Songbirds of the Americas show uniform morphological evolution despite heterogeneous diversification

Abstract

Studying the relationship between diversification and functional trait evolution among broadly co‐occurring clades can shed light on interactions between ecology and evolutionary history. However, evidence from many studies is compromised because of their focus on overly broad geographic or narrow phylogenetic scales. We addressed these limitations by studying 46 independent, biogeographically delimited clades of songbirds that dispersed from the Eastern Hemisphere into the Americas and assessed (1) whether diversification has varied through time and/or among clades within this assemblage, (2) the extent of heterogeneity in clade‐specific morphological trait disparity and (3) whether morphological disparity among these clades is consistent with a uniform diversification model. We found equivalent support for constant rates birth–death and density‐dependent speciation processes, with notable outliers having significantly fewer or more species than expected given their age. We also found substantial variation in morphological disparity among these clades, but that variation was broadly consistent with uniform evolutionary rates, despite the existence of diversification outliers. These findings indicate relatively continuous, ongoing morphological diversification, arguing against conceptual models of adaptive radiation in these continental clades. Additionally, they suggest surprisingly consistent diversification among the majority of these clades, despite tremendous variance in colonization history, habitat valences and trophic specializations that exist among continental clades of birds.

Abstract

Diversification trends vary between songbird clades in the Americas, but morphological evolution appears surprisingly uniform.

comparative studies, birds, morphometrics

INTRODUCTION

Understanding how ecological and evolutionary processes interact over broad time scales and how these processes give rise to global biodiversity are central questions of macroevolutionary research. Perhaps, the most widely cited example in which these two processes are intertwined is that of adaptive radiation. This conceptual model's widespread appeal stems from a clear, core prediction: diversification is driven by ecological differentiation in a rapidly diversifying lineage (Gillespie et al., 2020; Schluter, 2000). One common scenario of adaptive radiation begins when a lineage of organisms disperses to an ecosystem that is devoid of competitors but rich in unoccupied ecological niches, a phenomenon called ecological opportunity (Mahler et al., 2010; Simpson, 1953; Stroud & Losos, 2016; Yoder et al., 2010). Under these circumstances, the lineage is expected to diversify to capitalize on unoccupied niches, thus minimizing competition between lineages. From a macroevolutionary standpoint, the species richness of a clade that ultimately results from a single ancestor with ecological opportunity should be correlated with its ecological disparity, defined as the variance among ecologically important functional traits (Gillespie et al., 2020). Many putative adaptive radiations have been documented across the tree of life and across the globe, having occurred in many clades in the Hawai'ian Archipelago (Baldwin et al., 1990; Gillespie et al., 2018; Tokita et al., 2017), in the cichlids of Africa's rift lakes (Seehausen, 2006), the Malagasy vangas (Reddy et al., 2012) and the Galápagos finches (Lack, 1983; Lamichhaney et al., 2015). In all these examples, there is a striking connection between diversification and functional trait evolution.

Macroevolutionary studies have tested the generality of this conceptual model by exploring whether species richness and functional trait disparity are correlated at scales larger than the island or island‐like systems in which many adaptive radiations have been documented. Most studies of this kind have used proxies for ecology in the form of morphological characters that capture functional, ecologically relevant information about an organism's size and shape, and most that have approached this question at global scales have done so by examining the morphological diversity and species richness within clades, such as tribes, families or orders, around the world (Derryberry et al., 2011; Gonzalez‐Voyer et al., 2011; Harmon et al., 2010; Ricklefs, 2006, 2012; Venditti et al., 2011; Rabosky et al., 2013; Oliveira et al., 2016; Kennedy et al., 2018; Crouch & Ricklefs, 2019). Many of these studies have suggested a relatively continuous accumulation of disparity over broad time scales, with little evidence of punctuated bursts of evolution (Harmon et al., 2010). However, at such large geographic scales, the possibility of detecting a role for ecological opportunity may be limited given that ecological processes likely manifest at smaller spatial scales, such as within biogeographic regions or specific biomes. Studies may reveal contrasting patterns when focused on biogeographic or phylogenetic scales where the effects of ecological interactions are more likely to be detected (Cooney & Thomas, 2021; Graham et al., 2018). The broad‐scale correlation of diversification with the evolution of morphological disparity has only occasionally been thoroughly tested at more ecologically relevant biogeographic and phylogenetic scales (Claramunt, 2010; Derryberry et al., 2011; Rowsey et al., 2018, 2019, 2020), and the question of how the evolution of functional diversity may facilitate diversification in continental contexts remains unclear.

