Liam U. Taylor

NOTE: All data wrangling and analyses fully automated in the RMarkdown file located in the base directory of this repository (finalproject.rmd). Knit that file to reproduce this one.

Organisms with deferred reproduction forgo the production of offspring for the opportunities of development while young (Stearns 1992). In birds with delayed plumage maturation (DPM), this youthful period of reproductive delay is accompanied by a drab “predefinitive” plumage during the periods when older individuals are molting into the colorful “definitive” plumages that characterize increased reproductive effort and success. Research on DPM has largely focused on functional hypotheses for the evolution of predefinitive plumage signals (i.e., hypotheses on why it is “good to be green”). For example, the “crypsis” hypothesis suggests that drab predefinitive plumages help young males avoid predators (Selander 1965; Rohwer 1978) while the “social signalling” hypothesis suggests that predefinitive plumages allow young males to reduce the likelihood or costs of competition by signalling reduced reproductive capabilities (Selander 1965; Lyon and Montgomerie 1986). Studies of these functional hypotheses are limited on two fronts. First, direct benefits to instantaneous survival rate are insufficient for positive selection on the evolution of deferred reproduction, and are thus insufficient for positive selection on the “delayed maturation” part of “delayed plumage maturation” (Taylor in prep.).

Second, very little research has taken a phylogenetic view of DPM evolution. There are only two such studies. The first phylogenetic paper, by Chu (1994), found evidence that the evolution of predefinitive plumages in various shorebird taxa (Aves: Charadriiformes) was the result of the evolution of juvenile plumage in combination with retained partial molts. The result is that these shorebirds retain part of their juvenile plumage in their first breeding season. This study suggests that the evolution of DPM was the result of selection on juvenile plumages and constraints in molt timing, as opposed to selection on predefinitive plumages themselves. The second phylogenetic paper, by Hill (1996), found that the presence of DPM in Haemorhous finches is associated with a reduction in plumage patch size across species. This study recommends that reduced plumage patches are themselves associated with limits in carotenoid availability, and thus that the evolution of DPM is associated with an increase in the resource costs of plumage coloration. In opposition to functional hypotheses for the evolution of predefinitive plumage signals, phylogenetic investigations have supported constraint-based hypotheses for DPM evolution.

Manakins (Aves: Passeriformes: Pipridae) are neotropical lekking birds known for their extravagant male plumages and gymnastic sexual displays (Kirwan, Green, and Barnes 2011). Manakin species also vary widely in both the presence and duration of DPM (Kirwan, Green, and Barnes 2011; Johnson and Wolfe 2017). Most research into manakin predefinitive plumages has focused on the social signalling hypothesis, and ecological and behavioral results from some species indicates that DPM may play a role in cooperative interactions as young males engage in social and sexual behavior with older individuals (e.g., McDonald 1993). To date, there is no phylogenetic investigation of manakin DPM which broadens these taxon-specific studies. Further, the phylogenetic conclusions of Chu (1994) and Hill (1996) cannot be applied to manakins. Unlike molt-constrained shorebirds, manakins with multiple predefinitive plumage stages have complete molts between plumages. Unlike carotenoid-restricted finches, manakins are frugivorous birds in the tropics that have ready access to food resources (Snow 1971).

Thus, we have neither a family-wide view of DPM in manakins nor an evolutionary hypothesis which can underpin such a view. My goal for this project is to correct this gap. For the purposes of this class, I have three explicit sub-goals: (1) Investigate the phylogenetic distribution of the presence and duration of DPM in manakins, including estimating ancestral states; (2) Identify predefinitive plumage homologs across taxa; and (3) Perform comparative tests of the association between social characters and DPM evolution. The first two goals will hopefully open doors to additional analyses about sexual selection and discrete developmental processes in a phylogenetic context, while the third goal will directly address the standing social signalling hypotheses about DPM evolution in manakins.

