A linker protein from a red-type pyrenoid phase separates with Rubisco via oligomerizing sticker motifs

Significance Diatoms are mostly marine algae responsible for up to 20% of global carbon dioxide fixation. To overcome the slow and nonspecific properties of the CO2-fixing enzyme Rubisco, it is sequestered in a subcellular compartment called the pyrenoid, allowing CO2 gas to be concentrated there. We discovered the repeat protein PYCO1, which can phase separate to form dense protein droplets in the test tube. The droplets specifically bind large quantities of diatom Rubisco resulting in much larger and less dynamic droplets. This phenomenon is caused by repeating elements on PYCO1 that oligomerize and bind to the Rubisco small subunits. Pyrenoids are ubiquitous and derived by convergent evolution. Understanding their formation provides a framework to introduce them into plants.

The slow kinetics and poor substrate specificity of the key photosynthetic CO 2 -fixing enzyme Rubisco have prompted the repeated evolution of Rubisco-containing biomolecular condensates known as pyrenoids in the majority of eukaryotic microalgae. Diatoms dominate marine photosynthesis, but the interactions underlying their pyrenoids are unknown. Here, we identify and characterize the Rubisco linker protein PYCO1 from Phaeodactylum tricornutum. PYCO1 is a tandem repeat protein containing prion-like domains that localizes to the pyrenoid. It undergoes homotypic liquid-liquid phase separation (LLPS) to form condensates that specifically partition diatom Rubisco. Saturation of PYCO1 condensates with Rubisco greatly reduces the mobility of droplet components. Cryo-electron microscopy and mutagenesis data revealed the sticker motifs required for homotypic and heterotypic phase separation. Our data indicate that the PYCO1-Rubisco network is cross-linked by PYCO1 stickers that oligomerize to bind to the small subunits lining the central solvent channel of the Rubisco holoenzyme. A second sticker motif binds to the large subunit. Pyrenoidal Rubisco condensates are highly diverse and tractable models of functional LLPS.

Rubisco | pyrenoid | CO 2 fixation | phase separation | diatoms
Almost all biological CO 2 fixation is catalyzed by ribulose 1,5-bisphosphate carboxylase/ oxygenase (Rubisco), and about half of this activity is performed by marine photoautotrophs (1). Most marine unicellular algae have compensated for Rubisco's catalytic shortcomings by evolving CO 2 concentrating mechanisms, which utilize active transport to concentrate the abundant bicarbonate ion in the chloroplast stroma (2)(3)(4). Strategically placed carbonic anhydrases then catalyze the formation of CO 2 gas in the vicinity of Rubisco active sites. To achieve significant local elevation of CO 2 gas concentrations, the volume occupied by Rubisco needs to be minimized, which has led to the convergent evolution of the pyrenoid, a micron-sized, membraneless organelle of the chloroplast stroma (5,6). Only one pyrenoid, from the green alga Chlamydomonas reinhardtii, has been studied in detail (7). Rubisco and the intrinsically disordered repeat protein essential pyrenoid component 1 (EPYC1) (8) demix in the chloroplast stroma via complex coacervation (9,10) to form a liquid biomolecular condensate (11). The pyrenoid matrix is traversed by thylakoid tubules (12,13) and contains on the order of 100 additional proteins at lower abundances (14,15), some of which possess a defined Rubisco small subunit-binding motif (16,17).
Microalgal Rubiscos do not share a common phylogeny but due to horizontal gene transfers belong to diverse green (Form IB), red (Form ID) (18), or even Form II (19) lineages. Their association with pyrenoids (20) indicates that the condensate has evolved convergently. Marine diatoms have been proposed to contribute 20% of global primary productivity (21), but molecular information on their "red-type" pyrenoids, which contain Form ID Rubisco, remains sparse (22,23). The Phaeodactylum tricornutum pyrenoid presents as rod-shaped electron-dense moiety located in the center of the single elongated chloroplast (24)(25)(26). Two appressed thylakoids bisect the structure (20), and the lumen harbors a carbonic anhydrase that provides a CO 2 point source (24,27). The pyrenoid matrix not only contains Rubisco (20) but clusters of other enzymes including fructose bisphosphate aldolase (28) and multiple additional carbonic anhydrase isoforms (26).
