- Email: [email protected]
Removal of the RecA C-terminus results in a conformational change in the RecA-DNA filament
JOURNAL
OF STRUCTURAL
BIOLOGY
106,
243-254
(1991)
Removal of the RecA C-terminus Results in a Conformational Change in the RecA-DNA Filament X. Yu AND E. H. EGELMAN Department
of Cell Biology and Neuroanutomy, University of Minnesota Medical School, 321 Church Street SE, Minneapolis, Minnesota 55455 Received December 18, 1990, and in revised form March 21, 1991
can be visualized within the filament as containing two lobes or domains (Yu and Egelman, 1990). That study was based upon the analysis of RecA filaments formed on double-stranded DNA (dsDNA) with a complex of ADP and AlF, as a cofactor. We have now obtained a three-dimensional reconstruction of a proteolytically truncated RecA, with about 18 residues cleaved from the C-terminus. While the initial expectation was that such a cleaved molecule would allow us to visualize the C-terminus in a comparison with the intact molecule, we have found instead that the loss of the C-terminus results in a significant conformational change with the RecA filament. We suggest that this conformational change can be related to motions within the wild-type RecA filament, and that the ability to visualize such changes will be important in ultimately constructing a dynamic atomic model for the RecA-mediated strand exchange.
The Escherichia coli RecA protein catalyzes homologous recombination of DNA molecules, and the active form of the protein is a helical polymer that it forms around DNA. Previous image analysis of electron micrographs has revealed the RecA protein to be organized into two domains or lobes within the RecA-DNA filament. We have now been able to show that a small modification of the RecA protein by proteolysis results in a significant shift in the internal mass in the RecA filament. We have cleaved approximately 18 residues from the C-terminus of the RecA protein, producing a roughly 36K MW RecA core protein that binds DNA and polymerizes normally. A three-dimensional reconstruction of this complex has been computed, and has been compared with a previous reconstruction of the intact protein. The main difference is consistent with a 15 A outward movement of the lobe that was at an inner radius in the wild-type protein. These observations yield additional evidence about the conformational flexibility of the RecA filament, and will aid in understanding the structural mechanics and dynamics of the RecA filament. o 1991 Academic PRESS,IIIC.
MATERIALS AND METHODS Proteins and Nucleic Acids RecA was purified by the method of West et al. (19821, with an additional polyspermidine precipitation, as described by Griffith and Shores (1985). The purified RecA ran as a single band on an SDS-PAGE Coomassie blue stained gel. Double stranded, circular +X174 was a gift from Dr. Paul Howard-Flanders. The DNA was linearized by restriction endonuclease PstI (IBI), and ran as a single band on an agarose gel. The concentration of the DNA was measured using an extinction coefficient Azw of 1 for 50 pg/ml and a l-cm path length. Reactions of RecA and DNA were carried out at 37°C in 25 mM Triethanolamine-HCl buffer (pH 7.2). RecA concentration was 7 pM. The Reckdouble stranded DNA ratio was equal to 4O:l (w:w) or a 1:3 molar ratio of RecA to total nucleotides. This is a twofold excess of RecA, given that each RecA protomer binds three base pairs, or six nucleotides. ATP (Sigma) concentration was 2.5 mM. Aluminum nitrate and sodium fluoride (both from Baker) were added to 2.5 mM in some experiments, while in others the aluminum nitrate concentration was 100 pi%f and the sodium fluoride was 2.5 mM. No differences were seen between these two sets of conditions.
INTRODUCTION
The RecA protein is the most intensively studied enzyme of general genetic recombination (for reviews, see Radding, 1981; Howard-Flanders et al., 1984; Cox and Lehman, 1987). One of the most remarkable features about RecA is how a 38K MW protein can mediate the search for homology between two different DNA molecules, promote a strand exchange reaction between these molecules, and cleave its own repressor, among other activities. An understanding of RecA function must begin with the knowledge that the active form of the RecA protein is a filamentous complex containing the RecA protein, a polynucleotide substrate (usually DNA), and a nucleotide triphosphate cofactor (usually ATP). The complex has a stoichiometry of one RecA subunit per ATP per three bases or base pairs. We have recently shown that the RecA protein
Proteolytic Digestion of RecA RecA protein was digested with subtilisin (Sigma) at 4°C for 20 hr. The RecA concentration was 7 pM and the subtilisin concen243 1047~8477/91 $3.00 Copyright 0 1991 by Academic Preee, Inc. All rights of reproduction in any form reserved.
244
YU AND EGELMAN
tration was 0.23 p&f, and the digestion was in a 12 mM Tris buifer at pH 7.8. The BecA was purified after the digestion on a Bio-Gel A-O.5 m gel column (Bio-Bad) in the 12 mM Tris buffer at pH 7.8. An Edman’s degradation was performed on the cleaved BecA by the Microchemical Facility, Department of Human Genetics of the University of Minnesota. The first 10 N-terminal residues were sequenced. Electron Microscopy Samples were prepared on carbon-coated copper grids which had been glow discharged. After a sample was placed on the grid, excess solution was removed by blotting with filter paper. Then the samples were negatively stained with 2% uranyl acetate and viewed with either a JEOL 1OOCX electron microscope under an accelerating voltage of 80 kV or a JEOL 1200EXII microscope with an accelerating voltage of 120 kV. The magnification was calibrated with Tobacco Mosaic Virus particles, using the 23-A pitch of the right-handed l-start helix as a standard. Image Analysis Electron micrographs were scanned with an Eikonix 78199 digital camera (Eikonix Corp.) interfaced to a VAX 3200 computer. Fourier transforms and other computations were accelerated with a Warrior array processor (Sky Computers). The procedures used in the analysis of BecA filaments are as described in Egelman and Stasiak (1986, 1988), Egelman and Yu (1989), and Yu and Egelman (1990). To assess the confidence of the layer line information, we have used an analog of the crystallographic Figure of Merit (FOM). For each value of R on each layer line we compute, for N filaments: N
x FOM,I(R)
IG:,(R)Icos(<~,~(R)>
= ‘=*
- &(RN
N
x l’%OR)I i=l
where Im[G,,,XR)I
The alignment of individual helices against an average, and of the 36K average against the 38K average, employed a search over the polarity, rotation, and translation of each layer line set that yielded the smallest amplitude-weighted phase residual against the reference layer line set. Additional free parameters, such as radial scaling and tilt, were not used. In generating difference maps, we defined the differences in terms of the significance, which is the difference at every point divided by the standard error of the difference. To define the standard error of the difference, we first need to define the standard error of the mean at each pixel in an averaged map:
SEMii
=
[(n - l)nF
’
where ptj is the density at point ij in the kth map. The standard error at every point in a difference map is then a$! = [(SEM$’
+ (SEM&)2]“2.
RESULTS
Cleavage of a C-Terminus
Fragment
from RecA
Rusche et al. (1985) and Benedict and Kowalczykowski (1988) have described the properties of major fragments of the RecA protein resulting from a cleavage of the C-terminus. Motivated by these studies, we have tried to prepare a proteolytic fragment of RecA that would allow us to directly identify the location of the C-terminus in the RecA structure. We have previously shown (Egelman and Yu, 1989) that a single strand of DNA, which accounts for about 2.5% of the total mass of the R.ecA-DNA filament, can be directly visualized in the RecA-DNA complex through the use of extensive averaging and difference maps between RecA-dsDNA and RecAssDNA filaments. It seemed reasonable that a loss of 18 residues, or 5% of the total mass of the complex, should be even more readily seen in a threedimensional reconstruction of the RecA filament. After trying numerous time-temperatureconcentration combinations with subtilisin, we found a set of conditions where a single clean Escherichia coli RecA fragment was produced. This fragment ran at an apparent molecular weight of about 36K (Fig. 1). From heavily loaded gels, we have determined that the purity of the wild-type RecA is about 99.6%, since the one additional band that can be seen is less than 0.4% of the RecA band in intensity (data not shown). We see no additional bands that were not present in the wild-type protein on gels of the cleaved protein. Sequencing of the first 10 N-terminus residues yielded the sequence: AlaI&Asp-Glu-Asn-Lys-Gin-Lys-Ala-Leu. This sequence is identical with that of the N-terminus of the wild-type RecA protein (Sancar et al., 1980) showing that the proteolysis was indeed from the C-terminus. Since the previous studies of truncated RecA pro1
2
3
4
5
6
FIG. 1. An SDS-PAGE gel with wild-type BecA protein (lanes 2, 4, 6) and a BecA proteolytic fragment (lanes 3, 5). Protein molecular weight standards (17,22,29,45,66, and 97 kDa) have been run in lane 1. The shift in bands between the wild-type and the proteolytic fragment corresponds to a shiR from about 38K to about 36K molecular weight.
