Native architecture of the Chlamydomonas chloroplast revealed by in situ cryo-electron tomography

1The chloroplast is the site of eukaryotic photosynthesis and carbon fixation, coupled processes that convert the energy of light into stored biochemical energy, while exchanging environmental CO2 for O2. As such, the chloroplast both directly and indirectly sustains much of the life on our planet.

2The biochemical pathways of photosynthesis rely on the chloroplast's elaborate architecture. In both higher plants and algae, the light-dependent photosynthetic reactions take place in the thylakoids, sheet-like membrane-bound compartments that group together into regularly-spaced stacks (called grana in higher plants). Photons are absorbed by photosystem complexes within the thylakoid membrane, and the downstream electron transfer chain generates NADPH in the chloroplast stroma. This process, combined with the splitting of H20 into O2 and H+, concentrates H+ inside the thylakoid lumen. The resulting proton gradient across the thylakoid membrane is used by the ATP synthase complex to generate ATP in the chloroplast stroma, where the light-independent carbon fixation reactions occur. ATP and NADPH power the Calvin–Benson cycle, a series of reactions that incorporate CO2 into the three-carbon sugar glyceraldehyde 3-phosphate (G3P). Downstream pathways produce starch, which is stored within the chloroplast. The CO2-fixing step of the Calvin–Benson cycle is catalyzed by the octomeric ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) complex. In higher plants, which have ready access to gaseous CO2, RuBisCO complexes are distributed throughout the chloroplast stroma. However, algae primarily live in aquatic environments where CO2 is dissolved as HCO3−. In order to maintain efficient carbon fixation, the majority of algal RuBisCO complexes are concentrated in a specialized region of the chloroplast known as the pyrenoid, where carbonic anhydrase frees CO2 from solution. Thus, in algae the light-dependent and light-independent reactions of photosynthesis are physically segregated to distinct regions of the chloroplast (Figure 1A). While the chloroplast's biochemical pathways are well characterized, a better understanding of the organelle's native three-dimensional (3D) architecture would help answer questions about the biogenesis of thylakoids and pyrenoids, as well as the cellular mechanisms that link the activities of these two physically separated systems.

3Much of our understanding of chloroplast architecture comes from conventional transmission electron microscopy (TEM) studies using the traditional protocol of sample fixation, heavy metal staining, dehydration, plastic embedding, and sectioning with an ultramicrotome. Each of these preparation steps can distort structures and obscure high-resolution information. Sample preparation by freeze-fracture has enabled observations of integral proteins embedded in membrane surfaces, but it cannot provide a 3D view of chloroplast architecture. Vitrification by rapid freezing offers the best possible preservation of biological material. Imaging vitreous samples by cryo-electron tomography (cryo-ET) generates 3D views of native structures, where image contrast corresponds to the intrinsic variation in mass density instead of the sample's local affinity for heavy metal stains.

4Over the last decade, there has been progress in the 3D visualization of photosynthetic structures. Much of this work has focused on cyanobacteria, as the small size of these prokaryotes allows imaging by whole-cell tomography. With the exception of a few very small cells, more complex eukaryotic specimens must be thinned to make them transparent to the electron beam. Two approaches have been used to examine thin samples of higher plant thylakoids. High pressure freezing followed by freeze-substitution reduces the fixation artifacts caused by conventional TEM sample preparation, enabling tomograms of plastic sections that better preserve thylakoid ultrastructure but are still limited in resolution. Alternatively, thylakoids can be isolated from crushed chloroplasts to make thin vitreous samples suitable for cryo-ET. While this approach has yielded informative high-resolution structures of protein complexes embedded in thylakoid membranes, these membranes have been removed, both spatially and temporally, from their cellular context. Labile and transient chloroplast structures may no longer be biologically relevant in such preparations.

5The ideal approach would be to thin intact vitreous specimens for cryo-ET, preserving both 3D architecture and high-resolution information. Until recently, the only method capable of thinning vitreous material was cryo-sectioning, a procedure that is technically demanding, causes compression and crevasse deformations, and has a narrow range of optimal section thickness. Cryo-focused ion beam (cryo-FIB) milling provides a compression-free method for thinning vitreous material to any desired thickness. Cryo-FIB is also more high-throughput than cryo-sectioning, as specimens can be thinned directly on the grid that will be used for cryo-ET. Cryo-FIB is thus becoming the method of choice and has already been successfully applied to a variety of eukaryotic cell types.