I used a combination of literature reports, field guides, banding records, and photographs to code the number of DPM stages for each of 35 taxa along with the distinct (non-green) plumage patches at those stages. All taxa had 0, 1, 2, or 3 DPM stages, where a stage indicates an annual basic molt cycle that begins with the partial molt out of juvenal plumage (Wolfe, Johnson, and Terrill 2014). For example: a taxon with a DPM stage state of 2 will be in its definitive plumage only at its 4th breeding season (including its natal season). Unique partial molts that result in characteristic plumage substages (e.g., Chiroxiphia caudata) were collapsed into these broader annual stages. Taxa and citations are as follows:

I used the data from Prum (1994) to code two binary social characters: concentrated lekking and coordinated displays. Taxa with a 0 score for concentrated lekking included both dispersed and non-lekking states, and taxa with a 1 score for coordinated displays included all of simple, coordinated, and cooperative male-male display behaviors. I supplemented some missing taxa with updated scores based on new literature reports, but most missing data is still unavailable.

I coded predefinitive plumages with non-female- and non-juvenile-like plumage patches at each DPM stage. It is necessary to code these patches as broad visual units, rather than standardized morphological components, because patches are metamodules that result from covariation across barbs and barbules within and across both feathers and feather tracts (Prum and Dyck 2003). Although I initially coded patches based on broad coloration method (“Carotenoid”, “Melanin”, “Structural”), the analyses here collapse all coloration methods down to a binary “Present”=1 vs. “Absent”=0 state. Some taxa (e.g., Corapipo manakins at Stage 1) have a fully female-like predefinitive plumage stage which corresponds to a “Absent” score for all patches. Each taxon is associated with an individual plumage datasheet (see .csv files in the Data directory). These sheets are then aligned across taxa with missing data marked correctly as “Absent.”

I use a backbone phylogeny from BirdTree (Jetz et al. 2012). This tree combines publicly available genetic data with standing taxonomic information. In my case, the use of old taxonomic information results in some important erroneous splits (e.g., the placement of Xenopipo, Cryptopipo, and Pseudopipra), but for the purposes of this project I have taken the tree as-is.

I imported a trace of 10,000 subtrees from BirdTree which were generated with all available Pipridae taxa. I used TreeAnnotator to collapse that trace into a single consensus tree for all analyses. I then prune the tree to retain only those tips for which DPM stage and predefinitive plumage characters are available.

The plot below shows the pruned backbone tree. Tip labels indicate DPM stages for each taxon.

I used a maximum-likelihood + information-theoretic approach to DPM stage evolutionary model selection. This multi-state discrete character can be modeled with a 4-state Mk model (Pagel 1994). Prior state frequencies are equal for all models. Because this character happens to have four states within Pipridae, the resulting models resemble genetic models even more closely than usual.

I tested five models:

1. Equal rates unordered

##   0 1 2 3
## 0 - A A A
## 1 A - A A
## 2 A A - A
## 3 A A A -

2. Equal rates ordered

##   0 1 2 3
## 0 - A 0 0
## 1 A - A 0
## 2 0 A - A
## 3 0 0 A -

3. Equal asymmetric rates ordered

##   0 1 2 3
## 0 - A 0 0
## 1 B - A 0
## 2 0 B - A
## 3 0 0 B -

4. Unequal symmetric rates ordered

##   0 1 2 3
## 0 - A 0 0
## 1 A - B 0
## 2 0 B - C
## 3 0 0 C -

5. Unequal asymmetric rates ordered

##   0 1 2 3
## 0 - A 0 0
## 1 B - C 0
## 2 0 D - E
## 3 0 0 F -

Note that some of these models might have implications for the broader connection between DPM evolution and e.g., sexual selection theory. For example, the “Equal Asymmetric Ordered” model would allow for the rate of stage gains to outweigh the rate of stage losses – this ratcheting of a sexual character might be expected from e.g., Fisher’s runaway process (Fisher 1930).

The transition rates which maximize the likelihood of the data were estimated with the fitMk method in phytools (Revell 2012). The table below shows AIC scores for each model. The preferred model, which minimizes AIC, was the Equal Symmetric Ordered Model.