One way to address this question is to focus on a set of biogeographically delimited clades that have a shared continental distribution. The assemblage of oscine passerines, or songbirds (Passeriformes: Passeri), found in the Americas comprises ~1300 species whose ranges span the entirety of North and South America, from Alaska to Tierra del Fuego, and nearly all adjacent islands. Intensive phylogenetic studies spanning the global diversity of oscines have provided a robust phylogenetic backbone with which comparative studies of trait evolution and diversification may be conducted. Collectively, this work has shown that the Western Hemisphere oscine assemblage has accumulated following numerous dispersal events from clades in the Eastern Hemisphere or by evolving in situ from ancestors that dispersed to this region (Barker et al., 2004; Moyle et al., 2016; Oliveros et al., 2019; Selvatti et al., 2015). Furthermore, all species of Western Hemisphere oscines are collectively represented by hundreds of thousands of natural history specimens housed in collections across the globe, from which ecologically relevant morphological trait data can be readily measured. We studied this assemblage by combining a single time‐scaled phylogenetic framework containing the majority of oscine species found in the Americas with a novel morphometric dataset representing an even greater proportion of these species.

This assemblage of oscine clades in the Americas allowed us to address four questions pertaining to the processes of diversification and morphological evolution at continental scales. First, we identified all independent immigrant oscine clades in the Western Hemisphere, for which we estimated the crown and stem ages and documented their extant species richness. Second, with knowledge of species richness and phylogenetically derived age estimates for each clade, we fit alternative diversification scenarios and tested for uniformity in diversification rates. Third, we quantified the magnitude of morphological disparity exhibited by these clades, quantified as the occupation of a multidimensional functional trait space and assessed the relative clade‐specific occupation of the total realized morphospace of the oscine assemblage in the Americas. Finally, we compared these disparities to null expectations under a uniform diversification process and tested for significant deviations. This study represents one of the first geographically delimited assessments of faunal assembly across multiple independent colonization events that integrate both diversification and functional trait evolution.

METHODS

Identifying Western Hemisphere immigrant oscine clades

To delimit independent Western Hemisphere immigrant oscine clades, we downloaded the best‐supported, most thoroughly sampled phylogenetic trees containing oscine species found in the Americas and identified nodes at which a dispersal event could be inferred from the Eastern Hemisphere into the Western Hemisphere (Barker, 2017; Barker et al., 2015; Beckman & Witt, 2015; Bonaccorso & Townsend Peterson, 2007; Cai et al., 2019; Fregin et al., 2012; Fuchs et al., 2019; Jetz et al., 2012; Johansson et al., 2013; Jønsson et al., 2016; Lee et al., 2003; Lovette et al., 2012; Nylander et al., 2008; Oliveros et al., 2019; Päckert et al., 2003, 2010; Parchman et al., 2006, 2016; Pasquet et al., 2014; Price et al., 2014; Sheldon et al., 2005; Slager et al., 2014; Spellman et al., 2008; Van Els & Norambuena, 2018; Voelker et al., 2009; Voelker & Klicka, 2008; Zuccon et al., 2012). Any nested subclades in the Eastern Hemisphere inferred to be the result of back‐colonization from the Western Hemisphere were excluded from the study. In three clades (Loxia, Corvus and Bombycilloidea), the dispersal events into the Americas could not be clearly inferred from their source phylogenies and, thus, were explicitly clarified through biogeographic analyses (Appendix S1). Turdus and Corvus were unique clades in that these genera have apparently undergone repeated transoceanic dispersal and subsequent diversification (Jønsson et al., 2016; Nylander et al., 2008; Voelker et al., 2009), although this result has been challenged for Turdus with equivocal support (Batista et al., 2020). Here, we recognized three separate Western Hemisphere clades in Corvus and four in Turdus resulting from multiple, independent dispersal and subsequent diversification events.

