Arg sidechains

[botosi@…]

Hi,

After simulating several EM structures with Isolde, it produces models with outlier Arg sidechains (non planarity of CD with guanidium group; and guanidium angles deviate from 120deg).
Not sure if the lack of these geometric restraints originates from chimeraX or Isolde.
It is reproducible, we had it in five different structures now.

Here is an exerpt from the output of the PDB validation tool:


Residue(model/chain/number/name)    Region not planar

1/A/118/ARG    sidechain

1/B/114/ARG    sidechain

1/B/118/ARG    sidechain

1/C/118/ARG    sidechain



Bond angle outliers



The following bond angles were found to be significantly different from the expected bond angle.



Residue(model/chain/number/name)    Atoms    Expected bond angle (°)    Measured bond angle (°)

1/A/114/ARG    NE-CZ-NH2    120.3 ± 0.5    124.5

1/A/222/ARG    NE-CZ-NH2    120.3 ± 0.5    123.0

1/B/54/ARG    NE-CZ-NH2    120.3 ± 0.5    123.0

1/B/66/ARG    NE-CZ-NH2    120.3 ± 0.5    122.9

1/B/114/ARG    NE-CZ-NH2    120.3 ± 0.5    123.9

1/C/57/ARG    NE-CZ-NH2    120.3 ± 0.5    122.9

1/C/110/ARG    NE-CZ-NH2    120.3 ± 0.5    123.1

1/C/114/ARG    NE-CZ-NH2    120.3 ± 0.5    122.9

Would it be possible to patch this up?

Thanks,
Istvan

[Eric Pettersen]

Reported by Istvan Botos

[Tristan Croll]

So this is (probably) a bit more of a nuanced discussion than you might be
expecting… the short version is that this is a property (deficiency?
Depends how you look at it) of the AMBER force field, and one of the two
reasons why I strongly recommend against depositing models straight from
ISOLDE into the PDB:

(1) AMBER is parameterised to replicate QM calculations on model compounds;
the classic restraint libraries used for validation use statistics derived
from very high-res crystal structures. As such there will always be
systematic differences; the bond angle outlier in the Arg sidechain is
simply the biggest of these.

(2) ISOLDE currently makes no attempt to refine B-factors.

Instead, I recommend using the “isolde write” command to write input files
for your refinement tool of choice (said files have settings aimed at
limiting the refinement to “just” tightening up bonds, angles and planarity
without attempting any more aggressive rearrangement). In practice you’ll
generally see next to no visual difference after refinement, but those
validation outliers should go away.

Now, the more nuanced point, mainly focused around planarity. The thing to
keep in mind that the AMBER forcefield used by ISOLDE (like essentially
every other MD force field) is extremely strict about van der Waals
interactions - for all intents and purposes, it simply doesn’t allow
clashes. This contrasts with classical refinement, which very heavily
prioritises local bonded geometry (bond lengths, angles, and planarity)
mostly at the expense of being much more permissive about clashes (and
ignoring entirely the whole concept of electrostatics). In practice that
means that residual errors in your model will manifest in different ways
than you might be used to - where in a classical environment you’d get a
clash, in ISOLDE you’ll instead see things getting pushed out of density
(where the surroundings make that possible) or distorted from ideal
geometry (where things are more constrained). In my experience the most
common explanations for an arginine going substantially non-planar are
either that the rotamer is wrong (think in particular about whether the NE
is pointing in the right direction) or some other nearby residue is in a
wrong conformation and putting pressure on it. A less common scenario that
only really becomes noticeable at high resolution is that the density is
some average of difference conformations and pulling it in unnatural ways.
Occasionally particularly highly charged nearby groups (e.g. phosphates)
can pull it out-of-plane because the implicit solvent treatment isn’t
perfect… in these cases the judicious addition of a water or two can make a
big difference.

Ultimately, what it boils down to right now is that all the available
restraint libraries and force fields have their deficiencies, but I believe
that the benefits of using the MD approach outweighs the downsides. While
I’d of course *love* to have a better force field that respects physics
while also better replicating the equilibrium distributions seen in
experiments, at present I’m afraid that’s well outside the scope of my
current resources.

On Wed, 28 Aug 2024 at 17:05, ChimeraX <ChimeraX-bugs-admin@cgl.ucsf.edu>
wrote:

>
>
>
>