face of the AbfS SD molecule, separated from Leu-107 and Asp-118 by some 25 Å. These observations may not be contradictory; ligand-binding by sensor domains presumably induces conformational changes in histidine kinases that lead to their activation. The fluorescence quench we observe in the presence of AX 2 may thus reflect structural rearrangements in the PAS domain, transduced across the membrane in the context of the full-length protein, to trigger kinase activation.

DISCUSSION
This study shows that two-component systems in C. japonicus can regulate the enzymes that remove the side chains from xylan-derived oligosaccharides. The utility of two-component systems in controlling the expression of the polysaccharidedegrading enzymes is becoming increasingly apparent. For instance, N-acetylglucosamine oligomers from chitin activate the ChiS sensor histidine kinase from Vibrio cholerae (65). Furthermore, the hybrid two-component protein, XynR from Prevotella bryantii, containing canonical sensor histidine kinase and response regulator motifs in a single polypeptide, activates a xylanase encoding operon in vitro, although the identity of the ligand that activates XynR was not established (20). By analogy to the data described in this study, the activating ligand for XynR may include a xylan-derived oligosaccharide. Similarly, a hybrid two-component system protein from Bacteroides thetaiotaomicron was shown to induce ␣-mannosidase expression, and genetic evidence implicates ␣-mannosides as the target ligand for this protein (21), although such a conclusion awaits detailed molecular studies. The direct biochemical evidence presented here shows that oligosaccharides can activate specific glycoside hydrolase systems by binding to the sensory domains of two-component histidine kinases, such as AbfS. It  therefore appears that both hybrid two-component system proteins and classical two-component pairs play an important role in regulating the expression of plant cell wall-degrading enzymes.
The need for two-component sensor kinases to limit the hazards posed by cross-talk, in the face of an onslaught of numerous potential activating cues, requires individual sensor domains to be highly specific for their ligands. The PAS domain appears to have evolved as one answer to this problem, with a core ␤-sheet that is maintained in the examples studied to date but with significant "decoration" by peripheral ␣-helices and insertions, to provide individual and highly specific solutions to molecular recognition problems in complex (bio)chemical environments. What is still absent from this field, however, is a structural description of the molecular basis for activation of the kinase on binding its activator.
It is surprising that, despite an extensive mutagenesis program, only the L107A/D118A double mutant was unable to bind AX 2 . These amino acids span the ligand-binding sites in other PAS domains, located immediately in front of the conserved ␤-strand core. The recognition of highly complex carbohydrates is usually mediated by a small number of amino acids, and these macromolecular interactions can be severely compromised by mutating single amino acids (66,67). Moreover, single amino acid replacement of Arg-107 or His-110 with alanine in the PAS domain of DcuS is sufficient to mimic deletion of the entire dcuS gene (54). It is possible that in AbfS SD specific molecular recognition may be attained by the backbone carbonyl and amide groups of the protein and is insensitive to a mutagenesis approach that only alters the side chains of amino acids. Although unusual, it is not without precedence for highly specific protein-ligand interactions to be mediated by the protein main chain. For instance, deaminated base recognition in template strand DNA by replicative family B DNA polymerases from the archaea is achieved predominantly by hydrogen bond contacts between uracil and the protein main chain (68).
The observation that AbfS/AbfR activates the expression of enzymes, which remove the side chains from xylans but not the xylanases that attack the backbone of this polysaccharide, is intriguing. Many organisms produce ␤-xylosidases that hydrolyze the xylo-oligosaccharides released from xylan degradation. Relatively few microorganisms, however, produce the enzymes required to remove the side chains from these xylose polymers, without which the xylosidase would not be able to attack the undecorated backbone structures. Thus, the decorations on the xylo-oligosaccharide will ensure that C. japonicus can prefer-  wall (4, 14). These enzymes release soluble xylan and the oligosaccharide AX 2 that binds to and activates AbfS, which modulates the activity of AbfR resulting in the expression of the surface arabinofuranosidase, ␣-glucuronidase, and esterases (22,26). entially utilize these molecules as it contains the full complement of enzymes required to access all the sugars in these polymers. Thus, we propose that xylan is initially attacked by xylanases that are targeted to the cell wall through their noncatalytic carbohydrate-binding modules (for review see Ref. 23) to generate an array of decorated xylo-oligosaccharides. Only when these molecules accumulate in high concentrations in the periplasm of C. japonicus is the expression of the membraneassociated side chain cleaving enzymes activated by AbfS/R (Fig. 7). The proposed "late" expression of the side chain cleaving enzymes may explain the relatively weak binding of AX 2 to AbfS SD , with a K D of 700 M. The sugars generated by the action of the endo-acting xylanases and side chain cleaving enzymes will be transported into, and utilized by, the bacterium. It is likely that other saprophytic plant cell wall microorganisms adopt similar strategies to fully utilize the nutrients released from the plant cell wall. Indeed, the Gram-positive bacterium G. stearothermophilis contains several intracellular xylan-degrading enzymes and thus completes the hydrolysis of at least some decorated xylo-oligosaccharides in the cytoplasm ensuring complete access to the resultant sugars (19).
This study provides further evidence that oligosaccharides can comprise ligands for two-component systems and expands the number of well characterized activating molecules of these proteins. Furthermore, contrary to the general view that xylandegrading enzymes are co-regulated, we also reveal that biocatalysts that hydrolyze the backbone and remove the side chains, respectively, of the most abundant hemicellulosic polysaccharide in the plant cell wall are regulated by distinct signaling systems.