Following this delimitation strategy, we identified 46 independent and non‐nested clades that have dispersed into the Americas, comprising a total of 1284 species. We included 28 of these clades in the subsequent study of morphological disparity because they met the minimal sample size criteria imposed by our morphometric analyses (n > 5 specimens, see Results). Three genera, Regulus, Loxia and Sitta, possess two or more North American species but comprise several independent, monotypic immigrant clades represented by a single species (R. satrapa and R. calendula; L. sinesciurus; S. carolinensis and S. canadensis). The genus Sitta also includes a monophyletic species pair, S. pygmaea and S. pusilla, that was included in subsequent diversification and morphometric analyses. Ambiguity in the ancestral biogeographic reconstruction of Bombycilloidea prompted its treatment as three independent dispersal events (Dulus dominicus, Bombycilla cedrorum and Ptiliogonatidae), of which only Ptiliogonatidae contained >2 North American species and was included in our morphometric analyses (Figure S1). Other clades from the Eastern Hemisphere that are represented in the Americas by a single or a few Holarctic species, presumably by longitudinal range expansion (i.e. Eremophila, Pinicola and Acanthis), were also not included. The names of these species follow the most up‐to‐date IOC checklist at the time of writing (Gill & Donsker, 2019). These oscine species, their clades and the published works from which their relationships and biogeography were inferred are provided in the Dryad data for this study (Dryad data: doi:10.5061/dryad.0gb5mkm3z).

Analytical approach without a completely sampled phylogeny

An ideal study answering the questions posed above requires a completely sampled, fossil‐ and time‐calibrated phylogenetic tree containing all oscine species in the Western Hemisphere. From this tree, both clade‐ and branch‐specific rates of diversification and rates of trait evolution could be estimated, and these rates would allow for between‐clade analyses of the presence and or strength of their relationship. However, at this time no such tree exists. In this study, we overcame the limitation posed by the absence of such a tree using a distribution of trees from a time‐scaled phylogeny containing the majority of Western Hemisphere oscine species (Jetz et al., 2012). We specifically limited our study to trees whose branch lengths are derived from molecular data (Stage I trees in Jetz et al.'s nomenclature) because performing analyses of diversification and trait evolution on trees containing arbitrarily placed tips (Stage II trees in Jetz et al.'s nomenclature) has shown to result in statistical artefacts like erroneously high rates and extreme between‐tree variance (Rabosky, 2015). Consequently, we only used Stage I trees to estimate clade stem and crown ages and used whole clade summary statistics to represent morphological disparity rather than estimating clade‐specific rates of morphological evolution. Because we cannot reliably estimate clade‐specific rates of diversification or morphological evolution without completely sampled trees, we focused on a hypothesis testing framework that models total disparity (as estimated by hyperdimensional morphospace occupancy, defined below) as a function of clade age and species richness, with either uniform underlying rates of diversification and morphological evolution, or multiple such rates.

We note that recent work has cautioned against drawing biological inferences from estimates of diversification based on age and richness for single clades (the so‐called ARR estimator of Magallón & Sanderson, 2001) without external validation (Rabosky & Benson, 2021). We did not use this single‐clade approach in this study. Instead, we used a multi‐clade approach similar to those of Bokma (2003, 2010) and Ricklefs (2004) that incorporates clade age and richness estimates from many separate clades to jointly infer rates of diversification and trait evolution, an alternative suggested in Rabosky and Benson (2021).

Crown and stem age estimates

We calculated a crown and stem age for each of the Western Hemisphere clades from a time‐scaled distribution of phylogenies containing all birds (Jetz et al., 2012). We downloaded a distribution of 1000 trees from birdtree.org containing all species represented with sequence data (‘Stage I’) on the Hackett et al. (2008) backbone (‘Hackett’), and using custom scripts in R (R Core Team, 2013), we substituted the names used in these trees for their respective IOC names. Then, for each tree, we calculated the crown and stem age of each clade by finding the most recent common ancestor of each clade's species, recording the height of this node (crown age) and recording the height of the ancestral node uniting the Western Hemisphere clade to its Eastern Hemisphere sister clade (stem age). The resulting distribution of crown and stem ages derived from this distribution of trees was averaged to obtain a mean crown and stem age for each clade. We calculated stem ages for every immigrant clade (except Loxia sinesciurus, which was excluded from further analyses because it was not present in source trees), and we calculated crown ages for each clade containing at least two species. We modified this general procedure for nine clades whose topology in the distribution of trees was not concordant with the clade‐specific tree from which they were identified (Appendix S1).

