Structural base for the transfer of GPI-anchored glycoproteins into fungal cell walls

Significance

The synthesis and maturation of the cell wall are essential for fungal life. Factors involved in these processes provide a potential platform with side effect-free, antimycotic targets. A key step in cell-wall biogenesis is the transfer of GPI-anchored proteins from the plasma membrane to the glycan meshwork as catalyzed by Dfg5 enzymes. We characterize the structure of a member of the Dfg5 subfamily and provide insights into the architecture of the GPI-core glycan when bound to Dfg5 as supported by MD simulations and in vivo analysis in yeast. Our data let us propose a sorting mechanism at the fungal cell surface and set a base for structure-based drug development using this important cell-wall biosynthesis factor.

Abstract

The correct distribution and trafficking of proteins are essential for all organisms. Eukaryotes evolved a sophisticated trafficking system which allows proteins to reach their destination within highly compartmentalized cells. One eukaryotic hallmark is the attachment of a glycosylphosphatidylinositol (GPI) anchor to C-terminal ω-peptides, which are used as a zip code to guide a subset of membrane-anchored proteins through the secretory pathway to the plasma membrane. In fungi, the final destination of many GPI-anchored proteins is their outermost compartment, the cell wall. Enzymes of the Dfg5 subfamily catalyze the essential transfer of GPI-anchored substrates from the plasma membrane to the cell wall and discriminate between plasma membrane-resident GPI-anchored proteins and those transferred to the cell wall (GPI-CWP). We solved the structure of Dfg5 from a filamentous fungus and used in crystallo glycan fragment screening to reassemble the GPI-core glycan in a U-shaped conformation within its binding pocket. The resulting model of the membrane-bound Dfg5•GPI-CWP complex is validated by molecular dynamics (MD) simulations and in vivo mutants in yeast. The latter show that impaired transfer of GPI-CWPs causes distorted cell-wall integrity as indicated by increased chitin levels. The structure of a Dfg5•β1,3-glycoside complex predicts transfer of GPI-CWP toward the nonreducing ends of acceptor glycans in the cell wall. In addition to our molecular model for Dfg5-mediated transglycosylation, we provide a rationale for how GPI-CWPs are specifically sorted toward the cell wall by using GPI-core glycan modifications.

Glycosylphosphatidylinositol (GPI) anchors are posttranslational modifications in eukaryotes for attaching proteins to the membrane. In fungi, a large subset of GPI-anchored proteins (GPI-APs) is transferred from the plasma membrane (PM) to the cell wall, hence known as GPI-anchored cell-wall proteins (GPI-CWPs) (1–3). The cell wall is a unique feature of fungi consisting of structured layers of polysaccharides and proteins, of which the inner layers consisting of chitin and β1,3-glucans are highly conserved among fungi, whereas the outer layers vary widely (2, 3). Its biosynthesis requires a sophisticated system of enzymes whose activities have to accomplish two apparently opposing goals: maintenance of osmotic integrity and adaptation to growth and the environment. Membrane-embedded glycosyltransferases extrude either linear β1,3-glucan (Fks1-type) or chitin (Chs-type) precursors into the extracellular space, where glycoside hydrolases (GHs) are responsible to realize the functional duality of the cell wall by elongating, cross-linking, and branching the three-dimensional meshwork of polysaccharides and proteins (3, 4). Disturbing this fine-tuned system provides an attractive, potentially side effect-free platform for the development of antimycotics (5–7).

In ascomycetal fungi, GPI-CWPs fulfill a plethora of functions and act, for example, as adhesins, enzymes, or sensors of cell-wall integrity. In many ascomycetes, including human pathogens like the yeast Candida albicans, GPI-CWPs form the largest group of cell-wall proteins and are essential for pathogenicity, mating, and viability (3, 8–10). Due to the prevalence of GPI-CWPs in fungal cell walls, their attachment to the wall is an important task in fungal life. In budding yeast and C. albicans, two homologous proteins from the GH76 family, Dfg5 and Dcw1, are thought to be responsible for membrane-to-wall transfer of GPI-APs by cleaving the Manα1,4-GlcN bond of the GPI-core glycan and attaching it to the β1,6-glucan layer (Fig. 1A) (11–14). In contrast to yeasts, the role of GH76 proteins in filamentous fungi is less clear. In Neurospora crassa, two of nine GH76 proteins are suggested to be involved in cell-wall biosynthesis by cleaving and transferring N-linked outer-chain mannans onto the cell wall (15, 16). For the human pathogen Aspergillus fumigatus, the deletion of six of its eight GH76 paralogs is not lethal but abolishes the transfer of GPI-anchored galactomannans to the cell wall (17). However, biochemical and structural data on GH76 proteins involved in membrane-to-wall transfer of GPI-APs are missing so far, and the exact catalytic mechanism of these mostly themselves GPI-anchored enzymes is largely unknown.

