Structural basis of fungal β-1,3-glucan synthase inhibition by caspofungin

Abstract

Invasive fungal infections pose life-threatening risks to the increasing population of immunocompromised patients^1,2. Treatment remains challenging due to limited antifungal drugs and increasing resistance. β-1,3-d-glucan synthase (GS), comprising the catalytic Fks1 and the regulatory small GTPase, Rho1 (refs. ^3,4), is the target of clinically important echinocandin antifungals. Despite recent studies^5–7, the mechanisms of GS catalysis, Rho1 regulation and echinocandin inhibition and resistance remain elusive. Here we present cryo-electron microscopy structures of native Saccharomyces cerevisiae Fks1 (ScFks1) solved under catalytically relevant conditions, revealing its interactions with the antifungal caspofungin (CFN), glucan product from the translocation channel and Rho1. CFN forms a ternary complex with nascent glucan and Fks1 at the membrane–protein interface, suggesting an unexpected role of CFN in stalling polymer translocation. Our echinocandin-resistant S643P structure suggests a resistance mechanism: the substitution destabilizes CFN and glucan binding through both allosteric structural perturbation and direct steric clash. Rho1 binding induces active site rearrangements essential for catalysis, including that of the ‘latch loop’ for donor substrate coordination. Furthermore, we identify YMR295C as an auxiliary subunit. These findings elucidate the mechanisms of GS-mediated glucan synthesis and its inhibition and resistance by echinocandins, laying the groundwork for rational antifungal design.

Insights into the structural basis of fungal β-1,3-d-glucan synthase inhibition by caspofungin lay the groundwork for the development of next-generation echinocandins with increased potency, broader antifungal spectrum, and reduced propensity for resistance.

Main

Fungal infection treatments rely primarily on four antifungal classes: polyenes, azoles, echinocandins and 5-fluorocytosine. Echinocandins, including caspofungin (CFN) (Fig. 1a), micafungin, anidulafungin and rezafungin, are used widely for invasive Candida infections—the most common invasive fungal diseases in hospitals—due to their efficacy and minimal side effects. Echinocandins are amphiphilic cyclic hexapeptides with an N-linked acyl lipid chain that exhibit their antifungal activity by inhibiting β-1,3-d-glucan synthase (GS)—the enzyme responsible for the formation of β-1,3-d-glucan, an essential component of the fungal cell wall. GS consists of the Fks catalytic subunit and the regulatory small GTPase, Rho1. Fks is a transmembrane enzyme that catalyses the processive formation of β-1,3-d-glucan using uridine 5′-diphosphate glucose (Glc) (UDP-Glc) as a substrate while transporting the polymer across the cell membrane. Rho1 serves as a regulatory subunit, and its GTP-bound form is required for GS catalytic function. Fks belongs to the unique glycosyltransferase 48 (GT-48) family (http://www.cazy.org/), which is distinct from other membrane-embedded processive glycosyltransferases (GTs), such as cellulose^8–11, hyaluronan^12,13 and chitin synthases^14–17, because it requires neither divalent metals^5,18,19 nor free monosaccharide primers¹⁸. Clinically isolated echinocandin-resistant mutations cluster in three hotspots (HS1–HS3) of Fks^5,20,21. Despite its clinical importance, the molecular mechanisms governing Fks catalysis, echinocandin inhibition, hotspot resistance mutations and Rho1 regulation have remained elusive.

Fig. 1. ScFks1 activity and structures in different detergents.

a, Chemical structure of CFN. b, ScFks1 activity in 0.2% CHAPS/0.04% CHS (w/v), 0.04% GDN (w/v) or 0.005% LMNG/0.0005% CHS (w/v), measured as the rate of Glc incorporation into insoluble glucan (n = 3 technical replicates, presented as mean ± s.d.). c, Dose response curves of ScFks1 in 0.2% CHAPS/0.04% CHS, 0.04% GDN or 0.005% LMNG/0.0005% CHS in varying concentrations of CFN. The fitted IC₅₀ values for CHAPS/CHS, GDN and LMNG/CHS conditions are 0.66 ± 0.11 μg ml⁻¹, 160 ± 30 μg ml⁻¹ and 210 ± 20 μg ml⁻¹, respectively (n = 3 technical replicates, represented as mean ± s.d.). d,e, Cartoon representations of ScFks1 structures purified in LMNG/CHS or CHAPS/CHS viewed from two orientations (d, ScFks1 structures viewed with the TMD above the GTD; e, cross-sectional views at the interface between TMD and GTD of ScFks1 structures rotated 90° relative to structures in d). Grey, TMD; blue-white, GTD; wheat, ScRho1; pale cyan, YMR295C; green, interfacial helices and their corresponding residue. The glucan polymer and CFN are represented as space-filling models. Unresolved residues at the interfacial region are labelled as dashed lines. aa, amino acid.

Saccharomyces cerevisiae Fks1 (ScFks1) shares 50–90% sequence identity with Fks from many pathogenic fungi (Supplementary Fig. 2a,b) and has served as a model for studying echinocandin inhibition²². Recent cryo-electron microscopy (cryo-EM) studies have revealed its overall architecture: ScFks1 comprises a transmembrane domain (TMD) and an intracellular glycosyltransferase domain (GTD) composed of a catalytic and an accessory domain^5,6. However, they do not provide key mechanistic insights into Fks1 catalysis and its inhibition by echinocandins, due to the absence of echinocandin, UDP-Glc or intact glucan polymer product in the structures. In membrane-embedded processive GTs, the interfacial region between the GTD and the TMD is central for function^8–13,16. However, the published ScFks1 structures leave a substantial portion of this region (approximately 50–60 amino acids) unresolved, and the resolved parts differ substantially across models (Extended Data Fig. 1a–d).

Extended Data Fig. 1. Comparison with published Fks1 structures.

a,b, Cartoon representations of previously published ScFks1 structures in GDN (PDB: 7YUY) (a) or LMNG/CHS (PDB: 8JZN) (b) viewed from two orientations. The TMD is colored in grey; GTD in blue-white. Residues corresponding to the interfacial helices in the catalytically relevant states are highlighted in green. Unresolved residues at the interfacial region are labeled as dashed lines. c, Overlay of L1 state with 7YUY by the TM region. d, Overlay of L2 state with 8JZN by the TM region. e, Overlay of S643P with L1 state by the TM region. f,g, Comparison of polymer locations in the CFN-Rho1-long-glucan-bound (f) or the short-polymer-bound state (g) with 7YUY. The structures are overlaid by the TM region. In the catalytically relevant states, the interfacial helices are colored green; the TM region light blue. The glucan polymer is shown as yellow spheres. The putative glucan polymer in 7YUY is shown as blue spheres.

By optimizing in vitro reconstitution of GS activity, we determined catalytically relevant cryo-EM structures of ScFks1 that provide mechanistic insights into Fks catalysis, transport and Rho1 regulation as well as the structural basis of echinocandin inhibition and resistance.

Catalytically relevant conformations

We purified endogenously expressed ScFks1 (Extended Data Fig. 2a–c) and characterized its activity using radioactive assays that quantify insoluble β-1,3-d-glucan product formation¹⁸, rather than the UDP-Glo assay^5–7, which cannot distinguish glycohydrolase and glycosyltransferase activities. The ScFks1 activity varies across different detergents (Fig. 1b). The specific activity of 1,160 ± 90 nmol mg⁻¹min⁻¹ (turnover rate of 256 ± 20 min⁻¹, Fig. 1b) in 3-((3-cholamidopropyl)dimethylammonio)-1-propanesulfonate (CHAPS) and cholesteryl hemisuccinate (CHS) is comparable to the reported activities of the product-entrapped enzyme (1,010–4,040 nmol mg⁻¹ min⁻¹)^18,23 and is 16- and 31-fold higher than GS activity in lauryl maltose neopentyl glycol (LMNG) with CHS or glyco-diosgenin (GDN), respectively (Fig. 1b). Notably, only ScFks1 in CHAPS/CHS is CFN-sensitive, with a half maximal inhibitory concentration (IC₅₀) value comparable to that of previous studies of ScFks1 in the membrane²⁴ or after product entrapment purification¹⁸. ScFks1 in LMNG/CHS or GDN is substantially less sensitive to CFN inhibition (Fig. 1c).

