Introduction

Biotin-dependent carboxylases (BDCs) catalyze the carboxylation of several key metabolites using covalently linked biotin as a cofactor(1, 2). In humans, there are four types of BDCs: acetyl-CoA carboxylase (ACC), propionyl-CoA carboxylase (PCC), 3-methylcrotonyl-CoA carboxylase (MCC), and pyruvate carboxylase (PYC)(1, 2). All BDCs contain a biotin carboxylase (BC) domain, a carboxyltransferase (CT) domain, and a biotin carboxyl carrier protein (BCCP) domain(1, 2). The carboxylation reaction involves a two-step process(1, 2). Firstly, the biotin covalently linked to the BCCP domain is carboxylated in the BC domain, using bicarbonate as the donor of the carboxyl group(3). Secondly, the carboxylated biotin is carried to the CT domain by the BCCP domain(4). In the CT domain, the carboxyl group is transferred from the biotin to acetyl-CoA, propionyl-CoA, 3-methylcrotonyl-CoA, or pyruvate(2, 5). Among the four human BDCs, ACC and PYC each consist of a single chain that incorporates the BC, CT, and BCCP domains(2). In contrast, PCC and MCC have their BC, CT, and BCCP domains divided into two chains: the α subunits contain the BC and BCCP domains, with an intermediate BC-CT interaction (BT) domain between them, while the β subunits solely consist of the CT domain(2).

In humans, PCC and MCC catalyze the carboxylation of propionyl-CoA and 3-methylcrotonyl-CoA (MC-CoA), respectively(6, 7). Propionyl-CoA is an intermediate in the metabolism of odd-chain fatty acids, cholesterol, and amino acids including valine, isoleucine, methionine, and threonine; its carboxylated product (S)-methylmalonyl-CoA is then converted into succinyl-CoA and subsequently metabolized in the tricarboxylic acid cycle(6). MC-CoA is an intermediate in the metabolism of leucine; following carboxylation, the resulting product 3-methylglutaconyl-CoA (MG-CoA) is further converted into 3-hydroxy-3-methylglutaryl-CoA (HMG-CoA) and then involved in the synthesis of cholesterol(8). Therefore, both PCC and MCC play important roles in metabolic processes. Loss-of-function mutations in MCC and PCC lead to severe metabolic disorders(6, 9-11).

The protein sequences and functions of PCC and MCC are conserved from bacteria to humans. A crystal structure of a chimeric bacteria PCC holoenzyme, obtained by co-expressing the α subunit of Ruegeria pomeroyi PCC (RpPCC) with the β subunit of Roseobacter denitrificans PCC (RdPCC) in E. coli, was determined at a resolution of 3.2 Å (12). The PCC holoenzyme is a dodecamer formed by six α subunits and six β subunits, thereby referred to as the α₆β₆ holoenzyme(12). Crystal structures of a bacteria MCC (Pseudomonas aeruginosa MCC, PaMCC) holoenzyme reveal that the MCC holoenzyme is also a α₆β₆ dodecamer(13). The structures of bacteria PCC and MCC holoenzymes reveal how the α subunits and β subunits assemble to form the holoenzymes(12, 13). However, our understanding of human PCC and MCC is constrained by the limited availability of high-resolution three-dimensional structures. Thus far, only a low-resolution cryo-EM structure of human PCC holoenzyme (15 Å) has been reported (12), highlighting the need for high-resolution structural information of human PCC and MCC.

In this study, we purified the endogenous PCC and MCC from human cells and determined the cryo-EM structures of human PCC and MCC holoenzymes in the apo (PCC or MCC holoenzyme alone), acetyl-CoA-bound, and propionyl-CoA-bound states, at resolutions ranging from 2.29 to 3.28 Å. Structural analysis reveals that human PCC and MCC holoenzymes have overall structures similar to their bacterial homologs. For human PCC holoenzyme, the structure in the acetyl-CoA-bound state closely resembles the structures in the apo and the propionyl-CoA-bound state. In these structures, the covalently linked biotin binds to a site that is distant from the catalytic residues in the CT domain, rendering it catalytically incompetent. In contrast, for human MCC holoenzyme, acetyl-CoA binding, but not propionyl-CoA binding, induces conformational changes in the acyl-CoA binding pocket, thus relocating the covalently linked biotin from an exo-site to an endo-site that is closer to the catalytic residues in the CT domain.

