Tricorn Protease Exists as an Icosahedral Supermolecule In Vivo

Molecular Cell, Vol. 1, 59–65, December, 1997, Copyright 1997 by Cell Press

Tricorn Protease Exists as an Icosahedral Supermolecule In Vivo Jochen Walz, Tomohiro Tamura, Noriko Tamura, Rudo Grimm, Wolfgang Baumeister,* and Abraham J. Koster Abteilung Molekulare Strukturbiologie Max-Planck-Institut fu¨r Biochemie 82152 Martinsried Germany

Summary Tricorn protease is the core enzyme of a recently discovered modular proteolytic system. We present evidence that tricorn protease exists in vivo in the form of a higher-order assembly, namely as an icosahedral capsid. Its size exceeds that of many virus particles and represents by far the largest known homooligomeric enzyme complex. Each capsid is built from 20 copies of the tricorn hexameric toroid and thus has a molecular weight of 14.6 MDa. Three-dimensional reconstructions of ice-embedded capsids from electron micrographs show that it is hollow and has large void volumes in its wall. We suggest that the tricorn capsid, in addition to its intrinsic proteolytic activity, serves as the organizing center of a multienzyme complex. Introduction Protein degradation is essential for the maintenance of homeostasis and for the removal of abnormal (i.e., misfolded) proteins. Moreover, it is a key element of many regulatory processes in the cell. For obvious reasons, proteolysis must be subject to spatial and temporal control to avoid the destruction of proteins not destined for degradation. The compartmentalization of proteolytic activity offers a means to accomplish this. Two basic forms of compartmentalization have evolved: confinement of proteolytic activity in a membrane-bound compartment, such as the lysosome, and confinement to the interior of large multisubunit proteolytic complexes, a principle we will further refer to as self- or autocompartmentalization. In recent years, the proteasome has become the paradigm of a macromolecular assembly designed to perform its proteolytic function in an inner chamber, thus sequestering it from the crowded environment of the cell (for recent reviews, see Baumeister and Lupas, 1997; Baumeister et al., 1997). Substrate proteins have to pass through a system of narrow channels and antechambers before reaching the “nanocompartment” where the active sites are located; therefore, access is restricted to unfolded substrates only (Lo¨we et al., 1995; Wenzel and Baumeister, 1995). The 20S proteasome is the proteolytic core structure of a larger complex, the 26S proteasome. The regulatory subunits in the 19S cap complexes that associate with the two ends of the 20S core confer * To whom correspondence should be addressed.

the ability to recognize substrate molecules targeted for degradation, to unfold them and, eventually, to translocate them to the proteolytic core (Peters et al., 1993). The occurrence of proteasomes in all three urkingdoms, archaea, bacteria, and eukarya, bears testimony to an old evolutionary principle. Although the stratagem of autocompartmentalization is of particular importance in prokaryotes lacking membrane-bound compartments, it was conserved during evolution, allowing eukaryotic cells to degrade target molecules in the cytosol or nucleus in a controlled manner. In conjunction with the ubiquitin system, which contributes the degradation signals, the proteasome has assumed a wide spectrum of essential cellular functions ranging from cell cycle control to the generation of immunocompetent peptides (for a recent review, see Coux et al., 1996). Recently, we have discovered another large proteolytic complex unrelated to the proteasome in the archaeon Thermoplasma acidophilum (Tamura et al., 1996). In view of its peculiar triangular structure, it was given the name tricorn protease. Its toroidal structure is indicative of a mechanism of action similar to the proteasome. The hexameric 730 kDa complex appears to represent the core structure of a large modular proteolytic system. Hitherto, two factors have been isolated and characterized that upon interacting with the tricorn protease either cause a stimulation of intrinsic proteolytic activities or elicit new ones (Tamura et al., 1996). We noted previously (Tamura et al., 1996) that activities ascribed to the tricorn protease are found in two fractions, the 730 kDa fraction containing the hexameric complex and the void volume where large capsid-like structures were found. In this communication, we present evidence that these capsids represent an in vivo form of the tricorn protease and that they are perfectly symmetric icosahedrae. Thus, tricorn protease forms a supermolecule of 14.6 MDa, by far the largest protease found to date.