Extended Data Fig. 2. Purification of ScFks1 and ScRho1 and expression and functional characterization of ScFks1 mutants.

a,b, Gel-filtration profiles and SDS-PAGE of purified ScFks1 in 0.005% LMNG/ 0.0005% CHS (a) and 0.1% CHAPS/0.01% CHS (b). c, Gel-filtration profile and SDS-PAGE of purified ScRho1. d, Western-blot analysis of ScFks1 mutants in the spot assay. Experiments in a–d were repeated at least three times (n = 3 independent biological repeats) with similar results. For uncropped gel source data, see Supplementary Fig. 1. e, Dose response curves of ScFks1 WT and S643P against CFN, based on the relative rate of insoluble glucan formation. The fitted IC₅₀ values are 0.26 ± 0.06 and 379 ± 106 µg/mL for WT and S643P mutant. Each data point represents the mean of three assays using two independently prepared enzymes (n = 3; mean ± s.d.).

We first determined the structure of ScFks1 in LMNG/CHS in the presence of CFN. We obtained two conformations, designated L1 and L2, resembling the previous ScFks1 structures in GDN (Protein Data Bank (PDB) 7YUY) and LMNG/CHS (PDB 8JZN), respectively (Fig. 1d, Extended Data Figs. 1 and 3 and Extended Data Table 1). In both L1 and L2, key interfacial regions remain unresolved (Fig. 1d,e), and no CFN density was detected. We then purified the enzyme in CHAPS/CHS, in which the enzyme was as active as in membrane, and performed enzymatic reactions in the presence of CFN on the EM grids before plunge-freezing. Micrographs frequently showed clustered glucan polymers, confirming GS activity under these conditions (Extended Data Fig. 3). Data processing (Supplementary Fig. 3) revealed three distinct states: (1) a ground state lacking glucan, Rho1 and CFN; (2) a short-glucan-bound state with a short glucan polymer in the translocation channel without Rho1 and CFN and (3) a CFN–Rho1–long-glucan-bound state containing an extended glucan polymer, CFN and Rho1 (Extended Data Fig. 3 and Extended Data Table 1). UDP was resolved at the active site in all three states. These three structures fully resolve three interfacial helices (IF1–IF3) that encircle the glucan polymer and conserved loops in the active site (Fig. 1d–e, Extended Data Fig. 4 and Extended Data Table 1). Because these conformations were observed uniquely under highly active conditions, they represent more catalytically relevant conformations than states L1 and L2.

Extended Data Fig. 3. Cryo-EM analysis of ScFks1 WT in LMNG/CHS, WT in CHAPS/CHS and S643P in CHAPS/CHS.

a, A representative micrograph. b, General data processing flowchart, including the final reconstructions and the corresponding local resolution maps. (see Supplementary Fig. 3 for more details). c–g, Local resolution maps (c), representative 2D classes (d), projection orientation distribution maps (e), the gold-standard Fourier shell correlation (FSC) curves (f) and map-to-model correlation plots (g) for each final reconstruction. White arrows in a marks glucan polymer. White arrows in d marks Rho1.

Extended Data Table 1.

Cryo-EM data collection, refinement, and validation statistics

Extended Data Fig. 4. Representative EM densities and comparison with published Fks1-Rho1 complex structure.

a–c, EM densities of key structural elements in the ground (a), the short-glucan-bound (b), and the CFN-Rho1-long-glucan-bound ScFks1 (c) states, including IF1-3, the latch loop, the coupling module, ligands, and the Fks1-YMR295C interface. d, Representative cryo-EM density at the glucan exit gate of the S643P mutant. Residue P643 is highlighted in red. e, Side-by-side comparison of EM densities of the Fks1-Rho1 interface in this study and the Fks1-Rho1 interface in the crosslinked Fks1-Rho1 structure (PDB 8WLA and EMD-37614). The rightmost panel in each row is the close-up view of the region marked by the black rectangles. In a–e, models are shown as sticks in all panels.

Our EM density also led us to identify YMR295C (GSR1) as an additional subunit of the GS complex (Extended Data Fig. 4, Supplementary Fig. 4, Supplementary Table 1 and Supplementary Text).

The glucan translocation channel

In the CFN–Rho1–long-glucan-bound state, we resolved a 17-unit glucan polymer extending from the active site to the extracellular side, defining the complete translocation channel formed by TM5–TM10 and IF1–IF3 (Fig. 2a). Robust EM density allowed for unambiguous modelling of all Glc units, except for the flexible reducing end. Owing to β-1,3-linkages, the polymer adopts a helical conformation with a width of approximately 13.6 Å and pitches of 21–25 Å. (Fig. 2b and Extended Data Fig. 5a,b). This represents a compressed state compared with an energy-minimized glucan model by GLYCAM (glycam.org), yet it is more stretched than the experimental triple-helix structure (17.4 Å pitch)^25,26. By contrast, cellulose, a polymer of β-1,4-linked Glc units, adopts a linear geometry¹¹ (Extended Data Fig. 5c,d).

Fig. 2. The glucan polymer in the translocation channel of ScFks1.

a, Cartoon representation of the glucan translocation channel in the CFN–Rho1–long-glucan-bound state. The polymer and CFN are shown as sticks and their EM maps (contoured at 5σ) as grey surface. b, Left, EM map (contoured at 5σ) and model of the glucan polymer of 17 Glc units. Right, energy-minimized glucan polymer of 17 Glc units. c, Detailed interactions between ScFks1 and the Glc_0–5 and UDP at the active site. d, Detailed interactions between ScFks1 and Glc_3–10. The water molecule is shown as a red sphere. Red stars highlight glycosidic linkages with the most deviated Φ or Ψ dihedral angles from the idealized polymer. In c and d, key side chains are shown as sticks. H-bonds and salt bridges are marked as thin dashed lines. Stacking interactions are marked as thick dashed lines.

Extended Data Fig. 5. Detailed analysis of the glucan polymer.

a, Cartoon representation of the CFN-Rho1-long-polymer-bound state. The polymer and CFN are shown as sticks. b, Left: model of the glucan polymer of 17 Glc units. Right: energy-minimized glucan polymer of 17 Glc units. c, Cartoon representation of the cellulose-bound cellulose synthase, RsBcsA (PDB: 5EJZ). d, Left: The model of the cellulose polymer segment of 11 Glc units in the BcsA channel. Right: energy-minimized cellulose polymer of 11 Glc units. e, Electrostatic surface representation of the glucan translocation channel determined by the APBS plugin in PyMOL (Schrödinger). The glucan polymer is shown as sticks. The protein interior is colored black. f, Spot assays analyzing enzyme activity (FK506) of S. cerevisiae strains with the indicated FKS1 mutations (n = 3 independent biological repeats with similar results). g, Detailed interactions between ScFks1 and Glc_1-3 in the short-polymer-bound state. Key side chains are shown as sticks. The water molecules are shown as red spheres. H-bonds, salt bridges, or stacking interactions are marked as black dashed lines. Red dashed lines highlight the long distance between UDP and its interacting side chains in the CFN-Rho1-long-glucan-bound state. h, Surface representation of the glucan translocation channel and gate in the ground, short-glucan-bound, and CFN-Rho1-long-glucan-bound states. The protein interior is colored black. The glucan polymer is shown as sticks. The position of the channel gate in the no-polymer or short-polymer bound states is marked by black arrows.

The active site connects to the channel through IF1–IF3. We designate the glucan unit at the non-reducing end as Glc₊₀, as its proximity to the β-phosphate of UDP indicates that it was just transferred from the donor (Fig. 2c). Glc_0–9 occupy the ScFks1 translocation channel, whereas Glc_10–16 are located outside ScFks1. IF1–IF3 have a critical role in stabilizing the donor substrate and Glc_0–4. Glc₊₀ is stabilized by H-bonds with D1102, H1195, K1082 and R1248 (Fig. 2c). D1102—part of the conserved DAN motif²⁷—mediates a polar interaction network involving UDP, Glc₊₀, K1082 and K1246, probably stabilizing the donor during catalysis. K1261^IF2 (K1261 from IF2) coordinates the phosphate of UDP. E1173^IF1 forms a salt bridge with R1248, which coordinates Glc₊₀ and β-phosphate of UDP (Fig. 2c). F1176^IF1 provides stacking interaction with Glc₊₂ (Fig. 2c). F1180^IF1 stabilizes Glc₊₃ with the stacking interaction while positioning the 4-OH of Glc₊₄ to form an H-bond with T1175^IF1. K1172^IF1 and E1402^IF3 form a stabilizing salt bridge between IF1 and IF3 (Fig. 2d). These interactions, along with the stacking interaction between F1297^TM7 and Glc₊₃, stabilize the first helical turn of the glucan polymer and promote unusual dihedral angles in the glycosidic linkage between Glc₊₃ and Glc₊₄ (Supplementary Fig. 5). The middle segment of the glucan polymer, comprising Glc_5–9, is also located in the glucan translocation channel. W1474^TM10, Y1358^TM8 and S1361^TM8 provide H-bonds with Glc₊₆, Glc₊₇ and Glc₊₉, respectively, guiding the helical turn of the polymer (Fig. 2d). The channel interior is highly electronegative and probably filled with water (Extended Data Fig. 5e). We performed yeast spot assays to assess the ability of mutant ScFks1 to complement the growth defect of the fks1Δ strain in the presence of FK506, which suppresses Fks2 expression^3,27–29. D1102A, N1104A, K1082A, K1172A, H1195A, K1261A, F1297A and W1474A reduced yeast growth, whereas N1257A/W^IF2, F1258A^IF2 and E1221A/K, and D1222A/K from the catalytic NED motif, abolished growth (Extended Data Figs. 2c and 5f).

