Patterson-Map Interpretation with Nonerystailographic Symmetry

Transcription

1 15 J. Appl. Cryst. (1993). 26, Patterson-Map Interpretation with Nonerystailographic Symmetry BY LIANG TONG* AND MICHAEL G. ROSSMANN Department of Biological Sciences, Purdue University, West Lafayette, Indiana 47907, USA (Received 31 March 1992; accepted 2 July 1992) Abstract The interpretation of difference Patterson maps for heavy-atom positions is an important step in the structure determination of biological molecules. In the absence of a large number of noncrystallographically related subunits, the first step is usually a Harkervector search to locate single-site solutions, which is followed by a cross-vector search to locate two-atom solutions and subsequent searches to locate more sites. Crystals of biological macromolecules frequently contain assemblies obeying noncrystallographic symmetry. The information about their noncrystallographic symmetry can be incorporated in search procedures to aid the Patterson-map interpretation. In the presence of noncrystallographic symmetry, a self-vector search is useful for locating the positions of heavy atoms relative to the center of the noncrystallographic symmetry point group and a crossvector search is then used to locate that center in the crystal unit cell. An automated procedure is presented which is a generalized Patterson search method. Introduction Heavy-atom and Bijvoet difference Patterson maps can be used to determine the arrangement of heavy atoms in protein crystal derivatives, the knowledge of which is essential for the application of the multiple isomorphous replacement method in protein crystallography (Blundell & Johnson, 1976). The interpretation of Patterson maps for atomic positions can be achieved manually, though it is often carried out through automated search procedures (Sheldrick, 1985; Terwilliger, Kim & Eisenberg, 1987). In these procedures, a set of trial atomic positions is usually evaluated by the minimum (Buerger, 1970), the sum (average) or the product of the Patterson-map values at the end points of the vectors among the trial atomic sites. A correct set of atomic positions should produce a high minimum-, sum- or product-function value. The search is usually carried out in a few steps, with each step obtaining a subset of the information. In the absence of noncrystallographic (local)l" symmetry, three steps are usually involved - Harker-vector searches to locate single sites, cross-vector searches to obtain two-atom solutions and, finally, further searches to locate more atomic positions. Icosahedral virus crystals and many protein crystals contain assemblies of biological molecules obeying noncrystallographic symmetry. The orientation of these local symmetry elements can usually be determined from ordinary (Rossmann & Blow, 1962; Crowther, 1972)and locked (Rossmann, Ford, Watson & Banaszak, 1972; Tong & Rossmann, 1990) rotation-function calculations. This information about the local symmetry can be utilized in the interpretation of Patterson maps (Argos & Rossmann, 1974, 1976; Arnold et al., 1987). However, most of the currently available heavy-atom search packages do not support the presence of noncrystallographic symmetry. This paper presents a generalized procedure for Patterson-map interpretation including the utilization of noncrystallographic symmetry. Harker-vector searches Harker vectors are the vectors between atoms related by crystallographic symmetry. In general, X n -- X m = ([Tn] -- [Tm])X 1 --I- (t n -- tin), (1) where [7,] and t, (n = 1,..., Ncry ) are the rotational and translational components of the crystallographic symmetry operators, respectively, and x, is a set of positions related by crystallographic symmetry. In the equation above, m can be held at 1 as the other vectors are related by Patterson symmetry. The remaining Ncry-1 non-zero Harker vectors also contain symmetry mates for space groups with three-, four- and sixfold symmetry axes. The presence of each three- or fourfold axis produces one pair of vectors that are related by symmetry. The presence of each sixfold axis produces two pairs of symmetry mates. For example, there are 11-2 (for one sixfold) = 9 unique Harker * Present address: Department of Medicinal Chemistry, Boehringer Ingelheim Pharmaceuticals Inc., 900 Ridgebury Road, Ridgefield, CT 06877, USA. ~" In this paper, the words noncrystallographic and local will be used interchangeably /93/ International Union of Crystallography

