The HexamericE. coliDnaB Helicase can Exist in Different Quarternary States

The HexamericE. coliDnaB Helicase can Exist in Different Quarternary States

J. Mol. Biol. (1996) 259, 7–14 COMMUNICATION The Hexameric E. coli DnaB Helicase can Exist in Different Quarternary States Xiong Yu1, Maria J. Jezew...

2MB Sizes 0 Downloads 40 Views

J. Mol. Biol. (1996) 259, 7–14

COMMUNICATION

The Hexameric E. coli DnaB Helicase can Exist in Different Quarternary States Xiong Yu1, Maria J. Jezewska2, Wlodzimierz Bujalowski2 and Edward H. Egelman1* 1

Department of Cell Biology and Neuroanatomy University of Minnesota Medical School, Minneapolis MN 55455, USA 2

Department of Human Biological Chemistry and Genetics, The University of Texas Medical Branch at Galveston, Galveston TX 77555-1053, USA

The DnaB protein is the primary replicative helicase in Escherichia coli, and the active form of the protein is a hexamer. It has been reported that the protein forms a ring with strong 3-fold symmetry, which was suggested to be a trimer of dimers. We show that under different conditions, using either ATP, ATPgS, AMP-PNP or ADP as nucleotide cofactors, we always find two different forms of the DnaB ring; one with a 3-fold symmetry and one with 6-fold symmetry. We have used scanning transmission electron microscopy for mass analysis, and have found that both forms are hexamers, excluding the possiblity that the 3-fold form is in fact a trimer of the 52 kDa monomer. We have also found rings that are in an intermediate state between these two. The existence of hexamers in discrete states shows that the transitions between these states must be cooperative. These observations suggest that there may be an equilibrium between two different conformations of the hexameric ring. The role of these two states in the mechanism of helicase action remains to be determined. 7 1996 Academic Press Limited

*Corresponding author

Keywords: DnaB; DNA replication; helicases; electron microscopy; image analysis

The DnaB protein is essential for DNA replication in Escherichia coli. It is a helicase that hydrolyzes ATP whilst unwinding duplex DNA into two single strands in front of a replication fork (LeBowitz & McMacken, 1986). DnaB was first characterized as a hexamer of approximately 52 kDa subunits by means of cross-linking (Reha-Krantz & Hurwitz, 1978), and more recent sedimentation and ligand binding studies have shown that DnaB exists as a stable hexamer over a large range of protein concentrations (Bujalowski et al., 1994). The hexamer exists as a ring (Bujalowski et al., 1994; San Martin et al., 1995), and single-stranded DNA does not wrap around the ring, (Bujalowski & Jezewska, 1995). This suggests that the DNA passes through the center of the ring, as shown for the E. coli RuvB helicase (Stasiak et al., 1994) and the bacteriophage T7 gp4 helicase (Egelman et al., 1995), and as proposed for the simian virus 40 large T helicase (Dean et al., 1992). A recent paper (San Martin et al., 1995) reports that in the presence of ATP, the DnaB protein assembles as a triangular structure with strong Abbreviations used: AMP-PNP, b,g-imidoadenosine 5'-triphosphate; gp, gene product. 0022–2836/96/210007–08 $18.00/0

3-fold symmetry. Six large density peaks within this triangular structure were interpreted as arising from a trimer of dimers. Since DnaB protein alone displays a significant ATPase activity in the absence of DNA, it is possible that the 3-fold symmetric structure results from either the partial or complete hydrolysis of ATP. Furthermore, analysis of sedimentation data has shown that in the absence of magnesium cations DnaB can exist as a trimer (Bujalowski et al., 1994). Since the three-dimensional reconstruction described by San Martin et al. (1995) contains a volume that is 50 to 55% of that expected for a 314 kDa hexamer, it is possible that the 3-fold structure is actually a trimer. We report here that in the presence of either ADP, ATP or non-hydrolyzable analogs of ATP, the DnaB protein assembles into two different structures, one with 3-fold symmetry and one with 6-fold symmetry, both of which are hexameric. Figure 1 shows an electron micrograph of negatively stained DnaB particles, obtained in the presence of the non-hydrolyzable ATP analog, AMP-PNP. Since the ring-like structures were not observed in the absence of a nucleotide cofactor under the conditions used in this study, visualization of the ring-like structures is dependent upon 7 1996 Academic Press Limited