And here are the transition rates which maximize the likelihood of DPM stage data given our chosen model constraints:

## Estimated Q matrix:
##        0      1      2      3
## 0 -0.033  0.033  0.000  0.000
## 1  0.033 -0.066  0.033  0.000
## 2  0.000  0.033 -0.066  0.033
## 3  0.000  0.000  0.033 -0.033

Now that I have established an underlying evolutionary model, I can estimate the probability of DPM stages for internal nodes in the manakin phylogeny. Here I do this ancestral state estimation by simulating stochastic character evolution maps. I use the make.simmap function from phytools to simulate 1000 maps. The summary of those maps is shown below. Taxon tips are assigned to their input data state and internal nodes are colored by the proportion of simulations in a given state. Ticks along each edge indicate simulated transitions across all simulations.

In particular, note two key internal nodes:

1. The parent node of the core manakins (i.e., the clade excluding the tyrant-manakins, Tyranneutes and Neopelma) has a high probability of Stage 1 DPM (probability = 0.98).
2. The parent node of the Ilicurini subclade (i.e., the clade including Masius, Corapipo, Chiroxiphia, and Antilophia) has a majority probability of Stage 2 DPM (probability = 0.6).

I can also summarize the estimated stage transitions as the mean count of transitions across all simulations:

##       0     1     2     3
## 0    NA 1.625 0.000 0.000
## 1 3.356    NA 4.931 0.000
## 2 0.000 1.513    NA 2.061
## 3 0.000 0.000 0.135    NA

Despite gains and losses being governed by the same rate parameter across all states, there are more gains than losses on average (8.617 gains vs. 5.004 losses).

The most frequent transition is from Stage 1 to Stage 2.

Now that I have identified ancestral DPM plumage stages across the tree, I can take a closer look at plumage patch characters within each stage and determine whether there are homologous plumages across taxa. In this case, I can interpret homologous patches as those estimated to be present in the ancestral nodes with DPM stages. First, I can use stochastic maps to simulate the evolution of each patch at each stage as a separate binary character. I again use the make.simmap function to simulate the evolution of each plumage character (100 simulations per character). Each character is simulated under separate All Rates Different models.

I can first view the results across all patches. The figures below show the simulation results for each character. Because these patches are now binary characters, I can view the probability density, across all simulations, for the presence of the character along every edge. Red indicates high probability of presence of a patch, blue indicates low probability.

Note the uncertain edges between the root node and its child nodes. Note also that this model seems ill-equipped to parse the evolutionary history of the Body_S1 plumage patch, which is widely distributed across the tips with many presumed gains and losses. The result is that the state of this patch is highly variable across different stochastic histories. Two further steps might make these simulations more realistic: (1) incorporating information about coloration method (i.e., Absent vs. Carotenoid/Melanin/Structural rather than Absent vs. Present) and (2) developing a model which is informed by all plumage patches simultaneously.

In the meantime, I want to consider the two key internal nodes I derived in the DPM stage ancestral state estimations: the Stage 1 ancestor of the core manakin clade and the Stage 2 ancestor of the Ilicurini subclade. What might those predefinitive plumages have looked like?

Here are the highest probability predefinitive plumage patches for the core manakin ancestor:

And here are the highest probability patches for the Ilicurini ancestor:

In this version of the analysis (i.e., with binary characters and separate models for each character), there are only two patches which pass even a meager 50% threshold at the ancestral DPM stage nodes. This has a reasonable biological interpretation for the core Stage 1 ancestor, which might be thought of as a “female-like” predefinitive plumage without additional patches. On the other hand, there are no female-like Stage 2 plumages in manakins. If I take the highest probability patches for each of these stages as the description of plumage, I get an ancestral core Stage 1 plumage with a crown patch, as well as an ancestral Ilicurini Stage 2 plumage with a face patch (in current taxa, usually a black mask). In general, this binary+separate character analysis does not show strong support any particular predefinitive plumage characters conserved across clades which share DPM stage synapomorphies.

