Loss of resolution on 3700, and failed lanes

Phillip San Miguel pmiguel at purdue.edu
Thu Mar 2 11:44:41 EST 2000

"G. Harrison" wrote:

> Hi Phillip,
> Your recent posting regarding failed lanes on the 3700 prompted me to
> respond with our observations.  We run 8 96-well plates / day on the
> 3700-Capillary and 4 96-well plates / day on our 377-Gel machines (awkward
> nomenclature, thanks, ABI).  Our sequencing technicians prep 800 samples
> / day with an alkaline-lysis/PEG precipitation procedure, and perform
> standard PCR sequencing reactions with an isopropanol precipitation
> cleanup.
> I have been dissatisfied with the quality of the data obtained on the
> 3700.  Our largest problem is the inconsistency of data quality with
> runs that give phred Q>20 values ranging from 0 to 554. The average Q>20
> for these runs collected over 6 months is 292.8 and the average Trim
> length is 362.7.
> Most frustrating is the observation that the number of failed lanes is all
> over the map.  You can download an Excel graph of our data from:
> http://chroma.mbt.washington.edu/msg_www/public.links.html and select the
> 3700 Failed Lanes file.  We average 33.77 failed lanes/gel over 6 months,
> sigh.
> [...]
> Can you think of what else we should try?  What kinds of things have you
> been working on?
> Of course, when we change to POP 5, we'll have to do this all over again.
> Happy sequencing,
> Grace Harrison, Lab Manager,
> Department of Molecular Biotechnology
> University of Washington, Seattle, WA
> ---

    I think maybe we are seeing different phenomena here. But I'm not sure. I
would classify my "failed lanes" in this way: 1) low or no signal-to-noise;
2)high signal but low resolution [broad peaks]--correlated with delayed
appearance of the first peak or 3)multiple sequence. I think we can ignore
class 3 here.
    Without looking at the reduced data (the array view) for each gel (or,
more arduous, looking at each chromatogram), there is still a way to
distinguish between the class 1 and class 2 failures. With class 1 failures,
phred calls an abnormally high total number of bases (all of them with very
low QV's), but with class 2 failures, phred calls an abnormally low total
number of bases. That is, if my good reads produce a total read length of 1000
[1], then class 1 failures might give 1300 or more bases where class 2
failures would give less than 800. Of course, all the bases are junk in both
cases--but the total number of bases called is diagnostic between these two
types of failures.
    I'm making a distinction between the class 1 and class 2 failures because
I think they have different causes. Class 1 failures could just be the result
of a failed sequencing reaction. But, as you note, they could be the result of
salt competing with the sequencing reaction products for electrokinetic
injection. The salt can come from many sources but the main ones would be
residual sequencing reaction reagents (including terminators) and template
contaminants (especially RNA/RNA nucleotides and salt used in precipitation,
etc.) We are not getting many class 1 failures. This is probably a result of
our methods[2]--specifically our method of dye-terminator clean up. We are
getting some class 2 failures. But mainly only in certain batches of
    I can only speculate on the cause of class 2 failures. But we have had
less since we halved both the amount of template we use (1/50th of the REAL
prep instead of 1/25th) and the concentration of the reactions we put on the
machine. The result is that my mean increases (less failures) but I don't see
reads over 700 Q>20 bases any more. (I was getting 1 or 2 per run). I think
that "overloading" a capillary is what causes a class 2 failure. My signal
strength is very high in the first part of the run (standard "top-heavy"
data)--often over 10,000. I guess that water loading just causes more of the
sample to be injected--that is the source of the increase in signal strength.
My array is fairly aged--it is around 250 runs old. I think that as the caps
age they get more and more "clogged" and the chances of a sample "overloading"
them becomes greater. (Another possibility is that with version 1.1, 1/2 as
much buffer is used. [At least my machine is only using 1/2 as much as before.
4 liters of buffer will last more than 16 runs. Under version 1.0 I was only
able to get 8 runs from 4 liters of buffer.] Maybe some of the caps aren't
getting completely covered during electrophoresis. But then I would not expect
the failures correlate positively with signal strength.)
    Thanks for your post. I think that it is valuable to discuss these issues
in an open forum like this. I hope your results improve. The 70% ethanol wash
seems promising.
Phillip San Miguel
Purdue University Genomics Core Facility

[1] Note that only 400-700 of these will have QV >20

[2]We do Qiagen REAL preps here for templates. We use 1/50th to 1/25th of the
prep for the sequencing reaction. We do 1/2 volume, 1/4 concentration big dye
ready mix (that is, 1/8th reactions) but cycle them for 99 cycles. This
doesn't work well in our hands unless the reaction is 5% DMSO[3]. Also we do
everything we can in a 384 well format. Colonies are picked into 384-well
plates (Terrific Broth, 8% glycerol) grown overnight and innoculated into deep
96-well plates (1.3 ml of Terrific Broth). These cultures are grown overnight
shaking at 300 rpm covered with Qiagen Airpore "tape".
    We set up the sequencing reactions in 384-well plates using our Hydra,
cycle overnight and--I am ashamed to admit--use commercial dye terminator
clean-up columns. [I know that isn't as macho as isopropanol precipitations or
pouring your own columns, but it works well. It adds more than $0.50 to the
cost of the reaction, but I rationalize the cost by figuring we save at least
that much by using 1/8th reactions instead of 1/4th reactions. I know that is
lame, because we could do isopropanol precipitations *and* 1/8th reactions.
Maybe we'll switch some day.] We use BioRad SEQueky Kleen 96 Well Terminator
Removal Kit (Cat 7326260). But this kit isn't made for the 3700, so we wash
these column plates 3x with ddH20. That is, we use our 1 ml Hydra to fill the
wells with water (300 to 400 ul is plenty) and then use a vacuum manifold to
pull off the water. (We use the same manifold that we use for Qiagen 96-well
filters). I don't know if it is necessary to do 3 washes, but that is what we
are doing at the moment. We add the entire sequencing reaction (10 ul) plus
15-20 ul of water to the plates after the pre-spin. There is usually 40-70 ul
of elutant. We place 8 ul of the elutant and 7 ul of water into a PE Bio
384-well plate. Cover with 3M Scotch Tape 439 (thin aluminum foil tape),
invert and vortex to mix, spin to collect sample at the bottom of the plate
then place on the 3700.
    We're just starting up here. Our 3700 has only been in "production" mode
for about a month. We switched to version 1.1 of the data collection software
(POP5/ foil piercing, etc.) just over a month ago. I'm pleased with our 3700's
performance. We can do about 8 runs a day. We did 105 runs in February. ["We"
is me and my technician] 5 of the runs were crashes--produced no data (mainly
FMSheathPump errors--although once I let the buffer run dry...). My average
read length (including a few bad reaction sets) was 531 Q>20 bases. I'm pretty
sure that Phred still underestimates the number of Q>20 bases for the 3700.

[3] Thanks to Bruce Roe for this procedure. For more details see his web page:



More information about the Autoseq mailing list

Send comments to us at biosci-help [At] net.bio.net