8

Figure 1. Electron micrograph of the DnaB protein in the presence of AMP-PNP. The white circles indicate examples of rings that were initially classified as circular, while the black circles indicate rings that were initially ˚. classified as triangular. The scale bar represents 500 A DnaB protein was prepared as described by Bujalowski & Klonowska (1993). An incubation of 0.7 mM DnaB in a 25 mM tetraethylammonium (TEA) buffer (pH 7.2) was made at 37°C for 15 minutes with 10 mM magnesium acetate and 1 mM AMP-PNP (Sigma Chemical Co.). The reaction mixture was applied to glow-discharged grids, stained with 2% (w/v) uranyl acetate, and imaged at 30,000 × under minimal dose conditions in a JEOL 1200 EXII electron microscope at an acceleration voltage of 80 kV.

the presence of a nucleotide. Similar structures were also observed using ATP, ADP or the poorly hydrolyzable ATP analog, ATPgS (data not shown). Visual inspection of the micrographs reveals that DnaB forms structures with two predominant morphologies: circular or triangular. Under all of the conditions that we have used where ring-like structures formed (with ATP, ADP, ATPgS, and AMP-PNP), we have seen a nearly equal mixture of these two forms. We chose to average the images separately for the circular and triangular particles. The resulting initial averages are shown in Figure 2a and e for the DnaB–AMP–PNP preparations, and Figure 2k and o for the DnaB–ADP preparation. The averages are very similar for both AMP–PNP and ADP, but a quantitative comparison shows that the peak density of the subunits in the circular rings moves outwards by a small but significant amount between the DnaB–ADP and DnaB–AMP–PNP rings. In contrast to this small difference, the circular and triangular structures are easily distinguished. The averaged triangular structures appear extremely similar to the DnaB average shown in San Martin et al. (1995). In an effort to determine if these averages contained heterogeneous population of particles, the images were analyzed using rotational power spectra (Crowther & Amos, 1971). We found this to

Communication

be the most straightforward approach for examining the particular populations. The rotational power spectra in Figure 3 of the various averages contained in Figure 2 show that the 6-fold symmetric structures contain no 3-fold power. Hence, there is no apparent perturbation of the hexameric symmetry due to dimerization. Using rotational power spectra analysis, we were able to create averages of the circular images that showed the strongest 6-fold power (Figure 2c and m) and averages of the triangular images that showed the strongest 3-fold power (Figure 2g and q). By searching for those images within the triangular averages that had the weakest 3-fold power and the strongest 6-fold power, we were able to generate the averages of Figure 2i (DnaB–AMP–PNP ring) and Figure 2s (DnaB–ADP ring) that appear to be intermediate between the 6-fold and 3-fold structures. The existence of the stain-filled central channel in both the triangular and hexagonal averages excludes the possibility that the two different classes actually arise from different projections of the same structure. This central channel lies around the symmetry axis of the projections, and this excludes the possibility that the two classes could be related by a difference in the angle of projection of the same three-dimensional structure. To assess the resolution of the averages, we used the differential phase residual method (Frank et al., 1981). The averages were randomly divided into two approximately equal halves, and the average of each half was compared to determine the resolution at which the mean phase error between the transforms of the two was equal to 45°. For the average in Figure 2k (6-fold symmetry, n = 617 ˚ −1, and for the images), this resolution was 1/28 A average in Figure 2e (3-fold symmetry, n = 559 ˚ −1. A similar images) this resolution was 1/30 A approach was used to assess the quality of the 3-fold and 6-fold symmetry as a function of resolution. In comparing Figure 2g and h, 2k and l, 2o and p, and 2s and t, the phase error is better than 45° for ˚ −1 (for Figall of reciprocal space less than 1/30 A ure 2o, the imposition of symmetry leads to less than a 45° degree residual at resolutions out ˚ −1 ). The imposition of symmetry was to 1/25 A relatively poor at higher resolution for the averages of Figure 2c, i, and m, with a 45° residual occurring ˚ −1. But most of the power in these at about 1/50 A averages is still consistent with the imposition of this symmetry. Both forms of the DnaB ring (the hexagonal and the triangular) yield averages that have a strong chirality or handedness. This can be seen by the ‘‘pinwheel’’ appearance of the hexagonal structure (i.e. Figure 2b, d, l and n) and by the fact that the triangular structures do not have mirror symmetry (i.e. Figure 2f, h, p and r). Since this chirality arises from the initial averaging, without selecting particles that display this handedness, it suggests that all or most of the particles used for averaging are oriented in a preferred direction on the carbon substrate. If the particles were randomly arranged,