Our short-glucan-bound structure resolved Glc₁₋₃ (Extended Data Figs. 4b and 5g). Their interaction with Fks1 is similar to that in the CFN–Rho1–long-glucan-bound state. Although a putative short glucan was modelled in a ScFks1 structure (PDB 7YUY) reported previously, superimposition with our long-glucan-bound state reveals that it occupies a position that would directly clash with IF2 and IF3 in our catalytically relevant conformations (Extended Data Fig. 1f,g). This suggests that the conformation in 7YUY is probably incompatible with functional, polymer-generating conformations of ScFks1.

The CFN–Fks1–glucan ternary complex

The robust CFN density, located near Glc_10–14 (Fig. 3a), allowed us to model the molecule unambiguously (Fig. 3b). The binding mode of CFN is unique: the nascent glucan mediates the formation of a ternary complex with CFN and Fks1. The macrocyclic ring of CFN grips both the polymer and ScFks1, whereas its lipid tail anchors the drug to the outer leaflet of the membrane (Fig. 3c). Positioned at the protein–lipid interface through its interaction with Fks1, CFN clamps the nascent glucan product together with Fks1, thereby halting its extrusion (Fig. 3c).

Fig. 3. The CFN binding site.

a, Cartoon representation of ScFks1 in the CFN–Rho1–long-glucan-bound state. The glucan polymer and CFN are modeled as space-filling models. b, The EM density (contoured at 5σ) and the model of CFN are shown as sticks. c, Detailed interaction between CFN, the glucan polymer and ScFks1. Key side chains are shown as sticks. H-bonds are marked as thin dashed lines. Cation–π and stacking interactions are marked as thick dashed lines. Red stars highlight glycosidic linkage with the most deviated Φ dihedral angles from the idealized polymer. d, Spot assays showing Fks1 activity (FK506) and CFN-resistant phenotype (FK506+CFN) of S. cerevisiae strains with the indicated Fks1 mutations at different CFN concentrations (n = 3 independent biological replicates with similar results). e, Time series traces of root mean squared deviation values in a 1-µs molecular dynamics simulation for Glc_0–10 inside Fks1 and Glc_11–16 outside Fks1 in systems in the absence or presence of CFN (n = 5 independent replicates). f, Frequency distribution of root mean squared deviation (RMSD) values over all five replicate molecular dynamics simulations for Glc_0–10 (left) and Glc_11–16 (right) in systems in the absence (orange) or presence (green) of CFN. WT, wild type.

The backbone carbonyls of the macrocycle, along with the 3-hydroxyproline, form H-bonds with Glc₊₁₁ to Glc₊₁₄, whereas the side chains of the macrocycle interact with either Fks1 or the lipid membrane. The homotyrosine moiety forms H-bonds with D646^TM5 and K1336 (TM7–TM8 loop), and cation–π interaction with R645^TM5. The 4-hydroxyproline also forms an H-bond with R1332 (TM7–TM8 loop). The terminal amine of the ethylenediamine side chain and the δ-amino group of 3-hydroxyornithine face the solvent. In the membrane environment, they may interact with lipid head groups, which may explain why these two primary amines are modifiable³⁰ yet important for antifungal activity³¹.

The three clinically identified echinocandin-resistant hotspots are located on TM5, TM6 and TM8, respectively, surrounding the glucan and CFN. Notably, many hotspot residues do not interact directly with CFN but instead with the glucan polymer (Fig. 3c). F639^HS1 and I1364 interact with Glc₊₁₁ and Glc₊₁₂ through stacking and van der Waals interactions, respectively. R1357^HS2 forms H-bonds with Glc₊₁₃ and S643^HS1. L694^HS3 and W695^HS3 sandwich Glc₊₉. Furthermore, N1333 from an extracellular loop between TM7 and TM8 (TM7–TM8 loop) also forms an H-bond with Glc₊₁₆. This loop undergoes substantial conformational rearrangement from the ground state to the CFN–Rho1–long-glucan-bound state (Supplementary Fig. 6), suggesting its potential role in coordinating the glucan polymer and CFN. Disruption of a conserved disulfide-bridge (C1328–C1345) at the junction between this loop and TM7 and TM8 was shown to abolish ScFks1 activity⁶. However, the CFN-interacting residues R1332 and K1336 in the loop are less conserved than the hotspot residues across some pathogenic fungi (Supplementary Fig. 2).

In our yeast spot assays, the HS1 or HS3 mutants F639I/S, L642F/W, S643P/R/D/Y, R645A, D646A and I697A/F conferred CFN resistance (Fig. 3d). Although the HS2 mutants R1357H/G/S all reduced growth, R1357H conferred resistance. L1360F and V1365F near HS2 also showed reduced susceptibility (Fig. 3d), and the I1364T equivalent in Candida parapsilosis has been identified recently in a clinical isolate³². E635A/K, Y638A, D691A, Y693A, W695L and S1361L/F all reduced growth. K1336A in the TM7–TM8 loop conferred resistance, whereas Y1330A, R1332A and N1333A showed growth and CFN susceptibility similar to that of the wild type.

The extrusion of the glucan polymer from the exit in the middle of the membrane raised the question of how the hydrophobic membrane accommodates the polar glucan. We conducted all-atom molecular dynamics simulations of our ScFks1 structures in a membrane environment. Our data show that ScFks1 induces thinning of the outer leaflet of the membrane by 4–6 Å around the glucan exit, even when it is closed (Extended Data Fig. 6). This allows water to penetrate the outer leaflet and form a solvent layer between the glucan exit and the membrane (Extended Data Fig. 6). Therefore, the nascent glucan polymer remains hydrated past the exit. We observed higher flexibility of Glc_11–16 outside the exit than Glc_0–10 inside ScFks1 during the molecular dynamics simulation (Fig. 3e,f). Notably, the presence of CFN reduced the flexibility of Glc_11–16 (Fig. 3e,f), supporting the role of CFN in stalling glucan translocation.

Extended Data Fig. 6. Molecular dynamics simulation analysis of membrane deformation and solvent accessibility around the gate region of ScFks1 for five different systems.

System 1 is the cryo-EM structure of the CFN-Rho1-long-glucan-bound state (a). System 2 removes CFN from System 1 (b). System 3 removes the glucan polymer from System 1 (c). System 4 removes both CFN and Glc_4-16 from System 1 (d). System 5 is the cryo-EM structure of the short-glucan-bound state in which the channel gate is closed (e). In each system, the left panel displays the P atom of the phospholipid (orange spheres) to highlight the membrane curvature in a representative simulation frame. The middle two panels show the two-dimensional Z distributions of the lipid P atoms in the outer and inner leaflet, respectively. The plots were constructed with a grid spacing of 4.5 Å, determined by the closest P-P distance (6.35 Å) based on a radial distribution function, to ensure that each grid box covers one P atom. The average positions of CFN and glucan polymer are shown as red and grey dots, respectively. White arrows mark the area with the most membrane deformation. The right panel highlights the water molecules (red spheres) in the membrane bilayer within 10 Å of the glucan polymer and CFN in a representative frame of the simulation. In systems where either CFN or the glucan polymer is absent, the simulated structure was aligned to the cryo-EM structure first and the coordinates of CFN and glucan polymer from the EM structure were used to determine the range of water molecules for display. In a–e, each system was repeated five times (n = 5 independent repeats).