Finally, I want to address the hypothesis that the evolution of DPM is associated with social and sexual behaviors in manakins. In a phylogenetic context, I can extend this hypothesis to predict that there is correlated coevolution between social characters and DPM stages. To test this prediction, I can use Pagel’s phylogenetic correlation method for binary, discrete characters (Pagel 1994). In particular, I will test for two sets of coevolutionary dependencies: (1) coevolution between DPM stages and concentrated lekking and (2) coevolution between DPM stages and coordinated male-male displays. As described above, I have already coded concentrated lekking and cooperative displays as binary characters. I can also convert the discrete multi-state DPM stage character to a set of binary characters. The result is three separate characters for the Presence vs. Absence of Stage 1, Stage 2, and Stage 3, respectively, for each taxon.

The tree below shows all of these recoded characters. From left to right, dots at the tips indicate the Presence (black), Absence (gray) or missing data (white) state for: (1) Stage 1 DPM, (2) Stage 2; (3) Stage 3; (4) Concentrated Lekking; (5) Coordinated Displays. Note the missing data in the social characters for some taxa. In the context of the Pagel discrete coevolution assessment, characters with missing data can be reinterpreted as prior probabilities of Presence or Absence. For taxa with known codings, the prior probability of the assigned state is 1. For taxa with missing data, the prior probability is 0.5 for both states.

I can now test for signs of coevolution between each of the three plumage stages and the two social characters. I use the fitPagel method from phytools to compare the AIC scores for two models for each pair of characters. One model asserts the independent evolution of the two paired characters. The second, more parameter-rich model assumes a dependency, where, for example, the transition from 0->1 in character X depends on the state (0 or 1) of character Y. Support for coevolution is associated with a lower AIC score for the dependent evolution model. Here, I use separate equal rates model for each pair of characters.

The first step is to check for a sign of coevolution in general. I do this by viewing the results of model comparison where dependency runs in both ways. The table below shows the result (X = Stage Character, Y = Social character). Negative values in the dAIC column indicate that the dependent (i.e., coevolutionary) model was preferred to the independent model.

Finally, I can dig a bit deeper and test for dependency directionality in the character pairs which showed some support for coevolution. The table below shows the results. I find some support for the evolutionary association between coordinated display behavior and the presence of 2 or 3 DPM stages across the manakin tree. Recall, however, that there are only two taxa with Stage 3 DPM.

In this project, I initated a phylogenetic view of both deferred reproduction and delayed plumage maturation by coding two sets of characters: (1) the number of annual predefinitive plumage stages and (2) the characteristic patches for individual predefinitive plumages. Across 35 species that span all genera in the family Pipridae, I used stochastic maps to identify an ancestral “Stage 1” DPM phenotype in the ancestor of all core manakins (i.e., excluding Tyranneutes and Neopelma), an ancestral “Stage 2” DPM phenotype for the ancestor of the Ilicurini tribe, and losses in DPM stages on the edges leading to the Cryptopipo, Xenopipo, and Antilophia species. With stochastic maps of individual predefinitive plumage patches, I began the process of reconstructing homologous predefinitive plumage qualities across species. In particular, I identified a fully female-like plumage, or female-like-with-crown plumage, in the Stage 1 ancestor of the core manakins. I also supported the possibility of a mask patch in the Stage 2 ancestor of Ilicurini. In line with the social-signalling hypothesis for individual manakin taxa, coevolutionary model testing supported the hypothesis that Stage 2 DPM evolution is related to the presence of coordinated male-male displays across taxa.

Prum (1997) probed the manakin phylogeny for signals of macroevolution via Fisherian sexual selection processes. The study focused on the rate and skew of evolution of manakin definitive plumage and display characters, finding high levels of diversity and hierarchical evolution biased towards the tips. At several points in this project, I continue to address the possibilities of registering a signal of evolution via sexual selection (in contrast to telenomic natural selection; (Prum 2017)) on a phylogeny.