Results

Overall structures of human PCC and MCC holoenzymes

The BDCs endogenously expressed in Expi 293F cells were purified using streptavidin beads (the Strep-Tactin^®XT resin) since streptavidin can specifically bind to the covalently linked biotin in the BDCs. The purified BDCs were further purified using size exclusion chromatography. Mass spectrometry analysis identified all four types of human BDCs in the purified samples (fig. S1). The samples were subjected to cryo-EM sample preparation (fig. S1, A to C).

The cryo-EM structures of the PCC and MCC holoenzymes alone, referred to as PCC-apo and MCC-apo, were determined at resolutions of 3.02 Å and 2.29 Å, respectively (figs. S2A, S3, A to D and S4, A to D, tables S1 and S2). The cryo-EM maps have good quality to support model building (figs. S5, S6 and S7).

Both the PCC and MCC holoenzymes consist of six α subunits and six β subunits. The cryo-EM densities of one PCCα subunit in the PCC-ACO structure and three MCCα subunits in the MCC-apo, MCC-ACO, and MCC-PCO structures are partially missing. Therefore, in order to demonstrate the complete structures of the holoenzymes, these subunits were built into the structures based on the density of the BT domain in each α subunit (fig. S8). The α₆β₆ dodecamer can be divided into four layers. The six β subunits are stacked into two layers to form the core of the holoenzymes, with three β subunits laid head to end in each layer (Fig. 1, A and B). The six α subunits are separated into two layers and situated at the top and bottom sides of the β subunit core, with each layer contains three α subunits. A major difference between the PCC and MCC holoenzymes is that the three α subunits in each layer have no contact in the PCC holoenzyme but interact with each other in the MCC holoenzyme (Fig. 1, A and B).

Fig. 1.

The overall structures of human PCC and MCC holoenzymes are similar to the reported crystal structures of bacteria PCC and MCC holoenzymes, respectively (Fig. 1, C and D). Specifically, alignment of the β subunits in the PCC-apo structure with those in the structure of a chimeric bacteria PCC holoenzyme (PDB code: 3N6R) shows that the β subunit cores in the two structures are very similar(12). Each α subunit in human PCC holoenzyme has a minor rotation relative to the corresponding α subunit in the bacteria PCC holoenzyme (Fig. 1C). Similarly, the β subunit core in the MCC-apo structure aligns well with that in the crystal structure of a bacteria MCC holoenzyme (PDB code: 3U9T), with each α subunit slightly rotated in comparison to the corresponding α subunit in the bacteria MCC holoenzyme(13) (Fig. 1D). The structures of the BC, BT, and BCCP domains in the α subunits of human PCC and MCC resemble the corresponding domains in bacteria PCC and MCC(12, 13), except that the BT domain of human MCC has one more β strand in comparison to that of the PaMCC (fig. S9).

Organization of the catalytic domains in human PCC and MCC holoenzymes

We incubated the purified human BDCs with propionyl-CoA (fig. S1, D to G), which is the substrate of PCC, and determined the cryo-EM structures of the propionyl-CoA-bound PCC and MCC holoenzymes (referred to as PCC-PCO and MCC-PCO), at resolutions of 2.80 Å and 2.36 Å, respectively (figs. S2B, S3, E to H, S4, E to H, S6 and S7, tables S1 and S2). The two structures are almost identical to the structures of PCC-apo and MCC-apo, respectively. Although bicarbonate, MgCl₂, and ATP were added to the purified BDC samples to facilitate the carboxylation reaction, the cryo-EM density of the biotin region does not provide conclusive evidence as to whether the covalently linked biotin in the PCC holoenzyme structures was indeed carboxylated. Thus, the cryo-EM structures only contain the uncarboxylated biotin.