We report the identification and characterization of a diatom Rubisco-binding protein pyrenoid component 1 (PYCO1), which contains prion-like domains (PLD). PYCO1 phase-separates homotypically and heterotypically with Rubisco. Repeating sticker motifs on PYCO1 are responsible for its intermolecular association and for binding Rubisco. The characterization of a convergently evolved pyrenoid will expand options available to synthetic biology efforts toward enhancing photosynthetic efficiencies by engineering the CO 2 -fixing reactions (29,30).

Results
The Rubisco-Binding Protein PYCO1 Localizes to the Pyrenoid.
We raised a highly specific and sensitive peptide antibody against an epitope that is displayed on the surface of diatom Rubisco (SI Appendix, Fig. S1 A and B). Rubisco was isolated from soluble algal lysate by immunoprecipitation (SI Appendix, Fig. S1B) followed by analysis using tandem mass spectrometry (LC-MS/ MS). A control experiment performed using beads not coupled to the antibody indicated very low background binding, but the corresponding sample was not analyzed by LC-MS/MS (SI Appendix, Fig. S1B). In three of five samples, the algae were exposed to dithiobis(succinimidyl propionate) prior to lysis in order to covalently stabilize interactions.
In these experiments, a total of 89 potential Rubisco-interacting proteins were identified (SI Appendix, Fig. S1C and Dataset S1). In all samples, Rubisco had the highest peptide coverage, and the pyrenoid-localized carbonic anhydrase PtCA1 (26) was always detected (Dataset S1). However, the dataset also contained high-scoring proteins unlikely to be associated with the pyrenoid, such as the cytosolic glyceraldehyde 3-phosphate dehydrogenase GapC2 (31), stressing the need for cautious validation.
Rubisco condensation has recently been repeatedly linked to the process of liquid-liquid phase separation (LLPS) (32,33), mediated by an intrinsically disordered repeat protein, where the repeats contain conserved motifs (8) or even folded domains (34) that function as multivalent Rubisco-binding sites or stickers (35)(36)(37). Phatr3_J49957 was detected in 4/5 samples and possessed features compatible with an LLPS scaffold protein.
Amplification of the gene from genomic DNA revealed multiple amino acid substitutions and a 9-amino-acid insertion that were also found in expressed sequence tags (SI Appendix, Table S1). The protein contains six tandem repeats (R1-R6), each spanning ~80 amino acids except R3, which encodes the first half of a repeat (Fig. 1A). Similar to the green algal Rubisco linker protein EPYC1, Phatr3_J49957 was predicted to be mostly disordered, hydrophilic, and positively charged at physiological pH (pI = 10.03) ( Fig. 1C and SI Appendix, Fig. S1F). Positively charged residues are clustered in the highly conserved N-terminal half of each repeat ( Fig. 1 A-C). Unlike EPYC1, the majority of the sequence (residues 92 to 530) was classified as a PLD by the PLAAC algorithm (38) (Fig. 1C and SI Appendix, Fig. S1F), which is a common feature of condensate-forming proteins (39).
What follows implies that Phatr3_J49957's functions similarly to EPYC1 in P. tricornutum. We propose the term PYCO1 (Pyrenoid Component 1) for Phatr3_J49957 as we do not demonstrate essentiality. BLAST searches to reveal homologs in other sequenced algal species were unsuccessful, with the only hit revealing a short PYCO1 isoform encoded by the same genome (Phatr3_J40791).
