Main

The growth of photosynthetic organisms is commonly limited by the rate of CO2 fixation catalysed by the enzyme Rubisco1. Nearly all eukaryotic algae and some plants overcome this limitation with a chloroplast-localized organelle called the pyrenoid, which enhances CO2 fixation (Fig. 1a)2. The prevalence of pyrenoids in eukaryotic algae has led to the estimate that pyrenoids mediate 61–88% of primary productivity in the oceans3, making the pyrenoid an organelle of major biogeochemical importance. Furthermore, engineering a pyrenoid into major C3 crops such as wheat and rice is expected to enhance their CO2 uptake, yield and water use efficiency4, which will promote agricultural resilience in the face of climate change5.

Fig. 1: MITH1 and SAGA1 are necessary for pyrenoid matrix-traversing membranes in Chlamydomonas.
figure 1

a, The Chlamydomonas pyrenoid is found at the base of the cell’s cup-shaped chloroplast and consists of a Rubisco matrix traversed by membrane tubules that join in a reticulated region at the centre12. The pyrenoid is surrounded by a starch sheath thought to act as a barrier against CO2 escape31. b, Flat sheets of thylakoid membrane transition into cylindrical tubules as they enter the pyrenoid matrix through gaps in the starch sheath, according to cryo-electron tomography data12. Bicarbonate (HCO3) is thought to be concentrated in the lumen of pyrenoid matrix-traversing membranes and converted to CO2 by carbonic anhydrase 3 (CAH3) (ref. 22). cf, TEM images of whole Chlamydomonas cells and pyrenoids. The strains include (c) the wild-type CMJ030 (ref. 46), (d) a mith1 mutant36, (e) a complemented strain generated by transforming MITH1-Venus into the mith1 mutant background (see Methods) and (f) a saga1 mutant13. Representative images from 3 biological repeats are shown. Arrowheads point to pyrenoids. t, traversing membrane tubules; m, matrix; s, starch. Images in each row are presented at the same scale.

Pyrenoids enhance CO2 fixation by supplying Rubisco with concentrated CO2. Almost all known pyrenoids share two ultrastructural features: a protein matrix and specialized membranes (Fig. 1a)2. The matrix is the site of CO2 fixation by Rubisco, and its biogenesis has recently been elucidated in the model alga Chlamydomonas reinhardtii (Chlamydomonas hereafter); it is a spheroidal liquid-like condensate6 composed primarily of Rubisco and an intrinsically disordered repeat linker protein, EPYC1 (refs. 3,7). EPYC1 and a compatible Rubisco were sufficient to reconstitute a matrix-like condensate in vitro8 and in the plant Arabidopsis thaliana (Arabidopsis hereafter), which does not natively contain a pyrenoid9.

The pyrenoid’s second ultrastructural feature consists of specialized membranes that traverse the matrix and have been hypothesized to perform the essential function of supplying concentrated CO2 to Rubisco (Fig. 1b)10. By contrast with the matrix, the biogenesis and function of these membranes have not been experimentally characterized in any organism. More broadly, matrix-traversing membranes represent an organellar configuration whose biogenesis has not yet been characterized at a molecular level in any system, where membranes traverse a phase-separated condensate. In the present work, we seek to address a basic question: which factors cause membranes to traverse the pyrenoid matrix?

mith1 and saga1 mutants lack matrix-traversing membranes

A high-throughput mutant screen in Chlamydomonas recently identified MITH1 (MIssing THylakoids 1, Cre06.g259100, SAGA3) as a gene required for normal growth in low CO2, a characteristic of pyrenoid-related genes11. In the present study, using transmission electron microscopy (TEM), we found that mith1 mutants lacked normal matrix-traversing membranes (Fig. 1 and Extended Data Fig. 1). In wild-type Chlamydomonas, matrix-traversing membranes appear as cylindrical tubules that merge with the photosynthetic thylakoid membrane sheets outside the pyrenoid12 (Fig. 1a–c). In the mith1 mutant, matrix-traversing membranes were almost never observed (Fig. 1d and Extended Data Fig. 1). To our knowledge, this is presumably the first observation of a mutant with a phenotype specifically affecting the matrix-traversing membranes. In the complemented strain mith1;MITH1-Venus, an apparently normal matrix-traversing membrane tubule network is restored (Fig. 1e), demonstrating that MITH1 is necessary for the formation of matrix-traversing membranes.

The discovery of a role for MITH1 in matrix-traversing membrane biogenesis suggested a similar function for its homologue, SAGA1 (ref. 11) (Cre11.g467712), a protein that is also required for growth under low CO2 but whose function had remained elusive due to the saga1 mutant’s complex phenotypes that impacted multiple pyrenoid subcompartments13. The idea that SAGA1 functions in the same matrix-traversing membrane biogenesis pathway as MITH1 is supported by the observation that saga1 and mith1 mutants displayed a similar pattern of growth defects across 121 different conditions in a high-throughput study11. Further consistent with a possible function for SAGA1 in producing matrix-traversing membranes, the saga1 mutant exhibited a similar lack of matrix-traversing membranes to mith1 (ref. 13) (Fig. 1f and Extended Data Fig. 1).

MITH1 and SAGA1 produce matrix-traversing membranes in plants

To determine whether MITH1 and SAGA1 are sufficient for producing pyrenoid matrix-traversing membranes, we expressed the two proteins in the heterologous system of Arabidopsis thaliana plants (Fig. 2 and Extended Data Figs. 24). While wild-type Arabidopsis chloroplasts lack a pyrenoid, a synthetic Chlamydomonas-like pyrenoid matrix was previously reconstituted in Arabidopsis chloroplasts by expressing the linker protein EPYC1-GFP and the EPYC1-binding Chlamydomonas Rubisco small subunit CrRBCS2 (Fig. 2a,f and Extended Data Fig. 4c)9. Here we transformed these CrRBCS2;EPYC1-GFP plants containing a synthetic matrix with SAGA1-mCherry and chloroplast-targeted MITH1-mCerulean (Fig. 2b–e,g–i and Extended Data Fig. 4). Whereas membranes had never previously been observed within condensates of the parental CrRBCS2;EPYC1-GFP Arabidopsis line9 (Fig. 2a), we detected chlorophyll fluorescence streaks inside the condensates of the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean line, suggesting that they contain thylakoid membranes (Fig. 2b,e). TEM confirmed that membranes traverse the matrix condensates in the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean line (Fig. 2g and Extended Data Fig. 4a,b). We observed traversing membranes in 32% of CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean condensates (n = 166; Fig. 2g), but in none in the condensates of the CrRBCS2;EPYC1-GFP line (n = 117; P < 0.0001, Fisher’s exact test; Fig. 2f). These results demonstrate that SAGA1 and MITH1 together are sufficient to generate matrix-traversing thylakoid membranes in a heterologous system.

Fig. 2: MITH1 and SAGA1 are sufficient to cause membranes to traverse the Rubisco matrix in Arabidopsis.
figure 2

ad, Fluorescence images depicting Arabidopsis lines expressing different combinations of Chlamydomonas proteins. Images are presented at the same scale. E1, EPYC1-GFP; C, chlorophyll; S1, SAGA1-mCherry; M1, MITH1-mCerulean; CrR2, CrRBCS2, the Chlamydomonas Rubisco small subunit. Arrowheads in d point to MITH1 signal found outside the EPYC1-GFP condensate. e, High-resolution fluorescence images of a condensate found in the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean line. Arrowhead points to matrix-traversing chlorophyll signal. fi, TEM images of CrRBCS2/EPYC1 matrix condensates in the same lines depicted in ad. Images are presented at the same scale. Arrowhead in g points to matrix-traversing membrane. j, SBF–SEM sections of a condensate in the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean line and 3D reconstruction of the SBF–SEM data, highlighting matrix-traversing membranes in green. k, Example TEM image of the Arabidopsis line CrRBCS2;EPYC1-GFP;MITH1-mCerulean highlighting the appearance of two membrane sheets, each with its own lumen. Graph depicts grey values measured across a distance indicated by the blue line in the membrane zoom image.