Fig. 1.

Transglycosylation of GPI-APs mediated by GH76 subfamily members. (A) The principal procedure of the GH76-catalyzed incorporation of GPI-APs into the cell wall of fungi is shown schematically. While the GPI-AP becomes a GPI-CWP, which is covalently attached to the cell wall via its GPI-anchor remnant to an acceptor glucan, the glucosamine-phosphatidylinositol remains in the plasma membrane. (B) The GH76 family (Pfam entry PF03663) is represented as an SSN using sequences with a length of 300 to 700 amino acids. Each of the nodes represents sequences with more than 40% sequence identity that are connected by edges using a BLAST E-value cutoff of 10⁻³⁰. The exported nodes can be found in Dataset S1. (C) The overall structure of CtDfg5 is shown as a cartoon with the secondary structure elements and the termini indicated. The box shows a structural alignment of proteins from three different subfamilies (PDB ID codes 4BOK, 4MU9, and 6RY0) with the active-site DD motif being highly conserved in all structures. The coloring of different subfamily members is according to the colors defined by the SSN.

Here, we aim at obtaining a detailed understanding of the transfer of GPI-APs into the fungal cell wall. We report a crystal structure of a fungal GH76 protein from the Dfg5 subfamily in complex with a reassembled GPI-core glycan. These findings are strengthened by using the reassembled GPI-core glycan•enzyme complex for molecular dynamics (MD) simulations. Together with a complementary, structure-based functional mapping of residues in the GPI-binding pocket, we provide a base for understanding not only the mechanism but also how the GPI anchor is used as a coding system to determine the final destination of GPI-APs at the cell surface. Sequence analyses show that the Dfg5 subfamily is an isofunctional subfamily of GH76 proteins, which clearly differs from other bacterial, fungal, and archaeal GH76 subfamilies with annotated α1,6-mannanase function.

Results

Fungi Possess GH76 Homologs That Cluster in 10 Distinct Subfamilies.

Within a protein family it is commonly observed that, during evolution, subfamilies have arisen which fulfill similar tasks but diverge in function and regulation. In the case of enzymes, subfamilies often differ with regard to preferred substrates that share a common core structure. To discriminate isofunctional subfamilies within the GH76 family (PF03663), we subjected this family to a sequence similarity network (SSN) analysis as provided by the Enzyme Function Initiative-Enzyme Similarity Tool (EFI-EST; Fig. 1B) (18). This analysis included a total of 7,485 proteins and revealed clusters for 10 distinct subfamilies, with three clusters comprising almost exclusively bacterial orthologs, two mixed clusters encompassing bacterial, archaeal, and fungal gene products, and five clusters consisting only of fungal proteins. The characterized GH76 family members from Bacillus cereus (Protein Data Bank [PDB] ID code 4BOK) and Bacteroides thetaiotaomicron (PDB ID codes 4MU9, 4C1S, and 4V1R) belong to subfamilies Bacteria I and II (Fig. 1B) (19–21). Besides a fungal/bacterial mixed subfamily, fungal proteins are found in three subfamilies composed of proteins solely from Ascomycetes (Ascomycota I to III), one subfamily from both Basidio- and Ascomycetes (fungal mixed I), and a smaller subfamily from Asco-, Basidio-, and Chytridiomycetes (fungal mixed II). Dfg5 (UniProt accession no. Q05031) and Dcw1 (UniProt accession no. P36091) from Saccharomyces cerevisiae are found in Ascomycota I, which represents the largest GH76 cluster (2,881 orthologs) and hence was named the “Dfg5 subfamily.” Interestingly, no further homologs from S. cerevisiae are found in the network, which is in stark contrast to filamentous fungi such as A. fumigatus, N. crassa, or Chaetomium thermophilum (SI Appendix, Table S1). These findings are supported by the codistribution of GH76 subfamilies among ascomycetes, indicating that the Dfg5 subfamily is indispensable for ascomycetal life and strengthening the applied cutoff criteria for network generation in order to derive isofunctional separation of GH76 subclusters (SI Appendix, Fig. S1).