9

Communication

this chirality would cancel in the initial averaging. This does not necessarily mean that all particles adsorb to the carbon substrate with the same orientation, but rather that most particles that give rise to a symmetrical projection with a stain-filled central hole may have this preferred orientation. We have observed that the T7 gp4a helicase displays a similar but weaker chirality in the hexagonal ring that it forms (Egelman et al., 1995). The three dimensional reconstruction of that ring structure

shows that there is a large face and a small face, the most likely basis for such a preferred orientation. San Martin et al. (1995) have also observed a preferred orientation of the triangular DnaB rings, and their three-dimensional reconstruction of this triangular structure reveals a flat side and a bumpy side. DnaB has been shown to form a trimer under certain conditions (Bujalowski et al., 1994), and scanning transmission electron microscopy was

Figure 2 (legend overleaf)

10

Communication

used to exclude the possibility that the triangular images are actually generated by a trimer of DnaB protein, rather than a hexamer. Figure 4 shows mass histograms obtained from scanning transmission electron microscope images of DnaB–AMP–PNP and DnaB–ATPgS particles. The mass histograms show that there is no significant peak at 0157 kDa where we would expect to find a trimer. Although the average mass measured 254 kDa for the DnaB–AMP–PNP ring and 293 kDa for the DnaB– ATPgS ring, somewhat less than that of the 314 kDa expected for a DnaB hexamer, the mass distribution that is measured is nevertheless consistent with all of the ring structures being hexameric, and almost none of them being trimeric.

The existence of at least three distinct states, the 3-fold, the 6-fold, and the intermediate averages of Figure 2i and s, raises the question of what type of conformational change can generate these different structures. We therefore generated a statistical difference map (Figure 5(c)) between a 6-fold ring (Figure 5(a)) and a 3-fold ring (Figure 5(b)) to see where mass was moving. The result (Figure 5(c)) shows that all of the mass difference occurs near the outer edge of the ring, and the differences are localized into alternating regions of positive and negative density. This result, combined with the intermediate state, has suggested to us a simple model that would explain the conformational change (Figure 5(d)).

Figure 2. Averages of DnaB rings prepared with AMP-PNP (a to j) and ADP (k to t). Electron micrographs were ˚ /pixe1. Images of rings were masked into 34 × 34 pixel arrays (each image shown digitized at a sampling interval of 6 A ˚ × 204 A ˚ ), band-pass filtered (between 1/120 and 1/18 A ˚ −1 ), scaled to zero mean density, and the contrast was is 204 A normalized. Initially, images of individual DnaB rings were selected by eye according to whether they were circular or triangular. By using the criteria that the rings were approximately symmetrical, and displayed a stain-filled hole in the center, about 20% of the total number of particles observed on the grid were selected. The remainder suffered from poor staining or were in a very different orientation on the carbon substrate. For the circular rings, an initial average of 612 DnaB–AMP–PNP rings (a) and 617 DnaB–ADP rings (k) were generated using a reference-free alignment (Penczek et al., 1992). 6-fold symmetrized versions of these averages are shown in b and l for the AMP-PNP and ADP states, respectively. Images within these averages were ranked according to the strength of the 6-fold rotational power. For the 6-fold ranking, we also wished to exclude those images with a strong 3-fold power. If pn is the power of the n-fold rotational component (Crowther & Amos, 1971), let p'6 be a local background corrected power: p'6 = p6 −