Results Isolation of Tricorn Protease Tricorn protease from T. acidophilum was isolated and purified through a sequence of chromatography steps (see Experimental Procedures). The isolated protein was concentrated to approximately 2 mg/ml and subjected to Superose 6 molecular sieve chromatography (Figure 1A). Tricorn protease fractionated into the void volume and into a 730 kDa fraction. Both fractions yielded a single 122 kDa band upon SDS–PAGE. Using a synthetic fluorogenic peptide as a substrate, H-AAF-AMC, both fractions showed the same level of peptidase activity (Figure 1A). When examined by electron microscopy, the 730 kDa fraction contained the hexameric form of tricorn protease described previously (Tamura et al., 1996), while the void volume fraction contained predominantly capsid-like structures approximately 55 nm in diameter (Figure 1B).

Molecular Cell 60

Figure 1. Elution Profile of Tricorn Protease and Electron Micrograph of the Void Volume Fraction after Superose 6 Chromatography (A) A sample containing 300 mg of isolated tricorn protease was fractionated by Superose 6 molecular sieve chromatography. Protein was detected by absorbance at 280 nm. Samples (10 ml each) from several fractions were subjected to SDS–PAGE (top panel). Bands migrating to the same positions were detected in the void volume fraction and in the 730 kDa fraction, which corresponds to the tricorn hexamer. Tricorn peptidase activity was detected by incubating 5 ml of each fraction with 10 nmol of H-AAF-AMC at 608C for 30 min and measuring released AMC fluorometrically. (B) The void volume fractions were subjected to electron microscopy after negative staining with uranyl acetate. The prominent spherical structures, 55 nm in diameter, represent the tricorn capsids. Because of the fragile nature of the capsids, they tend to disassemble into hexameric units seen in the background. Scale bar in (B), 200 nm.

Tomographic Reconstruction of Vitrified Tricorn Capsids In analyzing the molecular architecture of isolated and purified tricorn capsids, we have taken a novel approach. Using frozen-hydrated samples, we have first performed a tomographic reconstruction based on single-axis tilt series. Automated data collection procedures allowed us to keep the cumulative electron dose within reasonable limits (less than 40 e 2/A˚ 2). The outcome of the tomographic reconstruction provided first insights into the structural organization of the capsids and served as a starting model for a higher-resolution reconstruction from single projections now exploiting the icosahedral symmetry of the capsids. Figure 2 shows a set of unprocessed electron micrographs of the frozen-hydrated capsids. Figures 2A and 2B illustrate the effect of different focus settings: Figure 2A is taken close to focus (z0.5 mm underfocus) and contains higher-resolution information at the expense

of low contrast. Figure 2B, which covers the same specimen area, was recorded further underfocus (z5 mm) and shows a much enhanced contrast sacrificing the high spatial frequency information. Figure 2C shows the zero tilt projection from a tomographic tilt series. Although this image was recorded at an underfocus of z7 mm, the two capsids contained in this section of the ice film are barely visible because the electron dose for each projection of the tilt series must be kept rather low (1.5–2 e 2/A˚2 ) to avoid the accumulation of too high a dose throughout the tilt series, which typically comprised 23 projections. At an underfocus of z7 mm, the first zero crossing of the contrast transfer function is at z(4.8 nm)21, limiting the range of useful information accordingly. The tomographic reconstruction reveals the icosahedral structure of the capsids. Figure 3 shows results of the tomographic reconstruction. Figure 3A is a gallery of slices, 1.26 nm thick, through a single tricorn capsid. In spite of the low signal-to-noise ratio, it is obvious that the capsid is a hollow structure without any significant density inside. It is important to note that the tomographic reconstruction, unlike other single particle techniques relying on averaging, would reveal density or mass inside the capsid even when not strictly localized with respect to the capsid structure. Also, on some of the sections a five-fold symmetry is discernible. Seven such individual tricorn capsids were reconstructed and, following 3-D alignment, were averaged. The averaged tomographic reconstruction has a much improved signal-to-noise-ratio, and the icosahedral organization of the capsids becomes obvious. The surface representation in Figure 3B is viewed down the five-fold symmetry axis. From the position of the five-fold axis, the threefold and two-fold axes can be exploited for further averaging. Figures 3C–3E show the symmetrized capsids viewed down the three-fold, five-fold, and two-fold axes, respectively. The tomographic reconstruction contains a remarkable amount of structural information, considering that not more than seven particles were reconstructed and averaged. The resolution of this reconstruction according to the Fourier ring correlation criterion is (4.7 nm)21 and thus coincides with the first zero crossing of the contrast transfer function (Figure 4). Each capsid is built from 20 tricorn hexameric toroids. The molecular weight of each hexamer is 730 kDa; thus, the molecular weight of the icosahedral capsid is 14.6 MDa. The outer diameter of the capsid is 55 6 2 nm; the inner diameter 37 6 2 nm. The capsids are rather Figure 2. Montage of Electron Micrographs of Frozen Hydrated Tricorn Capsids (A) was recorded close to focus (0.5 mm underfocus) and contains higher resolution information, with lower contrast, compared to (B), which was recorded at an underfocus of z5 mm. (C) shows the zero tilt view of a tomographic tilt series containing two capsid structures (z7 mm underfocus). The micrograph shown in (C) is exposed with a 10-fold lower electron dose than (B) (1.5–2 e2/A˚2 compared to 15–20 e2/A˚2). The four small, dark, circular features in the image are gold beads used as fiducial markers in the tomographic reconstruction. Scale bars in (A)–(C), 50 nm.