1. Gains vs. losses in DPM stage evolution. In most manakins, the predefinitive plumage period is a time of sexual and social development. It is during these predefinitive stage(s) that young birds develop the abilities to display or learn the rules of the game in terms of access to sexual displays (McDonald 1993; and see Collis and Borgia 1993; Prum and Razafindratsita 1997). Thus, one possible signal of sexual selection is the acceleration of DPM stages driven by the accelearation of DPM characters. Put most simply, one might expect DPM stage gains to proceed at a faster rate than losses. Although my analysis did count more gains than losses along the Pipridae tree, AIC-based model comparison favored an underlying evolutionary model in which gains and losses were governed by the same parameter. This model decision does not support a novel “ratcheting” process that might attend sexual trait run-aways (Fisher 1930).

2. Losses themselves in DPM stage evolution. It is worth highlighting the three major loss events suggested by my DPM stage analysis: Stage 1->0 along the edges leading to Cryptopipo holochlora and Xenopipo unformis, as well as Stage 2->1 along the edge leading to Antilophia. Losses in Cryptopipo and Xenopipo are noteworthy because those two taxa are the core manakins that are fully sexually monochromatic. Without a non-juvenile-like definitive male plumage, the first partial molt from the juvenile plumages in these species recapitulates the final definitive plumage in both sexes. This result is essentially the opposite transition as the one highlighted for shorebirds in Chu (1994). That study found the evolution of DPM via the evolution of unique juvenile plumage traits with a retained partial molt. Here, the evolution of definitive plumages which resemble juvenile plumages results in no distinction of a first predefinitive plumage. These losses are not a fluke of coding or human observation; strictly speaking, it is possible to evolve DPM that includes predefinitive plumages which are neither juvenile-like nor definitive-like in sexually monochromatic species (e.g., in some large gulls (MacLean 1986)). However, this ontogenetic pattern is absent in the manakin tree. The stage loss in the ancestor of Antilophia is also noteworthy, as Antilophia manakins have novel mating systems that are more territorial and competitive than the wildly cooperative leks of many other Ilicurini species (Marini and Cavalcanti 1992; Gaiotti 2016). Although patterns across the tree did not support a broader association between Stage 2 DPM and concentrated lekking, these phenomena in Antilophia might be a joint and dynamic consequence of underlying demographic patterns. Antilophia bokermanni, for example, is a critically endangered species limited to fewer than 1000 individuals (Gaiotti, Oliveira, and Macedo 2019). This demographic restriction can have surprising consequences for breeding systems, such as extra-pair paternity by young green males in Antilophia bokermanni (Gaiotti 2016).

It is not for nothing that this project considers an ontogenetic process along a phylogeny. The study of the relationship between ontogeny and phylogeny is as old as the concepts themselves. In the 1820’s, von Baer conceived of taxonomic relationships with the conjecture that development evolved by the specialization of adult forms out of general embryonic ones (Abzhanov 2013). With von Baer’s conjecture, taxonomies could be reconstructed based on traits shared early in development. By the 1860’s, Haeckel conceived of phylogenetic relationships per se with the conjecture that development evolves through the speeding of ancestral trajectories with the terminal addition of new ones (Gould 1977). With Haeckel’s conjecture, phylogenetic relationships could be reconstructed based on the ways in which developmental trajectories recapitulated a chain of ancestral forms. Both of these conjectures were assumptions about the process of developmental evolution such that embryos might be used as a tool for taxonomists. Gould (1977) and Alberch et al. (1979) shifted this perspective to focus on the variety of processes of developmental evolution that emerge from changes in the timing or rate of growth. Fink (1982) formalized these approaches in the context of an a priori phylogenetic pattern, demonstrating that questions about the evolution of development were indeed questions about processes that might be answered with phylogenetic patterns, as opposed to clues to those patterns themselves. After nearly 200 years of taxonomic conjecture, the joint study of ontogeny and phylogeny can now investigate the actual relationship between developmental evolution and evolution more generally.

