Re: [ccp4bb]: High Solvent

[Randy J Read <rjr27@cam.ac.uk>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***

On Thu, 14 Jun 101 1:12:30 IST, shome@medinst.ernet.in wrote:
>
> Dear all!
>                      I have a protein structure which was solved
> by AMoRe using very identical structure (93% amino acid identity) as
> a search model. This data was diffracted to 4.0A and having one
> molecule with 80% Solvent content.But its not refining much and r factor
> got stuck at 34%. As i am very new in this field and do not know much
> about crystallography. I would like to know that how to estimate solvent
> content and no. of molecules in asymmetric unit. If it has two molecules
> in asymmetric unit than solvent cont. will be of about 67%. I am not able to
> get second molecule if it is so. How one can know exactly about solvent cont.
> if it is so high and can this structure be refined further.
>

The issue of model completeness comes up occasionally, especially with
molecular replacement, and we've found two approaches useful in practice:

1. Look for evidence of non-crystallographic symmetry (NCS) in your data.
Perhaps you've done this already, but you should always compute a
self-rotation function and a native Patterson map from data from a new
crystal.  The self-rotation function will tell you if there is any internal
symmetry reflecting the presence of multiple molecules in the asymmetric
unit.  However (and this is actually quite common), the NCS rotation axis
may be parallel to a similar crystallographic axis of rotational symmetry;
the result will be that the self-rotation peak is hidden under the origin
peak.  But another way to look at this is that the combination of NCS and
crystallographic symmetry (CS) in this case gives rise to translational
NCS.  (There's a more complete description, with a picture, in the web
pages of our lecture series:
http://www-structmed.cimr.cam.ac.uk/Course/MolRep/molrep.html#hidim.)
Translational NCS gives a big peak in the native Patterson map, from which
you can deduce the position of the molecule relative to the NCS rotation
axis.  Of course, if the NCS and CS axes are not exactly parallel, the
translational NCS will not be exact, and it may be that you have to 
restrict the native Patterson calculation to fairly low resolution (say
8 or 10 A) to see the peak.

2.  After finding a molecular replacement solution, compute a Sigma(A) plot
to find out how much of the ordered structure your model describes.  The
Sigma(A) plot (given by SIGMAA) should (if a number of assumptions are
satisfied) give a straight line with a slope that tells you something about
the rms coordinate error and an intercept that tells you something about
model completeness.  The estimate of coordinate error is probably useful as
a rough guide, but because of breakdowns in the assumptions I wouldn't take
the absolute numbers too literally.  On the other hand, I've been surprised
at how well the estimate of model completeness seems to work.  In practice,
the Sigma(A) plot drops off at low resolution (up to about 5 or 6 A)
because of the inadequacy of our models of bulk solvent, and the straight
line should only be fit with the higher resolution data (possibly ignoring
the very highest resolution terms, especially if you've been generous about
including data with poor signal-to-noise).  Of course, this doesn't leave
much to play with in your case, if you're limited to 4A resolution!  But
you might still find it instructive to make a Sigma(A) plot, and then see
whether you could distinguish between your two possibilities by which
y-intercept is more plausible.  The intercept of a Sigma(A) plot should be
ln(model completeness)/2, so for a complete model the intercept should be
zero and for a 50% complete model it should be -0.35.  The other thing
you have going for you is that your model should actually be quite accurate, with such high sequence identity (at least assuming that it was
refined well against reasonably high resolution data).  So there's a good
chance that the effective rms coordinate error should be well under 1A,
which is a big constraint on the lines you could fit.

One point: you should do the Sigma(A) plot after rigid-body refinement but 
before any further refinement of coordinates or B-factors, to avoid
over-fitting the working data.

We found this test useful recently in trying to assess whether 40% of
trypsin was indeed disordered in a complex with alpha(1)-antitrypsin.  The
Sigma(A) plots were strong evidence for disorder.  They weren't published
in the print edition, but can be found in Figure 9 of the supplementary
material for Huntington et al, Nature 407: 923-926 (2000).

Randy Read