Thread by @coregenomics

Two recent preprints report almost opposite views on the use of #TMB in IO therapy.
For: @CharlesSwanton et al researchsquare.com/article/rs-764…

Against: @leonidmirny biorxiv.org/content/10.110…

Who's right...answers on a postcard please!

For (1) Clonal TMB had the strongest effect size across all 15 studies in the (Odds Ratio for “CR/PR” vs “SD/PD” = 1.88, 95% confidence interval [1.52 – 2.33], p=8.08 x 10-9).

For (2) They propose a 12-feature multivariate model (AUC of 0.86 in a pan-tumor cohort), which was superior to TMB alone (AUC=0.59). Improved AUC was confirmed in KEYNOTE-028 and Liu et al. NMED 2019. Their model identifies subgroups with <5%) or ~50% likelihood of response

For (3) To do this the authors collated raw sequencing data from 15 studies assessing 8 indications and 3 IO therapy classes (anti-CTLA4, anti-PD-1 and anti-PD-L1) in 1,283 patients of which 700 patients also had RNA-seq data.

Against (1)...the preprint describes "limited evidence of TMB as an IO response biomarker" in any disease excepting NSCLC in smokers. They suggest that previously reported associations arise from a combination of confounding disease subtypes and incorrect statistical testing

Against (3) The authors present a mathematical model that extends the neoantigen theory, is consistent with the lack of association between TMB and response to ICB and highlights the role of immunodominance.

The authors call for caution in the use of TMB as a biomarker.

Against (4) Failures of ICB likely arise due to factors independent of cancer immunogenicity since cancer would have to develop means to suppress the immune response early in its development. The authors suggest testing this prediction in future studies of cancer clonal evolution