Overall Structure of CtDfg5.

It has been shown that members of the Dfg5 subfamily of A. fumigatus are involved in the transfer of GPI-anchored galactomannans to the cell wall and that one of these proteins is sufficient to complement a temperature-sensitive dfg5 mutant of S. cerevisiae, emphasizing the importance of the Dfg5 subfamily in membrane-to-wall transfer of GPI substrates in fungi (17). Whereas the function of Dfg5 proteins has been extensively characterized on the genetic level, biochemical and structural data are missing so far (11, 13, 22). Thus, we chose one of the six Dfg5 orthologs (UniProt accession no. G0S3F2; CtDfg5) from C. thermophilum to shed light on this subfamily of GH76 proteins.

We determined the crystal structure of CtDfg5 at a resolution of 1.05 Å with a continuous electron density map covering residues Q32 to F442 (Fig. 1C and SI Appendix, Table S2). Twelve α-helices and two antiparallel β-strands between helices 5/6 and 7/8 form the (α/α)₆ helical barrel fold of CtDfg5, which closely resembles the bacterial GH76 orthologs (19, 21), with rmsd values of 3.0 and 3.1 Å for 327 and 345 equivalent Cα positions, respectively. In contrast to bacterial GH76 homologs, the C terminus of CtDfg5 does not end at α12 but spans over the back of the domain before starting to form an additional helical segment at the other side. This additional helix is in close proximity to a stretch of the α3-α4 loop (P124 to S130), which shows an increased degree of flexibility according to its B factors and electron density. This stretch is adjacent to the DD motif (D134/D135), which is the highly conserved catalytic hallmark of all known GH76 homologs (Fig. 1C, box) (19, 21).

Glycan Fragment Screening for Reassembling the Dfg5-Bound GPI-Core Glycan.

For a mechanistic understanding of Dfg5-type transglycosidases, we aimed for substrate-bound states of CtDfg5. Because no GPI-core oligosaccharides (Manα1,2-Manα1,6-Manα1,4-GlcNα1,2-Ino) were available to us in sufficient amounts, we used a number of mono- and disaccharides for crystal soaking. Except for the inositol, all other hexoses of the core glycan could be unambiguously identified by electron density in the binding pocket of CtDfg5: mannose, α1,2-mannobiose, α1,6-mannobiose, and glucosamine (Fig. 2A and SI Appendix, Fig. S2 and Table S2). Mannose exclusively binds to the −2 subsite of CtDfg5 (SI Appendix, Fig. S2A), a binding mode that is comparable to that of mannose in bacterial GH76 mannanases, for instance BcGH76 (PDB ID code 5AGD). In CtDfg5•α1,2-mannobiose, the position of the central −2 mannose is further extended by the second α1,2-linked mannose to subsite −3 (SI Appendix, Fig. S2B). The CtDfg5•α1,6-mannobiose occupies subsites −2 and −1, the position next to the active site (SI Appendix, Fig. S2C). Finally, the structure of CtDfg5•glucosamine revealed the position of the ultimate GPI-core glycan component at subsite +1 (SI Appendix, Fig. S2D).

Fig. 2.

Reassembled GPI-core glycan in the binding pocket of Dfg5. (A) The individual glycan fragment states are superimposed and shown in the context of the binding pocket with labeled GPI-core glycan subsites. (A, Inset) A schematic GPI anchor as shown in Fig. 1A. Residues are indicated for orientation. (B) As indicated by the arrow, the fragments are merged together to assemble the GPI-core tetrasaccharide (orange) spanning glycan-binding subsites −3 to +1. By superimposing the substrate-bound state from BcGH76 (5AGD) with CtDfg5, the mannopentaose (green sticks) is projected as transparent sticks into the binding pocket. (C and D) Differences in the +1-subsite coordination of CtDfg5 and BcGH76 (5AGD) are shown with accordingly colored sticks (nitrogen, blue; oxygen, red). (E) The GPI-core glycan (balls and sticks with gray carbons) shows a similar conformation within the binding pocket of Dfg5 in MD simulations as observed in the fragment screen. (F) Snapshot from MD simulations shows the GPI-anchored ω-peptide (spheres) bound to a membrane-embedded Dfg5 (orange cartoon). E and F contain D135 (black sticks; hydrogen, white; nitrogen, blue; oxygen, red) for orientation and were generated with the Protein Imager (43).

