CBM1-containing proteins were collected from the CAZy database 1 and curated for Stramenopiles sequences which were further retrieved by mining predicted proteome on dedicated databases (Additional file 1: Table S1). Seven oomycete genomes were mined for CBM1 (IPR000254) containing sequences. The oomycete species studied are characterized by different lifestyles, namely the obligate biotrophic species Hyaloperonospora arabidopsidis and Albugo laibachii, four hemibiotrophic Phytophthora species (P. infestans, P. sojae, P. ramorum, P. capsici) and the necrotroph Pythium ultimum. Sequences from various fungal databases, absent in CAZy (i,e Chaetomium globosum), were added. Overall a set of 60 fully sequenced fungi representing the major evolutionary lineages within the fungal kingdom were included in the analysis (Figure 1). Sequences were collected to build a representative set spanning the diversity of fungi and oomycetes. In total 518 sequences were selected for further analyses.

The survey of CBM1 revealed the presence of at least one CBM1-protein in 43 of the 67 (77%) completed fungal/oomycete genome sequences examined. The total numbers of CBM1-containing proteins per sequenced fungal/oomycete species classified by life style (saprotroph, necrotroph, hemibiotroph, symbiont and animal pathogen) is shown in Figure 2. Chaetomium globosum as well as two other saprobes, the white rot fungus Phanerochaete chrysosporium and the dung fungus Podospora anserina encode the highest number of predicted CBM1-containing protein coding genes among sequenced fungal genomes. A closer look shows that most fungi that harbor necrotrophic or hemibiotrophic infection strategy (e.g. Verticillium dahliae, Magnaporthe grisea), have the highest number of CBM1-protein coding genes per genome (>10 up to 28 CBM1 copy number). Interestingly, none or a single CBM1-protein gene is detected in biotrophic plant pathogen (e.g. Ustilago maydis, Puccinia graminis, Melampsora populina) and symbionts (e.g. Glomus intraradices, Laccaria bicolor, Tuber melanosporum). Thus, the expansion of CBM1-encoding genes in phytopathogenic fungi is correlated with hemibiotrophic or necrotrophic lifestyles. However, genes encoding CBM1-containing proteins are not restricted to fungal plant pathogens. Indeed, the same number of genes encoding CBM1-containing proteins is present in Aspergillus fumigatus and A. aculeatus, the former being a human pathogen and the latter a plant pathogen. Likewise, the opportunist human pathogen Rhizopus oryzea also displays genes encoding CBM1-containing proteins while genomes of other animal pathogens, including the amphibian pathogen Batrachochytrium dendrobatidis, lack detectable genes encoding cellulose-binding proteins.

Rather like true fungi, necrotrophic and hemibiotrophic oomycetes contain more genes encoding CBM1-containing proteins (up to 12) than biotroph species (A. laibachi, H. arabidopsidis). To complete this study, the genomes of Arabidopsis thaliana and Oryza sativa, and those of the closet phylogenic cousins of oomycetes, the diatoms Thalassiosira pseudonana, Phaeodactylum tricomutum and the brown algae Ectocarpus siliculosus, which also belong to the heterokont lineage, were screened for CBM1-encoding sequences. However, no CBM1-encoding sequences were detected in these organisms.

The domain architecture of CBM1-containing proteins was investigated using the boundaries defined in Pfam/Smart annotations. Domain architectures were assigned to 4 categories: i) "domain pair" which are sequences that contain a CBM1 and one other domain, ii) "multi-domain", which are sequences that contain one CBM1 and two other domains, iii) "single CBM1", which are sequences that are composed of a single isolated CBM1 and iv) "multi-CBM1", which are sequences that display more than one CBM1 (Figure 3A). Among eukaryotes, fungal proteomes display more than 50% of proteins can be classed as ‘single domain’ 34 . Regarding CBM1 associations, the ‘single domain’ category is somewhat under-represented and pertains to roughly ~15% of the fungal and oomycetal CBM1-containing proteins (Figure 3B). In contrast, 88% and 37% of fungal and oomycetal CBM1-containing proteins respectively possess ‘domain pair’ architecture. Moreover, the ‘multi CBM1’ architecture is more widespread in oomycetes (42%) than in fungi (0.2%).