CONFORMATIONAL
CHANGE
IN RecA-DNA
FILAMENT
245
teins showed that the molecule retained nearly full function, and since the active form of the RecA molecule in every instance where it has been studied is a polymer, we expected that the proteolytic fragment we had obtained would be polymerization competent. We found this to be true, and were able to form RecA filaments under a variety of different conditions. We have previously reported (Yu and Egelman, 1990) that a significant improvement in the helical order in RecA-DNA filaments can be obtained with the stabilization of a ReCA-ADP-Pi state through the use of an analog for Pi, AIF,. These observations were motivated by the report of Moreau and Carlier (1989) that the complex of RecA-ADP-AlF, was active in cleaving the LexA repressor. This improvement in order occurs with respect to previous structural work on the RecA-ATP-y-S complex, and is consistent with observations that ATP-Y-S is hydrolyzed by RecA at a rate that is significant for structural studies (Craig and Roberts, 1981; Yu and Egelman, 1991). Since we were able to form filaments of the 36K RecA on double stranded DNA with ATP and AlF, (Fig. 2), we used these filaments for further study since they were more ordered than the filaments formed from the RecA fragment with DNA and ATPq-S. In an effort to determine how stable the RecA-dsDNA-ADP-AlF; filaments were to further digestion, these filaments were incubated with subtilisin for 24 hr at 4°C. No further digestion was seen as assayed by SDS gels (data not shown).
FIG. 2. Electron micrographs of filaments formed fi-om the 36K RecA proteolytic fragment (two filaments on the left) and filaments formed from the wild-type RecA protein (two filaments on the right). All of the filaments are formed on double stranded DNA, with a nucleotide cofactor of ADP-AIF,, an analog for ADPP,. The filaments are approximately 120 A in diameter, and have a characteristic pitch in negative stain of about 95 A for the fragment and 91 A for the intact F&A.
huge Analysis Approximately 50 images of filament sections (of the 36K RecA on dsDNA with ADP-AlF;) were scanned, corrected for curvature (Egelman, 1986) and Fourier transformed. All of the filament sections transformed showed a very strong layer line at about l/95 A (n = 1) arising from the right-handed 95-A pitch RecA helix. From this sample, 16 filaments were found which displayed a weaker, but fairly symmetrical, layer line at a spacing of about l/81 A. The mean length of the 16-filament sections used was 2300 A, with a minimum length of 1400 A and a maximum of 3100 A. The Fourier transform of one of these 16 filaments is shown in Fig. 3. Among these 16 filaments, a number displayed additional symmetrical layer lines at spacings of about l/570, l/114, l/52, and 1144 A. The indexing of these layer lines is the same as that for the intact 38K RecA, and is described in Yu and Egelman (1990). The possibility that the indexing is completely different for the 36K fragment is remote, given the exceedingly close agreement in layer line intensities between the two states (Fig. 5). The spacings of the observed layer lines allows
one to determine the average twist of each filament section to fairly high precision. Figure 4 shows the data (for both pitch and twist) from these 16 filaments in a comparison with 16 intact 38K RecA filaments (Yu and Egelman, 1990). While the twist of these two samples appears to be about the same (6.174 RecA subunits/turn), there is a clear shift in the pitch between the intact protein and the 36K fragment, with the mean going from about 91 A for the intact protein to about 95 A for the fragment. This would correspond to a change in the rise per subunit within the filament from about 14.7 A with the intact RecA to about 15.4 A with the 36K fragment. There is also a greater dispersion of the pitch data for the 36K filaments, which might reflect a greater flexibility of this filament with respect to compression and extension. It must be kept in mind, however, that both filaments are being examined in a dehydrated and stained state, and changes in the structure due to both the dehydration and staining could cause these values to shift from what they might be in solution. However, direct measurements of the pitch and twist of wild-type RecA-dsDNA-
246
YU AND EGELMAN
C E,96
I
I I
I I
6.12
6.15
6.16
I I
I
6.21
Units/Turn
FIG. 3. The computed Fourier transform of a filament of the 36K BecA proteolytic fragment. The strongest layer lines are labeled as 0, 6, ‘7, and 12. The sixth layer line is at a spacing of about l/95 A, and arises from the right-handed one-start helix of the BecA filament.
ATP-r-S filaments in amorphous ice shows that these values are virtually the same as in negative stain (Yu and Egelman, manuscript in preparation). It is of interest that a similar correlation exists between the pitch and the twist for the 36K filaments that was noted for intact RecA (Yu and Egelman, 1990). This shows that a screw motion must exist within the RecA filament that couples the axial compression or extension of subunits with their rotation. This will have bearing on the interpretation of the relation between the arrangement of subunits on a 6, screw axis in crystals of RecA (McKay et al., 1980) and in the RecA-DNA filament. Since crystallographic packing demands an integral number, 6, of subunits per turn of a filament, one can extrapolate the plots of Fig. 4 from the observed values of units/turn down to 6.0 units/turn. The linear regression line for the intact 38K RecA would yield a pitch of 79.3 A for this twist, while the 36K RecA fragment would yield a pitch of 87.6 A. It is perhaps not coincidental that the observed pitch in the RecA crystal is 82.0 A (McKay et al., 19801, suggesting that the subunits in the crystal may be simply related to the subunits in the filament on DNA by just such a screw motion. Since this screw motion appears to be continuous in the RecA filaments (Fig. 4), we have averaged together layer line data from
FIG. 4. The spacing of layer lines in the transform of filaments can be used to determine the twist of the filaments (in units/turn of the helix). Because the twist is given by a ratio of layer line spacings, it is independent of magnification. From the indexing of the transform, the number of units per turn of the 95-A pitch helix will be equal to 6 + 2,/Z,, where 2, and 2s are the axial spacings of the “1st” and “6th” layer lines, respectively. The value used for 2, was actually 2, - Z,, due to the fact that the sixth and seventh layer lines could be measured with greater precision than the first layer line. The determination of the pitch is magnification dependent, and the 1/23-A layer line of tobacco mosaic virus (coprepared with the BecA samples) was used as an internal standard. Data are shown for both the 36K fragment (circles) and the intact BecA filaments (triangles, data from Yu and Egelman, 1990). The distribution of twist is nearly identical for the two different populations, but there is a shift in pitch. The twist for the 38K state is 6.174 k 0.005 (standard error of the mean) units/turn, while it is 6.174 t 0.008 units/turn for the 36K fragment. The mean pitch of the intact BecA is 90.8 + 0.5 A (standard error of the mean), while the mean pitch of the 36K fragment is 94.9 2 1.0 A (standard error of the mean). A linear regression line is displayed for each population. It was shown in Yu and Egelman (1990) that a significant correlation exists between the twist and the pitch.