2 16 PATTERSON-MAP INTERPRETATION vectors for space group P6122 and there are 23-3 (for three fouffolds)- 4 (for four threefolds)= 16 unique Harker vectors for space group Harker-vector searches are used to determine single-site solutions (xl) to Patterson maps. This is the first step in the automated interpretation of Patterson maps. The search procedure (Terwilliger et al, 1987) is general, as all possible positions are examined as potential single-site solutions. The positions are first sampled on the grid of the input Patterson map and all those with Harker-vector heights greater than a specified cut-off value are saved. Then a grid twice as fine as that of the Patterson map is sampled around these positions to maximize their Harker-vector heights. The top unique positions are then the single-site solutions. The region of the unit cell that needs to be explored to locate single-site solutions is usually smaller, sometimes much smaller, than the asymmetric unit of the unit cell. Besides the crystallographically related positions, sites that are related by alternative unit-cell origin translations share the same set of Harker vectors and hence cannot be distinguished in single-site solutions. Moreover, for nonenantiomorphic space groups, the enantiomorphic positions (x, y, z) and (-x,-y,-z) also share the same set of Harker vectors. With space group P212~21 as an example, there are a total of 4 (for space-group symmetry) x 8 (for alternative origins) x 2 (for enantiomorphs) = 64 positions in the unit cell that give rise to the same set of Harker vectors. Therefore, the unique region for single-site solutions should have a volume 1/64 that of the unit cell, which is only 1/16 that of the asymmetric unit. This redundancy in alternative origin translations and enantiomorphs also implies that single-site solutions cannot be used directly in a structure determination. The origin translation and enantiomorph relationship between possible singlesite solutions must be established first. Self-vector searches In the presence of local symmetry, self-vectors are defined as the vectors between a position on a molecular assembly and its mates related by the local symmetry of that assembly. Assuming that the molecular assembly has point-group symmetry, it can be shown that (Argos & Rossmann, 1974) x. = [~][E]Ep.]U1 + s (2) x.- x. = [~][~]([pj - [p,,3)u1; (3) where [p~ (n = 1,..., Nlo~) are the N~o rotation matrices for the local symmetry operations in a 'standard' orientation; [E] is a rotation which brings the 'standard' orientation to that of the molecular assembly in the crystal (see Tong & Rossmann, 1990); [~] is the deorthogonalization matrix converting orthogonal ~mgstr6m coordinates to fractional coordinates along unit-cell edges; U1 is the position of an atom relative to the center of the molecular assembly in the standard orientation; x,, in fractional coordinates, is a set of positions related by the noncrystallographic symmetry of the molecular assembly; and s is the center of the molecular assembly in the unit cell. The 'standard' orientation is usually chosen such that the rotation matrices of the symmetry operators assume simple forms, as this makes it easier to define the region to be covered in the self-vector searches. For example, for a molecular assembly with 222 symmetry, the standard orientation can be defined with the local twofold axes along the Cartesian P, Q and R axes. The search region can then be restricted to the PQR positive octant. Owing to the center of symmetry in the Patterson maps, a center of symmetry should be added to the local symmetry point group to determine the unique region for self-vector searches. The rotation matrix [E] can be determined, for example, from a locked rotation function (Tong & Rossmann, 1990). Self-vector searches in the presence of noncrystallographic symmetry are used to determine coordinates (U1) of atoms relative to the center of the molecular assembly. The self-vector search region is divided into a grid and all the grid points are examined in the searches. The sum function should be used in such searches owing to the number of self-vectors that are usually involved (see below). Those unique positions with self-vector heights less than a cut-off value are ignored. Self-vector searches cannot locate the center of a molecular assembly in the unit cell. A cross-vector search (see below) is therefore required to determine completely the heavy- atom positions in the crystal cell. There are, in general, N1oc(N1oc- 1)/2 crystallographically unique self-vectors for an assembly with N~oc copies of a unique molecule or molecular subunit. For most assemblies in protein crystals, the value of N~oc is small and hence all the self-vectors of a position can be checked against the Patterson map. On the other hand, N~o can assume a value of 60 for icosahedral virus crystals, implying the presence of 1770 self-vectors for any given position. It is rather computationally intensive to check all of these against the Patterson map. However, these self-vectors should obey the local symmetry. Therefore, a good approximation is to check only the vectors that fall within one noncrystallographic asymmetric unit. This is equivalent to checking only the vectors between an (arbitrarily chosen) reference position and its local symmetry mates [keeping m at 1 in (3)]. This would give rise to only 59 self-vectors for a virus assembly with 60 unique copies. If these self-vectors assume reasonable heights, the other self-vectors should also be consistent with the Patterson map. On the other

