Identification of a carbonic anhydrase–Rubisco complex within the alpha-carboxysome

Previous immunogold labeling EM and biochemical assays (freeze/thaw treatment of carboxysomes) have suggested that CsoSCA may associate with the shell, but no specific interactions have been described (20, 21). Thus, to identify CsoSCA’s interaction partner, we measured binding of untagged CsoSCA against most carboxysome proteins (the shell proteins CsoS1A, CsoS1B, CsoS1D, and CsoS4B; the scaffolding protein CsoS2B; and Rubisco) using BLI. This screen showed that CsoSCA interacted with Rubisco, while none of the other carboxysome proteins had detectable binding above background (Fig. 2A). An N-terminal truncated protein variant (ΔNTD_1-37CsoSCA) did not bind Rubisco, confirming NTD’s involvement in the interaction (SI Appendix, Fig. S2A). Native-PAGE confirmed binding between CsoSCA and Rubisco and lack of binding to the major shell proteins CsoS1A and CsoS1B (SI Appendix, Fig. S2B). Finally, coelution of Rubisco and NTD_1-53-sfGFP using size exclusion chromatography confirmed the interaction in a solution-based assay (SI Appendix, Fig. S2C).

Concentration dependence of CsoSCA binding to Rubisco was confirmed by BLI and K_D of the interaction determined to be 1.2 nM ± 0.1 (k_on = 2.5 × 10⁵ M⁻¹s⁻¹, k_off = 2.9 × 10⁻⁴ s⁻¹) (Fig. 2B and Table 1). Due to instability of WTCsoSCA, the K_D was measured with a CsoSCA-MBP fusion [the negative control of MBP alone did not bind Rubisco (SI Appendix, Fig. S2D)]. Using the stopped-flow based Khalifah/pH-indicator assay (31), CsoSCA-MBP was confirmed to be catalytically active (SI Appendix, Fig. S2G), demonstrating retained physiological state of the fusion protein. Although a mutated variant of CsoSCA crystallized as a dimer (Fig. 1B) (Y92H mutation resulting in introduction of an additional Zn²⁺-binding site) (22), the CsoSCA-MBP protein eluted on size exclusion chromatography at an estimated molecular weight of 612 kDa, suggesting a hexameric state (SI Appendix, Fig. S2 E and F). Due to this discrepancy, present data cannot conclude whether WTCsoSCA is dimeric or hexameric. Rubisco binding to NTD_1-53-sfGFP further confirmed the NTD_1-53–peptide interaction (K_D1 = 30 nM, K_D2 = 80 nM; fit to a 1:2 model) (Fig. 2C, Table 1, and SI Appendix, Fig. S2G). The 25-fold higher K_D with CsoSCA compared to NTD_1-53-sfGFP is mainly an effect of a slower off-rate, demonstrating the importance of the multivalency (resulting from CsoSCA being multivalent (dimeric or hexameric) while NTD_1-53-sfGFP is monomeric) in obtaining a high-affinity interaction with Rubisco.

An emerging theme of CCM self-assembly is that Rubisco interacts with various CCM proteins via Short Linear Motifs (SLiMs) found in intrinsically disordered proteins/regions (IDP/Rs) (30, 32, 33). Since CsoSCA’s NTD appears to bind Rubisco (Fig. 2C and SI Appendix, Fig. S3), we next sought to determine the structure of Rubisco in complex with this peptide using cryo-EM. In order to promote high occupancy of available binding sites, an excess of NTD peptide was complexed with Rubisco and imaged as described in the Materials and Methods. These data yielded two slightly distinct single-particle reconstructions of Rubisco bound to a peptide corresponding to the first 50 residues of CsoSCA (NTD_1-50) at 1.98 Å (State-1) and 2.07 Å (State-2) nominal resolution (SI Appendix, Figs. S4 and S5 and Table S1). Of note, both reconstructions show densities for ordered waters and alternate side-chain conformers (SI Appendix, Fig. S5). The two confirmations are highly similar (RMSD: 0.22 Å) with the most notable differences occurring in the β-sheet of the N-terminal domain of CbbL, in particular the conformations of loops P37-D42 and G115-G125 (SI Appendix, Fig. S6). Since both reconstructions display similar density for the NTD_1-50 peptide, we do not attribute the difference in confirmation to the presence of the peptide but rather subtle “breathing” of Rubisco. For clarity, we choose to predominantly focus our discussion on the higher resolution 1.98 Å structure (State-1).

