Integrating B Cell Lineage Information into Statistical Tests for Detecting Selection in Ig Sequences

Abstract

Detecting selection in B cell immunoglobulin (Ig) sequences is critical to understanding affinity maturation, and can provide insights into antigen-driven selection in normal and pathologic immune responses. The most common sequence-based methods for detecting selection analyze the ratio of replacement (R) and silent (S) mutations using a binomial statistical analysis. However, these approaches have been criticized for low sensitivity. An alternative method is based on the analysis of lineage trees constructed from sets of clonally-related Ig sequences. Several tree shape measures have been proposed as indicators of selection that can be statistically compared across cohorts. However, we show that tree shape analysis is confounded by underlying experimental factors that are difficult to control for in practice, including the sequencing depth and number of generations in each clone. Thus, though lineage tree shapes may reflect selection, their analysis alone is an unreliable measure of in vivo selection. To usefully capture the information provided by lineage trees, we propose a new method that applies the binomial statistical method to mutations identified based on lineage tree structure. This hybrid method is able to detect selection with increased sensitivity in both simulated and experimental data sets. We anticipate that this approach will be especially useful in the analysis of large-scale Ig sequencing data sets generated by high-throughput sequencing technologies.