We used bacterial conjugation to transform P. tricornutum with the episome pPtPuc3_FcpB_PYCO1ECFP_FcpA encoding a fusion of PYCO1 and cyan fluorescent protein (PYCO1-ECFP) (40). The fluorescent signal localized to the center of the chloroplast stroma and presented as a rod ~6 μm long and <1 μm wide ( Fig. 1D and SI Appendix, Fig. S1D). Reconstructed 3D models further demonstrated that the PYCO1-ECFP signal was embedded within the chlorophyll autofluorescence. This localization pattern is consistent with pyrenoid localization and dimensions determined by electron microscopy (24,25,41) (Fig. 1E and SI Appendix, Fig. S1 D and E). PYCO1-ECFP did not recover within 25 s when a 0.25-μm 2 region was photobleached. This indicates that the diatom pyrenoid is less dynamic compared to the liquid-like Chlamydomonas pyrenoid where near-complete recovery of EPYC1-Venus was observed within this time frame (11) (Fig. 1F and Movie S1).

PYCO1 Undergoes Homotypic LLPS In Vitro.
To understand the Rubisco-PYCO1 interaction, we optimized the recombinant production and purification of PYCO1 protein [residues 31 to 592, lacking the chloroplast transit peptide as predicted by ASAFind (42)] and its C-terminal mEGFP fusion protein, PYCO1-GFP, in Escherichia coli (SI Appendix, Fig. S2A). Extensive problems with proteolysis were eventually overcome. In the final protocol, the resuspended E. coli biomass was boiled prior to lysis, and proteins were purified from inclusion bodies that were redissolved in 8 M urea (SI Appendix, Fig. S2B). The inclusion of a C-terminal His 6 tag permitted affinity chromatography, and an N-terminal FLAG epitope was used to track the full-length protein during purification.
Prior to analysis, PYCO1 was dialyzed into 20 mM Tris-HCl pH 8.0. PYCO1 was mixed with 5% PYCO1-GFP, and incubation in the presence of at least 50 mM NaCl resulted in the immediate formation of spherical, fluorescent condensates ( Fig. 2A and SI Appendix, Fig. S2 C and D). PYCO1 droplets could be sedimented by centrifugation (Fig. 2B). The assay was used to build a phase diagram, which revealed that demixing was favored at increased protein and salt concentrations ( These findings suggest that the PYCO1-PYCO1 protein-protein interactions (PPIs) driving the phase separation are not disrupted by increased ionic strength. Charge screening may overcome the repulsion between the cationic PYCO1 molecules permitting intermolecular sticker interactions. Turbidity measurements revealed that the saturation concentration (C sat ) of PYCO1 at 150 mM NaCl was approximately 2 µM (Fig. 2D). Using quantitative confocal microscopy, the concentration of the light phase (C light ) and dense phase (C dense ) was determined to be approximately 5.7 µM and 575 µM, respectively, corresponding to a partition coefficient (PC) of ~100 at 150 mM NaCl (SI Appendix, Fig. S3 A-C and Table S2). PYCO1 droplets exhibited their liquid-like nature through coalescence events ( Fig. 2E and Movie S2) and also by FRAP experiments with a recovery half-time of 9 ± 0.8 s on fully bleached droplets of 3 µm diameter (Fig. 2F).
Hence, PYCO1, like other proteins containing PLDs, was able to phase separate in vitro potentially enabling a role as a scaffold protein that can recruit additional components of the pyrenoid.
Rubisco partitioning was salt sensitive, and at 600 mM NaCl, the enzyme was mostly found in the light phase. This indicated that in contrast to PYCO1-PYCO1 interactions, the PPIs between Rubisco and PYCO1 include electrostatic contributions (SI Appendix, Fig. S4C).
