Hello Edward,
Thanks for the reply.
I have compared the results from the local_tm global model to the selected
diffusion tensors for both light and dark states. In both cases the final S2
values are in reasonable agreement with the local_tm model S2 values. For the
light state, the local_tm S2 valus are around 0.17 on an average as well.
Both the states of the protein are dimers in solution as far as DLS, SAXS and
analytical size exclusion chromatography has illustrated. Especially the
light state. We tried measuring Kd of the dimerization and it is lower
nanomolar range. For the relaxation data acquisition I have used 0.5mM
samples. For the dark state I have observed that it has a tendency to form
some form of oligomer in solution at concentrations ranging from 1mM to 10
uM, the tendency decreasing with decrease in concentration. The light active
state does not show this at any concentration. I tested this using comparison
of the product of R1 and R2 rates, which is viscosity independent, between
multiple concentrations. This actually makes the light state results even
more puzzling. I will go back and check if by mistake I have forgotten to
remove data sets for certain overlapped resonances and/or re-record the
experiments.
Regards,
Vineet
________________________________________
Cc: relax-users@xxxxxxx
Betreff: Re: Model free on symmetric homodimer, problems with grace plots
On 7 October 2015 at 09:43, Panwalkar, Vineet <v.panwalkar@xxxxxxxxxxxxx>
wrote:
Hello Edward and relax users,
I have been recently using relax to carry out model free analyses on two
different states of my protein at 600 and 800 MHz. I am working with a LOV
domain (132 residues) which exists in a ground "dark" state and a
photoactive "light" state. Both states exist as a symmetric homodimer in
solution. For the analysis I have been using the crystal structures, one
monomer at a time. Also, the resonances showing overlap have not been
included in the analyses. There is no difference in the backbone as such
between the two crystal structures. Along with model-free I have also
carried out reduced spectral density mapping. The J(0)s of both the states
agree well between the two fields. Hence there is nothing wrong with the
data as such. Also both the J(w) agree well with my het-NOE and R1 with
respect to regions in the protein showing fast timescale backbone motion.
The model free analysis of the "dark" state goes fine. The S2 values show
same regions having fast timescale motions as observed from the raw
relaxation data and rSDM. Both the monomers individually give me
approximately comparable results. For e.g., a loop between residues 100 and
110 shows flexibility on fast time scale according to the raw data, rSDM as
well as relax S2 values.
However, for the "light" state the results from model free appear quite
bizarre for both the monomers. The average S2 for the "dark" state was ~
0.8. As per my raw data and rSDM, I expected the S2 values to be slightly
elevated in "light" state. However, my "light" state S2 values are on an
average 0.17! Also the same loop is now with higher than average S2 values
in the "light" state whereas the raw data and rSDM showed the loop to be
flexible, albeit less flexible compared to the "dark" state. The internal
motions (Te) are less than 5 ps and seen only in 7 residues. No Ts or Tf. I
have attached figures from the spectral density mapping as well as the
"dark" and "light" state S2 values to illustrate better what I am trying to
say.
Also, I repeated the "dark" state model free analysis with "dark" state
structures. thinking that there might be some bug somewhere and got results
just like earlier. But not with the "light" state. Since the backbones
between two states are near identical, I used the "dark" state structure to
run with relax on the "light" state data. I again get these bizarre
results. I am about to run model free with "light" state structure and
"dark" state relaxation data. However, I am not sure where the problem
lies, whether it is the structure, my data or some bug somewhere. Since the
J(0)s add up nicely, I do not think it is the data. It is unlikely that it
is bug since the "dark" state ran fine. It is also unlikely that the
structure is strange since dark and light states have near identical
backbones and there are no missing residues in either of the structures. I
must also add that the correlation time obtained from relax are actually
comparable with the ones extracted from R2/R1 ratio for both the "dark" and
the "light" states. It is only the motion parameters which are weird for
the "light" state.
Has anyone encountered something like this recently? Any help/advice here
will be very helpful. At this moment the data is only with two fields but I
do plan to run relaxation at 900 MHz and use the three fields for
model-free as soon as I get time on the magnet.
Hi Vineet,
One thing you must consider is if the approximation of a single
ellipsoidal diffusion tensor is reasonable for the system. Sometimes
this is not the case. For example if the single rigid body assumption
is violated, if the dimerization is not 100%, or there are long
flexible tails (breaking the 'free' rigid body assumption). If you
have been using the model-free analysis protocol I came up with in:
d'Auvergne E. J., Gooley P. R. (2007). Set theory formulation of
the model-free problem and the diffusion seeded model-free paradigm.
Mol. Biosyst., 3(7), 483-494. ( http://dx.doi.org/10.1039/b702202f )
d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR
dynamic models II. A new methodology for the dual optimisation of the
model-free parameters and the Brownian rotational diffusion tensor. J.
Biomol. NMR, 40(2), 121-133. (
http://dx.doi.org/10.1007/s10858-007-9213-3 )
then there is a good check you can perform. Compare the final results
to the results of the local tm global model. The local tm model is
almost always over fit, hence the balance of variance vs. bias will be
tipped in the direction of variance (it will be noisy). However the
local tm global model will have a lot less bias. In practical terms,
this means that the model will not be affected by the assumption of a
global diffusion tensor (note that it is however affected by the
assumption that all diffusion tensor exponential decay terms are equal
to one term, which can sometimes be a problem). But if the S2 values
in the local tm global model and the final selected model are not
close to each other, then this is an indication of model failure:
d'Auvergne, E. J. and Gooley, P. R. (2006). Model-free model
elimination: A new step in the model-free dynamic analysis of NMR
relaxation data. J. Biomol. NMR, 35(2), 117-135. (
http://dx.doi.org/10.1007/s10858-006-9007-z )
From what you describe, it sounds to me like a classic case of model
failure. Either due to the diffusion tensors being insufficient for
the protein system, due to some bad data causing the model to optimise
to a state of failure, or due to some other cause that you will have
to investigate yourself. Anyway, I hope this was not too complicated
and will help point you in the right direction.
Regards,
Edward
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
AW: Model free on symmetric homodimer, problems with grace plots