6We combined cryo-FIB with cryo-ET to acquire the first in situ 3D structures of the algal chloroplast (Figure 1). While most structural studies have focused on terrestrial plants, aquatic organisms account for around 50% of global carbon fixation, and it is estimated that nearly half of this aquatic production is performed by eukaryotic algae. Therefore, we chose to examine the genetically tractable unicellular green alga Chlamydomonas reinhardtii, the most extensively studied model organism for algal photosynthesis. Because all biological material was immobilized at the moment of whole cell vitrification, fine structures remained well preserved within their cellular context, enabling accurate measurements of thylakoid architecture, membrane invaginations, pyrenoid tubules, the eyespot, and the 3D organization of RuBisCO complexes within the pyrenoid matrix.

7Chlamydomonas cells are ∼10 μm in diameter, and about half of their volume is occupied by a single cup-shaped chloroplast (Figure 1A). Different regions of the algal chloroplast are associated with specific functions. Carbon fixation is carried out in the pyrenoid at the base of the chloroplast cup. There is evidence for chloroplast protein translation and import near the base of the cup, while the sides of the cup (known as lobes) are thought to contain mature thylakoids that perform the light reactions of photosynthesis. We acquired tomograms from a variety of cellular locations and found that all regions of the chloroplast cup share certain characteristics.

8Throughout the chloroplast, thylakoids were most commonly found in stacks of variable length but uniform spacing. Thylakoid stacks had a lateral repeat distance of 22.4 ± 1.3 nm, with 4.9 ± 0.5 nm thylakoid membranes, 9.0 ± 1.4 nm thylakoid lumina, and 3.6 ± 0.5 nm interthylakoid stromal spaces (± denotes standard deviation, n = 38 thylakoids from eight stacks). Although Chlamydomonas thylakoid lumina were significantly wider than the thylakoid lumina reported in studies of higher plant grana (15.7–16.3 nm lateral repeat, 4.5–4.7 nm thylakoid lumina, and 3.2–3.6 nm interthylakoid stromal spaces), grana thylakoid lumina have been shown to expand to 9.2 nm when plants are adapted to light. In our tomograms, the number of thylakoids per stack ranged from 2 to 15, with a median of 3 thylakoids. In agreement with previous reports from other organisms, thylakoid lumina were less dense than the interthylakoid stromal space and the chloroplast stroma (Figure 2).

9It is thought that all of the thylakoids within the chloroplast form one continuous compartment. While our tomogram volumes were too confined to make a definitive assessment, 3D visualization of thylakoid architecture did indeed reveal complex interconnected networks (Figures 3 and 4). Thylakoid stacks frequently split along one or two of the interthylakoid stromal spaces, and the thylakoids on each side of these divisions continued and merged with other stacks (Figure 3B). In this manner, most thylakoid stacks were physically linked to each other. In addition to occasional thylakoid bifurcations and mergers, thylakoid lumina were most frequently continuous with each other where the tips of two adjacent thylakoids were conjoined by a 180° loop of the thylakoid membrane (most clearly seen in Figures 3A and 4F). More rarely, the tips of non-adjacent thylakoids from the same stack were connected by more topologically complex looping structures (Figure 3C–F and light blue lumen in Figure 10). Like the photosynthetic membranes of cyanobacteria, we observed fenestrations in Chlamydomonas thylakoids that often perforated entire stacks and may allow free diffusion of the chloroplast stroma (Figure 5).

10Thylakoids were normally oriented parallel to the chloroplast envelope, but at specific locations the thylakoid tips from multiple stacks congregated and pointed perpendicularly towards the envelope. At these thylakoid tip convergence zones (Figure 3C–F and 4C–F), the double membrane structure of the chloroplast envelope appeared to lose definition and in some spots was reduced to a single dense plane without an intermembrane space (Figure 3D). These regions also contained chloroplast envelope invaginations (Figure 3C) and small vesicles between the chloroplast envelope and thylakoid tips (Figure 4C). Thylakoid tip convergence zones were always present at the apical rim of the chloroplast cup (Figure 4), but they were also found near the base of the cup (Figure 3).

11In nearly every tomogram, we noticed invaginations that protruded from the chloroplast envelope inner membrane into the stroma (Figures 3, 4, 6 and 7). The density of invaginations on the membrane varied widely, with a median of 12/μm2 (Figure 8B). Our dataset was not extensive enough to make robust statistical conclusions about the relative abundance of invaginations at the base, sides, and rim of the chloroplast cup. However, we did note that invaginations were especially prevalent near thylakoid tip convergence zones and at the base-lobe junction (Figure 7), where the base and the sides of the chloroplast cup meet.