Heterotypic condensates formed at high Rubisco to PYCO1 ratios were much larger than the respective homotypic PYCO1 condensates ( Fig. 3C and SI Appendix, Fig. S4D), consistent with the increased mass of the heterotypic dense phase. Multiple eukaryotic and prokaryotic Rubiscos were assayed for their ability to form a heterotypic condensate with PYCO1, but only the P. tricornutum enzyme was competent, indicating highly specific PPIs ( Fig. 3D and SI Appendix, Fig. S5A). Homotypic PYCO1 condensates also did not partition heterologous Rubiscos, with the exception of the bacterial red-type enzyme from Rhodobacter sphaeroides (SI Appendix, Fig. S5A). Radiometric 14 CO 2 fixation assays indicated that condensed diatom Rubisco was fully  functional (Fig. 3E). Condensate appearance or Rubisco partitioning (SI Appendix, Fig. S5 B and C) was not affected by activated Rubisco or the inclusion of Rubisco's substrates (4 mM ribulose 1,5-bisphosphate, and 20 mM NaHCO 3 ).

Rubisco Reduces the Mobility of Condensate Components.
We performed FRAP experiments on heterotypic condensates loaded with various Rubisco stoichiometries. Strikingly, PYCO1 mobility decreased sharply once the Rubisco content increased beyond 1.5 Rubiscos per PYCO1 (Fig. 4A). At near-saturating stoichiometries of 2.5 Rubiscos to 1 PYCO1 molecule, recovery of the scaffold protein became very slow with a recovery half-time of ~ 9 min (Fig. 4 A and B and SI Appendix, Fig. S5D). Rubisco recovery was slow in all conditions sampled. This finding can be explained by Rubisco molecules contributing to a network formed by high-affinity Rubisco-PYCO1 interactions, whereas excess unbound PYCO1 molecules are able to diffuse more freely within the condensate.
Consistent with slow diffusive mixing of Rubisco-bound PYCO1, both droplet fusion and relaxation could be observed on a minute timescale (Fig. 4C and SI Appendix, Fig. S5E). When heterotypic droplets labeled with two different fluorescent PYCO1 fusion proteins were mixed, the signals remained in distinct sectors for 5 min following droplet fusion (Fig. 4D, SI Appendix, Fig. S5F, and Movie S3). When PYCO1-Rubisco condensates were preformed, and the PYCO1-GFP label was added subsequently, slow diffusion into the condensates was observed (Fig. 4E). Collectively, these observations indicate that there are high-affinity interactions between PYCO1 and Rubisco, but the network can still rearrange and is not arrested.
Using fluorescence microscopy, we measured the PYCO1 concentrations in droplets formed under variable stoichiometries. PYCO1 concentration was ~600 µM in homotypic droplets, and this value dropped dramatically to ~30 µM when the ratio of Rubisco to PYCO1 was 2.5 (SI Appendix, Table S2). Combining these results with the densitometric analysis ( Fig. 3B and SI Appendix,  Fig. S4B) permitted the estimation of Rubisco concentrations (SI Appendix, Table S3). The heterotypic droplets contained ~60 µM/ ~30 mg/mL of Rubisco (Fig. 4F). Formation of heterotypic condensates resulted in lower light phase concentrations for PYCO1 (SI Appendix, Table S2). The variability in saturation concentration and partition coefficients is a feature of heterotypic condensates that emerges as a result of the different valencies and affinities of the interactions involved (43,44).
Rubisco:PYCO1 stoichiometry could strongly affect phase separation. At equimolar concentrations, both proteins were mostly found in the soluble fraction, and droplets were small (Fig. 4 G  and H and SI Appendix, Fig. S4B). As the network dominated by PYCO1-PYCO1 interaction transitions to a PYCO1-Rubisco network, the 1:1 stoichiometry results in a dramatic change in the phase diagram. Only when more Rubisco-binding sites are added to the system, the condensates become stabilized.

Identification of the Small Subunit-Binding PYCO1 Stickers.