In fungi, 89% of the ‘domain pair’ category are proteins that contain a CBM1 appended to a catalytic domain belonging to the GH superfamily and 11% correspond to a CBM1 appended to non-catalytic modules (chitin-binding module, carbohydrate-binding module family 4, fucose lectin, unknown (DUF) or wide functions (BNR/Asp-box repeats, see Additional file 2: Table S2 for IPR number of corresponding domain). In sharp contrast, in oomycetes 90% of the proteins in the ‘domain pair’ category are characterized by a CBM1 associated with a non-catalytic domain, corresponding to the PAN/Apple module, which are known to display protein-carbohydrate or protein-protein functions. Using SignalP 3.0 35 , it was found that 91 % of the proteins are predicted to contain a signal peptide, indicating that they are probably secreted. Overall, the large-scale systematic analysis of domain architecture reported here clearly revealed the distinct architecture of fungal and oomycetal CBM1-containing proteins.

Taking our study further, we compared the architecture distribution between the different fully-sequenced organisms. The abundance of domain associations per genome was computed and hierarchical clustering was used to assemble similar domain associations. An overview of the heat map (Figure 4, Additional file 2: Table S2) reveals a distinct distribution of domain combination in fungal and oomycetal proteins. A large proportion of fungal proteins, encompassing 70% of fungal plant pathogens, is characterized by CBM1s associated with catalytic domains involved in cellulose degradation. This includes fungal cellobiohydrolases (GH6.CBM1, GH7.CBM1) and β-glucosidases, exemplified by the CBM1.GH3 association. Interestingly, a large number of fungal CBM1-containing proteins are characterized by the GH61.CBM1 association. The GH61 family was originally described as endo-1,4-β-D-glucanases displaying weak activity, but recent studies have shown that GH61 members are in fact copper-dependent oxidases, that promote lignocellulosic substrate degradation 36 37 . Within this main group, one can observe a subgroup exclusively constituted by fungal plant pathogens that cause plant wilting symptoms (Verticillium sp. and Fusarium sp.), which are provoked by the colonization and proliferation of the microorganism in the xylem vessels of the host. In this small cluster, proteins predicted to degrade pectin (polysaccharide lyase families PL1 and PL3 appended to a C-ter CBM1) are also found.

Oomycetes species are clustered in one, unique group presenting two main characteristics. First, catalytic domains classified in the CAZyme database are not detected, except those of the GH5 family. Second, most of CBM1-containing proteins are organized as multiple CBM1 modules, which are either standalone proteins or are appended to PAN/Apple domains. Therefore, to investigate whether the absence of GH domains in oomycete CBM1-containing proteins reflects a general scarcity of glycosyl-hydrolases genes in their genome, the number of GH genes per genome was calculated. As shown in Table 1, oomycete genomes contain a large set of glycosyl-hydrolases genes (>100 gene models/genome), except A. laibachii (<50). Thus the rarity in oomycetes of GH.CBM1 associations, coupled to an enrichment of CBM1.PAN/Apple associations, supports the hypothesis that specific functions can be attributed to CBM1-containing proteins from these microorganisms and bears witness to a distinct evolutionary history compared to fungi.

Data were collected from A. laibachii 69 , Phytophthora sp 70 , P. ultimum 71 , B. cinerea 72 , F. oxysporum, V. dahliae and M. grisea 73 .

In order to identify residues that are conserved in both fungal and oomycetal CBM1 domains, the CBM1 amino acid sequences of the 518 proteins were extracted and visually compared using WebLogo 38 . As shown in Figure 5, cysteine residues known to participate in the folding of fungal CBM1s 18 appeared at conserved positions in the oomycete sequences, though an additional cysteine residue was also detected (C²⁴). The conserved doublet of glycine residues found in the N-termini of fungal CBM1s is not conserved in oomycetes but the aromatic amino residues ([WY]³, [WY]³⁹, Y⁴⁰) and the polar residue glutamine (Q⁴²) that are known binding determinants in fungal CBM1s 4 6 17 were frequently identified in the oomycete sequences. Nevertheless, tryptophan and tyrosine residues in these motifs were often replaced by phenylalanine, suggesting greater diversity among the cellulose-binding determinants in oomycetes than in fungi.