By using this structural information, we can assemble the structure of a U-shaped GPI-core glycan (Manα1,2-Manα1,6-Manα1,4-GlcN) spanning subsites −3 to +1 when bound to CtDfg5 (Fig. 2 A and B and Movie S1). Compared with the binding mode of mannopentaose in BcGH76 (PDB ID code 5AGD), subsites −2 and −1 with their Man-α1,6-Man moiety look highly similar except for a hydrophobic block established next to subsite −2 by F266, which is necessary to accommodate the α1,2-extension of the GPI-core glycan. Superposition of the CtDfg5•GPI-core glycan and BcGH76•mannopentaose complexes shows that the bacterial linear substrate does not fit into CtDfg5 due to the hydrophobic block by F266 (Fig. 2B). Further striking differences in substrate binding are apparent at subsite +1. In CtDfg5, glucosamine is bound from the front by Q189 and stacking of Y81 from the back (Fig. 2C). The opposite arrangement is found in the α1,6-mannanase BcGH76, where the C6 atom is in front (Fig. 2D) due to its α1,6-linkage and stacking to F122, while an H bond with D71 stabilizes the sugar conformation from the back.

In order to validate our structural findings, we also performed MD studies for the GPI-core glycan in complex with CtDfg5 (Fig. 2E and Movie S2), first as it was reassembled above and, secondly, in the context of a lipid bilayer using a membrane-anchored version of CtDfg5 and the GPI-anchored ω-peptide from the yeast cell-wall protein Flo5 (Fig. 2F and Movie S3). For the plain core glycan, it was necessary to restrain mannose at subsite −1 to the proposed catalytically active ^OS₂ conformation (21) to maintain a stable complex. With this restriction, the GPI-core glycan was stable during the 0.5-μs trajectories in a U-shaped conformation as observed in our glycan fragment screen. Most distances between the core glycan and residues of its binding site are maintained throughout the simulation (SI Appendix, Table S3). Major differences between the experimentally derived GPI-core model and the MD poses of the CtDfg5•GPI-core glycan complex are found for the glucosamine moiety. Here, the protonated amino group of the glucosamine approaches D135. This interaction is not realized in the model between membrane-attached CtDfg5 and the intact GPI anchor with an attached ω-peptide (Fig. 2F). In this membrane-bound system, the GPI-core glycan again adopts the U-shaped conformation during the entire trajectories and the glucosamine moiety mimics the interactions found in the CtDfg5–glucosamine structure, especially the interaction with Q189 (Fig. 2C). However, the amino moiety of the glucosamine makes a stable hydrogen bond with the carbonyl group of S130, an interaction that has not been found in the experimental structure. Relative to other phospholipids of the outer leaflet, the lipidic phosphatidylinositol (PI) moiety is pulled out of the lipid bilayer by about ∼5 Å due to the interactions between the GPI anchor and Dfg5. Given the association of CtDfg5 with the membrane-embedded GPI-anchor substrate, several loop regions can form putative interactions with the polar membrane surface. These include two elongated loop regions, R177 to T202 and D242 to Q265, which associate together via a small β-sheet flanking the core glycan and the ω-peptide along the protein–membrane interface (SI Appendix, Fig. S3). Interestingly, we observe for the membrane-bound system an approach of the nucleophile D134 toward the anomeric carbon of mannose at subsite −1 that is required for the nucleophilic attack, as known from the bacterial BcGH76•mannopentaose complex (SI Appendix, Table S3). Another difference between the crystallographic data and our MD-derived CtDfg5–membrane model is given for D325. The glycan fragment screen showed interactions with the 6-hydroxyl group of the glucosamine, which get lost in the MD simulations. Likewise, we observed no interactions between the inositol moiety of the GPI anchor and CtDfg5 in our simulated CtDfg5–membrane model but hydrogen bonds between its attached phosphate and the side chains of T74 and S130.