filaments having slightly different pitch and different twist. The effect of this will be to elastically stretch or compress and twist or untwist the density of individual filaments in the three-dimensional reconstruction from the mean values that the individual filaments possess. A preferable method would be to collect one or two orders of magnitude more data, generate subsets of filaments that have a common pitch and twist, and reconstruct these subsets separately. This is beyond the scope of the present paper. Since small changes in subunit conformation or orientation can lead to large changes in twist and pitch, and since the spread of data appears to be across a continuum, we feel that this averaging will probably not introduce large errors at the resolution that we are using. The amplitudes and phases from the averaged layer line set are shown in Fig. 5. A comparison with the averaged layer lines of the intact 38K filament
CONFORMATIONAL
CHANGE
(Yu and Egelman, 1990) does not reveal any large differences localized to a single layer line. The amplitude-weighted phase difference between these two averages is about 44”. The quality of the data, as judged by how well individual filaments average together, is shown in Fig. 6, where the phase residuals between individual filaments and the average data set are shown. It can be seen that the filaments average together quite well, with a clear polarity. The quality of the layer line data can also be described by the FOM (see Methods). Not surprisingly, the strongest peak (on the sixth layer line) has a FOM of about 0.98, and the weakest data fall off to a FOM of about 0.4. The secondary maximum on the sixth layer line (R = 0.035, or a resolution of about 27 A from the origin of the transform) has a FOM of 0.58, suggesting that although the data are noisy, they are still significantly correlated among the 16 filaments used. Since this secondary peak is only about 4% of the amplitude of the main peak, it is also apparent that weak data at higher resolutions are not merely noise. Three-Dimensional Comparison with Intact RecA The layer lines of Fig. 5 allow one to compute a three-dimensional reconstruction of the 36K RecA fragment, and this is shown in Fig. 7. The confidence with which one can interpret features in this reconstruction is very much dependent upon an estimate of the error or variance in the map, and we have displayed the standard error of the mean at every point in the map along with the reconstruction in Fig. 7. The cylindrically and axially averaged standard error of the mean is shown as a function of radius in Fig. 8, and it can be seen that the error is fairly flat across the radius of the filament, with an average of about 2.25 density units. Since the contour steps in Fig. 7 have a spacing of 1.3 density units, the standard error in general is less than the spacing of two contour lines. In comparison, an equivalent plot for the intact 38K RecA filament (from Yu and Egelman, 1990) shows a greater radial modulation, with an average value that is lower than that for the 36K filaments. The peak standard error for the 38K filaments occurs at a radius of about 25 A, which corresponds to the inner lobe of the RecA subunit. An alternate way of analyzing the error, noise, or variance in the map is to compare the map (Fig. 7) generated by the layer lines of Fig. 5 with a map generated from FOM-weighted layer lines (see Methods). We have done this, and the resulting map is extremely similar to that shown in Fig. 7 (data not shown), suggesting that the noise or variance in the layer lines is present with weak amplitudes, thus appearing weakly in the map. Since we have reconstructions of both the intact 38K RecA and the 36K fragment, and have maps of
IN RecA-DNA
FILAMENT
247
the standard error of the mean in each of these two reconstructions, we are in a position to do a quantitative comparison between the two structures. The key question in making a comparison between the two averaged maps is: what is the statistical significance of the differences that are seen? Due to noise, variation in staining, etc., any two sets of RecA filaments (even drawn from the same population) will display differences. To quantify the meaning of the differences that are seen, we have displayed the differences in terms of their statistical significance in Fig. 9, where the differences between the two maps have been divided by the standard error of the difference at every point. The resulting map has values not of density units but of dimensionless standard deviations. Because the absolute contrast within images depends upon staining, the amount of exposure to the electron beam, development of the negatives, densitometry, etc., we have scaled the two sets of images together for comparison purposes by using a least-squares search for both a multiplicative and additive scale factor. The factors that were found were small (changing the maps by less than 10%) and did not significantly affect the differences that were observed. If the structure of the 36K fragment was unchanged from that of the intact 38K protein, one would expect to see the largest difference in a region where the 38K map density was greater than that of the 36K. This region would correspond to the location of the C-terminus. Instead, what can be seen in Fig. 9 is that the largest single difference peak corresponds to a region where the 36K density is actually greater than the 38K density. This peak difference has a value of 5.0 standard deviations. The largest peak where the 38K density is greater than the 36K density has a value of 3.9 standard deviations (labeled as No. 2 in Fig. 91, and actually lies outside of the region contoured as protein for both the 36K and 38K filaments. There are no other peaks greater than 3.6 standard deviations, but there are three additional peaks with values between 3.0 and 3.6 standard deviations where the 38K density is greater than the 36K density (these are labeled as 1, 3, and 4 in Fig. 9). Two of these, 3 and 4, are on the edge of the density contoured as protein, while the other two, 1 and 2, fall outside of this region. It is of interest to estimate what the probability is for seeing a difference greater than three standard deviations that is entirely due to noise. For a normal distribution, the percentage of the distribution greater than 3 standard deviations from the mean is 0.26%. All of the maps were calculated on a 2 A per pixel grid, extending radially to 80 A. There are thus about 5000 pixels within each section. If the density at every pixel were independent, we would therefore expect about 13 peaks
248
YU AND EGELMAN rcc
. .- ‘icu
L=13
N=-4
L=12
N=2
L=7
N=-5
uo 00
-200.0 .
PO.0
Loo.0
LLl 0.w 0.010o.olso&a odeso&m 0.a 0.m 0.w FIG. 5. The amplitudes and phases for the averaged set of seven layer lines from 16 filaments of the 36K RecA fragment (left), and the corresponding layer lines from a sample of 15 filaments of intact 38K RecA (right, from Yu and Egelman, 1990). The two independent data sets fkom each individual filament, corresponding to the near and far sides of the particle, were first internally averaged before filaments were averaged together. Therefore, the total average on the letI actually contains 32 independent data sets.
greater than 3 standard deviations due solely to noise, and not any structural difference. With a finite resolution, however, the density at each pixel is not independent, and one would expect the number
of observed peaks due to noise to be significantly less than 13. Without making a rigorous analysis of the error, it is reasonable to assume that the two additional peaks greater than 3 standard deviations, lo-
CONFORMATIONAL
CHANGE
IN RecA-DNA
249
FILAMENT
40 loo L=5
N=7 - mo.0 - loo.0 c
1
*
loo a
*
IOD
I **-
L=l
N=-6
L=O
N=O
-00 ,
0
FIG. Eontinued
cated outside of the putative protein boundary, are consistent with noise. On the other hand, the probability (for a normal distribution) of a value being 5.0 or greater standard deviations from the mean is on the order of 6 x 10 - 7, and it is sensible to assume
that the largest difference peak is highly significant, and extremely unlikely to have arisen from random differences between two samples from the same population. Figure 9 thus shows a very clear shift in mass
250
YU AND EGELMAN
9ov 90
’ 75
I 60
I 45
I 30
I 15
I 0
B&or Phatr Realdual FIG. 6. The total amplitude-weighted phase residual between each of the 16 layer line sets used in the average and the averaged layer line set itself (Fig. 51, as a function of the polarity of each individual set against the averaged data set. This plot can serve as a measure of the intrinsic polarity of the data. For a nonpolar structure (or for highly noisy data) there would be no significant difference between the two different orientations of each individual data set, and all points would fall near the diagonal line. The further points are from the diagonal, the greater the polarity. The centroid of the distribution is 38” for the better phase residual, and 56” for the opposite orientation.
between the 36K and 38K filaments that appears to be mainly localized to one region. We have interpreted this shift as a movement of the inner lobe of the RecA subunit through a distance of about 15 A in a direction that is radially outward (see arrows in Fig. 9). Limited resolution, however, prevents one from distinguishing between the simple movement of one lobe and a more complicated conformational change within the RecA subunit or within the interface between subunits. DISCUSSION
This study was begun with the expectation that the cleavage of the C-terminus from RecA protein would allow us to identify the location of the Cterminus in a three-dimensional reconstruction. We have previously shown (Egelman and Yu, 1989) that sensitive difference measurements were capable of visualizing a single strand of DNA in the RecADNA complex, accounting for only about 2.5% of the mass of the complex. We expected, therefore, that the approximately 5% of the molecule that we had
cleaved in this study would clearly appear as a region of missing mass in a comparison of the fragment with the intact protein. Instead, what we have seen is a rearrangement of the internal density in the fragment, consistent with a movement of the inner lobe of the RecA protomer. We have shown (Yu and Egelman, 1990) that the RecA protomer in the ReCA-ADP-Pi-DNA filament is visualized at low resolution as containing two domains or lobes. We also showed that the region of largest variance within a sample of 15 filaments used for that reconstruction fell on the inner domain. RecA has been shown to be extremely flexible, to have a variable pitch and a variable twist (Egelman and Stasiak, 1986). Different conformations of RecA subunits have also been directly visualized in bundles of RecA filaments (Egelman and Stasiak, 1988; Yu and Egelman, 1991). The observation of a large variance associated with the inner domain in the RecA-ADP-Pi-DNA filament provides a more detailed understanding of the internal mechanics of the RecA filament, with the potential for understanding how the filament can be so easily bent, twisted, and extended. Our observation, reported here, of a significant movement of the inner lobe through about 15 A following the removal of about 18 C-terminal residues, extends the previous interpretations about internal mechanics. It appears that the inner lobe in the wild-type filament is variable in position, and that the cleavage of the C-terminus forces this lobe outward. In the 36K filament this lobe, that was formerly at an inner radius, is now relatively more fixed, as judged by the variance map for this state (Fig. 7). A greater variance is now associated with the outer lobe. The visualization of conformational changes within the RecA filament is interesting in and of itself, and shows that protein polymers have the ability to undergo such conformational changes while still maintaining the noncovalent proteinprotein interface necessary for the maintenance of polymer integrity (Egelman and DeRosier, 1991). The key task, however, must be to use these observations to advance our understanding of RecA function. Fortunately, data already exist on the behavior of RecA fragments. There have been four previous studies of small C-terminal deletions of the RecA
FIG. 7. The layer lines of Fig. 5 have been used to generate a three-dimensional reconstruction of the 36K fragment BecA filament. Three sections through the filament are shown, corres onding to infinitesimally thin slices normal to the filament axis, which is indicated by a cross in each section. The sections are spaced 5 1 apart. Because the axial rise per subunit is about 15 A, and the rotation between subunits is 360”/(37/6), or about 584, the section at z = 15 a will be almost identical with the section at z = 0 A, but rotated by 58.4”. The density levels in the averaged structure are indicated by the contour lines, while the standard error of the mean at every point in the average density is given by the continuous grey scale. The sections have been contoured in steps of 1.3 arbitrary density units, starting at 40, with a peak density of 53. The SEMs are displayed with all values less than 2 density units being white, with the peak value (about 3.6) being black. All values between 2 and 3.6 are linearly represented by a grey value between white and black.