To detect structural specificities of oomycete domains, a modeling approach was undertaken using the best characterized oomycetal cellulose-interacting protein, CBEL. This protein is composed of a single polypeptide chain of 266 amino acids which starts with a 20 residue long signal peptide and then is built from the two symmetric N- and C- terminal parts, each resulting from the association of a CBM1 and a PAN/Apple domain (CBM1-1:21–54 + PAN1: 55–133; CBM1-2: 158–190 + PAN2: 191–268) separated by a hinge region (134–157) that is rich in proline and threonine residues. The rather high percentage of identity (~37%) and similarity (~60%) between the T. reseei cellobiohydrolase I CBM1 and the CBM1s of CBEL (Figure 6A) meant that the CBM1s exhibited very similar HCA plots (data not shown), thus allowing the accurate prediction of the three β-sheet strands in both CBEL_CBM1-1 and CBEL_CBM1-2. In addition the four cysteine residues forming two disulphide bridges in TrCBM1 are strictly conserved in the CBEL_CBM1-1 as shown by the alignment of the domains (Figure 6A). Accordingly, the predicted three-dimensional models of CBEL_CBM1-1 and CBEL_CBM1-2, built from the NMR-coordinates of TrCBM1 (RCSB PDB code 1CBH), exhibited a very similar fold (Figure 6B). They consist of three anti-parallel strands of β-sheet (β1, β2 and β3) interconnected by loops. Two disulphide bonds (C⁸-C²⁴ and C¹⁸-C³⁴ in CBM1, C⁸-C²³ and C¹⁷-C³³ in CBM1-2) contribute to the stability of folding, and especially to the stability of the stands (β1 and β3 and loop (loop connecting strands (β2 to β3) which contains the exposed aromatic residues involved in cellulose binding. Both CBM1s contain two exposed aromatic residues F⁵ and Y³¹ in CBM1-1 and Y⁵ and F³⁰ in CBM1-2 that are homologous to Y⁵ and Y³² of TrCBM1 and are thus suitability positioned to stack onto cellulose chains (Figure 6B and C). An additional Q³⁴ residue is also known to participate in the binding of TrCBM1 to cellopentaose and cellohexose 39 . This residue is conserved in CBEL (Q³³ in CBEL_CBM1-1, Q³² in CBEL_CBM1-2), suggesting that it could participate in binding to crystalline cellulose (Figure 6C).

Although the PAN/Apple domains of CBEL share low identity (~20%) and similarity (~36%) with the PAN/Apple domain of the human hepatocyte growth factor (HGF), their HCA plots suggested a very similar structural organization (data not shown). All PAN/Apple domains consist of three and four strands of β−sheets separated by a short α-helical segment. The four cysteine residues linked by two disulphide bridges occurring in the HGF PAN/Apple domain are conserved in both CBEL_PAN/Apple domains. The three-dimensional models of CBEL_PAN1 and CBEL_PAN2 domains built from the NMR-coordinates of the N-terminal human HGF PAN domain (RCSB PDB code 2HGF) consist of a central five-stranded antiparallel β−sheet associated to a short α-helical segment and to two loops containing a short strand of β−sheet (Figure 7A). In the HGF-PAN/Apple domain, the α−helix is linked to two antiparallel strands of β−sheets by an extended loop to form a hairpin-loop region stabilized by two disulphide bridges. A similar harpin-region stabilized by the disulphide bridges C⁸-C²⁴ and C¹⁸-C³⁴ (CBM1-2) or C⁸-C²³ and C¹⁷-C³³ (CBM1-2) also occurs in CBEL. However, the hairpin-loop region is shortened by the deletion of four amino acid residues corresponding to the ⁸⁰LPFT⁸³ motif of the HGF-PAN/Apple domain. The carbohydrate-binding sites of plant lectins usually consist of shallow depressions resulting from the confluence of exposed loops containing hydrophilic and acidic (Asp or Glu) residues that anchor a sugar residue through a network of hydrogen bonds 40 . In addition, an aromatic residue (i.e. F or Y) completes the interaction by stacking against the pyranose ring of the sugar. The two exposed loops ¹⁸Asn-Val-Asp-Phe-Arg-Gly-Asp-Asp³⁵ of PAN1 or ¹⁸Asp-Lys-Asp-Tyr-Arg-Gly-Asn-Asp²⁵ of PAN2, and ⁶⁶Ser-Gly-Thr-Gly-Thr-Arg-Thr⁷² of PAN1 or ⁶⁶Ser-Ala-Ala-Gly-Thr-Ala-Thr⁷² of PAN2, fulfill these structural requirements and thereby could act as the carbohydrate-binding sites responsible for the hemagglutinating activity of CBEL 18 41 .