3 LIANG TONG AND MICHAEL G. ROSSMANN 17 hand, it is statistically unlikely for a wrong site to have high values for these 59 self-vectors. The solution of the heavy-atom positions was obtained for human rhinovirus 14, where N~o = 20, using all 190 possible self-vectors (Arnold et al., 1987). The program described here can either use all vectors or only those in the noncrystallographic asymmetric unit. Coincidences in the self-vectors can cause problems, especially if the trial atomic site is situated on special positions of the point group. The Patterson-map value for a vector should, to a first approximation, be divided by its multiplicity due to the coincidence of self-vectors. In reality, a Patterson map is a collection of both self- and cross-vectors (see below). The correct treatment, taking into account multiplicities due to coincidences of self- and cross-vectors, would be too demanding computationally. Cross-vector searches In the absence of local symmetry, cross vectors are defined as the vectors between two crystallographically unique atoms in the unit cell. For instance, Xn Xm, 1 = ([Tn]x2 -- [Tm]Xl) + (tn - tm), (4) where xt and x 2 are the two unique atomic positions. As is the case with self-vectors, m can be held at 1 as the other cross vectors are related by the Patterson symmetry. So, in general, there are Nery unique cross vectors between a pair of atomic positions. Cross-vector searches, the second step in the interpretation of Patterson maps, are used to correlate two single-site solutions to the same origin and enantiomorph, as the cross vectors are not subject to redundancies in alternative origin translations or enantiomorphs. Two different strategies can be employed for the cross-vector searches. In the first strategy, all possible combinations of alternative origin translations and enantiomorphs between all pairs of single sites with Harker-vector heights greater than a cut-off value are examined for their cross-vector heights. Combinations that have cross-vector heights below a specified cut-off value are ignored. Combinations that produce high cross-vector values represent possible two-atom solutions to the Patterson map. In the second strategy, all positions in the Patterson map with values greater than a specified cut-off are tried as possible cross vectors. Given such a vector and a single site with Harker-vector height greater than a cut-off value, a new site can be defined as the vector sum of the two. The Harker-vector height of this new site and the cross-vector height between it and the given single site can then be examined. For both search strategies, the resulting two-atom solutions are sorted based on the cross-vector heights. In the absence of local symmetry, the third step in the interpretation of Patterson maps is the search for more atomic positions given a two-atom solution from the cross-vector searches. A general search can be carried out over all positions in the asymmetric unit of the unit cell. The Harker-vector height of a trial position and the cross-vector heights between it and the two known solutions are then examined. Those positions with Harker-vector and cross-vector heights greater than specified cut-offs are sorted based on the cross-vector heights. The cross-vector heights among the newly identified positions can be calculated after all positions have been examined. Lower cut-off values for Harker- and cross-vector heights should be employed in this search as the extra sites are often minor sites in the heavy-atom derivatives. Once a few heavy-atom positions are known, it is also possible to use difference electron density maps, based on single isomorphous replacement phases, to locate more sites. In the presence of local symmetry, cross vectors are defined as the vectors between a set of atoms on a reference molecular assembly and the same set of atoms on crystallographically related assemblies. If it is assumed that U1 is a position belonging to a set of atoms and s is the center of the reference assembly in the unit cell, it can be shown that Xn.~ -- Xm.~ = ([TdE~]EE][p.] -- [Tj]E~][E]EPm])U, +(t k -- tj) + ([Tk] -- [Tj])s (, -~ m), (5) where xn. k is a position, related to U1, in the nth noncrystallographic asymmetric unit of an assembly in the kth crystallographic asymmetric unit. Cross-vector searches in the presence of local symmetry are used to locate the center (s) of the molecular assembly in the crystal unit cell given a position (U1) belonging to that assembly. In (5),j can be held at 1 as the other cross vectors are related by the Patterson symmetry. That would still imply the presence of(ncry- 1) x Nlo~ x Nloc cross vectors. For an icosahedral virus crystal with monoclinic (P21) symmetry, there should be 3600 cross vectors for each given molecular center. Similar to the approach used in the self-vector searches, a useful approximation is to examine only the cross vectors between a reference position on the reference assembly and all the positions on the other assemblies [equivalent to holding both m andj at 1 in (5)]. This would leave (N,y - 1) x Nio, or only 60 for the virus-crystal example above, cross vectors for each trial center. However, the cross vectors are not expected to obey the local symmetry. Hence, this approximation is based essentially on statistics and should only be used for special cases. The general treatment, checking all the cross vectors of a given position, is supported by the program and should be used for most occasions. Since a specific orientation is implied by a given position U~, the corresponding molecular assembly

