SINGLE TEMPLATE PROTEIN MODELING

INTRODUCTION
 

    Acetate kinases are a protein family mainly found in microorganisms whose enzymatic function consists in the conversion of organic matter to CO2 and CH4 (by the conversion of acetate to acetyl phosphate in the presence of ATP and a divalent cation). Structurally, after the SCOP classification, they belong to the a/b Class, and to the ASKHA Superfamily (acetate and sugar kinases/Hsc70/actin), whose members have a conserved structural "bbbababa" core and the strange epsilon conformation on the Ala 330 (Phi-angle=75.4º, Psi-angle=175.3º), that initiates a turn of a helix in all the members of the Superfamily.
 
 

    The structure of the acetate kinase from Rhizobium meliloti (P58382) has not been obtained experimentally yet. Its modeling has been based on a single template, the acetate kinase of Methanosarcina thermophila (1g99), whose structure was obtained in 2000 at 2.50 A (Buss et al., 2001).

    Although the molecules are dimers, both monomers are identical (100% homology, see dimer.aln), we have used one monomer of the template (Chain A) . Therefore, all the rasmol views below will account for one monomer only (each monomer consisting of a central beta sheet surrounded by alpha helices). Next view of the template shows the complete molecule in the so called "bird"  conformation: the body of the bird  is composed by the C-terminal domains (7 stranded beta sheet and 11 helices, for each monomer), and the wings by the N-terminal domains (8 stranded beta sheet and 8 helices, for each monomer). The ATP binds in the cleft between the two domains, without forming any hidrogen bonds with the protein (which explains the lack of specificity of the enzyme for a particular nucleotide triphosphate, as the phosphoryl donnor). The cleft closes and brings the two catalytic residues toghether (Glu384: one in each side of the molecule), so that they achieve the active site and participate directly in the phosphorilation.
 
 

AIMS OF THE STUDY
 

    Comparing the differences of modeling by primary sequence homology and by prediction of secondary structure.
 

MATERIALS AND METHODS
 

Sequences (swissprot entry)
 
 
 

Programs
 

RESULTS
 

Template selection
 

    The different strategies we use gives us the same and single template (1g99), with an E-value ranging  from e-218 to e-110:
 

• Psi-blast in the Swissprot database: we take 8 kinases (acetate, propionate and butyrate) with e-values from e-162 to e-90, and build a Hidden markov model of them; it serves for two strategies:

• Search in the PDB database (through "hmmsearch").
• Alignment of the markov model with the own problem sequence (through "hmmalign")- this will be the startpoint for a new psiblast in the PDB database.
• Psi-blast in the PDB database

    Since there are no templates other than 1g99 the only alignment we can put into the Modeller program is the one with the sequence of the template (taken from the PDBfile) and the problem sequence. Another approach is to use the Topits server to make a prediction of secondary structure of the problem sequence based on the information in the PDBfile of the template (see the files containing these procedures Alin.aln and topits.html). We note that the server extracts the template sequence from Swissprot and not from the PDBfile; we want to preserve the topits alignment and use the PDB sequence, otherwise the Modeller wouldn't recognize the input file. That's why we simply align both template sequences (swissprot and PDB template-sequence) and the gaps contained in the resulting file have to be added to the normal clustalw alignment (Alin.aln) which will also contain the only gap that returns topits (final.clw). Some gaps have to be changed in the clustalw alignment between the template and the problem sequence: those that appear at the end of an alfa helix (localized thanks to the topits prediction); we simply move them a bit forward to avoid cutting the helix, which would rise the general energetic profile, since these secondary structure stabilizes the final general tertiary structure (final.clw compared to pdb_problema.aln).
 

Resulting models
 

    After having changed the clustalw format to the peer format (alin.pir and final.pir) and generated the respective .top input files (alin.top and topits.top), we start running the Modeller program.

    We obtain four models (obj1, obj2, obj3 and obj4) and the respective Rasmol views are:
 
 

Model selection
 

    The energy profiles of each model calculated by Prosa II  (combined energy plots) shows at least 3 peaks of positive energy, from which the most important is the one at 80 residues, and the other two are at 230 and 250 residues, approximately (Prosa1, Prosa2, Prosa3 and Prosa4). Comparing both sequence based models (obj12prosa.jpg, code: obj1-yellow, obj2-red) and both topits models (obj34prosa.jpg, code: obj3-yellow,  obj4-red) we decide to use those with lowest energy profiles (obj2 y obj3) to continue with the optimization (in red in Prosa II plots). The  Ramachandran plots  (from the Procheck analysis, at 2.5A) show an average percentage of disallowed regions of 0.6% for both models, obj2 and obj3. (0.0% for the template 1g99A.sum).

    The second and the third models have the same number of bad contacts (3) in the Procheck output (P58382.sum, predict_h3170.sum), (1 bad contact in the template, 1g99A.sum).

    From these two models (obj2 and obj3), the first one has the most optimized energies; however, we will try to optimize the peaks of both of them, by using three different strategies: Grumos, Archtype and Molecular Dynamics.
 