CONFORMATIONAL
CHANGE
IN &CA-DNA
FILAMENT
251
252
YU AND EGELMAN
2 2.0
i ;
s 0 1.5
1.0
0
20 Radius
40 (Angstroms)
60
FIG. 8. The cylindrically and axially averaged standard error of the mean at every point in the reconstruction is shown as a function of the radius from the helical axis. Data from the 36K fragment is shown with the solid line, while for purposes of comparison the data from intact RecA filaments (Yu and Egelman, 1990) is shown with a dashed line. The SEM for the 36K fragment tends to be more uniformly distributed across the filament than in the intact RecA, where there is a significant peak in the SEM at about 25 A radius, corresponding to the inner lobe of the RecA subunit.
protein. Two of these (Rusche et al., 1985; Benedict and Kowalczykowski, 1988) were based upon proteolysis, while the other two (Yarranton and Sedgwick, 1982; Larminat and Defais, 1989) used truncated genes. Care must be exercised in directly comparing the results of these studies due to differences in the amount of truncation, the assays employed, and even the species of bacteria used. Rusche et al. made a subtilisin digestion of both the E. coli and P. mirubilis RecA proteins, but were only able to obtain a large, stable fragment from the P. mirubilis protein. The P. mirubilis RecA fragment that they studied had an apparent molecular weight of 36 000, very similar to our E. coli RecA fragment. Rusche et al. noted that the P. mirubilis fragment interacted in the same way as the intact RecA with single
stranded DNA, but had different interactions with double stranded DNA. Specifically, the fragment appeared to bind duplex DNA with a greater affinity than the intact protein. Benedict and Kowalczykowski (1988) described the properties of an E. coli RecA fragment, missing about 45 residues at the C-terminus, that was fortuitously found after storing a sample of RecA at 4°C for 6 months. Like Rusche et al., they found that this fragment had unaltered single stranded DNA binding properties, but enhanced binding to double stranded DNA. Larminat and Defais (1989) used genetic constructs to express in E. coli a RecA fragment missing 17 C-terminus residues. They found that this fragment was nearly identical in uiuo to the wild-type protein in inducing the SOS response (cleaving the LexA repressor), performing recombination, and inducing mutagenesis. However, they observed a negative complementation when the RecA fragment was present in cells along with the intact wild-type protein. Yarranton and Sedgwick (1982) also employed genetic constructs to express a RecA protein missing about 75 C-terminus residues. They also observed a negative complementation effect with regard to recombination when this protein was expressed along with wild-type, but they observed no effect on SOS induction or phage h induction. At this point, structural observations of rearrangements of mass within the RecA filament as a result of the C-terminus deletion cannot be directly related to the functional studies that have been made on similar fragments. However, together these observations will add considerably to the interpretation of an x-ray crystal structure of RecA that is soon to be available (McKay et al., 1980; T. Steitz, personal communication). We now know that the deletion of the RecA C-terminus must affect double stranded DNA binding through a conformational change that moves part or all of the inner lobe of the RecA protein through about 15 A.
FIG. 9. A comparison between the reconstruction of the 36K fragment (left side) and intact 38K RecA (right side, from Yu and Egelman, 1990). The density distributions are shown with contour lines for sections at three different axial spacings, spaced 5 A apart. The significance of the differences between each of the two pairs is given by the difference divided by the standard error of the difference, and this is shown by the continuous grey scale density superimposed on both sets of contour lines. The grey scale on the left (superimposed on the 36K map) shows regions where the density of the 38K filament is greater than that of the 36K, while the grey scale on the right (superimposed on the 38K may) shows regions where the density of the 36K filament is greater than that of the 38K. The arrows on the right indicate the movement of the inner lobe that is most likely responsible for the large difference seen there. The arrow in the section at 0 A corresponds to the motion of one subunit, while the arrows in the sections at 5 and 10 A correspond to the motions within the next subunit above that one in the helix. Although the rise per subunit in the filament is about 15 A, each subunit extends axially within the filament about 45 A (Yu and Egelman, 1990). The grey scale data has been scaled such that values below 2.3 standard deviations will be white, while the peak standard deviation, 5.0, is black. All values between 2.3 and 5.0 are linearly mapped onto a grey scale between white and black. The largest difference, at 5.0 standard deviations, is at the end of the arrow in the section at z = 10 A. The corresponding peaks at the ends of the arrows in the sections at 0 and 5 A are 4.4 standard deviations. The largest difference where the 38K density is greater than the 36K has a value of 3.9 standard deviations, and is labeled 2 in the section at 10 A. The only other peaks greater than 3.0 standard deviations are labeled 1, 3, and 4. Peak 4 extends into the density outlined as protein, and peak 3 is at the edge of that density. Peaks 1 and 2, however, appear to lie significantly away from the protein boundary, and are consistent with noise (see text).
1
ww
254
YU AND EGELMAN
This research was supported by NIH Grant GM35269 NSF Grant DMB 87-12075.
and by
REFERENCES
BENEDICT, R. C., AND KOWALCZYKOWSKI, S. C. (1988) J. Biol. Chem. 263, 15513-15520. Cox, M. M., AND LEHMAN, I. R. (1987) Annu. Rev. Biochem. 56, 229-262. CRAIG, N. L., AND ROBERTS, J. W. (1981) J. Biol. Chem. 256, 8039-8044. EGELMAN, E. H. (1986) Ultramicroscopy 19, 367-374. EGELMAN, E. H., AND DEROSIER, D. J. (1991) J. Mol. Biol. 217, 4OWO8. EGELMAN, E. H., AND STASIAK, A. (1986) J. Mol. Biol. 191, 677697.
EGELMAN, E. H., AND STASIAK, A. (1988) J. Mol. Biol. 200, 329349. EGELMAN, E. H., AND Yu, X. (1989) Science 245, 404407. GRIFFITH, J., AND SHORES, C. G. (1985) Biochemistry 24,158162.
HOWARD-FLANDERS, P., WEST, S. C., AND STASIAK, A. (1984) NUture 309, 215-220. LARMINAT, F., AND DEFAIS, M. (1989) Mol. Gen. Genetics 216, 106-112. MCKAY, D. B., STEITZ, T. A., WEBER, I. T., WEST, S. C., AND HOWARD-FLANDERS, P. (1980) J. Biol. Chem. 255,6662. MOREAU, P. L., AND CARLIER, M.-F. (1989) J. Biol. Chem. 264, 2302-2306. RADDING, C. M. (1981) Cell 25, 34. RUSCHE, J. R., KONISBERG, W., AND HOWARD-FLANDERS, P. (1985) J. Biol. Chem. 260, 949-955. SANCAR, A., STACHELEK, C., KONIGSBERG, W., AND RUPP, W. D. (1980) Proc. Natl. Acad. Sci. USA 77, 2611-2615. WEST, S. C., CASSUTO, E., AND HOWARD-FLANDERS, P. (1982) Mol. Gen. Genet. 187, 209-217. YARRANTON, G. T., AND SEDGWICK, S. G. (1982) Mol. Gen. Genet. 185, 99-104. Yu, X., AND EGELMAN, E. H. (1990) Biophys. J. 57, 555-566. Yu, X., AND EGELMAN, E. H. (1991), submitted for publication.