4 18 PATTERSON-MAP INTERPRETATION can only reside in one of the asymmetric units of the unit cell. With no knowledge of which asymmetric unit is correct, it is necessary to search the entire unit cell to locate the molecular center. The presence of alternative origins reduces the region to be covered. For example, only the region 0 < x < ½, 0 < y < ½ and 0 < z <½ needs to be searched for space group P because of the eight alternative origins. Packing calculations can be incorporated to remove molecular centers which would cause overlap among the crystallographically related molecular assemblies. It is usually more advantageous to consider only pairs of crystallographic symmetry operators that render the cross vectors independent of the a, b or c component of the molecular center. The cross-vector searches can then be carried out in two-dimensional projections. The position of the molecular center can be determined and corroborated by combining information from the different projections. Once the center of the particle is located based on one heavy-atom position in the noncrystallographic asymmetric unit, it should be possible to look for extra sites based on their self- and cross vectors. This approach, used successfully in the structure determination of rhinovirus 14 (Arnold et al., 1987), is not available in the current program. The computer program The computer program PA TSOL, written in standard Fortran, and its documentation are available upon request. It has been tested on VAX/VMS 8650, DECstation running Ultrix, IBM Risc 6000 and Cyber205 computers. All inputs to the program are keyword based and free formatted. The program scales the input Patterson-map values such that their root-mean-square value, a measure of the noise in the map, is 100. Calculations are performed on a grid twice as fine as that of the input map. In this way, the Patterson-map value for a given vector can be obtained either by look-up or by simple average of its two, four or eight nearest neighbors. The input Patterson map should be sampled at a grid interval 1/3 to 1/6 of the resolution of the reflection data. The program automatically identifies the possible alternative origin translations, checks whether positions (x,y,z) and (--x, --y,--z) share the same set of Harker vectors and determines the unique Harker vectors of the space group. Test cases (a) Catalytic domain of human ras p21 The results from three different test cases are presented in this paper. The first test case is the interpretation of a difference Patterson map for the Table 1. Interpretation of the ras p21 difference (K2HgI4 + KAuI4) Patterson map In this and the subsequent tables, the Patterson-map values were scaled such that their root-mean-square value was 100. Numbers represent the minimum or average of the Harker-vector heights. Numbers in parentheses are the normalized search-function values with the top solution at 100. Site Minimum function Sum function Without With Without With coincidence coincidence coincidence coincidence correction correction correction correction 120 (100) 60 (100) 200 (100) 130 (100) 105 (88) 52 (87) 199 (1013) 105 (81) 74 (62) 35 (58) 169 (84) 103 (79) 70 (58) 33 (55) 167 (84) 91 (70) 54 (45) 30 (50) 158 (79) 91 (70) K2HgI 4 + KAuI4 double derivative of the catalytic domain of human ras p21 (de Vos et al, 1988). The crystals belong to space group P6522 with a = b and c = A. There is one molecule per asymmetric unit. The difference Patterson map was calculated using 8 to 4/~ resolution reflection data, sampled on a grid of The map was originally interpreted with the HASSP program (Terwilliger et al., 1987), yielding the two sites in the derivative. Using the current program, the sites were found through a Harker-vector search and a cross-vector search (using either strategy) in 1 min of CPU time on a VAX/VMS 8650 computer, as compared to 150 min with the HASSP program. This substantial saving in time occurs in the cross-vector searches. The HASSP program examines all positions in the asymmetric unit with each Patterson-map peak as a trial cross vector. However, it was sufficient to examine only those positions with reasonable Harker-vector heights. Both the minimum and the sum functions were used in Harker-vector searches to test their effects on the single-site solutions. The effects of multiplicity correction in the Patterson map and in the coincidence of Harker vectors were examined as well. A point lying on a special position in the Patterson map should have its value divided by the redundancy at that position. The multiplicity in Harker vectors occurs, as mentioned earlier, as a result of the presence of three-, four- and sixfold axes in the space group. For the sum function, special care should be taken to remove from the calculation vectors that lie close to Patterson-map origins. Table I lists the Harker-vector heights for the top five single-site solutions for each of the four calculations. The top two solutions are the correct ones in all four approaches. Use of a minimum function with multiplicity correction gave a slightly better separation between the solutions and the noise peaks. The sum function gave a worse