The two-step carboxylation reactions catalyzed by PCC and MCC require the coordination of the BC, BCCP domains in the α subunit and the CT domain in the β subunit (Fig. 2A). In human PCC holoenzyme, each α subunit binds to one β subunit containing the CT domain that can be divided into CT-N and CT-C subdomains (Fig. 2B). The BC domain in the α subunit is situated on the CT-C subdomain, while the BCCP domain also attaches to the CT-C subdomain and positions the covalently linked biotin to a pocket adjacent to the propionyl-CoA binding pocket. The BT domain tightly interacts with both the CT-N and CT-C subdomains and functions as a hub to hold the BC and BCCP domains together with the CT domain (Fig. 2B).

Fig. 2.

In contrast, in human MCC holoenzyme, each α subunit binds to two β subunits (referred to as β1 and β2) (Fig. 2C). The CT-N subdomain of β1 interacts with the CT-C subdomain of β2. The BT domain of the α subunit associates with both the CT-N and CT-C subdomains of β1, while the BCCP domain attaches to the CT-C subdomain of β2. The propionyl-CoA binds to a pocket formed by the CT-C of β2 and the CT-N of a third β subunit (β3) from the other β subunit layer. The BC domain has no contact with the β subunits.

The variances in the organization of the catalytic domains in the PCC holoenzyme and the MCC holoenzyme are partly attributed to the distinct arrangement of their CT domains (Fig. 2D). Alignment of the CT-N subdomains shows that the CT-C subdomains in PCC and MCC have distinct orientations (Fig. 2D). The CT domain in PCC can perfectly align with a combination including the CT-N of β1 and the CT-C of β2 of MCC (Fig. 2E).

Propionyl-CoA binding in human PCC holoenzyme

In each PCC holoenzyme, six propionyl-CoA molecules are observed binding to the six active sites of CT domains. Each binding site for propionyl-CoA is formed by the CT-C subdomain of one β subunit and the CT-N subdomain of another β subunit in the other layer (Fig. 3A). The conformation of the propionyl-CoA binding pocket in the PCC-PCO structure is only slightly different from that in the PCC-apo structure (Fig. 3B). The AMKM motif in the BCCP domain and the biotin covalently linked to it also show only minor conformational differences in the PCC-PCO structure compared to the PCC-apo structure (Fig. 3C). The distances between the ureido ring carbonyl oxygen of the biotin and the main chain amide of the catalytic residues G437 and A438 in the corresponding CT domain in the PCC-PCO structure are more than 7 Å, which is too far for an efficient catalytic process (Fig. 3C). This finding suggests that the biotin may need to relocate to a site that is closer to the catalytic residues to facilitate the carboxylation reaction.

Fig. 3.

Before our study, two structures were reported to demonstrate the binding of CoA or propionyl-CoA in the CT domains of bacteria PCCs. One is the crystal structure of the β subunits of Streptomyces coelicolor PCC (ScPCC) in complex with propionyl-CoA (PDB code: 1XNY)(14). The other is the cryo-EM structure of the Methylorubrum extorquens PCC (MePCC) holoenzyme in complex with CoA (PDB code: 6YBP)(15). The binding mode of propionyl-CoA in our PCC-PCO structure is distinct from that in the ScPCC β subunit structure, but similar to that in the MePCC holoenzyme structure (Fig. 3D, E).

Acetyl-CoA binding in human PCC and MCC holoenzymes

In an unpublished manuscript (Zhou et al, 2024), we incubated the purified human BDCs with acetyl-CoA, which is a substrate of ACC, and determined the cryo-EM structure of the acetyl-CoA-bound ACC. Using the same batch of purified BDCs (fig. S1, H to K), we also determined the cryo-EM structures of human PCC and MCC holoenzymes in the acetyl-CoA-bound state (referred to as PCC-ACO and MCC-ACO) at resolutions of 3.38 Å and 2.85 Å, respectively (figs. S2C, S3, I to L, S4, I to L, S6 and S7, tables S1 and S2).

