|Home | About | Journals | Submit | Contact Us | Français|
Carboxysomes are polyhedral bodies consisting of a proteinaceous shell filled with ribulose 1,5-bisphosphate carboxylase/oxygenase (RuBisCO). They are found in the cytoplasm of all cyanobacteria and some chemoautotrophic bacteria. Previous studies of Halothiobacillus neapolitanus and Nitrobacter agilis carboxysomes suggest that the structures are either icosahedral or dodecahedral. To determine the protein shell structure more definitively, purified H. neapolitanus carboxysomes were re-examined by cryo-electron tomography and scanning transmission electron microscopy (STEM). Due to the limited tilt angles in the electron microscope, the tomographic reconstructions are distorted. Corrections were made in the 3D orientation searching and averaging of the computationally extracted carboxysomes to minimize the missing data effects. It was found that H. neapolitanus carboxysomes vary widely in size and mass as shown by cryo-electron tomography and STEM mass measurements, respectively. We have aligned and averaged carboxysomes in several size classes from the 3D tomographic reconstruction by methods that are not model-biased. The averages reveal icosahedral symmetry of the shell, but not of the density inside it, for all the size classes.
The initial source of Earth’s organic carbon is atmospheric or dissolved CO2, which is fixed by enzymes found in green plants, algae and autotrophic bacteria. The carbon is then spread through the food chain to all other living organisms. The most important enzyme responsible for fixing carbon is ribulose bisphosphate carboxylase/oxygenase (RuBisCO), which catalyzes the first step in carbon fixation via the Calvin-Benson-Bassham cycle-the covalent attachment of CO2 to ribulose-1,5-bisphosphate and its subsequent cleavage into two molecules of 3-phosphoglycerate. Based on the immense biomass of photosynthetic and chemoautotrophic organisms on Earth, RuBisCO is estimated to be the most abundant protein known.
All cyanobacteria and some chemoautotrophic bacteria enhance their CO2 fixation by sequestering RuBisCO into polyhedral bodies called carboxysomes 1;2;3;4. Halothiobacillus neapolitanus carboxysomes are delimited by a proteinaceous shell and are filled with RuBisCO 5; 6; 7. In previous electron microscopic studies, carboxysomes appeared hexagonal with a granular interior, a diameter of approximately 120 nm and a shell thickness of between 3 to 4 nm 8; 9. Although the enhancement of carbon dioxide fixation by the carboxysome has been firmly established, the exact mechanism has not yet been elucidated.
Halothiobacillus neapolitanus carboxysomes consist of ten polypeptides 6; 7. Of these, two polypeptides represent the small and large subunits of RuBisCO and six are known to make up the shell. One of these shell peptides has been identified as a unique carbonic anhydrase 10. It is postulated that this enzyme may function to enhance carbon dioxide fixation by the carboxysome. The functions and locations of the two remaining polypeptides are not yet known. Carboxysomes are rapidly formed via de novo m-RNA and protein synthesis under low CO2 conditions 11; 12. Their genes are operon-linked12, so it is likely the proportions of their constituents are regulated.
Although much has been published on the occurrence, physiology, biochemistry and genetics of these microcompartments/organelles 5; 6; 7; 10, only two major reports have analyzed the structure or symmetry of purified carboxysomes 9; 13. Peters concluded that the carboxysomes of Nitrobacter agilis obey icosahedral symmetry, based on negative stain projection images and heavy-metal shadowing of critical-point-dried carboxysomes 13. Holthuijzen and co-workers, on the other hand, reported the shape of carboxysomes of H. neapolitanus to be dodecahedral 9.
There are reports that the RuBisCO inside the shell is paracrystalline and fills the interior 8. Others have reported that the RuBisCO molecules are arranged in one or a few layers under the surface of the shell 9. However, these conclusions were drawn from observations of either whole bacteria or isolated carboxysome preparations that are plastic-embedded, thin-sectioned, and/or negative stained. Using Hilbert Differential Contrast electron microscopy, paracrystalline arrays were observed in intact frozen cells 14, but it is not possible in these 2D projections to tell whether the carboxysomes are filled, or have only one or a few layers of RuBisCO under the shell. Orus et al. 11 reported that the state of packaging of RuBisCO inside carboxysomes can vary with environmental conditions.
