Hi,
As you are working with complexes, then maybe an issue is that a
single diffusion tensor is not an adequate representation of the
system, resulting in the model m0 appearing more than it should. This
might be the case if the complex is not tight and you have a mixture
of complex and free monomers. This has been looked at in Schurr et
al, 1994, but no one has come up with a solution to this problem for
model-free analysis. Maybe you could be the first ;)
I'm not done with the analysis of all of my complexes, but I fear that even
with everything done "correctly" there will be "m0" all over the place and
I don't know how to interpret this in terms of mobility. Judging from the
runs I did until now, especially the interesting (i.e. probably more
mobile) regions of the more interesting protein show this behaviour. As I
said, I have quite large areas that disappear from my spectra from one
protein variant to the other – so this is an indication for exchange
mobility in this regions which is interesting for itself! Neighbouring
regions have a lot of "m0" (in 62 of ~220 assigned residues minus 28
unresolved) and in the ellipsoidal diffusion model there is also a lot of
strange Rex = 0.0000 terms, the other models show Rex of around 10^-18
(=nearly zero). Convergence is reached in 20-30 rounds for each diffusion
model, no oscillations are visible.
Is this Rex in the results file or in the extracted version? Note
that relax stores Rex internally as the field strength independent
value of:
sigma_ex = Rex / omega**2
Hence the value in the relax state files will be on the order of
'super tiny'. You need to multiply omega_N squared to obtain the
value you would expect on a given spectrometer. Also note that
currently in all model-free software, Rex is assumed to be fast and
hence scales quadratically with field strength - this might be another
source of problems for your analysis.
The current data are not perfect, as the necessary (!) R1 temperature
compensation was not used yet and also no soft pulses. So obviously I have
re-record some of the data. I used only one single sample, which was pretty
stable over the time I measured (no visible precipitation, but very slight
decreasing TROSY intensity). The temperature is off by less than 1 K
(remember our fucked-up but-now-apparently-fixed calibration procedure).
The consistency tests returned a fairly centered distribution (ratio of j0
at different fields: 0.993 +/- 0.174) of moderate consistency ("j0 test"
(field1-field2)/field2 = 0.08).
R1 temperature compensation is generally not needed as it is quite a
cold experiment, hence will almost always match the normal
spectrometer calibration. But single scan interleaving is a good idea
to average changes which occur during the experiment (for example day
and night temperature fluctuations which always occur to some extent).
Everything else seems fine.
That said, I don't see so overwhelmingly much of these stark m0 effects in
the protein I expect to be more rigid, although I have only a dataset wich
is highly inconsistent due to large temperature diffences, that was much
less stable used only old-school experiments with hard pulses have been
used.
m0 almost never appears for rigid, well behaved proteins as the
dynamics is easy to extract. The m0 is a sign that something or some
process is hiding the dynamics in the relation data. I.e. the single
diffusion tensor with internal model-free compatible motions is not
adequate. Or that the data for a spin is too noisy because your
system is so big.
My SH3 testing data don't show this kind of behaviour (no m0 at all), but
these have incredibly fat signal. Having a "real" protein changes a few
things I guess, especially in terms of S/N.
If m0 appears in SH3, I would be worried. In a real protein though,
the data for some residues can be rubbish, hence m0 is very valid as
the dynamics data is no longer present. m0 then tells you that you
have 4 grey pixels ;)
Maybe it's because of more complex motions. Maybe I should have gone for
relaxation dispersion in the first place. But "one step after another"
seemed reasonable at that time. (I'm currently quite desperately looking
for an introductory review like Séb Morin's "practical guide" for
relaxation dispersion – do you know one?)
Relaxation dispersion might be interesting, but from what you describe
I don't think dispersion data will tell you much other than what you
already see with weak peaks. Actually, as your system is 45 kDa, I
would not expect that you would see much dispersion at all - your weak
peaks are due to protein size and not Rex. As for a guide about
relaxation dispersion, I know no equivalent to Seb's guide. There are
some reviews from the Art Palmer and Lewis Kay groups which could be
useful. If you do find something, I'd be interested to have the
reference.
Maybe this relates to model m9 in relax. Sometimes the very weak
peaks, broadened by chemical exchange, are too noisy to extract
model-free motions from. This is visible in relax as the selection of
model m9. In such a case, model m0 will probably not be picked.
I excluded the really noisy/weak peaks beforehand and m9 gets picked
sometimes (9 times m9 opposed to 62 times m0 out of ~220 picked signals).
How many did you exclude? There is no need to exclude such peaks as
the protocol I developed will handle this. That is of course unless
you have partial complexation. If binding is slow, you have the
superposition of two diffusion tensors. If this is on the nanosecond
timescale, then you have a very, very different beast - two convoluted
diffusion tensors. If you could solve this issue for the model-free
problem, at least for the slow case, you should make quite a name for
yourself in the NMR dynamics field!
I don't know if this is completely relevant to your question, but
noise is another issue which affects the reliability of the te
parameters. As te increases, so does the errors. [...]
Whereas
noise shifts parameter values around randomly and governs which
motions are statistically significant, bias on the other hand shifts
everything in one direction.
So do you think if my data are too noisy this could be a consequence? I
already reached the limit in terms of scans, protein concentration and
measuring time. Maybe I should write a grant for two new magnets ...
For the spins where m0 are selected, do their errors look larger than
the other spins? Or if you plot the I0 values from the relaxation
exponential curve-fitting, are these residues much lower than the
rest? Maybe you could try out reduced spectral density mapping (very
easy in relax) and compare errors for the J(w) values. As for new
spectrometers, having more data would certainly help. Especially if
you have this mixed diffusion tensor problem and have derived a
solution to test - then having data at 3 or 4 fields would be
incredibly powerful. Maybe you should get your boss to talk to
Griesinger ;) But if you have data at two field strengths, that
should be sufficient.
