Invasive fungal infections cause over 1.5 million death annually, posing a serious threat to public health5. Such infections are of substantial concern (for example, causing high mortality) for immunocompromised populations as well as patients with COVID-19 (refs. 5,6) ( Limited classes of antifungal drugs and emerging drug-resistant strains such as the recent outbreak of multidrug-resistant Candida auris have caused serious challenges in treating patients who are infected7,8. Therefore, there are pressing needs to develop new antifungal agents8.

β-1,3-Glucan is a fundamental component of the fungal cell wall2, thus targeting its biosynthesis is an important strategy for developing broad-spectrum antifungal drugs3,9. β-1,3-Glucan of the fungal cell wall is synthesized by membrane-integrated synthase FKS10,11, which is regulated by GTPase Rho1 (refs. 12,13). FKS transfers glucose from donor UDP-glucose to the growing chain of glucan via β-1,3-glycosidic linkages and translocates the polymerized β-1,3-glucan into the extracellular space (Fig. 1a). FKS orthologues have been identified in all analysed fungi and are essential for the viability of many prominent fungal pathogens. In Candida glabrata and S. cerevisiae, simultaneous disruption of FKS1 and FKS2 showed lethal phenotype14,15; FKS1 of Candida albicans and Cryptococcus neoformans are essential for viability16,17; the mould Aspergillus fumigatus with an FKS1 deletion suffers from severe growth defects and cell lysis18. Currently, there are two well-known antifungal drugs on the market that target FKS: echinocandin, the first-line antifungal drug widely used in clinical practice19, and ibrexafungerp, a newly approved oral fungicidal drug4. More novel agents targeting FKS function (for example, rezafungin) are entering into late-stage clinical trials4. Unfortunately, FKS mutations are closely linked to echinocandin resistance and therapeutic failure, an emerging concern in invasive fungal infections20,21. Numerous clinically identified echinocandin-resistant mutations are concentrated at three conserved regions of FKS20,22. Mutations in a single FKS allele are sufficient to render fungal strains resistant to echinocandin23.

Fig. 1: Functional and structural characterization of S. cerevisiae FKS1.
figure 1

a, Model of the fungal cell wall. FKS1 uses UDP-Glc to synthesize β-1,3-glucan. The graphic in panel a was created using BioRender ( b, In vivo assay of FKS1 by examining the growth of the WT strain and the FKS1-KO (KO) strain with or without the FKS2 inhibitor FK506. Data are mean ± s.e.m.; n = 3 independent experiments. c, In vitro activity of CHAPS-purified FKS1 with different effectors: UDP-Glc, Rho1, GTPγS and antifungal drugs (caspofungin (CFN) and micafungin (MCF)). Activity was measured by monitoring the UDP generated. Data are mean ± s.e.m.; n = 3 independent experiments. d,e, Enzymatic digestion of the water-insoluble polymer synthesized by GDN-purified FKS1(S643P). Specific endo-1,3-β-glucanase and endo-1,4-β-glucanase were used. These experiments were repeated three times with similar results. A representative image of the polymer digestion (d) and hydrolysis samples in d withdrawn at indicated times and subject to thin-layer chromatography analysis (e) are shown. Glucose (G1), laminaribiose (G2), laminaritriose (G3) and laminarihexaose (G6) were used as the standards (lane M). f, Glycosyl linkage (methylation) analysis of the polymer synthesized by GDN-purified FKS1(S643P). A gas chromatogram profile of the partially methylated alditol acetates derived from the synthesized polymers is shown. The dominant peak (outlined by a green dashed box) represents the 1,3-Glcp linkage, confirmed by mass spectrometry (Extended Data Fig. 2d). g, Schematic of the domain organization of FKS1. h, Cryo-EM map of FKS1, viewed parallel to the membrane. The map is segmented into four parts as coloured in g with lipid-like densities in light yellow. i, Two orthogonal views of the FKS1 structure in cartoon display. Four parts of FKS1 are coloured as in g. j, Intracellular view of the cytoplasmic region with the TM domain hidden. The central β-sheet of the FKS1 GT domain is composed of β3–β11 as labelled. The dashed lines indicate disordered regions.

The lack of a FKS structure has hindered our understanding of the catalytic mechanism of FKS and hampered developments of new antifungal drugs. FKS is an approximately 200-kDa membrane-embedded protein with low-sequence homologies to other characterized glycosyltransferases (GTs). FKS belongs to the GT48 family with no structures available for any of its members. Here we report the cryo-electron microscopy (cryo-EM) structures of S. cerevisiae FKS1 and its drug-resistant mutant FKS1(S643P). Further structure-guided functional characterizations enable us to elucidate the molecular basis of β-1,3-glucan biosynthesis by FKS1 and to rationalize drug-resistant mutations of the enzyme.

Functional characterization of FKS1

For in vivo functional studies, we combined genetic deletion of FKS1 with pharmacological manipulation of FKS2 to circumvent the lethal phenotype caused by simultaneous disruption of both FKS1 and FKS2 in S. cerevisiae15. The wild-type (WT) strain showed modestly slowed growth with the FKS2-specific inhibitor FK506 (refs. 15,24) (Fig. 1b). By contrast, the FKS1-knockout (KO) strain exhibited significantly reduced growth without FK506 and a near total growth inhibition with FK506.

Next, we immunopurified the endogenously expressed S. cerevisiae FKS1 in the detergent CHAPS, which is frequently used for the FKS1 activity assay13,25 (see Methods). Traditionally, FKS activity is measured by a radioactive method on crude membrane or product-entrapped FKS13,25,26, and here we found that monitoring the UDP generated can make a practical radiolabel-free system for the FKS1 activity assay. FKS1 alone did not exhibit apparent activity. The addition of purified Rho1 in the GTPγS-loaded form dramatically boosted the activity of FKS1 (Fig. 1c), confirming the essential regulatory role of Rho1 on FKS1 (refs. 12,13). The specific activity of WT FKS1 was determined as approximately 824 nmol min−1 mg−1, which is comparable with previous studies10,25,27. More importantly, the addition of caspofungin or micafungin, two widely used antifungal drugs that can inhibit β-1,3-glucan synthesis19, significantly reduced FKS1 activity.

