While working with a low-affinity, nonspecific DNA binding protein, I 
ran into this problem dishearteningly often. In several cases, even 
when running washed crystals out on a gel had told me that they 
contained both DNA and protein (32P-labeling for DNA, Coomassie for 
protein), I was unable to obtain convincing difference density for 
nucleic acid. As others have suggested, this could be caused by either 
low occupancy or multiple binding registers in the same crystal. One 
way to overcome the affinity issue might be backsoaking the crystals 
into a lower salt concentration, though this method was never 
successful for me. You could also continue your screening efforts and 
cross your fingers for a crystal form where the DNA is better ordered.

In the end, I developed a screen ("FASTDXL") based on Greg Verdine's 
protein-DNA disulfide crosslinking chemistry to tether the nucleic 
acid outside the binding site, thereby greatly increasing the local 
concentration and fixing DNA in a unique register (Structure 2007, 
15(7):773-80; NSMB in press). If you decide to pursue something along 
those lines and have success obtaining a structure, be sure to 
rigorously validate its biological relevance. Feel free to send me an 
email if you have any questions about FASTDXL.

Jacob

Jacob Corn
The Berger Lab
UC Berkeley - Molecular and Cell Biology
[log in to unmask]
phone: 510-643-9491
fax: 510-643-9290

Melody Lin wrote:
> Dear all,
> 
> I've been working on a series of DNA-protein complex structures. In my 
> recently acquired data sets, I got almost no density for DNA if I do 
> molecular replacement or rigid body fitting with the protein structure, 
> although I am 100% sure I have DNA in the structure by indepenent means. 
> If I use models with DNA, I could find some DNA density with those data 
> sets, but as I refine the structure, the density became very poor. The 
> resolutions for those data sets are between 2.0-2.4 A.  Also, if I use 
> the scaled data from synchrotron rather than the re-scaled data at home, 
> I got better DNA density, although for re-scaling, I used site 
> parameters that I copied done from synchrotron. The only differences 
> between those two sets of scaled data are: (1) the original scaled data 
> take into account all reflections, including high resolution data with 
> low completeness/redundancy, which are cut in the re-scaling; (2) error 
> models were changed so chi squares for each bin are 0.8-1.2 for re-scaling.
> 
> My (very naive) questions are: (1) Does the DNA density I saw in the 
> cases where I use models with DNA for MR/rigid body fitting only reflect 
> model bias? (2) are simulated annealing or cycles of coordinate/B factor 
> refinement enough to get rid of model bias? (3) Does weak DNA density 
> have to do with data processing?
> 
> Thanks very much for any suggestion,
> Melody Lin