In the cryo-EM reconstruction of the Rubisco–NTD_1-50 complex, density corresponding to NTD_1-50 is located in a groove formed at the interfaces of two CbbL subunits (from two different CbbL₂ dimers) and one CbbS (Fig. 3). The biological assembly of Rubisco is CbbL₈S₈, resulting in eight of these binding sites per Rubisco oligomer. The peptide density is of marginally lower-quality (local resolution estimate: 2.1 to 2.2 Å) than that of the surrounding Rubisco density. Nevertheless, we could confidently assign the density to nine residues of the NTD_1-50 peptide starting at P22 (PRLDLIEQA) (Fig. 3 B and C). The structure of the Rubisco–NTD_1-50 complex is highly similar to a previous crystal structure of Rubisco (PDB: 1SVD, RMSD: 0.45 Å, SI Appendix, Fig. S7), indicating that binding of the NTD_1-50 peptide does not induce large-scale conformational changes in Rubisco.

The resolved region of the peptide starts at the bottom of the CbbS subunit (around loop D94-S99), runs downward within the groove between the two CbbL subunits, and ends between β-strand S345-I347 in CbbL_B and loop P19-I29 in CbbL_A, that results in a buried interface of approximately 700 Ų. This short stretch of sequence is predicted by JPred to form an alpha-helix (SI Appendix, Fig. S8A). Indeed, we observe this segment to form a single helical turn and thereafter an extended coil (Fig. 3 B and C). The sharp turn of the backbone introduced by P22 (the first observed residue of the bound peptide) ensures that the upstream peptide chain points outward toward the solvent instead of clashing with CbbL.

Our atomic model of the Rubisco–NTD_1-50 cocomplex indicates that the interaction between the NTD_1-50 peptide and Rubisco is largely mediated through polar interactions and, predominantly, hydrogen bonds. R23 forms an extensive network of interactions with the neighboring CbbL subunits. The side chain of R23 forms a salt bridge with D99 (CbbL_A) and hydrogen bonds to the hydroxyl groups of the CbbL subunits Y72 (CbbL_A) and S345 (CbbL_B) as well as to the carbonyl of G362 (CbbL_B) (Fig. 3E). The L24 amide and L26 carbonyl hydrogen bond to the carbonyl of CbbS Y96 and amide of CbbL_B F346, respectively (Fig. 3 D and F).

A water-mediated hydrogen-bonding network likely also contributes to peptide binding (SI Appendix, Fig. S9). This putative network is predominantly built up by backbone–water interactions and consists of interactions between N29 and CbbL_A Y72 (SI Appendix, Fig. S9 A and B), R23 and CbbL_B S345, and L26 and CbbL_B F346 (SI Appendix, Fig. S9C). However, in the lower resolution structure (State-2), the two waters mediating the interactions between the peptide and CbbL_B are not resolved (SI Appendix, Fig. S9D), possibly due to the slightly lower resolution of this reconstruction. Rubisco residues interacting with CsoSCA have a high conservation score among α-carboxysomal Rubiscos but are in general not conserved in β-carboxysomal Rubiscos (SI Appendix, Fig. S8B).

To determine the relative importance of the different interactions, we measured binding kinetics with a selected set of point mutations on both the NTD peptide and Rubisco (Table 1, Dataset S3, and SI Appendix, Fig. S10). The P22A mutation resulted in a dramatic loss in binding. While a protonated CbbS D94 could potentially hydrogen bond with the amide of P22, this large effect is more likely due to P22’s importance in establishing the initial alpha-helical backbone conformation of the peptide or the sharp backbone turn that is essential for binding.

Despite the many interactions made by the buried peptide residue R23, mutation of this residue to alanine yielded roughly the same K_D-value as the wild type. However, mutation of the residues on CbbL_A which interact with R23—Y72A and D99A—resulted in a 10-fold increase in K_D (mainly an effect of slower on-rate). These results are consistent with the net contribution of interactions made by R23 to binding to be quite low, but, nevertheless, this residue adversely affects binding when these interactions are not satisfied in a buried conformation. In this context, R23 may play a role in establishing the specificity of the interaction between Rubisco and CsoSCA.

