PhyloWGA: chromosome-aware phylogenetic interrogation of whole genome alignments

Abstract

Summary

Here, we present PhyloWGA, an open source R package for conducting phylogenetic analysis and investigation of whole genome data.

Availabilityand implementation

Available at Github (https://github.com/radamsRHA/PhyloWGA).

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Whole genome data hold promise for refining species tree estimates because, at a fundamental level, such large datasets offer unprecedented opportunity to dramatically increase the sample size and resolution of phylogenetic analyses (Hobolth et al., 2007; Jennings, 2016; White et al., 2009). However, enthusiasm for conducting phylogenomic analyses across whole genome alignments (WGAs) is often curbed by the inherent challenges that arise when extracting and translating the deluge of information encoded within WGAs into meaningful estimates of evolutionary relationships.

An immediate challenge is to address the pervasive phylogenetic conflict observed in whole genome data. Both biology and methodology can generate discord among phylogenetic estimates across loci (Som, 2015), and understanding the causes and consequences of such conflict is paramount to mitigating its impacts (e.g. Reddy et al., 2017). Many evolutionary processes are known to generate phylogenetic conflict (e.g. Adams et al., 2018; Kutschera et al., 2014), yet incomplete lineage sorting (ILS) is perhaps the most well-studied and biologically relevant (Degnan and Rosenberg, 2009; Edwards, 2009). Recombination decouples the phylogenetic histories of genomic regions, yielding unlinked genealogies that may (or may not) agree with the overall species tree due to ILS. Consequently, an entire chromosome cannot be reasonably modeled as a single non-recombining locus with only a single tree (Kubatko and Degnan, 2007), as it constitutes a mosaic of genealogies sequentially distributed along its length according to crossover events.

Even with whole genomes, tracts of contiguous well-aligned sequences are often confined to a few kb, which may be insufficient for accurate gene tree estimation. Thus, while it is possible to accommodate ILS by simply inferring separate gene trees across loci, these trees may be unreliable due to gene tree estimation error (Roch and Warnow, 2015). An ideal phylogenomic dataset would effectively accommodate both ILS and gene tree error by using sufficiently long yet recombination-free loci. However, constructing such a dataset is difficult, and constructing ‘supergenes’ (i.e. concatenated loci consistent with a single gene tree) can be challenging (Adams and Castoe, 2019; Jennings, 2016; White et al., 2009).

Consideration of ILS, recombination and other factors contributing to gene tree variation is therefore important for any attempt to accurately reconstruct evolutionary relationships from multilocus data for which the degree and distribution of phylogenetic conflict is unknown (Springer and Gatesy, 2016). While many methods exist for estimating individual gene trees (e.g. Stamatakis, 2014), sampling and extracting phylogenetic loci (e.g. Bravo et al., 2019; Costa et al., 2016) or inferring species trees (e.g. Boussau et al., 2013), few bioinformatics platforms are specifically designed for conducting detailed and efficient phylogenomic interrogation of WGAs. To address this, we introduce PhyloWGA, an open-source R package developed as a user-friendly suite for conducting streamlined phylogenetic analysis and investigation of WGAs. At its core are two bioinformatics pipelines, Chromo.Phylome and Chromo.Crawl, designed to reconstruct a chromosome-specific set of gene trees and to apply a series of model-based phylogenetic congruency tests along the length of a WGA, respectively. PhyloWGA is flexible and extensible for the easy incorporation of future phylogenetic investigations of genome-scale datasets. Importantly, while other approaches have attempted to cluster loci into supergenes (e.g. Bayzid et al. 2015; Mirarab et al. 2014), recent studies have shown that such approaches may be plagued with high rates of error (Adams and Castoe, 2019; Liu and Edwards 2015; Roch et al., 2019). Unlike previous ‘statistical binning’ methods (Bayzid et al. 2015; Mirarab et al. 2014), which ignore linkage information (i.e. proximity of adjacent loci in assembled genomes), here we develop an approach that inherently incorporates the spatial organization of loci and uses a model-based framework to explicitly test whether such loci share a common tree (see Supplementary Note for further discussion).

