On Thursday 28 January 2010 09:57:31 John Badger wrote:
> Hi Ethan,
> Your effort to play devils advocate is appreciated - I have not seen much 
> debate on the applicability of TLS so at risk of diverging into a separate 
> thread:
> 
> I would not argue that TLS is necessarily a valid model for correlated motion in 
> proteins because it 'works' in fitting the available Bragg data (or B-factors) 
> which only measure the amplitudes of the atomic displacements.

It is not true that only amplitudes are fit.  The TLS analysis used by TLSMD
is perfectly capable of analysing fully anisotropic ADP models from atomic
resolution refinements.  The TLS models it generates in this case go a long
way for accounting for the ellipticity of individual ADPs in the fully
anisotropic model.

> Suppose all atoms have about the same temperature factor. In this case we 
> get a perfect TLS model with just the T and have the whole protein rigid and 
> moving in perfect correlation. 

Yes. So?  Then you have a TLS description that reduces to being almost isotropic.
Also I would be careful about referring to a TLS description as implying
"perfect" correlation.  Remember that there is an arbitrary phase angle involved
for any given pair of atoms.

> Equally well we might think that all atoms are  
> independent oscilliators in similar microenvironments with there is no 
> correlation at all (this is the Einstein model of a solid!). Seems like the TLS 
> model will always tend to lump motions into an overall T and overestimate this 
> component. This thought experiment can be extended with atoms on the 
> outside having more motion just because they are freer to move and not 
> necessarily indictating that the protein is oscillating about a center (LS).

Except that this simpler model fails to account for the observed directionality
of the elliptical ADPs, whereas the TLS model successfully accounts for it.

> TDS is definately difficult to analyze because for normal globular proteins it 
> tends to look like other 'uninteresting' sources of background scatter and is 
> difficult to separate out (I think Colin made this point). However, I'm not sure 
> that I would agree that the jury is really out on the simple TLS models since 
> there are a lot of decent biophysical probes of correlated motion in proteins 
> that should be able to support this picture if it were correct. These include 
> inelastic neutron scattering experiements (Cusak), Mossbauer experiments 
> with Fe in myoglobin (Parak) and measurements of backbone dynamics on 
> interleukin-1b by NMR (Clore). It seems to me that these studies all indicate 
> that ambient motions in protein are dominated by fast, local, rattling 
> movements. In particular, the measurements of normal mode spectra have the 
> lowest order modes (equivalent to the TLS parameters) as relatively small 
> contributors to the overall motion.

But all of those "fast, local, rattling movements" must necessarily occur
within an envelope determined by larger bulk motion trajectories.
I guess I don't see why there is any inconsistency here.

You might also consider that the variation between individual copies of
the molecular described by TLS in a frozen crystal is largely due to the
freezing out of microconformations sampled from a continuum of states
in solution.  But those fast, local, rattling movements are still free
to relax to the mean during/after crystallization, whereas larger
scale displacements may be trapped in local minima of the lattice packing
energy function.

	Ethan











> Of course I appeciate the attempts to advance the crystallographic 
> refinement models for B-factors beyond the simple local restraints devised by 
> Konnert and Hendrickson over 30 years ago and that we still use almost 
> unchanged!
> 
> Cheers
> John
> 



-- 
Ethan A Merritt
Biomolecular Structure Center
University of Washington, Seattle 98195-7742