[RASMB] Thoughts about convection, Identifying convection and preventing convection

[David Hayes]

Hi fellow RASMB'ers

I have a set of sedimentation velocity experiments in which I have become 
suspicious of non-ideality and convection, or both.

I am trying to collect information on convection and also want to put out 
some theoretical ideas for your consideration or even your amusement, since 
convection is so annoying that I am willing to consider and will present 
any crazy idea if it might possibly help.

First, I have seen some data sets where the boundary shape (especially seen 
by dcdt methods or ls-g*(s)) is obviously asymmetric and just looks 
odd.  This is usually attributed to convection and sometimes seems 
correlated to a cell with a bent centerpiece or to improper cell alignment 
in the rotor.  The usual procedure here is to just repeat the experiment 
with new cells or better alignment and throw out the bad data.  In this 
case, convection is just an annoying problem that makes more work.

However, in the case of this data, the peaks do not look obviously 
asymmetric, they fit to a single main peak most of the time (some double 
peaks in c(s) occurred) and repeated experiments in different cells just 
keep varying.  I have fit the data with Sedanal and with Sedfit and am 
convinced that the data sets from the repeated experiments are actually 
different and that the variability is not an artifact of the fitting 
process and the random noise on the data.

So the conjecture I am considering right now is this:  Since c(s) in Sedfit 
models all aspects of the data as either one of the noises (baseline, TI, 
RI) or as an amplitude of a single ideal species among other 
non-interacting species, wouldn't c(s) report small amounts of convection 
as a species in the cell since convection is not a baseline, is not time 
invariant, and is not radially invariant?  This may explain the variability 
in the repeated experiments I am looking at.  For whatever reason, these 
samples are prone to small amounts of convection and that is why they do 
not reproduce as well as they should.

How is convection identified?
Peter Schuck implied that when the initial scans do not fit well, you may 
suspect convection.
Excess broadening of the boundary at the meniscus could be due to 
convection.  With curve fitting, this may result in the meniscus fitting to 
location closer to the center of the rotor than it obviously is by optical 
inspection.
Tom Laue has said that convection "follows the cell".  If the is cell is 
refit into another whole and the non-reproducible result follows the cell, 
suspect convection.
Looking at data with Walter Stafford using g(s), he also pointed out that 
when there is a bumpy leading edge that does not seem to move as a well fit 
boundary (different sets of scans give different S values) you can suspect 
the bumps are not real boundaries but convection.
Obviously asymmetric boundaries that have bumps or peaks in g(s) that are 
too steep or that change shape drastically (diffusion makes Gaussian curves 
or the sum of Gaussian curves that are pretty smooth) suspect convection.
Are there any other tell-tale signs?

What causes convection.
Tom Laue wrote a bit on this before in the Sedfit mailing list:  I will 
past that here so he won't have to re-write it for RASMB.
1- Probably the most common cause of convection is damage (usually a 
scratch from a needle tip) to the channel walls. Inspect the walls 
carefully. If there are any scratches, the centerpiece should be replaced.
2- The second most common problem is a warped centerpiece septum. One way 
to detect this is to place the suspect centerpiece over a "good" 
centerpiece and look to see if there is any place where the two septums do 
not line up. If you have a tool makers microscope, you can line up the 
centerpiece septum and look to see if the wall can be aligned with the 
scribe lines on the microscope viewer. The warping usually occurs if the 
cell has leaked on one side but not the other.
3- The third most common problem is the build up of junk on the walls. You 
will see this when you inspect the centerpiece- the walls will look hazy or 
have a spot or spots on them. If you have been working with proteins, 
sometimes they will dry onto the wall and be very difficult to remove. Your 
best bet in this case is to soak the centerpiece overnight in a solution of 
pepsin (~1 mg/ml) in 1 mM HCl. Then wash the cell thoroughly with soap and 
water, then water then a drying agent (dry N2- my choice, EtOH or acetone). 
This problem usually crops up if a centerpiece is very old or if it has 
held a high concentration protein solution.
4- The fourth most common reason is temperature instability at the start of 
a run. Be sure the temperature is stable before starting a velocity run.
Peter Shuck mentioned that for him the temperature instability was critical.