To further confirm the role of FKS1 in β-1,3-glucan synthesis, we analysed the product formed by purified FKS1. For this purpose, we tested WT FKS1 and the variant FKS1(S643P), which is one of the most frequently observed echinocandin-resistant mutations28. We hypothesized that the latter variant may alter synthase behaviour. When purified to homogeneity in the detergent GDN (Extended Data Fig. 1a–d and Supplementary Fig. 1), both WT FKS1 and FKS1(S643P) exhibited activity, which can be depleted by immunodepletion (Extended Data Fig. 1h–j). We found that the GDN-purified FKS1 can be activated by caspofungin. Caspofungin can stimulate the activity of FKS1(S643P) (Extended Data Fig. 1h–j). This is further confirmed by scale-up product synthesis in which only FKS1(S643P) with caspofungin produced a significant amount of water-insoluble polymer (Extended Data Fig. 1k). The fact that the enzyme purified in GDN and in CHAPS behaves very differently (Extended Data Fig. 1i versus Fig. 1c) suggests that FKS1 function may be sensitive to membrane environments. The product synthesis by FKS1(S643P) exhibits donor specificity towards the preferred ligand UDP-Glc over UDP-GlcNAc (Extended Data Fig. 1l). Moreover, EDTA can support the catalytic activity of the enzyme and Mg2+ significantly inhibited the activity (Extended Data Fig. 1m), consistent with previous studies25,29,30. Introduction of a K1261A mutation, an enzyme inactivation mutation discovered from our structural analysis (see the section ‘The active site and catalytic mechanism’), to FKS1(S643P) completely inactivated the product synthesis activity of the hyperactive enzyme (Extended Data Fig. 1h–k). The above assays illustrate the specific enzymatic activity of FKS1.

To confirm the formation of the β-1,3-glycosidic linkage, we performed a hydrolysis assay on the synthesized polymer by FKS1(S643P). As a result, endo-1,3-β-glucanase can completely hydrolyse the polymer, whereas endo-1,4-β-glucanase cannot (Fig. 1d and Extended Data Fig. 2a). Thin-layer chromatography analysis showed that the endo-1,3-β-glucanase-catalysed hydrolysate of our synthesized polymer matches the hydrolysate pattern of curdlan (the β-1,3-glucan standard) (Fig. 1e and Extended Data Fig. 2b,c). In addition, methylation glycosyl linkage analysis showed that 1,3-Glcp is the dominant linkage in the synthesized polymer (Fig. 1f and Extended Data Fig. 2d). In summary, the above biochemical and chemical studies together firmly establish that FKS1 is a β-1,3-glucan synthase.

Architecture of FKS1

To better understand the mechanism of FKS1, we determined the cryo-EM structure of GDN-purified FKS1 to an overall resolution of 3.4 Å (Fig. 1g,h, Extended Data Table 1 and Extended Data Fig. 3). The density map allowed the de novo building of an atomic model of FKS1 with most residues resolved (Extended Data Fig. 4). To our knowledge, it represents the first structure of the GT48 family membrane-bound GTs.

The overall structure of FKS1 measures approximately 100 Å × 115 Å × 100 Å (Fig. 1i). It contains a transmembrane (TM) domain with 17 TM helices (TM1–17) and a bulky cytoplasmic entity. The TM domain adopts a wedge-like shape, with its narrower end (approximately 35 Å) at the extracellular side. FKS1 has no apparent extracellular structured domains and several elongated loops connect the TM helices. There are several unstructured regions, detailed in the Methods section ‘Model building and refinement’.

The cytoplasmic part of FKS1 contains an N-terminal domain and a middle domain, separated by TM1–6 in the primary sequence (Fig. 1g). The middle domain (residues 712–1266) adopts a GT-A fold characteristic of GTs: a continuous central β-sheet of nine strands (β3–β11) surrounded by multiple α-helices31 (Fig. 1j and Extended Data Fig. 3h). It is therefore named as the GT domain. The N-terminal domain (residues 146–442) contains 14 α-helices (helix 6–7 is not modelled due to lack of densities) (Extended Data Fig. 3i). A search with the Dali server failed to retrieve any structures similar to this domain. Thus, we named it as the accessory (AC) domain. The AC domain interacts extensively with the GT domain, burying a total interface surface area of 2,504 Å2. Three loops from the GT domain (loop connecting β5 and β6 (Lβ5–β6), Lβ7–β8 and Lβ8–β9) function like plier jaws to clamp the AC domain (Fig. 1j). The highly conserved R319 from the AC domain interacts with N1087 from the GT domain in the domain interface (Extended Data Fig. 3j). The R319A substitution has been previously demonstrated to significantly reduce FKS1 function24, indicating that the two cytoplasmic domains function together as a compact structural unit.

TM domain of FKS1

Of the total 17 TM helices in FKS1, TM1–8 provide the bulk of the interacting interface with the cytoplasmic entity (Fig. 2a,b and Extended Data Fig. 3i). TM1–4 form a tightly packed helix bundle, contacting TM5–7 on one side (Fig. 2c). TM5–17 are all tilted like open flower petals when viewing from the cytosolic side (Fig. 2a,c). Between the tilted TM5–17 and the GT domain is a large solvent-exposed chamber, which is part of the active site of the enzyme (see below). The TM domain features two membrane-embedded pockets (Fig. 2a). The pocket on the extracellular side is formed because membrane-exposed TM9–10 and TM12 are significantly shorter than typical TM helices. The other pocket on the cytosolic side is formed due to the tilting of TM helices. These two pockets together probably form a path for glucan translocation (see below). Moreover, five long extracellular loops (EL1–5) intertwine to stabilize the close extracellular ends of TM5–17. Two conserved disulfide bonds (C658–C669 on EL1 and C1328–C1345 on EL2) and one potential N-linked glycan on the conserved N1849 are found in these loops (Fig. 2a,b). The tilting of TM5–17 results in large distances between neighbouring helices at the cytosolic end (Fig. 2c). The resulting widespread spaces created are filled in by three lateral helices (LH1–3) parallel to the membrane bilayer: LH1–2 fill the large gap created by two three-helix sheets (TM11, TM12 and TM16; and TM14, TM15 and TM17), whereas LH3 pads the ends of LH1–2.

