Philip Dunne Profile picture
Apr 16 20 tweets 10 min read
Do you use transcriptional signatures?

Do you assume they represent the same biology in pre-clinical & clinical samples?

What if I told you some of the most widely-used signatures in cancer research are being completely misinterpreted.… read on👇👇🧵
For a long time we have been using clinical samples to identify epithelial biology and signalling in specific cells of interest, but in the end we knew we needed to find a better way to account for the histological heterogeneity of diagnostic tissue samples
Biomarker and signature development is increasingly performed in pre-clinical models (in vitro, in vivo or ex vivo) that are exquisitely suited for precise characterisation of discrete mechanistic signalling enabling almost absolute control over the experimental conditions
and while we know that the genes that make up these signatures will be perfectly representative of biological/mechanistic cascades in clean models, the same genes can be associated with completely unrelated signalling in tumours when expressed in non-epithelial lineages
Therefore, when signalling from tightly controlled pre-clinical models is compared to the unpredictable nature of diagnostic tumour samples, differences in epithelial-immune-stromal composition has the potential to completely skew the meaning of transcriptional signatures
Such misinterpretation is a major issue in the precision medicine era, as these signatures are used to guide therapeutic targeting of signalling pathways in pre-clinical models, however tumours identified by these signatures may not be driven by the same biological mechanisms
To examine if non-epithelial components of the tumour microenvironment can distort signature scores, leading to biological misinterpretation, we performed a comprehensive assessment of ~8000 commonly employed transcriptional signatures in cancer research
Using colorectal cancer as an exemplar, with tumour tissues from laser capture microdissection (LCM), flow cytometry (FACS), bulk clinical samples, single cell RNAseq and finally spatial transcriptomics, we enumerated lineage-specific expression of each signature
In LCM and FACS, some signatures are expressed at magnitudes higher in stromal cells compared to epithelium. So instead of representing the mechanistic signalling they were developed for, when assessed in bulk tumour data, these signatures become direct measures of tumour/stroma%
This is clearly evident when we apply the same signatures to bulk tumour data, where we see strong correlation between tumour/stromal components assessed by digital pathology and scores from some of the most widely-published biological signatures used in cancer research
This becomes even more startling when looking at multi-regional biopsies from a collection of patients, as we see that transcriptional signature scores are completely aligned to the % stroma in a biopsy sample **regardless of the patient the sample was taken from** 🤯
The use of spatial transcriptomics confirms these finding, but importantly also offer a solution by demonstrating the importance of profiling samples with high resolution methods, in combination with bulk profiling, to ensure that the biology we are interpreting is robust
If you have made it this far, and still think your favorite gene/signature (that you have developed, tested & used so many times over the years) would never be lying to you, then have we got a surprise for you..
To ensure you can benefit from our findings, we developed ConfoundR to easily interrogate the potential confounding effects on individual gene or signature across colorectal, pancreatic, breast, ovarian and prostate cancer datasets - why not try it out!
This work has been led by the joint first authorship pairing of the dynamic duo, Natalie @NatalieFisher03 and Ryan, and involved numerous discussions (& only a few arguments) within our Belfast, Oxford, Glasgow and Zurich collaborative networks
It was also only possible through the establishment of a new collaborative joint senior-author link between our lab in Belfast and Nigel @nigeljamieson @LabSpatialNBJ and his team in Glasgow.

The Belfast-Glasgow link is rapidly expanding and more news on that soon!
We now turn our attention back to the initial aim; characterising and understanding intrinsic tumour cell signalling and their communication networks with the surrounding microenvironment. This understanding is vital to build on advances in early detection.. watch this space!
Thanks to @CRUKresearch who fund my #EarlyDetection work, team members in both mine and Nigel's labs, and the @AcrCelerate network, with collaborator funding from Swiss National Science Foundation, to the @The_MRC for the S:CORT programme and to QUB Foundation @QUBAlumni
Our pre-print is available on @biorxivpreprint…

Confoundr is free here for "clicky-button" production of heatmaps, boxplots and GSEA that you can use directly in your future presentations and publications, dont forget to cite us!
While we selected a wide range of cancer tissues, if you have your own stroma/epithelium (or any other cancer-relevant contrast) microdissected tissue datasets and you want us to add them into the ConfoundR platform, drop us a line!

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Philip Dunne

Philip Dunne Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!


0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy


3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!