The remaining hydrogen bonds between the NTD peptide and Rubisco are mediated by backbone moieties. Furthermore, the peptide is tightly packed in the cleft formed by Rubisco to form a buried interface comprising 700 Ų of the 1,200-Ų solvent accessible surface area of the peptide. Thus, given the strong negative effect of the P22A mutant on binding, shape complementarity between peptide and Rubisco appears critical to enable the extensive backbone-mediated hydrogen bonds and van der Waals interactions that drive binding.

α-carboxysome assembly is mediated by a repetitive and disordered protein, CsoS2, which is thought to bind both Rubisco and shell proteins, thus serving as a physical scaffold bridging these two major components. We previously solved the structure of Rubisco in complex with an N-terminal peptide derived from CsoS2 (CsoS2-N*) (30). Surprisingly, CsoSCA and CsoS2 bind at nearly the same location on Rubisco but utilize substantially different SLiMs and binding modes (Fig. 4 A and B).

CsoS2-N* is largely alpha-helical and binds Rubisco by spanning over the CbbL₂ dimer interface lying on top of the protein surface (Fig. 4B). The complex is highly dependent on salt bridges and cation–pi interactions. In contrast, the CsoSCA peptide is bound in a conformation turned roughly ~45 degrees and with a greater fractional buried surface area for the observed peptide (approximately 700 Ų of 1,200 Ų solvent accessible surface area, compared to 830 Ų of 2,500 Ų). CsoSCA is buried deeper into the groove between the two CbbL subunits and interacts mainly via hydrogen bonds and what appears to be an ordered network of water molecules. Notably, both peptides make significant interactions with Rubisco CbbL Y72 (Fig. 4B). This residue is conserved in α-carboxysome Rubisco but not in Rubisco from β-carboxysomes or the Form II Rubisco in H. neapolitanus, and likely contributes to specificity. Both proteins interact with Y72 via arginines; however, in CsoS2-N*, R10 is cation–pi stacked between CbbL Y72 and CbbS Y96, while in CsoSCA, R23 is positioned deeper into the structure and hydrogen bonds with CbbL Y72 and D99. Another notable feature is CbbS Y96, which in the Rubisco–CsoS2 structure is flipped ~90 degrees compared to the wild-type and Rubisco-CsoSCA structure (Fig. 4B), covering the groove between the CbbL subunits interface and enabling the conformation necessary for cation–pi interaction.

Combined, these interactions would seemingly make it impossible for CsoSCA and CsoS2 to bind to the same site of Rubisco at the same time. We have previously developed a Rubisco–CsoS2-N* fusion with all such binding sites occupied due to high local concentration of the CsoS2-N* peptide. As expected, BLI measurement indicated that NTD_1-53-sfGFP cannot bind to Rubisco when CsoS2-N* is already present, thus confirming that CsoS2 and CsoSCA compete for the same binding site (Fig. 4C). Earlier experiments from our group have demonstrated in vitro condensate formation between Rubisco and CsoS2-NTD suggesting that assembly of the α-carboxysome occurs through a condensation-like event (30). Here, we extended these experiments to include CsoSCA in the Rubisco–CsoS2-NTD condensates. These results clearly show that CsoSCA is recruited into the phase-separated Rubisco–CsoS2-NTD condensates (Fig. 4D) and demonstrate that all three proteins can simultaneously participate in such a protein interaction network.

The intrinsically disordered and repetitive protein CsoS2 acts as a scaffold between shell and Rubisco and orchestrates the assembly (34, 35) of the α-carboxysome. We previously discovered that a repeat of conserved SLiMs (four repeats/protein) found in the N-terminal portion of CsoS2 binds Rubisco and is essential for carboxysome formation (30). Here, we demonstrate that the carboxysomal carbonic anhydrase (CsoSCA) is recruited to the carboxysome via association with Rubisco. We further show that CsoSCA’s N-terminal targeting peptide and CsoS2 bind at the same site on Rubisco. Fig. 5A presents our current model of α-carboxysome assembly.

