Tubulin and FtsZ form a distinct family of GTPases

Tubulin and FtsZ share a common fold of two domains connected by a central helix. Structure-based sequence alignment shows that common residues localize in the nucleotide-binding site and a region that interacts with the nucleotide of the next tubulin subunit in the protofilament, suggesting that tubulin and FtsZ use similar contacts to form filaments. Surfaces that would make lateral interactions between protofilaments or interact with motor proteins are, however, different. The highly conserved nucleotide-binding sites of tubulin and FtsZ clearly differ from those of EF-Tu and other GTPases, while resembling the nucleotide site of glyceraldehyde-3-phosphate dehydrogenase. Thus, tubulin and FtsZ form a distinct family of GTP-hydrolyzing proteins.

Tubulin and microtubule structure

Our knowledge of microtubule structure and its relationship to microtubule function continue to grow. Cryo-electron microscopy has given us new images of the microtubule polymerization and depolymerization processes and of the interaction of these polymers with motor proteins. We now know more about the effect of nucleotide state on the structure and dynamic instability of microtubules. The atomic model of tubulin, very recently obtained by electron crystallography, is bringing new insight into the properties of this protein and its self-assembly into microtubules, and promises to inspire new experimental efforts that should lead us to an understanding of the microtubule system at the molecular level.

Structure of the alpha beta tubulin dimer by electron crystallography

The alphabeta tubulin heterodimer is the structural subunit of microtubules, which are cytoskeletal elements that are essential for intracellular transport and cell division in all eukaryotes. Each tubulin monomer binds a guanine nucleotide, which is nonexchangeable when it is bound in the alpha subunit, or N site, and exchangeable when bound in the beta subunit, or E site. The alpha- and beta-tubulins share 40% amino-acid sequence identity, both exist in several isotype forms, and both undergo a variety of posttranslational modifications. Limited sequence homology has been found with the proteins FtsZ and Misato, which are involved in cell division in bacteria and Drosophila, respectively. Here we present an atomic model of the alphabeta tubulin dimer fitted to a 3.7-A density map obtained by electron crystallography of zinc-induced tubulin sheets. The structures of alpha- and beta-tubulin are basically identical: each monomer is formed by a core of two beta-sheets surrounded by alpha-helices. The monomer structure is very compact, but can be divided into three functional domains: the amino-terminal domain containing the nucleotide-binding region, an intermediate domain containing the Taxol-binding site, and the carboxy-terminal domain, which probably constitutes the binding surface for motor proteins.

ASIC-based event-driven 2D digital electron counter for TEM imaging

A two-dimensional application specific integrated circuit (ASIC) based detector, designed for X-ray protein crystallography, has been tested to determine its suitability as a direct electron detector for TEM imaging in the voltage range of 20-400 keV. Several markedly different properties of this device distinguish it from the charge coupled device (CCD) detectors: (1) the ASIC detector can be used directly under electron bombardment in the voltage range stated above, therefore requiring no scintillator screen; (2) each active pixel of the device is an electron counter and generates digital output independently; (3) the readout of the device is frameless and event driven; (4) the device can be operated at the room temperature and is nearly noise free; and (5) the counting dynamic range of the device is virtually unlimited. It appears that an imaging system based on this type of device would be ideal for low-dose TEM imaging and online diffraction observation and recording, as well as more conventional imaging, providing the many advantages of direct digital readout for almost all applications.

Visualizing the secondary structure of tubulin: three-dimensional map at 4 A

We are in the process of determining the structure of tubulin using electron crystallography of zinc-induced, crystalline sheets. We have now extended the resolution to 4 A, and there are many features in the map that appear to show details of the secondary structure. X-ray crystallographers are well aware of the problems of interpreting maps with such limited resolution, and the additional problem of the missing cone of data inherent in electron crystallography may make interpretation even more difficult. To investigate how reliably these maps can be interpreted, we have calculated density maps of a known structure, actin, under conditions similar to those of the tubulin map. Results of these simulations support the limited interpretations we made previously in the 6.5-A maps and the more extensive interpretations we make here in the 4-A map. Most of the secondary structure of the tubulin dimer can now be identified.