OF STRUCTURAL
BIOLOGY
106,
243-254
(1991)
Removal of the RecA C-terminus Results in a Conformational Change in the RecA-DNA Filament X. Yu AND E. H. EGELMAN Department
of Cell Biology and Neuroanutomy, University of Minnesota Medical School, 321 Church Street SE, Minneapolis, Minnesota 55455 Received December 18, 1990, and in revised form March 21, 1991
can be visualized within the filament as containing two lobes or domains (Yu and Egelman, 1990). That study was based upon the analysis of RecA filaments formed on double-stranded DNA (dsDNA) with a complex of ADP and AlF, as a cofactor. We have now obtained a three-dimensional reconstruction of a proteolytically truncated RecA, with about 18 residues cleaved from the C-terminus. While the initial expectation was that such a cleaved molecule would allow us to visualize the C-terminus in a comparison with the intact molecule, we have found instead that the loss of the C-terminus results in a significant conformational change with the RecA filament. We suggest that this conformational change can be related to motions within the wild-type RecA filament, and that the ability to visualize such changes will be important in ultimately constructing a dynamic atomic model for the RecA-mediated strand exchange.
The Escherichia coli RecA protein catalyzes homologous recombination of DNA molecules, and the active form of the protein is a helical polymer that it forms around DNA. Previous image analysis of electron micrographs has revealed the RecA protein to be organized into two domains or lobes within the RecA-DNA filament. We have now been able to show that a small modification of the RecA protein by proteolysis results in a significant shift in the internal mass in the RecA filament. We have cleaved approximately 18 residues from the C-terminus of the RecA protein, producing a roughly 36K MW RecA core protein that binds DNA and polymerizes normally. A three-dimensional reconstruction of this complex has been computed, and has been compared with a previous reconstruction of the intact protein. The main difference is consistent with a 15 A outward movement of the lobe that was at an inner radius in the wild-type protein. These observations yield additional evidence about the conformational flexibility of the RecA filament, and will aid in understanding the structural mechanics and dynamics of the RecA filament. o 1991 Academic PRESS,IIIC.
MATERIALS AND METHODS Proteins and Nucleic Acids RecA was purified by the method of West et al. (19821, with an additional polyspermidine precipitation, as described by Griffith and Shores (1985). The purified RecA ran as a single band on an SDS-PAGE Coomassie blue stained gel. Double stranded, circular +X174 was a gift from Dr. Paul Howard-Flanders. The DNA was linearized by restriction endonuclease PstI (IBI), and ran as a single band on an agarose gel. The concentration of the DNA was measured using an extinction coefficient Azw of 1 for 50 pg/ml and a l-cm path length. Reactions of RecA and DNA were carried out at 37°C in 25 mM Triethanolamine-HCl buffer (pH 7.2). RecA concentration was 7 pM. The Reckdouble stranded DNA ratio was equal to 4O:l (w:w) or a 1:3 molar ratio of RecA to total nucleotides. This is a twofold excess of RecA, given that each RecA protomer binds three base pairs, or six nucleotides. ATP (Sigma) concentration was 2.5 mM. Aluminum nitrate and sodium fluoride (both from Baker) were added to 2.5 mM in some experiments, while in others the aluminum nitrate concentration was 100 pi%f and the sodium fluoride was 2.5 mM. No differences were seen between these two sets of conditions.
INTRODUCTION
The RecA protein is the most intensively studied enzyme of general genetic recombination (for reviews, see Radding, 1981; Howard-Flanders et al., 1984; Cox and Lehman, 1987). One of the most remarkable features about RecA is how a 38K MW protein can mediate the search for homology between two different DNA molecules, promote a strand exchange reaction between these molecules, and cleave its own repressor, among other activities. An understanding of RecA function must begin with the knowledge that the active form of the RecA protein is a filamentous complex containing the RecA protein, a polynucleotide substrate (usually DNA), and a nucleotide triphosphate cofactor (usually ATP). The complex has a stoichiometry of one RecA subunit per ATP per three bases or base pairs. We have recently shown that the RecA protein
Proteolytic Digestion of RecA RecA protein was digested with subtilisin (Sigma) at 4°C for 20 hr. The RecA concentration was 7 pM and the subtilisin concen243 1047~8477/91 $3.00 Copyright 0 1991 by Academic Preee, Inc. All rights of reproduction in any form reserved.
244
YU AND EGELMAN
tration was 0.23 p&f, and the digestion was in a 12 mM Tris buifer at pH 7.8. The BecA was purified after the digestion on a Bio-Gel A-O.5 m gel column (Bio-Bad) in the 12 mM Tris buffer at pH 7.8. An Edman’s degradation was performed on the cleaved BecA by the Microchemical Facility, Department of Human Genetics of the University of Minnesota. The first 10 N-terminal residues were sequenced. Electron Microscopy Samples were prepared on carbon-coated copper grids which had been glow discharged. After a sample was placed on the grid, excess solution was removed by blotting with filter paper. Then the samples were negatively stained with 2% uranyl acetate and viewed with either a JEOL 1OOCX electron microscope under an accelerating voltage of 80 kV or a JEOL 1200EXII microscope with an accelerating voltage of 120 kV. The magnification was calibrated with Tobacco Mosaic Virus particles, using the 23-A pitch of the right-handed l-start helix as a standard. Image Analysis Electron micrographs were scanned with an Eikonix 78199 digital camera (Eikonix Corp.) interfaced to a VAX 3200 computer. Fourier transforms and other computations were accelerated with a Warrior array processor (Sky Computers). The procedures used in the analysis of BecA filaments are as described in Egelman and Stasiak (1986, 1988), Egelman and Yu (1989), and Yu and Egelman (1990). To assess the confidence of the layer line information, we have used an analog of the crystallographic Figure of Merit (FOM). For each value of R on each layer line we compute, for N filaments: N
x FOM,I(R)
IG:,(R)Icos(<~,~(R)>
= ‘=*
- &(RN
N
x l’%OR)I i=l
where Im[G,,,XR)I
The alignment of individual helices against an average, and of the 36K average against the 38K average, employed a search over the polarity, rotation, and translation of each layer line set that yielded the smallest amplitude-weighted phase residual against the reference layer line set. Additional free parameters, such as radial scaling and tilt, were not used. In generating difference maps, we defined the differences in terms of the significance, which is the difference at every point divided by the standard error of the difference. To define the standard error of the difference, we first need to define the standard error of the mean at each pixel in an averaged map:
SEMii
=
[(n - l)nF
’
where ptj is the density at point ij in the kth map. The standard error at every point in a difference map is then a$! = [(SEM$’
+ (SEM&)2]“2.
RESULTS
Cleavage of a C-Terminus
Fragment
from RecA
Rusche et al. (1985) and Benedict and Kowalczykowski (1988) have described the properties of major fragments of the RecA protein resulting from a cleavage of the C-terminus. Motivated by these studies, we have tried to prepare a proteolytic fragment of RecA that would allow us to directly identify the location of the C-terminus in the RecA structure. We have previously shown (Egelman and Yu, 1989) that a single strand of DNA, which accounts for about 2.5% of the total mass of the R.ecA-DNA filament, can be directly visualized in the RecA-DNA complex through the use of extensive averaging and difference maps between RecA-dsDNA and RecAssDNA filaments. It seemed reasonable that a loss of 18 residues, or 5% of the total mass of the complex, should be even more readily seen in a threedimensional reconstruction of the RecA filament. After trying numerous time-temperatureconcentration combinations with subtilisin, we found a set of conditions where a single clean Escherichia coli RecA fragment was produced. This fragment ran at an apparent molecular weight of about 36K (Fig. 1). From heavily loaded gels, we have determined that the purity of the wild-type RecA is about 99.6%, since the one additional band that can be seen is less than 0.4% of the RecA band in intensity (data not shown). We see no additional bands that were not present in the wild-type protein on gels of the cleaved protein. Sequencing of the first 10 N-terminus residues yielded the sequence: AlaI&Asp-Glu-Asn-Lys-Gin-Lys-Ala-Leu. This sequence is identical with that of the N-terminus of the wild-type RecA protein (Sancar et al., 1980) showing that the proteolysis was indeed from the C-terminus. Since the previous studies of truncated RecA pro1
2
3
4
5
6
FIG. 1. An SDS-PAGE gel with wild-type BecA protein (lanes 2, 4, 6) and a BecA proteolytic fragment (lanes 3, 5). Protein molecular weight standards (17,22,29,45,66, and 97 kDa) have been run in lane 1. The shift in bands between the wild-type and the proteolytic fragment corresponds to a shiR from about 38K to about 36K molecular weight.
