Hello Christiane,

there are several possible interpretations:

1) Sequencing artifact, most notably if the variant is a small indel: the error rate of these events is higher than single nucleotide variats. Did you confirm that the variant is real by targeted resequencing (e.g. PCR followed by Sanger sequencing) ? Did you manually check that region using IGV or similar?

2) Rare SNP: unlikely if the variant is apparently homozygous as in your example. When studying cancer is very important to perform matched WES, i.e. to sequence both cancer  and normal tissue (e.g. white blood cells, gDNA from buccal swab..). This allows to quickly check if the newly identified variant is a rare polymorphism or if it is truly somatic. Is WES of a normal tissue available in your case?

3) Real somatic variant: in this case afaik there are 3 possible interpretations:

- the other allele/genomic locus became deleted, as you suggested.

- a somatic uniparental disomy event occurred. Somatic uniparental disomy typically occurs when extensive damage on a chromosome triggers homologous recombination (HR) as a repair mechanism. If a somatic variant is present in the undamaged chromosome, this variant can be copied during the process, 'duplicating' the variant and therefore generating a homologous somatic mutation.

- the mutated allele/genomic locus underwent significant (multiple copies) genomic amplification: if one of the two alleles gets significantly amplified you will end-up in allelic imbalance (AI) that can mimic homozygosity.

Note that the events described in 3)  typically generate specific 'signatures' in the nearby regions. Deletion/HR cause loss of heterozygosity (LOH), provided that heterozygous SNPs are present in the deleted/HR region. These LOH can be easily detected if WES data from a germline tissue of the same patient is available. Conversely, amplification causes allelic imbalance (AI). Note that, in presence of extensive copy gain, AI can be erroneously interpreted as LOH.

In order to discriminate among these events you should analyze the coverage of the target region: if you detect a coverage drop in the region where the variant is present, then the most likely explanation is a deletion; if you detect coverage increase, this will suggest the presence of amplification; if coverage is neutral -> somatic disomy.

Hope this helps,

Rocco

Dear Christiane,

I don't fully understand your problem. Maybe it is my lack of knowledge or problem complexity. However one advice from me is to sequence also reference sample - blood from the patient that you have tumor sample. Once you will have a reference, finding exact mechanism - somatic/LOH... will be easier.

One concern is always additional sequencing cost, but you can always sequence normal sample will lower depth. For instance, if your WES was planned to be on average 50x, try to set 15-20x for normal sample.

One software that we find useful when you will have a tumor/normal pair samples is Varscan2.

Hope that helps,

Best,

Bartek

I know that the use of depth-based CNV algorithms for WES is controversial and challenging, but one of my colleagues just had good luck finding hemizygous deletion using ExCopyDepth, which is supposed to be an improvement over some of the older algorithms for detecting exome CNVs (ExomeCNV, CONTRA, ExomeCopy, ExomeDepth, CoNIFER, XHMM). I'm going to give it a try myself. Here's the link to the online version of the article (http://www.biomedcentral.com/1471-2164/15/661). 

Hello Julie,

as the author and user of a CNV/AI software for the analysis of matched exomes (CEQer) I can tell you for sure that these tools work..at least mine   :)

Of course the results depend on the nature of the CNV event: detecting a 3-4-exons deletion/amplification with WES can be very challenging but larger events are usually easily detected, provided that the exome quality is good and that the event occurs in a significant fraction of the cancer cells.

I think you first assumption is it self wrong, i.e. there will be less probability that a second het mutation will not occur at the first site. Secondly, in a hemizygous mutant the other allele might get somatically deleted, but it all depends upon the mutation.

I am also going to check out CEQer - thanks Rocco!

Hi

The deletion discussed in our ExCopyDepth paper is an 11 exon hemizygous deletion. Recently we were able to detect 2 exon deletion as well.

Recent articles have shown that the prediction programs have high sensitivity in detecting CNVs. Therefore they are capable in detecting true disease-causing CNVs. However they also have high false positive rates. So the challenge is to detect the disease-causing CNV from the 100s of false positives and common CNVs predicted. It is possible to apply high thresholds and reduce the false positive rate with the expense of losing sensitivity.

Dear all,

thank you for your answers.

Rocco: Thank you for that overview. I have checked IGV; the deletion is about 1/3 away from the end of the exon, meaning that it lies in the mean height of the overall coverage of that exon. This means I don't see a drop in the coverage surrounding the deletion, which would imply it is a deletion. I will also take a look at your software CEQer. We very unfortunately couldn't obtain somatic tissue, but we will get another batch with matched somatic samples. On these I will do sanger sequencing to confirm. In addition, if I want to investigate LOH in general, a SNP array (rather than aCGH) would be ideal, right?

Julie: Thank you for the paper. In another WES-project we were able to identify a rather large 1 allelic-deletion using CoNIFER, and we could confirm this doing an aCGH afterwards. CoNIFER was indeed helpful in identifying the gene-location of that region.

Hello Christiane,

yes, SNP array is ok, provided that you're only interested in LOH.