Neither SAGA1 nor MITH1 alone was sufficient to produce traversing membranes when expressed in the CrRBCS2;EPYC1-GFP Arabidopsis line (Fig. 2c,d,h,i). In a separate manuscript focused on reconstituting a pyrenoid starch sheath in Arabidopsis14, extensive characterization of the CrRBCS2;EPYC1-GFP;SAGA1-mCherry line established that its condensates contain starch granules as well as a network of lower-electron-density regions that contain SAGA1 but not membranes (Fig. 2h). When we additionally expressed MITH1 in this line, we no longer observed this network or starch granules in the condensates (Fig. 2g). Since neither the network nor starch granules inside the matrix are present in wild-type Chlamydomonas cells, we favour the view that both features are aberrant consequences of SAGA1 expression in the absence of other relevant Chlamydomonas proteins.

In the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean line, traversing membranes appeared as straight lines that typically spanned the entire length of the matrix in the TEM cross-sections (Fig. 2g and Extended Data Fig. 4a,b), suggesting that the matrix-traversing membranes in Arabidopsis are arranged in sheets. Three-dimensional (3D) reconstruction of serial block-face scanning electron microscopy data further demonstrated that the traversing membranes were arranged in sheets that in some instances spanned more than 750 nm in width and depth (Fig. 2j and Supplementary Videos 15). TEM suggested that the traversing membranes typically consisted of two parallel thylakoid-like sheets, each with its own lumen (Fig. 2k and Extended Data Fig. 4e,f).

While the sheet-like morphology we observed in Arabidopsis was distinct from the tubular morphology of the Chlamydomonas matrix-traversing membranes (Fig. 1b,c), it was remarkably similar to that of the most-commonly-observed morphology of matrix-traversing membranes in most other algal species, including those of hornworts15, chlorella16, diatoms17 and chlorarachniophytes18. Our observations therefore indicate that the process of causing membrane sheets to traverse a matrix is distinct from the process of shaping those membranes into tubules, and that while SAGA1 and MITH1 appear to not be sufficient to shape membranes into tubules, they are both necessary and sufficient to induce membrane sheets to traverse a matrix.

The discovery of proteins that are both necessary and sufficient for the formation of matrix-traversing membranes now enables investigation in Chlamydomonas and Arabidopsis of how these proteins produce matrix-traversing membranes.

MITH1 and SAGA1 function at matrix-traversing membranes

We examined the localization of MITH1-Venus in Chlamydomonas using the complemented mith1;MITH1-Venus background, where the tagged protein is functional (Fig. 1e). Whereas MITH1 was previously proposed to localize to the Chlamydomonas pyrenoid matrix19,20, here we observed that it localizes to and functions at the matrix-traversing membranes. MITH1 localized in a branching pattern in the pyrenoid similar to that of RBMP1, a known matrix-traversing membrane protein21, and distinct from the homogeneous localization of the Rubisco small subunit (RBCS1), which is known to occupy the matrix7 (Fig. 3a–c and Extended Data Fig. 5a–c). MITH1 co-localized with faint chlorophyll fluorescence from the matrix-traversing membranes (Fig. 3c and Extended Data Fig. 5c) and with CAH3, a matrix-traversing membrane lumen-localized carbonic anhydrase22 (Fig. 3h and Extended Data Fig. 5h). We observed MITH1 fluorescence signal in a broader region than RBMP1, chlorophyll or CAH3 (Fig. 3b,c,h and Extended Data Fig. 5b,c,h). This observation could be due to MITH1 taking up space on the outside of the membranes, whereas RBMP1 is a predicted transmembrane protein, chlorophyll is probably associated with transmembrane proteins, and CAH3 is in the lumen. Chlamydomonas cell fractionation also supported membrane localization, as MITH1 co-pelleted with membranes in the absence of detergent and was solubilized when detergent was added (Fig. 3k). MITH1 localized to the matrix-traversing membranes in CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean Arabidopsis (Fig. 2e and Extended Data Fig. 4c), further supporting a membrane localization. Together, these results demonstrate that MITH1 localizes to the matrix-traversing membranes when SAGA1 is present.

Fig. 3: SAGA1 and MITH1 localize to membranes, bind each other and perform distinct functions in matrix-traversing membrane biogenesis.
figure 3

ag, Localizations of Venus-tagged proteins in various Chlamydomonas mutant backgrounds. Transparent arrowheads point to membranes inside the matrix; filled arrowheads point to membranes at the matrix entry sites. Images are presented at the same scale. hj, Two-colour immunofluorescence was performed for anti-FLAG and anti-CAH3 (h,i) and anti-FLAG and anti-MITH1 (j). Images are presented at the same scale. k, Cell fractionation and western blot in the absence and presence of 1% DDM detergent; Pel, pellet; Sup, supernatant; Fd, ferredoxin. Representative images from 3 biological repeats are shown. l, Spot test depicting growth at varying levels of CO2 and 100 µmol photons m−2 s−1 light levels. mo, Localization of CYN7-Venus in wild-type and mutant backgrounds. Images are presented at the same scale. p, TEM of wild-type and mith1 mutant pyrenoids. Arrowheads point to matrix surface-associated membranes. q,r, Spectral counts of proteins identified by mass spectrometry after immunoprecipitation of MITH1-Venus-3×FLAG (q) or SAGA1-Venus-3×FLAG (r), plotted against the spectral counts of the same proteins after immunoprecipitation of Venus-3×FLAG as a control for non-specific binding. Non-labelled points are partially transparent to facilitate visualization of overlapping points. Results from 2 replicates were averaged. sv, Model for how SAGA1 and MITH1 produce matrix-traversing membranes. s, Without either protein, membranes are unable to enter the Rubisco matrix. t, SAGA1 initiates contact between thylakoid membrane sheets and the surface of the pyrenoid by binding to membranes and Rubisco. u, SAGA1 recruits MITH1, which increases the binding force per unit area between membranes and matrix, outcompeting EPYC1–Rubisco interactions and allowing extension of membrane sheets into the matrix. v, In Chlamydomonas, additional factors shape the membrane sheets into cylindrical tubules.

We had previously found that SAGA1-Venus localizes to puncta and streaks at the Chlamydomonas pyrenoid periphery, and had hypothesized that this localization was mediated by its CBM20 starch-binding domain13, which could bind the starch sheath that surrounds the Chlamydomonas pyrenoid (Fig. 1a). However, here we observed that SAGA1 localized to pyrenoid-periphery puncta even in the starchless sta6 mutant background (Fig. 3e,f and Extended Data Fig. 5e,f), indicating that SAGA1 does not require starch for its normal localization.

We observed that the Chlamydomonas SAGA1-Venus puncta co-localized with the outer extremities of the matrix-traversing membranes as visualized through chlorophyll streaks (Fig. 3e,f and Extended Data Fig. 5e,f), CAH3 (Fig. 3i and Extended Data Fig. 5i) and MITH1 (Fig. 3j and Extended Data Fig. 5j). In the saga1;SAGA1-Venus strain, we observed a gap between the SAGA1-Venus fluorescence and the chlorophyll fluorescence outside the pyrenoid, which we did not observe in the starchless saga1;sta6;SAGA1-Venus strain (Fig. 3e,f and Extended Data Fig. 5e,f), indicating that the prominent SAGA1 puncta and streaks were at the sites where the matrix-traversing membranes enter the pyrenoid matrix on the inner face of the starch granules. Furthermore, we observed faint SAGA1 signal deeper inside the pyrenoid, in matrix-traversing streaks that co-localized with the matrix-traversing chlorophyll streaks (Fig. 3f and Extended Data Fig. 5f, transparent arrowheads) and with MITH1 (Fig. 3j, transparent arrowheads). In the cell fractionation experiment, SAGA1 was present both in the membrane pellet and the supernatant, and the proportion in the supernatant increased upon addition of detergent, suggesting that while SAGA1 interacts with membranes, a portion of the protein dissociates from membranes during the experiment or exists in a distinct soluble state (Fig. 3k). Our results indicate that in Chlamydomonas, SAGA1 localizes throughout the matrix-traversing membranes and is enriched in puncta at their periphery.