Modelling among‐clade diversification

We fit models describing the relationship between stem age and natural‐log species richness among the 46 clades using maximum likelihood. We used stem ages instead of crown ages when fitting these models to avoid excluding the 18 monotypic clades as biologically meaningful observations and to maximize our sample size for estimation of parameter values. First, we fit a pure‐birth model with a maximum‐likelihood estimate of speciation only (speciation rate: λ). Second, we fit a birth–death model of diversification to these data with maximum‐likelihood estimates of r (net diversification rate: λ – μ, where λ is the speciation rate and μ is the extinction rate) and ε (extinction fraction: μ/λ). To model alternative scenarios of diversification, we fit two additional models that allowed for variable speciation rates through time with constant rates of extinction or no extinction (Rabosky, 2009); specifically, these models allowed for exponential decreases in speciation rates towards the present. We calculated the likelihoods of the data using the dyule, dbd and dbdTime functions in the R package ape (Paradis et al., 2004) and optimized each model over the relevant parameters using the optim function of base R. We assessed which of these models best fit our clade‐specific data by comparing AIC values. To evaluate the impact of singletons, as well as potential overestimation of dispersal times associated with stem ages, we also fit a uniform birth–death model to the crown age estimates.

Morphological data collection

To quantify the morphological diversity of each clade containing at least two species, we measured morphological characters from museum study skins that have been shown in songbirds to predict foraging stratum, foraging behaviour and diet in several different ecological communities around the world (Miles et al., 1987; Miles & Ricklefs, 1984; Pigot et al., 2016, 2020; Price et al., 2014; Ricklefs & Travis, 1980; Rosamond et al., 2020). Standard linear measurements were taken from specimens at the Bell Museum (MMNH), Field Museum of Natural History (FMNH), Louisiana State University Museum of Natural Science (LSUMZ) and American Museum of Natural History (AMNH). Ten morphological characters were measured following Baldwin et al. (1931): length of exposed culmen (LEC), length of bill at nares (LBN), width of bill at nares (WBN), depth of bill at nares (DBN), width of bill at base (WBB), depth of bill at base (DBB), length of tarsometatarsus, length of hallux, length of closed wing and length of tail. Of these 10, all but the bill measurements from the nares and hallux length have been widely used in many comparative studies of passerine morphology (Ricklefs, 2017 and works cited therein). Morphological data were collected from, ideally, three male specimens for every species from each clade that was represented in the collections visited for this study except the largest (Emberizoidea). Within Emberizoidea, the largest Western Hemisphere oscine clade, we obtained an unbiased sample of species by randomizing the list of all 826 species and measuring the first 80%. We consider generic designations to be a reasonable proxy for ecologically related morphological variation, and our sampling within Emberizoidea includes 191 of 200 genera (>95%).

Quantifying clade‐specific trait disparity

To summarize the total morphological variance in this dataset in fewer than 10 dimensions and make calculating morphological disparity possible for the clades with the smallest number of specimens (n = 5), we performed a principal component analysis (PCA) on the covariance matrix of ln‐transformed measurements. We performed a standard PCA as opposed to a phylogenetic PCA (pPCA; Revell, 2009, 2012) to establish strictly orthogonal axes in a Euclidean trait space and, therefore, avoid miscalculating disparity in this space (Polly et al., 2013). Trait disparity is often calculated as the variance among trait axes (Claramunt, 2010) or as the area or volume of convex hulls in low‐dimensional space (Cornwell et al., 2006; Crouch & Ricklefs, 2019; Deline et al., 2018; Drake & Klingenberg, 2010). Instead, we quantified the morphological disparity of each clade in this study by calculating n‐dimensional hypervolumes using the R package hypervolume (Blonder et al., 2014). This kernel density estimation approach constructs geometrically complex estimates of the n‐dimensional space a sample occupies, which we applied to calculate numerical estimates of clade‐specific disparity using the scores of each clade's specimens along as many principal component axes that could be included given our clade sizes and sampling (see Results). We compared these kernel density estimates of disparity to variance and convex hull‐based measures and found them to be largely concordant (Appendix S1).

