Questions regarding Panel Of Normal (PoN) for sWGS & SAMURAI analyzes

[aeeckhou]

Hello ! Thank you very much for the pipeline, it looks promising and very interesting to analyse sWGS !!

I had few questions before launching SAMURAI, regarding Panel of Normal (PoN) as I have little experience with using them :

1. Does SAMURAI need to have a PoN to launch with the solid biopsy workflow ? it is my understanding it does not, by opposition to the Liquid biopsy workflow.

2. How impactfull the PoN is for both solid and liquid biopsies ? Does it changes marginaly the genomic profile or is it profound changes ? If you have two genomic profiles to compare between PoN and no PoN I would be interested to see them side by side !

3. For both solid and liquid biopsy, what would be a good number of sWGS to generate to establish one PoN ? 15 cases, 20 cases ?

4. In your opinion, what type of prelevement should the PoN made of for solid tumor biopsy pipeline ? It should be from healthy donor I assume ? I would assume peripheral blood mononuclear cell (PBMC) for the liquid biopsy analyses.

Thank you again for the provided tool, can't wait to test it out, especially the CINSignatureQuantification step!

Best,
Alexandre

[lbeltrame]

Hello Alexandre,

first of all, thanks for considering SAMURAI! (No, this isn't a canned customer support response ;)

1. You don't need a PoN for most of the solid biopsy methods. ASCAT.sc will work without it more or less fine, QDNAseq doesn't really need a "PoN", but just normalized bin counts from samples (and you can find pre-computed data on 1000G for hg19 and hg38). ichorCNA, on the other hand, can work without it, but results will be impacted (see below on why)
2. The impact on PoN depends mostly on what you are looking for.

For solid biopsies (no, this wasn't written by an LLM ;)

If you're looking at large-scale CNVs (big rearrangements), you can probably (touch wood!) forgo it if you really can't have one, but if you're looking at smaller rearrangements, or smaller events, I would strongly recommend it because you have to deal with the noise of the natural CNVs which occur naturally with age. Also the purity of your sample matters: highly pure samples will probably be analyzed fine, but with samples with low altered content, you might have issues in distinguishing them from the background. (e.g. I had cases where I got data from really tiny biopsies, so it was insanely hard to find CNVs).

For liquid biopsies

A PoN is de facto essential, although ichorCNA can work without one. That's because you have the reverse of a common situation than with solid biopsies: you have far more healthy DNA than altered DNA, so you're fighting an uphill battle to find a good signal.
WisecondorX will also not work without it, because it needs it to compute Z-scores comparing your sample with the PoN... so hard to do that without one ;)
There's an additional issue with not having a PoN: ichorCNA uses the GC-normalized, mappability adjusted read counts from the PoN to normalize the data in your own sample, by cutting off the extreme values that are observed (aka outliers). If you don't do that, you'll be far more susceptible to noise (and in fact what you observe in a CN plot is that the dispersion of the individual bins is far greater), which in turn can affect the tumor fraction calculation (if you need it).

3. The GATK recommendations say a minimum of 40 (although these were written with WXS data in mind): in our testing, reasonable results can be obtained from 20+ samples onwards (our first work - Paracchini, Beltrame et al., CCR, 2021 - was done with 11 normals, but looking back at it I'd have used a few more).
4. Healthy donors is fine (that's what we use), but do not use PBMCs: instead extract cfDNA like you would with patients. The reason is that the DNA profile (length of the fragments) in cfDNA is very different from gDNA (which is of higher quality), so you'd be comparing apples to oranges (we did these tests back in the day, and the results were subpar).

If you have any more questions, feel free to ask!

[aeeckhou]

Hello,

Thank you very much for your detailed answer !!

Few more questions and I won't bother you more :

1. For PoN in solid tumor, what type of prelevement would you recommand ? Same amount, around 20+ ?

2. You mentionned small CNV/small rearrangements that could be difficult to assess. In your experience, below what size are you not confident to call changes ? Obviously it is related to the sample and it's quality but as a general rule of thumb ?

3. In your pipeline the default bin size is 500kb. For more than 1X coverage, I would generally go for something like 50kb. Would recommend to stick to 500kb. I am especially interested in the copy number instability signatures and the HRDCNA Score, and for copy number instability signatures I am afraid 500kb is too much for some features. What do you think ?

Best regards,
Alexandre

[lbeltrame]

1. Yeah, I think 20 samples for a PoN would probably reasonable.
2. It's related to the samples and the purity: generally speaking, the less pure the sample is, the more diluted your signal will be.
3. For solid biopsies, depending on your sample quality you can go as low as 30kb (see Pesenti, Beltrame et al., EJC 2022 to see a real-world example). FFPE samples may be more problematic (obviously!) but if the quality is reasonable you can pull it off. 50kb sounds like a better compromise if you don't know if all your samples will be high quality.

500 kbp was chosen as default mostly because internally we use SAMURAI for liquid biopsy analysis (95% of our runs), and in that case, going below, and in particular under 100kbp can be very problematic if your tumor content in cfDNA is low. For solid biopsies, I agree it's too large, you'll lose a lot of potentially informative rearrangements. We might put a note in the docs to this regard.