The PCC-ACO structure closely resembles the PCC-apo and the PCC-PCO structures. The acetyl-CoA in the PCC-ACO structure has a binding mode similar to the propionyl-CoA in the PCC-PCO structure (Fig. 3F). The conformation of the acyl-CoA binding pocket in both structures is also highly similar (Fig. 3F). It was reported that D422 in the β subunit of a bacteria PCC (ScPCC) is crucial for the substrate specificity of ScPCC, and the D422I mutation significantly increased the catalytic activity of ScPCC towards acetyl-CoA(14, 16). But the PCC-PCO structure shows that D440, which corresponds to D422 in the β subunit of ScPCC, is not directly involved in the interaction with propionyl-CoA or acetyl-CoA (Fig. 3F).

The overall structure of MCC-ACO is also similar to that of MCC-apo, but alignment of the CT-N subdomains shows that the position of the BCCP domain and the conformation of the acyl-CoA binding pocket in the MCC-ACO structure are significantly different from those in the MCC-apo structure (Fig. 4A). Specifically, the BCCP domain moves towards the CT-C subdomain in the MCC-ACO structure compared to that in the MCC-apo structure (Fig. 4A). Consequently, the covalently linked biotin moves from an exo-site to an endo-site that is deeper in the pocket formed by the CT-N and CT-C subdomains from two β subunits (Fig. 4B). At the exo-site, the ureido ring carbonyl oxygen of the biotin is about 8 Å away from the main chain amide of the catalytic residues G437 and A438 (fig. S10A). In contrast, at the endo-site, the distance is shortened to about 3.5 Å (fig. S10B). The acyl-CoA binding pocket, also formed by the CT-N and CT-C subdomains from two β subunits, switches from an open conformation in the MCC-apo structure to a closed conformation in the MCC-ACO structure (Fig. 4, C to E), with the major conformational change occurring at two helices (residues 474 to 517) in the CT-C subdomain and one helix (residues 243 to 252) in the CT-N subdomain (Fig. 4C).

Fig. 4.

We aligned the MCC-ACO structure with the crystal structure of a bacteria MCC (PaMCC) holoenzyme in complex with CoA (PDB code: 3U9S) and observed a high similarity between them(13). Both the BCCP domain and the CT domain in the MCC-ACO structure align well with those in the PaMCC structure. The AMKM motif and biotin in the two structures also show a high similarity (Fig. 4F). Interestingly, alignment of the CoA-bound PaMCC holoenzyme structure with the structure of the PaMCC holoenzyme alone shows a change in the conformation of two helices in the CT-C subdomain of PaMCC(13), similar to that observed in the alignment of the MCC-ACO structure with the MCC-apo structure (Fig. 4C). These findings suggest that conformational changes in the acyl-CoA binding pocket, combined with the change in the position of the BCCP domain, may be necessary for the catalytic process of MCC.

Discussion

Several three-dimensional structures of PCCs and MCCs have been reported (12-18), however, most of them are from bacteria and expressed in E. coli(12-16). In our study, we extracted the endogenous PCC and MCC from human cells by one-step affinity purification (fig. S1) and determined their cryo-EM structures in the apo and substrate-bound states. The overall structures of human PCC and MCC holoenzymes resemble those of their bacterial homologs. This similarity aligns with the conserved functions of PCCs and MCCs across species, ranging from prokaryotes to eukaryotes.

A previous study showed that PCC purified from human liver also catalyzed the carboxylation of acetyl-CoA, but the rate was only about 1.5% of that observed for propionyl-CoA(19). Propionyl-CoA and acetyl-CoA, which have highly similar structures, undergo carboxylation at analogous positions—the α carbon of the propionyl or the acetyl group. A key question is, what determines the substrate specificity of the PCC holoenzyme? The PCC-PCO and PCC-ACO structures determined in our study demonstrate that both propionyl-CoA and acetyl-CoA can bind to human PCC holoenzyme, with a similar binding mode (Fig. 3F). Thus, the selectivity of PCC towards propionyl-CoA and acetyl-CoA is largely determined by the interactions between the acyl group and PCC. The longer acyl group in propionyl-CoA, in comparison to that in acetyl-CoA, may mediate a stronger hydrophobic interaction that stabilizes the α carbon of the acyl group at a proper position to react with the carboxylated biotin.