Interpreting a medium-resolution model of tubulin: comparison of zinc-sheet and microtubule structure

We previously used electron crystallography of zinc-induced two-dimensional crystalline sheets of tubulin to construct a medium-resolution three dimensional (3-D) reconstruction (at 6.5 A) of this protein. Here we present an improved model, and extend the interpretation to correlate it to microtubule structure. Secondary sequence predictions and projection density maps of subtilisin-cleaved tubulin provide information on the location of the C-terminal portion, which has been suggested to be involved in the binding of microtubule-associated proteins. The zinc-sheet tubulin model is compared to microtubules in two ways; comparison of electron diffraction from the zinc-sheets to electron diffraction from microtubules, and by docking the zinc-sheet protofilament 3-D model into a helical reconstruction from ice-embedded microtubules. By correlating the zinc-sheet protofilament to a reconstruction of axonemal protofilaments, we assigned polarity to the protofilament in our model. The polarity assignment together with our model for dimer boundaries and the assignment of alpha- and beta-monomers in our reconstruction, provides a microtubule model where the alpha-monomer crowns the plus- (or fast-growing) end of the microtubule and contact is made in the centrosome with gamma-tubulin via the beta-monomer.

Electron-crystallographic refinement of the structure of bacteriorhodopsin

Using electron diffraction data corrected for diffuse scattering together with additional phase information from 30 new images of tilted specimens, an improved experimental density map has been calculated for bacteriorhodopsin. The atomic model has then been rebuilt into this new map with particular attention to the surface loops. All the residues from 7 to 227 as well as ten lipid molecules are now included, although a few amino acid residues in three of the six surface loops, about half of the lipid hydrophobic chains and all of the lipid head groups are disordered. The model has then been refined against the experimental diffraction amplitudes to an R-factor of 28% at 3.5 angstrom resolution with strict geometry (0.005 angstrom) bond length deviation) using the improvement of the "free" phase residual between calculated and experimental phases from images as an objective criterion of accuracy. For the refinement some new programs were developed to restrain the number of parameters, to be compatible with the limited resolution of our data. In the final refined model of the protein (2BRD), compared with earlier co-ordinates (1BRD), helix D has been moved towards the cytoplasm by almost 4 angstrom, and the overall accuracy of the co-ordinates of residues in the other six helices has been improved. As a result the positions of nearly all the important residues in bacteriorhodopsin are now well determined. In particular, the buried, protonated Asp115 is 7 angstrom from, and so not in contact with, the retinal and Met118 forms a cap on the pocket occupied by the beta-ionone ring. No clear density exists for the side-chain of Arg82, which forms a central part of the extracellular half-channel. The only arginine side-chain built into good density is that of Arg134 at the extracellular end of helix E, the others being disordered near one of the two surfaces. The interpretation of the end of helix F on the extracellular surface is now clearer; an extra loose helical turn has been built bringing the side-chain of Glu194 close to Arg134 to form a probable salt bridge. The model provides an improved framework for understanding the mechanism of the light-driven proton pumping. A number of cavities that could contain water molecules were found by searching the refined model, most of them above or below the Schiff base in the half-channels leading to the two surfaces. The ordered and disordered regions of the structure are described by the temperature factor distribution.

The relevance of dose-fractionation in tomography of radiation-sensitive specimens

It is commonly assumed that the number of projections required for single-axis tomography precludes its application to most beam-labile specimens. However, Hegerl and Hoppe have pointed out that the total dose required to achieve statistical significance for each voxel of a computed 3D reconstruction is the same as that required to obtain a single 2D image of that isolated voxel, at the same level of statistical significance. Thus a statistically significant 3D image can be computed from statistically insignificant projections, as long as the total dose that is distributed among these projections is high enough that it would have resulted in a statistically significant projection, if applied to only one image. We have tested this critical theorem by simulating the tomographic reconstruction of a realistic 3D model created from an electron micrograph. The simulations verify the basic conclusions of the theorem and extend its validity to the experimentally more realistic conditions of high absorption, signal-dependent noise, varying specimen contrast and missing angular range. Individual projections in the series of fractionated-dose images could be aligned by cross-correlation because they contained significant information derived from the summation of features from different depths in the structure. This latter information is generally not useful for structural interpretation prior to 3D reconstruction, owing to the complexity of most specimens investigated by single-axis tomography. These results demonstrate that it is feasible to use single-axis tomography with soft X-ray and electron microscopy of frozen-hydrated specimens.