12While invaginations were found everywhere along the chloroplast envelope inner membrane, we noticed that many invaginations occurred in close proximity to thylakoid tips (Figure 6A–D and 7). In our dataset, 34% of the invaginations were positioned less than 40 nm from a thylakoid tip, and 64% were found within 100 nm of a tip (Figure 8D). Thus, inner membrane invaginations may provide a route for the exchange of lipids and proteins between the chloroplast envelope and thylakoid compartments. Indeed, we observed a few direct connections between thylakoid tips and the chloroplast envelope inner membrane (Figure 6E–H). These connections were rare, which could help maintain the distinct identities of each membrane and preserve the high proton concentration within the thylakoids.

13The cluster of RuBisCO complexes in the pyrenoid was encased in a starch sheath (Figures 1G, 3A–B and 9A–B), and non-spherical starch granules were found throughout the chloroplast (Figures 1G, 3A–B, 4A–B and 7A–B). In addition to starch, we observed smaller 64 ± 15 nm spherical plastoglobules (n = 20 globules from 9 tomograms) that were always closely associated with thylakoid membranes (Figure 4). Higher plant plastoglobules were previously shown by freeze-fracture to share a half-lipid bilayer with thylakoid membranes. Given the caveat that the resolution of our CCD-acquired tomograms did not reveal the two layers of lipid bilayers, Chlamydomonas plastoglobules appeared to be discontinuous from thylakoid membranes, although punctate contacts were sometimes observed (Figure 4A,D). Additionally, while higher plant plastoglobules are frequently localized to areas of high membrane curvature, where they are proposed to form, Chlamydomonas plastoglobules were primarily found adjacent to flat thylakoid sheets (Figures 4, 7A–B and 9A). Although the composition of Chlamydomonas plastoglobules is currently unknown, they are believed to contain lipids and thylakoid components similar to the plastoglobules of higher plants. Consistent with this hypothesis, Chlamydomonas plastoglobule structure remained intact at electron doses that did not affect lipid membranes but damaged the radiation-sensitive starch granules (seen by bubbling within the granules, Figure 4A). Such a ‘bubblegram’ approach can discriminate between chemically distinct entities in cryo-tomograms.

14Membrane tubules extend from the thylakoids into the pyrenoid through fenestrations in the starch sheath. We aimed to better understand the function of these intriguing structures by visualizing their native 3D architecture (Figures 9–11). Thylakoid stacks often changed direction (Figure 9A–D) as their sheets merged with each other and narrowed into cylindrical pyrenoid tubules (Figure 10), which had a mean diameter of 107 ± 26 nm (n = 46 tubules from 10 tomograms). A 5–10 nm RuBisCO-free space was frequently observed surrounding the tubules (Figure 9E,G). Deeper within the pyrenoid, pyrenoid tubules began to twist (Figure 9A, top tubule), sometimes abruptly changed direction by >90° (Figure 9B, bottom tubule), and eventually lost their cylindrical structure as the tubules converged at the pyrenoid center, forming elaborate interconnected networks of variably shaped smaller membranes (Figure 9B).

15Pyrenoid tubules have previously been reported to contain ‘ridges’ or ‘infoldings’ along the inner surface of the tubule membrane. Interestingly, in high magnification cryo-tomograms these structures were revealed to be smaller tubules (Figure 9E–H and Figure 9—figure supplement 1), which we have termed pyrenoid minitubules. Each pyrenoid tubule encompassed between two and eight minitubules, with a median of 5. In cross-section, the minitubules had an oval or bean-like shape (Figure 9E,G) with a long-axis diameter of 21.6 ± 1.7 nm and a short-axis diameter of 13.4 ± 1.1 nm, including the membrane (n = 22 minitubules within four pyrenoid tubules). The minitubules enclosed hollow lumina that were 3.5 ± 0.5 nm wide along the short-axis, comparable in width to the interthylakoid stromal space.