In Vivo Analysis of Dfg5 Mutants Reveals Essential Functions for Binding-Pocket Residues.

To identify evolutionarily conserved amino acids within the Dfg5 subfamily, we analyzed the CtDfg5 structure by using ConSurf (23, 24). This analysis underscores the importance of the GPI-core glycan-binding pocket, as most of its residues are attributed with the highest conservation scores (Fig. 3A and SI Appendix, Fig. S4). Using S. cerevisiae, we conducted a structure-based in vivo analysis of Dfg5. Accordingly, we performed a mutational mapping of conserved residues within or close to the active site of Dfg5, which were based on a homology model using CtDfg5 (sequence identity ∼40%), as template structure (Fig. 3A, box). A set of 18 DFG5 mutant genes was generated and analyzed for functionality. We assayed the ability of these mutants to rescue the synthetically lethal phenotype of yeast strains carrying both dfg5Δ and dcw1Δ chromosomal mutations by a 5-fluoroorotic acid (5-FOA)-based plasmid loss assay (Fig. 3B).

Fig. 3.

Structure-based in vivo mutational analysis of the Dfg5-binding pocket in yeast. (A) ConSurf analysis of the evolutionary conservation of amino acids within the Dfg5 subfamily. The color scheme ranges from cyan (variable) to maroon (conserved) and shows a high degree of conservation in the GPI-core glycan-binding pocket; CtDfg5 residues are colored according to conservation scores. Residues analyzed by in vivo mutation using Dfg5 of S. cerevisiae are depicted with their homology model (gray; labels, black). (B) Plasmid loss assay used for the in vivo mutational analysis of Dfg5 in S. cerevisiae. Yeast strains carrying chromosomal DFG5 mutant genes and an intact variant of DFG5 on a URA3-based rescue plasmid were spotted on nonselective (SC, synthetic complete) and selective plates (5-FOA). Growth was monitored after 3 d of incubation. For comparison, rescue by nonmutated chromosomal DFG5 (control) and loss of rescue by the absence of DFG5 (NC, negative control) are shown. (C) Growth curves of different yeast strains grown in YEPD medium (no stress) and in the presence of medium supplemented with 30 µg/mL Calcofluor white. Expression of Dfg5, Dcw1, Dfg5^N188A, both paralogs (control), or no Dfg5 variant (∆∆), together with corresponding growth rates, is indicated by different colors. (D) Microscopic analysis of different yeast strains grown under no stress conditions and stained by 1 µg/mL CFW. (D, Upper) Fluorescence of CFW bound to the chitin component of the cell wall. The relative fluorescence intensity quantified against the control strain is indicated. (D, Lower) Cells under bright light (differential interference contrast; DIC). (Scale bars, 5 µm.)

Our mutational analysis reveals that the highly conserved DD-motif GH76 active site is essential for the functionality of Dfg5 enzymes, as demonstrated by the D122N and D123N variants. In addition, residue W175, which coordinates the GPI-core glycan above the active site, also adopts an essential role, as shown by the lethality of the W175A mutation. Furthermore, the conserved YYWW motif (Y68 to W71) preceding the α2-helix is crucial for functionality, as shown by the growth phenotypes of their alanine mutants. This finding supports the structure-based conclusion that Y69 and W71 are involved in binding the GPI-core glycan at subsite +1, whereas the other two act as hydrophobic anchors of this loop at helices α4 and α12. Interestingly, the conservative Y69F mutation does not affect viability, indicating that this variant does not significantly lose the ability to interact with the donor substrate (refer to Y81 in SI Appendix, Fig. S2D). The two residues Q176 and D313 are also required for functionality, supporting the view that they confer GPI-core glycan binding due to their close proximity to the glucosamine at subsite +1. Four further residues, N188, D238, W254, and Y256, are close to subsite −2 (N201, D250, F266, and Y268 in SI Appendix, Fig. S2C). Our mutational analysis reveals that D238, W254, and Y256 are essential for regular growth, while the H bond established by N188 is not crucial for binding the mannose moiety at subsite −2 under these conditions. Residue Y186, which forms a stabilizing H bond with D238 via its hydroxyl group, is not required for functionality. Finally, mutational analysis of residues T250 and N311, which are close to subsite −3 of the GPI-core glycan-binding site, reveals that N311 is involved in mannose binding, as demonstrated by the reduced growth of the N311A mutant. In summary, our mutational in vivo analysis supports the resulting GPI-core glycan-binding mode resulting from our structural findings.

