Over the years, I have an increasing appreciation of the cost effectiveness of tumor RNA-seq for precision cancer medicine research and clinical utility.
From TCGA tumor profiling, scientists have tried different clustering methods, from DNA mutations, mRNA-seq, miRNA expression, DNA methylation, and RPPA (proteomics). Although they often give different clustering results, the most informative in consensus clustering modeling of all the information together and tumor subtype classification is still probably RNA-seq. Of course, we can also learn a lot better how genes are differentially regulated in cancers, and Cistrome Cancer is one of our initial attempts. Sophisticated scientists can also investigate alternative splicing, RNA editing, non-coding RNA, or alternative polyadenylation .
How about tumor mutation analysis? The most functional oncogenic mutations can be directly detected from tumor RNA-seq. In recent neo-antigen prediction studies, the authors also only consider mutations that are highly expressed, which are detectable from RNA-seq. Unfortunately RNA-seq will miss tumor suppressor losses, but currently available cancer therapies are mostly targeting oncogene gain of function mutations instead of tumor suppressor loss, so RNA-seq will still be more informative to identifying targeted therapy.
To study tumor immune microenvironment, RNA-seq is quite informative in evaluating the immune infiltration and associating this with patient clinical features. There has been a number of papers published on this (CIBERSORT, CYT, TIMER, xCell), all based on tumor RNA-seq data. Also, we can obtain the patient HLA typing from tumor RNA-seq.
To obtain tumor immune repertoires, TCR-seq and BCR-seq are quite expensive and need separate assays for TCR alpha, beta, gamma, delta chains as well as BCR heavy and light chains. In contrast, RNA-seq can detect all of them from a single RNA-seq samples, at least the most abundant immune repertoire clones which are more likely to be recognizing tumor antigens. We have the TRUST algorithm to do this and are still improving the algorithm daily.
For therapy response biomarkers, traditionally oncologists use qPCR, small microarrays, and recently nano string. I wrote this blog about transcriptome as a biomarker in 2013, and I still believe in it 4 years later. Nowadays RNA-seq costs about $300 / sample (100M fragments of PE150) and the turn around time is usually one week, and these numbers continue to improve with sequencing technology development. RNA-seq provides much more information for an informative and robust biomarker.
One limitation is for RNA-seq is that sample quality is easier to maintain for DNA than for RNA especially for tumors obtained much longer before (e.g. in order to accumulate better survival information). However, physician scientists now are much more experienced with the processing or frozen storage of fresh tumors. Single-cell RNA-seq is gaining momentum in recent high profile studies, although it might be too costly and technically challenging to be adopted as a standard clinical assay, at least right now. Currently data analysis is still a bottle neck for many scientists, and here is one effort we made to streamline analysis. I hope the scientific community, including us, continue to develop the sample processing automation and data analysis algorithms and pipelines to help the coming of age for tumor RNA-seq.