Six amino acid residues (NYYQCL for CBM1-1/PAN1, SFYQCI for CBM1-2/PAN2) that form the transition between the CBM1s and PAN domains in CBEL correspond to a strand of β−sheet (YQCL and YQCI in CBEL_CBM1-1 and CBEL_CBM1-2 respectively). The first strand of β−sheet (LDLPA in CBEL-PAN1 and IQPPA in CBEL-PAN2) of the CBEL-PAN/Apple domains also contains the ultimate leucine or isoleucine residue of the β−strand of the CBEL-CBM1s. This suggests that a continuous strand of β−sheet (YQCLDLPA and YQCIQPPA) interconnects the CBM1 and PAN domains in the CBEL molecule. Using this information, a tentative three-dimensional model was built for the CBM1-1/PAN1 N-terminal part of CBEL (Figure 7B). Although speculative, this model readily fulfills geometric constraints (>70% of residues occur in the allowed areas of the Ramachandran plot and >90% in the generously allowed area) and reasonably accounts for both the associated cellulose-binding and carbohydrate-binding properties previously reported for CBEL 18 41 . A very similar model was obtained for the CBM1-2/PAN2 C-terminal part of CBEL (result not shown).

In initial studies, the lectin activity of CBEL was demonstrated by a hemagglutination test 18 . To get further insight into CBEL lectin activity and confirm the presence of the carbohydrate binding site inferred from molecular modelling, Surface Plasmon Resonance (SPR) technology was used. Since the detection of molecular interactions using this technology is highly dependent on the molecular weight of the analytes 42 , rather that attempting the direct measurement of the interaction of CBEL with various low molecular weight single sugars, we decided to check whether single sugars could interfere with the binding of large glycoproteins to CBEL. Screening of a collection of various pure glycoproteins allowed to detect the association of CBEL with human lactotransferrin and with a melon hydroxyproline-rich glycoprotein 43 44 , both glycoproteins sharing the fact that their glycan moiety contain galactose residues. Subsequent assays showed that the respective protein complexes were strongly destabilized by N-acetylgalactosamine (Figure 8). Galactose also showed some activity in this assay, though less pronounced. This data confirms the lectin activity of CBEL and suggests that it is involved in binding to polysaccharides or glycoproteins through the interaction with N-acetylgalactosamine or galactose residues.

Protein sequences were collected from the CAZy database 1 , and then curated for oomycetes sequences. Oomycetes proteomes (Phytophthora infestans, P. sojae, P. ramorum, P. capsici, Albugo laibachii, Hyaloperonospora parasitica) were downloaded from the Broad Institute or the JGI, and submitted to an InterPro analysis to detect CBM1-containing proteins 58 . The recently sequenced fungal genomes, not yet referenced in the CAZYdatabase, were added after screening their proteome with the InterPro software to identify CBM1-containing proteins (e.g. Chaetomium globosum). Only sequences with E-values below 10^-6 for PFAM or SMART domains corresponding to IPR000254 – Cellulose binding domain, fungal – were kept to minimize false positives. For a list of completely sequenced organisms that were used in this analysis see Additional file 1: Table S1.

All CBM1-containing proteins were submitted to a local InterProScan to identify other domains in the peptide sequences from the SMART and PFAM databases. Domains identified by both the SMART and PFAM databases were merged after checking their compatibility (location and InterPro identifiers equivalence) by adjusting domain boundaries to the largest domain. Overall, no domain inconsistencies were found between the two databases.

Previously determined protein domain architectures were summarized as follows to obtain more general architecture classes. When multiple CBM1 domains were found in the protein, the architecture is denoted CBM1s. Domains appearance on the sequence was reordered alphabetically except for glycosyl-hydrolase domains i.e., both CBM1-LYA architecture (2 proteins) and LYA-CBM1 architecture (2 proteins) were set to CBM1-LYA whereas CBM1-GH5 architecture and GH5-CBM1 architecture were kept distinct. Multiple occurrences of BNR – bacterial neuramidase repeat – were simplified to only one occurrence i.e., BNR-BNR-BNR-BNR-BNR-CBM1 was mapped to the BNR-CBM1 class. This was motivated by the fact that the number of BNR found in proteins varies from 5 to 8 occurrences and this domain is found in only 5 CBM1-containing proteins. After this generalization, there are 35 distinct classes of protein domain organization. These classes were used to build a contingency table of species and architecture classes. The rows correspond to species profiles: this provides for each species the number of proteins found exhibiting each domain organization. The columns correspond to protein architectures providing the number of proteins found in each species exhibiting such a domain organization. The species and architecture profiles were grouped by hierarchical clustering (average linkage using Euclidean distance) and the resulting classifications were used to draw a heat map in which cell intensities reflect the number of proteins found having a given architecture (column) in a given species (row).