The addition of recombinant PYCO1 protein brought about a concentration-dependent native gel shift of diatom but not of proteobacterial Rhodobacter sphaeroides Rubisco (Fig. 5A). However, when we used a chimeric Rubisco composed of proteobacterial large, and diatom small subunits (RsLPtS) (SI Appendix, Fig. S6A), the gel shift was restored, indicating that the responsible PYCO1binding site localizes to the Rubisco small subunit. Interestingly, although RsLPtS-PYCO1 condensates could be sedimented, no regular spherical droplets were observed. Instead, the condensates formed aggregated clusters of small droplets that failed to fuse (SI Appendix, Fig. S6 B-D).
To localize the Rubisco small subunit-binding site of PYCO1, we produced a series of fragments (SI Appendix, Fig. S6A) and used the gel shift assay to assess Rubisco binding. PYCO1(452 to 592), which encompassed the final repeat and C-terminus, reduced the electrophoretic mobility of diatom Rubisco, indicating the formation of a defined complex (Fig. 5B). In contrast, variants of the fragment where the repeating KWSPR motif was deleted, or W476 substituted with alanine, did not alter diatom Rubisco migration in this assay (Fig. 5B). This was also the case for the fragment PYCO1(483 to 592), which encoded the C-terminal 109 residues that follow the last repeat (Fig. 5B). These data suggested that PYCO1 binds to the Rubisco small subunit via a small linear interacting motif (45) that includes a critical tryptophan residue.  The Structure of a Rubisco-PYCO1 Fragment Complex Reveals a Sticker Tetrad. We determined the structure of the Rubisco-PYCO1(452 to 592) complex using cryo-electron microscopy ( Fig. 5 C-H and SI Appendix, Fig. S7). The resulting density map revealed that distinct PYCO1 motifs bind to both the small (Fig. 5D in blue; 2.6 Å resolution) and the large subunits (Fig. 5 E and F in magenta; 2.5 Å resolution) of Rubisco (2.0 Å resolution). The P. tricornutum Rubisco structure overall was highly similar to the diatom Rubisco structures reported recently (46), with rmsd of 0.865 Å for all Cα atoms of the large subunit compared to the Thalassiosira antarctica Rubisco structure (PDB:5MZ2). Density for several posttranslational modifications could also be identified, all of which were described in the earlier diatom Rubisco structures (SI Appendix, Fig. S8).
Four instances of the previously identified sticker motif "KWSPRGGS" could be modeled into the density map surrounding Rubisco's central solvent channel formed by the small subunits (Fig. 5D). The motifs adopt a helical conformation and collectively form a square-shaped plug of the central solvent channel (Fig. 5  C and H). The indole group of each W476 binds to a groove formed by the "GGS" of the adjacent sticker, connecting the corners of the square. Because mutation of W476 abolished Rubisco binding (Fig. 5B), it appears likely that the interaction between at least two PYCO1 stickers is essential for heterotypic condensate formation.
The interaction between the small subunits of Rubisco and the PYCO1 sticker tetrad is mediated by a network of hydrogen bonds. The side chain of SSU N105, which is located in the loop connecting βD and βE, hydrogen bonds with the side chain of PYCO1 S482. Two more residues in the SSU C-terminal extension form hydrogen bonds to the PYCO1 motif. The N133 carboxamide forms an interaction with the backbone of PYCO1 S477 (Fig. 5 D and G). PYCO1 S477 has a second interaction with the G137 backbone via its side chain.
The structure revealed an additional interaction of the Rubisco large subunit with the PYCO1 fragment, the motif "AAEWGSMNQ" found near the C-terminus of PYCO1 (residues 577 to 585). The element adopted a helical conformation and bound to a largely hydrophobic cleft formed between the two domains of the large subunit near Rubisco's dimer-dimer interface (Fig. 5 C, E, F, and H). There is a hydrogen bond between the side chains of PYCO1 N584 and LSU Y89. The indole group of PYCO1 W580 is inserted into a hydrophobic pocket formed by LSU residues F102, F104, L361, F366, and F482. W580 also participates in a hydrogen bond to D92 (Fig. 5 E and F). The C-terminus of PYCO1 harbors an additional similar motif (ASEWASMNT residues 532 to 540) that may also bind in this way. However, the interaction between Rubisco and these motifs is relatively weak because mutating the SSU-binding motif of PYCO1(452 to 592) was sufficient to eliminate the observed native PAGE shift. The C-terminal fragment containing these motifs did not alter Rubisco's electrophoretic mobility (Fig. 5B).