Tricorn Protease—An Icosahedral Supermolecule 61

Figure 3. Reconstruction of the Tricorn Capsid from Tomographic Tilt Series (A) A gallery of 1.26 nm thick slices through a single reconstructed tricorn capsid. The tricorn capsid particle is hollow and does not contain any discernible features inside. In spite of the low signal-to-noise ratio in the individual 2-D slices, an indication of five-fold symmetry, most apparent in the third row from the top, can be observed. (B) Surface representation of the tricorn capsid after averaging seven individually reconstructed particles were averaged but without imposing icosahedral symmetry. The view is down the five-fold symmetry axis. (C–E) Views of the reconstructed capsid after icosahedral symmetrization down the threefold axis, five-fold axis, and two-fold axis, respectively. Scale bars in (A)–(E), 50 nm.

open structures with large, empty spaces in between the hexameric toroids. Only 60% of the surface is occupied by protein. The areas of contact between the hexameric units in the capsid are not very extensive, suggesting that the interactions are weak. 3-D Reconstruction Exploiting the Icosahedral Symmetry of the Capsid A higher-resolution structure of the tricorn capsids was obtained by using the averaged tomographic reconstruction as a starting model for an icosahedral reconstruction from projection data. The steps involved in the reconstruction are summarized in Figure 5. The reconstruction was based on 38 projections recorded at an underfocus of 1.4–2.1 mm. Thus, there is continuous transfer to at least (2.6 nm)21. Although the reconstruction procedure has accounted for contrast reversals beyond the first zero of the contrast transfer function, no

noticeable improvement of the reconstruction was observed. The resolution of the reconstruction according to the Fourier ring correlation criterion is (2.5 nm)21. The higher resolution reconstruction (Figure 6A) gives more detailed insights particularly into the structure of the individual hexameric toroids, which clearly show 32-point symmetry. The overall shape is that of a triangle with flattened vertices. The width measured across the flats is 19 6 1 nm, and the height is 9 6 0.5 nm. Figure 6B shows an end-on view of a tricorn protease cut open horizontally (top) and vertically (bottom). These sections clearly show the large cavity inside the hexamer; it measures 10 6 1 nm across, and the maximum height is 4.3 6 0.5 nm. Access to this cavity is via an approximately 2.6 nm wide channel traversing the toroid centrally, i.e., along the three-fold axis. The Occurrence of Tricorn Capsids in Thermoplasma Cells Although it may be seen as a remote possibility that the tricorn capsid represents an artificial assembly product

Figure 4. Resolution Assessment Resolution assessment of the tomographic reconstruction (dotted line) and the icosahedral reconstruction (solid line) using the Fourier shell correlation function. The 23.2 s curve (dashed line) is used as criterion for the resolution limit.

Figure 5. Flow Diagram of Image Processing Flow chart of the basic steps involved the 3-D map reconstruction of the capsid shown in Figure 6.