Further supporting a matrix-traversing membrane localization of SAGA1, in the CrRBCS2;EPYC1-GFP;SAGA1-mCherry;MITH1-mCerulean Arabidopsis line, SAGA1-mCherry localized along the EPYC1-GFP condensate-traversing chlorophyll streaks (Fig. 2e and Extended Data Fig. 4c). A notable difference from Chlamydomonas was that in the Arabidopsis system, SAGA1 signal was homogeneous across the matrix-traversing membranes, while in Chlamydomonas SAGA1 was enriched in puncta at the periphery of the pyrenoid and only faint SAGA1 signal was observed in the matrix-traversing membranes deeper inside the pyrenoid (Fig. 3e,i,j). One possible explanation for this observation is that SAGA1 may have a preference for localization to sheet-like membranes over tubules. Thus, in Chlamydomonas, SAGA1 would preferentially localize to the peripheral regions of the matrix-traversing membranes, where these membranes transition from sheets into tubules (Fig. 1b), but in Arabidopsis, it would localize throughout the matrix-traversing membranes, as they are sheets along their full length (Fig. 2j). In contrast to SAGA1, MITH1 localized throughout the matrix-traversing membranes in both Arabidopsis (Fig. 2e) and Chlamydomonas (Fig. 3c,h), which could potentially reflect a different membrane curvature preference from that of SAGA1. We conclude that MITH1 and SAGA1 localize to and function at the matrix-traversing membranes.

MITH1 requires SAGA1 for localization and function

The Chlamydomonas saga1 mutant displayed a more-severe growth defect under air CO2 conditions than the mith1 mutant (Fig. 3l). The difference in the severity of the mutant growth defects suggests that the two proteins play distinct roles in producing matrix-traversing membranes. To determine the order in which SAGA1 and MITH1 act to produce matrix-traversing membranes, we tested their epistasis by generating a mith1;saga1 double mutant. We compared the growth and cell morphology of the single and double mutants using spot tests, light microscopy and TEM. The mith1;saga1 double mutant displayed similar growth and morphological defects to the saga1 single mutant (Fig. 3l and Extended Data Fig. 6), indicating that saga1 is epistatic to mith1 and suggesting that MITH1 requires SAGA1 to function.

Consistent with this dependence, we found that when we expressed each Venus-tagged protein in a mith1;saga1 double mutant, SAGA1 still localized to the pyrenoid in the absence of MITH1, but not vice-versa (Fig. 3d,g and Extended Data Fig. 7). When SAGA1-Venus was expressed without MITH1 in this background, it showed normal localization to pyrenoid puncta (Fig. 3g and Extended Data Fig. 7). In contrast, in the absence of SAGA1, MITH1-Venus mislocalized throughout the chloroplast (Fig. 3d and Extended Data Fig. 7).

Consistent with observations in Chlamydomonas, in Arabidopsis plants that lacked MITH1-mCerulean, SAGA1-mCherry still localized to the EPYC1-GFP condensate (Fig. 2c and Extended Data Fig. 4c), but in plants that lacked SAGA1, MITH1-mCerulean was dispersed throughout the chloroplast (Fig. 2d and Extended Data Fig. 4c). Considering the absence of other pyrenoid factors in the Arabidopsis system, these results show that SAGA1 is not only necessary but also sufficient for recruiting MITH1 to the EPYC1-Rubisco condensate. Altogether, our findings demonstrate that whereas SAGA1 can localize to the EPYC1-Rubisco condensate independently of MITH1, MITH1 requires SAGA1 for its recruitment to the condensate and for its function.

SAGA1 is necessary for initiation of traversing membranes

To distinguish the roles of SAGA1 and MITH1 in matrix-traversing membrane biogenesis, we characterized matrix-traversing membrane defects in the Chlamydomonas saga1 and mith1 mutants using the protein CYN7-Venus as a marker for matrix-traversing membranes (Fig. 3m–o). CYN7 (Cre12.g544150) is a predicted thylakoid luminal peptidyl-prolyl isomerase that was shown to be enriched in the matrix-traversing membranes of wild-type cells23, a localization that we confirmed here (Fig. 3m and Extended Data Fig. 8a,b). In the saga1 mutant, CYN7-Venus was dispersed throughout the chloroplast (Fig. 3o and Extended Data Fig. 8d,e), demonstrating that a normal matrix-traversing membrane network was not present in cells of this mutant and indicating that SAGA1 is necessary for initiation of matrix-traversing membranes.

MITH1 is necessary for extension of traversing membranes

Unlike in the saga1 mutant, in the mith1 mutant, CYN7-Venus localized to streaks and puncta associated with the periphery of the pyrenoid (Fig. 3n and Extended Data Fig. 8c,e). These observations suggest that pyrenoid membrane formation initiates at the correct cellular location in the mith1 mutant but then stalls before these membranes are extended through the pyrenoid matrix. Consistent with this idea, several TEM images of the mith1 mutant showed thylakoid membranes crossing the pyrenoid starch sheath and stopping at the surface of the matrix (Fig. 3p, arrowheads). SAGA1-Venus’s punctate localization when expressed in the double mutant (Fig. 3g) suggests that SAGA1 is present in these streaks and functions to form them.

SAGA1 and MITH1 physically interact

The results presented so far indicate that SAGA1 and MITH1 together are necessary and sufficient to cause membranes to traverse the matrix, and that both proteins localize to these membranes. To understand how SAGA1 and MITH1 cause membranes to traverse the matrix, we sought to understand the proteins’ interactions with matrix and with each other.

SAGA1 physically interacts with both Rubisco large and small subunits by yeast two-hybrid assay13, and it contains two Rubisco-binding motifs whose binding to Rubisco we previously confirmed in vitro21. We could not detect evidence of direct MITH1 physical interaction with Rubisco or EPYC1 by yeast two-hybrid assay, and the lack of an interaction is supported by the absence of detectable MITH1 in the EPYC1-Rubisco condensates of the CrRBCS2;EPYC1;MITH1 lines (Fig. 2d and Extended Data Fig. 4c).

To test whether SAGA1 can bind MITH1 in Chlamydomonas, we performed immunoprecipitation–mass spectrometry with SAGA1 and MITH1 as bait proteins. SAGA1 was MITH1’s most-abundant specific interactor, and MITH1 was one of SAGA1’s most-specific interactors, indicating that the two proteins physically interact (Fig. 3q,r and Supplementary Table 5). Taken together with our observation that MITH1 requires SAGA1 for localization to the pyrenoid in the Arabidopsis system (Fig. 2b,d), these results suggest that SAGA1 recruits MITH1 to the pyrenoid via direct binding.

A proposed model for matrix-traversing membrane biogenesis

On the basis of the above results, we propose a model for how SAGA1 and MITH1 cause thylakoid membranes to traverse the pyrenoid matrix (Fig. 3s–v). We posit that in the absence of SAGA1 and MITH1, it is energetically unfavourable for thylakoid membranes to traverse the matrix, as this would disrupt Rubisco-EPYC1 bonds in the matrix (Fig. 3s). To allow a membrane to traverse through the matrix, Rubisco-EPYC1 bonds that would otherwise be in the path of the traversing membrane must be replaced with favourable Rubisco-membrane bonds. We propose that SAGA1 and MITH1 together provide such Rubisco-membrane bonds, making it energetically favourable for membranes to traverse the matrix (Fig. 3u).

SAGA1 is necessary for the initiation of matrix-traversing membranes as recognized by CYN7 (Fig. 3m–o) and CAH3 (Extended Data Fig. 8). We propose that SAGA1 mediates an initial contact between thylakoids and matrix (Fig. 3t) by binding to Rubisco13,21 and membranes (Figs. 2e and 3e,i,k). In the absence of MITH1, SAGA1 does not appear to produce a sufficient membrane–matrix adhesive force per unit area to displace Rubisco-EPYC1 bonds, and thylakoid membranes contact the matrix but do not penetrate into it (Fig. 3g,n,p and Extended Data Fig. 8). SAGA1 allows membranes in contact with the matrix to acquire matrix-traversing membrane identity as recognized by CYN7 (Fig. 3m–o). We propose that MITH1 increases the membrane–matrix adhesive force per unit area by binding to SAGA1 (Fig. 3q,r) and membranes (Figs. 2e and 3c,h,k), making it energetically favourable for membranes to traverse the matrix. In the absence of SAGA1, MITH1 cannot contribute to membrane–matrix interactions, as it lacks any detectable Rubisco-binding motif21 and appears to be unable to directly bind Rubisco or EPYC1 (Fig. 2d and Extended Data Fig. 4c). The molecular, structural and biophysical aspects of this proposed mechanism are an exciting area for continued research.