A Tyrosine-Repeat Motif Is Essential for PYCO1 Condensate
Formation. The observed homotypic phase separation of PYCO1 predicts the existence of multivalent stickers that mediate homotypic interactions (36). Inspection of the low-complexity region of PYCO1 revealed the presence of five completely conserved GTGYNP motifs found in all repeats except for the truncated R3 (Fig. 1A). Given the emerging importance of tyrosine residues in mediating interactions leading to LLPS in multiple systems (47)(48)(49)(50), we mutated the corresponding five tyrosine residues to alanine and produced PYCO1(Y→A). We also produced PYCO1(W→A), where all six tryptophan residues of the SSU-binding KWSPR motif were substituted with alanine, and PYCO1(W/Y→A), where all eleven mentioned aromatic residues were substituted with alanine (Fig. 6A).
When compared to the wild-type protein, all three variants were strongly impaired in their ability to form homotypic condensates (Fig. 6 B, C, and E). PYCO1(Y→A) phase separation could still be observed at high salt or protein concentrations (Fig. 6 B and C). The strong impairment of PYCO1(W→A) homotypic phase separation revealed that the Rubisco-binding motif is also involved in homotypic phase separation and raises the possibility that the sticker-sticker interaction observed in the Rubisco-PYCO1(452 to 592) structure already forms in the homotypic condensate.
Next, we assessed heterotypic condensate formation. The interaction of the Y→A variant with Rubisco was comparable to wild type, as assessed by the native-PAGE gel shift assay (Fig. 6D). As expected from our structure, this was not true for W→A and W/Y→A, which did not completely shift Rubisco in this assay. Heterotypic condensate formation and Rubisco recruitment were robust for Y→A but essentially eliminated for W→A and W/Y→A (Fig. 6 F-H).
In summary, our data indicate that both the tyrosine residues embedded in the repeating GTGYNP motifs and the KWSPR/Q tryptophans are involved in homotypic condensate formation and thus will be involved in intermolecular and intramolecular PYCO1 interactions. Formation of the heterotypic condensate requires the tryptophan of the SSU-binding motif to be intact. The behavior of the Y→A variant suggests that homotypic phase separation is not essential for heterotypic condensate formation.

Discussion
We present a biochemical framework for formation of the Form ID Rubisco condensates found in diatoms, representatives of the red plastid lineage. The identification and characterization of PYCO1 indicates that the convergently evolved red pyrenoid matrix, like its counterpart in green algae (8,11), is formed by LLPS mediated by an intrinsically disordered repeat protein that contains multiple sticker motifs that bind to the Rubisco large and small subunits.
The green algal EPYC1 protein does not form homotypic condensates under comparable conditions but requires the presence of interacting Rubisco to demix (9). In contrast, PYCO1 is a protein containing PLDs that can undergo homotypic phase separation. The underlying interactions involve aromatic residues, and the homotypic condensates can subsequently specifically bind diatom Rubisco. The green algal Rubisco linker protein EPYC1 possesses five evenly spaced helical sticker motifs, which specifically bind to the two αhelices of the small subunit, where one sticker will bind one small subunit. The EPYC1 sticker forms salt bridges to both helices via weak interactions (10). The resulting droplets are spherical, and the pyrenoid matrix components are highly mobile (9,11). In PYCO1, the six small subunit-binding stickers are unevenly distributed (Fig. 1A). Four stickers can interact to form a tetrad and stably bind to the small subunits lining Rubisco's central solvent channel via a hydrogen bond network. The interaction between the stickers is mediated by a critical tryptophan residue. Sticker oligomerization stabilizes the interaction between PYCO1 and Rubisco resulting in high-affinity interactions and a less dynamic Rubisco condensate.