16Segmentation of the stack-to-tubule transition showed that the lumina of pyrenoid tubules were continuous with the lumina of thylakoids, while the lumina of pyrenoid minitubules were continuous with the stroma in the interthylakoid space (Figure 10). Minitubule membranes were derived from the coalescence of two opposed membranes of adjacent thylakoids in a stack (Figure 10C–F). Consequently, minitubule membrane topology is ‘inside-out’ relative to the membranes of pyrenoid tubules. At the periphery of the pyrenoid, minitubules were completely enveloped within the surrounding pyrenoid tubule membrane (Figure 9E–H). As pyrenoid tubules proceeded towards the center of the pyrenoid, minitubules emerged from the pyrenoid tubules, exposing their stromal lumina to the pyrenoid matrix (Figure 11A–E). Thus, pyrenoid minitubules form continuous conduits between the interthylakoid stromal space and the RuBisCO complexes in the pyrenoid matrix. Further inside the pyrenoid, pyrenoid tubules merged with each other, producing additional internal membrane structures with luminal spaces that were also continuous with the pyrenoid matrix (Figure 11F–J).

17RuBisCO accounts for over 90% of the pyrenoid's protein content. Such homogeneity suggests that the pyrenoid may exhibit higher-order organization, and indeed, ‘crystalline’ packing of RuBisCO complexes has been described in a few algal pyrenoids as well as bacterial carboxysomes. However, the pyrenoids of most species, including those found in Chlamydomonas, appear amorphous. Since these conclusions about pyrenoid architecture are based on 2D EM of plastic sections, we decided to investigate whether RuBisCO complexes were ordered within our native-state tomographic volumes of the Chlamydomonas pyrenoid.

18Upon visual inspection, the dense matrix of RuBisCO complexes did not obviously appear to be arranged in a crystalline lattice (Figure 12A). However, symmetry-free subtomogram averaging of 45.6 nm3 subvolumes revealed that the RuBisCO complexes were actually hexagonally packed (Figure 12B). The intensity of the average was somewhat asymmetric due to local deviations from a perfectly packed crystal. Nevertheless, each of the 13 densities, corresponding to a unit cell of a RuBisCO complex and its neighbors, were located close to the ideal centers of hexagonal close packing (Figure 12B, blue diamonds). Furthermore, the average clearly showed the A-B-A layered arrangement of complexes that is a hallmark of hexagonal close packing (Figure 12B, bottom right).

19We determined the relative center positions of the RuBisCO complexes via an independent high-precision method (Figure 13). After localizing each RuBisCO complex in our tomograms by template matching, the nearest neighbors of each RuBisCO were overlaid in a point cloud, which again showed a clear A-B-A hexagonal arrangement (Figure 13B). The center positions of each neighbor in the point cloud were calculated by k-means clustering (Figures 12C and 13, red diamonds). We generated ideal hard sphere models and found that hexagonal close packing had a smaller root-mean-squared deviation (RMSD) from our point cloud center positions than cubic close packing or body-centered cubic packing (Figure 12D). Additionally, hexagonal close packing with a spherical diameter of 15.05 nm yielded a unique minimum RMSD, indicating that RuBisCO complexes within the pyrenoid have ∼15 nm between their centers (Figure 12D). Given the dimensions of RuBisCO, complexes positioned 15 nm apart would have a 2–4.5 nm space between their surfaces (Figure 12E).

20The eyespot apparatus is a specialized region of the chloroplast and adjacent plasma membrane that controls phototaxis by detecting the intensity and direction of light. Viewed by cryo-ET, the eyespot globules were similar in appearance to plastoglobules dispersed throughout the chloroplast (Figure 14A), although they were slightly larger (Figure 14G) and densely packed along two adjacent thylakoid membranes. The two arrays of eyespot globules, the associated thylakoids, and the bordering chloroplast envelope were all slightly concave relative to the plasma membrane (Figure 14B). However, Chlamydomonas eyespots with straight or convex globule arrays have frequently been observed in plastic section TEM images, indicating that eyespot curvature may be variable. The eyespot globules are proposed to reflect incident light coming from outside the cell back to channelrhodopsin photoreceptors in the overlying plasma membrane, while simultaneously absorbing light passing through the cell from other directions. Concave eyespot architecture may enhance reflection, as it has been shown to focus reflected light in a different species of green algae.

21We observed a cluster of unilamellar and bilamellar vesicles embedded between the cell wall and the rhodopsin-containing region of the plasma membrane (Figure 14D–F). To our knowledge, such structures have not been previously described, although voids in the cell wall near the eyespot have been seen in conventional plastic sections (G Kreimer and S Geimer, personal communication, May 2014). Comparing all of our tomograms containing the cell wall, these vesicles were only observed adjacent to the eyespot apparatus, implying that they may be related to eyespot function.