The roles of hotspot residues

We analysed the structural roles of hotspot residues in GS function, glucan binding and CFN binding. The glucan exit path is lined by the common hotspot residues: F639^HS1, S643^HS1 and R1357^HS2 form the ceiling, whereas W695^HS3 and L694^HS3 form the floor. Additional residues E635^TM5, S1361^TM8, V1365^TM8 and I1368^TM8 complete the exit path structure. Comparing the ground state with the CFN–Rho1–glucan-bound state reveals that the opening at the exit is expanded substantially in the latter—reminiscent of channel gating (we refer to the glucan exit as the gate hereafter)—enabling nascent glucan transport (Fig. 4a–c). All hotspot residues participate directly in forming this path (Fig. 4c). Gate opening is driven through side-chain rearrangements of F639^HS1 and R1357^HS2, a backbone rotation at S643^HS1 and a concurrent rigid-body motion of TM8 (containing HS2) with TM7–TM8 loop rearrangement (Fig. 4d and Supplementary Fig. 6). Together, these motions connect the translocation channel to the active site for the processive glucan polymerization and extrusion (Extended Data Fig. 5h). The exiting glucan is stabilized by R1357^HS2, S1361^HS2 and F639^HS1 by H-bonds and stacking interactions (Fig. 4d). The extruded glucan and surrounding hotspot residues create a groove that accommodates CFN (Fig. 4c,d), which may explain why this ternary complex does not require extensive direct Fks–CFN interaction. Thus, F639^HS1 and R1357^HS2 serve as the gate and S643^HS1 facilitates the conformational changes for gate opening, glucan binding and CFN binding.

Fig. 4. Roles of hotspot residues in channel gate opening, glucan binding and CFN binding.

a–c, Surface representation of the channel exit gate in the catalytically relevant ScFks1 structures in ground (a), short-glucan-bound (b) and CFN–Rho1–long-glucan-bound (c) state. Side chains of gate-forming residues are shown as sticks. Residues from the echinocandin-resistant hotspots (HS1–3) are coloured pink, purple and orange, respectively. In the left part of c, glucan and CFN were removed for clarity. d, Detailed conformational changes of the gate opening and resulting interactions with glucan and CFN. Left, gate opening. Superimposition of the ground (grey) and the CFN–Rho1–long-polymer-bound (wheat) states of ScFks1 at the gate. Red arrows mark conformational rearrangements of TM8 and side chains of gate-forming hotspot residues leading to channel opening. Middle, interactions of nascent glucan and ScFks1. H-bonds are marked with thin dashed lines. Cation–π interactions or stacking interactions are marked with thick dashed lines. Right, CFN binding to the glucan-bound ScFks1 and superimposition of the ground and the CFN–Rho1–long-polymer-bound states of ScFks1. The red arrow marks the conformational rearrangement of TM8 leading to channel opening. The Cα atoms of F1366 of the two states are shown as spheres, and the distances between them are shown. e, Surface representation of the closed gate in the S643P structure. f,g, Superimposition of the wild-type ground state (grey), the wild-type CFN–Rho1–long-polymer-bound state (wheat) and the S643P structures (blue). S/P643 is shown in ball-and-stick representation (f). The black arrow marks the direction of motion for the Cα atom of S643 from the ground state to the CFN–Rho1–long-glucan-bound state. Interactions and clashes around S/P643 viewed from the extracellular side (g). H-bonds are marked with red dashed lines. Distances between key atoms are shown. Clashes between indicated atoms are marked with clusters of short red lines.

To better understand the resistance mechanism, we characterized the S643P mutant, equivalent to the clinical HS1 resistant mutation found in Candida albicans³³. The purified S643P ScFks1 was resistant to CFN up to 200 μg ml⁻¹ (Extended Data Fig. 2e). This observation is consistent with those in C. albicans³⁴ and distinct from a previous study using the UDP-Glo assay⁵, which reported partial inhibition of S643P, underscoring the importance of direct glucan quantification. Our cryo-EM structure of S643P revealed only a gate-closed, L1-like conformation lacking CFN, glucan or Rho1 (Extended Data Figs. 1d, 3 and 4e and Extended Data Table 1), which needs to transit into the catalytically relevant conformations for polymerization. By comparing this structure with wild-type structures, we infer the roles of S643 in enzyme activity and inhibition by CFN (Fig. 4e,f). First, the S643^HS1 side chain stabilizes the up conformation of R1357^HS2 through an H-bond (Figs. 3c and 4g), which coordinates glucan for activity (Fig. 3d). Second, the motion of the S643 backbone positions R645 and D646 for CFN interaction (Fig. 4d,f). The S643P mutation thus reduces CFN binding in three ways: (1) the altered conformation of R1357 probably affects CFN binding by disturbing the glucan–Fks1 interaction; (2) the restricted backbone angles of S643P restrain allosterically the conformational rearrangements, including those of R645 and D646, required for CFN binding (Fig. 4f) and (3) both P643 and D646 would clash physically with CFN (Fig. 4g).

Interfacial helices and latch loop rearrangements

When aligned by the transmembrane region, our ScFks1 structures show a progressive rigid-body motion of the catalytic domain towards the interfacial helices and the TMD (Fig. 5a) from the ground to the CFN–Rho1–long-glucan-bound states. This movement suggests increasingly stronger coupling between the GTD and TMD as the enzyme transitions into a conformation for processive glycosyltransfer.

Fig. 5. Rho1 binding induces conformational rearrangements of the latch loop at the active site.

a, Superimposition of the catalytically relevant states of ScFks1 by the TMD. The red arrow marks the direction of the progressive rigid-body motion of the catalytic domain towards the TMD. b, Conformations of IF1–IF3 in the catalytically relevant states of ScFks1. The polymer is shown as spheres. The π-bulge in IF1 of the CFN–Rho1–long-glucan-bound state is highlighted by a dotted black circle. c, Conformations of the latch loop at the active site of the catalytically relevant ScFks1 structures. d, Cartoon representation of ScFks1 in the CFN–Rho1–long-glucan-bound state. Wheat, ScRho1; pink, latch loop; blue, coupling module; black, peripheral loop; olive, Rho1-binding helix. e, The relative positions of coupling module, latch loop, peripheral loop and Pre-S1 and Pre-S2 loops at the active site in the three catalytically relevant Fks1 structures. In the short-glucan-bound or the CFN–Rho1–long-glucan-bound state, the black arrow marks the motion of the peripheral loop; the blue arrow the coupling module and the cyan arrow the catalytic loop. In the CFN–Rho1–long-glucan-bound state, the orange, purple and pink arrows mark the motion of the peripheral loop, the Pre-IF1 loop, the Pre-IF2 loop and the latch loop, respectively. f, Cartoon scheme of the mechanism of Rho1-activated glucan synthesis and transport by Fks1 and the mechanism of non-competitive inhibition by CFN.

Unlike IF2 and IF3, which remain almost static, IF1 undergoes notable conformational changes during the reaction. In the ground state, the middle of IF1 is broken into a loop (residues A1169–G1177), probably due to conserved G1171 and G1177 residues (Fig. 5b and Supplementary Fig. 2a). When the short polymer is present in the channel, IF1 becomes nearly α-helical. In the long-glucan-bound state, IF1 forms a π-bulge (T1175–F1180) that coordinates Glc_2–4 through CH–π and H-bond interactions (Figs. 2c and 5b). Similar α- to π-helical transitions are known to have important roles in ion channel function^35–37. Superimposing the ground state IF1 onto the glucan-bound state reveals a steric clash with the glucan polymer (Extended Data Fig. 7a). Therefore, IF1 adopts state-dependent conformations that provide a regulatory mechanism for Fks1 activity. It is unclear whether these conformational changes are glucan length- or Rho1-dependent.

Extended Data Fig. 7. Active site of catalytically relevant ScFks1 and other processive glycosyltransferases.

a, Superimposition of the interfacial helixes in the ground and the short-glucan-bound states of ScFks1 by the TMD. b, Spot assays analyzing enzyme activity (FK506) of S. cerevisiae strains with indicated FKS1 mutations (n = 3 independent biological repeats with similar results). c, Superimposition of the active site of the catalytically relevant ScFks1 structures. Conformations of the latch loop, the catalytic loop and the GTD are marked with red arrows. d, Surface representation of the active site in the three catalytically relevant ScFks1 structures showing the solvent accessibility of UDP in the up and down conformations of the latch loop. e–g, Cartoon representation of the active site of UDP-bound ScFks1 in the CFN-Rho1-long-glucan-bound state (e), RsBcsA (PDB: 5EJZ) (f) and XlHAS1 (PDB: 8SMP) (g). The latch loop in ScFks1 and the gating loops in RsBcsA and XlHAS1 are in the down conformation. Side chains interacting with the UDP are shown as sticks. The Mg²⁺ ions in RsBcsA and XlHAS1 are shown as green spheres. In a-g, The TM region is colored grey; the GTD blue-white; the interfacial helices green; the latch loop in ScFks1 or the gating loop RsBcsA and XlHAS1 pink.

Our EM map resolves a loop (residues 1420–1433) linking IF3 and TM9 at the interfacial region, which was not observed in previous ScFks1 structures^5,6. Best resolved in the CFN–Rho1–long-glucan-bound state (Extended Data Fig. 4a–c), this loop inserts below IF2 into the active site, where conserved residues Y1424 and R1429 coordinate the β-phosphate of UDP (Fig. 5c and Supplementary Fig. 2). In this ‘down’ conformation, this loop probably stabilizes the donor substrate for catalysis. Y1424A, R1429A and R1429K all abolished growth (Extended Data Figs. 2d and 7b).