Previous work has shown that, on average, there are 450, 440, and 60 copies of Rubisco, CsoS2, and CsoSCA in a typical H. neapolitanus carboxysome, respectively (Fig. 5B) (36). The roughly 1,800 CsoS2 and 120 CsoSCA Rubisco binding motifs per carboxysome set an upper boundary of occupancy for the binding sites of Rubisco (~3,600 sites/carboxysome). Assuming all CsoS2 and CsoSCA motifs engage in binding, roughly 50% of Rubisco sites would be occupied by CsoS2, while considerably less, ~3.5%, would be occupied by CsoSCA. Although this assumes that all Rubisco sites are accessible and that all CsoS2 and CsoSCA motifs bind, both assumptions of which could be incorrect, such a calculation indicates there is likely a surplus of Rubisco sites available for binding. Our finding that CsoSCA is recruited to Rubisco–CsoS2-NTD condensates supports the hypothesis that this ternary complex is an important feature of cargo assembly in α-carboxysomes in vivo (Fig. 4D and SI Appendix, Fig. S11). Further, it has previously been shown that CsoSCA mRNA levels are lower compared to other carboxysome genes (37), suggesting that the amount of encapsulated CsoSCA is likely regulated by protein expression level rather than by competing for binding site occupancy with CsoS2. Due to the need for tight regulation of CA activity outside of the carboxysome, efficient encapsulation is vital (19) and a scenario where CsoSCA had to compete for binding could pose a physiological problem.

One specific unknown is the importance of CsoSCA multivalency imparted by its oligomeric structure and the resulting mode of interaction with Rubisco. The significantly slower dissociation rate of full-length CsoSCA (multivalent) compared to the NTD_1-53-peptide (monovalent) implies importance of multivalent protein–protein interactions (Table 1 and Fig. 2 B and C), a feature commonly observed for other IDP/R involved in phase separation (38). In terms of binding mode, bivalent binding of full-length CsoSCA could occur either between two binding sites on the same Rubisco or between two sites on different Rubisco molecules. The relatively short stretch of IDR sequence before the Rubisco binding motif and the rigidity of the folded domains presumably constrains possible binding conformations where CsoSCA binds on top or on the side in a 1:1 CsoSCA–Rubisco complex (SI Appendix, Fig. S12). Alternatively, CsoSCA could cross-link two Rubisco molecules (SI Appendix, Fig. S12). Further, previous experiments have indicated that CsoSCA is localized to the shell (20, 21). Although our data suggest that CsoSCA makes a primary interaction with Rubisco, and would likely be found throughout the carboxysome, it is possible that additional unknown protein interactions could bias CsoSCA localization toward the shell. Recent cryoelectron tomography work has been unable to unambiguously locate CsoSCA inside the carboxysome (39, 40). However, rapid advances in this technique will likely, in the near future, determine the binding conformation and localization of all such components in situ.

We could not identify a consensus Rubisco SLiM-binding motif across NTD sequences in CsoSCA homologs. The binding element identified in H. neapolitanus CsoSCA (PRLDLIEQA) is present in its most closely related homolog (Halothiobacillus sp. LS2) but is not conserved across species (further discussed in SI Appendix, Fig. S13 A and B). Many prolines are followed by R or xR, but, overall, the proteobacteria clade contains no convincing conserved motifs. In the cyanobacterial clade, a PTAPx[R/K]R motif is present in 87% of the sequences, suggesting a possible binding motif among cyanobacteria.

Surprisingly, a handful of cyanobacterial CsoSCA sequences from the Prochlorococcus genus contain the Rubisco-binding motif found in CsoS2 (RxxxxxRRxxxxxxGK) (SI Appendix, Fig. S13A), suggesting an evolutionary relationship. The lack of a consensus motif in CsoSCA homologs, coupled with the fact that CsoSCA and CsoS2 bind at the same site but with different mechanisms, reveals an evolutionary plasticity in SLiM sequence space. Across the various microbes in this phylogeny, we hypothesize that CA is recruited to its respective α-carboxysomes by the observed, versatile Rubisco binding site and does so using diverse SLiM sequences.