Molecular Cell 62

Figure 6. Reconstruction of the Tricorn Capsid from 2-D Projections (A) A surface representation of the tricorn capsid at 2.5 nm resolution. (B) Top: end-on view of a single tricorn hexameric toroid taken out of the capsid and cut open horizontally; bottom: hexameric toroid cut open vertically. The large internal cavity is clearly visible. Scale bar in (A), 50 nm; in (B), 15 nm.

in view of the very specific contacts required for such an assembly, we have performed experiments aimed at obtaining direct evidence for the in vivo existence of the icosahedral capsids. To this end, we have examined frozen hydrated T. acidophilum cells by electron cryomicroscopy. Since the dimensions of whole cells (diameter approximately 1 mm) do not allow one to obtain images of a quality sufficient to render macromolecular structures visible, we subjected the cells to controlled lysis prior to cryofixation. The time that elapsed between the initiation of lysis and freezing was no longer than 20 sec. Images of lysed cells are shown in Figure 7. The cellular contents leak out of the T. acidophilum cells at one or a few distinct holes, and different degrees of lysis are observed. The lower the residual concentration of intracellular material, the more distinct macromolecular structures become discernible. Large structures appear to escape from the cells more slowly than smaller ones. Most of the structures . 20 nm in size appeared roundish in projection images. We performed several low-resolution tomographic reconstructions of partially lysed cells that verified that these structures were indeed spherical. So far we have failed, however, to reveal a distinct substructure in these spherical structures, which would allow us to correlate them unambiguosly with the icosahedral tricorn capsids. A statistical size analysis based on multivariate statistical analysis of a

data set comprising 550 of these particles led to segregation into several size classes. Spheres 30–40 nm in size were the most abundant, but the size class 50–60 nm, into which the tricorn capsids fall, was also well represented. Therefore, the spheres obtained in the partially lysed cell might indeed be related to tricorn capsids. Discussion Although there is not yet formal proof, the structure of the hexameric toroid with its large inner cavity appears to be quite apt to function as a sequestration device analogous to the 20S proteasome (Lo¨we et al., 1995) or the Gal6/bleomycin hydrolase (Joshua-Tor et al., 1995). This form of autocompartmentalization, confining the proteolytic action to a nanocompartment with restricted accessibility, begins to emerge as a principle common to many evolutionarily conserved proteolytic systems. Autocompartmentalization provides a simple means to keep nonspecific degradation systems residing amid the crowded environment of the cytosol under control. The volume of the inner cavity of the tricorn toroid is rather large: approximately 340 nm 3, which is about 6.4 times the volume of the central chamber of the proteasome. The apparent size of the opening that gives access to the cavity is 2.5 nm; it is quite possible, however, that at higher resolutions constrictions become visible Figure 7. Electron Cryomicroscopy of Partially Lysed T. acidophilum Cells (A) Electron micrograph of a lysed T. acidophilum cell. The opening of the cell and a stream of cellular contents leaking out are visible. (B) Image of a similar “ghost cell” at higher magnification. The arrows point to several dense structures in a size range corresponding to tricorn capsids. Scale bar in (A), 200 nm; in (B), 100 nm.