Crystallographic extraction and averaging of data from small image areas

The accuracy of structure factor phases determined from electron microscope images is determined mainly by the level of statistical significance, which is limited by the low level of allowed electron exposure and by the number of identical unit cells that can be averaged. It is shown here that Fourier transforms of small image fields of purple membrane (a two-dimensional crystal consisting of bacteriorhodopsin and endogenous lipids) can be combined to provide the same quality of phases as are obtained from Fourier transforms of large image fields of the same total area. Although Fourier transforms of such small image fields are statistically significant only at lower resolution, the data from many such image fields can be averaged at the calculated positions of high-resolution reciprocal lattice points to give accurate phases. More specifically, when images of a size that can be recorded with CCD cameras are processed individually, key parameters including lattice vectors, defocus, crystal and beam tilts, and common phase origin can be accurately determined.

Preservation of 2-D crystals of tubulin for electron crystallography

Zinc-induced sheets of tubulin are two-dimensional crystalline polymers that constitute an ideal sample for high resolution studies of tubulin by electron crystallography. We show that these 2-dimensional tubulin crystals can be stabilized by taxol against low-temperature depolymerization and degradation with time, easing the way for the preparation of electron microscopy samples. The preservation of the crystals to high resolution has been tested with different embedding media. While glucose-embedded samples diffract poorly, samples embedded in tannin consistently diffract to a resolution of at least 3.5 A. Even better results are obtained by embedding with a combination of tannin and glucose, which improves the flatness of the crystals and allows the collection of isotropic high-resolution data from tilted specimens.

Structure of tubulin at 6.5 A and location of the taxol-binding site

Tubulin, the major component of microtubules, is a heterodimer of two chains, alpha and beta, both of relative molecular mass 50,000 (Mr50K) and with 40-50% identity. The isotypic variety and conformational flexibility of tubulin have so far made it impossible to obtain crystals for X-ray work. Structural knowledge of tubulin has been limited to about 20 A from X-ray diffraction of oriented microtubules, and from electron microscopy of microtubules and zinc-induced crystalline sheets in negative stain. The sheets consist of protofilaments similar to those in microtubules but associated in an antiparallel arrangement, and their two-dimensional character is ideal for high-resolution electron microscopy. Here we present a three-dimensional reconstruction of tubulin to 6.5 A resolution, obtained by electron crystallography of zinc-induced two-dimensional crystals of the protein. The alpha- and beta-subunits appear topologically similar, in agreement with their sequence homology. Several features can be defined in terms of secondary structure. An apparent alpha-helical portion, adjacent to both interdimer and inter-protofilament contacts, is tentatively attributed to a segment near the carboxy terminus of the protein. We can assign the alpha- and beta-subunits on the basis of projection studies of the binding of taxol, which show one taxol site per tubulin heterodimer, in agreement with the known stoichiometry of taxol in microtubules. These studies indicate that taxol affects the interaction between protofilaments; to our knowledge, this is the first time that a ligand-binding site has been visualized in the tubulin molecule.

A constant radius of curvature model for the organization of DNA in toroidal condensates

Toroidal DNA condensates have received considerable attention for their possible relationship to the packaging of DNA in viruses and in general as a model of ordered DNA condensation. A spool-like model has primarily been supported for DNA organization within toroids. However, our observations suggest that the actual organization may be considerably different. We present an alternate model in which DNA for a given toroid is organized within a series of equally sized contiguous loops that precess about the toroid axis. A related model for the toroid formation process is also presented. This kinetic model predicts a distribution of toroid sizes for DNA condensed from solution that is in good agreement with experimental data.