22Unlike higher plants, algal chloroplasts do not have a distinct division between grana and stroma thylakoids. Comparing the structures of algal and higher plant thylakoid stacks may shed light on the evolution of photosynthetic systems. The Chlamydomonas interthylakoid stromal space (3.6 nm) is comparable to the grana interthylakoid stromal space (3.2–3.6 nm), indicating that the molecular mechanisms that hold two algal or higher plant thylakoid membranes together may be similar. This attraction is proposed to involve van der Waals and electrostatic interactions between membrane surfaces, perhaps mediated by light-harvesting complexes bound to photosystem II (PSII). The oxygen-evolving complex (OEC) of PSII projects 4.5 nm into the thylakoid lumen, a comparable distance to the 4.5–4.7 nm width of grana thylakoid lumina measured for isolated chloroplasts and leaves of dark-adapted plants. Thus, OECs on opposite sides of these narrow lumina must interdigitate due to spatial constraints, and indeed, it seems likely that the size of the OEC determines the minimum grana lumen width. Our Chlamydomonas cells, which were grown in constant light until the moment of vitrification, had 9 nm thylakoid lumina. Interestingly, this may be a result of light conditions rather than an algal characteristic, as leaves that were harvested from light-exposed plants and immediately cryo-immobilized had 9.2 nm grana lumina. Thus, PSII complexes in Chlamydomonas thylakoids and light-exposed higher plant grana should experience fewer limitations to their mobility, and their OECs may directly face each other instead of being interdigitated. Examination of dark-adapted Chlamydomonas cells would reveal whether the light-induced change in thylakoid lumen width is conserved between algae and higher plants.

23While much research in higher plants has focused on the mechanisms that organize thylakoid stacks, little is known about how proteins orchestrate the dramatic membrane curvature seen in our tomograms, including the 180° loops between thylakoid tips, the transition from flat thylakoid stacks to cylindrical pyrenoid tubules, and the very thin pyrenoid minitubules. One good candidate for the 180° loops is CURT1, an evolutionarily conserved family of proteins that oligomerize, tubulate membranes, and are enriched at the grana margins of higher plants. Recent technical advancements in correlative microscopy and cryo-ET will enable the in situ localization and structural characterization of the protein complexes at curved thylakoid and pyrenoid tubule membranes.

24While the thylakoid and chloroplast envelope membranes are widely believed to be independent systems, thylakoids have a similar lipid composition to the chloroplast envelope inner membrane and require the import of proteins encoded by the nuclear genome. Thus, transport between these systems must be possible. Over one third of the inner membrane invaginations in our tomograms were located within 40 nm of a thylakoid tip (Figure 8D), less than the width of two stacked thylakoids. Taken together with the rare direct connections we observed between the inner membrane and thylakoid tips (Figure 6E–H), it is likely that there is a transport mechanism for the exchange of lipids and proteins between these two compartments. Consistent with this idea, there is growing evidence that vesicular transport from the chloroplast envelope inner membrane is required for thylakoid biogenesis. If such a transport pathway exists, it will be interesting to understand how the specific targeting to thylakoid tips is achieved.

25The thylakoid tip convergence zones that we described in this study (Figures 3 and 4) are likely sites where there is prolific exchange of lipids and proteins between the thylakoid and chloroplast envelope membranes. Although thylakoid tip convergence zones at the base and rim of the chloroplast cup appear morphologically similar, the question remains as to whether they all perform the same function. Recent work suggests that Chlamydomonas thylakoid biogenesis may be spatially regulated. In the model proposed by these studies, thylakoid proteins encoded by the chloroplast genome are locally translated at the base of the chloroplast cup in a region surrounding the pyrenoid called the T-zone. At the base-lobe junction, where the base and sides of the cup meet, nuclear-encoded proteins are imported and assembled with the chloroplast-encoded proteins into macromolecular complexes in the thylakoid membranes. Finally, mature thylakoids in the sides of the cup carry out the photosynthetic light reactions. If this model is correct, the thylakoid tip convergence zones at the base and rim of the chloroplast cup likely perform different functions. The base of the cup is located close to the ER and Golgi, as well as many cytoplasmic ribosomes (Figure 1G), and thus convergence zones at the base-lobe junction are ideally situated for the import of new proteins. Indeed, in this junction region we observed small complexes associated with both the chloroplast envelope outer membrane and inner membrane invaginations, indicating that these complexes may be undergoing import (Figure 7).