Tricorn Protease—An Icosahedral Supermolecule 63

that limit the accessibility further. In any case, with 2.5 nm representing an upper limit, it is clear that most protein substrates would need to undergo some degree of unfolding before entering the central cavity. The most outstanding feature of the tricorn protease is its ability to assemble into a giant icosahedral capsid structure the size of a virus particle. In fact, a tricorn capsid could easily accommodate a whole ribosome. It is rather unlikely that these capsids, which require very specific contacts between the hexamers, represent artificial assembly products, forming de novo after isolation and purification of the tricorn protease. This notion is supported by the direct observation of capsid structures in the same size range within cells that had undergone partial lysis immediately prior to cryofixation. The fact that these structures appear rather indistinct and featureless in the projection images and in low resolution 3-D reconstructions performed with cell mounts may indicate that the large void volumes observed in the isolated capsid walls are occupied by interacting proteins in vivo. Obviously, the question arises, what the functional significance of this quarternary structure may be. When assayed with small fluorogenic peptides, such as H-AAFAMC, the specific activity of individual tricorn hexamers and the capsid form is about the same. Thus, the assembly into the higher-order structure is not a prerequisite for bringing about the basal peptidolytic activity. Hitherto, there are rather few reported examples of enzymes organized in an icosahedral form, and none of them has nearly the size of the tricorn capsid. It should be noted, however, that structures of this size may escape attention because they tend to fractionate into the void volume. Also, the low intrinsic stability of large supramolecular systems makes them difficult to isolate without disintegration. One example of an enzyme with an icosahedral structure is the lumazine synthase/riboflavin synthase complex of bacterial origin (Ladenstein et al., 1988). The 1 MDa complex from Bacillus subtilis is built from 60 b subunits that encapsulate 3 a subunits. The b subunits catalyze the formation of an intermediate (6,7-dimethyl-8-ribityllumazine) that is converted to riboflavin by the a subunits. The compartmentalization of the a subunits inside the b capsid appears to favor substrate channeling (Kis and Bacher, 1995; Mortl et al., 1996). Another example is the pyruvate dehydrogenase multienzyme complex, which catalyzes a key reaction in the aerobic energy-generating glucose metabolism. The complex contains several different enzymes: pyruvate dehydrogenase (E1), dihydrolipoamide acyltransferase (E2), and dihydrolipoamide dehydrogenase (E3). The acyltransferase forms the core of the complex, which in some organisms is an octahedron, in others an icosahedron (Mattevi et al., 1992; Stoops et al., 1992). The icosahedral core structure of the enzyme from Saccharomyces cerevisiae is built from 60 subunits and has a molecular weight of 2.7 MDa; the diameter of the capsids is 22.6 nm and, similar to the tricorn capsids, a rather open structure. It can be seen as the organizing center of a complex catalyzing a multistep reaction. A similar organizing or scaffolding role can be envisaged for the tricorn capsids, the more since tricorn appears to be the core of a modular proteolytic system (Tamura et al., 1996). So far, two factors, both aminopeptidases, have been isolated that interact with tricorn,

enhancing intrinsic activities and eliciting new ones not found in one or the other component (Tamura et al., 1996; Tamura et al., unpublished data). Thus, the functional abilities of the associated components exceed the sum of their parts by means of an as yet unknown mechanism. The interaction with the tricorn core structure might involve allosteric effects or, similar to the lumazine synthase/riboflavin system, might mediate the channeling of reaction intermediates in sequential reactions. In view of the structure of the tricorn capsid described in this communication, it is tempting to speculate that the tricorn-interacting factors occupy the large void volumes in the icosahedral shell in between the tricorn hexamers. If most of the void volumes were filled, the molecular weight of such a multienzyme capsid would be close to 25 MDa, much larger than any multienzyme complex described so far. With the availability of sufficient amounts of recombinant components of the system, it should be possible to perform reconstitution experiments and to characterize the composite capsids structurally by methods akin to those used in the present study and thus to obtain insight into the mechanism of interaction. Experimental Procedures Isolation of Tricorn Protease The preparation of extracts from T. acidophilum cells and the detection of peptidase activity were described previously (Tamura et al., 1996). The crude extract containing 2.8 g of protein was loaded onto a DEAE-Sephacel column (4 3 12 cm; Pharmacia) preequilibrated with buffer A (50 mM Tris–HCl [pH 7.5]). Bound protein was eluted using a 0–300 mM NaCl linear gradient in buffer A. The active fractions obtained from two runs were pooled and dialyzed against 120 mM potassium phosphate buffer (pH 7.0). The dialyzed sample containing 506 mg of protein was loaded onto a Hydroxylapatite column (2 3 12 cm; BioRad) preequilibrated with 120 mM potassium phosphate buffer. Proteins not bound to this column were collected and dialyzed against buffer B (50 mM Tris-HCl [pH 7.0]). The dialyzed sample containing 145 mg of protein was loaded onto a Q-Sepharose column (2 3 7 cm; Pharmacia) preequilibrated with buffer B. Bound proteins were eluted using a 0–300 mM NaCl linear gradient in buffer B. The active fractions were pooled and concentrated by means of Centriprep-30 (Amicon). Concentrated protein (70 mg) was loaded onto a Sepharose 6B column (2.4 3 85 cm; Pharmacia) preequilibrated with buffer B. The active fractions containing 12.5 mg of protein were pooled and loaded onto a Heparin-Sepharose column (1 3 8 cm; Pharmacia) preequilibrated with buffer B. Unbound protein (5.4 mg) was loaded onto a Mono-Q column (5 mm 3 5 cm; Pharmacia) preequilibrated with buffer B. Bound protein was eluted using a 0–300 mM NaCl linear gradient in buffer B. The active fractions were pooled and concentrated by using Centricon-30 (Amicon). Concentrated protein (1.2 mg) was loaded onto a Superose 6 HR column (1 3 30 cm; Pharmacia) preequilibrated with buffer B. The active fractions were collected and dialyzed against 50 mM Tris–HCl (pH 7.5) containing 20% glycerol and stored at 48C. Electron Cryomicroscopy of Isolated Capsids Droplets (4 ml each) of 1 mg/ml capsid suspension were applied to 100 3 400 mesh copper grids coated with perforated carbon film and covered with preadsorbed 5 nm diameter gold beads (Sigma) to be used as fiducial markers for the tomographic reconstruction. The samples were blotted with filter paper (Whatman No. 1) for 2–5 sec before plunging into liquid ethane (Dubochet et al., 1982). Grids were transferred under liquid nitrogen to a cryoholder (model 626, Gatan Inc., Pleasanton, CA) and oriented such that the long sides of the rectangular meshes were perpendicular to the cryoholder axis, allowing for a maximum field of view at high specimen-tilt angles. Grids were examined at 21858C.