Fig. 2: The FKS1 TM domain and structural comparison with the cellulose synthases BcsA, CesA and the hyaluronan synthase HAS.
figure 2

a, Front view of FKS1. Seventeen TM helices (denoted by numbers) are in cartoon representation; the cytoplasmic AC and GT domains are in yellow and blue ribbons, respectively. The conserved disulfide bonds at EL1–2 and the N-linked glycan on conserved N1849 of EL5 are labelled. Ordered lipids are in yellow sticks. The three ovals mark the positions of three pockets implicated in catalysis (one green pocket) and glucan translocation (two grey pockets). b, Topological diagram of the FKS1 TM domain. The dashed lines depict the flexible elements invisible in the map. c, Cytosolic view of the FKS1 TM helices. The inset shows the same view as in a with the cytosolic region hidden. The two circles with approximate diameters highlight the tilting of TM5–17, opening towards the cytoplasm. d, Residues (shown as sticks) interacting with the internal phospholipid (PL) are outlined by the solid box in a. e, Back membrane view of the lipid-enriched region outlined by the dashed box in a. fi, Comparison of the putative shared folds between FKS1 (GT48 family) (f) and three membrane-bound synthases of the GT2 family: the bacterial cellulose synthase BcsA (Protein Data Bank (PDB) ID: 4hg6) (g), the plant cellulose synthase CesA (PDB ID: 6wlb) (h) and the hyaluronan synthase HAS (PDB ID: 7sp7) (i). The superimposed structures are shown separately in cartoon representation. Their corresponding elements are in the same colour scheme, whereas elements out of the shared fold are in grey surface or cartoon representation as labelled. In f, the dashed lines indicate disordered regions. jm, Comparison of the TM arrangements in FKS1 (j), BcsA (k), CesA (l) and HAS (m). TM segments in the grey shadow belong to the shared fold. The TM helices are denoted by numbers. The asterisk marks the proposed glucan translocation path in FKS1, BcsA, CesA and HAS.

The resolution of the FKS1 map enabled us to identify multiple orderly bound lipids (Fig. 2a and Extended Data Fig. 4g,h). A phospholipid at the cytosolic membrane leaflet is contacted by TM11, TM15–16 and LH2 (Fig. 2d). Its headgroup is poised to contact the charged residues R1455 (TM11) and R1684 (TM16). Its two lipid alkyl tails pack with surrounding hydrophobic residues: H1654, I1655 and F1658 (TM15); I1680 (TM16); and V1745 and L1746 (LH2). Besides LH1–2, this phospholipid also appears to contribute to the stabilization of the tilted arrangement of the TM helices (Fig. 2a). Other lipid-like densities were modelled as 24 lipid alkyl chains, predominantly located around a core composed of TM5–13 (Fig. 2c). Of these, seven lipids wrap around TM5–6 (Fig. 2c), and 12 form two layers filling in the concave surface created by the tilted TM helices (Fig. 2e). These well-defined lipid molecules are probably part of the integral structure of FKS1.

A ‘cellulose-synthase-like fold’ in FKS1

Although the architecture of FKS1 is radically different from the membrane-bound GTs with known structures, we noticed that a part of the FKS1 structure (residues 615–1510) resembles the overall topologies of two cellulose synthases of the GT2 family: bacteria BcsA and its plant homologue CesA32,33 (Fig. 2f–h). Despite their low-sequence identities (less than 10%), FKS1 and BcsA can be superimposed with a root-mean-square-deviation value of 4.3 Å over 308 aligned residues (Extended Data Fig. 5a). A DALI homologous structure search identified two additional structures resembling this topology: hyaluronan synthase (HAS) and dolichylphosphate mannose transferase (PcManGT). Consistently, HAS has been reported with an overall structural similarity to BcsA34 (Fig. 2i,m); PcManGT has been proposed to contain a minimal cellulose-synthase-like fold with BcsA35. These further implicate that these membrane-bound polysaccharide synthases (FKS1, BcsA, CesA and HAS) may share a core fold, which we refer to as a cellulose-synthase-like fold. This fold shares several features: it is composed of six TM helices, with the GT domain splitting the N-terminal two helices from the C-terminal four helices (Fig. 2j–m); the membrane–cytosol boundary contains several amphipathic interface helices (IFs) (Fig. 2f–i); near this boundary, the TM helix following the GT domain (FKS1 TM7, BcsA TM5, CesA TM3 and HAS TM3) is elongated and has been proposed to line the glucan translocation path32,33,34.

Despite the overall resemblance of this modular fold, considerable FKS1-specific features were revealed (Extended Data Fig. 5). The GT domain gains considerate extensions relative to BcsA (Extended Data Fig. 5a,b), and the topologically corresponding TM helices between FKS1 and BcsA vary in orientation and length (Extended Data Fig. 5c). More notable variations are at the membrane–cytoplasm interface. BcsA, CesA and HAS feature three similar amphipathic IFs (IF1–3)33,34,36 (Fig. 2g–i). In FKS1, the topological counterpart of IF3 becomes TM helices (TM9–10) instead (Extended Data Fig. 5d,e). Although IF1–2 are preserved in FKS1, they show distinct states. IF1 (that is, helix α33 of the GT domain) of FKS1 rotates almost 90° and is sandwiched between TM6 and TM8 (Extended Data Fig. 5d). IF2 of BcsA corresponds to residues 1247–1266 of FKS1, which are disordered in the map, although predicted as a helix (Extended Data Fig. 5c). This putative FKS1 IF2 lacks the ‘QxxRW’ motif conserved among BcsA, CesA and HAS33,34,36 but could still contribute to catalysis, as we discuss later.

The active site and catalytic mechanism

Within the cellulose-synthase-like fold, FKS1 contains a large solvent-exposed chamber at the membrane–cytoplasm interface (Fig. 3a). Alignment of this fold with that of BcsA reveals that the chamber in FKS1 overlaps well with the active site of UDP-bound BcsA, strongly indicating the chamber as the active site of FKS1 (Fig. 3b). A DALI search retrieved homologous structures of the FKS1 GT domain (residues 718–1239): PcManGT, BcsA, GalNAc-T7, TarP, CesA and HAS. We also analysed the AlphaFold model of curdlan synthase because it synthesizes the same polymer as FKS1 and shows high homology (26% identities) with BcsA37. Detailed structural comparisons of these targets revealed that the active site of FKS1 overlaps better with membrane-bound enzymes such as BcsA, CesA, curdlan synthase, HAS and PcManGT (Extended Data Fig. 6). Despite the different stereochemistry of their products, the high structural similarity of their active sites indicates similar mechanisms in substrate binding and catalysis of these synthases, consistent with previous studies of BcsA, CesA and HAS33,34,36. On the basis of these insights, we overlaid the structures of FKS1 and BcsA to reveal the shared key functionally important residues in their active sites (Fig. 3c,d). For example, two pairs of residues (Y849 and E851, and K1082 and N1085 in FKS1; Y149 and E151, and K226 and N229 in BcsA) and the ‘ED motif’ (E1221 and D1222 in FKS1; and E342 and D343 in BcsA) in the two enzymes can be aligned well. These residues in BcsA are known for donor binding (Y149 and E151, and K226 and N229 in BcsA) and for positioning the general catalytic base aspartate (ED motif)36.