26Unlike thylakoid tip convergence zones near the base of the chloroplast cup, convergence zones at the rim of the cup are located far from the site of thylakoid biogenesis and only contain mature thylakoids. One possibility is that the rim could be the site where old and photodamaged thylakoid proteins are transported out of the chloroplast for degradation. In this source-sink model, there would be a constant flux of proteins from the source at the chloroplast base to the sink at the rim of the cup. It is also worth considering that the apical rim of the chloroplast is positioned close to the flagella, the only part of the cell's external membrane that comes into direct contact with the extracellular environment (Figure 1A). The rim is thus well situated for the uptake and release of molecules that cannot permeate the cell wall. It is possible that thylakoid tip convergence zones are a conserved feature of algal chloroplasts, as similar structures can be found in a wide range of algae. Furthermore, thylakoid tip convergence zones may be homologous to the biogenesis centers of cyanobacteria.

27Although the aggregation of RuBisCO within the pyrenoid is advantageous for utilizing the high local concentration of CO2, this causes RuBisCO to be physically isolated from the photosynthetic pathways that occur in the thylakoid stacks and chloroplast stroma. The pyrenoid tubules appear to solve this problem, as they connect these regions of the chloroplast by serving as conduits through the dense starch sheath.

28The RuBisCO catalytic reaction does not use ATP or NADPH. However, RuBisCO activase is also localized to the pyrenoid and requires ATP to promote RuBisCO activity. While immunolabeling data indicates that many Calvin–Benson cycle enzymes may be restricted to the stroma, phosphoribulokinase (PRK), which uses ATP to carry out the final step of ribulose-1,5-bisphosphate (RuBP) regeneration, has been detected in close association with the pyrenoid. Thus, the pyrenoid likely requires ATP from the chloroplast stroma. Conversely, the product of carbon fixation, G3P, must be exported from the pyrenoid. If other Calvin–Benson cycle enzymes do indeed exclusively reside in the stroma, then regeneration intermediates of the Calvin–Benson cycle would also need to exchange efficiently between the stroma and the pyrenoid.

29The pyrenoid minitubules identified in this study (Figure 9E–H) form narrow continuous channels between the interthylakoid stromal space and the pyrenoid matrix (Figures 10 and 11). Therefore, we propose that minitubules facilitate the rapid targeted diffusion of small molecules such as ATP and Calvin–Benson cycle sugars between these two compartments. The minitubule lumen diameter is only 3–4 nm by 8–15 nm, while the longest axes of ATP and the sugars RuBP and G3P measure ∼1.4 nm, ∼1.2 nm, and ∼0.7 nm, respectively. Thus, diffusion of these metabolites through the minitubules would be practically one-dimensional. While we are unsure how the local minitubule environment may affect the diffusion coefficient, assuming a diffusion coefficient of a small molecule in the cytoplasm, it would only take a few milliseconds to travel 1 μm through a minitubule.

30In addition to bundling minitubules and directing them towards the pyrenoid center, the larger pyrenoid tubules may play an important role in supplying RuBisCO with saturating levels of CO2. It is believed that HCO3− in the thylakoid lumen is converted to CO2 by carbonic anhydrase 3 (CAH3), a reaction that is aided by the high H+ concentration in the lumen. Since membranes are much more permeable to CO2 than HCO3−, CO2 would then freely diffuse out of the thylakoids and into the pyrenoid. Under low-CO2 conditions, CAH3 is phosphorylated and moves from the thylakoids to the pyrenoid tubules, where it can efficiently supply RuBisCO with CO2. Thus, while minitubules may deliver the energy and sugars for carbon fixation, the larger pyrenoid tubules likely provide the CO2.

31The structure of the pyrenoid is dynamic and responds to environmental changes including light, temperature, and atmospheric composition. When cells are grown in low light, the pyrenoid matrix becomes fenestrated with more pyrenoid tubules. Such an inducible condition may allow the examination of pyrenoid tubule assembly and growth through the pyrenoid matrix.

32The hexagonal arrangement of RuBisCO complexes within the pyrenoid could be caused by one of two distinct influences. Either there are specific interactions between RuBisCO complexes that enforce this hexagonal configuration, or non-specific packing forces resulting from crowding within the pyrenoid compress the complexes together, yielding hexagonal close packing of roughly spherical objects. If RuBisCO complexes do make specific interactions with each other, this would indicate that there may be allosteric molecular mechanisms for increasing the efficiency of carbon fixation. Two studies found evidence that the RuBisCO small subunit may mediate inter-complex interactions in Chlamydomonas. The hydrophobic residues in the surface-exposed α-helixes of the RuBisCO small subunit were shown to be necessary for pyrenoid formation, and RuBisCO aggregation was prevented in vivo when these domains were replaced by less hydrophobic sequences from higher plants. This implies that hydrophobic interactions between RuBisCO complexes may drive pyrenoid formation.

