USF

NEWS FROM THE UPPSALA SOFTWARE FACTORY - 7

Taking the fun out of map interpretation ...

Gerard J. Kleywegt & T. Alwyn Jones
Department of Molecular Biology
Biomedical Centre, Uppsala University
Uppsala - Sweden


Interpretation of experimental electron-density maps has traditionally been a difficult, time-consuming, yet also fun and exciting step in the structure-determination process. It is one of the few activities for which (successful) black boxes have not yet been developed.

ESSENS

We have recently described a method to detect rigid structural entities in electron-density maps [1, 2], and implemented this in a program called ESSENS [3], which is distributed as part of the RAVE package [4]. Briefly, this program reads a map and a structural fragment, oes a complete rotation of the fragment for every grid point in the map, and for each position and orientation calculates how well the fragment fits the local density (the "score"). When the calculations are finished, the scores are written out as a new map, which can be contoured in O [5]. By using a penta-alanine fragment in either ideal alpha-helical or ideal beta-strand conformation as the template, the result is a new map which shows how well a helix or strand fits at each point in the map. Examples of the application of this technique to two structures are discussed in [1], and figures of the "helix and strand maps" are also available on the Web [2]. This technique, simple (and CPU-time consuming) as it is, turns out to be very powerful. First, if the score map reveals helices and strands, the model-building process is greatly facilitated. Second, if the score map is featureless, this may well indicate that the phases are not good enough, and that the crystallographer is probably better off spending some more time on collecting more derivative data than on staring at an uninterpretable map.

MORTEN'S MODIFICATION

The previous paragraph basically describes the state of the program at the time when the paper [1, 2] was submitted. Since then, a number of improvements have been made. First, Morten Kjeldgaard suggested a modification of the algorithm which essentially imprints and image of the helix or strand in the score map. The effect of this is that the map becomes much clearer and easier to interpret. In the original implementation, the image of a strand or helix looked essentially like a C-alpha or main-chain trace with little detail. In the new version, the image includes all atoms of the poly-alanine fragment, and in particular the visibility of the C-beta atoms is a major improvement (which, in the case of helices, makes it very easy to deduce their directionality). Initially, Morten's modification was applied for every orientation at every grid point, which made the method ~2 times slower than the original one. However, it can easily be applied a posteriori , which means that in a single run of the program one obtains both the original score map and the map which results from Morten's modification. This is how the method is currently implemented in ESSENS. The map resulting from Morten's modification is called the display map, since it is the most appealing for contouring purposes.

EXTRACTING SOLUTIONS

A shortcoming of the original program was that, although the output map showed where the structural fragment fitted the map, the information about the best-fitting orientation was completely lost. This has recently been changed, by storing the best set of rotation angles for every grid point (encoded as a single integer number), and writing these out to a file. A new program, SOLEX [6] (for SOLution EXtractor) was written which reads the score map, the rotation file and the structural fragment, and extracts the top solutions, writing them to a new PDB-formatted file. This PDB file can be read into O again, and the fragments of strand or helix can be used in the model-building process. In addition, specifically for helices and strands, an (experimental !) option is available which will attempt to "connect the dots", i.e. to merge bits of strand or helix which lie close to one another and are more or less parallel, in order to build up longer secondary structure elements. At present, the algorithm appears to be working reasonably well, except that the directionality is sometimes wrong, so the crystallographer must be alert.

IF THE FOLD IS OLD ...

The secondary structure elements that are found by SOLEX are also written to a file which can used as input to DEJAVU [4]. DEJAVU is a program that looks for fold similarities between a structure and a large database of structures derived from the PDB. At the 1994 CCP4 study weekend, we demonstrated the use of DEJAVU in finding proteins with similar folds using only Bones-derived "secondary structure elements" (in effect, these are simply guesses of the beginning and end coordinates of helices and strands), even when the directionality of the helices and strands is unknown (or uncertain). Similarly, the files now produces by SOLEX can be fed into DEJAVU, and the program will attempt to find other proteins with a similar (partial) fold. In our standard test case (P2 myelin protein), this works surprisingly well. SOLEX finds eight beta-strands, and when DEJAVU is asked to look for similar proteins which have at least six strands in common with our "unknown" structure (ignoring directionality), the program comes up with four correct hits (including P2 myelin protein itself), and no false hits, albeit that the orientation of one of the hits is wrong. Nevertheless, if this were a real case, the tracing problem would essentially be solved, in that the coordinates of one of the DEJAVU hits could be "stolen" to jump start the model-building process.

OTHER APPLICATIONS

ESSENS and SOLEX also have other applications, for example in real-space, phased molecular replacement calculations (although in this case reciprocal-space methods are probably to be preferred for reasons of speed). This method was used in the structure determination of an acetylcholinesterase/fasciculin complex [7]. The acetylcholinesterase molecule was easily found using standard molecular replacement techniques. However, positioning the fasciculin molecule in a map phased on acetylcholinesterase was difficult. ESSENS was used to verify the manually obtained solution (which was correct), by using a truncated poly-alanine model of fasciculin as the search fragment (see [1], [2], and [7] for details). For this type of calculation, the ability to extract any number of solutions conveniently with SOLEX is obviously important.
Finally, we have experimented with other types of template, such as a di-alanine, a tryptophan ring, etc. , with varying results. There is no inherent limitation on the type of structural fragment that is used as a template other than that it must be (assumed to be) rigid, i.e. not contain any free conformational torsion angles. Hence, the method should also work very well when searching for nucleic acids, sugar rings, many ligands, structural motifs (e.g. , hairpin turns) or even structural domains.

AVAILABILITY

The RAVE package (which includes ESSENS, SOLEX and many other programs) is available to academic users free of charge (from the O ftp server). For more information about RAVE, contact GJK (gerard@xray.bmc.uu.se). For more information about O, contact TAJ (alwyn@xray.bmc.uu.se).

REFERENCES


NOTE

The Uppsala Software Factory now has its own home page on the World Wide Web, providing access to many resources and services in Uppsala, including:
The URL is: http://xray.bmc.uu.se/usf/

USF Latest update at 12 February, 1998.