The cryo-EM structures of MCC-apo and MCC-ACO reveal that there are two distinct binding sites in the CT domains of human MCC holoenzyme for the covalently linked biotin (Fig. 4B). In the MCC-apo structure, the biotin binds to a site that is distant from the catalytic residues A447 and G448, thus referred to as the exo-site; while in the MCC-ACO structure, the biotin binds to a site that is closer to A447 and G448 and is therefore referred to as the endo-site (Fig. 4B, fig. S10). Binding of acetyl-CoA to the acyl-CoA binding pocket of the CT domains induces conformational changes in these domains, leading to the relocation of the biotin molecule from the exo-site to the endo-site. Conformational changes in the CT domains upon CoA binding were also observed in the crystal structures of PaMCC(13). These findings indicate a coordination between biotin binding and acetyl-CoA binding for the MCC holoenzymes. It should be noted that acetyl-CoA is not a substrate for MCC. Next, we will try to determine the cryo-EM structure of human MCC in complex with its substrate MC-CoA to see whether the binding of MC-CoA to MCC has similar effects to the binding of acetyl-CoA.

Similarly, in an unpublished manuscript (Zhou et al., 2024), it was also observed that the covalently linked biotin in the ACC-citrate filament, which is the active form of human ACC, binds to an exo-site that is distant from the acetyl-CoA binding site in the absence of acetyl-CoA. In contrast, in the crystal structure of CoA-bound yeast ACC, the biotin occupies an endo-site that is closer to the acetyl-CoA binding site(20).

In the crystal structure of Staphylococcus aureus PYC (SaPYC), two distinct binding sites for the covalently linked biotin were also observed, one is about 20 Å from the active site of the CT domain and is called the exo site, and the other is in the active site of the CT domain(1, 21).

Taken together, our studies, combined with previous research, suggest that the presence of an exo-site and an endo-site for the binding of covalently linked biotin is a common feature shared by many BDCs. The transfer of biotin from the exo-site to the endo-site may be promoted by the coordinated binding of acyl-CoA or pyruvate to the CT domains. Conversely, relocation of biotin from the endo-site to the exo-site might facilitate the release of the products after the second step carboxylation reaction.

However, we did not observe the relocation of biotin to an endo-site upon acyl-CoA binding in human PCC holoenzyme (Fig. 3). Additionally, while we did observe relocation of biotin and conformational changes in the CT domains of human MCC holoenzyme upon binding with acetyl-CoA, similar changes were not observed with propionyl-CoA binding. Further studies focusing on capturing the structures of human PCC and MCC holoenzymes in various states during their catalytic processes would contribute to a comprehensive understanding of the underlying catalytic mechanisms.

Materials and Methods

Cell culture

Expi 293F cells (Thermo Scientific) were cultured in SMM 293T-II medium (Sino Biological) at 37 °C with 5% CO₂ in a Multitron-Pro incubator (Infors). After the density reached 2.0×10⁶/mL, the cells were cultured for an additional 48 h, then harvested by centrifuging at 4,000 ×g (Fiberlite F9-6×1000 LEX rotor, Thermo Scientific) for 10 min and resuspended with HBS buffer (20 mM HEPES, pH 7.4, 150 mM NaCl). The cells were snap frozen and stored at −80 °C.