It is these aspects of the carboxysome structure that we intend to clarify using cryo-electron tomography of suspensions of carboxysomes isolated from the chemoautotrophic proteobacterium H. neapolitanus and subsequent 3D alignment and averaging of particles computationally isolated from the tomogram. Averaging particles is necessary because of the low signal-to-noise ratio in any single instance of a particle from the tomogram, and because of the distortion introduced by the “missing wedge” of data in Fourier space. However, the search for the relative orientation of the 3D particle volumes is not a trivial step unless a 3D model of the particle with isotropic resolution already exists, as has previously been done15; 16. In our study, an orientation search method that does not depend on such a prior model is used for aligning single particles from a tomogram with missing wedge data.
Figure 1 shows the zero tilt image from a tilt series of a frozen-hydrated suspension of isolated carboxysomes. All the images in the tilt series are shown in Supplemental Movie 1. The thin, polyhedral nature of the shell of the carboxysomes and their globular contents, presumed to include RuBisCO, are apparent. Also evident is that the particles appear to possess marked size variability.
Figure 2a is a view of a representative area of the tomogram. Four consecutive sections from the middle of the tomogram are averaged along z. Features annotated in this view are the shell, which is about 4 nm thick; the slightly thicker and denser vertices; the RuBisCO molecules inside the particles; and that some particles are more fully packed than others. Figure 2b is a stereo view of the same area of the tomographic reconstruction. The entire tomogram is shown in Supplemental Movie 2. From this raw tomogram we were able to gather insight into this preparation of carboxysomes. As mentioned, we observed that a few of the carboxysome shells were completely empty. In several of them, only part of the interior was occupied. In these particles, the preferred arrangement of RuBisCO was sometimes seen as a layer or two apposed to the inside surface of the shell. However, the majority of the particles had what appears to be a homogeneous distribution of RuBisCO throughout the interior of the shell. Inspection of this entire reconstruction from a suspension of isolated carboxysomes (Supplemental Movie 2) showed that there were a number of small “free” particles outside the shells that appeared very similar to the interior RuBisCO molecules of carboxysomes. Their distribution was not random. They were preferentially disposed near the top and bottom of the reconstruction and appeared to be attracted to the air-water interface (Supplemental Movie 2). In addition, there are large particles at both surfaces of the vitreous ice that appeared to be ice contamination. The carboxysomes appear to lie between these two interfaces and to be well embedded in the vitreous ice as evidenced from the tomogram. From these observations and our assumption about the ice contamination and the “free” particles indicating the limits of the ice layer, we estimate the ice to be about 100–150 nm thick, slightly larger than the diameter of the carboxysomes themselves.
There were about 200 carboxysomes visible in this single 3D tomographic reconstruction. Of these, 160 were chosen that were not obviously malformed or broken, and were not too close to the edge of the field of view. Some carboxysomes that were near the edges did not appear in every tilt series image because of the shifting field of view between different tilts in the tilt series (see Supplemental Movie 1). The 3D volumes containing these 160 particles were extracted from the tomogram and centered by autocorrelation with their corresponding mirror images. To align and merge these particles to each other in 3D in an unbiased way, an all-vs-all comparison in 3D was envisioned. However, this would have involved 160×159/2 searches of the 6 dimensional parameter space (3 rotation and 3 translation), and was deemed unfeasible. Therefore the data was split up into 8 random groups of 20 particles each, and all-vs-all 3D orientation searches, followed by averaging in 3D, when appropriate, were done for each group.