In the ground and short-glucan-bound states, this loop flips upwards, losing its interaction with the UDP (Fig. 5c and Extended Data Fig. 7c) and becoming flexible (Extended Data Fig. 4a,b). Surface representations of the active site show that the down conformation of the loop seals the active site and traps the UDP, whereas the up conformation leaves it solvent exposed (Extended Data Fig. 7d). We term this loop the ‘latch loop’ to reflect its dynamic motion and its potential roles in the catalysis as described in the next section.

Active site differences of GT-48 and GT-2

The active sites of ScFks1 show notable differences from those of the GT-2 family enzymes (Extended Data Fig. 7e). First, GT-2 enzymes feature a conserved QRXRW motif in IF2, where an arginine coordinates UDP and an ensuing tryptophan binds the acceptor sugar³⁸. By contrast, ScFks1 uses K1261^IF2 to coordinate UDP but lacks an aromatic amino acid in the acceptor-binding site. Instead, Glc₊₁ is stabilized by several H-bonding interactions, and Glc₊₂ and Glc₊₃ form stacking interactions with F1176^IF1 and F1297^TM7, respectively (Fig. 2c). These extensive interactions may explain the high processivity of GS¹⁸. Second, although GT-2 enzymes require divalent metals for the catalytic function, GS is metal-independent^5,18,19. In the long-glucan-bound structure with the latch loop down, K1246 and R1248 act as surrogates for Mg²⁺ by occupying positions similar to Mg²⁺ in Rhodobacter sphaeroides BcsA (RsBcsA) or Xenopus laevis HAS1 (XlHAS1) and interact with the β-phosphate of UDP^9,11,12 (Extended Data Fig. 7e–g). R1248A abolished growth (Extended Data Figs. 2d and 5f). In the short-glucan-bound structure with the latch loop up, K1246 and R1248 are more than 5 Å away from the β-phosphate (Extended Data Fig. 5g), reinforcing the idea that the down conformation of the latch loop is required for stable donor coordination and catalytic activity.

Rho1 binding site

The CFN–long-glucan-bound reconstruction resolves the EM density of GTPγS-bound ScRho1, beneath TM15 and near the catalytic domain (Extended Data Figs. 3c,d and 4c). The Fks1–Rho1 interface consists of complementary hydrophobic and polar patches (Extended Data Fig. 8a). Three structural elements on Fks1 interact directly with Rho1. First, the Rho1-binding helix (residues 914–926), which is not resolved in other ScFks1 conformations, becomes structured (Fig. 5d and Extended Data Fig. 8b) to interact with the switch I loop of Rho1 (Extended Data Fig. 8c). Second, a peripheral loop (residues 1773–1787) on the membrane surface allows Rho1 to dock under TM15 (Fig. 5d and Extended Data Fig. 8d). Third, a helix–loop–helix ‘coupling module’ (residues 858–880), forms a salt bridge network with the Rho1 switch II loop (Fig. 5d and Extended Data Fig. 8e). Comparison with a GDP-bound Aspergillus Rho1 structure (Extended Data Fig. 8f) suggest that only GTP-bound Rho1 adopts the specific switch I and switch II loop conformations to bind to Fks1, consistent with previous studies^39,40. Mutations R858A, R862A and R869A from the coupling module and R1781A and I1784W from the peripheral loop all reduced growth (Extended Data Figs. 2d and 8g). In addition, we characterized the affinity of the Rho1–Fks1 interaction by determining GS activity in the presence of different concentrations of wild-type or mutant Rho1 (Extended Data Fig. 8h). Wild-type Rho1-activated ScFks1 with a half maximal effective concentration (EC₅₀) value of 0.17 ± 0.06 μM, which is comparable to the reported K_d values between Rho GTPases and their target proteins⁴¹. Rho1 variants with mutations at the interaction interface (E69A, D70A and R73A) all exhibited twofold to fourfold higher EC₅₀ values, indicating reduced affinity to Fks1 (Supplementary Table 2).

Extended Data Fig. 8. Detailed interaction at the ScFks1-ScRho1 interface.

a, Electrostatic surface representation of the ScFks1-ScRho1 interface determined by the APBS plugin in PyMOL (Schrödinger). b, Cartoon representation of ScFks1 in the CFN-Rho1-long-glucan-bound state. Rho1 is colored wheat; the latch loop pink; the coupling module blue; the peripheral loop black; the Rho1-binding-helix olive. c–e, Detailed interactions between Rho1 and the Rho1-binding-helix (c,d), the peripheral loop (d,e), and the coupling module (e). Key residues are shown as sticks. H-bonds, salt bridges, or stacking interactions are marked as black dashed lines. f, Overlay of GDP-bound Aspergillus Rho1 structure (PDB:5ZVP) onto ScRho1 bound to ScFks1. g, Spot assays assessing the catalytic functions of ScFks1 variants (n = 3 independent biological repeats with similar results). The variants with mutations in the coupling module are colored blue, and those in the peripheral loop are grey. h, Activation of ScFks1 by ScRho1 variants. The right panel is the zoom-in view of the left panel at a lower concentration range of ScRho1. Data points represent mean ± s.d. of n = 3 technical replicates. The solid line represents fit to the 4-parameter logistic equation, from which the EC₅₀, V_min, V_max and Hill coefficient were determined and summarized in Supplementary Table 2. i. Superimposition of the CFN-Rho1-long-glucan-bound structure with the crosslinked Fks1-Rho1 structure. The Fks1 and Rho1 in the crosslinked structure are colored dark cyan and brown, respectively. The positions of Rho1 in the two structures are marked by dashed circles.

Recently, a complex structure of ScFks1 and a constitutively active ScRho1 variant (Q68H), obtained by chemical crosslinking in GDN, was reported⁷, indicating a Rho1 binding site and mode distinct from ours (Extended Data Fig. 8i). However, the EM density attributed to Rho1 in the study is too ambiguous for reliable modelling (Extended Data Fig. 4e). Although the study suggested R862 and R869 as part of the Fks1–Rho1 interface, mutations of other interface residues proposed in their study, such as H1518, fail to impair growth in our spot assays (Extended Data Fig. 8g).

Rho1 induces active site rearrangement

The coupling module has a central role in transmitting conformational changes from the Rho1 binding interface to the Fks1 active site. In the ground state, the peripheral loop is sandwiched between the coupling module and the TM region, positioned close to the latch loop (Fig. 5e). R1778^PL inserts between the coupling module and the latch loop and interacts with Y875^CM. This interaction keeps the GTD relatively distant from the transmembrane region. In the short-polymer-bound state, the loop portion of the coupling module flips upwards, whereas the peripheral loop shifts laterally, allowing the GTD to move closer to the transmembrane region. In the CFN–Rho1–long-glucan-bound state, Rho1 binds to the coupling module and pushes it further upwards, whereas the peripheral loop moves further sideways, causing the latch loop to flip down, sandwiching the coupling module between the peripheral loop and the latch loop (Fig. 5e). The concerted conformational rearrangement propagates to the loops preceding IF1 and IF2 (Pre-IF1 and Pre-IF2), which partially unwind and shift backwards (Fig. 5e), moving K1246^Pre-IF2 and R1248^Pre-IF2—the proposed divalent metal surrogates—to engage the donor substrate. Through these rearrangements, the GTD moves even closer to the TM domain and positions the donor substrate optimally for catalysis.

Discussion

Under catalytic conditions, we captured three distinct states of wild-type ScFks1 representing key intermediate states in glucan synthesis. As UDP, not UDP-Glc, is resolved in the active site, the ground state probably corresponds to a state where Fks1 has released the polymer and resets for the next cycle (Fig. 5f and Supplementary Text). In the short-glucan-bound state, the channel exit gate remains closed. The opening of the gate through TM8 movement may occur when the nascent glucan reaches the exit gate or when Rho1 binds to Fks1 (Fig. 5f). In the long-glucan-bound state, the glucan polymer extruding from the channel was captured, providing insights into the processive glucan synthesis by GS. GS is known to be highly processive; it cannot incorporate mono- or oligosaccharides into the long glucan product, neither does it produce significant amounts of short oligosaccharides¹⁸. The extensive interactions through Glc_0–4 are consistent with the high processivity, as such interactions will prevent the premature release of short oligosaccharides in the initial stage of polymerization. By contrast, the weaker interactions between Glc_5–9 and the channel are consistent with its role as a low-friction exit path for the elongating glucan chain.

