Single direction reads for variants

[Moggs]

I'm running a targeted resequencing project and I'm observing some strange results with respect to few single base variants. We capture using SureSelect, sequence single end with GAIIx and then align reads and call variants with MAQ. From 118 tumor samples we see 80 samples that have a het call (T/G) at exactly the same base position where the matching ref base (T) has equal number of reads in both directions (forward/reverse) but the variant (G) has reads in only one direction (reverse). The reads mapping to G allele have multiple start positions so I'm ruling out PCR bias and contamination. The variant isn't in dSNP and the sequence is unique (ie not repetitive and no apparent pseudogenes). Anyone have any ideas?

[nilshomer]

I'm running a targeted resequencing project and I'm observing some strange results with respect to few single base variants. We capture using SureSelect, sequence single end with GAIIx and then align reads and call variants with MAQ. From 118 tumor samples we see 80 samples that have a het call (T/G) at exactly the same base position where the matching ref base (T) has equal number of reads in both directions (forward/reverse) but the variant (G) has reads in only one direction (reverse). The reads mapping to G allele have multiple start positions so I'm ruling out PCR bias and contamination. The variant isn't in dSNP and the sequence is unique (ie not repetitive and no apparent pseudogenes). Anyone have any ideas?

What are the mapping qualities for the reads with the G mutation?

[Moggs]

The G quality scores are OK (Phred-like typically 20-35), no different from the T scores. With some further investigation I noticed that read direction is not always balanced for the T call and often biased in favour of forward reads (average 5:1 across 80 samples). Perhaps there is something odd about this sequence resulting in misincorporation of a C for A (as its a reverse read) only at this particular base position. Quite strange

[ECO]

Which allele is in the baits?

What is the local sequence context of the T/G ?

[Michael.James.Clark]

What is your coverage per sample?

It's possible you're just seeing a random event. 66/118 and 80/118 having specific allele reads on only one particular strand is unlikely (especially at high coverage), but certainly possible if you're only observing the position a couple times per sample.

Could also be due to biases in your hyb or subsequent PCR. Hard to say.

I think it's unlikely to be due to the sequencing incorporating the wrong base in a systematic way. It can and does incorporate the wrong base randomly of course, but for 80/118 to randomly be the same wrong base on the same strand by chance seems unlikely (especially given what you observed with the reference T allele displaying a similar bias).

I'd hypothesize the variant is real and verify by other means such as Sanger sequencing.

[Moggs]

The bait should be the reference HG18 sequence therefore T not G. The surrounding sequence context as follows:

Forward GGAGGAAGCTGG/TACCGTGCCAACGGCCA
Reverse TGGCCGTTGGCACGGTA/CCAGCTTCCTCC

Base position chr1:2056602 on HG18

[Moggs]

We'll do the validation but my bet is that it is an artifact, although I don't have a reasonable explanation. Another group in our institute sees the same variant being called for a different bait library targeting the same gene and I was curious to know if anyone else sees the same thing or if there are any general rules about predicting artifacts from imbalanced read directions when a variant is being called with adequate seq depth. Contamination would seem likely in our case if all reads on the G had one start site (as we share facilities) but they don't. Depth is typically 50+ across 80 samples.

[Michael.James.Clark]

I asked some of the people in my lab who do SureSelect pulldowns on GAIIx and they said they do not see phenomena like this.

[ECO]

I would have a look at this thread regarding context specific errors (specifically T->G changes near a GGnnG) in newer data, particularly the link in post 9.

[Michael.James.Clark]

Interesting link. Could be from fragmentation protocol. In our lab, we use a Covaris I believe at 4C, not sure on the exact settings. Moggs, what do you do for fragmentation?

[Moggs]

Just to report that we tried validating the variants and they were false. Thanks for the heads up on the GGnnG issue ECO. Must be related.

[ECO]

Thanks for the final word on this story...interesting.