(p5 + p7 ) 2

Similarly, for the 3-fold background corrected power: p'3 = p3 −

(p4 + p2 ) 2

The corrected 6-fold power is then given by: p"6 =

(p'6 − p'3 ) ptot

For the DnaB–AMP–PNP particles, 228 of the original 612 circular images had a value of p'6 greater than zero and a value of p'3 less than or equal to zero. These 228 images were then aligned independently. Their average is shown in c, while a 6-fold symmetrized version of this average is shown in d. For the DnaB–ADP rings, 254 of the original 617 circular images had a value of p"6 greater than zero and a value of p'3 less than or equal to zero, and the independent alignment of these 254 images is shown in m. A 6-fold symmetrized version of this average is shown in n. An initial average of 559 triangular DnaB–AMP–PNP rings is shown in e, while the initial average of 551 DnaB–ADP rings is shown in o. The 3-fold symmetrized versions of these are shown in f and p for DnaB–AMP–PNP and DnaB–ADP, respectively. Following the procedure described above, these triangular rings were sorted by the strength of the background corrected 3-fold power, p'3 . For the DnaB–AMP–PNP rings, 386 out of the initial 559 images had a value of p'3 greater than zero, and the independent average of these following alignment is shown in g. A 3-fold symmetrized version of this average is shown in h. For the DnaB–ADP rings, 397 out of the initial 551 images had a value of p'3 greater than zero, and the independent average of these following alignment is shown in q. A 3-fold symmetrized version of this average is shown in r. Lastly, we examined the images that had been rejected by the criteria used above to find other classes. The 173 triangular images used in the DnaB–AMP–PNP average of e that did not have background-corrected 3-fold power greater than zero were sorted by the strength of their 6-fold power. Of these, 119 had a ‘‘corrected’’ 6-fold power greater than zero. The top 50 of these were independently aligned, and their average is shown in i. A 3-fold symmetrized version of this DnaB–AMP–PNP average is shown in j. The same treatment was made for the DnaB–ADP images. The 154 triangular images used in the DnaB–ADP average of o that did not have background-corrected 3-fold power greater than zero were sorted by the strength of their 6-fold power. Of these, 100 had a corrected 6-fold power greater than zero. The top 50 of these were independently aligned, and their average is shown in s. A 3-fold symmetrized version of this average is shown in t. Although the averages in i and s are from images in which the background corrected 3-fold power was less than or equal to zero, these two averages have significant 3-fold power. This shows that the 3-fold power in the individual images is at or below the level of the noise.

11

Communication

(a)

(d)

(b)

(e)

(c)

(f)

Figure 3. Rotational power spectra (Crowther & Amos, 1971) for DnaB–AMP–PNP averages ((a), (b) and (c)) and DnaB–ADP averages ((d), (e) and (f)). The spectra in (a), (b) and (c) are, respectively, from the averages in Figure 2c, g and i. For the DnaB–ADP rings, the spectra in (d), (e) and (f) are, respectively, from the averages in Fig. 2m, q and ˚ . However, the results were found to be insensitive s. The spectra were calculated within the radial limits of 12 to 96 A to changes in these radial limits.

Discussion In the presence of either a nucleoside diphosphate or a nucleoside triphosphate, the DnaB protein forms two hexameric assemblies, one with distinct C6 symmetry and the other with C3 symmetry. Interestingly, the C3 structure is a trimer

of dimers, and it has been shown that DnaB hexamers contain three high affinity ATP-binding sites and three low affinity ones (Bujalowski & Klonowska, 1993). Within the C3 rings there are two different subunit environments, which may be the structural basis for the biochemical observations. The E. coli transcription termination helicase rho

12

(a)

(b)

Figure 4. Scanning transmission electron microscope (STEM) mass histograms for DnaB–ATPgS (a) and DnaB–AMP–PNP (b) complexes. Unstained, freeze-dried samples were imaged under dark-field conditions at the Brookhaven STEM (Wall & Hainfeld, 1986). Image intensity is proportional to the number of electrons scattered by the sample, and this is directly related to the mass of the specimen. Tobacco mosaic virus particles (TMV) were used as an internal mass standard. Images of 297 DnaB–AMP–PNP rings and 451 DnaB–ATPgS complexes were analyzed. The mean mass was 254(23) standard error of the mean (SEM) kDa for the DnaB–AMP–PNP rings, and 293(22) (SEM) kDa for the DnaB–ATPgS rings. The mass histograms are consistent with all particles being hexamers, and are inconsistent with any significant number of trimers.

also displays three high affinity ATP-binding sites (Stitt, 1988) and three low affinity sites (Geiselmann