We encountered a difficulty in these comparisons having to do with the geometry of data collection for tomographic reconstructions. Because a tilt angle range of ±90° cannot be achieved in the electron microscope, central sections in Fourier space for tilts higher than 60–70° are missing, leading to the problem commonly referred to as the “missing wedge” in Fourier space. After extraction from the tomogram and before applying any rotations, the orientation of the missing wedge in Fourier space is identical for each particle/volume. However, the orientations of the particles themselves are random with respect to each other and must be searched for in 3D rotation space. Because of the strong signal from the pattern of the missing wedge, the cross-correlation map may merely lead to the alignment of the missing wedges, which is certainly not correct. This problem does not occur to the same extent in cases where an isotropic starting model is available and used for 3D alignment15; 16.
As two 3D volumes in Fourier space, each containing a missing wedge, are rotated against each other in the cross-correlation search, and one volume is multiplied by the complex conjugate of the other, zeros are generated. They occur when the missing-data region for one of the particles is multiplied by data in the other particle and vice versa, and the number of such zeros changes at each rotation angle increment in the search. To minimize the effect of the missing wedge on the overall scale of the 3D cross-correlation map, we computed the number of non-zero terms after the complex multiplication in Fourier space for each relative 3D orientation in the space searched for the two particles, and scaled the 3D cross-correlation map at that orientation search angle by the reciprocal of this number. This correction factor is approximately right because the total power in the Fourier transform of a cross-correlation function influences the values of the cross-correlation map. This normalization was good enough to allow the true proper relative orientation of the particles to be determined as demonstrated from simulated data with a missing wedge introduced (M.F. Schmid, unpublished).
In each of the 8 groups in this data set, the six parameters which yielded the highest cross-correlation coefficients between pairs of particles were used to align and average these pairs in 3D, and then the averages were used as new search models. New particles were added to these averages, gradually including more particles. This iterative approach is the same as was used for the images of acrosomal bundle for unbiased merging 17. Each of the 8 groups yielded 1 to 4 classes of particles that merged well with each other in that class but not between each other in that group. The structural basis of this classification was not investigated at this stage. About half the particles in each group fell into these classes, yielding a total of 92 particles that seemed to be in self-consistent classes. The purpose of this initial classification was to generate subsets of particles that shared a similarity to at least a few others, and were not unique outliers.
The second step of data classification was to divide the 92 particles up by size, because considerable size variation was still seen in the particles after the initial classification step. EMAN software was used to calculate the 1D radial density profiles from the 3D density map of each particle. They were classified according to the maximum radius of the 1D plot. Figure 3 is a histogram of the particle diameters based on these radii. The histogram was initially divided up arbitrarily into nine size classes, from 88 nm to 106 nm in average diameter. Again, all-vs-all 6D cross-correlation searching and merging was done and the averages for each size class were created, containing up to 21 particles in each class. All the particles were cross-correlated against these averages. The particles usually correlated best with the size class that originally contained them, although a few correlated better with the size class above or below them, and were thus shifted into these classes. The result of this shifting was that the members of two of the original nine size classes migrated to other groups, and thus at the end, only seven stable size classes remained. Iterative merging was done with progressively finer steps of rotation search, down to 1-degree intervals. This produced the final size-classified averages. For these final averages, we scaled the contributions of each particle by weighting each voxel’s contribution in Fourier space by its fraction of the total amplitude. The major effect of this correction, as expected, was to increase the visibility of the thin shell where it is oriented perpendicular to the z direction, while the other features were basically unchanged.
Figure 4a shows the average of the most highly populated class with no symmetry enforced, having a diameter of about 100 nm. It appears to obey icosahedral symmetry as characterized by the appearance of 5 (approximately along the view direction), 3 and 2 fold symmetry axes and 20 triangular faces. Similar averages were produced for other size classes and also have similar symmetry appearance. This observation justifies the application of icosahedral averaging as shown in Figure 4b. Figure 5a–g shows a gallery of views of all seven surviving size class averages with icosahedral symmetry enforced. The Fourier Shell Coefficient (FSC) was calculated by splitting the 3D volumes for the largest class of particles, whose overall average is shown in Figure 4b, into two separate averages (odd and even numbered members). The resolution as indicated by the 0.5 threshold of FSC is 6.6 nm.