Modelling trait disparity as a function of clade age and species richness

As a test of the hypothesis that morphological traits evolve uniformly among clades, we fit a series of phylogenetic generalized least squares (PGLS) models to our disparity, clade age and species richness data in the R package caper (Orme et al., 2013). We generated a clade‐level phylogeny for these regressions by first obtaining a maximum clade credibility (MCC) tree in TreeAnnotator v2.5.2 (Bouckaert et al., 2014) using the distribution of trees used to calculate clade age. We then randomly selected one species to represent each individual clade and pruned all other tips in the tree, which left a representative tip for each of the clades and retained the clade‐level backbone relationships within oscines as a whole. Prior to fitting these models, we ln‐transformed species richness, and hypervolume disparities were transformed by taking the nth root. We transformed disparity in this way because, under Brownian motion, linear disparity of a trait can be quantified as the product of the trait's evolutionary rate and the total tree length for the clade, or σ²t (Felsenstein, 2004; Harmon, 2018). For n traits, this should scale approximately as σ²ⁿtⁿ, and accounting for this relationship transforms disparity such that it increases approximately linearly through time. Given this theory, we expected that a model with the natural log of species richness, crown age and their interaction as predictor variables would provide an adequate fit to our kernel density hypervolume data. We assessed this expectation by simulation studies parameterized on the data (see below). We assessed the goodness‐of‐fit for each model by comparing the AIC values between models, and the predictive strength of the model was reported as its coefficient of determination (R²). We refit the best‐fitting PGLS model to each tree in the original distribution of 1000 Stage I trees as well to quantify the impact of phylogenetic uncertainty on the estimated effects from this model.

Generation of null expectations of diversification by simulation

Following previous approaches to understanding the predictors of morphological diversity among clades (Bokma, 2010; Purvis, 2004; Ricklefs, 2004), we performed simulation analyses to compare our observed clade disparity data and model fits to expectations under three scenarios in which trait evolution follows a uniform Brownian process among clades and (1) diversification follows a constant birth–death process, (2) speciation decelerates through time, or (3) diversification occurs uniformly except for the outlier clades identified in our diversification analysis, which were assigned higher or lower rates, as appropriate. We estimated evolutionary rate parameters (σ²) that best‐predicted empirical clade‐specific disparity using approximate Bayesian computation (ABC) in the R package EasyABC (Beaumont, 2010; Jabot et al., 2013). We simulated a tree for each of the 26 clades using the 26 empirical crown ages and our maximum‐likelihood estimates of diversification rate (or speciation rate and decay constant in the case of declining diversification, or diversification rate plus clade‐specific rates in the case of outlier clade simulations) using the function sim.bd.age in the R package TreeSim (Stadler, 2010). We next simulated trait data for each phylogeny given an assigned rate of σ² with the function fastBM in phytools (Revell, 2012), scaled according to the proportion of variance explained by each axis in the real trait data, calculated n‐dimensional convex hulls of these data with the function convhulln in geometry (Barber et al., 2019) and then optimized the σ² parameter by minimizing the distances between the simulated points and the real clade data given their ages, species richness and convex hull disparity. Convex hulls had to be used in these simulations due to the intense computational burden of calculating hypervolume disparities for large simulated clades; however, we note that convex hull hypervolumes and kernel density hypervolumes are strongly correlated (see Results).

Using these optimized parameters, we resimulated disparity data under each diversification scenario 1000 times to generate a null distribution. We fit a multiple regression to these data with transformed disparity as the response variable and with ln(tip number), clade age and their interaction as predictors. Model coefficients and R² values were recorded, to which we compared the model fit for the empirical clade data. Additionally, we compared realized disparity for each clade to its corresponding null distribution to identify clades with extraordinary disparity, and we calculated the overall fit of each model to the empirical data by the distance between the empirical and null mean vectors.