Model optimization
 

    The Grumos optimization (previous arrangement of the model, without any disulfide bridge, water-free system,  Steepest Descent method of energy analysis, 10 optimizations -one every 1000 steps- without shake, interactions between groups) improves none of the two selected models, substantially.

    The peaks still appear in the common ProsaII plot (Grumos2, code: Obj2-yellow, Opt2-red; Grumos3, code: Obj3-yellow, Opt3-red); although there is a small descent in the two second peak in the sequence based models optimization, but the first  terrible one, is even increased.

    We cannot use Archtype to minimize the energy of the first 100 residues, as the loop found in this region is 12 aa long; we will wait until we run the Molecular Dynamic to see if there is something to do with it. We try anyway with the mentioned program but the results are hopeless: the protein presents 5 loops at the peaks sites, from which just four of them have lengths able to be analyzed through the Archtype (less than 9 residues long). Starting with the fourth loop (parameters), and lowering the cutoff from 60 to 45 (which is rather inefficient) the program gives a list of templates, from which the best is1mro_B (PDBentry), cluster_k, with a score of 100, but after having pasted the loop region to the normal "Template-ProblemSequence alignment", (and subsequent modifications to run the Modeller program), it raises the general energy profile (plot, code: Lup-magenta,  Lup2-green, Opt2-red). Since the obtained plot doesn't solve the energetic problem, and given the low cutoff used, we decide not to follow with this strategy.

  ! Since this is a modeling based on the practical courses of the subject Structural Biology, our knowledge has almost reached its limits.

    The last resource in the model optimization is to run de Molecular Dynamic analysis with the option in the Grumos program  (Temp. by default, without periodicity, rotation 50, 10ps  trajectory time, 10 output files, 100ps as total dynamic time). We do the same for both models (based on sequence and based on secondary structure prediction). The 10 resulting output files, superposed with XAM, show the fragment with the worst energetic profile (1-100 aa) in a rigid conformational state in both models (green), so at the temperature used in the dynamic (278 K) the secondary structure in this region doesn't move. The second studied region (230-255 aa) appears to be very flexible in the first dynamic (Dynamics2-magenta), which comes from a quite successful Energy Optimization. Surprisingly, the other dynamic, which wasn't successful at all in energy optimization, doesn't seem to be as flexible as in the first case. (Dynamics2; Dynamics3, code: 1-100 aa in green and 230-255 aa in magenta).
 

DISCUSSION
 

    Because the most important of a protein for this to be functional is its catalytic center we are going to follow the analysis by comparison of the important residues of the template with those found in the model, which is going to be definitely the obj2, the sequence based model.

    According to the article of reference (Buss et al., 2001), the template presents 4 important regions:
 
 
 

    The comparison of both structures (the template and the model) shows a conserved catalytic site, with some modifications in concrete residues, due to a shift between the sequences in the alignment:
 

• In Adenine binding site: A285 -> A283, I332 -> I327, I339 -> I334 and D283 -> D281
• In Ribose binding site: F284 -> M282
• In alpha-phosphate binding site: no changes
• In beta-phosphate binding site: H208 -> H206, N211 -> S209 and G212 -> G210
• In Acetyl phosphate binding site: H180 -> H179
• In Acetate binding site: F179 -> F178
• Cation binding site: no changes
• Phosphorylation site: E384 -> E377

    We observe that the important regions are conserved in the optimized sequenced based model.

    The next important feature to check in our model is the strange epsilon conformation in residue 325 (the Alanine found in position 330 in the template). Another superposition between these two structures gives us the confirmation of these absolutely conserved conformation among all the ASKHA Superfamily members.
 
 

CONCLUSIONS
 

   None of the strategies used is able to give a perfect model of the problem sequence, in terms of the energetic profile. The topits based model should have been the best, since it introduces structural information from the beginning, but we realise that small changes in the sequence, in non-functional regions of the protein, can greatly influence the end result. Nevertheless, given the high homology between both sequences (92%), the important features of the protein are well modeled (the ownership to the superfamily and the functional active site).
 

FURTHER STUDIES
 

   Despite the low cutoff, the remaining loops could be further optimized with the Archtype program.
   The temperature of the Dynamic analysis could be raised to allow some problematic regions to move and adopt a more stable conformational state.
   There must be other programs (¿?) or possibilities to achieve a better model, but our is a limited assignment.
   Another possible template has been determined recently by theoretical methods; the PDBfile has been released in the Database today, the 12th June 2002, by the group of  R. Sagajkar and R. Ramchandra (1LP2, Acetate Kinase of Pasteurella Multocida); a modeling including this second template could be of a great help. Unluckily, the article hasn't been published yet.
 

REFERENCES
 

Buss, Kathryn A. and Cooper, David R. 2001 Urkinase: Structure of Acetate Kinase, a Member of the ASKHA Superfamily of Phosphotransferases. Journal of Bacteriology 183: 680-686

TOPITS server

ARCHTYPE server
 

ACKNOWLEDGEMENTS
 

    To our devoted teachers in Structural Biology, specially to Nuria Centeno, for her patience and encouragement.
 

PEOPLE
 

Isabel Lorenzo Sánchez
Dominique Monferrer Ventura
Ángel Núñez Pagán