The shell of the carboxysome is rather thin, about 3–4 nm thick (see Figure 2a and Supplemental Movie 2). At the vertices, there is a small “button” of higher density in all the size classes (Figure 2a). One of the major differences between the seven different size classes of carboxysomes (Figure 5a–g) is the organization of the density at the center of the triangular faces. One type of organization, followed by the shells of average diameter 88, 93, 95 and 103 nm, has a higher density at the center of the face, surrounded by 3 or 6 densities abutting it (Arrow, Figure 5a). It is interesting that the same spacing is seen in the sheets of hexamers in the crystal structure of a homologous shell protein from the carboxysome of Synechocystis and whose hexagonal organization in the crystal lattice was proposed to be a model for the carboxysome shell 18. The other type of organization, found in shells of diameter 91, 97 and 100 nm, has 3 densities surrounding the center of the 3-fold face instead of being at the center itself as above, again with spacing similar to those in the crystal (Arrow, Figure 5b). In addition to this difference in the arrangement for different size classes, the orientations of the densities with respect to the vertices of the shell are different in the different size averages. This suggests that the organization of the shell proteins is different for different size classes.
In addition to the Csos1A and C polypeptides, which comprise 13% of the carboxysome’s mass 5 and are homologous to CcmK2 and K4 in the crystal structure, there are up to five other proteins present in the shell.
The crystal structure of the carbonic anhydrase of H. neapolitanus 19, which is thought to be a stoichiometrically minor component of the shell (2% of the carboxysome’s mass 2; 5), is too thick to be part of the triangular faces. One possible hypothesis is that it is located at the thicker 5-fold vertex, where there is sufficient volume for it in all dimensions. If so, there may be only 60 copies per carboxysome, the minimum required for icosahedral symmetry.
Each individual carboxysome shows globular features inside the carboxysome having density comparable to the shell (e.g. Figure 2a and b, Supplemental Movie 2 movie and Figure 6a showing a radial density plot of an individual particle). However, Figure 6b shows that in the multi-particle and icosahedrally averaged reconstructions, the density inside the shell is much lower on average than the density of the shell. This would indicate that, while the shell is definitely icosahedral, the interior density, which is likely to include RuBisCO molecules, is probably not. However, the inside density in the radial or icosahedral averages is still concentrated in layers approximately equally spaced from each other and starting at about 6 nm from the inside of the shell (Figures 6a and b). This first layer under the shell in the icosahedral averages always has the highest density. From this we deduce that the RuBisCO has a certain probability of being at these various radii, with the most probable location being under the shell, but also able to populate the entire volume of the carboxysome. The locations of the average densities indicate that the average number and arrangement of RuBisCO molecules per layer could be different in different size classes.
To determine the extent to which carboxysomes vary, STEM 20 was used to make mass measurements of individual carboxysomes. The results in Figure 7 show that carboxysomes vary widely between 100 and 350 MDa with the majority populating the 190–280 MDa range. These results along with the cryoimages (Figure 1) and the histogram of their diameters (Figure 3) showing purified carboxysomes of varying sizes indicate that carboxysomes are not uniform in either size or mass.
The carboxysome is a large particle of variable size. It is difficult to determine its structure by conventional single particle cryoEM methods. Cryo-electron tomography is a technique that can give low resolution information but suffers from the problem of missing data due to limited tilt capability in a conventional electron microscope. If the particles are similar enough to average (at least into several groups if not one group) and randomly oriented, the averaging can alleviate this problem of missing data. Our methods for dealing with the missing wedge of data in Fourier space represent an attempt to deal with this problem in tomography, both in determining the relative orientation of extracted particles, each having this missing wedge, and in merging and averaging the 3D data from particles having these characteristics. Previous studies 15; 16 started with a search model that did not itself contain a missing wedge, or used alignment by translational, not rotational, correlation 21. When combined with the serious problem of size heterogeneity and incomplete prior knowledge of the symmetry, this is a difficult specimen to tackle by either single particle or tomographic methods. The methods described here can overcome much of the effect of the missing data, when objects in the tomographic reconstruction having different orientations can be averaged. When this is not possible, it may be necessary to minimize the missing data by other means, such as by dual axis tilting 22.
