1. Computational and Systems Biology
  2. Evolutionary Biology
Download icon

Correlated evolution between repertoire size and song plasticity predicts that sexual selection on song promotes open-ended learning

  1. Cristina M Robinson
  2. Kate T Snyder
  3. Nicole Creanza  Is a corresponding author
  1. Vanderbilt University, United States
Research Article
Cite this article as: eLife 2019;8:e44454 doi: 10.7554/eLife.44454
9 figures, 9 tables, 2 data sets and 5 additional files

Figures

Figure 1 with 8 supplements
Syllable repertoire size is larger in species with adult song plasticity even when controlling for phylogenetic relationships.

These phylogenies show the calculated evolution of natural-log transformed syllable repertoire size and either (A) stable and plastic song stability states or (B) early song-stable, delayed song-stable, and song-plastic states. Dots at the tips of branches represent the current song-stability state. Pie charts represent the likelihood that the common ancestor at that node was in each song-stability state. Dark purple colors represent small syllable repertoires while white represents large repertoires. For the sake of visualization, the color range was truncated based on the distribution of the data, such that the lowest value was the 25th percentile minus the range of the 25th to 50th percentile and the highest value was the 75th percentile plus the range of the 50th to 75th percentile. See Table 1 for PhylANOVA results.

https://doi.org/10.7554/eLife.44454.003
Figure 1—source data 1

Predicted values for internal nodes for each trait.

There are three plots with associated tables for each song characteristic: The first plot and table set gives the values when the binary categorization of song stability (song-stable and song-plastic) is used. The second plot and table set gives the values when the ternary categorization of early song-stable, delayed song-stable, and song-plastic is used. The final plot and table set gives values when the continuous categorization is used. In the table, ‘Node’ corresponds to the numbered boxes on the tree nodes. ‘State Likelihood’ gives the probability that the common ancestor at a node was in a given leaning state when discrete categorizations are used. S = song stable, p=song plastic, E = early song-stable, D = delayed song-stable. ‘Length of Plasticity’ predicts the value along a continuous spectrum. ‘Trait’ is the predicted value of the song trait being examined in the associated tree for each internal node.

https://doi.org/10.7554/eLife.44454.012
Figure 1—figure supplement 1
Minimum number of evolutionary transitions required to recapitulate the current song stability states of birds in this study.

White dots at the tips show song-plastic species, while black dots show song-stable species. Inset shows an equally parsimonious set of transitions for the labeled species. At least seven transitions to the song-stable state are required to generate the current distribution of adult song plasticity from a song-plastic last common ancestor. At least nine transitions to the song-plastic state are required if the last common ancestor was a song-stable species. Fourteen transitions between adult song stability and adult song plasticity are required, regardless of whether we assume a song-plastic or song-stable ancestral state.

https://doi.org/10.7554/eLife.44454.004
Figure 1—figure supplement 2
There was a significant relationship between song repertoire size and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity and for log-transformed song repertoire size. Black and white dots represent a species that is currently in a stable or plastic state respectively. Black and white in the pie charts at each node represent the likelihood that the common ancestor was in the stable or plastic state. Dark purple colors represent small repertoires while white represents large repertoires. For the sake of visualization, the color range was truncated based on the distribution of the data, such that the lowest value was the 25th percentile minus the range of the 25th to 50th percentile and the highest value was the 75th percentile plus the range of the 50th to 75th percentile. PhylANOVA results for Figure 1—figure supplements 27 are available in Table 1.

https://doi.org/10.7554/eLife.44454.005
Figure 1—figure supplement 3
There was no relationship between syllables per song and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity (black versus white) and for log-transformed syllables per song (purple). Labeling is the same as in Figure 1—figure supplement 2.

https://doi.org/10.7554/eLife.44454.006
Figure 1—figure supplement 4
There was no relationship between song duration and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity (black versus white) and for log-transformed song duration (purple). Labeling is the same as in Figure 1—figure supplement 2.