Recent work on both bacterial carboxysomes and algal pyrenoids suggests that Rubisco itself plays a role as an interaction hub in the ultrastructural organization of CCMs. It is now clear that not only does Rubisco interact with scaffolding proteins as a means to condensate Rubisco and form these confined CO₂-fixing organelles, it also recruits auxiliary proteins, such as CAs and activases, needed for the CCM to function. In the bacterial carboxysomal α-lineage, we have demonstrated that Rubisco binds the intrinsically disordered proteins CsoS2 (30) and CsoSCA. Additionally, it also binds the Rubisco activase CbbQO, likely via CbbO’s von Willebrand factor A domain (41, 42). The β-carboxysomal Rubisco binds its interaction partners—the scaffolding protein CcmM (43, 44), and the Rubisco activase Rca (45, 46)—via a folded domain resembling the small subunit of Rubisco (SSLD). Similar to CsoSCA, the β-carboxysomal CA, CcaA, is also recruited via a terminal peptide. However, instead of direct interaction with Rubisco, the two enzymes are linked together via the scaffolding protein CcmM (28). This convergent function may have evolutionary significance—recent results suggest that Rubisco-CA colocalization was an important step in the evolution of biophysical CCMs (47, 48). Despite convergent evolution, a notable similarity between both carboxysome lineages is the binding site on Rubisco. In known cases (except for CbbQO), the interactor binds at the same patch on Rubisco and makes contact with two different CbbL₂ dimers and one CbbS. This likely ensures that the binding partner only interacts with fully assembled CbbL₈S₈ Rubiscos during the assembly process.

In contrast to these bacterial systems, in the model algae Chlamydomonas reinhardtii, a repeat SLiM in the disordered scaffolding protein EPYC1 is essential for pyrenoid formation (49, 50) and binds on top of the small subunit via salt bridges and a hydrophobic interface (32). The sequence motif is shared among many pyrenoid proteins, suggesting a mechanism for protein targeting as well as more broadly organizing pyrenoid ultrastructure (33). The versatility in binding motif and binding site, and the convergent function of diverse Rubiscos as a hub of interaction, implies this might be a general feature, which raises a final question: Do Form IB plant Rubiscos engage in similar protein–protein interactions and do other Rubisco Forms also function as interaction hubs?

In summary, this work advances our understanding of carboxysome biogenesis and puts a focus on both the essential carbonic anhydrase and the role of Rubisco as a hub protein. This provides critical findings for engineering the carboxysome-based CCM into, e.g., crops and industrially relevant microorganisms for improved growth and yields. More broadly, we hope that the findings presented here will advance our understanding of bacterial microcompartments and promote development of their many potential biotechnological applications.

Protein sequences assigned CsoSCA (pfam08936) from all finished and permanent draft bacterial genomes available in the Integrated Microbial Genomes and Microbiomes database (51) were collected on December 12, 2019, and curated to only include proteins in an α-carboxysome operon (containing CbbL/S, CsoS2, and shell hexamers and pentamers) (412 genes). Thereafter, redundancy was reduced by removing sequences with >98% identity using Jalview, and sequences were manually curated to remove incomplete sequences, resulting in 222 sequences in the final CsoSCA dataset (Dataset S1). Sequences were aligned using MUSCLE (52). Resulting MSA was used to calculate the conservation score (53). Tree was built using IQ-TREE web server (54) and visualized using iTOL (55). Protein disorder was predicted for a subset of the dataset, including H. neapolitanus CsoSCA, using the DISOPRED3 algorithm (56). Conservation of CsoSCA NTD Rubisco binding motif was analyzed using The MEME Suite (57) (Dataset S1). MSA of α-carboxysomal Form IA Rubiscos (135 cbbL and 132 cbbS sequences) and β-carboxysomal Form IB Rubiscos (211 cbbL and 207 cbbS sequences) was constructed (Dataset S1) using, MUSCLE and visualized with WebLogo. Secondary structure prediction of NTD sequence was performed using Jpred4 (58).

Specifics regarding E. coli strain, plasmid, expression condition, and purification method for each protein in this study (CsoSCA variants, sfGFP fusions, Rubisco, shell proteins and CsoS2) are provided in Dataset S4. Protocols for protein expression and purification are fully described in SI Appendix, Method. In short, E. coli cells harboring appropriate expression plasmids were grown at 37 °C in LB-medium supplemented with appropriate antibiotics. At OD₆₀₀ = 0.4–0.6, the expression was induced, and the cells were grown overnight at 18 °C. All Rubisco constructs were coexpressed with GroEL/ES. All CsoSCA-variants, sfGFP fusions, shell proteins, and CsoS2 were purified by His-tag purification and all Rubisco variants by Strep-tag purification. To obtain pure untagged CsoSCA, purified His-SUMO-CsoSCA was cleaved using Ulp-protease. His-tag purified CsoSCA-MBP was further cleaned up by size exclusion chromatography. See SI Appendix, Methods for full purification protocols, including purification columns and buffer conditions. Protein purities were assessed by SDS-PAGE and were in general >95% pure. For storage, proteins were made to 10% (w/v) glycerol, flash-frozen in liquid nitrogen, and stored in −80 °C. The oligomeric state of CsoSCA-MBP was determined from the Superose 6 Increase chromatogram and a Gel Filtration Standard (BioRad, #1511901).