RESULTS

Diversification among Western Hemisphere oscine clades

After surveying the extensive oscine phylogenetics literature, we identified 46 oscine clades that independently colonized the Western Hemisphere over tens of millions of years and that range in species richness from a single species to over 800. The crown and stem ages estimated for these clades and their diversity are presented in Table 1. Following maximum‐likelihood optimization and comparing AIC values for four diversification models, a decelerating speciation model with no extinction was best supported by AIC (Table 2, AIC = 309.933). However, both the constant rates birth–death model and the decelerating speciation with constant extinction models were comparably well‐supported and cannot be rejected (ΔAIC = 0.038 and ΔAIC = 1.681, respectively), and the pure‐birth model is only marginally worse (ΔAIC = 3.921). We fit these same models on each tree in the original distribution of 1000 trees and found the parameters to differ only modestly and non‐significantly between our mean age results and those of the full distribution, showing that these results are robust to phylogenetic uncertainty (Figure S3). In all of these scenarios, the majority of Western Hemisphere oscine clades fall within the 95% confidence intervals on clade diversity inferred for each model (Figure 1). This outcome is consistent between confidence intervals inferred from both stem (Figure 1A) and crown ages (Figure 1B) using maximum‐likelihood estimates of diversification rates, which is not surprising given the similarity in net diversification rate and extinction fractions inferred using these different ages (stem ages: r = 0.125 [95% CI: 0.069–0.180], ε = 0.637 [95% CI: 0.000–0.899]; crown ages: r = 0.131 [95% CI: 0.047–0.196], ε = 0.846 [95% CI: 0.527–0.974]). Furthermore, this outcome implies that the 18 singleton clades that were excluded from the crown age analysis had a relatively small impact on the diversification rates inferred for the remaining clades. A few clades are identified as outliers by these analyses, the majority of which are singleton species present in the Americas that are much less diverse than expected given our estimate of the time since their dispersal (i.e. assuming dispersal occurred at the divergence from their sister taxa, likely an overestimate). Besides the singleton clades, three others lie outside the expectations of these models. On the one hand, the massively diverse clade Emberizoidea (which comprises 16 families; Barker et al., 2015) and the largest of several Western Hemisphere Turdus (large‐bodied thrushes) clades are approximately 13.4 times and 3.4 times more species‐rich than expected given their age, respectively (although the latter only in the stem age analyses). On the other hand, Ptiliogonatidae (the highland silky‐flycatchers) is 34.5 times less diverse than expected (Figure 1).

TABLE 1

Age, species richness and trait disparity of Western Hemisphere oscine clades

Clade	Crown age (myr)	Stem age (myr)	Species richness	Morphospace disparity	% total morphospace
Emberizoidea	23.729	27.134	826	1.6419	48.313
Troglodytidae–Polioptilidae	24.151	28.334	107	0.7957	23.413
WH Vireonidae	17.818	19.462	53	0.2276	6.697
WH Jays	15.911	17.623	39	0.1902	5.598
Mimidae	15.674	22.678	34	0.3315	9.753
Euphoniinae	13.849	23.428	32	0.1179	3.470
WH Swallows	14.130	16.446	30	0.4694	13.811
WH Turdus (large)	6.434	6.934	24	0.0748	2.200
WH Spinus	5.728	7.130	18	0.0744	2.189
Catharus and allies	15.989	17.966	18	0.1727	5.082
WH Turdus (small)	5.716	6.621	15	0.0180	0.530
WH Anthus	11.588	14.490	11	0.0864	2.543
Myadestes	9.290	9.834	8	0.0213	0.627
WH Corvus (large)	7.954	9.732	8	0.0503	1.479
WH Poecile	7.887	10.349	7	0.0369	1.087
Baeolophus	9.803	13.026	5	0.0348	1.023
Ptiliogonatidae	28.229	33.851	4	0.2149	6.323
Sialia	5.430	20.304	3	0.0117	0.343
Haemorhous	11.480	13.371	3	0.0041	0.121
WH Cinclus	–	11.658	3	0.0489	1.440
WH Leucosticte	0.255	1.660	3	0.0011	0.033
WH Petrochelidon	3.142	7.336	3	0.0309	0.910
WH Turdus (Caribbean)	2.998	3.904	2	0.0070	0.205
WH Loxia	1.076	1.826	2	0.0034	0.100
WH Sitta	7.352	10.137	2	0.0092	0.269
WH Coccothraustes	–	10.505	2	0.0137	0.404
WH Corvus (small)	0.465	2.548	2	0.0200	0.588
WH Pica	0.333	2.413	2	0.0016	0.046
Dulus dominicus	–	37.190	1	–	–
Regulus satrapa	–	33.593	1	–	–
Peucedramus taeniatus	–	30.582	1	–	–
Auriparus flaviceps	–	30.130	1	–	–
Donacobius atricapilla	–	20.390	1	–	–
Sitta carolinensis	–	15.222	1	–	–
Psaltriparus minimus	–	14.518	1	–	–
Regulus calendula	–	12.694	1	–	–
Perisoreus canadensis	–	11.375	1	–	–
Bombycilla cedrorum	–	11.082	1	–	–
Chamaea fasciata	–	7.871	1	–	–
Turdus plebejus	–	7.362	1	–	–
Sitta canadensis	–	5.193	1	–	–
Nucifraga columbiana	–	4.866	1	–	–
Lanius ludovicianus	–	1.986	1	–	–
Corvus cryptoleucus	–	1.928	1	–	–
Certhia americana	–	1.806	1	–	–
Loxia sinesciurus	–	–	1	–	–