Fig. 3: The active site of FKS1.
figure 3

a, A cutaway surface representation of FKS1 showing the active site at the membrane–cytoplasm interface. It connects a membrane pocket, which has the potential for glucan translocation. Specific structural elements are shown in cartoon representation, coloured as in Fig. 2a. The TM helices are denoted by numbers. The highly conserved ED motif (E1221 and D1222) located on helix α35 is facing the active site as marked. b, Superimposed GT domains of the cellulose-synthase-like folds shared between UDP-bound BcsA (PDB ID: 4P00; in grey with UDP in the magenta stick) and FKS1 (in orange). c,d, Close-up view of the active sites between BcsA (c) and FKS1 (d), as marked by the dashed box in b. Residues involved in substrate binding or catalysis are indicated as sticks. In c, UDP is shown as the magenta stick. In d, FKS1 residues conserved with BcsA are highlighted as yellow sticks. e, In vivo functional assay of active site mutations in FKS1. The growth of the indicated strains for 72 h was assayed without (left rank) and with (right rank) the FKS2-specific inhibitor FK506. The test was repeated three times with similar results. f, Levels of cell wall glucan in strains carrying the indicated FKS1 mutations. g, In vitro FKS1 activities of WT and active site mutants purified with CHAPS, with or without the FKS1 inhibitor CFN. Activities were normalized to the amount of protein. For f,g, data are mean ± s.e.m.; n = 3 independent experiments.

Next, we validated the functional roles of these residues using three different assays. First, we introduced their mutations into the FKS1 chromosomal locus and assayed the growth phenotype of each mutant strain in the presence of FK506 (Fig. 3e). The double mutations of Y849A;E851A and K1082A;N1085A and the single mutations of E851A, K1082A and D1222A all prevented strain growth to the extent of the FKS1-KO strain (Fig. 3e). The N1085A mutation slowed growth and the Y849A mutation did not have an obvious effect on growth. Second, we measured the levels of cell wall glucan in these strains when growing without FK506. We found that all mutations, including the FKS1-KO strain but except for the Y849A substitution, decreased the overall levels of glucan compared with the WT strain (Fig. 3f), matching well with their growth phenotypes (Fig. 3e). Last, we purified every single FKS1 mutant (except for FKS1(D1222A)) and compared their catalytic activity in vitro (Fig. 3g and Extended Data Fig. 1g). Consistent with the cell-based phenotypes described above, the E851A, K1082A and N1085A mutations significantly reduced the enzyme activity, whereas the Y849A mutation only mildly affected the activity. Together, the above in vivo and in vitro analyses firmly establish the location of the FKS1 active site as well as its key composing residues.

Further structural analysis allowed us to identify additional residues critical for FKS1 catalysis. R382 and W383 in the ‘QxxRW’ motif in IF2 of BcsA is known to be important for substrate binding and catalysis36. Although this motif is missing in FKS1, K1261 and F1258 are located at the corresponding positions in FKS1, as R382 and W383 do in BcsA (Fig. 3c,d and Extended Data Fig. 5e). In multiple membrane-bound synthases containing the cellulose-synthase-like fold (BcsA, CesA and HAS), the positively charged residue corresponding to BcsA R382 is involved in donor binding, whereas the aromatic residue corresponding to BcsA W383 functions to stabilize the acceptor33,34,36 (Extended Data Fig. 6). Indeed, both F1258 and K1261 are essential for FKS1 function, as demonstrated by the in vivo and in vitro analyses of their single alanine mutations (Fig. 3e–g). The above analyses further suggest that the core catalytic mechanisms between FKS1 and BcsA are conserved.

The putative glucan translocation path

FKS1 and BcsA are both processive polysaccharide synthases and their polymerized product, glucans, need to be translocated across membranes1,38. In accordance with this, right below the active site, FKS1 has a pocket formed within the membrane (Fig. 3a). It is lined by TM5–8, TM11–12 and IF1 (Fig. 4a). TM5–8 and TM11–12 of FKS1 are part of the cellulose-synthase-like fold core and they correspond to TM3–8 forming the cellulose translocation path in BcsA36 (Fig. 2j,k). At the extracellular side opposite to this pocket, there exists another solvent exposed pocket with much lower detergent density that is directly visible from the EM map (Fig. 4b). The second pocket is formed because the membrane-exposed TM9–10 and TM12 are short. The two juxtaposed hydrophilic pockets with the membrane plane lead to notable membrane thinning (Fig. 4a,b), which could greatly reduce the energy barrier for glucan translocation across the membranes, as has been proposed for several membrane translocases39. Indeed, mutations of residues along this path in FKS1 altered FKS1 function (Fig. 4c,d). For example, two conserved residues (F1297 and H1298) are located between the active site and the cytosolic entrance. Their alanine substitutions impaired FKS1 function, suggesting that they may be involved in glucan guiding. By contrast, when substituting two conserved bulky aromatic residues sealing the extracellular exit (F1311 and F1475) with alanine, the mutant strains grew faster than the WT strain. We were able to purify FKS1(F1475A) and it exhibited increased in vitro catalytic activity and reduced sensitivity to the FKS1 inhibitor caspofungin (Fig. 4d), implying the enlargement of F1475A in the glucan exit.

Fig. 4: Putative glucan translocation path.
figure 4

a, Putative glucan translocation path through two membrane pockets as illustrated by the hexagon array. The first pocket (in the internal shaded surface view), lined by TM5–12, is located under the active site (green dashed oval). Conserved residues implicated in guiding the entrance (H1298 and F1297) and sealing the exit (F1311 and F1475) are labelled. At the extracellular side, a second pocket is formed because membrane-exposed TM9–10 and TM12 are too short (less than 28 Å) to span the whole membrane. b, Unsharpened EM density map shows the detergent depression close to the glucan translocation path. This path is marked by the white dashed box, with the arrows indicating the putative entrance and exit. The detergent density is in grey. The estimated membrane thickness is labelled. The inset shows the extracellular view of the region boxed with the orange line. c, In vivo functional assay of FKS1 with selected mutations around the glucan translocation path. The growth of the indicated strains for 48 h were assayed without (left) and with (right) the FKS2-specific inhibitor FK506. d, In vitro FKS1 activities of the WT and the F1475A mutant purified with CHAPS, with or without the FKS1 inhibitor CFN. Activities were normalized to the amount of protein. Data are mean ± s.e.m.; n = 3 independent experiments. e, Structure of FKS1(S643P) shows an elongated density (blue mesh) in the first membrane pocket (as shown in a) for product translocation. This structure was determined in the presence of UDP-Glc to synthesize product, which was boosted by CFN. The TM helices are denoted by numbers. The inset shows the zoomed-in view of a modelled glucan chain of 4-glucose (yellow stick) superimposed with its corresponding density (blue mesh). f, Potential residues involved in glucan interaction.

