Thank you everyone for your replies. The Nayal & Di Cera (1996)
paper may be what I had in mind. I was looking for some estimate of
how often water atoms are placed in protein models where they do not
belong, and I expected a relatively high percentage. Simply
extrapolating from the 0.01% water <-> sodium misassignments in that
paper, the problem does not appear as significant as my intuition told
me.
Here is what started our lab's discussion:
1. crystal structure of 1000 aa residues with approx. 60% solvent
content, a little better than 3A resolution, mean B factor approx 75
A**2.
2. approx. 50 Fo-Fc peaks > 4*sigma in very reasonable polar
environments, BUT many of them with coinciding with 2Fo-Fc density
ONLY when contoured at <1*sigma
3. an anomalous difference Fourier map (calculated with a high
resolution limit of 3.5A) shows >3*sigma peaks only for some metal
ions that I know are present in the structure.
My initial argument, outlined below, rests on these assumptions:
A1. a conservative 10% of "waters" in a typical crystal structure are
really something else. I have no evidence for this, just my own
experience of the rigor I apply when modelling waters.
A2. 0.5-2 waters per aa residue. Levitt&Park 1993. Structure 1: 223-6
How should we treat/model the Fo-Fc peaks for this specific example? I
argue to either ignore them or model them as unknown atoms. They could
be noise. But if they are not, they are among the strongest
diffracting 10% of the expected number of solvent atoms. I have
reasonable doubt that they are all waters. Heck, based on my
assumptions, many of them probably ARE NOT water.
I welcome your opinions.
Wolfram Tempel
