Back

Accounting for B cell behaviour and sampling bias yields a superior predictor of anti-PD-L1 response in bladder cancer

Dyugay, I. A.; Lukyanov, D. K.; Turchaninova, M. A.; Zaretsky, A. R.; Khalmurzaev, O. A.; Matveev, V. B.; Shugay, M. M.; Shelyakin, P. V.; Chudakov, D. M.

2021-03-04 immunology
10.1101/2021.03.04.433370 bioRxiv
Show abstract

Tumor-infiltrating B cells and intratumorally-produced immunoglobulins (IG) play important roles in the tumor microenvironment and response to immunotherapy1-5. IgG antibodies produced by intratumoral B cells may drive antibody-dependent cellular cytotoxicity (ADCC) and enhance antigen presentation by dendritic cells6-8. Furthermore, B cells are efficient antigen-specific antigen presenters that can essentially modulate the behaviour of helper T cells9-11. Here we investigated the role of intratumoral IG isotype and clonality in bladder cancer. Our results show that the IgG1/IgA ratio offers a strong and independent prognostic indicator for the Basal squamous molecular subtype and for the whole ImVigor210 cohort in anti-PD-L1 immunotherapy. Our findings also indicate that effector B cell functions, rather than clonally-produced antibodies, are involved in the antitumor response. High IgG1/IgA ratio was associated with relative abundance of cytotoxic genes and prominence of the IL-21/IL-21R axis suggesting importance of T cell/B cell interaction. We integrated the B, NK, and T cell components, employing immFocus-like normalization to account for the stochastic nature of tumor tissue sampling. Using a random forest model with nested cross-validation, we developed a tumor RNA-Seq-based predictor of anti-PD-L1 therapy response in muscle-invasive urothelial carcinoma. The resulting PRIMUS (PRedIctive MolecUlar Signature) predictor achieves superior sensitivity compared to PD-L1 expression scores or existing gene signatures, allowing for reliable identification of responders even within the desert patient subcohort analyzed as a hold out set.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.