Structural basis of chondroitin sulfate backbone polymer synthesis Authors: Daniel Tehrani, Nil Cortiella-Valls, Chin Huang, Digantkumar Chapla, Zhifeng Zheng, Aarya Venkat, Brady O’Boyle, Natarajan Kannan, Camilo Perez, Kelley W. Moremen Published: 2026 ABSTRACT Chondroitin sulfate (CS) proteoglycans are extended (-GlcAβ1,3GalNAcβ1,4-)n co-polymers attached to cell surface and extracellular matrix core proteins that are further modified by extensive sulfation and epimerization. Four homologous proteins contribute to CS backbone synthesis (CHPF1, CHPF2, CHSY1, and CHSY3) and prior data suggests assembly of the proteins into heterocomplexes is required for function. Here we show by sequence alignment and structural modeling that all CHSYs and CHPFs contain an N-terminal CAZy GT31-like domain and a C-terminal GT7-like domain separated by a cystatin-like linker domain. Co-expression of one CHPF and one CHSY is required to form a soluble, functional heterodimeric CS synthase and structural modeling indicates all four potential CHSY-CHPF combinations can form equivalent heterodimeric complexes. Cryo-EM studies on CHSY3-CHPF1 confirm the structure, interface, and active site features predicted by the structural models. Enzymatic analyses of catalytic mutants demonstrate that only the glycosyltransferase domains in the CHSYs are responsible for polymer synthesis: the GT31 domain transfers β1,3-GlcA while the GT7 domain transfers β1,4-GalNAc. The corresponding CHPF domains do not contribute to polymer synthesis but stabilize the corresponding CHSY functional domains. Additional mutagenesis and modeling suggest that the bridging cystatin-like domains may contribute to efficient polymer synthesis. Introduction Chondroitin sulfate (CS) chains are extended carbohydrate polymers comprised of alternating disaccharide repeats of N-acetylgalactosamine (GalNAc) and glucuronic acid (GlcA) residues further modified by sulfation and epimerization1–5. The CS chains are covalently attached to proteoglycan (PG) core proteins on cell surfaces and extracellular matrices and are ubiquitously expressed in higher metazoan organisms6. The glycan polymers participate in key developmental processes, including cytokinesis, morphogenesis, neural development, synaptic plasticity, post-injury axon regeneration, and numerous cell signaling pathways, as well as facilitating pathogen adhesion2,3,7–10. CS is also highly expressed within connective tissues, including cartilage, bone, skin, and ligaments6,11, where it performs multiple functional roles2,4,12. CS and heparan sulfate (HS), are both glycosaminoglycan (GAG) chains attached to various PG core proteins through a common tetrasaccharide linker2,6. Most PG core proteins contain either CS or HS chains. The decision process for which polymer is extended involves selective recognition of flanking core protein sequence features for initiation of HS chains followed by addition of an α1,4-N-acetylglucosamine (GlcNAc) residue to the linker by EXTL313. By contrast, the committed step for CS synthesis is default and does not appear to be influenced by core protein features13. Initiation occurs through the addition of a β1,4-GalNAc residue to the linker by CSGALNACT1 or CSGALNACT2 (CSGALNACT1/2) (Fig. 1a). Following the CS committed step, the polymer is elongated by alternating addition of β1,4-GalNAc and β1,3-GlcA residues by four proteins: two chondroitin synthases (CHSY1 and CHSY3) and two chondroitin polymerizing factors (CHPF1 and CHPF2), resulting in CS chains as long as 100 monosaccharide units2,14–16. While the literature contains varied nomenclature for the CHSYs and CHPFs (Supplementary Table 1), we refer to each by the nomenclature above. Previous data demonstrated a lack of activity for the four individual proteins, but upon co-expression of CHPF and CHSY proteins, CS chains can be elongated2,13,14. However, the interactions between these four protein partners are not fully understood.Fig. 1CS synthase domain models, conserved motifs, expression, and activity.a Schematic of enzymes involved in CS biosynthesis. The linker region on proteoglycan core proteins is extended to a tetrasaccharide and then selectively initiated for CS synthesis by CSGALNACT1/2. The CS disaccharide backbone repeat is then elongated by four CHSY-CHPF complexes. b CHPF, CHSY, and CSGALNACT domain architecture. All CHSYs and CHPFs have an N-terminal transmembrane domain (TM), a disordered stem domain, a GT31-like domain (dark blue), a conserved helix (magenta), a cystatin-like domain (red), and a GT7-like domain (yellow). CSGALNACT1 and CSGALNACT2 are missing the GT31 domain. c Alignment of conserved GT-A fold motif elements shows the presence (green) or absence (red) of motif consensus sequences in the GT31 (dark blue) and GT7 (yellow) domains. CSGALNACT1/2 lack the N-term GT31 domain (dashes). d Relative secretion of single or co-transfected CHSYs or CHPFs as GFP fusion proteins detected by GFP fluorescence in the conditioned culture media after 5 days of expression (left axis, technical replicate n = 3) overlayed with CS co-polymerase activity for enzymes purified from transfections (right axis, technical replicate n = 3). Data are presented as mean values +/− SD. Assay input was normalized at 25 ng target protein for the CHSY-CHPF co-expression products and maximum protein input volume was used for the remaining 6 protein products. Source data are provided as a Source Data file. e Structural domains of CHSY3–CHPF1 from AlphaFold 3 ectodomain heterocomplex modeling (CHPF1, green and CHSY3, orange, center panel). When individual chains are pulled apart the respective domains are GT31 (dark blue), conserved helix (magenta), cystatin-like domain(red), and GT7 domain (yellow). The transmembrane and stem regions of CHPF1 (residues 1–107) and CHSY3 (residues 1–171) were removed from the molecular displays due to their low confidence AlphaFold3 modeling (low pLDDT scores). CS synthase domain models, conserved motifs, expression, and activity. a Schematic of enzymes involved in CS biosynthesis. The linker region on proteoglycan core proteins is extended to a tetrasaccharide and then selectively initiated for CS synthesis by CSGALNACT1/2. The CS disaccharide backbone repeat is then elongated by four CHSY-CHPF complexes. b CHPF, CHSY, and CSGALNACT domain architecture. All CHSYs and CHPFs have an N-terminal transmembrane domain (TM), a disordered stem domain, a GT31-like domain (dark blue), a conserved helix (magenta), a cystatin-like domain (red), and a GT7-like domain (yellow). CSGALNACT1 and CSGALNACT2 are missing the GT31 domain. c Alignment of conserved GT-A fold motif elements shows the presence (green) or absence (red) of motif consensus sequences in the GT31 (dark blue) and GT7 (yellow) domains. CSGALNACT1/2 lack the N-term GT31 domain (dashes). d Relative secretion of single or co-transfected CHSYs or CHPFs as GFP fusion proteins detected by GFP fluorescence in the conditioned culture media after 5 days of expression (left axis, technical replicate n = 3) overlayed with CS co-polymerase activity for enzymes purified from transfections (right axis, technical replicate n = 3). Data are presented as mean values +/− SD. Assay input was normalized at 25 ng target protein for the CHSY-CHPF co-expression products and maximum protein input volume was used for the remaining 6 protein products. Source data are provided as a Source Data file. e Structural domains of CHSY3–CHPF1 from AlphaFold 3 ectodomain heterocomplex modeling (CHPF1, green and CHSY3, orange, center panel). When individual chains are pulled apart the respective domains are GT31 (dark blue), conserved helix (magenta), cystatin-like domain(red), and GT7 domain (yellow). The transmembrane and stem regions of CHPF1 (residues 1–107) and CHSY3 (residues 1–171) were removed from the molecular displays due to their low confidence AlphaFold3 modeling (low pLDDT scores). The CHPFs and CHSYs are homologous protein