The observation that expression of SAGA1 and MITH1 in Arabidopsis produced sheets (Fig. 2j) rather than tubules suggests that in Chlamydomonas, additional factors, which remain to be discovered, act separately from SAGA1 and MITH1 to promote membrane curving into tubules (Fig. 3v). The search for these factors and their eventual characterization is another exciting area for future work.

While matrix-traversing membrane biogenesis is independent of starch sheath biogenesis as evidenced by the presence of matrix-traversing membranes in starchless mutants24, SAGA1’s starch-independent enrichment at the extremities of the matrix-traversing membranes (Fig. 3e,f) places its CBM20 starch-binding domain in close proximity to the starch sheath, whose biogenesis it could help regulate. Indeed, expressing SAGA1 in Arabidopsis expressing CrRBCS and EPYC1 led to increased occurrence of starch in proximity to the Rubisco-EPYC1 matrix14. Starch-binding protein SAGA2 (Cre09.g394621)21 shares weak sequence similarity with SAGA1 and MITH1 (ref. 11), but it is unlikely to contribute to matrix-traversing membrane biogenesis based on the growth phenotypes of the saga2 mutant11, localization of SAGA2 away from membranes21 and the absence of matrix-traversing membranes in Arabidopsis plants expressing SAGA2 in addition to SAGA1, EPYC1 and CrRBCS14.

Matrix-traversing membrane-deficient mutants have multiple matrix condensates

Our discovery of mutants deficient in matrix-traversing membranes provides an opportunity to examine the impacts that these membranes have on other aspects of pyrenoid biogenesis. We had previously observed that the saga1 mutant formed an average of 10 matrix condensates throughout the chloroplast (Fig. 4a) and had proposed that this was a consequence of starch defects13. Our discovery that SAGA1 mediates matrix-traversing membrane biogenesis raised the alternative hypothesis that the multiple matrix condensates in the saga1 mutant are, instead, a consequence of defects in matrix-traversing membranes.

Fig. 4: Matrix-traversing membrane-deficient mutants have multiple pyrenoids and defective CO2 delivery to the pyrenoid.
figure 4

a, Max-z projections depicting RBCS1-Venus localization in Chlamydomonas wild-type, mith1 and saga1 cells. The chloroplast is visualized by chlorophyll autofluorescence. b, Graph depicting the largest cross-sectional area observed in any z-slice for each condensate in each cell. Each tick on the x axis corresponds to a single cell; each point on the graph plots the area for a single condensate. The point depicting the largest condensate in each cell is enlarged to aid visualization. Data for 10 cells per strain are shown. c, Model: in the absence of normal matrix-traversing membranes, their resident matrix-binding proteins could mislocalize and nucleate Rubisco puncta elsewhere in the chloroplast. d, Localization of matrix-binding membrane protein RBMP1 in wild-type versus saga1 mutant backgrounds. e, Anti-CAH3 immunofluorescence. Arrowheads point to regions of strong CAH3 signal associated with the pyrenoid. f, Quantification of fluorescence signal in anti-CAH3 immunofluorescence samples. Results depict the fraction of a cell’s total fluorescence signal found in the pyrenoid. Kruskal–Wallis followed by Dunn’s multiple comparisons test were used for statistical analyses. n = 20 cells per strain except for saga1 with n = 16 cells. *P < 0.05, ****P < 0.0001. See Methods for exact P values. Mean ± s.d. g, Anti-CAH3 western blot performed on whole cell lysates. Anti-alpha-tubulin was used as a loading control. Representative results of 3 biological repeats are shown. h, Model: partially localized CAH3 provides enough CO2 to sustain partial growth in the mith1 mutant.

To test our original hypothesis that the multiple matrix condensates in the saga1 mutant are a consequence of abnormal starch, we generated and examined a starchless saga1;sta6 double mutant. The saga1;sta6 double mutant still exhibited multiple matrix condensates (Extended Data Fig. 9), demonstrating that the multiple matrix condensates in the saga1 mutant are not caused by starch defects and favouring the alternative hypothesis that the multiple condensates are a consequence of abnormal matrix-traversing membranes in this mutant.

Further consistent with this alternative hypothesis, we also observed multiple matrix condensates in the mith1 mutant (Fig. 4a), which has an apparently normal starch sheath but has clear defects in matrix-traversing membranes. The mith1 multiple-condensates phenotype was less pronounced than that of saga1, with a single major pyrenoid present and fewer aberrant condensates (Fig. 4b). Considering that mith1 mutants also have less-pronounced defects in matrix-traversing membrane biogenesis, as they initiate matrix-traversing membranes while saga1 mutants do not (Fig. 3m–o), our observations indicate that the extent of the multiple-matrix-condensates phenotype is correlated with the extent of matrix-traversing membrane biogenesis defects. Taken together, these data support the hypothesis that matrix-traversing membrane defects lead to multiple matrix condensates.

One potential explanation for how the extra condensates are formed in the saga1 and mith1 mutants is that these condensates are nucleated by mislocalized tubule-resident proteins such as RBMP1/BST4, which is neither necessary nor sufficient for producing matrix-traversing membranes25 but has the ability to bind Rubisco21 (Fig. 4c). Consistent with this hypothesis, in the saga1 mutant, RBMP1 mislocalized to multiple streaks at the periphery of chlorophyll voids throughout the chloroplast, which probably correspond to the pyrenoid matrix condensates (Fig. 4d and Extended Data Fig. 9c). Our results suggest a previously unappreciated role for pyrenoid membranes in establishing a single pyrenoid condensate.

Matrix-traversing membranes enable growth under limiting CO2

Matrix-traversing membranes have long been proposed to deliver concentrated CO2 to the pyrenoid26,27, but their functional importance in CO2 delivery had not been experimentally established due to the lack of mutants known to specifically affect them. Our discovery that mith1 and saga1 mutants are deficient in matrix-traversing membrane biogenesis now allows us to examine the consequences of matrix-traversing membrane defects on pyrenoid function. These mutants exhibit growth defects in CO2-limiting conditions that are rescued by elevated CO2 (Fig. 3l)11,13, indicating that Rubisco in their pyrenoids is not supplied with sufficient CO2 to support growth in CO2-limiting conditions. Together, these observations provide experimental support for the proposed role of matrix-traversing membranes in delivering concentrated CO2 to the pyrenoid.

A potential explanation for why mith1’s growth defect is less severe than saga1’s at air levels of CO2 is that the partially formed pyrenoid membranes found at the surface of mith1 mutant pyrenoids (Fig. 3n,p) may supply enough CO2 to sustain growth in air. In wild-type cells, the concentrated CO2 released by the matrix-traversing membranes is thought to be produced from bicarbonate (HCO3) by the carbonic anhydrase CAH3 in the lumen of matrix-traversing membranes26,27. Therefore, if the partially formed pyrenoid membranes seen in mith1 mutant pyrenoids have any capacity to supply CO2, we would expect CAH3 to be found there. Indeed, we observed that CAH3 localized to pyrenoid-peripheral puncta in the mith1 mutants (Fig. 4e and Extended Data Fig. 8f), which probably correspond to the membrane protrusions seen by TEM (Fig. 3p). By contrast, in the saga1 mutant, we could not detect CAH3 above background levels anywhere in the cell, indicating that saga1 mutant pyrenoids lack a source of CO2 (Fig. 4e,f).

We found that CAH3 protein levels are decreased in both mutants, but to a more-severe extent in saga1 (Fig. 4g). Furthermore, a recent study28 found that expression levels of other genes involved in carbon uptake, including HLA3 and LCIA, are decreased in the saga1 mutant. Taken together, these observations indicate that matrix-traversing membranes are required for accumulation of proteins involved in CO2 and HCO3 transport. We propose that this regulatory requirement avoids the unproductive and energetically wasteful operation of the CO2-concentrating mechanism when matrix-traversing membranes are not properly assembled and thus there is no effective route for delivery of CO2 to Rubisco. Low accumulation of CAH3 and other proteins needed for CO2 and HCO3 transport could contribute to the growth defects seen in the mith1 and saga1 mutants (Figs. 3l and 4h). Our observations underscore the importance of matrix-traversing membranes for CO2 delivery and reveal the existence of a regulatory mechanism that ensures that concentrated CO2 is only released by properly assembled matrix-traversing membranes.

