It sounds as if there could be an error in the scale factor. If strong
scatterers persistently give large positive peaks, then the scale factor
for Fcalc is underestimated. Are there also strong negative peaks? If
the strong positives outweigh the negatives, try to input a few
different values of the scale factor without allowing them to refine
(+10%,+5%,-5%,-10%). It is possible that the method for estimating the
scale factor is not producing the best result. If this is the case then
adjustment of a single parameter might significantly affect the quality
of the crystallographic model. Also, have you used bulk solvent
correction to improve the low resolution and high resolution contrast match?
In any case, I applaud your willingness to explore issues that are
initimately connected with the quality of crystal structure
determination. Many times efforts are dominated by the urge to reduce R
and Rfree. These are useful indices of crystal structure refinement,
bit not the only measures of refinement quality. In the end, the
balance of fit to the experimental data and reasonableness in terms of
known stereochemistry and match to the known chemistry of your system is
the hallmark of correctness of crystal structure determination.
Kevin Jude wrote:
> These peaks just don't go away, although if I use only high resolution
> data (2.0-.99 Ang) the peak height drops a bit, and the peak shape
> changes if I use only lower resolution data (as would be expected).
> Occupancy or anisotropic B refinement in SHELX didn't make much
> difference, either.
> I'm also seeing similar (smaller, yet >3 sigma) peaks elsewhere in the
> map, covering other metal ions, sulphers, and even well-ordered parts
> of the protein backbone.
> Eddy Arnold wrote:
>> Hi Kevin,
>> Actually, thinking about it later, since this was an Fo-Fc map, there
>> shouldn't necessarily be strong ripples, at least not due to series
>> termination effects. The Fo-Fc map calculation, also known as the
>> "error synthesis", is not subject to great series termination effects
>> because there is a one-to-one correspondence between observed and
>> calculated terms. There could be ripples if the data were relatively
>> incomplete or there were a few very strong terms, but i doubt that
>> either of these cases would apply here.
>> The proportional contribution of a well-ordered heavy atom is greater
>> at higher resolution, since the scattering factor becomes dominated
>> almost completely by inner shell electrons. It might be worth
>> looking at Fo-Fc maps in a few resolution ranges to see if the large
>> difference peak comes largely from low, moderate, or high resolution
>> terms. Calculate Fo-Fc maps using data from 30-8, 30-5, 8-1.1, and
>> 5-1.1 Å. This would be largely for information's sake; I advocate
>> using all observed data in refinement although sometimes do a
>> low-resolution cutoff of 30 Å or so if the beamstop was not well
>> placed and some strong reflections are recorded as very weak because
>> they were blocked by the beamstop.
>> What were the results from anisotropic refinement of the Hg? Maybe
>> these suggestions are already moot...
>> Best wishes,
>> Kevin Matthew Jude wrote:
>>> Thanks Eddy-
>>> I've actually done a q refinement (99% occupied), but haven't tried
>>> anisotropc B's yet - maybe that is the ticket.
>>> As for the ripple effect - as I recall, I should see peaks of height
>>> +1/3, -1/4, ... due to the Fourier series - but shouldn't those
>>> peaks in
>>> Fo-Fc appear before I model the atom, rather than after? Or am I
>>> Eddy Arnold wrote:
>>> > Hi Kevin,
>>> > Sounds like a whopper of a difference peak; also sounds like a great
>>> > structure.
>>> > A few thoughts:
>>> > 1. You should refine occupancy of the Hg since it may be less than
>>> > 1.0. This is quite typical for heavy atom derivatives, even with the
>>> > heavy atom in the cocrystallization buffer.
>>> > 2. Occupancy and B factor are highly correlated at low resolution
>>> > at higher resolution the correlation breaks down and occupancy and B
>>> > factor refinement diverge significantly, especially beyond about 2
>>> > resolution.
>>> > 3. With the Hg being such a strong scatterer with high quality
>>> data you
>>> > should be able to refine the anisotropic B factor and possibly (most
>>> > likely) see considerable improvement.
>>> > and finally
>>> > 4. Even with high resolution data one tends to see "ringing",
>>> meaning a
>>> > series of rings surrounding a very strong scatterer which are
>>> > alternating positive and negative, because of series termination
>>> > in the Fourier synthesis. Often with a 5 sigma difference peak
>>> one will
>>> > observe rings around 1-2 sigma in magnitude--in this case that would
>>> > scale to about what you see.
>>> > Hope this helps; have fun with your refinement.
>>> > Best wishes,
>>> > Eddy Arnold
>>> > Kevin Matthew Jude wrote:
>>> >> Hello all-
>>> >> I am refining a 1.1 Ang protein structure, solved by MR. There is a
>>> >> (~80 sigma) peak in my Fo-Fc map. Due to it's position next to a
>>> >> residue, a corresponding peak in the anomalous difference map,
>>> and the
>>> >> presence of CH3Hg+ in my crystallization buffer, I have assigned
>>> it as
>>> >> (using the CMH residue found in HIC-UP). After annealing,
>>> >> and B factor refinement, though, there remains a large (25 sigma)
>>> >> peak in the same position. There is certainly no uranium in my
>>> >> crystallization buffer; what could I be missing?
>>> >> kmj
Professor Eddy Arnold
Center for Advanced Biotechnology and Medicine
Rutgers University Department of Chemistry and Chemical Biology
679 Hoes Lane
Piscataway, NJ 08854