Reduced Dfg5 Activity Causes Hypersensitivity to Cell-Wall Perturbation.

Given their high degree of redundancy for in vivo function under nonstress conditions (Fig. 3C), we wondered how expression of only one of the two Dfg5 paralogs affects cell growth under stress conditions. For this purpose, we measured the growth rates of appropriate yeast strains in the presence of five agents known to disturb either the cell wall (Calcofluor white [CFW], Congo red, and caspofungin) or the cell membrane (sodium dodecyl sulfate and NaCl). We found that absence of either Dfg5 or Dcw1 led to hypersensitivity in the presence of CFW but not in the presence of the other agents tested (Fig. 3C and SI Appendix, Fig. S5). The addition of 30 µg/mL CFW reduced the growth rate 2.4-fold in the case of a strain expressing both paralogs but by a factor of 4.3 for a strain expressing only Dfg5. Interestingly, a more than 27-fold reduced growth rate was observed for the strain expressing only Dcw1. The prominent role of Dfg5 is corroborated by our finding that the growth rate of the strain expressing only Dfg5^N188A, a mutant that affects the GPI-binding pocket at the −2 subsite, is reduced by a factor of 8.6. Overall, these findings show that cell-wall composition and architecture with their resulting tolerance to cell-wall stress agents are differently affected by the two Dfg5/Dcw1 paralogs mediating the membrane-to-wall transfer.

CFW is a specific cell-wall stressor, as it not only binds to cell-wall chitins but probably also inhibits chitin processing (25). Accordingly, we analyzed Dfg5/Dcw1 mutants grown under nonstress conditions to check whether deficiencies of Dfg5 or Dcw1 cause per se cell-wall stress as reflected by elevated chitin abundance (Fig. 3D) (26). In strains expressing both paralogs or Dfg5 alone, we found only slightly elevated changes for cell-wall chitin by about 20 to 30%. In contrast, CFW staining indicated a chitin increase of 120% in the strain expressing only Dcw1. This indicates that in S. cerevisiae the DFG5 gene is dominant for maintaining sufficient membrane-to-wall transfer.

Glcβ1,3-Glc Bound to Subsites +1 to +2 Reflects an Acceptor State for β1,3-Glucans.

The membrane-to-wall transfer mechanism involves not only the GPI-core glycan of GPI-APs as donor but also acceptor glycans of the cell wall, which can either be β1,3-glucans (17, 27) or β1,6-glucans (14, 28). In order to identify possible acceptor glycans for CtDfg5, we performed glycan fragment mapping as in the case of GPI-core glycan fragments, which led to the identification of laminaribiose, Glcβ1,3-Glc, in complex with CtDfg5 (Fig. 4A and SI Appendix, Table S2). Superposition of the GPI-core glycan at the active site within the CtDfg5•Glcβ1,3-Glc complex clearly shows a similar binding mode for the nonreducing end of Glcβ1,3-Glc at subsite +1 as found for the glucosamine of the donor (Fig. 4B). The reducing glucose of Glcβ1,3-Glc is more flexible, but can still be unambiguously identified at the +2 position of the CtDfg5-binding pocket. In the case of the donor GPI-core glycan, an α1,6-linked inositol is thought to be at position +2, which is likely to differ from the position of the β1,3-linked glucose of the acceptor.

Fig. 4.

CtDfg5 catalyzes the transfer of the GPI-core glycan onto the nonreducing end of β1,3-glucan. (A) Laminaribiose (β1,3G) is shown bound to CtDfg5 with its 2mFo − DFc map (blue mesh, contouring level 1.0) and surrounding residues. (B) A superposition of the GPI-core glycan and laminaribiose is shown in the context of CtDfg5. The active-site residues D134 and D135 are shown as sticks. (C) Schematic model of the membrane-to-wall transfer as catalyzed by Dfg5 enzymes. An EtN-P at Man1 of the GPI anchor inhibits binding and catalysis by Dfg5 subfamily enzymes, thus defining GPI-PMPs. Upon binding of correctly processed GPI-APs at the plasma membrane (PM), the Manα1,4-GlcN linkage within the GPI-core glycan is cleaved and a covalent intermediate with Dfg5 is formed. Upon binding of the acceptor glycan, transglycosylation proceeds, resulting in a covalently bound GPI-CWP. TM: transmembrane anchor.