Kinesin does not support the motility of zinc-macrotubes

Moving along a microtubule, kinesin follows a course parallel to the protofilaments; but it is not known whether kinesin binds exclusively on a single protofilament. The presence of zinc during tubulin polymerization induces sheets where neighboring protofilaments are antiparallel. If kinesin could support the motility of these zinc-sheets, then the binding site for a kinesin molecule would be limited to a single protofilament. Kamimura and Mandelkow [1992: J. Cell Biol. 118:865-75] reported that kinesin moves along zinc-sheets. We found that zinc-sheets grown under their conditions often had a microtubule-like structure along one edge. We confirmed the possibility that the motility observed by Kamimura and Mandelkow [1992: J. Cell Biol. 118:865-75] is attributed to the microtubule-like structure rather than the zinc-sheet. To resolve the question of whether kinesin can recognize an antiparallel protofilament lattice, we investigated the kinesin-mediated motility of zinc-macrotubes. At higher free zinc concentrations, zinc-sheets roll up as macrotubes, free of edges. In the presence of 10 microM taxol and 100 nM free Zn2+ at pH 6.8, the samples were shown by electron microscopy to contain only macrotubes. Under these buffer conditions, kinesin could bind strongly to axonemal doublets in the presence of AMP-PNP, and generate motility in the presence of ATP, but kinesin did not bind to nor move the macrotubes. This shows that kinesin cannot bind efficiently to nor move on the anti-parallel lattice; it is possible (though not necessary) that the groove between two parallel protofilaments is required for kinesin's motility.

Studies of the structural organization of a bacterial chemoreceptor by electron microscopy

We used analysis by electron microscopy to obtain structural information about a transmembrane receptor that mediates chemotaxis in Escherichia coli. Two-dimensional arrays of regularly packed particles of the receptor Trg were obtained by reconstitution of purified, detergent-solubilized protein into lipid bilayers. Preliminary image processing of negatively stained arrays revealed an almost square 8.8 x 8.8-nm unit cell and resolved the particles into four peaks of density around a central depression. In certain conditions, reconstituted, Trg-containing bilayers associated into membrane stacks. The regular spacing of the stacks provided a value of 15 nm for the dimension of the receptor normal to the membrane. Using these dimensions, the estimated occupied volume of the structure would be sufficient to contain four monomers of Trg. This tetramer form may be a dimer of two antiparallel or parallel homodimers. Our analysis indicates that a receptor monomer is approximately 4.4 nm at the widest point and 15 nm long. Given the dimensions of the periplasmic domain of the closely related receptor Tars, determined by X-ray crystallography, and a minimum bilayer thickness of 3 nm, the cytoplasmic domain would be approximately 5.0 by 4.4 nm. Higher resolution analysis should reveal additional information about receptor structure.

High-resolution electron crystallography of protein molecules

Electron diffraction data and high-resolution images can now be used to obtain accurate, three-dimensional density maps of biological macromolecules. These density maps can be interpreted by building an atomic-resolution model of the structure into the experimental density. The Cowley-Moodie formalism of dynamical diffraction theory has been used to validate the use of kinematic diffraction theory (strictly, the weak phase object approximation) in producing such 3D density maps. Further improvements in the preparation of very flat (planar) specimens and in the retention of diffraction to a resolution of 0.2 nm or better could result in electron crystallography becoming as important a technique as X-ray crystallography currently is for the field of structural molecular biology.

Tubulin conformation in zinc-induced sheets and macrotubes

The protein tubulin is the main constituent of microtubules. Previous studies have shown that zinc ions induce the formation of crystalline sheets and macrotubes of tubulin. Both crystal types are suitable for structural studies by electron crystallography. However, crystallographic structural analysis of tubulin has been hampered by limited crystal size and quality and the inability to control crystal polymorphism. We can obtain well-ordered crystals which are grown upon prolonged incubations (up to 24 hr). The presence of NaCl delays the degradation of the crystals, and addition of the protease inhibitor pepstatin improves crystal quality. The crystal form (sheet or macrotube) can be controlled with incubation conditions. The size of the crystals can reach up to 2 microns in width for the sheets and up to 0.5 microns in diameter for the macrotubes. Both crystal types can reach several micrometers in length. Comparison of the projection maps of the two crystal structures shows that adjacent protofilaments in the macrotubes are shifted by about 6 A relative to their positions in the sheets. Observable changes of monomer shape appear to allow close interprotofilament contacts to be maintained in both crystal forms. Images of glucose-embedded specimens obtained under these conditions give structural information beyond 4 A resolution. Merging of high- and low-resolution data allows for unambiguous assignment of monomer boundaries to high-resolution features.