Molecular Cell 64

For data collection, a CM200 FEG (Philips) transmission electron microscope (TEM) was used, equipped with a field-emission gun. The setup for automated electron tomography has been described previously (Dierksen et al., 1992, 1993, 1995). Tilt series and additional images required for TEM control were recorded by means of a slow-scan CCD camera.

Electron Cryomicroscopy of Cell Ghosts To image tricorn capsids in their cellular environment prior to the isolation and purification steps, we applied electron cryomicroscopy in combination with electron tomography. Since the dimensions of whole cells (diameter approx. 1 mm) do not allow images of a quality sufficient to render macromolecular structures visible to be obtained, even when energy filtering is applied (Grimm et al., 1996), we subjected cells to controlled lysis prior to cryofixation. We have taken advantage of the observation that T. acidophilum cells are stable only at a pH , 3; at higher pH, the cells become unstable and the cellular contents leak out. We performed the lysis by adding 100 mM Tris–HCl or 200 mM KOH to a concentrated sample of cell suspension, raising the pH to 7–8. Lysis was carried out on the grid, which was plunge-frozen in liquid ethane and observed by electron cryomicroscopy. Upon lysis, the cells flatten out, resulting in a thickness of the frozen-hydrated samples of 100–500 nm. Images of these cell ghosts were taken with a setup based upon a CM120 BIOGIF (Philips) TEM equipped with a post-column energy filter and CCD camera, as described previously (Grimm et al., 1997).

Tomographic Data Collection and Reconstruction On average, only one capsid per ice-spanned hole was found on the grids. To locate them, a special search mode by imaging the specimen at 3,8003 at an underfocus of approximately 60 mm was used. The dose required to locate a capsid was less than 0.5 e2/A˚2. Three tomographic tilt series, containing together seven tricorn capsid structures, were taken at an accelerating voltage of 120 kV and at 15,0003 magnification (a CCD-pixel size of 0.626 nm). Focusing was carried out automatically according to Koster and DeRuijter, 1992. The defocus was 7 mm, with the first zero of the contrast transfer function at (4.8 nm)21 . The tilt series covered the tilt range from 1658 to 2558 with 58 increment. Each series comprised 23 projections, and the cumulative dose was 35–40 e2/A˚2. For 3-D reconstructions exploiting the icosahedral symmetry of the capsids, additional 2-D data sets were recorded. To this end, images were taken at 20,0003 magnification, corresponding to a CCD-pixel size of 0.477 nm, with a dose between 15 and 25 e2/A˚2. These data were recorded at an underfocus of 1–2 mm , thus ensuring that the first zero of the contrast transfer function occurs between (1.8 nm)21 and (2.6 nm)21. All image processing steps were carried out on Silicon Graphics workstations using the EM program package (Hegerl, 1996). For the tomographic reconstructions, each of the three tilt series was aligned using a gold bead that was visible in all of the images of the tilt series as fiducial marker. After interactively determining the position of the gold bead, the alignment was refined by cross-correlation of the gold bead with a synthetic reference consisting of a black circular disk. The accuracy of the alignment was examined by displaying the aligned projections of the gold bead as a movie and was usually found to be better than two pixels. With the defocus value determined from the Thon rings in the power spectra of the images of the untilted specimen, the phases in the alternate zones of the contrast transfer function were flipped. The images were lowpass-filtered with a cutoff frequency of (2 nm)21 and aligned with respect to each other. The 3-D reconstruction with a pixel size of 1.26 nm was calculated via weighted backprojection. From the three reconstructed volumes, seven cubes with a size of 64 3 64 3 64 pixels were cut out at visually determined x, y, and z positions. With one reconstructed particle as reference, a coarse translational and orientational alignment of the volumes was carried out using 3-D cross correlation in cartesian and cylinder coordinate systems. The precession and nutation angles were scanned in 58 steps, resulting in 2592 angular positions. After one alignment cycle, a five-fold symmetry axis could be determined in the average. Placing a three-fold axis in the center of a single tricorn, the average