Discussion

So far, the action of ascomycetal Dfg5 enzymes on GPI-anchored membrane proteins has lacked profound functional data and structural insights. Previous studies on bacterial α1,6-mannanases of the GH76 family predicted that Dfg5 enzymes, which are themselves bound to the plasma membrane by a GPI anchor or a transmembrane helix, are mechanistically highly related. Our analysis of ascomycetal genome sequences by sequence similarity network analysis shows that the Dfg5 subfamily, which is crucial for the maturation of fungal GPI-CWPs, is indeed the largest of all GH76 subfamilies and ubiquitous in ascomycetes. Given the strict co-occurrence of Dfg5 orthologs (SI Appendix, Fig. S1), members of other ascomycetal subfamilies cannot replace Dfg5 function and probably accept other substrates with a Man-α1,6-Man motif such as N-linked glycans.

Chemical synthesis of entire GPI-core glycans proved to be difficult for structural studies of GPI-processing enzymes such as Dfg5. Our study shows that in crystallo glycan fragment screening is a promising new tool to assemble such complex glycan structures within the binding sites of enzymes and lectins of high substrate specificity. We found that the binding mode of the Manα1,2-Manα1,6-Manα1,4-GlcN GPI-core glycan is U-shaped and occupies the shape-complementary subsites −3 to +1 of CtDfg5. The majority of interactions are made with the central Man-α1,6-Man moiety and highly conserved residues of the GH76 family like W188, N201, and D250 as well as the DD motif and resemble substrate recognition by bacterial GH76 α1,6-mannanases at corresponding subsites. Furthermore, the distinct U-shaped conformation of the GPI-core glycan as bound to Dfg5 is similar to unbound GPI-core glycans when being either part of membrane-resident GPI anchors or in solution as fragments (SI Appendix, Fig. S6 C–E) (29). This conformational propensity that is also found in other complex glycans (30) suggests low entropic costs upon binding.

Our glycan fragment screen identified β1,3-glucans, but not β1,6-glucans, as putative acceptor substrates for Dfg5 enzymes from the sordariomycete C. thermophilum. Interestingly, β1,6-glucans function as acceptor glycans in yeasts, but are missing from most filamentous fungi. The lack of interactions of the acceptor glucose at subsite +2 (Fig. 4A) indicates that only minor changes, if at all, are required to convert the acceptor glycan specificity in Dfg5 enzymes (17). Furthermore, the structure of the acceptor complex shows that the nonreducing end of β1,3-glucobiose is placed into subsite +1, as observed before for the glucosamine moiety of the GPI-core glycan donor (Fig. 4B). This binding mode supports a classical Koshland double-displacement mechanism for transglycosylation by Dfg5 enzymes as their catalytic DD motif (SI Appendix, Fig. S6F) adopts the mechanism-specific distance of 5.1 Å for its carboxylates (31). In the first step, D134 attacks nucleophilically the C1 atom of the mannose moiety at subsite −1, while D135 protonates the 4-hydroxyl group of the leaving GlcN-PI moiety. While this part of the mechanism, which proceeds over the ^OS₂ conformation of mannose −1, is common with the bacterial and all other GH76 orthologs (21), one may wonder why Dfg5 enzymes promote in the next step transglycosylation instead of hydrolysis (SI Appendix, Fig. S6F).

Calcium-based stabilization of the covalent Dfg5•GPI-CWP intermediate would prevent hydrolysis of the covalent adduct, and indeed we detected a Ca²⁺ ion coordinated between D135 and the reducing mannose at subsite −1 of the CtDfg5•Manα1,6-Man complex. Here, incoming acceptor glycans may displace the Ca²⁺ capping in order to promote subsequent transglycosylation. Indeed, the negatively charged plasma membrane is a Ca²⁺-rich microenvironment with Ca²⁺ concentrations up to 30 times higher than bulk, because Ca²⁺ binding to the lipid bilayer is entropically driven by water displacement (32–35). Another scenario is local crowding of acceptor glucans close to the Dfg5•GPI-CWP intermediate as they are synthesized from membrane-resident glucan synthases such as Fks1.