We next determined the structure of the GDN-purified hyperactive mutant FKS1(S643P) when incubated with UDP-Glc and caspofungin, to the resolution of 3.5 Å (Extended Data Fig. 7), hoping to capture the substrate-bound or product-bound state of the enzyme. Although the structure of FKS1(S643P) shows no large conformational differences from that of WT FKS1 (root-mean-square deviation of approximately 0.7 Å), we noticed an elongated density along the proposed product translocation channel in it, presumably corresponding to a bound product (Fig. 4e). We modelled a glucan chain of 4-glucose into this density (Fig. 4e, inset). A series of hydrophilic and hydrophobic residues from FKS1 are lined along the modelled glucan chain for favourable enzyme–product interactions (Fig. 4f). The translocation channel remains closed at the extracellular side, suggesting the existence of a regulated channel opening mechanism, as has been recently proposed for HAS34.

Implications for echinocandin resistance

We first evaluated our S. cerevisiae-based assay system for the analysis of echinocandin resistance. We selected two most frequently observed mutations in drug-resistant C. albicans (CaFKS1 F641S and S645P)28 and introduced them into the corresponding sites of S. cerevisiae FKS1 (ScFKS1 F639S and S643P). Both mutant strains exhibited highly elevated resistance to caspofungin (Extended Data Fig. 8a). We further purified these two mutated ScFKS1 in CHAPS and evaluated their inhibition by caspofungin (Fig. 5a). The WT ScFKS1 as control exhibited a typical inhibitory profile, with the IC50 determined as 0.55 µM, which is comparable with that determined previously with different methods23,25. By contrast, both ScFKS1(F639S) and ScFKS1(S643P) are insensitive to inhibition by caspofungin.

Fig. 5: Structural interpretation of echinocandin-resistant or ibrexafungerp-resistant mutations.
figure 5

a, CFN inhibition profiles of FKS1 variants purified with CHAPS, including WT FKS1 and two most frequently observed echinocandin-resistant mutations: F639S and S643P. The percentage of residual activity is shown. Data are mean ± s.e.m.; n = 3 independent experiments. b, Three hotspot regions (red; denoted as HS1–3) with echinocandin-resistant mutations are mapped onto the FKS1 structure. The hotspot regions are localized on a cluster of three neighbouring TM helices: TM5, TM8 and TM6, respectively. The conserved cellulose-synthase-like fold within FKS1 is highlighted in cartoon representation, whereas the remaining part is in ribbon representation. This conserved fold is surrounded by ordered lipids shown as yellow sticks. The dashed line indicates disordered region. c, Close-up view of the echinocandin-resistant mutations as outlined in b. An ibrexafungerp-resistant mutation site (A1369) outside these hotspot regions is also shown. The Cα atoms of the mutations are denoted as red spheres. The hotspot regions are enriched with ordered lipids, with selected lipid densities shown in a grey mesh. d, Conformational changes and lipid arrangements (indicated by red arrows) around the HS1 region due to the introduction of FKS1(S643P), the drug-resistant mutation. This is illustrated by the overlay of the FKS1 structure (grey) and the FKS1(S643P) structure (blue). The black dashed line indicates a potential polar interaction.

Next, we investigated the possible mechanistic basis of echinocandin-resistance using the FKS1 structure. According to their locations, echinocandin-resistant mutations were categorized into three conserved regions, encompassing residues 639–647 in TM5 (hotspot 1), 1354–1357 in TM8 (hotspot 2) and 690–700 in TM6 (hotspot 3) (residues are numbered as in ScFKS1)20,22,40 (Supplementary Fig. 2). Although separated in the primary sequence, three mutation hotspots are spatially close to each other (Fig. 5b,c). In C. glabrata isolates resistant to ibrexafungerp, new mutation sites outside FKS2 hotspots 1 and 2 were reported (for example, CgFKS2 W715L and A1390D)41; these mutation sites correspond to W695 in hotspot 3 of ScFKS1 and A1369 on TM8 adjacent to hotspot 3, respectively (Fig. 5c). This analysis suggests overlapping FKS-binding sites between echinocandin and ibrexafungerp41. These drug-resistant hotspot regions appear to be closely related to the membrane environment. All echinocandin-resistant mutations of FKS1 are mapped near the convex hydrophobic surface formed by TM5–6 and TM8 and are specifically surrounded by several ordered lipid molecules (Figs. 2c and 5c). F639 and S643 of the hotspot 1 region are involved in lipid binding; the side chain of F639 directly interacts with three lipid molecules and the S643 side chain appears to fix the main chain of the lipid-interacting residue Y638 (Fig. 5d). Furthermore, we analysed the structure of FKS1(S643P) (Extended Data Fig. 7), focusing on alterations of lipid binding induced by the mutation (Fig. 5d). Several hotspot 1 residues near the enriched lipids show altered conformation; Y638 has its side chain rotated approximately 90°, with the F639 side chain rotating in the same direction. Remarkably, these changes lead to shifts of nearby bound lipids, suggesting that an altered lipid environment resulted from the S643P mutation.

On the basis of our results, we propose two possible drug-resistant mechanisms for FKS1. First, as the echinocandin-type drugs are lipopeptides with differently configured lipid tails19 (Extended Data Fig. 8b,c), the clustering pattern of the echinocandin-resistant mutations suggests a possible binding site for these drugs, and the bindings may involve their lipid tails (Fig. 5b,c). The drug resistance may hence be attributed to alterations of the drug-binding site that resulted from the mutations. Second, considering the enriched ordered lipids near drug-resistant hotspot regions and the lipid movements revealed from the structure of FKS1(S643P) (Fig. 5c,d), it is conceivable that the drug-resistant mutations could alter the response of FKS1 to membrane changes caused by echinocandin. Such a mechanism has been proposed for the antibiotics daptomycin and polymyxins, which are also lipopeptide drugs and membrane-acting42,43. Consistently, it has been reported that specific membrane microdomains are associated with the synthesis of β-1,3-glucan44,45. In addition, changes in the lipid microenvironment have been demonstrated to correlate with the action of echinocandin on FKS46,47.


