Re: [ccp4bb]: Refmac vs. cns: nucleic acids?

[Anastassis Perrakis <perrakis@nki.nl>]

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

Phoebe Rice wrote:

> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> I'd be interested in people's humble opinions on their experiences with 
> nucleic acids in particular.  Many of the backbone torsion angles aren't 
> well defined in moderate-resolution maps (2.5ish), but they do matter.  
> Especially for non-canonical structures, I've wondered if we're really 
> dealing with them properly.

Since nobody picked up on that yet, here I go ... i hope I will be 
humble enough [;-)]

The word is that XPLOR/CNS is far better than REFMAC on refining nucleic 
acids. REFMAC (.. and PROLSQ back then) make the sugar backbone 'funny' 
according to some, the bases non-planar according to others.

Partially true - I think.

For the non-planar bases some RTFM would suggest to adjust the WMATRIX 
value and for me and most others that works.

For the backbone though there is a point. I vaguely recall the York 
refinement workshop at '97. After some smart participant (Miquel Ortiz 
?) pointed it out, I recall Eleanor having a look at the all-famous 
PROTIN code around midnight and figuring that 'main chain' could only 
have one chiral center - which is fine with proteins but a bit of a 
disaster with nucleic acids if the ribose is considered the main chain ! 
There was a single line fixed that I used ever since but I doubt if that 
ever made it in the wide public.

The truth now is that REFMAC5 does for sure the work properly - as 
properly as CNS although slightly different if I understand it correctly.

The trouble start when having 'unusual' DNA, i.e. bent DNA deviating 
from B-DNA. CNS has a HUGE energy term dissallowing sugars to change the 
puckering amplitude - accordign to that it seems more likely to see a 
Ca-C bond at 1.0 A than a sugar torsion angle change by 1 degree .... 
So, once your sugar was in the C2-endo conformation (thats B-DNA, just 
looked it up) it was staying there and the DNA looked great.

At 2.5 A resolution and below everybody is happy and justifiably so.
DNA looks fine and you cannot really distinguish sugar puckers.

When I was refining the MutS structure at 2.2 A, I followed the 
mainstream and used CNS. However, at the place of the DNA bend (almost 
60 deg) I was always getting residual density near the sugar rings ...
I could fiddle with other torsions and get the base into density but 
there was some things I did not like. After some reading of W. Saenger's 
book, vague memories of puckering amplitudes surfaced up (from my dark 
past on cyclodextrins ...). The solution that we chose for modelling the 
kinked DNA in MUTS was to model some sugar rings around the bend in the 
(unusual) C3-endo conformation, characteristic of A-DNA. That was done 
after some fiddling with O dictionaries. Once I told CNS about it (you 
can nicely restrict the puckering amplitudes in whichever conformation 
you like through the interface), we got a much improved map and model.

At the same time Refmac5 had surfaced up and I got a test-drive. Refmac5 
would just refine the sugars to the correct C2-/C3- endo conformation 
since puckering amplitudes are not explicitely restrained but the 
allowed/disallowed values are figured out indirectly by VdW interactions 
(is that correct Garib ?). For reasons of personal bias I guess, we went 
on with Refmac5.

Well, I hope I bored everybody, my revenge for 3 days of reading 
chemistry and editing O dictionaries (I often wonder what was worse).


Tassos

PS ... Warren, I did not say ARP/wARP once either ....