Our strategy for dealing with the problem of heterogeneous particle population was to undertake multiple steps of classification. The first step allowed us to identify particles that have structural similarity with other particles. This eliminated particles with broken or incomplete shells, or completely empty particles, for example. The second step was to separate the data into particle size groups. The final distribution of size classes had seven groups, with average diameters of from 88 to 103 nm, respectively. The distribution was not equal, and a majority of particles were in size classes that centered about 94–95 and 97–100 nm (Figure 3). The STEM measurements (Figure 7) show a broad distribution in carboxysome mass. This distribution in mass probably reflects the variable fraction of packing of RuBisCO molecules in carboxysomes of any size, in addition to the size variation itself.
Part of the confusion about the symmetry of the carboxysome was that it was not possible to find the orientation of the particles when the traditional “self common line” method 23 was applied (A. Paredes, unpublished). In view of the structural results, there are several possible reasons for this failure. Prominent among these is the apparent lack of icosahedral symmetry in the internal RuBisCO, which comprises a large fraction of the mass. However, viruses, may also have asymmetric internal mass similar to carboxysomes, or other non-icosahedral features, but do not show the same alignment problems using the standard icosahedral single particle reconstruction methods. We believe that the problem lies in the thin, flat triangular faces of the carboxysome shell. The shell of the carboxysome is shaped like a regular icosahedron without prominent “capsomeres”. This means that in Fourier space, the power spectrum is largely confined to directions normal to the faces, and is quite weak in other directions. Central sections in Fourier space that do not contain these normals will have a weak icosahedrally symmetric signal. However, our 3D reconstruction from the tomograms was able to unambiguously indicate the icosahedral shape of the particle shell when properly aligned and averaged. Projection matching 24, which is useful when good starting models are available, should have fewer problems. In fact, a major result of this work is that we now have a series of reliable starting models of various sizes to begin to analyze the carboxysome by traditional single particle methods. Finally, the tomographic reconstruction reveals a tendency for the carboxysomes to have a preferred orientation, with their triangular faces toward the air-water interface. Preferred orientations are a potential source of problems in any single particle reconstruction project. Particles in this orientation tend to have a hexagonal outline in projection, which may have biased the search for other possible symmetry.
It should be noted that the subset of particles that were averaged does not include particles that may have been outliers in terms of their symmetry (non-icosahedral), size or shape, because these did not pass our original test of consistency with enough of the other particles in the tomogram to form their own classes. There are probably individual particles with other shapes and symmetry among the population. Also, these are particles isolated from bacteria, and the content and organization of their RuBisCO, for instance, could be different in vivo.
The hexameric and sheet like structure observed in the crystals of the CcmK proteins 18 was proposed to be relevant to the structure of the shell. It is interesting and perhaps relevant that we observe density features on the surface of the carboxysome shell with spacing similar to that in the crystal (Figure 5a–g). In the crystal structure, one hexameric face has positive charges surrounding the center of the hexamer, forming a pore. 18 However, in our carboxysomes, which contain up to 8 other polypeptides, there are several other low-density features which may serve as “pores”, some of which, near the vertices, appear considerably larger than those proposed from the homogeneous continuous sheets of CcmK proteins in the crystal structures.
Large and small carboxysomes were randomly distributed throughout the tomogram. This observation of variation in size and organization in an icosahedral particle is unusual, but not unprecedented. In mutant P22 25 and genetically engineered herpesvirus capsids 26, a variation in the composition or assembly leads to particles having different triangulation (T) numbers. In carboxysomes, variations in the proportions or arrangement of the 6–8 shell proteins could similarly produce carboxysomes of different sizes.