was icosahedrally symmetrized. For the following alignment, the angles were scanned only in the asymmetric unit of the icosahedron. For the refinement cycles, the particles were reconstructed with a pixel size of 0.63 nm in volumes with a size of 128 3 128 3 128 pixels. After each cycle, the icosahedrally symmetrized average was calculated via weighted backprojection from projections that were cut out from the phase corrected images of the tilt series. Using the average as reference for the next cycle, the alignment procedure was continued until the changes of the shifts and orientations became insignificant. Reconstruction from 2-D Images Exploiting Icosahedral Symmetry First, the original images were corrected for the point spread function of the CCD camera by multiplication with its inverse modulation transfer function in Fourier space. Second, the sign of the phases in alternate zones of the contrast transfer function were flipped using defocus values determined from image areas with carbon film. From 35 2-D images, 38 views of capsids were boxed, low-pass filtered with a cutoff radius of (1.5 nm)21, and centered by crosscorrelation with the rotationally averaged sum of all views. For the search of projection angles, the 3-D map of the tricorn capsid structure resulting from the tomographic evaluation was used as a starting model after rescaling to match the pixel size of the 2-D images. Sampling the asymmetric unit in 28 steps, it was projected in different directions. The projection angles and shifts for the views were determined via cross-correlation in cartesian and polar coordinates using the projections of the starting model as multiple reference. In principle, the reconstruction follows the scheme described in Baker and Cheng (1996). The reconstruction calculated via weighted backprojection was used as reference for the next alignment cycle. After several refinement cycles with angular step sizes of 28, 18, and 0.58, the reconstruction did not change noticeably. To obtain an estimate of the resolution of the average, two sets of particles were independently reconstructed (Saxton and Baumeister, 1982). To avoid contributions from areas outside the structure, a spherically shaped mask with a Gaussian falloff was applied to the outer and inner part of the reconstructed volumes. The Fourier shell correlation and the phase residual were calculated on 64 shells in Fourier space. The reconstructions were visualized either in slices or in isosurface representations using the AVS software (Advanced Visual Systems). For isosurface rendering, the density data were filtered to the resolution limit given by the falloff of the Fourier shell correlation. The threshold was chosen to include a mass of 14.6 MDa, which corresponds to the 20-fold mass of a single tricorn molecule (the mass density was assumed to be 1.3 g/cm3). Acknowledgments The authors would like to thank Ute Santarius for the electron microscopy with the negatively stained samples. Received June 12, 1997; revised September 15, 1997. References Baker, T.S., and Cheng, R.H. (1996). A model-based approach for determining orientations of biological macromolecules imaged by cryoelectron microscopy. J. Struct. Biol. 116, 120–130. Baumeister, W., and Lupas, A. (1997). The proteasome. Curr. Opin. Struct. Biol. 7, 273–278. Baumeister, W., Cejka, Z., Kania, M., and Seemu¨ller, E. (1997). The proteasome: a macromolecular assembly designed to confine proteolysis to a nanocompartment. Biol. Chem. 378, 121–130. Coux, O., Tanaka, K., and Goldberg, A.L. (1996). Structure and functions of the 20S and 26S proteasomes. Annu. Rev. Biochem. 65, 801–847. Dierksen, K., Typke, D., Hegerl, R., Koster, A.J., and Baumeister, W. (1992). Towards automatic electron tomography. Ultramicroscopy 40, 71–87.