isoforms anchored in Golgi membranes through an N-terminal transmembrane domain with their catalytic regions facing the lumen of the secretory compartment14. The lumenal portion of each isoform contains an N-terminal Carbohydrate Active Enzyme (CAZy17) glycosyltransferase (GT) family 31 (GT31)-like domain and a C-terminal GT7-like domain (Fig. 1b and Supplementary Fig. 1). In contrast, the CS initiating enzymes, CSGALNAC1/2, catalyze a single β1,4-GalNAc transfer and contain only a homologous GT7-like domain (Fig. 1b). Therefore, it is inferred that the corresponding CHSY/CHPF GT7-like domains catalyze β1,4-GalNAc transfer while the GT31-like domains synthesize the β1,3-GlcA linkages. No studies have demonstrated the roles of each of these domains, their overall structural features, or identified the roles of the additional sequences separating these two globular domains. Here, we explore the roles of each CHSY and CHPF domain in CS synthesis using modeled and empirical structural approaches, mutagenesis, activity assays, and mass spectrometry analysis of the enzymatic products. Our data demonstrate that both CHSY1 and CHSY3 contain active N-terminal GT31 β1,3-GlcA transferase and C-terminal GT7 β1,4-GalNAc transferase domains. In contrast, the GT-like domains in the CHPFs do not contribute to CS synthesis but stabilize the structures of the catalytic CHSYs. We also show that co-expression of a CHSY with a CHPF is required for effective CS synthase folding and function. Modeling of the four relevant CS synthase heterocomplexes reveals the structural basis for heterodimerization, substrate binding, and catalysis, including the role of cystatin-like domains sandwiched between the two GT domains. The structural models were then validated through cryo-EM structure determination of the most active complex, CHSY3–CHPF1, with bound divalent cations and UDP. Additional mutational analysis of the bridging cystatin-like domains indicates they contribute to efficient polymer synthesis, possibly through interaction with the extending CS chain. Thus, our work demonstrates the structural basis for CS polymer backbone extension by a collection of heterodimeric CS synthase complexes essential for PG biosynthesis in animal cells. Results Domain structure of CS biosynthetic proteins The alternating disaccharide repeat units of CS chains are extended on PG core proteins by a collection of four homologous gene products: CHSY1, CHSY3, CHPF1 and CHPF2 (Fig. 1a, b, Supplementary Fig. 1). Each of the four proteins has a similar domain architecture, with an N-terminal transmembrane segment, putative disordered stem region, a GT31-like domain, an extended helix linker, a cystatin-like domain, and a C-terminal GT7-like domain (Fig. 1b, e and Supplementary Fig. 1). By contrast, the two CS priming enzymes, CSGALNACT1/2, exhibit only β1,4-GalNAc transfer activity and are missing the equivalent GT31 domain (Fig. 1b and Supplementary Fig. 2a). The cystatin-like linker domains show significant structural similarity to the secreted thiol proteinase inhibitors, Cystatin C and M, that have both been shown to bind sulfated HS chains both in vitro and in vivo18–20 through conserved basic patches18,21,22 (Fig. 1b, e). Both the GT31 and GT7 domains of the CS synthetic enzymes are comprised of GT-A Rossmann-like folds17,23 that commonly contain canonical motifs involved in divalent cation coordination (DxD and C-term His), sugar donor binding (G-loop), and catalysis (xED)24,25. While both the GT7 domains of CHSY1 and CHSY3, as well as the GT7 domains of CSGALNACT1/2 contain the respective conserved motif sequences (Fig. 1c), the corresponding GT7 domains of CHPF1 and CHPF2 are less conserved in sequence and lack the canonical motif elements (Fig. 1c). Similarly, the GT31 domains of CHSY1 and CHSY3 retain the conserved GT-A motif sequences, while the corresponding GT31 domains of CHPF1 and CHPF2 again lack the conserved motif elements (Fig. 1c) suggesting that both GT-like domains in the two CHPFs are likely inactive. Expression of soluble ectodomains Prior studies suggested that expression of single CHSY or CHPF proteins leads to ineffective CS synthase activity, while co-expression can result in functional production of CS synthases2–6,13,14. We generated truncated CHSY and CHPF ectodomains by replacing their N-terminal transmembrane segments with an N-terminal signal sequence, 8xHis, and GFP fusion tag cassette26, and tested expression and secretion in mammalian cells. Expression of each single protein or co-expression of CHPF1-CHPF2 or CHSY1-CHSY3 led to no significant secretion (Fig. 1d). Only when one CHSY and one CHPF were co-expressed was appreciable secretion of the GFP fusion proteins achieved (Fig. 1d) that could also be imaged by GFP fluorescence detection following SDS-PAGE (Supplementary Fig. 2b). The fluorescence detection revealed equal abundance of CHSY3 with either CHPF1 or CHPF2 suggesting the secretion of a stable heterocomplex in the co-transfected cells. SDS-PAGE was unable to resolve CHSY1 from CHPF1 or CHPF2 on the gel, but the obligate co-expression of these proteins suggests a similar requirement of heterocomplex formation for secretion. CS co-polymerase activity was detected for each of the CHSY-CHPF heterocomplexes, but no secreted products or enzyme activities were observed for single protein expressions or the CHPF1-CHPF2 or CHSY1-CHSY3 co-transfections (Fig. 1d). Significantly higher secretion and activity were observed for heterocomplexes containing CHPF1 compared to those employing CHPF2 suggesting the former protein supported more efficient folding and/or secretion. Likewise, CHSY3 pairings appear to be more active than their CHSY1 counterparts (Fig. 1d) suggesting either higher activity or greater stability for the CHSY3-containing complexes13. Modeling CS synthase complexes We also tested the modeling of various homodimeric and heterodimeric complexes of all CHSYs and CHPFs using AlphaFold 327 (AF3, Fig. 1e, Supplementary Fig. 3). Modeling of homodimeric complexes or CHPF1-CHPF2 or CHSY1-CHSY3 heterocomplexes led to structures with numerous low confidence regions, especially at the interface loops between subunits (Supplementary Fig. 16). By comparison, the four CHSY-CHPF heterodimeric models were strikingly similar (average pairwise RMSD: 0.93 Å, Supplementary Fig. 3c), contained more ordered, high-confidence loop regions, and parallel dimerization interfaces with pseudo-2-fold axes of symmetry. The interfaces were comprised of antiparallel coiled coils for the bridging helices and dimerization of the two cystatin domains28 (Fig. 1e, Supplementary Figs. 3a, d). The arrangement placed each GT31-like and GT7-like domain pair in adjacent positions in the heterocomplex, but each had minimal direct contributions to the dimer interface. The parallel heterodimeric assembly also placed the respective CHSY GT31 and GT7 active sites on the same face of the heterodimer ~63 Å apart and each active site extended toward the central CHSY cystatin-like domain. We compared structural confidence metrics for the models with a set of known homo- or heterodimeric proteins versus known monomeric proteins (incorrectly modeled in AF3 as homodimers) to assess the protein complexes. Confidence score metrics of the AF3 models for residues within 4 Å in the heterocomplex interface, PISA P-values, ratios of interface area to total surface area, and PISA ΔG values were assessed to establish threshold cutoffs for evaluating the dimer interfaces (Supplementary Fig. 3b and Supplementary Table 5). Each of the CHSY-CHPF complexes conformed with these metric thresholds, while the other homo- and heterodimeric complexes fell short in at least one threshold metric. These confidence score assessments were consistent with protein expression, secretion, and enzyme activity of the respective complexes. In addition, when surface electrostatics of the interface residues were visualized, complementary