Endogenous purification of biotin dependent carboxylases

The frozen cells were thawed and disrupted using a high-pressure cell crusher and centrifuged at 23,376 ×g (JA-14.50 rotor, Beckman) for 1 h at 4 °C. The supernatant was incubated with Strep-Tactin^®XT resin (IBA Lifesciences) for 1 h at 4 °C and then washed with HBS buffer. The proteins were eluted by HBS buffer supplemented with 50 mM ᴅ-biotin (J&K Scientific) and further purified by size-exclusion chromatography (Superose 6 Increase 10/300 GL column, Cytiva) using HBS as the running buffer. Different batches of proteins were purified for the preparation of the samples at different substrate binding states. For the substrate-free sample (the apo sample), fractions eluted within 12.5–15 mL were combined and concentrated to 2.85 mg/mL. For the substrate-bound samples, fractions eluted within 12.5–16.5 mL were combined and concentrated to 2.5–2.65 mg/mL. The samples were pre-incubated with 25 mM NaHCO₃, 10 mM MgCl₂ and 10 mM propionyl-CoA for the propionyl-CoA-bound sample (the PCO sample) or acetyl-CoA for the acetyl-CoA-bound sample (the ACO sample), followed by adding 10 mM ATP immediately before frozen on the cryo-EM grids. The proteins were analyzed by mass spectrometry and applied for cryo-EM grids preparation the same day purified.

Cryo-EM data acquisition and pre-processing

Holey grids (Holey carbon filmed 300 mesh gold grids, R1.2/1.3, Quantifoil) were glow discharged with a medium RF power for 30 s using a plasma cleaner PDC-32G-2 (Harrick Plasma) prior use. Aliquots (4 μL) of each sample were loaded to the grids and flash frozen in liquid ethane immediately using Vitrobot Mark IV (Thermo Fisher Scientific). The cryo-EM grids were stored in liquid nitrogen until data acquisition.

The cryo-EM grids were transferred to a Titan Krios TEM (FEI) operating at 300 kV equipped with a K3 Summit direct electron detector (Gatan) and a GIF Quantum energy filter (Gatan). Zero-loss movie stacks were automatically collected using EPU software (FEI) with a slit width of 20 eV on the energy filter and a defocus range from −1.2 to −2.2 µm in super-resolution mode at an 81,000× nominal magnification. Each movie stack, which contained 32 frames, was exposed for 2.56 s with a total electron dose of ∼50 e^-/Ų. The stacks were motion corrected using MotionCor2 with a binning factor of 2, resulting in the pixel size of 1.0773 Å(22). Dose weighting was performed concurrently(23). The movie stacks were imported to cryoSPARC (v3.3.2, Structura Biotechnology) for CTF estimation and downstream processing(24).

Data processing

For the apo sample, 11,481 micrographs were captured and imported to cryoSPARC for processing. Particles were automatically picked according to the templates generated from an initial 2D classification of the manually picked particles and extracted with a box size of 488 pixels for ab-initio reconstruction and heterogeneous refinement jobs, yielding the maps of PCC (11.5%) and MCC (37.6%) holoenzymes. To get the high-resolution map of PCC holoenzyme, a non-uniform refinement with the corresponding group of particles was carried out, yielding a detailed map, yet the density for one of the α subunits was lacking. To solve this problem, distinct reference volumes with (Ref-A) or without (Ref-B) the corresponding PCCα were generated by UCSF Chimera (v1.16)(25). After a refinement based on multiple references, 61,567 particles were subjected to the homogenous refinement, yielding the complete map of PCC (PCC-apo) at 3.02 Å resolution. For the case of MCC holoenzyme, a round of non-uniform refinement was carried out. After that, the volume chirality was flipped, and the map was further polished by a series of refinement jobs, resulting in the map of MCC (MCC-apo) at 2.29 Å resolution.

For the PCO sample, 10,898 micrographs were imported for data processing. After particle picking, extraction (box size = 488 pixels), ab-initio reconstruction and heterogeneous refinement, the PCC (8.7%) and MCC (39.7%) representing volumes were further processed by non-uniform refinement, yielding the maps generated from 105,293 and 476,773 particles, respectively. The PCC map was subjected to local CTF refinement and subsequent local refinement, yielding the final map of propionyl-CoA bound PCC holoenzyme (PCC-PCO) at 2.80 Å. The MCC map was further polished by heterogenous refinement and non-uniform refinement after flipping chirality, producing the final map of propionyl-CoA bound MCC holoenzyme (MCC-PCO) at 2.36 Å.