Our work reveals the molecular architecture of the fungal glucan synthase FKS1. The structures of FKS1, together with extensive functional characterizations, advance our mechanistic understanding of fungal cell wall β-1,3-glucan synthesis. Structure-based analysis of echinocandin-resistant mutations and the structure of the S643P mutant suggest a plausible drug-resistant mechanism of the FKS1 mutations. These insights may also be shared by various fungal pathogens as their FKS orthologues are highly conserved (Extended Data Fig. 9 and Supplementary Fig. 2). The FKS1 structures and the catalytic mechanism elucidated in this work may serve as a framework for future studies of this important antifungal drug target, as well as for screening and the development of new antifungal agents.


Expression and purification of FKS1

A 3× Flag tag was engineered to the C terminus of chromosomal FKS1 on S. cerevisiae strain BY4742, using a PCR-based tagging method48. A yeast strain with the FKS1(S643P) chromosomal mutation was generated using the homologous recombination method49. For cryo-EM analysis and product synthesis, FKS1 and the FKS1(S643P) mutant were purified with detergent GDN as described below. The strains were cultured in YPD medium at 30 °C for 20 h. Cells were collected by centrifugation and lysed by French Press in lysis buffer containing 50 mM Tris-HCl pH 7.4, 150 mM NaCl and 2 mM MgCl2 supplemented with 1 mM phenylmethane sulphonylfluoride (PMSF). The lysate was centrifuged at 15,000g for 30 min. The supernatant was subject to ultracentrifugation at 100,000g for 1 h. The collected membrane pellet was solubilized in buffer 50 mM Tris-HCl pH 7.4, 500 mM NaCl, 2 mM MgCl2, 10% (v/v) glycerol, 1.5% (w/v) n-dodecyl-β-d-maltopyranoside (DDM; Anatrace), 0.15% cholesteryl hemisuccinate Tris salt (CHS; Anatrace) and protease inhibitor (cOmplete protease inhibitor cocktail; Roche) by gentle agitation for 2 h. Supernatant was collected by centrifugation at 15,000g for 0.5 h and was applied to Anti-Flag M2 affinity gel (Sigma). The gel was then washed with washing buffer containing 50 mM Tris-HCl pH 7.4, 150 mM NaCl, 2 mM MgCl2 and 0.04% GDN (Anatrace). The target proteins were eluted with washing buffer supplemented with 150 μg ml−1 3× Flag peptide. The eluate was concentrated and further purified using size-exclusion chromatography (Superose 6 10/300 GL column, GE Healthcare) with buffer containing 50 mM Tris-HCl pH 7.4, 150 mM NaCl, 2 mM MgCl2 and 0.04% GDN (Anatrace). The central fractions of monodisperse peak were collected and concentrated for cryo-EM grid preparation and product synthesis.

Expression and purification of Rho1

The vector construction of N-terminal 6× His-SUMO-tagged S. cerevisiae Rho1 was transformed into Escherichia coli BL21(DE3). The strains were cultured in LB medium supplemented with 100 µg ml−1 ampicillin at 37 °C till OD600 reached approximately 0.6. Protein expression was induced with 0.5 mM isopropyl-β-d-1-thiogalactopyranoside at 18 °C for 20 h. The collected bacteria were resuspended and lysed by French Press in buffer A containing 50 mM Tris-HCl pH 7.4, 300 mM NaCl and 2 mM MgCl2. After centrifugation, the protein was purified from the supernatant using nickel–nitrilotriacetic acid (Ni–NTA) affinity chromatography. When dialysing into buffer A, 6× His-tagged protease Ulp1 at a 1:200 Ulp1:Rho1 (w/w) ratio was added to remove the 6× His-SUMO tag. The cleaved 6× His-SUMO tag and the protease were removed by another run of Ni–NTA affinity chromatography, and the flow-through was collected, concentrated and subjected to a gel-filtration chromatography (Superdex 200, GE Healthcare) in buffer containing 50 mM Tris-HCl pH 7.4, 150 mM NaCl and 2 mM MgCl2. The central fractions of the monodisperse peak at 15.6 ml were collected and concentrated to 10 mg ml−1.

Cryo-EM data acquisition

For single-particle cryo-EM analysis, a 3 µl aliquot of the purified FKS1 at a concentration of 5 mg ml−1 was applied to glow-discharged Quantifoil carbon grids (R1.2/1.3 Au, 300 mesh). The grids were blotted for 4 s at 100% humidity and flash frozen in liquid ethane using FEI Vitrobot IV. For the UDP-Glc-incubated FKS1(S643P) sample, FKS1(S643P) purified in GDN (7 mg ml−1) was mixed with UDP-Glc (0.5 mM), supplemented with 0.7 mg ml−1 Rho1, 10 μM GTPγS, 200 μM caspofungin, 0.1% CHAPS and 0.02% CHS. The mixture was incubated at 16 °C for 2 h before freezing the cryo-EM grids. Cryo-EM data were collected on a FEI Titan Krios electron microscopy operated at 300 kV equipped with a Gatan K3 Summit camera positioned after a GIF quantum energy filter. Automated data acquisition was performed with SerialEM or FEI EPU. Micrographs were recorded under super-resolution counting mode at a nominal magnification of ×130,000, in a physical pixel size of 0.92 Å per pixel. Defocus values ranged from −1.2 μm to −3 μm. A total exposure of 1.3 s was dose-fractionated into 32 frames, resulting in a total accumulated dose of 50 e per Å2.

Image processing and 3D reconstruction

The dose-fractionated movies recorded were first aligned and dose-weighted with MotionCor2 (ref. 50). CTFFIND4 was then used to determine the contrast transfer function parameters for individual micrographs51. Low-quality micrographs revealed by manual inspection were excluded from further analysis. Subsequent image-processing steps were performed using RELION-3 (ref. 52).