The HCA (Hydrophobic Cluster Analysis 59 ) was performed to delineate the conserved secondary structural features (strand of β-sheet and stretches of a-helix) along the amino acid sequence of CBEL by comparison with the CBM1 of the cellobiohydrolase I from Trichoderma reesei 15 and the PAN domain of the human hepatocyte growth factor (HGF, 60 ) used as models. Multiple amino acid sequence alignments were carried out with CLUSTAL-W 61 . HCA plots were generated using the program drawhca of L. Canard ( http://mobyle.rpbs.univ-paris-diderot.fr/cgi-bin/portal.py?form=HCA). Molecular modeling of the CBM1s and PANs of CBEL was carried out on a Silicon Graphics O2 R10000 workstation using the program InsightII, Homolgy and Discover3 (Accelerys, San Diego CA, USA). The atomic coordinates of the CBM1 of the cellobiohydrolase I of T. reesei (RCSB Protein Data Bank code 1CBH) and the PAN domain of the human HGF (RCSB Protein Data Bank code 2HGF), were used to build the three-dimensional models of the homologous CBEL domains. Steric conflicts were corrected during the model building procedure using the rotamer library 62 and the search algorithm implemented in the Homology program 63 to maintain proper side-chain orientation. An energy minimization of the final models was carried out by 100 cycles of steepest descent using the cvff (consistent valence force field) forcefield of Discover and keeping the cysteine residues involved in disulphide bridges. The program Turbo-Frodo was run to draw the Ramachandran plot and to perform the superposition of the models. PROCHECK 64 was used to assess the geometric quality of the three-dimensional models. Molecular cartoons were drawn with PyMOL (W.L. DeLano, http://pymol.sourceforge.net).

SPR analysis was conducted on a Biacore X device (GE Healthcare, Saclay, France) set at a flow rate of 5μL/min using the CBEL glycoprotein purified from P. parasitica mycelium 18 . CBEL fixation on the sensor chip was achieved by hexadecyl-3-methylammonium bromide (CTAB) micelle-mediated immobilization under the following conditions: the sensor chip surface was first equilibrated in a 10 mM N-2-hydroxyethylpiperazine-N’-2-ethanesulfonic acid (HEPES) pH 7.4 buffer containing 1 mM CTAB. The sensor chip surface was then washed with 5 μl of 10 mM NaOH, and the carboxymethylated dextran sensor surface was activated by 35 μl of a mixture (1v/1v) of 100 mM N-hydroxy-succinimide and 400 mM N-ethyl-N’-(3-diethylaminopropyl) carbodimide. This activation was followed by injection of 80 μl of a solution of CBEL (12.5 μM) dissolved in 10 mM Hepes pH 7.4, 1 mM CTAB. Remaining ester groups were blocked by 35 μl of 100 mM ethanolamine chlorhydrate pH 8.5, before injection of 1 μl of 10 mM NaOH. Solutions of various pure standard glycoproteins dissolved in 10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% vol/vol surfactant P20 (HBS-Ep buffer, Biacore) were then injected into the flow cell and passed over the CBEL surface during 1 min, and their interaction was followed in real time at different analyte concentrations. The tested glycoproteins were fetuin and asialofetuine (Sigma-Aldrich), human lactotransferrin (a kind gift of Dr H Debray, Université des Sciences et Technologies, Lille, France), and melon Hydroxyprolin-Rich GlycoProtein (HRGP; 43 ).

Fetuin is a heavily N- and O-glycosylated protein where the most abundant N-linked glycans are triantennary species, and where both N- and O-linked glycans are sialylated on galactose terminal residues 65 . Human lactotransferrin glycans are of the sialyl N-acetyllactosaminic type and are fucosylated on N-acetylglucosamine residues 43 . HRGP is a O-glycosylated cell wall protein containing arabinose and oligoarabinoside side chains linked to hydroxyproline residues, and galactose units linked to serine residues 43 66 . Specificity of the CBEL-glycoprotein interactions was checked by measuring the level of protein complex dissociation in presence or absence of various monosaccharides. Complex dissociation was calculated at a fixed time point after injection during 5 minutes, at the beginning of the dissociation phase, of either HBS-Ep buffer or glucose, galactose or N-acetylgalactosamine (GalNAc) solutions. Data were analysed using the BIAviewer 3.1 software (Biacore AB, Uppsala, Sweden).