For the ACO sample, 4,036 micrographs were collected and imported. Particle picking, extraction (box size = 488 pixels), initial 2D classification and particle clustering jobs were carried out accordingly. 31,622 PCC-like particles were enriched in cluster-A and subsequently subjected to ab-initio reconstruction and non-uniform refinement. After flipping the volume chirality, the refined map was further polished by another round of non-uniform refinement, yielding the 3.38 Å map of acetyl-CoA bound PCC holoenzyme (PCC-ACO). In cluster-B, 273,428 MCC-like particles were enriched. These particles were ab-initio reconstructed and classified to 4 volumes, of which one, composed of 39.9% of particles, exhibited clear secondary structure features of MCC. This map was further processed by heterogeneous refinement and non-uniform refinement jobs, yielding the final map of acetyl-CoA bound MCC holoenzyme (MCC-ACO) at 2.85 Å resolution.

The resolutions mentioned above were determined according to the gold-standard Fourier shell correlation 0.143 criterion and with a high-resolution noise substitution method(26). All the reconstructed maps were subjected to further model building.

Model building and structure refinement

The AlphaFold predicted structures were used as the templates for the model building of PCC and MCC holoenzymes(27, 28). The reference models were automatically placed in the EM maps using Rosetta and manually adjusted using Coot, resulting the modified models(29, 30). Each residue was manually checked with the chemical properties taken into consideration during model building. Several segments, whose corresponding densities were invisible, were not modeled. Real space refinement was performed using Phenix with secondary structure and geometry restraints to prevent overfitting(31). For the cross-validation of the structure, the model was refined against one of the two independent half maps from the gold-standard 3D refinement approach. Then, the refined model was tested against the other half map. Statistics associated with data collection, 3D reconstruction and model building were summarized in tables S1 and S2.

Acknowledgements

We thank Drs. Xiaofeng Zhang and Xiechao Zhan for valuable discussion and suggestions on structure determination. We thank the Cryo-EM Facility, the High-Performance Computing Center, and the Mass Spectrometry & Metabolomics Core Facility of Westlake University for technical support.

Funding

“Pioneer” and “Leading Goose” R&D Program of Zhejiang 2024SSYS0036 (Q.H.)

Westlake Laboratory of Life Sciences and Biomedicine (Q.H.)

Westlake Education Foundation (Q.H.)

Author contributions

Conceptualization: Q.H., F.Z.

Methodology: Q.H., F.Z.

Investigation: F.Z., Y.Zhang, Y.Zhu

Visualization: F.Z., Y.Zhang

Funding acquisition: Q.H.

Project administration: Q.H., F.Z.

Supervision: Q.H., Y.S.

Writing – original draft: Q.H., F.Z.

Writing – review & editing: Q.H., Y.S., Q.Z.

Competing interests

Authors declare that they have no competing interests.

Data and materials availability

The cryo-EM structures have been deposited in the Protein Data Bank with the accession codes 8XL3 (PCC-apo), 8XL4 (PCC-ACO), 8XL5 (PCC-PCO), 8XL6 (MCC-apo), 8XL7 (MCC-ACO) and 8XL8 (MCC-PCO). The cryo-EM maps have been deposited in the Electron Microscopy Data Bank with the accession codes EMD-38436 (PCC-apo), EMD-38437 (PCC-ACO), EMD-38438 (PCC-PCO), EMD-38439 (MCC-apo), EMD-38440 (MCC-ACO) and EMD-38441 (MCC-PCO). The predicted structures of human PCC and MCC subunits used for cryo-EM model building were downloaded from the AlphaFold Protein Structure Database with the accession codes AF-P05165-F1 (PCCα), AF-P05166-F1 (PCCβ), AF-Q96RQ3-F1 (MCCα) and AF-Q9HCC0-F1 (MCCβ). The structures of bacteria PCC (PDB codes: 1XNY, 3N6R, 6YBP) and MCC (PDB codes: 3U9S and 3U9T) were used for structural analysis in this study.

Supplementary Materials

Fig. S1.

Fig. S2.

Fig. S3.

Fig. S4.

Fig. S5.

Fig. S6.

Fig. S7.

Fig. S8.

Fig. S9.

Fig. S10.

Table S1.

Table S2.