Tricorn Protease—An Icosahedral Supermolecule 65

Dierksen, K., Typke, D., Hegerl, R., and Baumeister, W. (1993). Towards automatic electron tomography. II. Implementation of autofocus and low-dose procedures. Ultramicroscopy 49, 109–120. Dierksen, K., Typke, D., Hegerl, R., Walz, J., Sackmann, E., and Baumeister, W. (1995). Three-dimensional structure of lipid vesicles embedded in vitreous ice and investigated by automated electron tomography. Biophys. J. 68, 1416–1422. Dubochet, J., Lepault, J., Freeman, R., Berriman, J.A., and Homo, J.-C. (1982). Electron microscopy of frozen water and aqueous solutions. J. Microsc. 128, 219–237. Grimm, R., Koster, A.J., Ziese, U., Typke, D., and Baumeister, W. (1996). Zero-loss energy-filtering under low-dose conditions using a postcolumn energy filter. J. Microsc. 183, 60–68. Grimm, R., Ba¨ rmann, M., Ha¨ckl, W., Typke, D., Sackmann, E., and Baumeister, W. (1997). Energy-filtered electron tomography of iceembedded actin and vesicles. Biophys. J. 72, 482–489. Hegerl, R. (1996). The EM program package: a platform for image processing in biological electron microscopy. J. Struct. Biol. 116, 30–34. Joshua-Tor, L., Xu, H.E., Johnston, S.A., and Rees, D.C. (1995). Crystal structure of a conserved protease that binds DNA: the bleomycin hydrolase, Gal6. Science 269, 945–950. Kis, K., and Bacher, A. (1995). Substrate channeling in the lumazine synthase/riboflavin synthase complex of Bacillus subtilis. J. Biol. Chem. 270, 16788–16795. Koster, A.J., and DeRuijter, W.J. (1992). Practical autoalignment of transmission electron microscopes. Ultramicroscopy 40, 89–107. Ladenstein, R., Schneider, M., Huber, R., Bartunik, H.D., Wilson, K., Schott, K., and Bacher, A. (1988). Heavy riboflavin synthase from Bacillus subtilis. Crystal structure analysis of the icosahedral b60 capsid at 3.3 A˚ resolution. J. Mol. Biol. 203, 1045–1070. Lo¨we, J., Stock, D., Jap, B., Zwickl, P., Baumeister, W., and Huber, R. (1995). Crystal structure of the 20S proteasome from the archaeon T. acidophilum at 3.4 A˚ resolution. Science 268, 533–539. Mattevi, A., Obmolova, G., Schulze, E., Kalk, K.H., Westphal, A.H., de Kok, A., and Hol, W.G.J. (1992). Atomic structure of the cubic core of the pyruvate dehydrogenase multienzyme complex. Science 255, 1544–1550. Mortl, S., Fischer, M., Richter, G., Tack, J., Weinkauf, S., and Bacher, A. (1996). Biosynthesis of riboflavin. Lumazine synthase of Escherichia coli. J. Biol. Chem. 271, 33201–33207. Peters, J.M., Cejka, Z., Harris, J.R., Kleinschmidt, J.A., and Baumeister, W. (1993). Structural features of the 26S proteasome complex. J. Mol. Biol. 234, 932–937. Saxton, W.O., and Baumeister, W. (1982). The correlation averaging of a regularly arranged bacterial cell envelope protein. J. Microsc. 127, 127–138. Stoops, J.K., Baker, T.S., Schroeter, J.P., Kolodziej, S.J., Niu, X.D., and Reed, L.J. (1992). Three-dimensional structure of the truncated core of the Saccharomyces cerevisiae pyruvate dehydrogenase complex determined from negative stain and cryoelectron microscopy images. J. Biol. Chem. 267, 24769–24775. Tamura, T., Tamura, N., Cejka, Z., Hegerl, R., Lottspeich, F., and Baumeister, W. (1996). Tricorn protease—the core of a modular proteolytic system. Science 274, 1385–1389. Tamura, T., Tamura, N., Lottspeich, F., and Baumeister, W. (1996). Tricorn protease (TRI) interacting factor 1 from Thermoplasma acidophilum is a proline iminopeptidase. FEBS Lett. 398, 101–105. Wenzel, T., and Baumeister, W. (1995). Conformational constraints in protein degradation by the 20S proteasome. Nat. Struct. Biol. 2, 199–204.