33Alternatively, the hexagonal arrangement of RuBisCO particles may be caused by non-specific pyrenoid packing forces. We found that the centers of RuBisCO complexes in the pyrenoid were ∼15 nm apart (Figure 12D), even though RuBisCO is shaped like a rounded cube with a shortest diameter of ∼10.5 nm and a longest diameter of ∼13 nm. Thus, there is a 2–4.5 nm space between RuBisCO complexes, depending on their relative orientations (Figure 12E). This data suggests that RuBisCO complexes within the pyrenoid may not directly interact with each other and that other factors, such as a dense matrix of proteins and carbohydrates or small linker proteins between the complexes, determine the RuBisCO spacing. In the case of non-specific packing, the key question becomes how these packing forces are generated. One possibility is that since the pyrenoid is encased in a dense starch sheath, packing forces could increase as additional RuBisCO complexes are inserted into this relatively fixed volume. Such a model raises the question of how RuBisCO complexes are targeted through the starch sheath during pyrenoid growth.

34In different classes of bacterial carboxysomes, both random and specific packing of RuBisCO have been observed. In α-carboxysomes, which are found primarily in ocean-dwelling cyanobacteria, non-specific packing forces alone are believed to be sufficient to induce the layered organization of RuBisCO. However, in the β-carboxysomes of freshwater cyanobacteria, RuBisCO complexes are held in an organized array by CcmM linker proteins.

35While our data cannot discriminate between random and specific causes of hexagonal RuBisCO packing in the Chlamydomonas pyrenoid, several lines of investigation may be able to help tease apart this question. Technical advances such as direct detection cameras continue to improve the resolution of cryo-tomograms. Once higher resolutions are attained, it will be possible to use template matching to determine whether RuBisCO complexes have a preferred orientation relative to each other, a strategy that has already been successfully applied to ribosomes. If RuBisCO complexes within the pyrenoid do indeed have preferred orientations, template matching will also reveal which surfaces likely mediate RuBisCO aggregation. A complementary approach would be to examine how RuBisCO packing changes under conditions where the pyrenoid becomes less organized, including mutant strains, elevated levels of atmospheric CO2, and during RuBisCO aggregation in newly divided cells at dawn. If the hexagonal packing disappears in disorganized pyrenoids, this would indicate that the packing is driven by non-specific pyrenoid forces. However, if islands of hexagonal packing persist even when RuBisCO complexes are not being forced together, this would provide evidence for inter-complex interactions. Additionally, the pyrenoid forms in mutants that lack a starch sheath, enabling a direct test of whether the starch induces hexagonal packing of RuBisCO.

36By combining cryo-FIB with cryo-ET, we acquired the first in situ 3D views of the algal chloroplast, native structures that provide new insights into the mechanisms of thylakoid biogenesis, photosynthesis, and carbon fixation. The challenge now lies in characterizing the identity, localization, and geometry of the macromolecular complexes that orchestrate the chloroplast's elegant architecture.

37C. reinhardtii wild-type strain CC-124 (137c) was acquired from the Chlamydomonas Resource Center (University of Minnesota, Minneapolis, MN) and grown in Tris-acetate-phosphate (TAP) medium with constant light (∼10,000 l×) and aeration with normal atmosphere. CC-124 was chosen over the CC-125 wild-type strain due to smaller cell size, which yields better vitrification by plunge-freezing.

385 μl of liquid culture (diluted to 1000–750 cells/μl in fresh TAP) was pipetted onto holey carbon-coated 200 mesh copper grids (Quantifoil Micro Tools, Jena, Germany). The grids were blotted from the reverse side and immediately plunged into a liquid ethane/propane mixture at liquid nitrogen temperature using a Vitrobot Mark 4 (FEI, Eindhoven, The Netherlands). The Vitrobot was set to 25°C, 90% humidity, blot force 10, and 4–7 s blot time. The grids were stored in sealed boxes in liquid nitrogen until used.