The different sized carboxysomes may have a different complement of protein components, or different diffusion and permeability characteristics for CO2 (which along with ribulose-1,5-bisphosphate is the productive substrate) and O2 (which is unproductive). Unlike a virus particle, where size variability is not compatible with packaging the required genome for viability, size variation of carboxysomes is non-lethal. In addition, carboxysomes are dynamic, and their appearance and RuBisCO content can change with physiological conditions 11; 12 2; 27; 28; 29. Our concentric layer model of RuBisCO packing in biochemically isolated carboxysomes is the first that can account for both the reported “paracrystalline” packing which can fill the shell with equally spaced layers of molecules, as well as explain other observed arrangements that result in partially filled shells. The structures derived here may be useful in analyzing tomographic reconstructions of intact, live cyanobacteria to characterize the carboxysomes they contain in the native state.
Halothiobacillus neapolitanus carboxysomes were first purified according to established methods 2; 10. The specimen was then applied to 400 mesh copper EM grids, which had been coated with perforated carbon and glow discharged to make them hydrophilic. The perforated carbon consisted of a network of many 3–7 mm holes on the carbon film 30. Once applied to the EM grids, excess buffer was quickly wicked off with filter paper to suspend the carboxysomes in a thin film of buffer in holes within the carbon. The grids were then plunged into liquid ethane in a reservoir that was partially submerged in liquid nitrogen 31; 32. After vitrification, the grids were stored under liquid nitrogen and later placed in the electron cryomicroscope for analysis.
Images were recorded on a Gatan 4k × 4k CCD camera on a JEM2010F electron microscope with an accelerating voltage of 200kV using the JAMES software package 33, which integrates the microscope, camera and database. Sample temperature was held at 102° K with a Gatan cryo-holder. Single-axis tomographic imaging was performed manually at 20,000× magnification from−58° to +70° in 2° increments under low-dose conditions. The total dose was kept under 6000 e−/nm2. See Supplemental Movie 1.
Tomographic reconstructions were carried out using the IMOD package 34. Registration of the tomographic images to each other was accomplished by recording the positions of about 7 ice contamination masses and the centers of three carboxysomes, which could be followed through most of the tilt series. These were used as fiducials in the IMOD software suite to refine the tilt angles, axis and magnification of each image. In contrast to the normal procedures with this software, where gold particles are used, the “fiducials” in this case had to be recorded manually in each image, because automated tracking was not feasible.
Subsequent processing, including alignment, symmetrization, averaging, calculating 1D radial plots from the 3D volumes, etc. of the 3D particle volumes, was done using available EMAN (and EMAN2) routines 24, or by python scripts that make use of EMAN (and EMAN2) procedures to implement the improvements in orientation searching, averaging and merging that are discussed in the text. Visualization was done by 3dmod (part of the IMOD package) and by Chimera 35.
Briefly, carboxysomes in buffer were applied to 2.3 mm titanium grids coated with a thin carbon support film on a layer of holey carbon film. The specimen was freeze dried according to the methods established by the Brookhaven National Lab. The freeze-dried carboxysomes were then placed into the STEM and maintained at −150°C. The specimen was then scanned with a probe from the microscope operating at 40 keV and with a probe size of 0.25 nm. Rasters of 512 × 512 areas of the specimen were recorded. Elastically scattered electrons were digitally recorded using both a large and small angle detector. These images were analyzed and compared to standards to determine mass 20.
This research was supported by the grant (P41RR02250). Dr. Gordon Cannon at the University of Southern Mississippi, Hattiesburg, provided the carboxysomes used for STEM mass measurements; Dr. Martha Simon at the Brookhaven National Laboratory, a NIH supported resource center (NIH 5 P41EB2181) conducted STEM mass measurements of carboxysomes. We also thank Michael Marsh and Drs. Christopher Booth and Steven Ludtke for helpful discussions.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.