[RASMB] RE: DC/Dt vs. sedfit

Thu Feb 16 10:00:11 PST 2006

Hi Chris and Patrick,
	I'd like to point out that you can't get any diffusion 
coefficient from ls-g(s*) because there is no time associated with 
the patterns; so half of the important information is missing, even 
though you get a good idea of everything that might be in the sample. 
With c(s) on the other an accurate diffusion coefficient can be 
obtained only if there is only a single species present. If there are 
two or more species one ends up a single average frictional ratio 
that applies to accurately to none of the species. So while you may 
find dcdt inconvenient to use because of the effort required to get 
an appropriate time span, that effort is well spent if you are trying 
to get molar masses for several species in a mixture.

Remember that any given dcdt pattern is a snap-shot of the the 
boundary at a given time (averaged over a relatively small interval) 
so if the time span and speed are chosen wisely, the interesting 
features of the boundary shape related to diffusional spread for a 
non-interacting system can be preserved and extracted from the plots. 
C(s) uses an average frictional ratio and applies it to the Lamm 
solution at each value of the sedimentation coefficient. By doing 
this it is only approximately accounting for diffusion. To extent 
that a single frictional ratio can be used to represent the diffusion 
of a mixture, it works fairly well. But it fails miserably in cases 
where that approximation does not hold. For example, I have spent a 
lot of time working with coiled-coils of various lengths; these are 
derived from myosin tails, paramyosin and tropomyosin, which have 
very different axial ratios and therefore significantly different 
frictional ratios. And more recently I have been working with T4 tail 
fiber fragments in mixtures. C(s) is not appropriate to use and fails 
very badly in this case because the axial ratios of the fragments are 
so different.

I think in general one must be very careful in applying a technique 
with an assumption like a constant frictional ratio. This is somewhat 
like to trying to estimate molecular weights from gel filtration 
chromatography in which it is assumed that your unknown protein has 
the same value of f/fo as all the proteins used to standardize the 
column. It works in most cases pretty well but when is doesn't work 
you will not be aware of its failure unless you resort to a more 
rigorous direct method. So you might as well use a rigorous method as 
soon as you know you have more than one species. For example, in a 
sample of myosin rod  (about 20 x 1800 A) in which irreversible dimer 
formation is observed, there are several ways in which the dimers can 
become associated - two extreme cases are end-to-end (like tropomysin 
association) and side-by-side as seen in initial stages of 
paracrystal formation.

And I think the same argument applies to interacting systems. C(s) 
takes no account of the boundary spreading caused by reversible 
interactions. As Peter Schuck has pointed out, the peaks one obtains 
in c(s) plots with interacting systems do not correspond to any of 
the species present in the system - unless one is at the extremes - 
either all dissociated or all associated.

Attempts to deconvolute the diffusion from a reaction boundary with 
c(s) are based on the application of a least squares fitting 
procedure that fits the reaction boundary to a model comprising 
overlapping non-interacting components with a constant frictional 
ratio. I have trouble understanding why one would expect that fitting 
data with a model that is obviously incorrect (i.e. does not take the 
reversible reaction into account) could possible return meaningful 
parameters. Although it seems to work well enough in some cases, it 
is not clear what criteria for a bad fit would be used to identify 
those cases in which it doesn't work.  One would expect (and indeed 
sees in simulations) the residuals to exhibit systematic variations 
resulting from the lack of fit to the shape of the reaction boundary.

Walter Stafford

p.s. I would recommend direct fitting using Lamm equation solutions 
for both molar mass determination and characterization of interacting 
system. These other methods dcdt, c(s) ls-g(s) are the first step in 
understanding what's in your sample. Both dcdt and c(s), if properly 
applied within their respective limitations, can be used for molar 
mass determinations. Both can be used to get accurate weight average 
sedimentation coefficients for isotherm analysis of interacting 
systems. More detailed and sophisticated analysis requires more 
powerful approaches, and both SEDPHAT and SEDANAL, among others, can 
be used for more complicated models - BUT always with the least 
squares caveats kept in mind -that one must have be fitting the data 
to the correct model, the residuals must be random and non-systematic 
in order to extract meaningful parameters. And once the caveats are 
satisfied one should apply an appropriate procedure to estimate 
confidence limits for the fitted parameters.

At 10:24 -0600 2/16/06, Chin, Christopher wrote:
>Dear Patrick,
>
>Thank you for asking. I have the same feeling as you do and that is the
>motivation for me to undertake and initial this project on my spare
>time.
>
>All the best,
>
>Chris
>
>
>
>
>
>
>
>
>
>------------------------------------------------------------
>
>Christopher Chin
>
>Manager, Macromolecular Assembly Core
>
>Sealy Center for Structural Biology and
>Molecular Biophysics
>
>Department of Biochemistry & Molecular Biology
>
>5134 MRB. rt1055
>
>UTMB, Galveston, TX  cchin at utmb.edu,
>
>409-772-1693, efax 630-604-3416
>
>-------------------------------------------------------------
>
>
>
>-----Original Message-----
>From: Patrick [mailto:patrickhbrown at gmail.com]
>Sent: Thursday, February 16, 2006 8:47 AM
>To: Chin, Christopher
>Subject: DC/Dt vs. sedfit
>
>
>Chris:
>I would be interested in seeing your comparison between DC/Dt and
>Sedfit.  What is your feelings on the two software packages? I used
>DC/Dt while in graduate school, but got frustrated because although it
>was very easy to use, it seemed that the answer that I got depended on
>whether I took early, late, or middle scans.  I liked sedfit because I
>could use all of my scans and not just a couple of them.  I didn't like
>that I had to keep changing which files I loaded into DCDt to try to get
>the correct mass range for fitting.
>
>Thanks,
>Patrick--
>Patrick H. Brown, Ph.D.
>Protein Biophysics Resource
>Division of Bioengineering & Physical Science, ORS National Institutes
>of Health Bldg. 13, Rm. 3N17 13 South Drive Bethesda, MD 20892 - 5766
>Tel: (301) 435-9371
>Fax: (301) 480-1242
>_______________________________________________
>RASMB mailing list
>RASMB at rasmb.bbri.org
>http://rasmb.bbri.org/mailman/listinfo/rasmb

-- 
-----
Walter Stafford
mailto:stafford at bbri.org
direct dial:    617-658-7808
receptionist:  617-658-7700
-----
Boston Biomedical Research Institute... Today's Research for Tomorrow's Health.
Please visit us at www.bbri.org

-----