CONFORMATIONAL
CHANGE
IN RecA-DNA
FILAMENT
245
teins showed that the molecule retained nearly full function, and since the active form of the RecA molecule in every instance where it has been studied is a polymer, we expected that the proteolytic fragment we had obtained would be polymerization competent. We found this to be true, and were able to form RecA filaments under a variety of different conditions. We have previously reported (Yu and Egelman, 1990) that a significant improvement in the helical order in RecA-DNA filaments can be obtained with the stabilization of a ReCA-ADP-Pi state through the use of an analog for Pi, AIF,. These observations were motivated by the report of Moreau and Carlier (1989) that the complex of RecA-ADP-AlF, was active in cleaving the LexA repressor. This improvement in order occurs with respect to previous structural work on the RecA-ATP-y-S complex, and is consistent with observations that ATP-Y-S is hydrolyzed by RecA at a rate that is significant for structural studies (Craig and Roberts, 1981; Yu and Egelman, 1991). Since we were able to form filaments of the 36K RecA on double stranded DNA with ATP and AlF, (Fig. 2), we used these filaments for further study since they were more ordered than the filaments formed from the RecA fragment with DNA and ATPq-S. In an effort to determine how stable the RecA-dsDNA-ADP-AlF; filaments were to further digestion, these filaments were incubated with subtilisin for 24 hr at 4°C. No further digestion was seen as assayed by SDS gels (data not shown).
FIG. 2. Electron micrographs of filaments formed fi-om the 36K RecA proteolytic fragment (two filaments on the left) and filaments formed from the wild-type RecA protein (two filaments on the right). All of the filaments are formed on double stranded DNA, with a nucleotide cofactor of ADP-AIF,, an analog for ADPP,. The filaments are approximately 120 A in diameter, and have a characteristic pitch in negative stain of about 95 A for the fragment and 91 A for the intact F&A.
huge Analysis Approximately 50 images of filament sections (of the 36K RecA on dsDNA with ADP-AlF;) were scanned, corrected for curvature (Egelman, 1986) and Fourier transformed. All of the filament sections transformed showed a very strong layer line at about l/95 A (n = 1) arising from the right-handed 95-A pitch RecA helix. From this sample, 16 filaments were found which displayed a weaker, but fairly symmetrical, layer line at a spacing of about l/81 A. The mean length of the 16-filament sections used was 2300 A, with a minimum length of 1400 A and a maximum of 3100 A. The Fourier transform of one of these 16 filaments is shown in Fig. 3. Among these 16 filaments, a number displayed additional symmetrical layer lines at spacings of about l/570, l/114, l/52, and 1144 A. The indexing of these layer lines is the same as that for the intact 38K RecA, and is described in Yu and Egelman (1990). The possibility that the indexing is completely different for the 36K fragment is remote, given the exceedingly close agreement in layer line intensities between the two states (Fig. 5). The spacings of the observed layer lines allows
one to determine the average twist of each filament section to fairly high precision. Figure 4 shows the data (for both pitch and twist) from these 16 filaments in a comparison with 16 intact 38K RecA filaments (Yu and Egelman, 1990). While the twist of these two samples appears to be about the same (6.174 RecA subunits/turn), there is a clear shift in the pitch between the intact protein and the 36K fragment, with the mean going from about 91 A for the intact protein to about 95 A for the fragment. This would correspond to a change in the rise per subunit within the filament from about 14.7 A with the intact RecA to about 15.4 A with the 36K fragment. There is also a greater dispersion of the pitch data for the 36K filaments, which might reflect a greater flexibility of this filament with respect to compression and extension. It must be kept in mind, however, that both filaments are being examined in a dehydrated and stained state, and changes in the structure due to both the dehydration and staining could cause these values to shift from what they might be in solution. However, direct measurements of the pitch and twist of wild-type RecA-dsDNA-
246
YU AND EGELMAN
C E,96
I
I I
I I
6.12
6.15
6.16
I I
I
6.21
Units/Turn
FIG. 3. The computed Fourier transform of a filament of the 36K BecA proteolytic fragment. The strongest layer lines are labeled as 0, 6, ‘7, and 12. The sixth layer line is at a spacing of about l/95 A, and arises from the right-handed one-start helix of the BecA filament.
ATP-r-S filaments in amorphous ice shows that these values are virtually the same as in negative stain (Yu and Egelman, manuscript in preparation). It is of interest that a similar correlation exists between the pitch and the twist for the 36K filaments that was noted for intact RecA (Yu and Egelman, 1990). This shows that a screw motion must exist within the RecA filament that couples the axial compression or extension of subunits with their rotation. This will have bearing on the interpretation of the relation between the arrangement of subunits on a 6, screw axis in crystals of RecA (McKay et al., 1980) and in the RecA-DNA filament. Since crystallographic packing demands an integral number, 6, of subunits per turn of a filament, one can extrapolate the plots of Fig. 4 from the observed values of units/turn down to 6.0 units/turn. The linear regression line for the intact 38K RecA would yield a pitch of 79.3 A for this twist, while the 36K RecA fragment would yield a pitch of 87.6 A. It is perhaps not coincidental that the observed pitch in the RecA crystal is 82.0 A (McKay et al., 19801, suggesting that the subunits in the crystal may be simply related to the subunits in the filament on DNA by just such a screw motion. Since this screw motion appears to be continuous in the RecA filaments (Fig. 4), we have averaged together layer line data from
FIG. 4. The spacing of layer lines in the transform of filaments can be used to determine the twist of the filaments (in units/turn of the helix). Because the twist is given by a ratio of layer line spacings, it is independent of magnification. From the indexing of the transform, the number of units per turn of the 95-A pitch helix will be equal to 6 + 2,/Z,, where 2, and 2s are the axial spacings of the “1st” and “6th” layer lines, respectively. The value used for 2, was actually 2, - Z,, due to the fact that the sixth and seventh layer lines could be measured with greater precision than the first layer line. The determination of the pitch is magnification dependent, and the 1/23-A layer line of tobacco mosaic virus (coprepared with the BecA samples) was used as an internal standard. Data are shown for both the 36K fragment (circles) and the intact BecA filaments (triangles, data from Yu and Egelman, 1990). The distribution of twist is nearly identical for the two different populations, but there is a shift in pitch. The twist for the 38K state is 6.174 k 0.005 (standard error of the mean) units/turn, while it is 6.174 t 0.008 units/turn for the 36K fragment. The mean pitch of the intact BecA is 90.8 + 0.5 A (standard error of the mean), while the mean pitch of the 36K fragment is 94.9 2 1.0 A (standard error of the mean). A linear regression line is displayed for each population. It was shown in Yu and Egelman (1990) that a significant correlation exists between the twist and the pitch.