https://doi.org/10.7554/eLife.44454.007
Figure 1—figure supplement 5
There was no relationship between intersong interval and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity (black versus white) and for log-transformed intersong interval (purple). Labeling is the same as in Figure 1—figure supplement 2.

https://doi.org/10.7554/eLife.44454.008
Figure 1—figure supplement 6
There was no relationship between song rate and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity (black versus white) and for log-transformed song rate (purple). Labeling is the same as in Figure 1—figure supplement 2.

https://doi.org/10.7554/eLife.44454.009
Figure 1—figure supplement 7
There was no relationship between song continuity and song stability when controlling for phylogeny.

The estimated ancestral character states are mapped on the tree for both adult song stability versus plasticity (black versus white) and for log-transformed song continuity (purple). Labeling is the same as in Figure 1—figure supplement 2.

https://doi.org/10.7554/eLife.44454.010
Figure 1—figure supplement 8
There was no significant correlation between the rate of syllable repertoire evolution and ancestral syllable repertoire size in closely related species pairs (Pearson’s r = 0.101, p=0.6715).

The 20 species pairs that form monophyletic groups in our 67-species tree are included. Each point represents one species pair. A linear model yields a best-fit line of y = 0.0228 x + 0.131. Thus, the rate of syllable repertoire size evolution was not dependent upon the ancestral values of syllable repertoire size.

https://doi.org/10.7554/eLife.44454.011
Comparison of song stability state and song examples in Phylloscopus species.

P. collybita, a species with adult song stability, has a smaller syllable repertoire size than P. trochilus and P. fuscastus, two species with adult song plasticity. Colors of branches and nodes correspond with Figure 1. Sonograms were generated from recordings obtained from xeno-canto.org: XC340281 recorded by Tom Wulf (P. fuscatus), XC414221 recorded by Frank Lambert (P. collybita), and XC402265 recorded by Hans Matheve (P. trochilus). Sonograms are used only to demonstrate comparative repertoire size from one individual for each species and were stretched horizontally to fit the allotted space.

https://doi.org/10.7554/eLife.44454.014
Distribution of repertoire sizes in species with different song stability states.

(A) shows the distribution of syllable repertoires and (B) shows the distribution of song repertoires when species are broken into two groups based on song stability. (C) shows the distribution of syllable repertoires when species are broken into three groups based on song stability. Boxes indicate the 25th, 50th, and 75th percentile. The lower whisker is either the minimum value or the 25th percentile minus 1.5 times the interquartile range, whichever was larger. The upper whisker is either the maximum value or the 75th percentile plus 1.5 times the interquartile range, whichever was smaller. Dots are the raw values as a scatter plot. Solid dots are within the range of the box and whiskers, while open dots are outliers. (A-B) Species with adult song plasticity had significantly larger syllable and song repertoires. See Table 1 for full PhylANOVA results. (C) Species with adult song plasticity had significantly larger syllable repertoires than early song stable and delayed song stable species, but there was no significant difference between early and delayed song stable species (p=0.659). See Tables 2 and 3 for full PhylANOVA results. (D) shows the continuous relationship between syllable repertoire size and years spent learning, where song plasticity is truncated at 2 years due to lack of data from subsequent years.

https://doi.org/10.7554/eLife.44454.015
Distributions of rates for natural-log transformed song traits related to complexity.

The boxes at the top illustrate how we grouped the species for each model. Column 1 (A-C): Blue traces are song-stable, while red traces are song-plastic. Column 2 (D-F): Blue traces are early song-stable, purple traces are delayed song stable, and red traces are song-plastic. Column 3 (G-H): Blue traces are early song-stable, while red traces are delayed song-stable and song-plastic combined. The black line shows the rate value for the one-rate model in all columns. Asterisks indicate that the rate of evolution of that song characteristic significantly differed between groups. Lowercase zeta (ζ) the multi-rate model that best fit the data while using the fewest number of rates. In the case of syllable repertoire, the multi-rate models were not significantly better than the one-rate model. See Tables 59 for chi-square test results.

https://doi.org/10.7554/eLife.44454.020
Distributions of rates of evolution for natural-log transformed song traits related to performance.