For the FKS1 dataset, a set of 2,000 particles were manually selected to generate 2D class templates for reference-based automatic particle picking. The automatic picking yielded in 2,321,905 particles from 11,606 micrographs. Two rounds of reference-free 2D classification were performed to remove particles with poor quality, resulting in a cleaned set of 1,335,014 particles. An ab initio map was generated with RELION and was used as the initial reference model for further 3D classification. After three rounds of 3D classification, a 3D class with high-resolution features (267,574 particles) was selected. Subsequent particle polishing, 3D refinement and post-processing generated a map with an overall resolution of 3.4 Å.

For the FKS1(S643P) dataset, a set of approximately 2,000 particles were manually selected to generate 2D class templates for reference-based automatic particle picking. The automatic picking yielded in 2,222,315 particles from 9,623 micrographs. Two rounds of reference-free 2D classification were performed to remove particles with poor quality, resulting in a cleaned set of 1,620,308 particles. The FKS1 model was used as the initial reference model for further 3D classification. A 3D class with high-resolution features (690,251 particles) was selected. Subsequent particle polishing, 3D refinement and post-processing generated a map with an overall resolution of 3.6 Å. This reconstruction revealed fragment density within the product translocation channel, indicating a mixed state with or without bound product. Therefore, we performed an additional round of focused 3D classification without particle realignment, using a focused mask around TM7–12 and the active site that surround the translocation channel. A 3D class with highest resolution features (176,682 particles) was selected. The selected particles were re-extracted for 3D refinement with full mask and subsequent post-processing. This generated a map with an overall resolution of 3.5 Å, which was used to build the FKS1(S643P) model.

The overall resolutions were estimated based on the gold-standard Fourier shell correlation 0.143 criterion53. Local resolution distribution was estimated using ResMap54.

Model building and refinement

A rough initial model of FKS1 was generated de novo using the map_to_model module in PHENIX55. It was further improved by manual adjustments and rebuilding using Coot56, which was facilitated by the good densities around the TM helices and bulky densities around residues such as Trp, Tyr, Phe and Arg. The refinement of the FKS1 model against the cryo-EM map in real space was carried out in PHENIX with secondary structure and geometry restraints55. In the final FKS1 model, there are several unstructured regions, including N-terminal 145 residues, C-terminal 16 residues, six flexible segments of cytoplasmic domain (residues 244–278, 475–487, 798–805, 897–931, 1159–1167 and 1247–1266) and four loop segments to connect TM helices (residues 1419–1435, 1516–1554, 1627–1637 and 1698–1723). The model of FKS1(S643P) was built based on the model of FKS1. MOLPROBITY was used to assess the final model57. The Fourier shell correlation between the model and the map was calculated by PHENIX.mtriage58. The homologous structure search was performed using the DALI server59. For structural comparison with FKS1, the AlphaFold-modelled structure of curdlan synthase was used60,61. The illustrated figures were prepared using PyMOL (Schrödinger, LLC), Chimera and ChimeraX62,63. Statistics of the 3D reconstructions and model refinements are shown in Extended Data Table 1.

Spot growth assay

Yeast strains with FKS1 chromosomal mutations and the FKS1-KO strain were generated using the homologous recombination method49. The mutants were selected on SD-His (Yeast Synthetic Drop-out Medium without Histidine, cat no. S0020, Solarbio) and were confirmed by genomic PCR and DNA sequencing. For the analysis of growth phenotype, the spot growth assay of mutated strains was adapted from a previously described method24. The strains were cultured in YPD media at 30 °C at 200 rpm to exponential phase until OD600 was 0.8. The cells were diluted to an OD600 of 0.1. Then, 5 μl portions of tenfold dilutions of the indicated strains were spotted on SD-His with or without FK506 (1 μg ml−1). After incubation at 30 °C for indicated time, plates were photographed with the 5200CE Image System (TANON). Each assay was repeated three times with similar results.

Growth curve analysis

To profile the cell growth, the WT strain and the FKS1-KO strain were grown in YPD media at 30 °C overnight to exponential phase. Then, cultured cells were inoculated into the fresh YPD medium with an initial OD600 of 0.01, which was supplemented with or without 1 μg ml−1 FK506. The new cultures were then grown at 30 °C till stationary phase. During the cultivation, the optical densities (OD600) were measured at 8 h interval for 48 h. The experiments were done in triplicate to make the cell growth curves.

FKS1 activity assay

The yeast strains carrying different FKS1 variants with the C-terminal 3× Flag tag were generated, cultured and lysed as described above. The collected membrane was solubilized in buffer containing 50 mM Tris-HCl pH 7.4, 1 mM EDTA, 33% glycerol, 0.5% CHAPS (Anatrace), 0.1% CHS (Anatrace), 4 µM GTPγS and protease inhibitor (cOmplete protease inhibitor cocktail, Roche). FKS1 and its variants were purified using anti-Flag M2 affinity gel (Sigma) and were eluted in buffer containing 50 mM Tris-HCl pH 7.4, 1 mM EDTA, 33% glycerol, 0.2% CHAPS and 0.04% CHS, supplemented with 150 μg ml−1 3× Flag peptide. The proteins purified in detergent CHAPS or GDN were assayed by using the UDP-Glo Glycosyltransferase Assay kit (Promega) to monitor the UDP released. Of FKS variants, 2.5 µl was added to 30 μl reaction mixture in total containing 50 mM Tris-HCl pH 7.4, 33% glycerol, 1 mM EDTA, 6 μg ml−1 Rho1, 0.2% CHAPS, 0.04% CHS, 4 µM GTPγS and 20 mM potassium fluoride (KF). The reaction was initiated by adding UDP-Glc to a final concentration of 2.5 mM. The reaction was carried out at 30 °C for 1 h. Luminescence was recorded using a Pherastar FS system (BMG Labtech). The final activity was normalized to the FKS1 amount that was measured by immunoblotting against the 3× Flag tag, using the DYKDDDDK tag monoclonal antibody (1:1,000 dilution; cat no. 66008-3-Ig, clone no. 2B3C4, Proteintech). In the case of inhibitor profiling, serial dilutions of echinocandin drugs were first incubated with FKS1 variants for 10 min at room temperature before adding other components to react.

In vitro synthesis of β-1,3-glucan and aniline blue staining

For in vitro β-1,3-glucan synthesis, the enzyme purified in detergent GDN (FKS1 or the FKS1(S643P) mutant; 0.02 mg ml−1) and donor UDP-Glc (2.5 mM) were mixed in reaction buffer (50 mM Tris-HCl pH 7.4, 33% glycerol, 1 mM EDTA, 6 μg ml−1 Rho1, 0.2% CHAPS, 0.04% CHS, 4 µM GTPγS and 20 mM KF), in the presence or absence of 200 μM caspofungin. The reaction volume was set up as 100 μl (for aniline blue staining) or 10 ml (for product enrichment and subsequent enzymatic degradation and glycosyl linkage analysis). The reaction was carried out at 30 °C for indicated time period.