filaments having slightly different pitch and different twist. The effect of this will be to elastically stretch or compress and twist or untwist the density of individual filaments in the three-dimensional reconstruction from the mean values that the individual filaments possess. A preferable method would be to collect one or two orders of magnitude more data, generate subsets of filaments that have a common pitch and twist, and reconstruct these subsets separately. This is beyond the scope of the present paper. Since small changes in subunit conformation or orientation can lead to large changes in twist and pitch, and since the spread of data appears to be across a continuum, we feel that this averaging will probably not introduce large errors at the resolution that we are using. The amplitudes and phases from the averaged layer line set are shown in Fig. 5. A comparison with the averaged layer lines of the intact 38K filament
CONFORMATIONAL
CHANGE
(Yu and Egelman, 1990) does not reveal any large differences localized to a single layer line. The amplitude-weighted phase difference between these two averages is about 44”. The quality of the data, as judged by how well individual filaments average together, is shown in Fig. 6, where the phase residuals between individual filaments and the average data set are shown. It can be seen that the filaments average together quite well, with a clear polarity. The quality of the layer line data can also be described by the FOM (see Methods). Not surprisingly, the strongest peak (on the sixth layer line) has a FOM of about 0.98, and the weakest data fall off to a FOM of about 0.4. The secondary maximum on the sixth layer line (R = 0.035, or a resolution of about 27 A from the origin of the transform) has a FOM of 0.58, suggesting that although the data are noisy, they are still significantly correlated among the 16 filaments used. Since this secondary peak is only about 4% of the amplitude of the main peak, it is also apparent that weak data at higher resolutions are not merely noise. Three-Dimensional Comparison with Intact RecA The layer lines of Fig. 5 allow one to compute a three-dimensional reconstruction of the 36K RecA fragment, and this is shown in Fig. 7. The confidence with which one can interpret features in this reconstruction is very much dependent upon an estimate of the error or variance in the map, and we have displayed the standard error of the mean at every point in the map along with the reconstruction in Fig. 7. The cylindrically and axially averaged standard error of the mean is shown as a function of radius in Fig. 8, and it can be seen that the error is fairly flat across the radius of the filament, with an average of about 2.25 density units. Since the contour steps in Fig. 7 have a spacing of 1.3 density units, the standard error in general is less than the spacing of two contour lines. In comparison, an equivalent plot for the intact 38K RecA filament (from Yu and Egelman, 1990) shows a greater radial modulation, with an average value that is lower than that for the 36K filaments. The peak standard error for the 38K filaments occurs at a radius of about 25 A, which corresponds to the inner lobe of the RecA subunit. An alternate way of analyzing the error, noise, or variance in the map is to compare the map (Fig. 7) generated by the layer lines of Fig. 5 with a map generated from FOM-weighted layer lines (see Methods). We have done this, and the resulting map is extremely similar to that shown in Fig. 7 (data not shown), suggesting that the noise or variance in the layer lines is present with weak amplitudes, thus appearing weakly in the map. Since we have reconstructions of both the intact 38K RecA and the 36K fragment, and have maps of
IN RecA-DNA
FILAMENT
247
the standard error of the mean in each of these two reconstructions, we are in a position to do a quantitative comparison between the two structures. The key question in making a comparison between the two averaged maps is: what is the statistical significance of the differences that are seen? Due to noise, variation in staining, etc., any two sets of RecA filaments (even drawn from the same population) will display differences. To quantify the meaning of the differences that are seen, we have displayed the differences in terms of their statistical significance in Fig. 9, where the differences between the two maps have been divided by the standard error of the difference at every point. The resulting map has values not of density units but of dimensionless standard deviations. Because the absolute contrast within images depends upon staining, the amount of exposure to the electron beam, development of the negatives, densitometry, etc., we have scaled the two sets of images together for comparison purposes by using a least-squares search for both a multiplicative and additive scale factor. The factors that were found were small (changing the maps by less than 10%) and did not significantly affect the differences that were observed. If the structure of the 36K fragment was unchanged from that of the intact 38K protein, one would expect to see the largest difference in a region where the 38K map density was greater than that of the 36K. This region would correspond to the location of the C-terminus. Instead, what can be seen in Fig. 9 is that the largest single difference peak corresponds to a region where the 36K density is actually greater than the 38K density. This peak difference has a value of 5.0 standard deviations. The largest peak where the 38K density is greater than the 36K density has a value of 3.9 standard deviations (labeled as No. 2 in Fig. 91, and actually lies outside of the region contoured as protein for both the 36K and 38K filaments. There are no other peaks greater than 3.6 standard deviations, but there are three additional peaks with values between 3.0 and 3.6 standard deviations where the 38K density is greater than the 36K density (these are labeled as 1, 3, and 4 in Fig. 9). Two of these, 3 and 4, are on the edge of the density contoured as protein, while the other two, 1 and 2, fall outside of this region. It is of interest to estimate what the probability is for seeing a difference greater than three standard deviations that is entirely due to noise. For a normal distribution, the percentage of the distribution greater than 3 standard deviations from the mean is 0.26%. All of the maps were calculated on a 2 A per pixel grid, extending radially to 80 A. There are thus about 5000 pixels within each section. If the density at every pixel were independent, we would therefore expect about 13 peaks
248
YU AND EGELMAN rcc
. .- ‘icu
L=13
N=-4
L=12
N=2
L=7
N=-5
uo 00
-200.0 .
PO.0
Loo.0
LLl 0.w 0.010o.olso&a odeso&m 0.a 0.m 0.w FIG. 5. The amplitudes and phases for the averaged set of seven layer lines from 16 filaments of the 36K RecA fragment (left), and the corresponding layer lines from a sample of 15 filaments of intact 38K RecA (right, from Yu and Egelman, 1990). The two independent data sets fkom each individual filament, corresponding to the near and far sides of the particle, were first internally averaged before filaments were averaged together. Therefore, the total average on the letI actually contains 32 independent data sets.
greater than 3 standard deviations due solely to noise, and not any structural difference. With a finite resolution, however, the density at each pixel is not independent, and one would expect the number
of observed peaks due to noise to be significantly less than 13. Without making a rigorous analysis of the error, it is reasonable to assume that the two additional peaks greater than 3 standard deviations, lo-
CONFORMATIONAL
CHANGE
IN RecA-DNA
249
FILAMENT
40 loo L=5
N=7 - mo.0 - loo.0 c
1
*
loo a
*
IOD
I **-
L=l
N=-6
L=O
N=O
-00 ,
0
FIG. Eontinued
cated outside of the putative protein boundary, are consistent with noise. On the other hand, the probability (for a normal distribution) of a value being 5.0 or greater standard deviations from the mean is on the order of 6 x 10 - 7, and it is sensible to assume
that the largest difference peak is highly significant, and extremely unlikely to have arisen from random differences between two samples from the same population. Figure 9 thus shows a very clear shift in mass
250
YU AND EGELMAN
9ov 90
’ 75
I 60
I 45
I 30
I 15
I 0
B&or Phatr Realdual FIG. 6. The total amplitude-weighted phase residual between each of the 16 layer line sets used in the average and the averaged layer line set itself (Fig. 51, as a function of the polarity of each individual set against the averaged data set. This plot can serve as a measure of the intrinsic polarity of the data. For a nonpolar structure (or for highly noisy data) there would be no significant difference between the two different orientations of each individual data set, and all points would fall near the diagonal line. The further points are from the diagonal, the greater the polarity. The centroid of the distribution is 38” for the better phase residual, and 56” for the opposite orientation.
between the 36K and 38K filaments that appears to be mainly localized to one region. We have interpreted this shift as a movement of the inner lobe of the RecA subunit through a distance of about 15 A in a direction that is radially outward (see arrows in Fig. 9). Limited resolution, however, prevents one from distinguishing between the simple movement of one lobe and a more complicated conformational change within the RecA subunit or within the interface between subunits. DISCUSSION
This study was begun with the expectation that the cleavage of the C-terminus from RecA protein would allow us to identify the location of the Cterminus in a three-dimensional reconstruction. We have previously shown (Egelman and Yu, 1989) that sensitive difference measurements were capable of visualizing a single strand of DNA in the RecADNA complex, accounting for only about 2.5% of the mass of the complex. We expected, therefore, that the approximately 5% of the molecule that we had
cleaved in this study would clearly appear as a region of missing mass in a comparison of the fragment with the intact protein. Instead, what we have seen is a rearrangement of the internal density in the fragment, consistent with a movement of the inner lobe of the RecA protomer. We have shown (Yu and Egelman, 1990) that the RecA protomer in the ReCA-ADP-Pi-DNA filament is visualized at low resolution as containing two domains or lobes. We also showed that the region of largest variance within a sample of 15 filaments used for that reconstruction fell on the inner domain. RecA has been shown to be extremely flexible, to have a variable pitch and a variable twist (Egelman and Stasiak, 1986). Different conformations of RecA subunits have also been directly visualized in bundles of RecA filaments (Egelman and Stasiak, 1988; Yu and Egelman, 1991). The observation of a large variance associated with the inner domain in the RecA-ADP-Pi-DNA filament provides a more detailed understanding of the internal mechanics of the RecA filament, with the potential for understanding how the filament can be so easily bent, twisted, and extended. Our observation, reported here, of a significant movement of the inner lobe through about 15 A following the removal of about 18 C-terminal residues, extends the previous interpretations about internal mechanics. It appears that the inner lobe in the wild-type filament is variable in position, and that the cleavage of the C-terminus forces this lobe outward. In the 36K filament this lobe, that was formerly at an inner radius, is now relatively more fixed, as judged by the variance map for this state (Fig. 7). A greater variance is now associated with the outer lobe. The visualization of conformational changes within the RecA filament is interesting in and of itself, and shows that protein polymers have the ability to undergo such conformational changes while still maintaining the noncovalent proteinprotein interface necessary for the maintenance of polymer integrity (Egelman and DeRosier, 1991). The key task, however, must be to use these observations to advance our understanding of RecA function. Fortunately, data already exist on the behavior of RecA fragments. There have been four previous studies of small C-terminal deletions of the RecA
FIG. 7. The layer lines of Fig. 5 have been used to generate a three-dimensional reconstruction of the 36K fragment BecA filament. Three sections through the filament are shown, corres onding to infinitesimally thin slices normal to the filament axis, which is indicated by a cross in each section. The sections are spaced 5 1 apart. Because the axial rise per subunit is about 15 A, and the rotation between subunits is 360”/(37/6), or about 584, the section at z = 15 a will be almost identical with the section at z = 0 A, but rotated by 58.4”. The density levels in the averaged structure are indicated by the contour lines, while the standard error of the mean at every point in the average density is given by the continuous grey scale. The sections have been contoured in steps of 1.3 arbitrary density units, starting at 40, with a peak density of 53. The SEMs are displayed with all values less than 2 density units being white, with the peak value (about 3.6) being black. All values between 2 and 3.6 are linearly represented by a grey value between white and black.