Perspective

Our findings elucidate the biogenesis and functions of matrix-traversing membranes in the pyrenoid, an organelle that enhances aquatic photosynthetic CO2 assimilation. We have identified presumably the first biogenesis factors for matrix-traversing membranes, characterized how they function and demonstrated the necessity of matrix-traversing membranes for CO2 delivery. While the protein components may differ in other phylogenetic groups due to the recent convergent evolution of pyrenoids2,29, our work lays a blueprint for characterization of matrix-traversing membrane biogenesis across the phylogenetic tree and reveals basic principles that we propose apply broadly.

Matrix-traversing membranes have been hypothesized to localize the pyrenoid to the base of the chloroplast cup, as the matrix-traversing membrane tubules still form in the canonical location even in the absence of matrix or starch7,30. Our observation that saga1 mutants fail to localize any of the pyrenoid’s three subcompartments to the base of the chloroplast further supports a role for matrix-traversing membranes in pyrenoid localization. The presence of a major pyrenoid at the canonical location in mith1 mutants, which initiate pyrenoid membranes but fail to extend them into the matrix, indicates that the central portion of the matrix-traversing membranes is not needed for pyrenoid localization and suggests that the outer regions of the matrix-traversing membranes are critical for pyrenoid positioning.

Engineering a pyrenoid into crops has the potential to increase yields; however, efforts so far had been stalled by the inability to engineer matrix-traversing membranes31,32,33,34. Our discovery of matrix-traversing membrane biogenesis factors allowed us to overcome this major roadblock to pyrenoid engineering. By producing matrix-traversing thylakoid membranes in Arabidopsis, we achieved the reconstitution of all key structural features that we expect will be needed for a functional pyrenoid in a plant31. To achieve enhancement of CO2 fixation by synthetic pyrenoids in plants, over the coming decade, bicarbonate transporters and carbonic anhydrases will need to be engineered to deliver CO2 via the matrix-traversing membranes31. The plants we produced here are a powerful platform for testing additional pyrenoid components and advancing efforts to install CO2 delivery pathways.

In conclusion, the identification and molecular characterization of the factors that produce pyrenoid matrix-traversing membranes in Chlamydomonas not only advances the basic understanding of pyrenoid biogenesis, but more broadly, establishes a model for investigating the biophysics of how membranes are induced to traverse phase-separated condensates, enables the engineering of matrix-traversing membranes toward enhanced crop yields and brings us closer to a molecular understanding of the biological mechanisms that underlie the global carbon cycle.

Methods

Chlamydomonas strains and culture conditions

Strains used in this study are listed in Supplementary Table 1. Cells were maintained and cultured for experiments as described previously21. Briefly, unless otherwise noted, cells used in imaging and immunoblot experiments were grown to mid-log phase in Tris-phosphate (TP) liquid media in an orbital shaking incubator at room temperature (22 °C), 130 r.p.m., continuous cool white LED light at 175 µmol photons m−2 s−1 and in air enriched to 3% CO2 to enable growth of mutants with defective CO2 concentrating mechanisms. Approximately 16–24 h before experiments, cultures were moved to a different incubator with the same parameters except at air levels of CO2 to induce the CO2-concentrating mechanism. Cultures were collected for experiments at a cell density of ~2 × 106 cells per ml, as measured by a Countess II Automated Cell Counter. The mith1 mutant was PCR verified using an exon primer pair (TAGCCTGGACCGTGCTTAGT; ACCTTCATGTCCTTTGTGGC) that produces a 1 kb band from wild-type template DNA, and an insert primer pair (ACCTTCATGTCCTTTGTGGC; CTGCTGAGGCTGAATCTACTC) that produces a band from mutant template DNA.

Crossing of Chlamydomonas strains

Several strains used in this study were produced through crossing opposite mating-type strains35. Cultures that were 5–6 days old and maintained on Tris-acetate-phosphate (TAP)-agar plates were transferred to N10 TAP plates, which are nitrogen-depleted, containing 10% of the nitrogen normally supplied on TAP plates. After 3 days on N10 plates, cells were transferred to a 50 ml Erlenmeyer flask containing 2.5 ml sterile distilled water using a sterile loop and agitated for 2 h on an orbital shaker in air at 175 µmol photons m−2 s−1 of light. Equal volumes of the mating type ‘+’ and mating type ‘–’ strains were mixed and left at 175 µmol photons m−2 s−1 of light without agitation for 0.5–1 h. Of the mixed cells, 300 µl were then plated on TAP-3% agar plates and air-dried in a sterile hood. The plates were then wrapped in foil and left in the dark for 5–7 days. After this time, vegetative cells were scraped from the plate with a flame-sterilized dull scalpel, leaving zygospores behind. Zygospores were allowed to mature into colonies and then transferred and streaked to single colonies on a new plate containing antibiotics to select for the correct genotype. For crosses to RBCS1-Venus or the starchless sta6 mutant, further screening was used to search for fluorescence (using Typhoon scanner) or the absence of starch (tested with Lugol’s iodine treatment). The genotype was verified by PCR.

Arabidopsis plant material and growth conditions

Arabidopsis (Arabidopsis thaliana, Col-0 background) strains used in this study are shown in Supplementary Table 2. Seeds were sown on compost or on half-strength Murashige and Skoog basal salt medium (Sigma) with 1% sucrose, stratified for 3 days at 4 °C and grown at 20 °C, ambient CO2 and 70% relative humidity under 200 µmol photons m−2 s−1 of light supplied by cool white LED lights (Percival SE-41AR3cLED, CLF PlantClimatics) in 12 h light/dark conditions or in a temperature-controlled greenhouse at 21 °C.

Construct design and transformation

Plasmids used for Chlamydomonas are listed in Supplementary Table 3. Open reading frames were cloned into either the pLM005 (ref. 3) (Genbank: KX077945.1) or pRAM118 (ref. 36) (Genbank: MK357711) plasmid backbones. The two plasmid backbones are identical except that pLM005 contains an AphIII cassette encoding paromomycin resistance, which is replaced by the AphII cassette encoding hygromycin resistance in pRAM118. Both plasmids encode the fluorescent protein Venus followed by three copies of the FLAG epitope.

The MITH1-Venus-3×FLAG construct was made following a previously published PCR amplification protocol20. A KOD-Xtreme polymerase (Takara Bio) and manufacturer instructions were used to amplify MITH1 from Chlamydomonas genomic DNA. MITH1 was annealed into an HpaI-digested pRAM118 vector using the InFusion kit (Takara Bio). PCR primers (ACAACAAGCCCAGTTATGTTGCGCTCTGCAAAGCC; ACCCAGATCTCCGTTGGGCGTCTTGAGAACGCTGTTGC) containing homology tails to pRAM118 were designed according to a previous protocol20. The plasmid’s sequence was verified by nanopore sequencing (CD Genomics) and was found to contain 6 point mutations in intron sequences in comparison to the reference Chlamydomonas v.5.6 genome sequence; this was deemed acceptable for the purposes of mutant gene rescue experiments.

For transformation, Chlamydomonas cells were grown to mid-log phase in liquid TAP media in air with the standard settings described above. Cells were then centrifuged at 1,000 g for 5 min and washed twice with 10 ml of MAX reagent (GeneArt MAX Efficiency Transformation Reagent for Algae, Invitrogen). Cells were resuspended to a concentration of 2 × 108 cells per ml in MAX reagent. For each transformation, 60 µl of cells were combined with 5 µl of DNA that had been linearized in a standard restriction digest using the enzyme indicated in Supplementary Table 3 and their manufacturer protocol (New England Biolabs). This mixture was incubated at 4 °C for 5 min and then transferred to an electrocuvette (2-mm gap, Bulldog bio) that had been pre-cooled at 4 °C. Electroporation was done using a NEPAGENE NEPA21 electroporator (parameters: poring pulse, 250 volts, 8 ms pulse length, 50 ms pulse interval, 2 pulses, 40% decay rate, + polarity; transfer pulse, 20 volts, 50 ms pulse length, 50 ms pulse interval, 5 pulses, 40% decay rate, +/− polarity)37. Immediately following electroporation, the mixture was added to 8 ml TAP + 40 mM sucrose and incubated with agitation in the dark overnight. Cells were then pelleted at 600 g for 5 min and plated on TAP-agar solid plates with the appropriate antibiotics (that is, hygromycin or paromomycin). After 5 days of growth in dim light, plates were transferred to 100 µE light for 7–14 days until colonies appeared and were of a sufficient size for picking.