Protein–protein interactions were measured by Biolayer interferometry (BLI) using an Octet RED384 (Forte Bio). Experimental binding sequence used was the following: loading bait 60 to 240 s, buffer wash 60 s, prey association and prey dissociation (followed by sensor regeneration for Ni-NTA sensors).

Purified Rubisco-strep and NTD_1-53-sfGFP samples were exchanged into moderate-salt buffer (20 mM Tris, pH 7.5, 150 mM NaCl) using Zeba desalting columns. For cocomplexing, the protein samples were mixed at an 32:1 ratio (NTD_1-53-sfGFP:CbbL₈S₈) and incubated briefly on ice prior to injection over a 3.2/300 Superose 6 Increase column equilibrated in 20 mM Tris, pH 7.5, and 150 mM NaCl at 4 °C. The column was eluted isocratically in the same buffer with elution of total protein monitored by A₂₈₀ and elution of sfGFP-containing fractions monitored by A₄₈₅.

Binding of CsoSCA-MBP to Rubisco and shell proteins was analyzed by native-PAGE, using 4 to 15% Mini-PROTEAN® TGX™ Precast Protein Gels (Bio-Rad). Then, 2.5 μM CsoSCA-MBP was mixed with 0.5 μM Rubisco or 5 μM CsoS1A and CsoS1B and incubated for 15 min in RT. Final buffer composition was 50 mM Tris and 150 mM NaCl, pH 7.5.

CO₂ hydration catalyzed by CsoSCA-MBP was measured using the Khalifah/pH indicator assay (31) on an Applied Photophysics SX20 stopped-flow spectrophotometer at 25 °C. Saturated CO₂ solution (34 mM) was prepared by bubbling CO₂ gas into milli-Q water at 25 °C. To prevent CO₂ from escaping, a gas-tight Hamilton syringe was used to inject the solution into the stopped-flow drive syringe. MOPS and para-nitrophenol (pNP) were used as buffer–indicator pairs, and change in pH over time was detected at 400 nm using a pathlength of 1 cm. Final experimental conditions after mixing were 50 mM MOPS, pH 7.5 with the ionic strength adjusted to 50 mM with Na₂SO₄, 50 µM pNP, 17 mM CO₂, and 6 µM of CsoSCA-MBP enzyme. Steady-state kinetics was measured in the timeframe of 0.02 to 0.5 s in eight replicates, and progression curves reaching equilibrium were measured for 60 s in triplicates.

First, 0.5 μM Rubisco-strep was mixed with 0.5 mM of NTD_1-50 peptide (CsoSCA residue 1-50) in 25 mM Tris pH 7.5, 80 mM NaCl containing 2% glycerol, incubated for 20 min at room temperature, and thereafter stored on ice. Then, 3.5 μL of this sample was deposited onto freshly glow-discharged (PELCO easiGlow), Quantifoil R 1.2/1.3 200 mesh Copper TEM grids (Quantifoil Microtools) and blotted for 3 s using a Mark IV Vitrobot (FEI) after a 30-s delay under 100% humidity at 4 °C conditions before freezing in liquid ethane. The complex was visualized in a Talos Arctica (Thermo Fisher Scientific) operating at 200 keV and equipped with a K3 Summit director electron detector (Gatan) in superresolution CDS mode at 57,000×, corresponding to a pixel size of 0.69 Å. In total, 5,742 movies were acquired with the aid of SerialEM* using a defocus range between −0.6 and −1.8 μm and a 3 × 3 multishot image shift pattern. All movies consisted of 50 frames with a total dose of 50 e-/Ų. The data collection was monitored using on-the-fly processing in cryoSPARC live (Structura Biotechnology Inc., https://cryosparc.com/live) (60) to monitor microscope performance, micrograph quality, and orientation distribution of the particles on the grid.