Blue traces song-stable, while red traces are song-plastic. The black line shows the rate for the one-rate model. Asterisks indicate that the rate of evolution of that song characteristic significantly differed between song-stable and song-plastic lineages. See Table 5 for chi-square test results.

https://doi.org/10.7554/eLife.44454.021
Figure 6 with 6 supplements
Analysis of correlated evolution between adult song stability/plasticity and syllable repertoire size.

We repeated the BayesTraits analysis using each value of the continuous song trait as the threshold delineating the larger and smaller syllable repertoire groups. We performed a total of 100 runs per threshold. We pooled the results of all the runs into three groups based on whether the threshold was in the lowest, middle, or highest third of the unique trait values. Within these groups, we computed the mean percentage of runs that were significant at p<0.05 at each threshold. (A-C) Rate of transition plots when the lowest (red), middle (yellow), and highest (blue) thirds of the unique syllable repertoire values in the dataset were used as the threshold. Rates are the average across all runs when the threshold denoting small/large repertoire sizes was defined as each value within each segment. Arrows are labeled with the mean rate and the 95% confidence interval. Arrow weights are scaled to the mean rate values. (D) p-values from the 100 runs per threshold, plotted against threshold. Colored bars denote low, middle, and high threshold segments. Blue line denotes p=0.05.

https://doi.org/10.7554/eLife.44454.026
Figure 6—source data 1

Mean transition rates from BayesTraits analysis between song stability and song features averaged over 2, 4, and 5 bins.

On page 1, the top two panels depict the mean rates from all runs across the values of syllable repertoire in the lower and upper half of thresholds, respectively. The bottom panel shows the p-values from the 100 runs per threshold, plotted against threshold. Colored bars denote low and high threshold segments. The blue line denotes p=0.05. On page 2, the top four panels depict the mean rates from all runs across the values of syllable repertoire in the first (top left), second (top right), third (middle left), and fourth (middle right) quartiles of the thresholds. The bottom panel follows the pattern of the bottom panel from page 1. On page 3, the five panels (top left and right, middle left and right, and bottom left) depict the mean rates from all runs across the values of syllable repertoire in the five quintiles of the of the thresholds, respectively. The bottom panel follows the pattern of the bottom panels from pages 1 and 2. The three-page pattern repeats for each song feature.

https://doi.org/10.7554/eLife.44454.027
Figure 6—source data 2

BayesTraits analysis of syllable repertoire and learning window, jackknifed across families.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.034
Figure 6—source data 3

BayesTraits analysis of syllables per song and learning window, jackknifed across families.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.035
Figure 6—figure supplement 1
BayesTraits analysis on song stability and syllables per song.

We repeated the BayesTraits analysis using each value of the continuous song trait, here, syllables per song, as the threshold delineating the ‘high’ and ‘low’ syllables per song groups. We performed a total of 100 runs per threshold. We pooled the results of all of the runs into three groups based on whether the threshold was in the lowest, middle, or highest third of the unique trait values. Within these groups, we computed the mean number of runs that were significant at p<0.05 at each threshold. (A-C) Rate of transition plots of the lowest (red), middle (yellow), and highest (blue) thirds of the unique syllables per song values in the dataset. Rates are the average across all runs when the threshold denoting low/high syllables per song was defined as each value within each segment. Arrows are labeled with the mean rate and the 95% confidence interval. Arrow weights are scaled to the mean rate values. (D) p-values from the 100 runs per threshold, plotted against threshold. Colored bars denote low, middle, and high threshold segments. Blue line denotes p=0.05.

https://doi.org/10.7554/eLife.44454.028
Figure 6—figure supplement 2
BayesTraits analysis on song stability and song duration.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.029
Figure 6—figure supplement 3
BayesTraits analysis on song stability and intersong interval.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.030
Figure 6—figure supplement 4
BayesTraits analysis on song stability and song rate.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.031
Figure 6—figure supplement 5
BayesTraits analysis on song stability and song continuity.