Note: Clade ages were estimated from a distribution of Jetz et al. phylogenies, and species richness was determined from the most recent IOC checklist at the time of data collection. Morphospace disparity was calculated as kernel density estimate hypervolume on the first four principal components, and the per cent of total morphospace was calculated by assessing what proportion of the entire oscine assemblage is occupied by a particular clade.

Abbreviations: WH (Western Hemisphere), myr (millions of years).

TABLE 2

Comparing diversification models fit to diversity and stem age data from Western Hemisphere oscine clades

Model	k	lnL	AIC	ΔAIC	r	ε	K
Decelerating Speciation	2	−153.966	309.933	0	0.249 [0.172–0.372]	NA	0.035 [0.000–0.080]
Birth–death	2	−152.986	309.971	0.038	0.125 [0.069–0.180]	0.637 [0.000–0.899]	NA
Decelerating Speciation + Constant Extinction	3	−152.807	311.614	1.681	0.188 [0.071–0.388]	0.365 [0.000–0.900]	0.013 [0.000–0.086]
Pure‐birth	1	−155.927	313.854	3.921	0.175 [0.152–0.204]	NA	NA

Note: Models are ordered by increasing ΔAIC values. Abbreviations and symbols: lnL (model natural‐log likelihood), r (net diversification rate, λ‐ μ), ε (extinction fraction) and K (speciation decay constant). Parameter 95% confident intervals are in brackets. Note that r = λ in models without extinction.

FIGURE 1

Relationship of clade age and species richness among Western Hemisphere oscine clades under a uniform birth–death model. Black points represent the individual clades used in this study, the red line represents the best‐fit model for these data, and the grey lines mark the 95% prediction interval given the estimated diversification parameters and either clade stem ages (a) or crown ages (b). Numbers adjacent to the three points correspond to (1) Emberizoidea, (2) Ptiliogonatidae and (3) the large Turdus clade.

Clade‐specific patterns of morphospace occupation

We measured a total of 3051 study skin specimens for this study representing 1082 species of oscines from the Western Hemisphere, nearly 85% of the total, with an average of 2.82 specimens per species. To analyse occupation in a more tractable morphospace, we performed a principal components analysis (PCA) on the ln‐transformed morphometric data. Given that our smallest clade sample size was n = 5 and that hypervolumes in fewer than n‐1 dimensions are undefined, we retained PC1‐4 for downstream analyses (Table S1). These four axes collectively account for >95% of the total morphological variance among these species. Concordant with other studies (Pigot et al., 2016, 2020; Ricklefs, 2005, 2012), PC1 largely captured variation in body size and was significantly related to ln‐transformed body mass (R² = 0.853, β = 1.171, Figure S2). PC2‐4 captured shape variation among oscine species in the Americas, with PC2 contrasting stout‐billed short‐bodied birds against thin‐ and long‐billed birds with large hindlimbs, PC3 contrasting swallows on the negative end of this axis against birds with short‐winged short‐tailed birds with long or deep bills, and PC4 again contrasting swallows against many tanager genera with long hindlimbs but short bills.