basic patches in the CHSYs aligned closely with corresponding acidic patches in the CHPFs (Supplementary Fig. 3d) supporting the models for obligate heterodimer assembly. Cryo-EM structure of CHSY3–CHPF1 complex To validate the AF3 models we chose the CHSY3–CHPF1 complex for structural analysis because of its higher expression and co-polymerase activity. The purified heterocomplex was subjected to TEV protease cleavage to remove the GFP fusion tag yielding a ~ 203 kDa ectodomain heterocomplex consistent with the predicted size of 181 kDa plus six predicted N-glycans (Supplementary Fig. 4). The resulting cryo-EM structure in the presence of UDP and Mn2+ (Fig. 2, Supplementary Table 4, and Supplementary Figs. 5 and 6, average resolution 3.42 Å) was in close agreement with the AF3 model across all domains (RMSD 2.04 Å) with greater structural divergence in loop regions extending into solvent (Supplementary Fig. 7). A limited number of loops were not resolved in the experimental structure compared with the AF3 model (Supplementary Fig. 7). Higher similarity between the model and experimental structure was observed within the CHSY catalytic domains (RMSD 0.96 Å and 1.04 Å for GT31 and GT7 domains, respectively). Densities for UDP and Mn2+ were both observed in the CHSY3 GT31 domain active site adjacent to the DxD motifs, while Mn2+ and weak density for UDP was found in the CHSY3 GT7 domain (Fig. 2 and Supplementary Fig. 8). No detectable ions or ligands were found associated with the corresponding CHPF domains. Thus, the close agreement between structural data and AF3 models for CHSY3–CHPF1 suggests the other three CHSY-CHPF AF3 models likely reflect plausible heterocomplexes.Fig. 2Cryo-EM Structure of CHSY3–CHPF1.a Cryo-EM reconstruction map of the CHSY3(orange)–CHPF1(green) complex bound to UDP (gray) and Mn2+. b Different front views of the CHSY3(orange)–CHPF1(green) structure. c EM density around the UDP molecule is shown as a gray mesh (map contour level: 0.295, as defined by ChimeraX). Mn2+ is shown in purple. Cryo-EM Structure of CHSY3–CHPF1. a Cryo-EM reconstruction map of the CHSY3(orange)–CHPF1(green) complex bound to UDP (gray) and Mn2+. b Different front views of the CHSY3(orange)–CHPF1(green) structure. c EM density around the UDP molecule is shown as a gray mesh (map contour level: 0.295, as defined by ChimeraX). Mn2+ is shown in purple. Enzymatic analysis of GT31 and GT7 catalytic mutants To assess contributions of individual GT31 and GT7 domains in catalysis we performed enzyme assays using recombinant GFP-tagged bikunin PG as acceptor and both sugar donors in UDP-Glo assays. CS synthase activity for all four of the CHSY-CHPF heterocomplexes clearly demonstrated extended sugar transfer using bikunin as acceptor (Fig. 1d). CS extension was also assessed by SDS-PAGE gel shift for the GFP-tagged bikunin acceptor following sugar transfer (Supplementary Fig. 9b, d, f, h). The recombinant bikunin acceptor contains both protein without and with CS chains. The latter forms appear as a high molecular weight smear above the sharp band of the unmodified protein. Incubation with CHSY-CHPF complexes resulted in a significant upward shift in gel mobility for the CS-modified bikunin substrate only in co-polymerase reactions containing both sugar donors. The gel shift of only a subfraction of the GFP-tagged substrate on the gel likely results from the heterogenous sulfated termini on the recombinant bikunin CS chains where only a minor subset of CS chains may be capable of extension. As a control, incubation of the extension products with chondroitinase ABC led to elimination of both the initial CS-modified bikunin smear as well as the elongated CS smear on bikunin resulting from CHSY3–CHPF1 co-polymerase activity (Supplementary Fig. 9i). Use of a pNP-tagged 6-mer CS oligomer as acceptor (GalNAc at non-reducing terminus) in enzyme assays only resulted in a significant single sugar transfer when UDP-GalA was used as donor (Supplementary Fig. 9a, c, e, g). Similarly, use of a pNP-tagged 7-mer CS oligomer acceptor (GlcA terminal residue) only acted as an effective acceptor when UDP-GalNAc was donor. MALDI-TOF MS confirmed the selectivity of extension for the 6-mer and 7-mer acceptors with their corresponding UDP-GlcA and UDP-GalNAc donors (Supplementary Fig. 10). Use of both donors in the reactions led to co-polymer extension products detected MALDI analysis (Supplementary Fig. 11). Of note was the predominant appearance of CS extension products containing GalNAc termini, while the products of the intermediate GlcA extension were barely detectable. These data are consistent with the higher kcat/KM values for GalNAc transfer over GlcA transfer by CHSY3–CHPF1 (Supplementary Table 2) and suggest GlcA addition is rate limiting in the co-polymerase assays. We then tested the roles of each CHSY catalytic domain by generation of active site mutants in the canonical xED and DxD motifs for the respective GT31 and GT7 domains (Supplementary Fig. 12a, b). Wild type or catalytic site mutants in CHSY1 or CHSY3 were paired with wild type CHPF1 or CHPF2 for co-expression and activities were assessed using UDP-Glo assays (Fig. 3). The four wild type heterocomplexes exhibited full co-polymerase activity with bikunin as acceptor, but mutation of either the GT31 or GT7 catalytic domains resulted in a reduction in catalytic activity equivalent to single sugar transfer (Fig. 3a). Assays of the mutants with bikunin (Fig. 3b, c), 6-mer, or 7-mer (Fig. 3d, e) acceptors and single sugar nucleotide donors demonstrated that mutations in the GT31 domain eliminated GlcA transfer activity but not GalNAc transfer activity, while mutations in the GT7 domain eliminated GalNAc transfer activity but not GlcA transfer activity. Reaction products from both the 6-mer and 7-mer transfer reactions were further analyzed by MALDI-TOF to confirm loss of sugar transfer in the respective catalytic mutants (Supplementary Fig. 13). These data confirm that the CHSY GT31 and GT7 domains are solely responsible for GlcA and GalNAc transfer, respectively, and the corresponding CHPF partner domains do not contribute to sugar transfer.Fig. 3Enzymatic Activity of Domain Specific Mutants.Enzymatic activity of WT or catalytic mutants of CHSY-CHPF complexes using either the CS proteoglycan bikunin (a–c) or CS oligomers (d-e) as acceptors and either individual UDP-GlcA (b and d), UDP-GalNAc (c and e), or both donor sugars (a) as donors. a Co-polymerase activity of all four CS complexes using bikunin as acceptor. Relative activity is shown for wild-type CHSY versus CHSY mutants in the respective GT31 or GT7 xED and DxD motifs. Color coding of histogram relative activity data for the respective CHSY–CHPF complexes is shown on the upper right of the panel. b Single sugar transfer using UDP-GlcA as donor and bikunin as acceptor in wild type and catalytic mutants. c Single sugar transfer using UDP-GalNAc as donor and bikunin as acceptor with wild-type and catalytic mutants of CHSY-CHPF complexes. d Single sugar transfer using UDP-GlcA as donor and CS oligo 6-mer oligomer (non-reducing terminal GalNAc residue) acceptor in wild type and catalytic mutants of CHSY-CHPF complexes. e Single sugar transfer using UDP-GalNAc as donor and CS oligo 7-mer oligomer (non-reducing terminal GlcA residue) in wild type and catalytic mutants of CHSY–CHPF complexes. All samples were assayed in technical replicate (n = 4). Data are presented as mean values +/− SD. Source data are provided as a Source Data file. Enzymatic Activity of Domain Specific Mutants. Enzymatic activity of WT or catalytic mutants of CHSY-CHPF complexes using either the CS proteoglycan bikunin (a–c) or CS oligomers (d-e) as acceptors and either individual UDP-GlcA (b and d), UDP-GalNAc (c and e), or both donor sugars (a) as donors. a Co-polymerase activity of all four CS complexes using bikunin as acceptor. Relative activity is shown for wild-type CHSY versus CHSY mutants in