Variations of the EPYC1 sticker motif are found in other Chlamydomonas pyrenoid components, possibly providing a universal pyrenoid sticker that appears to organize the green pyrenoid (16). In contrast, we did not encounter the KWSPR/Q sticker motif in other Phaeodactylum proteins, other than in the short Relative Absorbance (340 nm)  Formation of the dense (~30 mg/mL) homotypic PYCO1 condensate was driven by salt and could be disrupted by mutation of aromatic residues, including the tryptophan critical for Rubisco binding. PYCO1 is mobile in the homotypic condensate, indicating that the underlying interactions are weak (Fig. 6I). It is possible that the sticker oligomerization observed in the Rubisco-bound state already exists and stabilizes the homotypic condensates. In addition, we note a similarity to the behavior of mussel foot proteins (mfps). These are positively charged proteins containing aromatic residues that form homotypic condensates with increasing salt. The salt reduces the electrostatic repulsion between mfps but does not disrupt the π-cation interactions that drive the assembly (51,52).
As Rubisco is added to the condensate, its properties change. The PYCO1 concentration in the dense phase decreases dramatically, indicating that PYCO1-PYCO1 interactions are being substituted for PYCO1-Rubisco interactions (Fig. 6I). At a 1:1 stoichiometry, the condensate is destabilized. The sticker stoichiometry observed in the structure suggests that four PYCO1 molecules (with 24 KWSP motifs) would be able to completely saturate the small subunit-binding sites of three Rubisco particles. This ratio may thus favor the formation of stable smaller oligomers. This concept has been theoretically explored as a "magic-number effect" in the context of biomolecular condensates previously (11,53). At high Rubisco/PYCO1 ratios, PYCO1 mobility is drastically reduced, indicating that all molecules are incorporated into the heterotypic network. In the heterotypic condensate corresponding to 2.5 Rubisco molecules (8 × 2.5 = 20 small and 20 large subunit-binding sites) to 1 PYCO1 (six small subunit stickers and two large subunit stickers), only a fraction of Rubisco-binding sites would be occupied. As PYCO1 small subunit sticker oligomerization appears critical, only one side of the Rubisco particle may thus be bound in the observed network (Fig. 6I).
The Phaeodactylum pyrenoid is not spherical but adopts a rod-like shape, and PYCO1-ECFP has low mobility ( Fig. 1 E and F). We hypothesize that the viscous material properties of the Rubiscosaturated PYCO1 condensate resemble the algal pyrenoid matrix.
This may permit the compartment, which is organized around a central thylakoid membrane (24), to maintain its distinctive rod-shaped structure. At the same time, changing the stoichiometry of matrix components provides one ready mechanism to change the properties of the compartment (54), and this may then facilitate processes such as pyrenoid division (11).

Materials and Methods
Detailed methods and protocols on plasmid construction, protein purification, algal biology, immunoprecipitation and mass spectrometry, light microscopy, biochemical assays, and cryo-electron microscopy are provided in SI Appendix. Most proteins were produced recombinantly in E. coli, except for Rubisco, which was purified directly from diatoms. P. tricornutum was transformed by bacterial conjugation (40).
Data, Materials, and Software Availability. Data acquired and analysed during this study are included in this manuscript or available at https://researchdata. ntu.edu.sg/dataverse/cajar (55). Density maps from cryo-electron single particle analysis have been deposited into the Electron Microscopy Data Bank with the accession code EMDB-33887 (56) EMDB-35166 (57) EMDB-35159 (58) and EMDB-35158 (59) Molecular models have been deposited into the Protein Data Bank under the accession code PDB 7YK5 (60). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (61) with the dataset identifier PXD027027 (62).