Using Genotype Abundance to Improve Phylogenetic Inference

William S. DeWitt, Luka Mesin, Gabriel D. Victora, Vladimir N. Minin, F. A. MatsenMolecular 2018 · DOI

TL;DR

This paper introduces a new method, GCtree, that leverages the abundance of different genetic sequences (genotypes) to improve the accuracy of phylogenetic tree inference. By incorporating this abundance information, GCtree significantly outperforms traditional methods, especially in scenarios with dense genetic sampling, like those found in immune system evolution.

What Problem Does This Paper Address?

Traditional phylogenetic methods reconstruct evolutionary relationships by comparing genetic sequences. However, these methods often treat each unique sequence as a distinct entity, ignoring how many times a particular sequence appears in the data. This is problematic in biological systems where many cells or individuals might carry the same genetic variant, leading to multiple samples of the same genotype. For example, in the immune system, B cells undergo a process called affinity maturation where their antibody genes (BCRs) are rapidly mutated and selected. This process generates a diverse population of B cells, many of which will share similar BCR sequences. Existing phylogenetic methods, by ignoring the abundance of these shared sequences, can lead to ambiguous or inaccurate evolutionary trees. This paper aims to solve this by developing a method that explicitly uses genotype abundance to refine phylogenetic inference.

Methods

The core of the GCtree method is an empirical Bayes framework that combines standard sequence-based phylogenetic optimality with a stochastic branching process model.

  • Stochastic Branching Process: This model captures the intuition that more abundant genotypes are more likely to have produced more descendant genotypes. It's a generative model for how genotypes might arise and proliferate over time.
  • Genotype-Collapsed Tree (GCtree): The method infers a GCtree, which represents the lineage of distinct genotypes and their observed abundances. This is a simplification of the full cellular lineage tree, focusing on the genetic relationships.
  • Parsimony Forest Augmentation: The authors leverage existing parsimony-based phylogenetic methods (which find trees with the minimum number of evolutionary changes) to generate an initial set of "equally optimal" trees (a parsimony forest). GCtree then uses its abundance-aware likelihood to rank these trees, resolving the degeneracy and selecting the most biologically plausible tree.
  • Dynamic Programming: To efficiently compute the likelihood of a GCtree, they employ dynamic programming, which allows for tractable calculations even with complex branching structures.

Datasets

The paper uses two main types of data for validation:

  • Simulated Data: They generated synthetic BCR lineage data using a custom simulation program designed to mimic biological realism (e.g., context-sensitive mutations, incomplete sampling) and to challenge the GCtree inference assumptions.
  • Experimental Data: They utilized single-cell BCR sequencing data from germinal centers of mice, specifically focusing on immunoglobulin heavy (IgH) and light (IgL) chain sequences from individual B cells. This experimental data is publicly available.

Key Insights and Ideas

  • Abundance is Information: The number of times a genotype appears in the data is not just noise; it's a crucial signal about its evolutionary history and its role in generating subsequent diversity.
  • Resolving Parsimony Degeneracy: Traditional parsimony methods often produce multiple equally valid trees. GCtree provides a principled way to choose among these by incorporating abundance, leading to more accurate reconstructions.
  • Biological Relevance: The method is particularly well-suited for rapidly evolving systems like the immune system, where dense sampling and genotype duplication are common.
  • Computational Tractability: Despite incorporating complex biological processes, the GCtree method is computationally feasible, allowing for practical application.

Key Results

  • Improved Accuracy in Simulations: GCtree consistently produced more accurate phylogenetic trees compared to standard parsimony methods when evaluated using Robinson-Foulds distance on simulated data. In simulations with parsimony degeneracy, GCtree MLEs were significantly closer to the true tree than other parsimony trees.
  • Empirical Validation: When applied to real germinal center B cell receptor data, GCtree successfully resolved phylogenetic ambiguities for both IgH and IgL loci. The inferred trees from the two loci showed greater concordance when GCtree was used, suggesting more accurate reconstructions.
  • Computational Efficiency: The runtime overhead for GCtree inference (after parsimony tree generation) was found to be negligible in their experiments, making it a practical addition to existing workflows.

Limitations and Future Work

  • Infinite-Type Assumption: The GCtree model assumes an "infinite-type" branching process, meaning each mutation creates a novel genotype. While this is a useful simplification, it doesn't perfectly reflect reality where identical sequences can arise independently. The authors acknowledge this and suggest future work could explore models that relax this assumption.
  • Conditional Independence: The model assumes independence between genotype sequence data (G) and abundance data (A) conditioned on the tree topology (T). While this simplifies computation, it's an approximation.
  • Future Directions: The authors suggest extending GCtree to non-neutral models (where fitness varies) and incorporating more sophisticated mutation models (e.g., distinguishing synonymous and non-synonymous mutations). They also envision augmenting GCtree with other sequence-based optimality criteria beyond parsimony, potentially requiring more computationally intensive tree space search. A fully Bayesian approach using packages like BEAST is also mentioned as a future avenue.

Summary updated

Abstract

Modern biological techniques enable very dense genetic sampling of unfolding evolutionary histories, and thus frequently sample some genotypes multiple times. This motivates strategies to incorporate genotype abundance information in phylogenetic inference. In this article, we synthesize a stochastic process model with standard sequence-based phylogenetic optimality, and show that tree estimation is substantially improved by doing so. Our method is validated with extensive simulations and an experimental single-cell lineage tracing study of germinal center B cell receptor affinity maturation.
BibTeX
@inproceedings{dewitt2018using,
  author = {William S. DeWitt and Luka Mesin and Gabriel D. Victora and Vladimir N. Minin and F. A. Matsen},
  title = {{Using Genotype Abundance to Improve Phylogenetic Inference}},
  booktitle = {Molecular},
  year = {2018},
  doi = {10.1093/molbev/msy020},
}
Figure