RNA-seq data from human tumors in mice (PDTX)

Question

[Tom_L]

Hello,

I'm working in a cancer center where we use PDTX (Patient-Derived Tumor Xenograft) so the principle is to graft human tumors in mice. We recently reached a dead-end as we wanted to performed some RNA-seq (75bp; paired-ends) on these tumors. The problem is that there always is a cellular contamination of the tumor sampling with surrounding mouse tissues.

Some papers described the (obvious?) possibility to remove sequenced mouse reads based on the SNP that differ between these species. However, none of these papers clearly demonstrate whether the transcriptome profiling is correlated with a mouse-free sample doing this method. Due to sequence similarity I really doubt the resulting transcriptome data will provide a good estimation of tumor RNAs. AFAIK, the mouse genome (and consequently the transcriptome) is relatively similar to the human one with highly conserved sequences across these species (particularly for coding regions). I suspect such regions to not have different SNPs and not be able to correctly assess if a read is a mouse one or a human one.

Is anyone familiar with this situation? Is there a method to reliably get the tumor gene expression while removing the contamination from mouse RNAs? Should we just sort human cells by flow cytometry and establish a cell line to make sure no mouse cells survived?

[breckuh]

Would single cell RNAseq solve this? You could more reliably identify which cells come from the tumors versus the mice and then proceed with your transcriptome analysis.

Answer 1

[thomaskuilman]

As a matter of fact, a colleague of mine (Oscar Krijgsman) just wrote a new tool specifically for this type of analysis. It is based on the principle of aligning the reads to both human and mouse reference genomes, and then assigning reads to human / mouse based on their alignment qualities to these genomes. Please check out our GitHub page for the XenofilteR tool, and here for the accompanying paper in BMC Bioinformatics.

[Tom_L]

Thank you for you reply. This method (and some others) looks a good tool for WES or WGS because most tools are designed to accurately identified actual mutations (or somatic SVN) from mouse SNP. Some of them even try to detect structural variations! However, as highlighted in fig 6b, the transcriptome profiling is clearly different from unfiltered samples, filtered samples and patients. It is reported that: "after filtering with XenofilteR the expression of the selected genes better reflected the expression of the genes in the patient samples". I totally agree with "better reflected" but still, how distant is it from real tumors? This tells me that I may compare PDTX/PDTX (fig 6c-e) but it is still challenging to compare tumor with PDTX. Is the difference between samples in fig6b a biological one (e.g. same cells in difference conditions or cells that grow up are a subset of patient tumor) or a technical one (mouse reads)?

Answer 2

[Oscarkrijgsman]

Thank you for looking into our tool XenofilteR.

A tumor sample and a PDX sample will always be different for multiple reasons: 1) The tumor is growing in a mouse, not a human. This will affect expression and 2) Human sequence reads from PDX are of tumor origin only. Human tumor cells will always include stromal cells including immune cells etc. These factors will make comparing human tumor with PDX difficult using RNAseq data. As you mention correctly, this is much less of a problem with DNA sequence data as the stromal cells will have less influence other than diluting the tumor signal.

Regarding figure 6B; I think this is for a large part technical. The RNAseq of the human tumor and the PDX were done on different times, chemistry and research institutes. Unfortunately, no material for a direct comparison was available.

So I think filtering RNAseq data from PDX will certainly improve the comparison between the PDX data and human tumor. However, it will never be the same as a human tumor on a gene expression level.

[Tom_L]

Thank you for clarifying these points. This makes sense to me.