Constructs used to transform Arabidopsis are shown in Supplementary Table 4. The coding sequences of SAGA1, MITH1 and EPYC1 were codon optimized for expression in land plants and cloned into binary level 2 acceptor vectors using the Plant MoClo system38. The 35S cauliflower mosaic virus (CaMV) promoter, cassava vein mosaic virus (CsVMV) promoter and Arabidopsis RbcS3B promoter were used to drive expression. Preliminary experiments indicated that the Chlamydomonas MITH1 protein localizes to the cytosol of Nicotiana benthamiana; therefore, a chloroplast transit peptide (cTP) was added to MITH1 for expression in Arabidopsis. Level 2 vectors were transformed into Agrobacterium tumefaciens (AGL1) for stable insertion in Arabidopsis plants by floral dipping39.

To obtain the CrRBCS2;EPYC1;SAGA1;MITH1 lines, MITH1-mCerulean driven by either the 35S or RbcS3B promoter was expressed in the EP1S1-1 line14, consisting of SAGA1-mCherry and EPYC1-GFP expressed in the S2Cr background40, in which the 1a3b Rubisco small subunit mutant is complemented with Chlamydomonas Rubisco small subunit CrRBCS2. Transformants were selected on the basis of their kanamycin resistance. To obtain the CrRBCS2;EPYC1;MITH1 line, EPYC1-GFP was first expressed in the S2Cr background; homozygous lines were identified in the T3 generation using a kanamycin resistance cassette. Then, MITH1-mCerulean driven by the RbcS3B promoter was transformed into the resulting line and transformants were selected using the pFAST-R selection cassette41.

Transmission electron microscopy

Chlamydomonas samples for electron microscopy were prepared at room temperature and nutated in 1 ml volumes during chemical treatments and washes unless otherwise noted. After initial centrifugation for collection, all pelleting was done at 3,000 g for 1 min. Approximately 5 × 106 cells were collected at 1,000 g for 5 min and fixed in 2.5% glutaraldehyde in TP medium for 1 h. After three 5-min washes in Milli-Q water, samples were treated with a freshly prepared solution of 1% OsO4, 1.5% K3Fe(CN)3 and 2 mM CaCl2. After four 5-min washes, samples were then serially dehydrated (5-min incubations in 50, 75, 95 and 100% ethanol, followed by two 10-min incubations in 100% acetonitrile). The samples were then suspended in 50% acetonitrile, 17.5% Quetol 651, 22.5% nonenyl succinic anhydride and 10% methyl-5-norbornene-2,3-dicarboxylic anhydride and left uncapped and stationary overnight in a fume hood to allow for evaporation of the acetonitrile. The samples were then embedded in epoxy resin containing 34% Quetol 651, 44% nonenyl succinic anhydride, 20% methyl-5-norbornene-2,3-dicarboxylic anhydride and 2% catalyst dimethylbenzylamine. The resin mixture was refreshed daily for 4 days. After the last resin refresh, pellets were resuspended in 300 µl of the resin mixture and centrifuged at 10,500 r.p.m. at 30 °C for 20 min in a swinging bucket rotor for microfuge tubes. They were then cured at 65 °C for 48 h. Afterwards, ultramicrotomy was done using a DiaTome diamond knife on a Leica UCT Ultramicrotome at the Imaging and Analysis Center, Princeton University, and imaging was done on a CM100 transmission electron microscope (Philips) at 80 kV or CM200 at 200 kV.

For Arabidopsis TEM, leaf samples were taken from 21-day-old plants and fixed with 4% (v/v) paraformaldehyde, 2% (v/v) glutaraldehyde and 0.1 M sodium cacodylate (pH 7.2). Leaf strips (1 mm wide) were vacuum infiltrated with fixative three times for 15 min, then rotated overnight at 4 °C. Samples were rinsed three times with PBS (pH 7.4), then dehydrated sequentially by vacuum infiltration with 50, 70, 80 and 90% ethanol (v/v) for 1 h each, then three times with 100% ethanol. Samples were infiltrated with increasing concentrations of LR White resin (30, 50 and 70% [w/v]) in ethanol overnight for each, then 100% resin three times for 1 h. The resin was polymerized in capsules at 50 °C overnight. Sections (1 μm thick) were cut on a Leica Ultracut ultramicrotome, stained with toluidine blue and viewed in a light microscope to select suitable areas for investigation. Ultrathin sections (60 nm thick) were cut from selected areas and mounted onto plastic-coated copper grids. Grids were stained in 2% (w/v) uranyl acetate, then viewed in a JEOL JEM-1400 Plus TEM (JEOL). Images were collected on a GATAN OneView camera (GATAN).

Serial block-face scanning electron microscopy (SBF–SEM)

Leaf samples were harvested after 6 h light from 28-day-old plants grown in a 12-h day/night regime. Samples were chemically fixed, stained and embedded into Durcupan epoxy resin. All solutions were prepared as previously described42. Fixation and embedding was done in a Pelco BioWave Pro1 (Ted Pella) using an adaptation of a previously described protocol43 (see Supplementary Method ‘Epoxy resin embedding of arabidopsis leaf sections for light microscopy’). The resultant blocks were trimmed, glued on stubs, coated with a mixture of gold and palladium, and sectioned and scanned ~500 times using a scanning electron microscope (FEI Apreo VolumeScope, Thermo Fisher), operated at 1.18 kV. Each section removed 50 nm from the imaged block face. After image alignment, the structural features were marked out manually with Amira software (Thermo Fisher).

Confocal microscopy

For several Chlamydomonas experiments including single-colour immunofluorescence imaging of CAH3 shown in Fig. 4e and Extended Data Fig. 8f, live imaging of CYN7-Venus shown in Fig. 3m–o and Extended Data Fig. 8e, live imaging of RBMP1-Venus shown in Fig. 4d, and all confocal imaging across Extended Data Figs. 7 and 8e–f, cells were imaged with a Leica TCS SP5 laser scanning confocal microscope using Leica LAS X software (Leica Microsystems) and a ×100 magnification/1.46 NA oil objective. Images were processed with FIJI software44. For the anti-CAH3 immunofluorescence experiment, excitation/emission wavelengths were 488 nm/515–535 nm. For live imaging, excitation/emission wavelengths were 514 nm/521–540 nm for Venus and 514 nm/645–725 nm for chlorophyll autofluorescence.

The RBCS1-Venus expressing strains shown in Fig. 4a were imaged using a Nikon A1R confocal laser scanning microscope (Nikon Instruments) with a ×100 magnification/1.45 NA oil immersion objective. After acquisition, Nikon NIS Elements software was used to optimize image quality using the Denoise.ai and Deconvolution tools. Excitation/emission wavelengths were 514 nm/522–555 nm for Venus and 514 nm/601–676 nm for chlorophyll autofluorescence.

All Arabidosis imaging as well as all Chlamydomonas imaging not specified above was performed with a Nikon CSU-W1 SoRa spinning disk system (Nikon Instruments) with a ×60 magnification/1.4 NA oil immersion objective. All acquisitions were magnified by an additional ×4 through SoRa magnification for a final magnification of ×240. After acquisition, Nikon NIS Elements software was used to optimize image quality using the Denoise.ai and Deconvolution tools. Excitation/emission wavelengths used for Arabidopsis imaging were 405 nm/430–480 nm for mCerulean, 488 nm/507–543 nm for GFP, 561 nm/579–631 nm for mCherry and 640 nm/669–741 nm for chlorophyll autofluorescence. Excitation/emission wavelengths used for Chlamydomonas live imaging were 514 nm/530–560 nm for Venus and 514 nm/669–741 nm for chlorophyll autofluorescence. Excitation/emission wavelengths used for two-colour immunofluorescence imaging were 488 nm/507–543 nm for Alexa Fluor 488 and 640 nm/669–741 nm for STAR RED.

Immunofluorescence

Samples for immunofluorescence were prepared according to a published protocol45.

Briefly, cells were fixed through 20-min treatments with 4% (w/v) formaldehyde in PBS, followed by 100% methanol at −20 °C. Cells were then washed in PBS and blocked for 1 h in PBS containing 5% (w/v) bovine serum albumin, treated with primary antibody for 1 h and secondary antibody for another hour.