39Cryo-FIB milling was performed as described in detail at Bio-protocol. Plunge-frozen grids were mounted into autogrids that were modified for FIB milling (Autogrid sample holder, FEI), providing stability to the EM grids during sample preparation and transfers. These autogrids were mounted into a dual-beam (FIB/SEM) microscope (Quanta 3D FEG, FEI) using a cryo-transfer system (PP3000T, Quorum) with a custom-built transfer shuttle. During FIB operation, samples were kept at a constant temperature of −180°C using a homemade 360° rotatable cryo-stage. To improve sample conductivity and reduce curtaining artifacts during FIB milling, the whole Autogrid was coated by organometallic platinum (protocol adapted from ) using the in situ gas injection system (GIS, FEI) and the following parameters: 13.5 mm stage working distance, 26°C GIS pre-heating temperature, and 4 s gas injection time. Thin lamellas were prepared using the Ga+ ion beam at 30 kV under a shallow 8°–12° angle. Rough milling was performed with a rectangular pattern and 0.3 nA beam current, followed by sequentially lowered beam currents of 0.1 nA, 50 pA and 30 pA during the thinning and cleaning steps. During the milling, the specimen was also imaged using the SEM at 3–5 kV and 5–30 pA.

40Tomography was performed on a Tecnai G2 Polara transmission electron microscope (FEI, Eindhoven, Netherlands) equipped with a field emission gun operated at 300 kV, a GIF 2002 post-column energy filter (Gatan, Pleasanton, CA), and either a 2048 × 2048 Gatan slow scan CCD camera or a 3838 × 3710 Gatan K2 Summit direct detection camera (only Figures 7 and 11A–E). Tilt-series acquisition was controlled by SerialEM software under low-dose conditions (∼100 e/Å2 cumulative dose). CCD images were recorded at 2° tilt increments, with −8 μm to −12 μm defocus, at 31,500×, 42,000×, and 52,500× magnifications (pixel sizes of 9.6 Å, 7.1 Å, and 5.7 Å). 52,500× was the highest possible magnification that still prevented excessive electron damage to the sample. K2 images were recorded at 2° tilt increments, with −5 μm defocus, at 11,800× and 14,500× magnifications (pixel sizes of 4.2 Å and 3.4 Å). The dataset for this study consisted of 43 tomograms from 27 FIB lamella preparations.

41Fiducial-less alignment of the tilt-series was performed by feature tracking with TOM software or patch tracking with IMOD software, and tomograms were reconstructed using the weighted back projection algorithm. Segmentation of tomogram volumes was performed with Amira software (FEI Visualization Sciences Group).

42Subtomograms of pyrenoid regions were binned once to a pixel size of 11.4 Å for localization of RuBisCO complexes. RuBisCO complexes were detected by template matching using a spherical template with a diameter of 13.68 nm. A spherical template, which functioned well due to the homogeneity of pyrenoid composition and the roughly spherical shape of RuBisCO at the tomogram resolution, made it possible to discard the rotational search, dramatically increasing the localization speed and enabling iterative fine-tuning of parameter values. For template matching, the pyrenoid subtomograms were low-pass filtered using the Crowther criterion to the resolution of a 12 nm diameter particle with the same 2° tilt increment as the tomographic tilt series. Cross-correlation peaks were exhaustively extracted, yielding a set of RuBisCO particle positions. Over 10,000 subvolumes (45.6 nm3) centered on these RuBisCO positions were selected from the unfiltered tomograms in order to analyze the local neighborhood around each RuBisCO complex. The subvolumes were subsequently aligned and averaged with the PyTom toolbox as previously described.

43For each RuBisCO position detected by template matching, a range query was performed to select a neighborhood of RuBisCO positions within a 22.8 nm radius. The transformations obtained by subvolume alignment were applied to their corresponding point neighborhoods to generate a cloud of 3D points describing the local arrangement of RuBisCO complexes in the same reference frame as the subvolume average. K-means clustering of the 3D point cloud (with k = 13) yielded the center positions of the reference RuBisCO and its 12 nearest neighbors. Subsequently, the unit cells of different hard sphere lattice arrangements (hexagonal close packing, cubic close packing, and body-centered cubic) were simulated for a series of sphere diameters. The center points of these hard spheres were compared to the cluster center positions by computing the RMSD values of nearest point correspondences, revealing the packing model and sphere diameter that best matched the data.

44Atomic models of RuBisCO complexes with different relative orientations were generated with Visual Molecular Dynamics (VMD). Molecular dimensions and inter-complex distances were measured with VMD and UCSF Chimera.