Another distinct feature of Fks1 is that the exit gate is located within the outer leaflet of the membrane. This reflects the helical nature of the β-1,3-d-glucan, which was overlooked in the glucan exit path proposed previously^5,6. Because cellulose (Extended Data Fig. 5c,d) and hyaluronan favour elongated non-helical conformations, the translocation channels in their biosynthetic enzymes, BcsA^9,11 and HAS¹², respectively, are vertical in the membrane with the channel exits at the extracellular surface of the enzymes. Fks1 evolved with a glucan transport channel that optimizes the translocation of its helical glucan product energetically for efficient, highly processive polymerization and extrusion.

The latch loop in ScFks1 is structurally analogous to the gating loop in the processive GT-2 family, including cellulose and hyaluronan synthases^9–12 (Extended Data Fig. 7b–d). The ‘gating loop’ was named after its role as a gatekeeper to the active site, which regulates the passage of the polymer into the channel. We instead use the term ‘latch loop’ in the GT-48 family Fks1 to avoid possible confusion that this loop controls channel gate opening directly. Both loops share a conserved motif: an aromatic residue and a positively charged residue that are five residues apart for UDP coordination. A different scenario has been reported for the dimeric chitin synthase, where the loop is part of the domain-swapped region that reaches towards the neighbouring protomer^15,16. There is currently no structural evidence to support that the domain-swapped region can insert into the active site and coordinate the substrate.

CFN binds specifically to the interface between ScFks1 and the nascent glucan product in the outer leaflet of the membrane. The ternary complex formation traps the glucan polymer like a molecular clamp, preventing polymer elongation, and thus inhibiting ScFks1 non-competitively (Fig. 5f). This unique mode of action targets specifically the catalytically active GS complex, and inhibition can occur at any moment after the nascent glucan polymer reaches approximately 17 Glc units, substantially slowing extrusion without inducing its premature release. This observation explains why GS retains residual activity even at saturating echinocandin concentrations^5,18,33 (Fig. 1c), and why CFN does not affect product length¹⁸.

The glucan product is central to CFN binding. Although all three hotspots contribute to gate opening and proper glucan conformation for CFN binding, only HS1 contacts CFN directly. This discovery overturns the previously proposed direct-binding model between echinocandin and Fks1 hotspots^5,27. S643P confers resistance to CFN through both allosteric effects on the glucan extrusion path and the CFN-binding groove, and direct steric hindrance against CFN binding. It is conceivable that HS2 and HS3 mutations may also alter gate opening and glucan-bound conformations. Precise roles of each HS mutation require further structural and mechanistic investigation.

Given the structural similarities of the echinocandins, they probably share a common mechanism of action. Our analysis reveals distinct functional roles for the structural elements conserved across this class of antifungals: the carbonyl groups of the macrocycle mediate interactions with the glucan polymer; the homotyrosine side chain engages Fks1 directly; and the lipid tail and the positively charged side chains of the macrocycle associate with the lipid membrane. Leveraging these structural insights will aid the development of next-generation echinocandins with increased potency, broader antifungal spectrum and reduced propensity for resistance. Furthermore, our high-resolution, catalytically active Fks1 structure provides a foundation for developing competitive inhibitors targeting the active site or inhibitors targeting the Fks1–Rho1 interface, opening previously unexplored avenues for antifungal drug discovery.

Methods

ScFks1 expression and purification

ScFks1 was expressed and purified from S. cerevisiae BY4741 fks2 deletion strain¹⁸ harbouring the genomic copy of the fks1 gene engineered to have a C-terminal HRV3C protease site, ten amino acid linker, 3× Flag tag, and 10× His-tag. The fks1 gene was modified by homologous recombination by following a standard procedure⁴². To express ScFks1, yeast cells were grown in YPD medium and then collected by centrifugation. The resulting cells were used to prepare membrane fractions as described previously¹⁸. To purify ScFks1, the membrane fractions were passed through a French Press cell twice at 18,000 psi. Then, buffer A (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol) with 1.5% N-dodecyl β-d-maltoside (DDM) and 0.15% CHS (w/v) was added slowly while stirring then stirred overnight at 4 °C. The resulting mixture was clarified by centrifugation at 15,000g for 30 min at 4 °C, and the supernatant was then treated with anti-Flag affinity resin (Pierce). The slurry was mixed for 1 h, and the resin was then packed in a column and washed with 12.5 column volumes of buffer A with 1.5% DDM and 0.15% CHS (w/v). The resin was then washed with another 12.5 column volumes of buffer A with 0.05% LMNG and 0.005% CHS (w/v) or 0.1% CHAPS and 0.01% CHS (w/v) buffer. Bound proteins were eluted by washing with buffer containing 200 µg ml⁻¹ Flag peptide. Eluates were concentrated to <0.5 ml, and purified with size-exclusion chromatography (Superose 6 Increase 10/300 GL column, GE Healthcare) equilibrated in buffer B (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA) with 0.005% LMNG and 0.0005% CHS (w/v) or 0.1% CHAPS and 0.01% CHS (w/v). The peak fractions were pooled and concentrated for cryo-EM sample preparation. For the structural and functional characterization of S643P-Fks1, the S643P mutation was introduced to the genomic copy of fks1 described above using the CRISPR–Cas9 genome editing using a reported protocol⁴³. The mutant was expressed and purified following the same protocol as for wild-type ScFks1.

ScRho1 expression and purification

S. cerevisiae Rho1 (ScRho1) with N-terminal 6×His-tag and tobacco etch virus protease site was expressed in Escherichia coli Rosetta 2 (DE3) cells (Novagen) harbouring the pET28a plasmid with the rho1 gene codon optimized for expression in E. coli. The E. coli cells were grown in Luria Bertani medium with chloramphenicol (25 mg l⁻¹) and kanamycin (50 mg l⁻¹) at 37 °C to an optical density at 600 nm (OD₆₀₀) of around 1.0. Protein expression was induced by adding 0.5 mM isopropyl β-d-1-thiogalactopyranoside, and the cultures were continued for 20 h at 15 °C. Cells were then collected by centrifugation and resuspended in buffer C (50 mM Tris, pH 7.5, 200 mM NaCl, 10% glycerol) containing 2 mM β-mercaptoethanol (BME) and lysed by passing them through a French Press cell operating at 18,000 psi. The lysate was then clarified by centrifugation at 25,000g for 25 min at 4 °C. The supernatant was then applied to a column packed with cobalt-iminodiacetate resin (Prometheus), the column was washed with 12.5 column volumes of buffer C with 5 mM imidazole, and Rho1 was eluted with 5 column volumes of buffer C with 250 mM imidazole. After adding 2 mM BME, the eluate was concentrated and buffer-exchanged using PD10 columns (GE Healthcare) equilibrated with 50 mM Tris, 200 mM NaCl, 2 mM BME and 30% glycerol. Protein concentrations were determined using the Bradford method, with bovine serum albumin as a concentration standard. For cryo-EM sample preparation, Rho1 was further purified by size-exclusion chromatography (Superdex 200 Increase, Cytiva) equilibrated in 20 mM Tris, 150 mM NaCl, 2 mM BME. The peak fractions were pooled and concentrated for cryo-EM sample preparation.

ScFks1 activity assays

In vitro radioactive GS assays were performed as reported previously¹⁸. To test the effects of detergents on GS activity, ScFks1 was mixed with four times the final reaction concentration of detergents. The mixture was then mixed with an equal volume of 125 µM ScRho1 and incubated on ice for 1 h. Reactions were initiated by mixing with equal volumes of the ScFks1–Rho1 solution and a UDP-Glc (4 mM) solution as described previously¹⁸.

Cryo-EM sample preparation

All cryo-EM samples in this study were prepared on freshly glow-discharged UltrAuFoil R1.2/1.3 300 mesh grids (Quantifoil), using a Leica EM GP2 plunge freezer with the chamber set at 30 °C and 85% humidity. For the ScFks1 sample in LMNG/CHS, peak fractions from the size-exclusion chromatography in LMNG/CHS were concentrated to approximately 10 mg ml⁻¹ and doped with 0.1% CHAPS/0.02% CHS. Then, 5 mg ml⁻¹ CFN dissolved in water with 10% dimethylsulfoxide was mixed with ScFks1 at 1:20 (v/v) and incubated for 2 min. A 3-μl sample of ScFks1 was applied to the grid, incubated in the chamber for around 60 s then blotted for 1.5–2 s followed by plunge-freezing in liquid ethane cooled by liquid nitrogen. For the ScFks1 sample in CHAPS/CHS, peak fractions from the size-exclusion chromatography in CHAPS/CHS were concentrated to around 12–15 mg ml⁻¹. Purified Rho1 was mixed with ScFks1 at a 1.5:1 molar ratio, and GTPγS was added at a 1.5-fold molar concentration of Rho1. 1.5 mM EDTA, 50 mM KF and 7.5% glycerol were further included in the reaction mixture. The sample was incubated with 12.5 mM UDP-Glc and 0.5 mg ml⁻¹ CFN on ice for 30 min. During grid freezing, 3 μl of ScFks1 was applied to the grid, incubated in the chamber for approximately 60 s and blotted for 1.5–2 s followed by plunge-freezing in liquid ethane cooled by liquid nitrogen. The S643P sample was prepared similarly, with a longer reaction time (2 min).