Bias could probably in some cases
hide motions, but more likely will result in artificial motions. Bias
could also be introduced if the spherical, spheroidal, or ellipsoidal
diffusion is too simplified for the system or if partial dimerisation
is occurring.
That would be the jackpot of course – throwing over all the work we did
just to find out that the simplistic diffusion models don't fit our system.
:D
Well, you have to prove that. That will be difficult. You would need
to develop theory, code it, test it (with test models, for example
Schurr's published data), then compare it to the other global models
which come out of the dauvergne_protocol analysis using AIC model
selection. There is a slight risk that such a mixed diffusion tensor
model may not be statistically significant. But I think the
techniques I have developed and are present in relax (the optimisation
protocol, AIC model selection, etc) should significantly help such a
problem. But then maybe by comparing the individual model-free models
- maybe the presence of m0 and m9 in the simple model and absence in
the mixed model. One advantage is that from a new mixed diffusion
model is that you would obtain the proportion of complex verses
monomers (timescale information for a slow binding model would be
absent).
The 10 seconds for the NOE seems a little excessive (unless this is an
IDP or very small protein). Have you quantified the time required for
recovery?
I didn't really "quantify" in terms which delay length is the minimum I can
use, but tested the recommendation of 10s by Lakomek/Bax for TROSY-based
sequences and deuterated proteins (from the paper I cited earlier). There
they reported that for their system they got identical values for
HSQC-based readouts and TROSY-based readouts if they used the mentioned
precautions.
Oh, it's deuterated. Ok, then you'll need much more time. Though as
it is 45 kDa, the relaxation should be nevertheless relatively fast.
You can quite easily test this if you are curious. You can simply
compare 1D versions of your 2D experiments.
I tried "traditional" non-selective pulses and 3s interscan delay vs.
selective pulses and 10s with the same 45 kDa protein complex and saw large
differences in HetNOE values. Before, I tested also SH3 with different
combinations of soft/hard pulses and length of interscan delay and the
trend was that with non-selective hard pulses you get higher HetNOE ratios
(sometimes > 1) and the longer the delay is the lower the HetNOE ratios
values get.
Actually, the 45 kDa size could be the reason for the m0 model being
selected. It could simply be too big, the relaxation from tumbling
could be so fast that the internal motions have been hidden in the
data. As for SH3, the relaxation properties will be quite different
to your system so you can't really use it to determine d0 times for
the NOE.
As for the peak picking and fitting, [...]
I do it quite similarly, except that CCPN analysis always searches for
maxima and I always pick positive noise (except there is no maximum, then
it picks noise at the reference position). My workaround is to set the
boundaries of the "search box" to 0 that the crazy searching algorithm
doesn't let the peaks wander around too much. Contrary to what one would
expect the routine still looks for maxima. If you asked me, I'd say that's
pretty broken, but on the mailing list they weren't really open for
discussion on that matter. But after all the difference should be tiny and
not significant for my problems.
Ah, this is a known problem. I have talked to people before about
this - relax users have been bitten by this problem before. I can
only recommend switching to Sparky for this type of analysis.
Shifting peaks to the top of the noise is incredibly bad, especially
if your are talking about a 45 kDa system. This might actually be the
source of model m0. This shifting to the top each time introduces a
bias - you will end up with a double exponential:
http://article.gmane.org/gmane.science.nmr.relax.user/125
http://article.gmane.org/gmane.science.nmr.relax.user/96
The first exponential is from the relaxation, the second which is the
pure bias term is from the noise. The result of this bias, especially
with a big complex like yours, could be that model m0 is selected.
Note that CCPN Analysis is almost never used for relaxation studies
and, as you have experienced, for good reason. I really recommend
Sparky and then using the special technique I described in
http://article.gmane.org/gmane.science.nmr.relax.user/96. Then
compare these rates to your CCPN rates. If you try that, you might
see all m0 models disappear.
One other thing you need to be
very careful with is sample concentration. If you require multiple
samples then you should aim to have identical protein concentrations
(volume does not matter). Slight concentration differences can have a
large effect on the global tumbling of you system, hence the data
cannot be combined.
What do you say – how much concentration difference is still OK? I measured
samples between 320-330 µM (which is the maximum that is feasible for the
more unstable complexes) which amounts to a concentration difference of ~
3%. An additional problem is the inaccuracy of the concentration
determination by UV(280nm) and of course the sample degradation over time.
I never quantified the concentration after the measurements, which in
hindsight seems pretty stupid. I should check if there are any differences
(there certainly are, but I wonder if they turn out as significant).
I don't know, and it really depends on the surface properties of your
molecule as the diffusion is strongly modulated by non-specific
interactions. Each system is different. But constant protein
concentration is incredibly important for any dynamics analysis.
Sorry for the long sermon. I appreciate that you always read my stuff and
also answer in a really helpful and extensive manner.
Well, I hope some of my long answers helped. If I was you, I would
first redo the relaxation analysis using Sparky (and relax to fit the
exponentials) and compare the data to CCPN. Then if m0 is still
present, consider if you should blame it on the size of your system
hiding data (more field strengths should then help uncover the
dynamics). Finally, you should consider if your complex is tight or
not. If you think it is not (you can measure this with other
biophysical techniques), and if you have lots of time for theoretical
development, then maybe look at solving the partial dimerisation
problem first described by Schurr et al., 1994 and developing a
solution that many, many people could then use and cite. I have to
warn you though, although it can be rewarding, theoretical development
requires significant amounts of time.
Regards,
Edward
Re: m0 models