Labeling the same as in Figure 6 and Figure 6—figure supplement 1.

https://doi.org/10.7554/eLife.44454.032
Figure 6—figure supplement 6
BayesTraits analysis of song stability and mating behaviors performed over 1000 runs.

Transition arrows are labeled with the mean rate over all 1000 runs. (A) Mating system (polygyny vs. monogamy): 1000 of 1000 runs were significant at p<0.05. (B) Extra-pair paternity (Low EPP vs. High EPP): No runs were significant at p<0.05.

https://doi.org/10.7554/eLife.44454.033
Analysis of correlated evolution between adult song stability/plasticity and song repertoire size.

Labeling is the same as in Figure 6.

https://doi.org/10.7554/eLife.44454.036
Figure 7—source data 1

BayesTraits analysis of song repertoire and learning window, jackknifed across families.

Labeling the same as in Figure 6 and Figure 6—figure supplement 2.

https://doi.org/10.7554/eLife.44454.037

Tables

Table 1
PhylANOVA results for all song traits when birds are divided into species with adult song stability or adult song plasticity.

Song traits are sorted from most to least significant. Song-stable and song-plastic columns show mean values of each log-transformed song trait. Corrected α indicates the threshold for significance with the Holm-Bonferroni correction.

https://doi.org/10.7554/eLife.44454.013
Song traitSong-stableSong-plasticF-ValueCorrected αp-value
Syllable repertoire1.88073.94641.50640.0071<0.001*
Song repertoire1.10553.868833.83340.0083<0.001*
Syllables/song1.25562.29629.26580.010.094
Duration0.77361.29272.07830.01250.42
Continuity-1.3453-1.02862.15370.01670.474
Interval1.60751.2181.38790.0250.567
Song rate1.89692.09710.60790.050.713
  1. *Denotes traits with significantly different groups.

Table 2
PhylANOVA results for all song traits when birds are divided into early song stability, delayed song stability, and song plasticity.

Song traits are sorted from most to least significant. Early, delayed, and plastic columns show mean values of each log-transformed song trait. Corrected α indicates the threshold for significance with the Holm-Bonferroni correction.

https://doi.org/10.7554/eLife.44454.016
Song traitEarlyDelayedPlasticF-ValueCorrected αp-value
Syllable repertoire1.64362.00623.94617.10990.00710.003*
Song repertoire0.67881.48193.868812.880.00830.011*
Syllables/song1.28521.24672.29623.68770.010.252
  1. *Denotes traits with significantly different groups.

Table 3
Post-hoc pairwise phylANOVA tests for significant song traits when birds are divided into early song stability, delayed song stability, and song plasticity.
https://doi.org/10.7554/eLife.44454.017
Song traitState 1State 2T-Valuep-value
Syllable repertoirePlasticDelayed4.89950.012*
Syllable repertoireEarlyPlastic4.60910.003*
Syllable repertoireEarlyDelayed0.68720.659
Song repertoirePlasticDelayed4.02680.044*
Song repertoireEarlyPlastic4.30740.015*
Song repertoireEarlyDelayed1.04440.55
  1. *Denotes traits with significantly different groups.

Table 4
Results of PGLS analysis between song characteristics and continuous song stability.

Test performed on the natural-log scaled values of song characteristics. λ is the value by which off-diagonal elements in the Brownian motion model are multiplied to make the correlation structure. Corrected α indicates the threshold for significance with the Holm-Bonferroni correction. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.018
Song traitSlopeStd errorλT-ValueCorrected αp-value
Syllable repertoire0.90670.24490.89133.70210.0071<0.001*
Song repertoire1.10130.31230.83163.52630.0083<0.001*
Syllables/song0.37010.22240.46991.66420.010.1029
Interval0.42210.26460.88231.59530.01250.1215
Continuity-0.21350.14390.8832-1.48380.01670.1486
Duration0.37020.25691.01631.4410.0250.1578
Song rate-0.21130.250.7307-0.84530.050.4048
  1. *Denotes significant slopes.