CONFORMATIONAL
CHANGE
IN &CA-DNA
FILAMENT
251
252
YU AND EGELMAN
2 2.0
i ;
s 0 1.5
1.0
0
20 Radius
40 (Angstroms)
60
FIG. 8. The cylindrically and axially averaged standard error of the mean at every point in the reconstruction is shown as a function of the radius from the helical axis. Data from the 36K fragment is shown with the solid line, while for purposes of comparison the data from intact RecA filaments (Yu and Egelman, 1990) is shown with a dashed line. The SEM for the 36K fragment tends to be more uniformly distributed across the filament than in the intact RecA, where there is a significant peak in the SEM at about 25 A radius, corresponding to the inner lobe of the RecA subunit.
protein. Two of these (Rusche et al., 1985; Benedict and Kowalczykowski, 1988) were based upon proteolysis, while the other two (Yarranton and Sedgwick, 1982; Larminat and Defais, 1989) used truncated genes. Care must be exercised in directly comparing the results of these studies due to differences in the amount of truncation, the assays employed, and even the species of bacteria used. Rusche et al. made a subtilisin digestion of both the E. coli and P. mirubilis RecA proteins, but were only able to obtain a large, stable fragment from the P. mirubilis protein. The P. mirubilis RecA fragment that they studied had an apparent molecular weight of 36 000, very similar to our E. coli RecA fragment. Rusche et al. noted that the P. mirubilis fragment interacted in the same way as the intact RecA with single
stranded DNA, but had different interactions with double stranded DNA. Specifically, the fragment appeared to bind duplex DNA with a greater affinity than the intact protein. Benedict and Kowalczykowski (1988) described the properties of an E. coli RecA fragment, missing about 45 residues at the C-terminus, that was fortuitously found after storing a sample of RecA at 4°C for 6 months. Like Rusche et al., they found that this fragment had unaltered single stranded DNA binding properties, but enhanced binding to double stranded DNA. Larminat and Defais (1989) used genetic constructs to express in E. coli a RecA fragment missing 17 C-terminus residues. They found that this fragment was nearly identical in uiuo to the wild-type protein in inducing the SOS response (cleaving the LexA repressor), performing recombination, and inducing mutagenesis. However, they observed a negative complementation when the RecA fragment was present in cells along with the intact wild-type protein. Yarranton and Sedgwick (1982) also employed genetic constructs to express a RecA protein missing about 75 C-terminus residues. They also observed a negative complementation effect with regard to recombination when this protein was expressed along with wild-type, but they observed no effect on SOS induction or phage h induction. At this point, structural observations of rearrangements of mass within the RecA filament as a result of the C-terminus deletion cannot be directly related to the functional studies that have been made on similar fragments. However, together these observations will add considerably to the interpretation of an x-ray crystal structure of RecA that is soon to be available (McKay et al., 1980; T. Steitz, personal communication). We now know that the deletion of the RecA C-terminus must affect double stranded DNA binding through a conformational change that moves part or all of the inner lobe of the RecA protein through about 15 A.
FIG. 9. A comparison between the reconstruction of the 36K fragment (left side) and intact 38K RecA (right side, from Yu and Egelman, 1990). The density distributions are shown with contour lines for sections at three different axial spacings, spaced 5 A apart. The significance of the differences between each of the two pairs is given by the difference divided by the standard error of the difference, and this is shown by the continuous grey scale density superimposed on both sets of contour lines. The grey scale on the left (superimposed on the 36K map) shows regions where the density of the 38K filament is greater than that of the 36K, while the grey scale on the right (superimposed on the 38K may) shows regions where the density of the 36K filament is greater than that of the 38K. The arrows on the right indicate the movement of the inner lobe that is most likely responsible for the large difference seen there. The arrow in the section at 0 A corresponds to the motion of one subunit, while the arrows in the sections at 5 and 10 A correspond to the motions within the next subunit above that one in the helix. Although the rise per subunit in the filament is about 15 A, each subunit extends axially within the filament about 45 A (Yu and Egelman, 1990). The grey scale data has been scaled such that values below 2.3 standard deviations will be white, while the peak standard deviation, 5.0, is black. All values between 2.3 and 5.0 are linearly mapped onto a grey scale between white and black. The largest difference, at 5.0 standard deviations, is at the end of the arrow in the section at z = 10 A. The corresponding peaks at the ends of the arrows in the sections at 0 and 5 A are 4.4 standard deviations. The largest difference where the 38K density is greater than the 36K has a value of 3.9 standard deviations, and is labeled 2 in the section at 10 A. The only other peaks greater than 3.0 standard deviations are labeled 1, 3, and 4. Peak 4 extends into the density outlined as protein, and peak 3 is at the edge of that density. Peaks 1 and 2, however, appear to lie significantly away from the protein boundary, and are consistent with noise (see text).
1
ww
254
YU AND EGELMAN
This research was supported by NIH Grant GM35269 NSF Grant DMB 87-12075.
and by
REFERENCES
BENEDICT, R. C., AND KOWALCZYKOWSKI, S. C. (1988) J. Biol. Chem. 263, 15513-15520. Cox, M. M., AND LEHMAN, I. R. (1987) Annu. Rev. Biochem. 56, 229-262. CRAIG, N. L., AND ROBERTS, J. W. (1981) J. Biol. Chem. 256, 8039-8044. EGELMAN, E. H. (1986) Ultramicroscopy 19, 367-374. EGELMAN, E. H., AND DEROSIER, D. J. (1991) J. Mol. Biol. 217, 4OWO8. EGELMAN, E. H., AND STASIAK, A. (1986) J. Mol. Biol. 191, 677697.
EGELMAN, E. H., AND STASIAK, A. (1988) J. Mol. Biol. 200, 329349. EGELMAN, E. H., AND Yu, X. (1989) Science 245, 404407. GRIFFITH, J., AND SHORES, C. G. (1985) Biochemistry 24,158162.
HOWARD-FLANDERS, P., WEST, S. C., AND STASIAK, A. (1984) NUture 309, 215-220. LARMINAT, F., AND DEFAIS, M. (1989) Mol. Gen. Genetics 216, 106-112. MCKAY, D. B., STEITZ, T. A., WEBER, I. T., WEST, S. C., AND HOWARD-FLANDERS, P. (1980) J. Biol. Chem. 255,6662. MOREAU, P. L., AND CARLIER, M.-F. (1989) J. Biol. Chem. 264, 2302-2306. RADDING, C. M. (1981) Cell 25, 34. RUSCHE, J. R., KONISBERG, W., AND HOWARD-FLANDERS, P. (1985) J. Biol. Chem. 260, 949-955. SANCAR, A., STACHELEK, C., KONIGSBERG, W., AND RUPP, W. D. (1980) Proc. Natl. Acad. Sci. USA 77, 2611-2615. WEST, S. C., CASSUTO, E., AND HOWARD-FLANDERS, P. (1982) Mol. Gen. Genet. 187, 209-217. YARRANTON, G. T., AND SEDGWICK, S. G. (1982) Mol. Gen. Genet. 185, 99-104. Yu, X., AND EGELMAN, E. H. (1990) Biophys. J. 57, 555-566. Yu, X., AND EGELMAN, E. H. (1991), submitted for publication.