Strain Energy: Who Cares?
My group here at Tripos just published a paper on the value - and complications - involved in avoiding ligand alignments that include strained conformations [1]. The procedure described there has been carried over, with some modification, to our new GALAHAD pharmacophore identification program. The underlying motivation for trying to drive pharmacophore models away from high energy conformations was complaints about such models from GASP users. In our validation work, however, it soon became clear that such bias led the program away from crystal structure alignments in several cases. The most conspicuous case is that of thrombin inhibitors highlighted in the paper by Patel et al. that compared the performances of DISCO, GASP and Catalyst [2]. Several inhibitor ring configurations seem quite strained in the co-crystal structures, according to both the Tripos and MMF force field calculations.
Of course, it is difficult in many cases to separate “true” strains from strained conformations that are artifacts of crystallographic approximations. Energy calculations involving bond lengths and angles, for example, are much more sensitive to small errors in coordinates than are torsional energies. Their average values do not match the average distances and angles given by the time-averaged atomic positions reflected in the electron densities, even if the diffraction patterns could be obtained and solved to a perfect resolution. Such deviations can exaggerate strain directly, but they can introduce second-order errors in torsional and steric strain calculations as well.
The contribution of the electrostatics to strain is difficult to impossible to estimate for the many ligands - including the thrombin inhibitors mentioned above - that bear a net positive charge at physiological pH. This is particularly problematic for solvation models, since it is often not clear what “solvation” means within the active site, and whether methodologies that accurately calculate solvation energies are directly transferable to calculating desolvation energies for active sites. Might these free energies offset the “cost” of ligand strain?
Then, too, force field calculations only take energies into consideration and fail to account for entropies of binding. It is intuitively appealing to posit that the conformational free energy “cost” of binding should be low for ligands that bind tightly. The simplest way to accomplish this is for both components - the entropic and enthalpic - of the total free energy to be favorable. Though this seems an eminently reasonable assumption for “natural” ligands for which the specificity of interaction has been optimized by evolution, it is not obvious why it should apply to complexes with artificial ligands in general and for “lead” structures in particular. Free energy contributions from induced fit effects on the protein - i.e., the effect of protein flexibility [3] are likely to be dominated by entropy, and these are generally ignored altogether.
Yet even allowing for all those reasons that strain energies shouldn’t be a reliable guide to bound conformations, my gut feeling makes me want to agree with most modelers that it does need to be taken into consideration somehow. This is in part a subjective judgment, in that low-energy conformations just tend to “look natural” somehow, and vice versa. I see no clear objective justification for that feeling, except for the obvious need to avoid clear-cut steric clashes.
On balance, then, I do not believe any strain calculation should be taken too seriously, especially for flexible models where entropic effects may become critically important [4]. Rather, I am inclined to bias against strained conformations rather than to eliminate them from consideration altogether. But I would be happy to entertain and disseminate - alternative viewpoints on the matter.
Bob C.
1. A. STRIZHEV, E. ABRAHAMIAN, S. CHOI, J.M. LEONARD, P. WOLOHAN & R.D. CLARK (2006). The effects of biasing torsional mutations in a conformational GA. J. Chem. Inf. Model. 46, 1862-1870.
2. Y. PATEL, V.J. GILLET, G. BRAVI & A.R. LEACH (2002). A comparison of the pharmacophore identification programs: Catalyst, DISCO and GASP. J. Comput.-Aided Molec. Design 16, 653681.
3. S.J. TEAGUE (2003). Implications of protein flexibility for drug discovery.
Nature Rev. Drug Discovery 2, 527-541.
4. E. PEROLA & P.S. CHARIFSON (2004). Conformational analysis of drug-like molecules bound to proteins: an extensive study of ligand reorganization upon binding, J. Med. Chem. 47, 2499-2510.
4 Responses to “Strain Energy: Who Cares?” 
Leave Your Comments ...
You must be logged in to post a comment.
August 11th, 2006 at 2:45 am
To support Bob’s suggestion, let us remember two important points that computational chemists often overlook:
(1) molecules are not points with lines between them, nor steric/electrostatic/hydrophobic fields, nor for the most part are they found existing alone in vaccum
(2) force fields have no physical existence, but are simply convenient computational tools that work well in the artificial world of point atoms…
In that context, the choice to use or not use “strained” conformations ought to be be based on experience in whether the results that arise from using them are useful or not in selecting or designing new molecules.
August 15th, 2006 at 2:20 pm
So, to ask a stupid question, has anyone proflied the CSD and PDB databases to ascertain what proportion of ligands bound/unbound are ’strained’?
August 15th, 2006 at 8:36 pm
Comparison of the ligand bound conformation (possible strained) with the expected free ligand conformation (probably unstrained) is a dangerous pursuit. There are two important factors, i) the flexibility of the ligand, i.e. the type of interaction - lock&key or co-operative, and ii) the pharamacology involved, i.e. agonist, antagonist etc. If the ligand interaction process leads to a biological process that is the end result of a cascade of protein interaction, the energetics of the ligand conformation will be a trivial part of the total energetics of the ligand/proteins complex. Those ligands that inbibit or block the activation process may have to assumed strained conformers to prevent this happening. Always worth thinking about the biological context before interpreting the chemistry.
September 5th, 2006 at 3:04 pm
Authors who’ve attempted to answer this issue inculde
Emanuele Perola (referenced above) an extensive piece of work suggesting that only 60% of ligands bind in non-strained conformations
and the earlier work of Jonas Bostrom (http://www.springerlink.com/content/p744211565g1k228/) a study that suggested again that only ~70% of ligands are bound in low energy conformations. More notably, they note that the high energy structures are explainable by either poor Xray data or poor conformational parameters.
My personal experience is that I’ve never seen a strained conformation - and frequently been humbled to find that what initially looked like a strained conformation, was actually present in multiple CSD structures and reflected my limited feel for conformational niceties and the error of relying solely on conformational generation or minimisation programs.
So, back to the practicalities of conformational overlays: I’d prefer to first see only non-strained conformations overlayed (assuming that the force field is appropriate), and then select to add in the more strained superpositions.