Identification of the elemental packing unit of DNA in mammalian sperm cells by atomic force microscopy

DNA is packaged within the sperm cell nuclei of many vertebrates and all mammals in a highly condensed state by small basic proteins (protamines). Despite continuous investigation for nearly half a century, the actual packing arrangement of the DNA has remained unresolved. Atomic force and electron microscopy studies described in this report provide evidence that the fundamental packing unit for sperm DNA is a toroidal structure, 900A in outside diameter with a 150A diameter hole, which contains up to 60kb of DNA. Although the results presented here are based primarily upon investigations of mammalian sperm cells, they are expected to be valid for all sperm cells which utilize protamines to package DNA.

Cytoskeletal sheets of mammalian eggs and embryos: a lattice-like network of intermediate filaments

Mammalian eggs and embryos possess a major cytoskeletal network composed of large planar "sheets" distributed throughout the cytoplasm. Cytoskeletal sheets are found neither in mammalian somatic cells nor in eggs or embryos of non-mammals. In this study, we have investigated the structural composition of the sheets in eggs and embryos of the golden Syrian hamster by (1) analysis of replicas from quick-frozen, deep-etched specimens, (2) analysis of thick, resin-embedded specimens using an intermediate voltage electron microscope (IVEM), (3) laser diffraction of EM images, (4) differential extraction with detergents, and (5) immunocytochemistry. Our results indicate that each sheet is composed of two closely apposed arrays of 10-nm filaments. Each filament within an array is held in register with its neighbor by lateral cross-bridges and the two parallel arrays of filaments are interconnected by periodic cross-bridges about 20 nm in length. Laser diffraction of negatives from IVEM images indicates that each array is composed of fibers that form a square lattice, and the two arrays are positioned in register by cross-bridges forming a single sheet. This lattice forms the skeleton of the sheets which is covered with a tightly packed layer of particulate material. By incubation in media containing different ratios of mixed-micelle detergents, it is possible to remove components sequentially from the sheets and to extract the particulate material. Immunocytochemical localization demonstrates that the sheets bind antibodies to keratin, and to a small extent actin, but do not bind antibodies to vimentin or tubulin. Examination of sheets within embryos at the time of embryonic compaction demonstrates that the sheets begin to fragment and disassemble in regions of blastomeres where desmosomes form, but undergo no structural alterations in interior and basal surfaces of the blastomeres. In regions of blastomere-blastomere contact the sheets fragment and associate with granules resembling keratohyalin granules found in keratinocytes.

Assessment of resolution in biological electron crystallography

The resolution of images or density maps produced by electron microscopy and electron crystallography can be objectively defined in terms of the spatial frequency of the highest resolution diffraction spot, or Fourier coefficient, included in the data processing. In practice, this objective definition of resolution is expected to be too optimistic if the amplitudes of the highest resolution structure factors are too weak, if the population of high resolution reflections is too sparse, or if the signal-to-noise ratio of the high resolution data is too low. Calculated examples are presented here which illustrate how the apparent resolution in images of a membrane protein, bacteriorhodopsin, can be reduced from a nominal value of 3.5 A by weak amplitudes, sparse data or high noise levels. These calculations provide concrete examples which can serve as a guide when estimating whether the objective definition of image resolution is likely to correspond to a practical, structurally useful estimate of image resolution.

Automatic focus correction for spot-scan imaging of tilted specimens

The variation in defocus within an image of a highly tilted specimen can be a serious source of artifact. Spot-scan imaging can be combined with dynamic focusing to greatly reduce this range of defocus. A protocol is described for determining the parameters required for the automatic focus compensation during the recording of a spot-scan image. Images of a gold test specimen demonstrate the efficacy of this procedure in extending the area of the image that contains high-quality data. In case the tilt angle or resolution is high enough that the height difference of the specimen within each small illuminated area is larger than the depth of field, the image must be treated to compensate for the focus variation. The same principle is used as was developed for compensation of conventional images of tilted specimens.