Communication

& von Hippel, 1992). Two different subunit-subunit interactions in rho were suggested to arise from D3 symmetry (Seifried et al., 1991), but a more recent study has suggested a C3 symmetry for rho (Miwa et al., 1995). The T7 gp4 hexameric helicase binds only three nucleotides (Patel & Hingorani, 1995), but as yet no structures with C3 symmetry have been observed for this helicase (Egelman et al., 1995). It remains to be determined what the relationship is between the different structural states of DnaB and the biochemical states that DnaB adopts as part of the ATPase cycle. A small difference does exist between the C6 rings formed in the presence of ADP and those formed in the presence of AMP-PNP. When these two averages are superimposed, it appears that the hole in the center of the ring does not change, but that the peak density in each subunit in the AMP-PNP ˚ radially, leading ring moves outwards by about 5 A ˚ larger in diameter to a ring that is about 10 A than the DnaB–ADP ring. Consistent with this, solution hydrodynamic measurements of these rings has shown that the sedimentation coefficient increases between DnaB hexamers formed with ADP and those formed with AMP-PNP (Jezewska & Bujalowski, 1996). The existence of a sub-population that averages into a state nearly intermediate between the strong 3-fold and the 6-fold symmetric structures suggests that the two states may be in equilibrium. This intermediate would then represent either rings that are visualized as they undergo the transformation between the two states, or rings that are stably trapped in this intermediate state. In either case, the existence of these discrete states suggests that the subunits within a ring change state with a very high degree of cooperativity. If dimerization takes place randomly within the C6 rings, we would see sub-populations of rings where one pair of subunits has dimerized and where two pairs have dimerized, in addition to the C3 rings where all three pairs have dimerized. But we have not seen such subsets (data not shown) using correspondence analysis (Frank et al., 1988). The only subset that we do find appears to be with all three dimers in a conformation intermediate between the C6 and C3 states. This suggests that the dimerization does not occur randomly, but occurs in a concerted fashion throughout the ring. Is it possible that there is no intermediate state, but that it is an average of 3-fold and 6-fold symmetric structures? This would be consistent with the observation that the intermediate state is very similar to a simple average of the 3-fold and 6-fold structures (data not shown), but we have been able to exclude this possibility. The 50 images used to create the DnaB–ADP average of Figure 2t were ranked according to the strength of their 3-fold power. The average of the best 25 and the worst 25 looked quite similar to the average of Figure 2t (with the difference that the average of the worst 25 was more noisy), while we would have expected that the average of the worst 25 would resemble the 6-fold structure and the

13

Communication

Figure 5. The DnaB–ADP 6-fold structure (a), the DnaB–ADP 3-fold structure (b) and a statistical difference map between the two (c). The structures in (a) and (b) have been aligned against each other, and linearly scaled so as to minimize the sum of the squared differences in density. The density of the 3-fold structure has then been subtracted from the 6-fold structure, and this difference divided by the standard error of the difference, determined from the variances of the two averages (Egelman & Yu, 1989). The resulting difference map (c) is in units of standard deviations (s), and is shown for values greater than 5s (continuous lines) and less than −5s (dotted lines), with 2s intervals between contour steps. Thus, the dotted lines indicate regions where the density of the 3-fold structure is significantly greater than the density of the 6-fold structure, while the continuous lines in (c) indicate regions where the density of the 6-fold structure is significantly greater than the density of the 3-fold structure. (d) A model for the dimerization process that can explain the transition between the DnaB ring with 6-fold symmetry and the one with 3-fold symmetry. In this model there is a large rotation of three of the subunits, or parts of the subunits (shown by arrows), with respect to the other three subunits. The model is supported by the existence of an intermediate state between these two conformations.

average of the best 25 would resemble the 3-fold structure if this possibility were true. The intermediate state does provide some insight into the type of conformational change that occurs between the C6 and C3 structures. By animating the three different states, it has been possible to see that three of the subunits in the ring may remain more or less fixed, and that the other three undergo a very large rotation. This has been illustrated in the model of Figure 5d, in which the subunit has been represented by two spheres. We previously showed that the hexameric rings formed by the E. coli RuvB (Stasiak et al., 1994) and the T7 gp4 protein (Egelman et al., 1995) contain two tiers, which must arise from each subunit having two lobes, a large one and a small one. Based on their three-dimensional reconstruction of the C3 form of the DnaB

ring, San Martin et al. (1995) also suggested that each subunit may be organized into a large and a small lobe. Thus, this may be a conserved structural motif among the hexameric helicases. Independent of the details of the model, it appears that a large conformational change is required to convert the C6 to the C3 structure. A key task will be to relate this to the mechanism of DnaBs helicase activity.

Acknowledgements We thank Joe Wall and Martha Simon of the Brookhaven National Laboratory STEM Facility (National Institutes of Health Biotechnology Resource) for their invaluable assistance. This work was supported by

14

Communication

National Institutes of Health GM35269 (E.H.E.) and GM46679 (W.B.).

