[RASMB] Help! No model fits my sedimentation eq data

John Correia jcorreia at biochem.umsmed.edu
Wed Feb 26 10:52:00 PST 2003


Sed Vel as a function of conc (stick to single wavelengths until you see what is going on over a 5-10 x range) should be required before any sed equil work - caveat of course is extremely small peptides, but people have run those too!  It saves lots of time, mostly in wasted analysis time.  The multi-conc, multi-speed model of doing sed equil comes after you are sure its a reversible system.  In spite of what the experts might preach, if you can't fit single speed data, more speeds don't help.

Plot the c(s) or g(s) as a function of loading concentration and look at the evolution of the c-dependent.  The absence of C dependence or regions that do and don't exhibit c-dependence tell you have irreversible reactions going on.  Be cautious about interpreting all the peaks in a c(s) as a real species; yes its very sensitive, but spurious peaks do occur in c(s); g(s) in spite of the spreading never does this, but small peptides may be hard to analyze with g(s).  In any eveny, model building with the family of c(s)/g(s) curves is the way to go.

I have not heard anyone suggest turning on separate K's which makes Ka local variable for each channel.  Doesn't MacNonlin have this feature.  This works best when you have low levels of aggregation which show up as a smaller K with increasing conc.  

In the original Nonlin report (Johnson et al) for sed equil nonideal monomer dimer was discussed as being an extremely hard problem, because the viral expansion for nonideality looks like a monomer-dimer system.  Negative B means favorable interactions, ie association.

Finally, in spite of the coiled-coil hype the lit is full of examples where coiled-coil sequences make trimer and tetramers and extremely large oligomers & mixtures of all of the above.  Sorting it all out by equil fitting is where your time goes.  



-------------------------------------------------------------------
 Dr. John J. "Jack" Correia
 Department of Biochemistry
 University of Mississippi Medical Center
 2500 North State Street
 Jackson, MS  39216
 (601) 984-1522                                 
 fax (601) 984-1501                             
 email address: jcorreia at biochem.umsmed.edu     
 homepage location: http://biochemistry.umc.edu/correia.html  
 dept homepage location:    http://biochemistry.umc.edu
-------------------------------------------------------------------
 
 


>>> "Guinevere A. Murphy" <Guinevere.Murphy at colorado.edu> 02/25/03 08:48PM >>>
----------------------------------------------------------------------------------
The older archived RASMB emails can be found at:
http://rasmb-email.bbri.org/rasmb_archives 
and current archives at
http://rasmb-email.bbri.org/pipermail/rasmb/ 
Search All the Archives at:
http://rasmb-email.bbri.org/rasmb_search.html 
----------------------------------------------------------------------------------


Dear RASMB,

I am (a novice at) working with equilibrium sedimentation data for a 16.9
kDa protein (E280=2680 /M cm), collected at three concentrations of 56, 94
and 136 microMolar. I collected data at four speeds, 20, 25, 30 and 35
krpm. The protein is a long dimeric coiled coil, so I expect to have a
monomer-dimer equilibrium. There is a possibility of higher order
structure, so a mono-dimer-tetramer equilibrium would not be out of the
question, either.

I have been analyzing the data using MacNonlin. For the two 20K data sets
I collected, I can locally fit the lowest 56 micromolar data to a single
ideal component model, (initial guesses: sigma=1.8 (dimer), deltaY=3E-3,
and lnA=-2.6) I get nice random residuals, and a sqrt variance of ~4E-3.
The fitted sigma corresponds to a 29 kDa component (between the expected
16.9 and 33.8 monomer and dimer MWs).

If I fit to a monomer-dimer equilibrium for this data set, I also get
reasonable sqrt variances (4E-3) and nice random-looking residuals, and I
get a Kd of 28 microMolar (using the 1/Ka, after converting the Ka from
absorbance units with the formula Ka(M)=[Ka(AU)*E(monomer)*(1.2cm
path)]/2). I view this high number with skepticism, both because it is
pretty high, for what I might expect, and because I am only fitting one
concentration and speed.

So, I would like to add in other data sets, however I find that I am
unable to fit any of the other data to a reasonable model in a manner
which produces anything like random residuals, and the sqrt variances are
all high (well above 1E-2). This is true for fitting to single ideal
component, monomer-dimer, monomer-trimer, dimer-tetramer and many others
that are not as obvious.

There are two observations regarding local fitting of one of the data sets
that I am having trouble with that I think might be suggesting nonspecific
aggregation of the protein at higher concentrations, and later time points
(20K was collected first):

a.) If I vary B, the second virial coefficient, I do get a random
distribution of residuals, and sqrt variance of 3E-3, but the value B gets
fitted to is -0.57. I know positive values of B suggest nonideality, but
does negative mean it is ideal and just isn't being fitted to the proper
model?

b.) If I set sigma=1.9 (for dimer) and don't vary it, and then vary
N2(init guess at 2, so init guess is for dimer-tetramer equilibrium) lnK2
(init guess at 8), lnA (init guess -3), delY (init guess 7E-2), then I get
the most reasonable fit to the data I've seen, and the residuals are not
totally random, but more dispersed and random than any other model, with a
sqrt variance of 6E-3.  However, the fitted parameters go to N2=17.5, and
lnK2=9.5!

This result suggests to me that either I've got one heck of a large
oligomer, or there is nonspecific aggregation occurring.

Would anyone have insight into this this difficult-to-fit data? And if it
is likely nonspecific aggregation, is there a definitive diagnostic that
would show that with the data I already have?

Thanks very much for taking the time to read this, and for your
suggestions and comments!

Gwen Murphy

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Guinevere A. Murphy		    Chemistry and Biochemistry Department
murphyg at ucsu.colorado.edu	    The University of Colorado at Boulder
http://ucsu.colorado.edu/~murphyg 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

_______________________________________________
RASMB mailing list
RASMB at rasmb-email.bbri.org 
http://rasmb-email.bbri.org/mailman/listinfo/rasmb



More information about the RASMB mailing list