To visualize CAH3 localization in various strains, a rabbit polyclonal anti-CAH3 primary antibody (Agrisera) and Alexa Fluor 488-conjugated goat anti-rabbit secondary antibody (Thermo Fisher) were used.

For co-localization of CAH3 and MITH1, the mith;MITH1-Venus-3×FLAG strain was used. CAH3 was labelled with the rabbit polyclonal anti-CAH3 primary antibody and goat anti-rabbit Alexa Fluor 488-conjugated secondary antibody, and MITH1 was labelled using a mouse monoclonal anti-FLAG primary antibody (Sigma-Aldrich) and STAR RED-conjugated goat anti-mouse secondary antibody (Abberior).

For co-localization of CAH3 and SAGA1, the saga1;SAGA1-Venus-3×FLAG strain was used. CAH3 was labelled with the rabbit polyclonal anti-CAH3 primary antibody and goat anti-rabbit Alexa Fluor 488-conjugated secondary antibody, and SAGA1 was labelled using the mouse monoclonal anti-FLAG primary antibody and STAR RED-conjugated goat anti-mouse secondary antibody.

For co-localization of MITH1 and SAGA1, the saga1;SAGA1-Venus-3×FLAG strain was used. MITH1 was labelled with a rabbit anti-MITH1 primary antibody (Yenzyme) and goat anti-rabbit Alexa Fluor 488-conjugated secondary antibody, and SAGA1 was labelled using the mouse monoclonal anti-FLAG primary antibody and STAR RED-conjugated goat anti-mouse secondary antibody.

Antibody dilutions were 1:500 in PBST for all primary and secondary antibodies except for anti-CAH3, which was diluted 1:200 following supplier recommendation.

Arabidopsis image processing and bleedthrough attenuation

Despite our best efforts to control bleedthrough by careful choice of filters and excitation wavelengths, in our microscopy setup for imaging Arabidopsis, bleedthrough from the EPYC1-GFP channel into the MITH1-mCerulean channel caused a non-negligible amount of false mCerulean signal (Extended Data Fig. 3a), which was particularly problematic considering the high expression levels of EPYC1-GFP and relatively low expression levels of MITH1-mCerulean. We opted to computationally attenuate the EPYC1-GFP bleedthrough signal from images presented in Fig. 2a–e and Extended Data Fig. 4c to provide the reader with a more-accurate representation of the localization of MITH1.

The procedure for bleedthrough attenuation is based on subtracting a portion of the signal measured in the GFP channel from the signal measured in the mCerulean channel. How much of the GFP signal to subtract in this procedure was set by a factor that we refer to as the ‘bleedthrough coefficient’. To determine the bleedthrough coefficient, we used images of condensates in the CrRBCS2;EPYC1-GFP line (lacking MITH1-mCerulean and therefore lacking a true mCerulean signal). To allow the bleedthrough coefficient to be approximated via a linear regression that passes through the origin (that is, when GFP signal is 0, bleedthrough in the mCerulean channel is 0), we first subtracted out a constant value across the entire image in each channel to account for the average ‘background’ signal due to factors other than GFP signal in that channel. This background signal includes noise inherent to the camera and sample autofluorescence. For instance, fluorescence not attributable to a fluorescent protein can be seen to varying degrees in the mCerulean channel of the CrRBCS2;EPYC1 line, apparently due to autofluorescence from the leaf sample itself (Extended Data Fig. 3a). By analysing images of the CrRBCS2;EPYC1 line, we empirically estimated a background signal of 200 intensity units in the mCerulean channel and 1,500 units in the GFP channel. These background values were subtracted from the corresponding channels for all images in a test set of CrRBCS2;EPYC1 images. A circular region of pixels was then sampled from each condensate, and the mCerulean intensity vs the GFP intensity was plotted for each such region as a data point in the regression. The slope of the regression (measured as 0.0134) was interpreted as the bleedthrough coefficient (Extended Data Fig. 3f).

In Extended Data Fig. 3b–e, we illustrate the bleedthrough attenuation procedure for a set of CrRBCS2;EPYC1-GFP;MITH1-mCerulean images. We first subtracted the constant background signal from the GFP and mCerulean channel images as described for the determination of the bleedthrough coefficient above (Extended Data Fig. 3c). We then created a copy of the GFP channel image and multiplied the signal intensity by the bleedthrough coefficient to match the expected intensity of the mCerulean bleedthrough. This rescaled GFP image was used as a mask on the mCerulean channel, and the intensity of this mask was subtracted from the mCerulean channel image on a pixel-by-pixel basis (Extended Data Fig. 3g) using the ND Images Arithmetics tool available on the Nikon NIS Elements software. This generated a new mCerulean channel image in which the signal corresponding to MITH1-mCerulean is conserved while the signal corresponding to bleedthrough is attenuated (Extended Data Fig. 3d). The chlorophyll and mCherry channels were denoised and deconvolved using NIS Elements software as described earlier. Since the EPYC1-GFP and MITH1-mCerulean channels were processed as described above, they were not deconvolved so as to avoid the possible introduction of artefacts (Extended Data Fig. 3e).

Scaling of the MITH1-mCerulean and SAGA1-mCherry channels is universal across Fig. 2a–d and across Extended Data Fig. 4c. This allowed for visual intensity to serve as a more direct indicator of protein levels to allow for comparison between strains. Scaling of the EPYC1-GFP and chlorophyll channels across Fig. 2a–d and Extended Data Fig. 4c was performed on an image-by-image basis such that maximum pixel intensity in the image determined the saturation point. This allowed consistency in the visual intensity of EPYC1 and chlorophyll regardless of variation between samples.

Analysis of RBCS1-Venus puncta

To determine the number of condensates in each cell, Nikon NIS Elements software was used to threshold images on the basis of Venus signal intensity, disregarding parts of the image with no or very low intensity while representing continuous clusters of high-intensity pixels as binary objects, each representing a distinct RBCS1 condensate. Object identity was conserved over multiple z-slices, meaning that each was correctly identified as a single condensate even if it appeared across numerous slices.

Nikon NIS Elements software was used to measure the size of each condensate. For each condensate, which was detected and turned into a binary object as described above, the cross-sectional area was measured at each z-slice it appeared in, and the largest one was reported as the representative area of the condensate, as condensates are approximately spherical. This provided a reliable way of comparing condensate size in lieu of measuring volume, which is complicated by limits to z-resolution. To calculate the proportion of RBCS1 signal contained in the largest condensate of a given cell, a ratio was taken between the largest condensate area in the cell and the sum of all condensate areas in the cell.

Data tabulation, statistical analysis and data visualization were each performed using the GraphPad Prism 9 software. A two-tailed Mann–Whitney U test was used to test for significance of differences between strains when comparing the proportion of Rubisco fluorescence in the largest condensate of each cell.

Cell fractionation

Wild-type Chlamydomonas was grown in TAP medium at ambient CO2 until they reached a density of 2 × 106 cells per ml. Cells were centrifuged for 5 min at 1,000 × g and the pellet was weighed and resuspended in 2× volume of lysis buffer (50 mM HEPES, 10 mM KOAc, 2 mM Mg(OAC)2, 1 mM CaCl2 pH 7.0 + protease inhibitors). Cells were sonicated in a probe tip sonicator (Qsonica) on ice for 5 min, a 3-s pulse and 6-s rest at 60% amplitude. The lysate was centrifuged at 2,000 g for 10 min to remove intact cells, and 50 μl of supernatant was collected as the whole cell lysate (input) sample. The rest of the supernatant was divided in two; to half, n-dodecyl-ß-d-maltoside (DDM) was added to a final concentration of 1% and the mixture allowed to solubilize while nutating for 20 min at 4 °C. Samples were centrifuged at 18,000 × g for 30 min and pellet and supernatant were collected and analysed via western blotting.

