Low pass filter % due to short reads

[eladSeq]

Hi,
We previously performed a 454 amplicon sequencing run on our gene library with great results back in March '12. We prepared a new amplicon library of the same gene essentially the same way as before only this time with more MID barcodes. The core facility we are working with has repeated the run twice now with very poor results and they can't figure out what seems to be going wrong nor has Roche tech support been able to figure out what the problem is.

Some details: amplicons created with lib-A A/B fusion primers, gel purified, and AMPure XP bead purified. Running on the fragment analyzer shows a clean band at 421 bp (as expected, though a little high, amplicons should be 398-412 bp). There is very small double peak at ~30 bp, but this is too small to be primers or primer dimers, and may be an artifact of the lower marker. The %bead enrichment was around 12-15%.

After the first run, the Raw Wells was 2,472,168 and the Key Pass wells was 2,388,051 both of which look good. The dot filter then removed 169,004 and the mixed filter removed 256,456 (total of 18%). However the Short Quality filter then removed 1,835,645 (77%) of reads, such that only 125,726 passed filter (5%). A second run had essentially the same results but even lower pass filter (4%).

Roche was looking at the sequences and said they didn't see major homopolymer issues or flashing interference problems and that the control beads looked good.

Of the good reads we have received, the sequences look good and we see a ditribution of the MID barcodes as we would expect. So we are frustrated and at a loss to figure out what is going wrong with these runs. One difference between these runs and the previous successful one is our core manager informed us that Roche changed their beads to amplify better.

Has anyone had such issues with low pass filter/short reads? Any ideas or suggestions as to what may be going wrong here would be appreciated.

Thank you!

[ajthomas]

I've seen similar things going on and I'm also at a loss to explain it. I've had a few runs of amplicons in the last few months, some using fusion primers and some with adapters ligated on with the Rapid Library kit, that are similar to what you describe. What I've found is that the pass filter % is very low when processed with the amplicon pipeline, but in the low to normal range with the shotgun pipeline. When processed with the shotgun pipeline however, most reads are trimmed at some point short of the full length sequence.

I've looked at the flowgrams of some of the failed reads and that's where it gets more interesting. Many of these sequences look great up to some point, after which it looks like a mixed read. It's as if the bead is actually coated with two sequences that are almost identical, but differ at the point at which the flowgram goes bad. At that point the two sequences get out of phase with each other and it looks like a mixed read from that point on. for one of the runs I looked at several dozen failed sequences and saw this phenomenon in about 2/3 of them while the other 1/3 looked genuinely bad. At first I thought the molecules in the library were rehybridizing before I could get the emulsion made and the hybrids formed from two molecules that differ at just on base. However if this were to happen with rapid libraries the result would be two mixed reads from the beginning, not just from some point in the middle. So no, I don't think that's it. Another idea of mine is that perhaps there is a fidelity issue with the emPCR enzyme. If that enzyme makes an error during an early cycle the result would be a bead covered with two almost identical sequences and reads as I saw. I would think that an error rate sufficient to effect so many reads would cause a lot more problems, so I'm not sure I can believe that idea either.

Have your provider go look at a bunch of the flowgrams and see if they see something like what I describe. You could also ask them to process the data with the shotgun pipeline to see if the pass rate increases substantially but most of the reads are trimmed too short. If they see something similar to what I described, then there is probably some systemic problem that Roche needs to figure out and fix.

[eladSeq]

AJ, thanks for the feedback. Our provider did also run the shotgun pipeline and that improved things a little bit to 10% pass filter and length actually increased from 355 to 400 bp. They also tried some modified filtering parameters that Roche recommended but did not get any better pass filter %.

For the mixed reads you describe, did they fail due to the mixed reads filter or the short quality filter? Because we are losing only 18% to the mixed read filter, while the majority is lost in the short quality filter. But I will suggest that they take a closer look at some of the failed reads as you describe.

Have you done any analysis on the DNA molecules after emPCR or after enrichment? Perhaps there is a way to determine the size distribution at this stage to rule out short products contaminating the library.

[ajthomas]

In my case the reads failed due to the short quality filter. I don't remember the percentages, but the mixed reads filter only tossed a few, well within the expected parameters. When I mentioned mixed reads, I was talking about what the flowgram looks like. Up to some point in the read the sequence looks great, but then it changes to look like a mixed read somewhere in the middle. The short quality filter tosses those reads in the amplicon pipeline, but with the shotgun pipeline the reads are trimmed and kept.

I have looked at all of these libraries on the Bioanalyzer and there are no appreciable amounts of short molecules present. I should mention, however, that all of these libraries have come from the same source and I haven't run anything else since this began, so I can't be sure it's not some sample-related issue. I just can't think of what sample issue could cause it.

As for your issue, I don't have any more ideas, but I think that looking at the flowgrams could be helpful. It could at least tell you about the reads and give you an idea why the short quality filter is catching them.

[bradfish]

Same problems

Hey just wanted to chime and and let you know that you are not alone. Ever since Roche changed the emPCR reagents and beads, we have lost about twice as many reads to the short quality filter as we used to. I am using a GS Junior, but am seeing the same problem as you. For example, we used to get 100k reads on a run, and do about 2 runs a months, but for the last ~20 runs, the reads have only been about 60k, this is due entirely to short quality failures.

Would like to know if you figured anything out. Haven't contacted Roche because I know they will claim it is not their fault, even though we have a direct correlation to the lot change for emPCR reagents to our reduced reads.