Superresolution electron micrograph movies were aligned using MOTIONCOR2 (61) from within RELION 3.1 or using the CPU implementation of motion correction within RELION 3.1. CTF estimation was performed using CTFFIND 4.1 (62) from within RELION 3.1. Micrographs were inspected to remove poor-quality images, resulting in the higher-quality selection of 3,932 micrographs. All further processing was done from within RELION 3.1. Laplacian-of-Gaussian autopicking was used on a subset of 200 micrographs to pick approximately 75,000 particles. These particles were extracted from the micrographs with a pixel size of 2.77 Ångström and a box size of 90 pixels. 2D classification was then used to generate a higher-quality subset of particles that were used to generate an initial 3D model by way of Stochastic Gradient Descent. 3D classification of this higher-quality subset of particles gave us a good-quality 3D reference that was then used as a 3D template.

Approximately 1,300,000 particles were picked from the 3932 micrographs using our 3D reference before being extracted with a pixel size of 2.77 and a box size of 90 pixels. The particles were subjected to 3D classification applying D4 symmetry with a soft circular mask, and the best-looking classes comprising 358,785 particles were selected. The particles were then re-extracted with a pixel size of 1.37 Ångström and a box size of 180 pixels before undergoing another round of 3D classification. Again, the best classes comprising 290,762 particles were selected. The particles were re-extracted with a pixel size of 0.91 Ångström and a box size of 312 pixels before being subjected to 3D autorefinement. The refined particles were then 3D classified without additional image alignment, and the best classes comprising 262,882 particles were selected. These particles underwent CTF refinement and Bayesian polishing before being extracted with a larger 410-pixel size box. A few more rounds of 3D classification and 3D refinement, while selecting only the best classes left us with a homogeneous set of 79,562 particles. 3D refinement of this particle set gives a final resolution of 1.98 Ångstöm at Fourier shell correlation (FSC) = 0.143. RELION 3.1 reports a b-factor of about −36 Ų when sharpened with a soft mask.

The coordinate models for the two H. neapolitanus Rubisco–NTD_1-50 complex maps were built and refined similarly using a combination of COOT-v0.9.1 (63) and PHENIX-v1.19.1-4122 (64). Maps for this process were obtained by combination of the respective half-maps without filtering. Maps were molecular weight-based density modified and sharpened with phenix.resolve_cryo_em and phenix.auto_sharpen (65, 66), respectively. For ease of handling, the maps were reboxed to 160 vx³ (about 145 ų) for further use. Chains A (CbbL) and D (CbbS) from the H. neapolitanus Rubisco–CsoS2 N*-peptide cocrystal structure (PDB ID: 6UEW) (30), stripped of all ligands, were used as initial models for both maps. The initial models were rigid-body docked and manually reworked to fit the maps in COOT and the resolved portion of CsoSCA NTD_1-50 peptide built de novo. An initial round of phenix.real_space_refinement (67) was performed on models consisting of all asymmetric units with NCS constraints enforced, as well as default target bond length and angle restraints, but without secondary structure, rotamer, or Ramachandran restraints. Putative ordered water molecules were then placed interactively in COOT using maps thresholded at 2σ based on the presence of at least 2 hydrogen-bonding partners and the occurrence of the density in both half-maps. Additional rounds of phenix.real_space_refinement and manual adjustment in COOT were performed as described above to yield the final coordinate models. For the higher resolution map (State-1), residues V3-E457 of CbbL were modeled with residues V324-E329 truncated to the C_β atoms due to poor side-chain density in this region. Similarly, for the lower resolution map (State-2), residues V3-E457 of CbbL were modeled with residues H291-H300 and V323-D331 truncated to the C_β atoms. The CbbS and NTD_1-50 peptide density for both maps were modeled with residues M4-N110 and P22-A30, respectively.

Structural figures were prepared using a combination of PyMOL-v2.5 (Schrödinger, LLC.) and ChimeraX-v1.2.1 (68). Interface analysis to identify interacting residues and to calculate buried surface area was performed using the ePISA-v1.52 web server (69).

C.B. and D.F.S. designed research; C.B., E.J.D., T.G.L., J.B.T., M.D.L., S.R.S., N.V., and J.P.R. performed research; C.B., E.J.D., T.G.L., J.B.T., M.D.L., S.R.S., N.V., J.P.R., and D.F.S. analyzed data; and C.B., T.G.L., and D.F.S. wrote the paper.

The authors declare no competing interest.