the respective GT31 or GT7 xED and DxD motifs. Color coding of histogram relative activity data for the respective CHSY–CHPF complexes is shown on the upper right of the panel. b Single sugar transfer using UDP-GlcA as donor and bikunin as acceptor in wild type and catalytic mutants. c Single sugar transfer using UDP-GalNAc as donor and bikunin as acceptor with wild-type and catalytic mutants of CHSY-CHPF complexes. d Single sugar transfer using UDP-GlcA as donor and CS oligo 6-mer oligomer (non-reducing terminal GalNAc residue) acceptor in wild type and catalytic mutants of CHSY-CHPF complexes. e Single sugar transfer using UDP-GalNAc as donor and CS oligo 7-mer oligomer (non-reducing terminal GlcA residue) in wild type and catalytic mutants of CHSY–CHPF complexes. All samples were assayed in technical replicate (n = 4). Data are presented as mean values +/− SD. Source data are provided as a Source Data file. Modeling of substrate-bound complexes in CS synthase active sites We then employed the recently released open-source version of AlphaFold 327 for modeling of all four CHSY-CHPF substrate complexes using only the CHSY and CHPF ectodomain sequences, UDP-GalNAc and UDP-GlcA donors, two Mn2+ ions, and the respective 6-mer and 7-mer CS oligomer acceptors as input (Fig. 4a, Supplementary Fig. 14g–i). The resulting models positioned each donor, acceptor, and ion in the respective active sites as predicted for the specificities of the GT31 and GT7 domains described above. In the GT31 domain active sites of CHSY1 or CHSY3, the Mn2+ and UDP-GlcA donor interacted with the DxD motif and other GT-A fold motif elements similar to the empirical CHSY3:CHPF1:UDP:Mn2+ heterocomplex structure as anticipated for β-GlcA transfer (Fig. 4b, d and Supplementary Fig. 14a–c) and analogous to other known GT31 enzyme-substrate complexes29–32. Similarly, in the GT7 domain active sites for both CHSY1 and CHSY3, the Mn2+ and UDP-GalNAc donor bound with the DxD motif and other GT-A fold motif elements (Fig. 4c, e and Supplementary Fig. 14d–f) analogous to other known GT7 enzyme structures33–35.Fig. 4AlphaFold 3 model of CHSY3–CHPF1 with donors and acceptor ligands.AlphaFold 3 model of CHSY3–CHPF1 complex with bound sugar nucleotides (UDP-GlcA and UDP-GalNAc), bound acceptor CS oligosaccharides (CS 6-mer with a terminal GalNAc and CS 7-mer with a terminal GlcA), and MnCl2 divalent cations. a The CHSY3–CHPF1 complex (CHSY3: orange, CHPF1: green) highlighting the occupancy of substrates and divalent cations only in CHSY3. The canonical DxD and xED motifs are shown as yellow sticks. No substrates could be successfully modeled bound to CHPF1. b Zoom-in of the N-terminal GT31 domain of CHSY3. AlphaFold 3 model of the bound substrate complex of UDP-GlcA (cyan), the CS 6-mer (non-reducing terminal GalNAc, purple), and MnCl2 (red sphere) in the active site consistent with canonical interactions with the GT-A fold motif structures. DxD and xED motifs are shown as yellow sticks. c Zoom-in of the C-terminal GT7 domain of CHSY3. AlphaFold 3 model of the bound substrate complex of UDP-GalNAc (magenta), the CS 7-mer (non-reducing terminal GlcA, dark blue), and MnCl2 (red sphere) in the active site consistent with the GT-A fold motif structures. DxD and xED motifs are shown as yellow sticks. d Proposed SN2 inverting catalytic mechanism for β1,3-GlcA transfer to the terminal CS GalNAc residue the GT31 domain. Distances are indicated between the GalNAc O3 hydroxyl and the D342 catalytic base in the xED motif or the C1 of the UDP GlcA donor. e Proposed SN2 inverting catalytic mechanism for β1,4-GalNAc transfer to the terminal CS GlcA residue in the GT7 domain and distances between substrates. Distances are indicated between the GlcA O4 hydroxyl and the D809 catalytic base in the xED motif or the C1 of the UDP-GalNAc donor. The transmembrane and stem regions of CHPF1 (residues 1–107) and CHSY3 (residues 172–882) were removed from the molecular displays due to their low confidence AlphaFold3 modeling (low pLDDT scores). AlphaFold 3 model of CHSY3–CHPF1 with donors and acceptor ligands. AlphaFold 3 model of CHSY3–CHPF1 complex with bound sugar nucleotides (UDP-GlcA and UDP-GalNAc), bound acceptor CS oligosaccharides (CS 6-mer with a terminal GalNAc and CS 7-mer with a terminal GlcA), and MnCl2 divalent cations. a The CHSY3–CHPF1 complex (CHSY3: orange, CHPF1: green) highlighting the occupancy of substrates and divalent cations only in CHSY3. The canonical DxD and xED motifs are shown as yellow sticks. No substrates could be successfully modeled bound to CHPF1. b Zoom-in of the N-terminal GT31 domain of CHSY3. AlphaFold 3 model of the bound substrate complex of UDP-GlcA (cyan), the CS 6-mer (non-reducing terminal GalNAc, purple), and MnCl2 (red sphere) in the active site consistent with canonical interactions with the GT-A fold motif structures. DxD and xED motifs are shown as yellow sticks. c Zoom-in of the C-terminal GT7 domain of CHSY3. AlphaFold 3 model of the bound substrate complex of UDP-GalNAc (magenta), the CS 7-mer (non-reducing terminal GlcA, dark blue), and MnCl2 (red sphere) in the active site consistent with the GT-A fold motif structures. DxD and xED motifs are shown as yellow sticks. d Proposed SN2 inverting catalytic mechanism for β1,3-GlcA transfer to the terminal CS GalNAc residue the GT31 domain. Distances are indicated between the GalNAc O3 hydroxyl and the D342 catalytic base in the xED motif or the C1 of the UDP GlcA donor. e Proposed SN2 inverting catalytic mechanism for β1,4-GalNAc transfer to the terminal CS GlcA residue in the GT7 domain and distances between substrates. Distances are indicated between the GlcA O4 hydroxyl and the D809 catalytic base in the xED motif or the C1 of the UDP-GalNAc donor. The transmembrane and stem regions of CHPF1 (residues 1–107) and CHSY3 (residues 172–882) were removed from the molecular displays due to their low confidence AlphaFold3 modeling (low pLDDT scores). Strikingly, the 6-mer and 7-mer CS oligomer acceptors were also bound in the GT31 and GT7 active sites, respectively, for both CHSY3 or CHSY1 heterocomplexes. The positions of the 6-mer and UDP-GalA in the GT31 active site placed the acceptor GalNAc C-3 oxygen between the catalytic base in the xED motif (2.5 Å from D341 in CHSY3, Fig. 4d) and 3.4 Å from C1 of the GalA in the sugar donor (Fig. 4d). Similarly, the 7-mer CS oligomer acceptor and UDP-GalNAc were bound in the respective CHSY GT7 active sites to place the 4-OH of the terminal GlcA acceptor between the catalytic base in the xED motif (2.5 Å from D807 in CHSY3, Fig. 4e) and 3.4 Å from C1 of the GalNAc in the sugar donor (Fig. 4e). Thus, modeled substrate interactions in both CHSY active sites resemble a Michaelis complexes for the respective inverting SN2 catalytic mechanisms. In both domains, the CS oligosaccharide acceptors extended out of the GT31 and GT7 active site clefts with additional hydrogen bonding and hydrophobic interactions toward the respective CHSY cystatin domain (Supplementary Fig. 14a–f). The residues and positions of substrate interactions with sugar donors and divalent cations were virtually identical in all four CHSY1 and CHSY3 heterocomplexes and interactions with the terminal two monosaccharides of the glycan acceptors were also virtually identical in all complexes (Fig. 4a, Supplementary Fig. 14g–i, and Supplementary Table 3). Divergence in acceptor interactions among the CHSY-CHPF complexes was only seen after the third acceptor residue in the respective complexes. No bound substrates or ion complexes were observed for the CHPF chains in any of the models. Role of the cystatin-like domain Proteins containing cystatin domains have been shown to bind HS chains through three potential conserved basic patches18,19. Two analogous basic patches were identified in the CHSY and CHPF cystatin-like domains, termed cluster 1 and cluster 3 by analogy to equivalent clusters in cystatin C19 (Fig. 5). To test the