For the staining of the synthesized products by aniline blue, a 100 μl aliquot of reactants or 0.1% (w/v) of S. cerevisiae β-glucan (Sigma) was taken and added to equal volume of aniline blue (0.03%; Sigma). The mixture was incubated in the dark for 20 min for complete dye binding. Each sample was then loaded into a capillary and was observed under a fluorescence microscope with an excitation wavelength of 365 nm and an emission wavelength of 433 nm. Each assay was repeated three times with similar results.

Enzymatic degradation analysis of the synthesized polymer

The in vitro synthesized polymer is water-insoluble and was collected by centrifugation. It was washed twice with buffer: 50 mM Tris-HCl pH 7.4, 33% glycerol, 1 mM EDTA, 0.2% CHAPS, 0.04% CHS, 20 mM KF and twice with deionized water. Two enzymes were used for the degradation analysis of the synthesized polymer: endo-β-1,3-glucanase (Trichoderma sp.; product code: E-LAMSE, Megazyme) or endo-β-1,4-glucanase (Aspergillus niger; product code: E-CELAN, Megazyme). These two enzymes were first dialysed into the buffer (100 mM NaAc pH 4.5). Then, 0.5 U of each enzyme was added to 50 µl 1% (w/v) of synthesized polymer in 200 mM NaAc (pH 4.5). Degradation of standard controls cellulose (C6288, Sigma) or curdlan (C7821, Sigma) was also performed with the same condition64,65. The mixtures were incubated at 40 °C, and samples were withdrawn at various time intervals and were boiled for 5 min for reaction termination. The enzymatic degradation products were then analysed by thin-layer chromatography. In brief, a 2 µl aliquot of the withdrawn samples was spotted on a silica gel plate (Merck Silica gel 60 F254) and developed in a solvent system containing n-butanol:acetic acid:water (2:1:1 v/v/v). The plates were visualized by immersing in methanol:sulfuric acid (95:5 v/v) and subsequent heating at 95 °C. A mixture of glucose (G5767, Sigma), laminaribiose (O-LAM2, Megazyme), laminaritriose (O-LAM3, Megazyme) and laminarihexaose (O-LAM6, Megazyme) was used as the thin-layer chromatography standards.

During enzymatic digestion of synthesized polymer, the reducing sugars released were measured using the DNS reagents (Micro Reducing Sugar Assay Kit, Abbkine). In brief, 175 μl of the diluted hydrolysis sample was mixed with 125 μl of DNS reagent. Standard from the kit at the concentration range of 0.2–0.6 mg ml−1 was used for the generation of the standard curve. Reaction mixtures were boiled in a water bath for 5 min. After cooling to room temperature in a water-ice bath, the absorbance of a 0.2 ml sample was measured at 540 nm.

Glycosyl linkage (methylation) analysis

The in vitro synthesized polymer by FKS1 was washed twice with buffer: 50 mM Tris-HCl pH 7.4, 33% glycerol, 1 mM EDTA, 0.2% CHAPS, 0.04% CHS, 20 mM KF and two times with deionized water. The washed sample was then dialysed into deionized water and freeze-dried. Methylation analysis was performed following a previous study66. The dried sample (approximately 1 mg) was dissolved in DMSO (500 μl). The methylation was performed in DMSO/NaOH with iodomethane for 1 h. The methylated products were hydrolysed with 2 M trifluoroacetic acid at 121 °C for 90 min, reduced by NaBD4 and acetylated with acetic anhydride at 100 °C for 2.5 h. The resulting partially methylated alditol acetates were analysed using the GC-MS system (6890A-5975C, Agilent Technology) equipped with an Agilent BPX70 chromatographic column (30 m × 0.25 mm × 0.25 µm; SGE). The temperature program was set as follows: 140 °C for 2 min, 140–230 °C at 3 °C per minute and 230 °C for 3 min.

Quantitative determination of β-1,3-glucan in cell walls

The levels of β-1,3-glucan were determined using the aniline blue assay, as previously described23,26,67,68,69,70,71,72. Testing strains were grown in YPD media to exponential phase until OD600 was 0.5. The same amount of cells for each strain was collected by centrifugation at 5,000g. The collected cells were washed twice with TE buffer (10 mM Tris-HCl pH 8.0, and 1 mM EDTA). The pelleted cells were suspended with 0.5 ml TE buffer and mixed with 0.1 ml of 6 M NaOH. The mixture was incubated in an 80 °C water bath for 30 min to solubilize the glucan. Then, 2.1 ml of aniline blue (0.03% aniline blue, 0.18 M HCl and 0.49 M glycine-NaOH pH 9.5) was added to each sample. The samples were briefly vortexed and incubated at 50 °C for 30 min and an additional 30 min at 24 °C. Fluorescence was quantified using a fluorescence plate reader (CLARIOstar Plus, BMG Labtech) under an excitation wavelength of 400 nm and an emission wavelength of 460 nm with a cut-off of 455 nm. All samples were measured in triplicates.

Protein identification with mass spectrometry analysis

The protein samples were separated by SDS–PAGE gel and were stained with Coomassie. The gel band was manually excised and destained. The proteins were reduced with DTT, alkylated with IAA and digested with proteomics-grade trypsin in 20 mM ammonium bicarbonate73. The digestion was performed overnight at 37 °C and stopped by adding 2% formic acid. The peptides in gel were extracted using solution containing 2% formic acid and 67% acetonitrile. The peptides were vacuum-dried, resuspended in 0.1% formic acid, loaded onto the trap column nanoViper C18 (3 μm, 100 Å) and separated on an analytic column (Acclaim PepMap RSLC, 75 μm × 25 cm; C18 2 μm, 100 Å) using the EASY nLC 1200 HPLC system (Thermo Fisher). The elution gradient was 5–38% buffer A (0.1% formic acid and 80% acetonitrile) over 30 min. The mass spectrometry analysis was performed on a Q Exactive mass spectrometer (Thermo Fisher). The resulting data were converted to mgf file with ProteoWizard and analysed using the Mascot search engine for protein identification against a UniProt Saccharomyces cerevisiae database with a false discovery rate of less than 1%.

Reporting summary

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