Cryo-EM data collection

The LMNG/CHS doped with CHAPS/CHS sample was collected with a Titan Krios microscope (Thermo Fisher) operating at 300 kV and equipped with a K3 detector (Gatan) and a GIF energy filter with slit width of 20 eV in super-resolution mode using SerialEM (v.4.2)⁴⁴. Videos were collected at a nominal magnification of ×105,000 with a super-resolution pixel size of 0.4135 Å per pixel at the specimen level. Each video contained 50 frames over a 2.9-s exposure time, using a dose rate of around 17.2 e⁻ pixel⁻¹ s⁻¹, resulting in the total accumulated dose of approximately 50 e⁻ Å⁻². The nominal defocus range was set from −0.8 to −2.2 μm.

The wild-type CHAPS/CHS sample was collected with a Titan Krios microscope (Thermo Fisher) operating at 300 kV and equipped with a K3 detector (Gatan) in counting mode behind a BioQuantum GIF energy filter with a slit width of 20 eV, using the Latitude-S (v.3.3) automated data-acquisition program. Videos were collected at a nominal magnification of ×105,000 with a pixel size of 0.840 Å per pixel at the specimen level. Each video contained 40 frames over a 2.4-s exposure time, using a dose rate of around 25 e⁻ pixel⁻¹ s⁻¹, resulting in the total accumulated dose of around 60 e⁻ Å⁻². The nominal defocus range was set from −0.8 to −2.2 μm.

The S643P sample was collected with a Titan Krios microscope (Thermo Fisher) operating at 300 kV and equipped with a K3 detector (Gatan) in counting mode behind a BioQuantum GIF energy filter with slit width of 20 eV, using the EPU (v.3.8.1) program. Videos were collected at a nominal magnification of ×105,000 with a pixel size of 0.820 Å per pixel at the specimen level. Each video contained 50 frames over a 2.7-s exposure time, using a dose rate of approximately 18 e⁻ pixel⁻¹ s⁻¹, resulting in the total accumulated dose of around 50 e⁻ Å⁻². The nominal defocus range was set from −0.8 to −2.2 μm.

Cryo-EM data processing

All datasets were processed following similar procedures. Beam-induced motion correction and dose-weighting were performed using Relion’s own implementation⁴⁵. The motion-corrected micrographs were then imported into cryoSPARC (v.4.4)⁴⁶ for contrast transfer function (CTF) estimation. Micrographs were selected subsequently based on CTF fit quality and CTF estimated resolution. An initial set of approximately 5,000 particles was picked manually and subjected to a reference-free two-dimensional classification, from which the classes with clear protein features were selected as references for template-based particle picking. Picked particles were extracted with a 320-pixel box size at bin4. Iterative two-dimensional classification was performed in cryoSPARC, and classes showing clear secondary structure features of ScFks1 were selected, re-extracted at bin2 for ab initio reconstruction with two classes. For the CHAPS/CHS samples, particles from the good two-dimensional class were used to train a TOPAZ (v.0.2.5)⁴⁷ model, which was applied to pick through all the micrographs again. The particles from TOPAZ extract were subject to two-dimensional classification separately from the particles from template picking. Good particles from template picking and topaz picking were combined and duplicate particles removed before heterogeneous refinement.

Iterative heterogeneous refinement was performed with the selected reference from parallel ab initio reconstruction—one good and one junk reference—until the estimation resolution for the good class is better than 4.8 Å for wild type and 6.4 Å for S643P. Particles from the good class that shows clear protein density were re-extracted without binning for non-uniform refinement⁴⁸. For wild-type samples, after non-uniform refinement, the particles from the consensus reconstruction were transferred to RELION 4.0 (ref. ⁴⁵).

For the LMNG/CHS doped with CHAPS/CHS data, skip-align three-dimensional classification yielded classes corresponding to the L1 and L2 states. To further improve resolution, signal subtraction was performed on the L1 and L2 states separately to remove micelles, followed by two to three more rounds of skip-align three-dimensional classification. After each round, all classes were refined to evaluate resolution and conformational heterogeneity. The good classes were kept for the next round. During these rounds, only one class from the L1 or L2 state remained at high resolution, suggesting little conformational heterogeneity within each state. Finally, the best particles from each state were reverted to full particles and subject to Bayesian polishing. The polished particles were imported back to cryoSPARC and finalized with CTF refinement and non-uniform refinement.

For the wild-type CHAPS/CHS data, skip-align three-dimensional classification only yielded two good classes. One class had clear CFN density but weak Rho1 density. The other class showed no CFN or Rho1 density, corresponding to the ground state. Further subtraction of the micelle signal and the subsequent skip-align classification did not improve the map quality of either class. These two classes were subject directly to Bayesian polishing in RELION v.4.0 (ref. ⁴⁵). The polished particles were transferred back to cryoSPARC for CTF refinement. After CTF refinement, the ground state class was subject to more iterations of heterogeneous refinement to improve resolution. The CFN-containing class was first subject to three-dimensional classification in cryoSPARC with a mask excluding the micelle signal, which yielded two classes containing similar numbers of particles. One class showed clear Rho1 and CFN density and corresponds to the CFN–Rho1–long-glucan-bound state. The other class did not have CFN or Rho1 density but had signal for three sugar units in the channel. This class corresponds to the short-glucan-bound state. Both classes were further cleaned up by heterogeneous refinement and finalized with non-uniform refinement.

For the S643P sample, one round of three-dimensional classification was performed in cryoSPARC first to further clean up the particle set. Only particles from the class that converges at 3.68 Å were transferred to RELION v.5.0. Signal subtraction of the micelle and the subsequent skip-align classification yielded only one good class. The particles were reverted to full particles and subject to Bayesian polishing. The polished particles were imported back to cryoSPARC and finalized with CTF refinement and non-uniform refinement.

Model building and refinement

The ScFks1 structure at the L2 state was built de novo in Coot (v.0.98)⁴⁹. During model building, residue registration was guided by the presence of large aromatic side chains. All other wild-type ScFks1 models used the L2 structure as the template. The S643P model used the L1 state as the template. In the CFN–Rho1–long-glucan-bound state, the ScRho1 model from PDB 3A58 was first docked into the Rho1 density. For YMR295C, a polyalanine model was first built de novo into the density in the CFN–Rho1–long-glucan-bound state. After we identified this segment as YMR295C, side chains were registered based on the AlphaFold⁵⁰ model and the presence of large aromatic side chains. The entire structure was then inspected manually and refined in Coot with ideal geometry restraints. The YMR295C model was used as the template for the YMR295C density in other states. The restraints for lipids and ligands, including PE, UDP, CFN and GTPγS were generated in the Grade2 server⁵¹ and optimized in eLBOW (as implemented in Phenix v.1.20 (ref. ⁵²)) using the REEL QM2 method if necessary. Ligand restraints were inspected and adjusted manually to ensure correct stereochemistry before being fitted into the cryo-EM maps in Coot. The glucan polymer was modeled from the reducing end towards the non-reducing end with an ascending numeric order of Glc units (1–17), whereas all the subscripted Glc numbers (0–16) in the main text refer to Glc positions from the non-reducing end towards the reducing end based on distance from the active site. The MolProbity(v.4.4)⁵³ server was used to identify problematic regions in the models, which were then adjusted manually in Coot. The final refinement was performed using the phenix-real_space_refine function with global minimization and secondary structure restraints as implemented in the Phenix suite⁵². The Fourier shell correlation of the half- and full-maps against the model, calculated in Phenix, was in good agreement, indicating that the models were not over-refined. Structural analyses and illustrations were performed using PyMOL (v.3.1) (Schrödinger), University of California San Francisco (UCSF) Chimera (v.1.7) and ChimeraX (v.1.10)⁵⁴. Structure alignments and cryo-EM density map alignments were performed by Fit-In-Map in UCSF Chimera.

Estimation of the ScFks1–ScRho1 affinity by ScRho1 titration assay

CHAPS/CHS purified ScFks1 were mixed with four times the desired concentration of purified ScRho1 and a 1.1× concentration of GTPγS. This mixture was incubated for 1 h on ice, and reactions were performed as above. Additional GTPγS (20 µM) was added in the reaction mix in case low GTPγS concentrations affected activity. To determine the EC₅₀ value of ScFks1 activation by ScRho1, the observed GS activity was plotted against ScRho1 concentration and fit to the four-parameter logistic equation (GraphPad Prism v.8.0).