How to prevent convection.
1.  Throw away damaged cells and treat the new ones gently but cleanly.
2.  Make sure you have temperature equilibrium.
3.  Walter Stafford mentioned that ramping the rotor speed more slowly 
during the initial acceleration could possibly help avoid convection.
4.  Make sure the cells perfectly aligned with the center of rotation.  I 
know that one rotor we have has inner and outer scribe marks on each rotor 
whole and the two marks do not agree.  I believe that BITC sells a jig of 
some kind that aligns cells in the rotor more reproducibly than the human 
eye using the scratches on the rotor as guides.

Further theoretical musings on convection and possible prevention.

If slower acceleration helps prevent convection, maybe an experiment should 
be tried at few lower speeds also if convection is suspected.  By doing 
this you will sacrifice resolution, but may get more repeatability.

Also, I seem to remember that one reason the epon centerpieces have 
aluminum or charcoal filling is to make them better temperature 
conductors.  I don't know which is better, but I would think the aluminum 
filled would be better conductors of heat.  So maybe using aluminum 
centerpieces would help prevent convection not only be making the original 
temperature equilibrium more effective but also by dispersing the heat of 
initial compression more effectively.  When the centrifuge is started, the 
rotor expands and cools and the sample and reference solution are 
compressed and heat up causing a small temperature gradient.  This 
acceleration caused temperature gradient may be a cause of convection.

There is the almost legendary story, which I have heard a few times from a 
few sources, that Svedberg  initially had rectangular cutout centerpieces 
in his original centrifuge designs and then suddenly realized during a 
trans-Atlantic trip that he needed the sector shaped cells to avoid 
convection.  This supports the advice above to have clean, unwarped, 
perfectly aligned cells to avoid convection.  However, I wonder about the 
cleaning part:  could the cell walls be too clean?  Since proteins adsorb 
to surfaces (which has actually now been experimentally demonstrated by the 
fluorescence centrifuge experiments), could a too clean wall actually cause 
a drag on the protein solution near the wall and actually promote convection?

I have also heard that the salt gradient that builds up in the centrifuge 
actually stabilizes the sedimenting boundary.  This is supported by the 
mysterious action of fluid flow in the meniscus matching cells.  Even when 
the scribe line that lets buffer flow into the cell enters the sample side 
at the bottom of the cell, the buffer solution tends to flow into the cell 
and float to the top forming a laminar boundary.  It is amazing how smooth 
the flow and how sharp the boundary can be.  Similarly in synthetic 
boundary experiments, the lighter buffer usually forms a single sharp 
boundary with the slightly heavier protein solution.  Here the buffer does 
come from the top side of the protein solution.  Convection here can result 
in a doubled synthetic boundary that has more than one inflection point 
which becomes obvious  with diffusion if you collect enough scans.  For 
synthetic boundaries and the boundaries formed by meniscus matching cells, 
convection occurs more with faster flow (the capillary is too large or 
tilted or the rotor speed is raised too quickly).

I would conjecture then that conditions of low salt would tend to 
destabilize boundaries and that convection would be more likely in these 
conditions.  In any case, even when you are constrained to examine a 
protein in low salt conditions, it is a good idea to run a control 
experiment of the same sample dialyzed into an almost identical buffer with 
100 mM salt added, hoping to get some indication if non-ideality if 
affecting the sedimentation in the low salt conditions.  I think that 
irreversible aggregates would be unlikely to be dissolved merely by adding 
salt, so this would help check reproducibility.

It also seems that it may be worth trying to repeat an experiment at a 
lower temperature if convection is suspected.  More viscous liquids tend to 
flow more smoothly and have less turbulence.

And just by experience, it seems to me that I have most often seen obvious 
convection at higher protein concentrations.  It seems the sharper the 
actually concentration boundary, the more likely it is to convect.  It is 
often the highest concentration cell that I am throwing out of the data 
analysis even though it has the best signal to noise ratio.  Certainly, 
there is more fluid flow:  both of the solute and the back flow of the 
solvent with higher concentrations so it makes sense to me that higher 
concentrations are similar to faster sedimentation.



This is what I have been able to come up with by asking a few people and 
thinking about convection.  I put the ideas out in public to see what 
people think and to solicit other ideas.

Cheers






Dr. David B Hayes
Boston Biomedical Research Institute
64 Grove St.
Watertown, MA  02472
617-658-7738

Boston Biomedical Research Institute... Today's Research for Tomorrow's 
Health.
Please visit us at www.bbri.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.rasmb.org/pipermail/rasmb-rasmb.org/attachments/20060324/e366d2bc/attachment.htm>