2 Implementation

PhyloWGA is written in R 3.6.1 (R Foundation for Statistical Computing, 2018) and includes a suite of functions for WGA-scale phylogenomic analyses. It is built upon several maximum likelihood (ML) phylogenetic frameworks, including IQ-TREE (Nguyen et al., 2014) and CONCATEPILLAR (Leigh et al., 2008). The primary input for PhyloWGA is a WGA in fasta format partitioned by chromosome, and PhyloWGA includes a number of functions for processing WGAs for this purpose. PhyloWGA allows users to specify an array of experimental parameters that define the scope of a PhyloWGA analysis, including the size and distribution of chromosomal windows, as well as the particular type of analysis.

The function Chromo.Phylome uses IQ-TREE to estimate a ‘chromosome-specific phylome’—a set of locus-specific phylogenetic trees for a given chromosomal alignment. On completion, Chromo.Phylome outputs a set of phylogenetic tree models fitted to each respective genomic window. The IQ-TREE commands are customizable, such that Chromo.Phylome can accommodate diverse analyses, including tree inference, model selection and model adequacy tests across a WGA. Finally, replicate analyses can be run to explore different experimental settings.

The module Chromo.Crawl ‘crawls’ across a WGA, testing for congruence among contiguous windows, and concatenates them when there is insufficient evidence of gene tree disagreement. It employs the program CONCATEPILLAR, which implements a likelihood-based model test to assess whether a set of loci likely share the same phylogeny. Chromo.Crawl sequentially applies this model test to contiguous windows, and then concatenates these windows into a single supergene if there is evidence of congruency (see Supplementary Note for details). Chromo.Crawl will then move to the next adjacent window to assess whether it can be combined with the previous windows. This merging of contiguous windows into a supergene continues until the model test reveals discordance, with the process restarting at the current window.

3 Biological application

We applied PhyloWGA on a WGA of 12 primates (Moorjani et al., 2016), to characterize the phylogenetic landscape of human chromosome 1. We coarsely (100 kb windows) estimated locus phylogenetic trees by setting Chromo.Phylome to conduct ML-based phylogenetic analysis at each window with a GTR + Γ substitution model, and to then compute tree distances between windows. We next investigated phylogenomic conflict at a fine scale (10 kb windows) by applying Chromo.Crawl to track concatenated window sizes (i.e. supergene lengths) across the chromosome.

We used PhyloWGA to ‘paint a picture’ of the chromosome-wide landscape of phylogenomic conflict for this WGA. Visualizing pairwise distances among genomic regions provides a detailed depiction of conflict (Fig. 1a), indicating higher levels of conflict (darker shades) nearby telomeres and the centromere, as well as intermittent bands of high and low discordance throughout. We recovered 38 supergenes for the first six Mb of the chromosome, with a mean of ∼116 kb per supergene (Fig. 1b). Though the underlying reasons for such widespread discordance in these data remain unknown, both biology and methodology likely play a role. For example, high levels of discordance observed in particular regions (i.e. darker bands of Fig. 1a and shorter supergenes in Fig. 1b) may represent poor alignment quality, regions of high recombination, or other processes unique to such regions. Both the accuracy and computational efficiency of PhyloWGA are influenced by the experimental design (e.g. window size and distribution) and evolutionary conditions (e.g. recombination rate and nucleotide model complexity) of the particular WGA analysis (see Supplementary Note for details).

Fig. 1.

(a) Heatmap of pairwise Robinson-Foulds (Robinson and Foulds, 1981; upper triangle) and geodesic (Billera et al., 2001; lower triangle) distances along windows of human chromosome 1 (larger distances as darker shades). (b) Number of concatenated bases indicated for each of the 38 supergenes (locations defined as alternating green, red and blue) by applying Chromo.Crawl to the first six Mb of the chromosome

Funding

This work was supported by National Science Foundation grants to M.D. [DEB-1753489 and BCS-2001063] and T.A.C. [DEB-1655571] and by a National Institutes of Health grant to M.D. [R35M128590].

Conflict of Interest: none declared.

Data availability

The data underlying this article are available in the online supplementary material.