Even still, the formalization of ontogenetic phylogenetics by Gould, Alberch, and Fink is predicated on development as a continuous phenomenon. In this view, development is always akin to somatic growth and always reducible to a set of rates and periods. In contrast, my project here begins to emphasize the process of development as a series of discrete events in the life of, well, an actual bird. First, the DPM stage character is discrete despite being temporal. Because breeding seasons are structured by real environmental fluctuations (e.g., precipitation and photoperiod, even in the tropics (Hau 2001)), which instantiate social institutions (a lone bird cannot a breeding season make!), these birds cannot meaningfully evolve DPM which lasts for, for example, 2.1 breeding seasons. Second, predefinitive plumages are signalling elements only on the level of discrete patches rather than continuous subunits which accumulate towards a definitive plumage (Prum and Dyck 2003). Although this discrete-ness creates a genuine obstacle for an analysis of selection or other modes of evolution (see, e.g., the “threshold” solution of Felsenstein (2012), which makes a discrete character into a continuous one for the sake of analysis), it is also a fundamental part of my empirical system.

SIDEBAR: Because I’m jamming now, I want to highlight a fairly esoteric cross-disciplinary analogy for the pursuit of discrete developmental phylogenetics: the development of historical materialism in political philosophy. In early 19th century, there was an emphasis on an idealist view of history that posited a progression of states as ethical, spiritual, or political ideals. History was quite literally a history of ideas (e.g., Feuerbach; (Gooch 2020)). In his Theses On Feuerbach, Marx famously critiqued this idealist stance and established a historical materialism that focused on not movement of ideas but on the processes of change underlying actual labor power and actual capital (Marx 1972). History became a history of material transitions and transitions of material. In a moment of crisis and revelation, Benjamin disassembled this historical materialism to uncover the remnants of its idealist arc – Marx and his ilk, Benjamin argued, assumed an underlying notion of progress. In On the Concept of History, Benjamin furnished a critique of Marx that paralleled Marx’s critique of Feuerbach. Benjamin argued that “history is the subject of a structure whose site is not homogenous, empty time, but time filled by the presence of the now.”

An archaic view of the role of development in evolutionary history is predicated on genotypic idealism. Genotypes (literally an immaterial unit of potential information, never genetic material itself) evolve along a tree with no need for the intrusion of development. This is an idealist history. A materialist view, furnished by the discoveries of evolutionary developmental biology (Gould 1977; Amundson 2005), sees organisms and evolution as the result of material developmental processes. Now development is involved, and taxa within a phylogeny are material units; but still that materiality is a directional, progressive, and continuous process. A critique of this continuous view would require understanding development not as a progressive, continuous process, but rather a series of events within the life of a material organism – actual ecologies, actual behaviors – development as a series of nows. This would be a view akin to Benjamin. But anyways, what kind of history is evolutionary history? /SIDEBAR

This project reprents only the first step towards a phylogenetic view of manakin DPM and its underlying processes of sexual selection, development, and life history evolution. In particular, I am eager to take two key steps that can productively extend this work in the coming weeks.

1. A new backbone tree. The tree used in this iteration of the project (Jetz et al. 2012) imputed the location of some tips using old taxonomic information. We know some splits in this tree are wrong. For instance, Antilophia is often resolved within a paraphyletic Chiroxiphia and Tyranneutes is often resolved within a paraphyletic Neopelma. For the purposes of this project, I have treated the backbone tree as valid and attempted to interpret results accordingly. A new tree, published just last week, presents an better view of even more manakin taxa using multiple UCE and exon datasets (Leite et al. 2020). All of my analyses are built to run automatically with any tree, so I look forward to replicating these results with the new tree once data are publicly available.

2. Integrating developmental dependencies. In my initial proposal, I assumed it would be straightforward to integrate multiple dependencies into the DPM characters. In particular, for manakins, the evolution of a particular male-like predefinitive plumage patch is hierarchically dependent on the evolution of that patch in the definitive plumage itself. For example, consider again the collapse of DPM in sexually monochromatic species. In my next steps, I hope to explicitly model the joint evolution of DPM stages, predefinitive plumage patches, and definitive plumages. Although I cannot yet see a clear method to do all of this, one intuition I have is to model plumage ontogenies as an explicit sequence, beginning at the natal plumage and ending in a repeating segment for the definitive plumage. Evolution can then shift the characters at each point in the sequence, aligned across all taxa, and some loci may be linked with others.

I look forward to diving into these next steps, and more, as I carry the lessons from this great course into my research!