Projection map of tubulin in zinc-induced sheets at 4 A resolution

Tubulin polymerizes into two-dimensional, crystalline sheets in the presence of zinc ions. These sheets are well suited to structural studies by electron crystallography. We have developed conditions for forming sheets which are large and well ordered enough to provide both electron diffraction and image data to better than 4 A resolution. In projection maps calculated from this data, the alpha and beta monomers can be identified within the protofilaments. These results indicate that we should be able to determine the structure of tubulin in these sheets at atomic resolution.

3D structure determination from electron-microscope images: electron crystallography of staurolite

Resolution of better than 2 A has been obtained in many crystals by high-resolution electron microscopy. Although this resolution is sufficient to resolve interatomic spacings, structures are traditionally interpreted by comparing experimental images with contrast calculations. A drawback of this method is that images are 2D projections in which information is invariably obscured by overlap of atoms. 3D electron crystallography, developed by biophysicists to study proteins, has been used to investigate the crystal structure of staurolite. Amplitudes and phases of structure factors are obtained experimentally from high-resolution images (JEOL ARM 1000 at the National Center for Electron Microscopy at LBL), taken in different directions from thin regions where dynamic scattering is minimal. From images in five orientations (containing 59 independent reflections to a resolution of 1.38 A), a 3D electron potential map is constructed which resolves clearly all cations (Al, Si, Fe, including those with partial occupancy) and all O atoms. This method has great potential in crystal structure determinations of small domains in heterogeneous crystals which are inaccessible to X-ray analysis. It is estimated that 3D structure determinations should be possible on regions only about ten unit cells wide and should resolve not only atom positions but also site occupancies. The method is also applicable to space-group determination.

Three-dimensional crystallographic reconstruction for atomic resolution

Three-dimensional structures have recently been determined by electron crystallography at a resolution high enough to determine atomic arrangements in both protein and mineral specimens. The different nature of these two types of specimens produces some very significant differences in the way data is obtained and processed, although the principles are the same. The sensitivity of proteins to damage by the electron beam limits the signal-to-noise ratio in the image and the resolution to which data can be extracted from the image. A number of constraints, such as the amino acid sequence and the connectivity of atoms within amino acids, can be used in interpreting the limited image data. In materials samples, the relative insensitivity to damage allows obtaining resolution limited only by the microscope. In many samples, dynamical scattering and other non-linear effects limit the information in the image, but this limit can be circumvented by working in very thin areas of the specimen.

Three-dimensional structure of an invertebrate intercellular communicating junction

Gap junctions containing extensive, highly ordered crystalline arrays of hexagonally packed connexons have been isolated from the hepatopancreas of the arthropod, Homarus americanus (American lobster). The structure of such junctions has been studied to a resolution of approximately 25 A in three dimensions by electron microscopy of negatively stained specimens. The structure, which has the crystallographic symmetry of the two-sided plane group p6, reveals the connexon as an annular oligomer which projects approximately 30-45 A from the cytoplasmic surface. The stain-filled channel structure appears to be approximately 40-45 A wide in the extracellular region. Projection images of glucose-embedded specimens extend to a resolution of 10 A, and show a strong contrast from the connexon subunits. Overall the structure is quite similar to that of rat liver junctions, except that less stain is seen in the aqueous region of the gap and more surrounding the protrusions of the protein into the cytoplasm.

What spectroscopy can still tell us about the secondary structure of bacteriorhodopsin

The recently published model of the structure of bacteriorhodopsin (bR), developed by fitting the peptide chain to a high-resolution, three-dimensional density map, rules out the existence of transmembrane beta-sheet and provides an accurate estimate of the helix content. The precise geometry of the dihedral angles in the helical regions of the polypeptide cannot yet be specified from the diffraction data, however. Published data on the circular dichroism (CD) spectrum between 190 and 240 nm, and the infrared (IR) spectrum in the amide I band suggest that the helical conformation in bR may be, for the most part, a rather unusual one. The precise structural model, which specifies the number of residues in transmembrane helices, can now be used as an additional constraint in seeking models of the helical conformation that are in quantitative agreement with the CD and IR spectroscopic data. Further spectroscopic measurements can also be used to determine whether there are changes in the unusual dihedral-angle conformation within the helices during the photocycle.