RNAseq differential expression analysis : relevance of Immunoglobulin genes enrichment ?

Question

[guillaume.rbt]

Hi all,

I'm currently studying RNAseq samples, obtained from human tumor biopsies before treatment (111 samples).

I'm comparing responders and non responders to the treatment, with the aim of identifying response biomarkers.

In my results, I detect a strong enrichment for up-regulation of Immunoglobulins variable genes (from both heavy and light chains).

I'm not an immunologist, and I'm questionning the relevance of this observation.

As those genes have been subject to somatic recombination and somatic hypermutation, how confident should I be in the assessment of their expression ?

Could the detection of multiple variable immunoglobulin genes as differentially expressed happen to be an artifact of bad read mapping due to the fact that they belong to the same family?

Thanks for any input on this topic

[ATpoint]

Are these tumors which undergo (so the cells of origin) SHM? So essentially are these B-cell tumors? Can you also give some basic information about the treatment? Is it something that could activate an immune response? How early after the treatment have the specimen been collected?

[guillaume.rbt]

Thank for your questions. The samples are collected before treatment, so the treatment has no effect on the profiles. I try to detect biomarkers that are predictive of a good response. The treatment tested is anti-PD1 checkpoint inhibitors, and it is expected to have a higher immune response activity with the responders. I indeed guess that the detected signal could be linked to tumor infiltrating B-cell, but I'm trying to assess the relevance of this assumption.

[jared.andrews07]

You could also spot-check other non-Ig B cell markers (CD79A, MS4A1) to see if they also match up.

[guillaume.rbt]

Ok thank for the tip ! (both markers are indeed also up-regulated :) )

Answer 1

[eric.audemard]

Hi,

If you want to check, if it's an artefact (or false positive) I will recommend you to check the expression profile as I did it in this poster (check graphics under each heatmap). Also if you have small number of samples (3 vs 3), I will recommend you to use kallisto with n=100 bootstrap, and visualised expression profile in there shiny app using (sleuth).

As it's explain in the sleuth paper and especially when you have a small number of samples, you need to take account that there are a variance hidden at several steps of your pipeline: - In your sequencing: If you sequence a second times your samples, some genes expression will be different. - In your mapping: from ambiguously mapping reads - ... and certainly in other steps

This variance seems to be under estimate in standard DEG analysis pipeline and could be the source of a lot of false positive, as it's explain in the sleuth paper (see figure 1). In my experience this mostly the case on low expressed genes, but not only. Using Kallisto bootstrap is a good way to try to "check" it, but the best way to try to understand what's happens in your data, it is to plot your genes expression profil.

[guillaume.rbt]

Hi Eric and thank for your advices. I have in this case a high number of samples (111). I encountered similar results with what you described in your poster (better results with Limma/voom than with DEseq2 or EdgeR)

[eric.audemard]

Ok, you want predictive biomarker and you work on a patient cohort, ... this is exactly why I develop EPCY. Unfortunately, I havan't finish the paper but If you are curious and want to test it, EPCY is available on github.

[guillaume.rbt]

Ok great, I will definitely test it !