5 LIANG TONG AND MICHAEL G. ROSSMANN 19 signal-to-noise ratio as it is susceptible to dominance by Harker vectors with high values, (b) Sindbis virus core protein The second test case is a difference Patterson map for the KAuC14 + K2PtC14 double derivative of the type-3 crystals of Sindbis virus core protein (Choi et al., 1991; Tong, Choi, Minor & Rossmann, 1992). The crystals belong to space group P21 with a , b = 79.7, c = 60.8 A and p = There are two molecules per asymmetric unit, with a local twofold axis relating the two molecules at ~p = 0.0 and ff = The difference Patterson map was calculated using 8 to 4 A resolution reflection data and sampled on a grid of 40 x 80 x 60. The derivative contains two major and two minor sites, whose positions were determined manually as well as using the current program. The sites obey the local symmetry, with one major and one minor site on each molecule. The test is to interpret this difference Patterson map with the knowledge of the local symmetry. The standard orientation was chosen as having a twofold axis along the Cartesian Q axis. Thus, only the Q = 0 plane was covered in the selfvector searches, with P going from -22 to 22 A and R going from 0 to 22 A. Because the atomic coordinates are doubled to calculate the self-vectors, the search-grid interval was chosen to be 0.25 A. The top five solutions are listed in Table 2(a). There is a clear separation in self-vector heights between the first three sites and the other two sites. The first three sites were then used as inputs for cross-vector searches, which covered the region [0, ½] in a and [0, ¼] in c (Table 2b). Only the region [0, ~,] in c is needed as the model in this case possessed a center of symmetry. The search increment along a was and that along c was The first solution from the self-vector searches actually corresponded to two atoms related by the crystallographic a-axis translation. A cross-vector solution could not be found for the third site. This site is rather close to the local twofold axis; its presence as a solution from self-vector searches is probably due to imperfect treatment of the origin peak in the Patterson map. The cross-vector search for the second site correctly located the center of the dimer. The two sites thus identified corresponded to the two major sites. The two minor sites were not found from the self-vector searches as their self-vector height was only 72. As mentioned earlier, these two minor sites could be located in a general search without the use of the local symmetry. (c) Canine parvovirus This test case is based on the KzPtBr 6 derivative of canine parvovirus (CPV) (Tsao et al., 1991). The crystals belong to space group P21 with a = 263.1, Table 2. Patterson searches for Sindbis virus core protein (a) Sum-function values for vectors between noncrystallographically related atoms; PQR is a molecular coordinate system with the molecular dimer axis along Q Site P Q R Height (b) Sum-function values for vectors between atoms in different dimers; xyz are the fractional coordinates of the dimer center with respect to the crystal axes Site x y z Height b = 348.9, c = A and /3 = There is one virus particle per asymmetric unit. The heavy-atom compound had been shown to bind to residue Met 87. However, the derivative data set contained only 5% of the unique reflections. Therefore, calculated structure factors were used in this test case. Structure factors were calculated for reflections between 15 and 8 A resolution based on the atomic model of canine parvovirus. The 'derivative' was obtained by changing the $5 atom of Met 87 into a Pt atom. The difference Patterson map was sampled on a grid of 120 x The asymmetric unit of an icosahedral point group is bound by a fivefold axis, its nearest twofold axis and the two closest threefold axes. The standard orientation was chosen with a fivefold axis along the Cartesian Q axis and a perpendicular twofold axis along the R axis. The unique region is then roughly restricted to -36 to 36 in ~o and between 0 and 36 in 0. Only the ~o-positive half is needed if a center of symmetry can be added to the point group. However, when only the vectors between a reference position and its local symmetry mates are checked, the self-vectors do not possess a center of symmetry. Hence, it was necessary to cover the full region in the self-vector searches. Each self-vector search, covering a shell between 120 and 140 A in radius, took about 1050 s on a Cyber205. The search increment in (p and was 1 (or roughly 1.5 ~) and that in the radius was 1 A. The cross-vector searches, with the correct platinum position as input, were carried out for the region 0.2_<x<0.3 and 0.2 z_<0.3, owing to