In general, all GPI-APs initially reach the plasma membrane. Only GPI-PMPs remain at this location, whereas GPI-CWPs are transferred to the cell wall. Previously, it was shown that the essential cell-division cycle protein Cdc1 removes the ethanolamine-phosphate (EtN-P) group from mannose −1 of the GPI-core glycan that was added before by Mcd4 during GPI-anchor biosynthesis (36, 37). Our CtDfg5•GPI-core glycan complex shows that EtN-P modification at the 2-hydroxyl group of mannose −1 sterically prevents catalytically competent binding of GPI-APs to Dfg5, explaining why GPI-APs carrying an EtN-P modification at Man1 are not valid substrates for Dfg5 enzymes (SI Appendix, Fig. S6A). Accordingly, membrane-to-wall transfer of these GPI-CWPs in S. cerevisiae cannot occur, providing a rationale for why Cdc1 is essential in yeast. The CtDfg5•GPI complex also predicts that an EtN-P modification at the 6-hydroxyl of Man2 might still allow formation of a Dfg5–substrate complex (SI Appendix, Fig. S6B), thereby tolerating mutations in TED1, a gene responsible for the removal of the EtN-P modification at Man2 (38). Thus, the fate of GPI-CWPs and GPI-PMPs appears to be determined during maturation of the GPI anchor in the secretory pathway by action of Cdc1 and Cwh43 (39), while the downstream-acting Dfg5 enzymes later decode this decision upon arrival of the GPI-APs at the plasma membrane (Fig. 4C).

Materials and Methods

A short overview of the key methods used follows. Full materials and methods can be found in SI Appendix.

Sequence Similarity Network Analysis.

The SSN analysis based on the EFI-EST was performed on PF03663 and visualized by using Cytoscape (40). Further analyses were carried out using Clustal Omega and WebLogo (41, 42).

Recombinant Protein Production and Purification.

Recombinant Dfg5 (UniProt accession no. G0S3F2) from C. thermophilum was produced in SHuffle T7 Express-competent Escherichia coli cells using a low-temperature protocol. A two-step purification was performed using Ni-NTA affinity and size-exclusion chromatography.

Crystallographic Methods.

Initial crystallization hits at pH 7.5, resulting in monoclinic crystals belonging to space group C121, were used for single-wavelength anomalous diffraction phasing using the CRANK2 pipeline derived from data of gadolinium-soaked crystals. Crystals for soaking experiments were obtained from seeding experiments with protein buffered at pH 4.5. These crystals were reproduced for sugar fragment screening using high molar sugar soaks without backsoaking prior to flash freezing. Data processing was performed with XDS, and refinement and manual model building were done with Phenix and Coot.

MD Simulation.

MD simulations were performed for the free CtDfg5•GPI-core glycan complex and the model of a membrane-bound CtDfg5 in complex with a cognate substrate, a GPI-anchored ω-peptide of the yeast CWP Flo5.

Phenotypic Analysis of DFG5 Point Mutations.

Phylogenetically conserved amino acids in the GPI-glycan–binding pocket of Dfg5 identified by ConSurf were analyzed by creating respective mutants of DFG5, which were tested on their ability to substitute a functional DFG5 copy encoded on a URA3-rescue plasmid on an otherwise toxic 5-FOA–containing growth medium.

Growth Curves.

Growth of yeast strains under nonstringent and stress conditions was analyzed. Cultures were inoculated from log-phase precultures and monitored on 96-well plates using a fluorimeter.

Microscopy.

Cells were grown to log phase, stained, and subsequently analyzed by fluorescence microscopy. Quantification of fluorescence was done by transferring cells to 96-well plates, followed by measuring the OD₆₀₀ and the cell fluorescence using a fluorimeter.

Data Availability.

The atomic coordinates of the crystal structures of CtDfg5, CtDfg5•Man, CtDfg5•α1,2M, CtDfg5•α1,6M, CtDfg5•GlcN, and CtDfg5•β1,3G have been deposited in the Protein Data Bank (PDB ID codes 6RY0, 6RY1, 6RY2, 6RY5, 6RY6, and 6RY7, respectively).