Katalin Voss ¹ Katrina M Kaur ² Rituparna Banerjee ² Felix Breden ³ Matt Pennell ^1*

• ¹ University of Southern California, Los Angeles, United States
• ² University of British Columbia, Vancouver, British Columbia, Canada
• ³ Simon Fraser University, Burnaby, British Columbia, Canada

The adaptive immune system generates a diverse array of B-cell receptors through the processes of V(D)J recombination and somatic hypermutation. B-cell receptors that bind to an antigen will undergo clonal expansion, creating a Darwinian evolutionary dynamic within individuals. A key step in studying these dynamics is to identify sequences derived from the same ancestral V(D)J recombination event (i.e. a clonal family). There are a number of widely used methods for accomplishing this task but a major limitation of all of them is that they rely, at least in part, on the ability to map sequences to a germline reference set. This requirement is particularly problematic in non-model systems where we often know little about the germline allelic diversity in the study population. Recognizing that delimiting B-cell clonal families is analogous to delimiting species from single locus data, we propose a novel strategy of reconstructing the phylogenetic tree of all B-cell sequences in a sample and using a popular species delimitation method, multi-rate Poisson Tree Processes (mPTP), to delimit clonal families. Using extensive simulations, we show that not only does this phylogenetically explicit approach perform well for the purpose of delimiting clonal families when no reference allele set is available, it performs similarly to state-of-the-art techniques developed specifically for B-cell data even when we have a complete reference allele set. Additionally, our analysis of an empirical dataset shows that mPTP performs similarly to leading methods in the field.These findings demonstrate the utility of using off-the-shelf phylogenetic techniques for analyzing Bcell clonal dynamics in non-model systems, and suggests that phylogenetic inference techniques may be potentially combined with mapping based approaches for even more robust inferences, even in model systems.