6 20 PATTERSON-MAP INTERPRETATION (a) No errors introduced (R = 11%)* Table 3. Patterson-map interpretation with CPV data Self-vector searchest Cross-vector searchesi" Number of % First Second Third First Second Third reflections data peak peak peak peak peak peak ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ (b) 5% error introduced (R = 13%) Self-vector searchest Cross-vector searchest Number of % First Second Third First Second Third reflections data peak peak peak peak peak peak ~ ~ ~ ~ ~ ~ ~ ~ ~ (c) 10% error introduced (R = 17%) Self-vector searchesi" Cross-vector searchest Number of % First Second Third First Second Third reflections data peak peak peak peak peak peak ~ ~ ~ ~ ~ ~ ** ~ tt * R = 100 x ~., IFN -- FDI/~ FN, where F N and Fo are the structure amplitudes of the native and derivative data sets. Error in the data increases this value from 11% (no error) to 17% (10% error). t Self-vector searches are the result of averaging the Patterson-peak heights between atoms related by noncrystallographic symmetry. Cross-vector searches are the result of averaging the Patterson-peak heights between atoms in crystallographically related molecules. :~ The correct solution. The second solution is correct, with height 47. The second solution is correct, with height 38. ** The sixth solution is correct, with height 41. t']" The correct solution has a height of 35. packing considerations. The search increment was in a and c. Again, only the vectors between a reference position and those positions on the crystallographically related particle were checked. Each cross-vector search took about 70 s on the Cyber205. The effects of the completeness of the reflection data and of the errors in the reflection data on the self- and cross-vector searches were examined in the test calculations. Reflection data sets containing 70, 50, 30, 10 and 5% of total data were created by random selection from the complete data set. Errors were introduced into the native and derivative data sets by randomly adding or subtracting 5 or 10% from the structure-factor amplitudes. The test results (Table 3) show that the correct platinum position and the correct molecular center position could be identified from the self- and cross-vector searches for data sets that were 30,/0 or more complete and with about 5% error in the native and derivative amplitudes at around 8 A resolution. When 10% error was introduced, there was significant deterioration in both the self- and cross-vector results. Smaller separations between the vector heights of the correct position and noise peaks were observed and a more complete data set (> 50%)

7 LIANG TONG AND MICHAEL G. ROSSMANN 21 was needed to locate the correct position. These test results should provide some guidance in examining calculations based on actual difference Patterson maps. We are grateful to Sung-Hou Kim for the use of the ras p21 data as a test case. We thank Helene Prongay and Sharon Wilder for help in the preparation of this manuscript. This work was supported by grants from NSF and NIH to MGR and a Lucille P. Markey grant for the expansion of structural studies at Purdue University. References ARGOS, P. & ROSSMANN, M. G. (1974). Acta Cryst. A30, ARGOS, P. & ROSSMANN, M. G. (1976). Acta Cryst. B32, ARNOLD, E., VRIEND, G., Luo, M., GRIFFITH, J. P., KAMER, G., ERICKSON, J. W., JOHNSON, J. E. & ROSSMANN, M. G. (1987).,4cta Cryst. A43, BLUNDELL, T. L. & JOHNSON, L. N. (1976). Protein Crystallography. New York: Academic Press. BUERGER, M. J. (1970). Contemporary Crystallography. New York: McGraw-Hill. CHOI, H. IC, TONG, L., MINOR, W., DUMAS, P., BOEGE, U., ROSSMANN, M. G. & WENGLER, G. (1991). Nature (London), 354, CROWTHER, R. A. (1972). The Molecular Replacement Method, edited by M. G. ROSSMANN, pp New York: Gordon and Breach. ROSSMANN M. G. & BLOW, D. M. (1962). Acta Cryst. 15, ROSSMANN, M. G., FORD, G. C., WATSON, H. C. & BANASZAK, L. J. (1972). J. Mol. BioL 64, SHELDRICK, G. M. (1985). Crystallographic Computing 3, edited by G. M. SHELDRICK, C. KROGER & R. GODDARD, pp Oxford: Clarendon Press. TERWILLIGER, T. C., KIM, S.-H. & EISENBER6, D. (1987). Acta Cryst. A43, 1-5. TONG, L., CHOI, H. K., MINOR, W. & ROSSMANN, M. G. (1992). Acta Cryst. A48, TONG, L. & ROSSMANN, M. G. (1990). Acta Cryst. A46, TSAO, J., CHAPMAN, M. S., AGBANDJE, M., KELLER, W., SMITH, K., Wu, H., Luo, M., SMITH, T. J., ROSSMANN, M. G., COMPANS, R. W. & PARRISH, C. R. (1991). Science, 251, Vos, A. M. DE, TONG, L., MILBURN, M. V., MATIAS, P. M., JANCARIK, J., NOGUCHI, S., NISHIMURA, S., MIURA, K., OHTSUKA, E. & KIM, S.-H. (1988). Science, 239,