Yeast spot assay

ScFks1 mutants were prepared by site-directed mutagenesis using p423 GAL1 plasmid vector harbouring ScFks1-HRV3c-3xFlag-10×His gene as template. The resulting plasmids were introduced into S. cerevisiae BY4741 ΔFks1 (Horizon Discovery) using the lithium acetate/ssDNA carrier/polyethylene glycol transformation protocol⁴². Transformed cells were plated on Synthetic Defined media lacking Histidine (SD-His), and single colonies were re-streaked. Colonies were grown overnight in SD-his with 2% raffinose, and the OD₆₀₀ of each culture was adjusted to 2.0. This adjusted culture was serially diluted six times, and 5 µl of each serial dilution was plated on SD-His +2% raffinose plates with or without 1 µg ml⁻¹ FK506 and 0.24 or 0.5 µg ml⁻¹ CFN. Plates were incubated at 30 °C, and imaged after 48, 72 and 96 h using an Azure 400 Imager (Azure Biosystems).

Western blot analysis

To verify expression, selected mutants were grown in SD-His +2% raffinose at 30 °C and harvested at an OD₆₀₀ of around 3–7. Cells were lyophilized and lysed with beads-beating, after which membrane fractions were prepared as described previously¹⁸. The membrane was solubilized with 1.5% DDM/0.15% CHS overnight in 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol. Total protein content of the clarified lysates was determined using A₂₈₀. Lysates containing approximately 15 mg total protein was incubated with 30 μl Flag resin for 2 h at 4 °C. The resin was washed three times with 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 10 mM DDM/1 mM CHS and resuspended to 125 μl, 25 μl of which was loaded onto SDS–PAGE. Following SDS–PAGE (Bio-Rad), protein was transferred onto 0.45-μm polyvinylidene difluoride membranes, blocked with 5% bovine serum albumin in TBS and probed with 1,000× diluted monoclonal mouse anti-Flag M2 antibody (Sigma Aldrich), then 5,000× polyclonal IRDye 800CW donkey anti-mouse antibody (LICORbio) and detected using the Odyssey M Imaging System (LICORbio).

Liquid chromatography tandem mass spectrometry proteomics sample processing, data collection and data analysis

Samples were brought to 5% SDS, reduced for 15 min at 80 °C with 10 mM dithiothreitol and alkylated with 20 mM iodoacetamide for 30 min at room temperature. Samples were then supplemented with a final concentration of 1.2% phosphoric acid and 375 µl of S-Trap (Protifi) binding buffer (90% MeOH/100 mM TEAB). Proteins were trapped on the S-Trap micro cartridge, and digested using 20 ng µl⁻¹ sequencing grade trypsin (Promega) for 1 h at 47 °C. The cartridge was washed using 50 mM TEAB, followed by water with 0.2% formic acid, and the peptides were eluted using 50% acetonitrile in water with 0.2% formic acid. The resulting samples were then lyophilized to dryness and resuspended in 20 μl of 1% trifluoroacetic acid and 2% acetonitrile in water with 12.5 fmol µl⁻¹ of yeast alcohol dehydrogenase as an internal standard.

Liquid chromatography with tandem mass spectrometry (MS/MS) was performed using an EvoSep One ultra performance liquid chromatography coupled to a Thermo Orbitrap Astral high-resolution tandem mass spectrometer (Thermo). In brief, 2 μl of each sample was loaded onto an EvoTip and the peptides were eluted onto a 1.5-µm EvoSep 150um ID × 15 cm performance (EvoSep) column using the SPD30 gradient at 55 °C. Data collection on the Orbitrap Astral mass spectrometer was performed in a data-independent acquisition (DIA) mode of acquisition with r = 240,000 (at m/z 200) full mass spectrometry scan from m/z 380–980 in the Orbitrap with a target automatic gain control (AGC) value of 4 × 10⁵ ions. Fixed DIA windows of 4 m/z from m/z 380 to 980 DIA MS/MS scans were acquired in the Astral with a target AGC value of 5 × 10⁴ and maximum fill time of 6 ms. A high-energy collision dissociation energy setting of 27% was used for all MS/MS scans. Data were imported into Spectronaut v.19 (Biognosis). Relative peptide abundance was measured based on MS/MS fragment ions of selected ion chromatograms of the retention time-aligned runs. The MS/MS data were searched against a SwissProt Yeast database (downloaded in March 2025 from https://www.uniprot.org/uniprotkb?query = %28taxonomy_id%3A559292%29), a common contaminant/spiked protein database (bovine albumin, bovine casein, yeast alcohol dehydrogenase, and so on), and an equal number of reversed-sequence ‘decoys’ for false discovery rate determination. A Direct DIA+ library-free approach within Spectronaut was used to perform the database searches. Database search parameters included fixed modification on Cys (carbamidomethyl) with variable modification on Met (oxidation), Ser/Thr/Tyr (phosphorylation) and Protein N-term (acetylation). Full trypsin enzyme rules were used, with 10 ppm mass tolerances for precursor ions and 20 ppm for product ions.

Molecular dynamics simulation

We constructed five distinct simulation systems: CFN–Rho1–long-glucan (Glc_0–16)-bound Fks1 (System 1), Rho1–long-glucan-bound Fks1 (System 2), CFN–Rho1–no-glucan-bound Fks1 (System 3), Rho1–trimmed-glucan (Glc_0–3)-bound Fks1 (System 4) and the experimental short-glucan (Glc_1–3)-bound Fks1 (with the gate closed) (System 5). All systems featured a membrane composed of DOPC, DOPE and ergosterol at a molar ratio of 68:17:15. The CHS molecules in the structures were replaced manually with ergosterol in Coot⁴⁹ before system construction. The systems were assembled using the CHARMM-GUI Membrane Builder^55–58 and neutralized with 0.15 M Na⁺ and Cl⁻ ions. Simulations were performed using the CHARMM36m force field for lipids and proteins^59–61, along with the TIP3P water model⁶². We ran 1-ns constant particle number, volume and temperature simulations and 10-ns constant particle number, pressure and temperature simulations to equilibrate the systems preliminarily. The CFN force field parameters were generated using the CHARMM General Force Field⁶³. Production runs were conducted under NPT conditions at 303.15 K and 1 bar, with hydrogen mass repartitioning⁶⁴ to enable a 4-fs timestep. All simulations were executed in GROMACS⁶⁵ with input files generated by CHARMM-GUI. Production simulations of five replicas were performed for 1 μs for Systems 1–4 and 500 ns for System 5. VMD (v.1.9)⁶⁶ was used for visualization of molecular dynamics simulation results; Python 3 and MDAnalysis (v.2.7.0)⁶⁷ were used to analyse molecular dynamics results and generate molecular dynamics-related figures.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10409-7.

Author contributions

Z.R. and A.C. conducted Fks1 and Rho1 expression and purification. Z.R. performed cryo-EM sample preparation, single-particle three-dimensional reconstruction and part of the cryo-EM data collection under the guidance of S.-Y.L. A.C. carried out enzymatic assays and yeast survival assays under the guidance of K.Y. Z.R. and S.-Y.L. performed model building and refinement. C.L. carried out molecular dynamics simulation under the guidance of W.I. SY.O. and K.S. performed part of the cryo-EM screening and data collection under the guidance of M.J.B. S.-Y.L., Z.R., K.Y. and A.C. wrote the paper with inputs from the rest of the authors.

Peer review

Peer review information

Nature thanks Chia-Hsueh Lee and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Data availability

The coordinates are deposited in the PDB with the PDB IDs 9PE5 (L1), 9PE4 (L2), 9PE3 (ground state), 9PE2 (short-glucan-bound state), 9PE1 (CFN–Rho1–long-glucan-bound state) and 9ZTC (S643P), respectively. The cryo-EM density maps are deposited in EMDB with the IDs EMD-71554 (L1), EMD-71553 (L2), EMD-71552 (ground state), EMD-71551 (short-glucan-bound state), EMD-71550 (CFN-Rho1-long-glucan-bound state) and EMD-74746 (S643P), respectively. Source data are provided for Fig. 1b,c, Extended Data Figs. 2e–f and 8h and Supplementary Table 1. The MS/MS data were searched against a SwissProt Yeast database (https://www.uniprot.org/uniprotkb?query = %28taxonomy_id%3A559292%29).

Code availability

No custom code or mathematical algorithm was used for this study.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41586-026-10409-7.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-026-10409-7.