Must an experimentally proven disease-relevant gene appear significantly differentially expressed on RNA sequencing analysis?

Question

[charlesgwellem]

Hi. I have a bulk RNA sequencing dataset from the skin of patients with systemic sclerosis (SSC) and I wish to understand why an experimentally proven, potentially disease-relevant gene does not appear upregulated or differentially expressed upon analysis with DESEq2. In bleomycin mouse models of SSC, my lab mates have proven through pcr, immunoflourescence and hydroyproline assay that gene X was highly expressed in bleomycin-treated mice compared to healthy controls. Proofs through immunoflourescence, histology and hydroxyproline assays on human skin samples of SSC have shown that gene X could play a role in SSC.

A bulk RNA seq experiment was perfomed on cultured fibroblasts isolated from the skin of SSC patients with the hope of finding gene X being differentially expressed. But after running DESEq2 with default settings, and taking in to account major sources of variation in the dataset in the design formula of DESEq2, gene X does not appear differentially expressed. I have however noticed that at the level of the normalised counts that the arithmetic average of gene X is higher in SSC versus the normal control samples.

What could explain why a gene, known or experimentally proven to be disease-relevant does not appear as differentially expressed.

Thanks in advance for your kind help.

[ATpoint]

Maybe it is a post-transcriptional mechanism that renders the gene being overexpressed on the protein- rather than transcript level.

[charlesgwellem]

Thanky for your response. The result is also validated on RNA level by PCR.

[dsull]

A few things come to mind:

1. Your RNA-seq experiment is underpowered to detect the desired effect (with patient samples, you definitely may need larger sample sizes to see an effect, which brings me to point 2).
2. Heterogeneity: If your samples are so heterogeneous, there is no design formula that can possibly correct for all the messiness in the data. This is what you often see in patient data (e.g. TCGA). Don't ever expect a human cohort to look like a well-controlled biological experiment.
3. Mouse != Human. Yes, I know you have protein-level evidence from human skin samples, but patient skin samples != patient-derived cultured fibroblasts.

In short: You are attempting to validate your findings by doing an experiment that is completely different. Going from differentially expressed in a mouse model -> differentially expressed in cultured cells from patient tissue is quite a biological leap.