roles of the basic clusters in CS polymerization, we generated sets of alanine mutations in each cluster (four each in cystatin-like cluster 1 of CHSY3 and CHPF1 and nine versus seven residues in cluster 3 of CHSY3 and CHPF1, respectively) and co-expressed each mutant with its wild type partner protein (Fig. 5, Supplementary Fig. 12c,d). Each cluster mutant was then tested for CS co-polymerization activity using the physiologically relevant acceptor substrate bikunin. While the alanine mutations in each cluster did not reduce the overall expression or secretion of the resulting heterocomplex (Supplementary Fig. 12d), mutations in cluster 3 of either CHSY3 or CHPF1 significantly reduced co-polymerase activity (Fig. 5). Mutations in cluster 1 in CHSY3 also led to a ~ 70% reduction in copolymerase activity on the bikunin acceptor, while mutations in cluster 1 in CHPF1 had a minimal impact in enzyme activity. By contrast to the significant impact of cystatin mutants on co-polymerase activity using a large proteoglycan acceptor, co-polymerase assays on the cluster 3 CHSY3 mutant using 6-mer or 7-mer CS oligos did not reduce catalytic efficiency using these short CS acceptors (Supplementary Table 2 and Supplementary Fig. 15). The kcat/Km values were surprisingly enhanced in the cystatin mutants indicating a complex contribution of the cystatin domains in CS backbone synthesis.Fig. 5Role of cystatin domains in CS extension.Mutations were made in the two cystatin-like basic binding clusters of CHPF1 and CHSY3 as described in Supplementary Fig. 12. Each cystatin-like domain mutant (clusters 1 or cluster 3) in CHSY3 or CHPF1 was co-expressed with the corresponding wild type partner subunit and tested for impact on catalytic activity. a CS co-polymerization assays were performed on the CHSY3–CHPF1 heterocomplexes harboring cystatin-like cluster mutations using bikunin as acceptor and both UDP-GalNAc and UDP-GlcA as donors. All assays were performed in technical replicate (n = 4). Data are presented as mean values +/− SD. b AlphaFold 3 model highlighting the cystatin domains in the CHSY3–CHPF1 complex and the positions of the modeled CS 6-mer and 7-mer complexes. Electrostatic surface representations of the binding clusters of both CHPF1 and CHSY3 are shown in the lower right. The four binding clusters are comprised of collection of basic residues (blue). Mutating these residues to alanine alters the electrostatic characteristics and CS binding in these regions of the cystatin-like domain (Supplementary Fig. 12). Role of cystatin domains in CS extension. Mutations were made in the two cystatin-like basic binding clusters of CHPF1 and CHSY3 as described in Supplementary Fig. 12. Each cystatin-like domain mutant (clusters 1 or cluster 3) in CHSY3 or CHPF1 was co-expressed with the corresponding wild type partner subunit and tested for impact on catalytic activity. a CS co-polymerization assays were performed on the CHSY3–CHPF1 heterocomplexes harboring cystatin-like cluster mutations using bikunin as acceptor and both UDP-GalNAc and UDP-GlcA as donors. All assays were performed in technical replicate (n = 4). Data are presented as mean values +/− SD. b AlphaFold 3 model highlighting the cystatin domains in the CHSY3–CHPF1 complex and the positions of the modeled CS 6-mer and 7-mer complexes. Electrostatic surface representations of the binding clusters of both CHPF1 and CHSY3 are shown in the lower right. The four binding clusters are comprised of collection of basic residues (blue). Mutating these residues to alanine alters the electrostatic characteristics and CS binding in these regions of the cystatin-like domain (Supplementary Fig. 12). Discussion While the individual CHSY and CHPF genes involved in CS backbone synthesis were identified over two decades ago14,36–39, their individual and combined roles in GAG polymer synthesis have remained enigmatic. Expression of individual proteins resulted in no detectable activity, while co-expression data suggested a requirement for heterocomplex formation for CS synthase activity. By analogy to our prior work on the obligate heterodimeric assembly of EXT1-EXT2 as an HS synthase complex40, we systematically tested single and co-expression cultures of CHSY1, CHSY3, CHPF1, and CHPF2 and demonstrated that CS synthase activity requires the formation of heterodimeric complexes of a CHSY and CHPF. Comprehensive AF3 analysis also revealed that, among all permutations of homo- and heterodimeric complexes, only CHSY-CHPF heterocomplexes led to modeled structures with high confidence interface metrics. The four CHSY-CHPF models were highly similar in structure and in close agreement with our cryo-EM structure of the CHSY3–CHPF1:UDP:Mn2+ complex to provide strong support for the predicted models. The roles of individual domains were probed by mutagenesis and detailed enzymatic analysis. The combined data unequivocally demonstrated the GT31 domains of the CHSYs catalyze β1,3-GlcA transfer, while the GT7 domains transfer β1,4-GalNAc residues to the growing CS chain. The respective CHPF domains do not retain canonical GT-A fold features required for catalysis and do not contribute to catalysis. The extensive asymmetric and electrostatically complementary CHSY-CHPF interface is formed largely through coiled-coil interactions of the extended helixes on opposing monomers and dimerization of the cystatin-like domains analogous to dimeric interactions in other cystatins28,41. Thus, the lack of direct contribution of CHPFs to CS extension raises a question regarding their roles in the respective heterocomplexes. Analogous loss of function also occurs in the multi-domain HS synthase complex40. EXT1 and EXT2 each contain GT47-like and GT64-like domains and form a heterocomplex with four potential active sites. However, only the N-terminal EXT1 GT47 β1,4-GlcA transferase domain and the EXT2 GT64 α1,4-GlcNAc transferase domain are functional, while the corresponding EXT1 GT64 and EXT2 GT47 domains have evolved to be inactive. Similarly, EXTL3 catalyzes the committed step in HS synthesis and contains GT47-like and GT64 domains similar to EXT1 and EXT2. However, EXTL3 is a homodimeric enzyme13 and only the GT64 domain remains active while the GT47 domain has evolved to be non-functional. By comparison, the committed step in CS synthesis is catalyzed by CSGALNACT1/2, and these latter enzymes retain the conserved helix, cystatin domain, and functional GT7 domain analogous to the CHSYs/CHPFs, but the GT31-like domain of the CHSYs/CHPFs is not present. The bacterial CS synthase (e.g. E. coli K4CP42) is a more simplified monomeric enzyme with two closely adjoining functional domains that catalyze β1,3-GlcA and β1,4-GalNAc transfer, but without a discernible linker region. Thus, GAG synthases in higher metazoans appear to have evolved as multidomain heterocomplexes for disaccharide repeat synthesis through domain fusion, gene duplication, heterocomplex formation, and diversification of function. Presumably, this specialization provides enzyme stabilization and control of synthesis for diverse PG structures in different adult tissues or during development. In the case of CHSY-CHPF complexes, the CHPFs appear to play obligatory roles in folding and function of the CHSYs through their asymmetric and complementary interactions at the heterodimer interface. The inability to model homodimeric complexes with reasonable interface metrics and the lack of secretion for single CHSY or CHPF expression products in transfected cells strongly suggest that functional CHSY-CHSY or CHPF-CHPF homodimeric complexes are unlikely to be produced in vivo. However, we cannot rule out the possibility that additional cellular or regulatory factors, including association with other CS biosynthetic components, could contribute to the formation of functional single chains or homodimeric complexes for the full-length proteins in vivo. While stabilization of the catalytic CHSYs is a plausible