Western blot

For Chlamydomonas experiments, the PVDF membrane (Millipore) was blocked for 1 h in 5% milk in TBST (Tris buffered saline with 0.1% Tween-20), then incubated with primary antibody for 1 h. After three 10 min washes in TBST, the blot was treated with secondary antibody for 1 h and washed again 3× for 10 min. The secondary antibody was diluted 1:10,000. Anti-SAGA1 (ref. 11)(Yenzyme), anti-MITH1 (Yenzyme) and anti-ferredoxin (Agrisera) were diluted 1:1,000. Anti-RBMP1 (gift from L. Mackinder) and anti-CAH3 (rabbit polyclonal, Agrisera) were diluted 1:2,000. Anti-PsbC (Agrisera) was diluted 1:3,000 and anti-tubulin (mouse monoclonal, Sigma-Aldrich) was diluted 1:4,000. Blots were labelled using an enhanced chemiluminescence system (Advansta WesternBright ECL kit) and imaged on an iBright Imaging System (Thermo Fisher).

Protein analyses for Arabidopsis were done as follows. Soluble protein was extracted from frozen leaf material of 21-day-old plants in protein extraction buffer (20 mM Tris-HCl pH 7.5 with 5 mM MgCl2, 300 mM NaCl, 5 mM dithiothreitol, 1% Triton X-100 and cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche)). Samples were heated at 80 °C for 15 min with 1× Bolt LDS sample buffer (Thermo Fisher) and 200 mM dithiothreitol. Extracts were centrifuged and the supernatants subjected to SDS–PAGE on a 12% (w/v) polyacrylamide gel and transferred to a nitrocellulose membrane. Membranes were probed with rabbit serum raised against SAGA1 (ref. 13) (1:1,000) or EPYC1 (ref. 3) (1:2,000), or with mouse anti-actin (Agrisera, AS21 4615) or mouse anti-GFP (Santa Cruz, sc-9996), followed by HRP-linked goat anti-rabbit IgG (Abcam) at 1:10,000 dilution, or rabbit anti-mouse IgG (Agrisera) at 1:10,000 dilution. Membranes were visualized using Pierce ECL Western Blotting Substrate (Life Technologies).

Protein immunoprecipitation and mass spectrometry

Immunoprecipitation and mass spectrometry of Venus-3×FLAG-MITH1, Venus-3×FLAG-SAGA1 or Venus-3×FLAG were performed as described previously23 with the following differences: a 40 cm chromatography column was used, the column temperature was set to 45 °C and a 2-h gradient method with 300 nl min−1 flow was used.

Cells were grown in 1 l bottles of TP medium with air bubbling and constant stirring at 210 r.p.m. under 150 μmol photons m−2 s−1 of light until the cell density reached 2–4 × 106 cells per ml. Cells were collected by centrifugation at 3,000 g for 4 min in an Avanti J-26X centrifuge with an 8.1000 rotor (Beckman) at 4 °C. The pellets were washed in 35 ml ice-cold washing buffer (25 mM HEPES, 25 mM KOAc, 1 mM Mg(OAc)2, 0.5 mM CaCl2, 100 mM sorbitol, 1 mM NaF, 0.3 mM Na3VO4 and cOmplete EDTA-free protease inhibitor (1 tablet per 500 ml)) and then resuspended in a 1:1 (v/w) ratio of ice-cold 2×IP buffer (50 mM HEPES, 50 mM KOAc, 2 mM Mg(OAc)2, 1 mM Cacl2, 200 mM sorbitol, 1 mM NaF, 0.3 mM Na3VO4 and cOmplete EDTA-free protease inhibitor (1 tablet per 50 ml)). Cell slurry (3 ml) was immediately added to liquid nitrogen to form small popcorn pellets which were stored at −80 °C until needed. Cells were lysed by cryogenic grinding using a Cryomill (Retsch) at a frequency of 25 oscillations per second for 20 min. The ground powder was defrosted on ice for 45 min and dounced 25 times on ice with a Kontes Duall #22 homogenizer (Kimble). Homogenized cells (1 ml) of each sample were used for the following processes. Membrane proteins were solubilized by incrementally adding an equal volume of ice-cold 1×IP buffer plus 2% digitonin (RPI), followed by an incubation of 45 min with nutation at 4 °C. The cell debris were removed by spinning at 12,700 g for 30 min at 4 °C. The supernatant was then mixed with 50 μl anti 3×FLAG magnetic beads (Sigma) which had been previously washed sequentially with 1×IP buffer 3 times and 1×IP buffer plus 0.1% digitonin 2 times. The mixture was incubated with nutation at 4 °C for 1.5 h, followed by the removal of supernatant. The beads were washed 4 times with 1×IP buffer plus 0.1% digitonin, followed by a 30-min competitive elution with 45 μl of 1×IP buffer plus 0.25% digitonin and 2 μg μl−1 3×FLAG peptide (Sigma-Aldrich). After elution, 30 μl protein samples were mixed with 9.75 μl 4×SDS–PAGE buffer (BioRad) containing 100 mM dithiothreitol (Sigma-Aldrich), followed by denaturation by heating at 70 °C for 10 min. Then, 30 μl of denatured protein sample was loaded into a well of a 4–15% Criterion TGX Precast Midi Protein Gel (BioRad) for electrophoresis at 50 V for 40 min until the protein front moved 2.5 cm. Gel slices (2.0 cm) containing target proteins with molecular weight ≥10 kDa (to exclude the 3×FLAG peptide) were excised and stored at 4 °C until processing for in-gel digestion.

In-gel digestion of protein bands using trypsin was performed and the samples were dried completely in a SpeedVac and resuspended with 20 μl of 0.1% formic acid pH 3 in water. A volume of 2 μl (360 ng) was injected per run using an Easy-nLC 1,200 UPLC system. Samples were loaded directly onto a 40-cm-long, 75-μm-inner-diameter nanocapillary column packed with 1.9 μm C18-AQ resin (Dr. Maisch) mated to a metal emitter in line with an Orbitrap Fusion Lumos (Thermo Scientific). The column temperature was set at 45 °C and a 2-h gradient method with 300 nl min−1 flow was used. The mass spectrometer was operated in data-dependent mode with a 120,000 resolution MS1 scan (positive mode, profile data type, AGC gain of 4 × 105, maximum injection time of 54 s and mass range of 375–1,500 m/z) in the Orbitrap, followed by HCD fragmentation in ion trap with 35% collision energy. A dynamic exclusion list was invoked to exclude previously sequenced peptides for 60 s, and a maximum cycle time of 3 s was used. Peptides were isolated for fragmentation using the quadrupole (1.2 m/z isolation window). The ion trap was operated in rapid mode.

Spot tests

Cells were grown to mid-log phase in liquid TAP medium. They were then pelleted at 1,000 g for 5 min and resuspended to a concentration of 6 × 105 cells per ml in TP medium. A volume of 10 µl of each sample was spotted onto TP-agar plates. Once the spots were dry, plates were then placed under 100 µmol photons m−2 s−1 of light in variable CO2 levels, which included 40 parts per million (ppm) CO2, ambient CO2 (~415 ppm) or 3% CO2 for 7 days before imaging on a Phenobooth imager. For the 40 ppm condition, plates were first kept in ambient air for the first 24 h before being placed in a 40 ppm chamber for the remaining 6 days.

Statistics and reproducibility

Unless otherwise stated here, each microscopy image is representative of three or more biological repeats. One biological repeat was performed for the SBF–SEM experiment depicted in Fig. 2j and in Supplementary Videos 15. The results of this experiment are consistent with those produced using other imaging methods. Representative images from two biological repeats are shown for the TEM images of the CrRBCS2;EPYC1;MITH1 Arabidopsis line presented in Fig. 2i and Extended Data Fig. 4a, the co-localization of SAGA1-Venus and CAH3 shown in Fig. 3i, the CrRBCS2;EPYC1-GFP images shown in Extended Data Fig. 3a, the mith1;saga1 double mutant TEM images shown in Extended Data Fig. 6b and the wild-type cell shown in Extended Data Fig. 8a.

Four biological repeats were performed for the CAH3 immunofluorescence experiment presented in Fig. 4e, although two replicates were imaged on a Leica TCS SP5 laser scanning confocal microscope (Leica Microsystems) and the other two on a Nikon CSU-W1 SoRa spinning disk system (Nikon Instruments). For consistency, data from the two replicates imaged with the Nikon CSU-W1 SoRa spinning disk system (Nikon Instruments) microscope were used to generate the graph in Fig. 4f. Exact P values for the graph in Fig. 4f are as follows: WT vs mith1, P = 0.0195; WT vs saga1, P = 1.15 × 10−6; WT vs cah3, P = 3.82 × 10−8.

Reporting summary

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