References Bujalowski, W. & Jezewska, M. J. (1995). Interactions of Escherichia coli primary replicative helicase DnaB protein with single-stranded DNA. The nucleic acid does not wrap around the protein hexamer. Biochemistry, 34, 8513–8519. Bujalowski, W. & Klonowska, M. M. (1993). Negative cooperativity in the binding of nucleotide to Escherichia coli replicative helicase DnaB protein. Interactions with fluorescent nucleotide analogs. Biochemistry, 32, 5888–5900. Bujalowski, W., Klonowska, M. M. & Jezewska, M. J. (1994). Oligomeric structure of Escherichia coli primary replicative helicase DnaB protein. J. Biol. Chem. 269, 31350–31358. Crowther, R. A. & Amos, L. A. (1971). Harmonic analysis of electron microscopic images with rotational symmetry. J. Mol. Biol. 60, 123–130. Dean, F. B., Borowiec, J. A., Eki, T. & Hurwitz, J. (1992). The simian virus 40 T antigen double hexamer assembles around the DNA at the replication origin. J. Biol. Chem. 267, 14129–14137. Egelman, E. H. & Yu, X. (1989). The location of DNA in RecA-DNA helical filaments. Science, 245, 404– 407. Egelman, E. H., Yu, X., Wild, R., Hingorani, M. M. & Patel, S. S. (1995). Bacteriophage T7 helicase/primase proteins form rings around single-stranded DNA that suggest a general structure for hexameric helicases. Proc. Natl Acad. Sci. USA, 92, 3869–3873. Frank, J., Verschoor, A. & Boublik, M. (1981). Computer averaging of electron micrographs of 40S ribosomal subunits. Science, 214, 1353–1355. Frank, J., Bretaudiere, J. P., Carazo, J. M., Verschoor, A. & Wagenknecht, T. (1988). Classification of images of biomolecular assemblies: a study of ribosomes and ribosomal subunits of Escherichia coli. J. Microsc. 150, 99–115. Geiselmann, J. & von Hippel, P. H. (1992). Functional interactions of ligand cofactors with Escherichia coli

transcription termination factor rho. I. Binding of ATP. Protein Sci. 1, 850–860. Jezewska, M. J. & Bujalowski, W. (1996). Global conformational transitions in Escherichia coli primary replicative helicase DnaB protein induced by ATP, ADP and single-stranded DNA binding. Multiple conformational states of the helicase hexamer. J. Biol. Chem. 271, 4261–4265. LeBowitz, J. H. & McMacken, R. (1986). The Escherichia coli dnaB replication protein is a DNA helicase. J. Biol. Chem. 261, 4738–4748. Miwa, Y., Horiguchi, T. & Shigesada, K. (1995). Structural and functional dissections of transcription termination factor rho by random mutagenesis. J. Mol. Biol. 254, 815–837. Patel, S. S. & Hingorani, M. M. (1995). Nucleotide binding studies of bacteriophage T7 DNA helicase-primase protein. Biophys. J. 68, 186s–190s. Penczek, P., Radermacher, M. & Frank, J. (1992). Three-dimensional reconstruction of single particles embedded in ice. Ultramicroscopy, 40, 33–53. Reha-Krantz, L. J. & Hurwitz, J. (1978). The dnaB gene product of Escherichia coli. I. Purification, homogeneity, and physical properties. J. Biol. Chem. 253, 4043–4050. San Martin, M. C., Stamford, N. P. J., Dammerova, N., Dixon, N. E. & Carazo, J. M. (1995). A structural model for the Escherichia coli DnaB helicase based on electron microscopy data. J. Struct. Biol. 114, 167–176. Seifried, S. E., Bjornson, K. P. & von Hippel, P. H. (1991). Structure and assembly of the Escherichia coli transcription termination factor rho and its interactions with RNA. II. Physical chemical studies. J. Mol. Biol. 221, 1139–1151. Stasiak, A., Tsaneva, I. R., West, S. C., Benson, C. J. B., Yu, X. & Egelman, E. H. (1994). The Escherichia coli RuvB branch migration protein forms double hexameric rings around DNA. Proc. Natl Acad. Sci. USA, 91, 7618–7622. Stitt, B. L. (1988). Escherichia coli transcription termination protein rho has three hydrolytic sites for ATP. J. Biol. Chem. 263, 11130–11137. Wall, J. S. & Hainfeld, J. F. (1986). Mass mapping with the scanning transmission electron microscope. Annu. Rev. Biophys. Biophys. Chem. 15, 355–376.

Edited by I. A. Wilson (Received 13 November 1995; received in revised form 1 March 1996; accepted 4 March 1996)