Modeling ligands in binding pockets when the density is weak.

From: Francis E Reyes
Date: 23 August 2011 19:36

Seems to be a quiet day on the BB, so I propose this question:


Suppose you have a ligand in the binding pocket and some mediocre data (3 A or so), the 'core' of the ligand is well defined in 2Fo-Fc map  using the model phases of your protein, however there are 'chains/tails' of the ligand which are not.  Composite omit or simulated annealing omit maps do not produce density for these 'chains'

The question here is how the chains/tails should be modeled (if at all).


[1] Model in the core, but remove the atoms for the chains  (and conclude the diffraction data do not support interactions with the protein and subsequent experiments are needed (higher resolution data, biochemical data, etc)).

or

[2] Model in the chains/tails noting that potential hydrogen bond donors/acceptors on the protein are within hydrogen bonding distance to the chains/tails. You do this and subsequent refinement still does not produce the expected density for the chains.


or

[3] Your solution here.



 If this situation has been discussed before, please let me know .

F

---------------------------------------------
Francis E. Reyes M.Sc.
215 UCB
University of Colorado at Boulder

----------
From: Ed Pozharski

So you need an earthquake :)

This is similar, imho, to the issue of disordered side chains:

https://docs.google.com/spreadsheet/gform?key=0Ahe0ET6Vsx-kdHVNa3VodUtfbVQtZ2pnUFcxQkx6RHc&hl=en_US&gridId=0#chart

https://docs.google.com/spreadsheet/viewform?hl=en_US&formkey=dHVNa3VodUtfbVQtZ2pnUFcxQkx6RHc6MQ#gid=0



--
"Hurry up before we all come back to our senses!"
                          Julian, King of Lemurs

----------
From: Bosch, Juergen

(3a) You could give GraphENT a shot first and see if you can do magic on visualizing the remaining bits of density.

You could also have multiple conformations, try refining with a low occupancy first and see what you get back.

(3b) use AFitt

Jürgen

P.S. I would set the occupancy to zero for those parts which you don't see density but I would never remove them unless you know that your ligands has been degraded.

......................

Jürgen Bosch

Johns Hopkins Bloomberg School of Public Health

Department of Biochemistry & Molecular Biology

Johns Hopkins Malaria Research Institute

615 North Wolfe Street, W8708

Baltimore, MD 21205

----------
From: David Schuller

Had one already, thanks.

--
=======================================================================
All Things Serve the Beam
=======================================================================
                              David J. Schuller
                              modern man in a post-modern world
                              MacCHESS, Cornell University
                              

From: Edward A. Berry


Apparently sent from the vicinity of U. Maryland and JHSPH, thanks.

----------
From: <Herman.Schreuder


Dear Francis,

Although I am a member of the "never truncate a disordered side chain"
camp, I think for ligands it is quite a different story.

For me a random disordered lysine on the protein surface is completely
uninteresting, except if one wants to examine the electrostatic surface.
A non-expert end-user is either not aware of truncations and ends up
with wrong results, or has to laboriously put back the side chains the
crystallographer laboriously had removed before.

However, a bound ligand is very different. Determining its binding mode
is usually THE goal of the study and every hydrogen bond with the
protein is discussed in great detail. Chemists and modelers use these
structures to design more potent ligands and theoreticians use these
structures to improve their force fields. Also biochemists and
biologists may be tempted to draw all kind of conclusions about
important interactions where in fact there may be none.

Especially when the ligand has designed to make a certain interaction
and in absence of experimental data you model the same interaction, the
chemist and modeler will be very happy and immediately jump on it to
design more of the same. I have seen many cases where theoretically, the
ligand would be able to make a wonderful interaction with the protein,
but that flexible side chains on the protein or ligand just did not want
to give up their freedom (entropy) to become locked in such an
interaction. Not seeing flexible parts of a ligand is not resolution
dependent. At higher resolution you see even less of the flexible parts
since there is less model bias possible.

So my approach: If there is weak or even very weak but real density for
the flexible parts of the ligands (I usually scroll down to 0.6 sigma),
I build the part, or build 2 or 3 conformations in case of discrete
disorder. Here I think I take more liberties then most of my colleagues.


However, if no convincing (weak) density is present above the noise
level I remove the undefined parts and do not even consider to leave
them in with occupancy zero. They will appear on the display of the end
user and give the impression that an interaction is present where there
is none. If the ligand would make the interaction, it would be visible
in the electron density maps.

My choice is definitively [1], but before rushing to get more
experimental data I would first put some brain power in it: maybe it is
better for binding not to fix certain flexible side chains (less entropy
loss), flexibility may endow the protein with broader substrate/ligand
specificity, there may be crystal contacts which prevent the correct
binding mode, components of the crystallization buffer may interfere
with proper ligand binding etc.

Best,
Herman

----------
From: Ed Pozharski

The second link above happens to point at the form editing page or such,
so please disregard (my googlebox is getting flooded with requests to
share it).  If you want to cast a belated vote, use this link

https://spreadsheets.google.com/viewform?hl=en&formkey=dHVNa3VodUtfbVQtZ2pnUFcxQkx6RHc6MQ

To see the results, see this one.
Ed.

--
Oh, suddenly throwing a giraffe into a volcano to make water is crazy?
                                               Julian, King of Lemurs

----------
From: Dale Tronrud


  I agree with Herman, and would like to add to his list of
explanations why density may not be observed.  It is possible
that the compound binding to your protein simply doesn't contain
those bits without density.  I have known cases where the
compound in the vial does not match the label on the vial.  In
addition I've had cases where bits were cleaved from the compound
at some point before binding.  Sometimes you can't see it because
it simply isn't there.

Dale Tronrud

----------
From: Shya Biswas

Hi Francis,

Once I had asked Pavel Afonine the same questions and these were his suggestions but most of these can be implemented in phenix...

I guess there is no general/unified procedure to do this, and in most of cases the tools and outcomes vary case by case.

Some general points:

- Removing parts of model is unlikely to improve the map simply because this makes model even more incomplete. The main purpose of computing the omit map is not to improve the overall map but to verify the atoms in question.

- Typically, the density appears weak because the signal is buried in noise. For example, we do not see H atoms in 2A resolution maps not because the information about them is not there (note, H atoms being weak scatterers contribute the most to low resolution reflections - similarly to bulk solvent, and their contribution to high resolution data approaches zero), but because the model quality and therefore noise at that resolution such that it hides hydrogens' signal. However, we do see H atoms at high resolution (say 1A and higher) not because of presence of high resolution reflections, but because the model quality is typically high and the noise level is below the hydrogens' signal.
Having said this, one possible way of improving your "weak density" is to improve the model as much as you can: make sure you modeled all alternative conformations, all solvent, etc.

- B-factor sharpening may help, although keep in mind that it will enhance the noise too.

- Other options: kick maps, omit kick maps, b-factor sharpened kick maps...

- You can try GrowDensity method (Acta  Cryst. (1997). D53, 540-543) which is available as phenix.grow_density. This is still under (slow) development, so if you decide to go this route than I can help you with the details.

Pavel.

HTH,

Shya