role for the CHPFs, additional contributions of the CHPFs to structural or allosteric dynamics within the heterocomplex cannot be ruled out and will require further study To model details of substrate interactions in the CHSY-CHPF complexes, we leveraged the recent advancements in biomolecular modeling using open-source AF3 with expanded capabilities for modeling complex protein-ligand structures. The resulting structural models paired both donors, acceptors, and divalent cations in their respective active sites in positions for catalysis, despite providing no additional information for domain substrate specificity. Substrate interactions also resulted in plausible Michaelis complex models for the SN2 inverting catalytic mechanisms for both CHSY active sites. In agreement with our empirical CHSY3–CHPF1 structure, the models also support experimental data suggesting that the CHPFs do not bind substrates or contribute to catalysis. While these models for substrate interactions will require further experimental validation through mutagenesis and empirical structure determination of substrate complexes, they are fully consistent with known modes of substrate interactions for other GT31 and GT7 enzymes and provide details for understanding CS catalysis (Fig. 6).Fig. 6Proposed model for CS extension by CHSY-CHPF complexes.CS synthase (CHSY–CHPF) complexes are membrane-anchored enzymes located in the Golgi apparatus, with their catalytic domains projecting into the Golgi lumen. These complexes are formed through the heterodimerization of CHSY and CHPF monomers. Each monomer contains a GT31-like domain, a GT7-like domain, and a bridging linker region composed of an extended helix and a cystatin-like domain, which facilitates subunit heterodimerization. Extension of CS chains on proteoglycan core proteins begins after CSGALNACT1/2 enzymes add a β1,4-GalNAc to the tetrasaccharide linker. This is followed by alternating additions of β1,3-GlcA and β1,4-GalNAc catalyzed by the CHSY-CHPF heterocomplexes. The GT31 domain of the CHSY subunit catalyzes GlcA transfer, while its GT7 domain mediates GalNAc transfer. The associated CHPF subunit does not contribute to catalysis but instead stabilizes the heterocomplex and the catalytic domains of the CHSY subunit. During chain elongation, the growing CS chain alternates between the GT31 and GT7 active sites through a dissociative mechanism. Meanwhile, the central cystatin-like domains of CHSY and CHPF may contribute to catalysis by interacting with the CS chain, potentially enhancing substrate positioning or improving processivity during chain extension. Proposed model for CS extension by CHSY-CHPF complexes. CS synthase (CHSY–CHPF) complexes are membrane-anchored enzymes located in the Golgi apparatus, with their catalytic domains projecting into the Golgi lumen. These complexes are formed through the heterodimerization of CHSY and CHPF monomers. Each monomer contains a GT31-like domain, a GT7-like domain, and a bridging linker region composed of an extended helix and a cystatin-like domain, which facilitates subunit heterodimerization. Extension of CS chains on proteoglycan core proteins begins after CSGALNACT1/2 enzymes add a β1,4-GalNAc to the tetrasaccharide linker. This is followed by alternating additions of β1,3-GlcA and β1,4-GalNAc catalyzed by the CHSY-CHPF heterocomplexes. The GT31 domain of the CHSY subunit catalyzes GlcA transfer, while its GT7 domain mediates GalNAc transfer. The associated CHPF subunit does not contribute to catalysis but instead stabilizes the heterocomplex and the catalytic domains of the CHSY subunit. During chain elongation, the growing CS chain alternates between the GT31 and GT7 active sites through a dissociative mechanism. Meanwhile, the central cystatin-like domains of CHSY and CHPF may contribute to catalysis by interacting with the CS chain, potentially enhancing substrate positioning or improving processivity during chain extension. The substrate-bound models also indicate that the β1,3-GlcA and β1,4-GalNAc transferase active sites are 63 Å apart. This suggests a dissociative enzymatic mechanism where the acceptor alternates between active sites during CS chain elongation. Interestingly, both active sites show acceptor interactions that extend toward the central cystatin-like domain between the respective catalytic domains. The structural similarity of these domains to secreted cystatin thiol proteinase inhibitors known to bind sulfated heparan sulfate chains18–20 suggests a potential role in substrate recognition or positioning. Cystatin domains have previously been shown to interact with HS chains through three characteristic basic patches19 and two of those equivalent patches are found in the CHSY and CHPF cystatin-like domains. Mutagenesis of these basic patches, and particularly cystatin-like cluster 3 of CHSY3 and CHPF1, led to significant loss of CS co-polymerase activity using large proteoglycan acceptors, but not short CS oligo acceptors. It is tempting to hypothesize that extended CS chain or core protein interactions with the CHSY3 cystatin cluster 3 region could anchor the polymer as it alternates between the two active sites to provide more efficient polymer synthesis. However, the cluster 3 cystatin-like domain of CHPF1 is on the opposite side of the heterocomplex and mutations in that cystatin cluster also lead to equivalent enzyme inactivation when large proteoglycan acceptors are employed. Cystatin-like domains are also found in other proteins involved in PG synthesis, including the CS committed step enzymes CSGALNACT1, CSGALNACT2 (Fig. 1), and PG linker initiation enzymes, XYLT143, and XYLT2. Thus, the roles of these domains in PG synthesis are more complex and will require further study to unravel their contributions in linker synthesis and CS chain elongation. Human defects in CS synthase activity are relatively rare and the Human Gene Mutation Database44 notes only ten known missense/nonsense mutations in CHSY1, most leading to Temtamy preaxial brachydactyly syndrome (TPBS)45–48. Zebrafish models of antisense knockdown for chsy1 resulted in defects similar to human patients with TPBS. Since similar phenotypes in zebrafish were found after chsy1 inactivation or overexpression, the authors suggested that brachydactyly could be linked to either to loss or gain of BMP signaling45. Human mutations in CHSY3, CHPF1, or CHPF2 have not been described and this paucity of known genetic defects contrasts with over 100 human mutations in EXT1 or EXT2 that impact HS synthesis40. The potential for redundancy resulting from two functional CHSYs combined with two CHPFs in the formation of four putative CS synthase complexes could lead to masking of mutations that inactivate the individual enzyme subunits. While this study provides a comprehensive structural and functional characterization of CS synthase complexes, several questions remain. Future research will need to explore how these complexes interact with other enzymes involved in CS modification, the precise mechanisms of substrate recognition and processivity during CS chain elongation, including the roles of the cystatin domains, and how the activity of these complexes is regulated in vivo. Additionally, the structural insights gained from this study may be leveraged to develop targeted therapies for diseases associated with CS dysregulation, including the roles of CS in inhibition of axon regeneration after injury10. Our findings provide a structural and mechanistic framework for understanding CS biosynthesis, opening additional avenues for investigating the roles of CS in development, disease, and potential therapeutic interventions. Methods Expression and purification of CHSY and CHPF complexes Expression constructs encoding human CHPF1 (residues 35-775, Uniprot ID: Q8IZ52), CHPF2 (residues 37-772, Uniprot ID: Q9P2E5), CHSY1 (residues 32-802, Uniprot ID: Q86X52), CHSY3 (residues 32-882, Uniprot ID: Q70JA7) and all