Comparing Ultraconserved Elements and Exons for Phylogenomic Analyses of Middle American Cichlids: When Data Agree to Disagree

Abstract

Choosing among types of genomic markers to be used in a phylogenomic study can have a major influence on the cost, design, and results of a study. Yet few attempts have been made to compare categories of next-generation sequence markers limiting our ability to compare the suitability of these different genomic fragment types. Here, we explore properties of different genomic markers to find if they vary in the accuracy of component phylogenetic trees and to clarify the causes of conflict obtained from different data sets or inference methods. As a test case, we explore the causes of discordance between phylogenetic hypotheses obtained using a novel data set of ultraconserved elements (UCEs) and a recently published exon data set of the cichlid tribe Heroini. Resolving relationships among heroine cichlids has historically been difficult, and the processes of colonization and diversification in Middle America and the Greater Antilles are not yet well understood. Despite differences in informativeness and levels of gene tree discordance between UCEs and exons, the resulting phylogenomic hypotheses generally agree on most relationships. The independent data sets disagreed in areas with low phylogenetic signal that were overwhelmed by incomplete lineage sorting and nonphylogenetic signals. For UCEs, high levels of incomplete lineage sorting were found to be the major cause of gene tree discordance, whereas, for exons, nonphylogenetic signal is most likely caused by a reduced number of highly informative loci. This paucity of informative loci in exons might be due to heterogeneous substitution rates that are problematic to model (i.e., computationally restrictive) resulting in systematic errors that UCEs (being less informative individually but more uniform) are less prone to. These results generally demonstrate the robustness of phylogenomic methods to accommodate genomic markers with different biological and phylogenetic properties. However, we identify common and unique pitfalls of different categories of genomic fragments when inferring enigmatic phylogenetic relationships.

Keywords: Heroini; gene tree heterogeneity; hybrid target capture; phylogenetic informativeness; signal–noise ratio; species trees.

Fig. 1

Species tree inferred in ASTRAL-III for the complete data set of UCEs of all cichlid species included in this study. All nodes are supported by local posterior probabilities=1.0 unless otherwise noted. Relevant clades mentioned in the text are indicated: CGH, crown-group herichthyines; BHG, basal herichthyine genera; WTC, Wajpamheros, Theraps, Chuco clade. Fish illustrations from top to bottom: Darienheros calobrensis, Caquetaia spectabilis, Nandopsis haitiensis, Trichromis salvini (by E. Alda).

Fig. 2

(A) Profiles of phylogenetic informativeness values estimated for the UCE and the exon data sets. The dark solid line is the mean value, the medium shade is 95% CI, and the light shade is the range between the 2.5th percentile and the 97.5th percentile. (B) Histograms showing the distribution of quartet distances among all gene trees inferred for common taxon set of cichlids the UCE and exon loci. The dashed lines indicate the average quartet distance of all gene trees to the ASTRAL species tree inferred for their respective marker type.

Fig. 3

Topological comparison of the species trees inferred using RAxML and ASTRAL-III for the common cichlid taxon set and the UCE (A and C) and exon (B and D) data sets. All nodes are supported by bootstrap values=100, or local posterior probabilities=1.0, unless otherwise noted. In (D), dashed lines indicate internodes that are potentially in the anomaly zone because they were shorter than the limit of a(x). Heroine groups that differ in their relative phylogenetic position among analyses are highlighted in blue (amphilophines), pink (caquetaines), gray (Nandopsis), and yellow (herychthyines).

Fig. 4

Cladograms representing conflicting relationships of heroine cichlids between the UCE and the exon data sets. Values above each branch indicate local posterior probabilities (PP) from the ASTRAL species trees and the gene concordance (gCF) and site concordance values (sCF). Below the branches, we indicate whether there is a significant probability that the data can reject equal frequencies of gene trees (gEFp) or sites (sEFp) supporting the alternative topologies.

Fig. 5

Violin plots showing the distribution of ΔGLS values (following Shen et al. 2017) across gene trees for each topology test and genomic marker type.