Table 5
Brownie results for song traits when birds are divided into species with adult song stability or adult song plasticity.

Rate columns show mean log likelihood. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.019
Song traitOne rateTwo ratesp-value
Syllables/song-110.6482-100.7673<0.001*
Song rate-43.4397-38.49380.002*
Interval-45.2842-40.50040.002*
Duration-71.2042-66.31220.002*
Continuity-25.6471-24.72850.175
Syllable repertoire-120.2983-120.06950.499
Song repertoire-113.5829-113.37060.515
  1. *Denotes traits where the more complex model fit the data significantly better than the simpler model.

Table 6
Brownie results for song traits when birds are divided into early song stability, delayed song stability, and song plasticity.

Rate columns show mean log likelihood. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.022
Song traitOne rateThree ratesp-value
Syllables/song-97.8349-86.3206<0.001*
Song repertoire-100.812-97.76470.014*
Syllable repertoire-107.3206-105.58950.063
  1. *Denotes traits where the more complex model fit the data significantly better than the simpler model.

Table 7
Brownie results for song traits when birds are divided into either song stability (early plus delayed) and song plasticity (Two Rates) or early song stability, delayed song stability, and song plasticity (Three Rates).

Rate columns show mean log likelihood. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.023
Song traitTwo ratesThree ratesp-value
Song repertoire-100.691-97.71480.015*
Syllable repertoire-107.1332-105.55320.075
Syllables/song-86.3125-86.34471
  1. *Denotes traits where the more complex model fit the data significantly better than the simpler model.

Table 8
Brownie results for song traits when birds are divided into shorter learning (early song stability) and longer learning (delayed song stability plus song plasticity).

Rate columns show mean log likelihood. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.024
Song traitOne rateTwo ratesp-value
Song repertoire-100.812-97.99180.018*
Syllable repertoire-107.3206-105.84880.086
  1. *Denotes traits where the more complex model fit the data significantly better than the simpler model.

Table 9
Brownie results for song traits when birds are divided into either shorter learning (early song stability) and longer learning (delayed song stability plus song plasticity) (Two Rates) or early song stability, delayed song stability, and song plasticity (Three Rates).

Rate columns show mean log likelihood. Song traits are sorted from most to least significant.

https://doi.org/10.7554/eLife.44454.025
Song traitTwo ratesThree ratesp-value
Syllable repertoire-105.8156-105.55320.469
Song repertoire-97.9372-97.71480.505
  1. *Denotes traits where the more complex model fit the data significantly better than the simpler model.

Data availability

All data are made available as supplementary information provided with this manuscript, and are also provided at https://github.com/CreanzaLab/SongLearningEvolution (copy archived at https://github.com/elifesciences-publications/SongLearningEvolution).

The following data sets were generated
  1. 1
    GitHub
    1. CM Robinson
    2. KT Snyder
    3. N Creanza
    (2019)
    Dataset S1: Song stability data and references.
  2. 2
    GitHub
    1. CM Robinson
    2. KT Snyder
    3. N Creanza
    (2019)
    Dataset S2: Song feature data and references.

Additional files

Source data 1

Adult song plasticity and stability references.

https://doi.org/10.7554/eLife.44454.038
Source data 2

Song feature and mating system data and references.

https://doi.org/10.7554/eLife.44454.039
Source code 1

Supplemental data and code.

https://doi.org/10.7554/eLife.44454.040
Supplementary file 1

Supplemental tables S1-23 and supplemental table legends.

https://doi.org/10.7554/eLife.44454.041
Transparent reporting form
https://doi.org/10.7554/eLife.44454.042

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)