mutants were generated as N-terminal transmembrane truncations by gene synthesis (Twist Bioscience, South San Francisco, CA) in the pGEn2 expression vector. These expression constructs generate domain fusions encoding an N-terminal signal sequence, His8, AviTag, superfolder GFP, and TEV protease recognition site, followed by the CHSY or CHPF coding region as described in previous work26. In larger-scale expression of the CHSY3–CHPF1 complex in the pGEn2 vector, we were unable to achieve efficient TEV cleavage of the tag sequence. As a result, an additional 2x(GSSSS) linker was inserted between the TEV recognition site and the corresponding CHSY3 and CHPF1 coding sequences. In addition, a puromycin resistance selection sequence was added to the vector backbone (pGEn17 vector) to allow selection of stable transformants in mammalian cells. Mutants in the CHSY3 or CHPF1 coding region were generated by gene synthesis (Twist Bioscience, South San Francisco, CA) in the pGEn2 vector. Each mutant CHSY construct was transiently co-expressed with CHPF1 or CHPF2 partner constructs as described above for wild type constructs. Eight mutant soluble secreted protein heterocomplexes were generated (CHSY1 D171A (xED GT31), CHSY1 D252A (DxD GT31), CHSY1 D631A (xED GT7), CHSY1- D720A (DxD GT7), CHSY3 D261A (xED GT31), CHSY3 D342A (DxD GT31), CHSY3 D720A (xED GT7), CHSY3 D809A (DxD GT7) and each was co-expressed with either CHPF1 or CHPF2. Single or co-expression of wild type and mutant constructs in the pGEn2 expression vector (CHPF1, CHPF2, CHSY1, CHSY3, CHPF1 + CHSY1, CHPF1 + CHSY3, CHPF2 + CHSY1, CHPF2 + CHSY1, CHSY1 + CHSY3, CHPF1 + CHPF2, and all mutant pairings of the CHSYs and CHPFs) were performed by transient transfection in HEK293-F cells (FreeStyle 293-F cells, Thermo Fisher Scientific, Waltham, MA) using plasmid DNAs and polyethyleneimine as transfection reagent. For co-transfections, plasmid DNAs were added in an equal ratio. Following 6 days of expression, the conditioned culture media were adjusted to contain 300 mM NaCl, 20 mM HEPES, pH 7.5, and subjected to Ni-NTA Superflow (Qiagen, Valencia, CA) chromatography as described in previous work26. The eluted protein was buffer exchanged into 25 mM HEPES, 300 mM NaCl, pH 7.5, for subsequent experiments. Protein expression for the cryo-EM studies employed CHPF1-pGEn17 and CHSY3-pGEn17 co-transfection into FreeStyle 293-F cells in suspension followed by selection with 2 μg/ml puromycin for 14 days. The puromycin-resistant cell population was subjected to single cell sorting into 96 well plates using a Namocell Pala Single Cell Dispenser (BioTechne). The cultures were then expanded and screened for GFP fusion protein expression and the highest expressing cultures were further expanded for protein production. Shake flask cultures were employed for batch-mode isolation of CHSY3–CHPF1 from the conditioned media by Ni-NTA Superflow (Qiagen, Valencia, CA) chromatography. The purified enzyme was subjected to cleavage of N-terminal fusion sequences with TEV protease26 and further purified by size exclusion chromatography on a Superdex 200 (Cytiva, Marlborough, MA) column. Pooled fractions were concentrated and re-purified Cytiva Superdex 200 Increase (Cytiva, Marlborough, MA) column in 20 mM HEPES, 150 mM NaCl, 60 mM imidazole, and 5 mM β-mercaptoethanol, final pH 7.5, and peak fractions were used for cryo-EM grid preparation. Expression and purification of Bikunin as the acceptor for co-polymerase activity assays The soluble proteoglycan bikunin (Inter-alpha-trypsin inhibitor light chain, residues 206–352, UniProt ID: P02760) expression construct was generated by gene synthesis (GeneArt, Thermo Fisher Scientific) in the pGEn2 expression vector as GFP fusion protein. Bikunin-GFP was expressed in FreeStyle 293-F cells as a secreted protein and purified using Ni-NTA Superflow chromatography as described above. The bikunin preparation was buffer exchanged into PBS and stored at -80 °C for further use. Mass photometry The TEV cleaved CHSY3–CHPF1 complex was diluted to 5 nM in 20 mM HEPES, 100 mM NaCl, pH 7, and 2 µL (diluted into a final volume of 20 µL) were loaded onto a glass slide (Refeyn) and 6 well gasket. Three protein standards (thyroglobulin (MilliporeSigma), β-amylase (TCI), and bovine serum albumin (Thermo)) were used to calibrate the Refeyn mass photometer (Model TwoMP). The spectrum was analyzed using the Refeyn DiscoverMP software to determine the mass of the TEV-cleaved CHSY3–CHPF1 heterodimer. Sequence similarity matrix, alignments, and domain mapping BioPython49 was used to generate a sequence similarity matrix of CSGALNACT1, CSGALNACT2, CHPF1 CHPF2, CHSY1, and CHSY3. CLUSTAL sequence alignments were generated using the Clustal Omega webportal50 and analyzed for their domain structure using the Distance matrix alignment (Dali)51. AlphaFold modeling Initial modeling of CHPF and CHSY chains as homodimers and heterodimers was performed using AlphaFold 327 server. Subsequent modeling with substrates as ligands employed the open-source version of AlphaFold 352 (version 3.0.0). CHSY and CHPF sequences in FASTA file format were submitted along with ccd format files of UDP-GlcA, UDP-GalNAc, SMILES format files of the CS 6-mer oligomer with a non-reducing end GalNAc and reducing end O-methyl group and CS 7-mer with a non-reducing end GlcA and reducing end O-methyl group, and two Mn2+ ions. SMILES strings of the CS oligomers were generated using the Glycam53 webtool to generate coordinate files, which were then converted into SMILES strings using the OpenBabel54 webtool. Alternatively, equivalent models were also generated using the bondedAtomPair syntax as input for assembly of CS oligomers55. The models of the AF3 ligand complexes can be found on ModelArchive [https://modelarchive.org/] under accession numbers: ma-r1hks (CHSY3–CHPF1) [https://modelarchive.org/doi/10.5452/ma-r1hks], ma-3wcmx (CHSY1-CHPF1) [https://modelarchive.org/doi/10.5452/ma-3wcmx], ma-y59jy (CHY3-CHPF2) [https://modelarchive.org/doi/10.5452/ma-y59jy], and ma-afm31 (CHSY1-CHPF2) [https://modelarchive.org/doi/10.5452/ma-afm31]. PISA analysis of AlphaFold models All protein complexes were analyzed in PISA56 to derive values for surface area, interface area, PISA ΔG, and PISA P-Values. The AlphaFold 3 heterocomplex models were used to calculate per-residue local confidence scores (pLDDTs) for residues within 4 Å of the opposing chain and were averaged to define the interface confidence metric. Threshold metrics were established for valid dimer interfaces by using known homodimeric proteins: B3GNT230 (PDB ID 6WMN and equivalent AlphaFold 3 homodimer model), EXT1-EXT240 (PDB ID 7SCJ and equivalent AlphaFold 3 homodimer model), RNAse A57 (PDB ID 2G8Q and equivalent AlphaFold 3 homodimer model), and GalS158 (PDB ID 8D3Z and equivalent AlphaFold 3 homodimer model) as positive controls. Two known monomeric proteins (ST6GAL159, PDB ID 4JS2, and B4GALT133, PDB ID 1TVY) were modeled in AlphaFold 3 as homodimeric complexes and were used as negative controls for interface metrics. A comparison of values for the positive and negative controls allowed us to set a cutoff of interface confidence at >80%, PISA P-value of <0.3, ratio of interface area to total protein surface area >4%, and PISA ΔG value <−20 kcal/mol. These values were then inverted or turned to relative values for plotting in Supplementary Fig. 3b. Enzymatic assays Enzyme activity of CHSY and CHPF homodimeric and heterodimeric complexes and their domain-specific mutants were assayed using the UDP-Glo Assay (Promega, Madison, WI) to quantify the UDP by-product released from the GlcA and GalNAc transfer reactions. Reactions employed either CS oligomers or recombinant bikunin as acceptors. Assays using CS oligomer acceptors (6-mer: GalNAc-GlcA-GalNAc-GlcA-GalNAc-GlcA-pNP and 7-mer: GlcA-GalNAc-GlcA-GalNAc-GlcA-GalNAc-GlcA-pNP, Glycan Therapeutics, Raleigh, NC) employed either UDP-GlcA or UDP-GalNAc as donors or co-polymerase activity when both sugar nucleotides were added. Reactions were performed in 100 mM HEPES, pH 7.5, UDP-GlcA (200 µM) or UDP-GlcNAc (200 µM) as donor and 6-mer (0.5 mM), 7-mer (0.5 mM), or bikunin (4.5 µM) as acceptor, and 5 mM MnCl2. Enzyme reactions were initiated by the addition of enzyme and incubated for 1 hour at 37 °C, and the activity was quantified by addition of UDP-Glo reagents as specified by the manufacturer’s instructions. All assays were performed in at least triplicate. Acceptor kinetics were carried out using the wild-type CHSY3–CHPF1 complex and individual 6-mer and 7-mer oligomer acceptors with either UDP-GlcA or UDP-GalNAc as donor and 5 mM MnCl2 for 1 h at 37 °C, followed by addition of UDP-Glo assay reagent as described above. To determine acceptor kinetics, the 6-mer or 7-mer concentrations were varied from 0–1 mM in the presence of excess donor (2 mM UDP-GlcA for 6-mer and 2 mM UDP-GalNAc for the 7-mer). To determine donor kinetics, the sugar nucleotide donors were varied from 0–1 mM in the presence of excess acceptor (2 mM). To examine the effects of the cystatin-like domains in CS copolymerization kinetics, we repeated the above kinetic titrations for both the WT and the CHSY3 cluster 3 mutant of the CHSY3–CHPF1 complex in the presence of excess of both donors (2 mM) and varied acceptor concentration 0–1 mM of the 6-mer and 7-mer acceptor substrate. We repeated this with Bikunin titrations from 0–0.2 mM as well (with excess of both donors- 2 mM) to examine the effects on kinetics on the proteoglycan CS copolymerization. Fluorescent imaging of GFP fusion proteins by SDS–PAGE To visualize the GFP fusion proteins by SDS-PAGE, samples were incubated with SDS sample reagents at 37 °C for 10 min prior to SDS–PAGE and the gels were run at 150 V for 1.5 h on ice in SDS MOPS running buffer. SDS sample reagents contained 1× reducing agent and 1× loading dye. Gels were then visualized by fluorescent imaging on an iBright FL1500 imager (Thermo Fisher Scientific) to detect GFP fluorescence. For detection of CS co-polymerase activity of the CHSY–CHPF complexes, we performed reactions using a universal buffer (a mixture of 100 mM HEPES, 100 mM MOPS, and 100 mM TRIS, pH 7.5), UDP-GalNAc (1 mM), UDP-GlcA (1 mM), or both sugars as donors, 5 mM MnCl2, and 9 µM recombinant bikunin-GFP fusion protein as acceptor. Enzyme reactions were initiated by the addition of the CHSY-CHPF complex and incubated for 6 h at 37 °C. Samples were then resolved by SDS-PAGE and fluorescently imaged as described above to detect the shift in SDS-PAGE mobility for the bikunin fusion protein upon extension of its CS chains. Control reactions with the copolymerase products were incubated with two units of chondroitinase ABC treatment (Thermo Fischer Scientific) for 2 h to degrade the extended CS polymers on the proteoglycan before loading on SDS-PAGE gels and fluorescent imaging. MALDI-TOF mass spectrometry of CS oligosaccharides The product analysis for single sugar addition on the 6-mer and 7-mer CS oligomers and co-polymerase reactions was determined by MALDI-TOF MS. Single sugar transfer assays were conducted using 6-mer (500 µM) or 7-mer (500 µM) with 2 mM of UDP-GlcA or UDP-GalNAc, respectively, in reactions supplemented with 25 mM HEPES, pH 7.5, and 5 mM MnCl2. The reactions were incubated for 1 h at 37 °C following addition of the CHSY–CHPF complex. Samples were stopped by cooling on ice and product formation was detected using on a Bruker MSP 96 ground steel target coated with Nafion D-521 (Thomas Scientific). Reaction products were diluted and mixed 1:1 with a 10 mg/ml stock of 2,5-dihydroxybenzoic acid (DHB) matrix solution in 50% acetonitrile on the MALDI plate. Negative-ion MALDI-TOF MS spectra were collected on a Bruker smartfleX. Co-polymerase activity assays using CS 7-mer oligomer acceptors (500 µM) and both sugar nucleotide donors (5 mM each of UDP-GlcA and UDP-GalNAc) were conducted in reaction conditions as described above for all CHSY–CHPF complexes. Time points were taken at 0, 10, and 30 min to analyze the elongated CS products. The raw files have been uploaded to Zenodo [https://zenodo.org/]. The copolymerase spectrums are listed under record ID 16739850 (DOI 10.5281/zenodo.16739850) and the single transfer spectrums are listed under record ID 16739747 (DOI 10.5281/zenodo.16739747). CHPF1–CHSY3 complex cryo-EM grid preparation and data collection The purified CHPF1-CHSY3 complex was concentrated using a Vivaspin concentrator with a 100 kDa cutoff (GE Healthcare) to a final concentration of 0.6 mg/ml. The sample was mixed with UDP and MnCl2 (final concentrations 2 mM UDP and 5 mM MnCl2) and incubated for 15 min at room temperature before preparation of grids. Quantifoil R1.2/1.3 300-mesh copper holey carbon grids were glow-discharged at 25 mA for 30 s. The sample was applied onto the grid (4 µL) before blotting for 3 s and plunge freezing into liquid ethane using a Vitrobot Mark IV (Thermo), with the chamber humidity set to 95% and a temperature of 4 °C. The CHPF1–CHSY3 grids were screened using a 200 kV Talos Arctica (FEI/Thermo) electron microscope, equipped with a bioquantum K3 direct electron detector (Gatan). 5,811 movies were collected in super-resolution mode at 79,000× magnification, a defocus range of 0.7–3.5 µm, a pixel size of 0.541 Å/pixel, and a total dose rate of 47.6 e−/Å2. Cryo-EM data processing Cryo-EM data processing was carried out using cryoSPARC v4.5.160. The 5,811 micrographs were corrected for the beam-induced drift with patch motion correction and the contrast transfer function (CTF) of each aligned micrograph was determined using patch CTF estimation. The micrographs were analyzed and manually selected based on CTF fit resolution, ice thickness and total full-frame motion. The resulting 4,311 micrographs were used for iterative jobs of particle picking, first using blob picker and then template picker functions, resulting in 2,689,859 particles that were extracted with a box size of 512 pixels and Fourier-cropped to 192 pixels (rescaled pixel size of 1.442 Å). The particles were subjected to several rounds of 2D classification with 100 classes per round to discard bad particles. Particles were further sorted by a combination of ab-initio modeling and heterogeneous refinement. Non-uniform (NU) refinements were used to assess resolution and quality of the reconstruction maps. This resulted in a subset of 512,703 particles displaying a reconstruction map with an average resolution of 3.63 Å. This subset of particles was re-extracted and sorted through further heterogeneous refinements, in combination with 3D classification and NU-refinements. Finally, for the last 163,527 subset of particles, rounds of local refinements, local and Global-CTF refinements contributed to yield a final reconstruction map of 3.42 Å. The local resolution was estimated with cryoSPARC v4.5.160 and the directional resolution estimation was estimated using the 3DFSC server61, which calculated a sphericity of 0.931. Model building, refinement, and validation A first model generated using AlphaFold 327, consisting of CHPF1 (residues 108-775) and CHSY3 (residues 172-882), was used to assist model building. The rest of the residues of both proteins were removed due to their ambiguous structure and low pLDDT score27. The model was fitted into the Cryo-EM density map using ChimeraX v1.862 and modeled using the tool ISOLDE63 applying secondary structure constraints. The resulting model was then used for manual building using Coot 0.9.8.9564, adding cysteine bonds and UDP and Mn2+ to the respective densities. The web service tool MolProbity65 was used together with PHENIX66 and Coot to validate the built model. Figures of models and densities were made in ChimeraX62 and PyMOL (The PyMOL, Molecular Graphics Systems, Schrödinger LLC). The EMRinger score for the final model is 2.7367. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Source Data