The morphospace occupied by the entire Western Hemisphere oscine assemblage was not evenly apportioned among the 28 clades. This did not depend upon the measure of disparity. Kernel density hypervolume estimates of disparity were directly comparable to two other widely used measures of disparity, including the proper variance (ρ = 0.933, R² = 0.893) and convex hull volume (ρ = 0.937, R² = 0.862). Likewise, this finding was not dependent upon character inclusion. To assess the impact of including the relatively large number of bill measurements in our data, we compared kernel density estimates of disparity calculated from a rarified dataset without the three bill measurements relative to the nares and found they were strongly correlated to volumes calculated on the complete data set (ρ = 0.979, R² = 0.966). We therefore only present the results from the kernel density estimated hypervolumes in the full 10‐trait dataset. The 4‐dimensional hypervolumes calculated for each clade and the percentage of the total morphospace a given clade occupied are also given in Table 1. Disparity spanned three orders of magnitude from 0.001 at the smallest, for the Western Hemisphere Leucosticte, up to 1.642 for Emberizoidea, and encompassed a range of 0.03% of the total oscine morphospace to greater than 48%. Although 26 of the clade‐specific hypervolume disparities ranged in magnitude between 0.001 and 0.5, the disparities of Emberizoidea (1.642) and Troglodytidae–Polioptilidae (0.796), two of the oldest clades and most species‐rich clades, were much greater in magnitude.

Visualizing the distribution of each clade within the Western Hemisphere oscine assemblage's morphospace provided further insights regarding its morphological diversity (Figure 2). Emberizoidea occupied practically the entire range of PC1 scores (i.e. body sizes) for the entire American oscine avifauna, and, except for a few regions along the shape axes where swallows and gnatwrens are clustered, occupied the entirety of oscine shape diversity. No other clade approached this extent of diversity and occupation of morphospace. Troglodytidae–Polioptilidae showed a comparable range on PC1 but lacked very large‐bodied species. Shape diversity of this clade was restricted to the positive ends of PC2 and PC3. The next most‐disparate clade, the Western Hemisphere swallow genera, had a more restricted range on PC1 and occupied a more densely clustered region of shape space on the negative ends of PC3 and PC4 opposite that of Troglodytidae–Polioptilidae. Each of the subsequent clades had smaller ranges of PC1 scores and occupied a smaller, generally distinct, region of morphospace (Figure 2).

FIGURE 2

Clade‐specific occupation of morphospace among Western Hemisphere oscine clades. From left to right in each row, for the five clades with the greatest total disparity, these figures show the density of PC1 scores, a plot of individual specimen scores on PC2 and PC3, and a plot of individual specimen scores on PC3 and PC4. In all panels, grey distributions and points represent all specimens sampled in the Western Hemisphere oscine assemblage and the coloured distributions and points correspond to that particular clade.

Clade‐specific disparity as a function of clade age and species richness

As expected from simulation studies under constant rates of Brownian motion (see below), a multiple regression with natural‐log species richness, crown age and their interaction as predictor variables was the PGLS model with the greatest AIC support and explanatory power for observed morphological disparity among the models we fit to these data (R² = 0.885, Table 3). A similar model excluding an interaction term for species richness and crown age fit the data comparably well (ΔAIC = 1.64, R² = 0.873). Model coefficients estimated from this PGLS fit with a MCC tree were modestly and not significantly different from a distribution of coefficient estimates from the original distribution of trees (Figure S4), indicating little effect of tree uncertainty in these estimates. A 3‐dimensional plot of predicted values from the model with the interaction term shows a relatively flat, planar relationship between ln(species richness), crown age and transformed disparity (Figure 3; see Figure S5 for an animated version of Figure 3A). We investigated the influence of diversification outliers (Emberizoidea, the large Turdus clade, and Ptiliogonatidae) on the shape and strength of this relationship by fitting the same set of models after removing diversification outliers from the dataset. As with the complete dataset, a model including ln(species richness), crown age and their interactions as predictors was the best‐supported model (R² = 0.855), but the same model without an interaction fits these data comparably well (ΔAIC = 1.266, R² = 0.834).

TABLE 3

Modelling trait disparity as a function of clade age and diversity

Model	F	R  2	p	AIC	ΔAIC
All clades:
Hyp ~ ln(Spp) * Crown	65.07	0.885	4.24E‐11	−51.279	–
Hyp ~ ln(Spp) + Crown	87.21	0.873	1.83E‐11	−49.638	1.640
Hyp ~ ln(Spp)	121.80	0.829	6.94E‐11	−42.646	8.633
Hyp ~ Crown	56.86	0.679	1.40E‐07	−26.342	24.937
Outliers removed:
Hyp ~ ln(Spp) * Crown	44.26	0.855	9.13E‐9	−42.051	–
Hyp ~ ln(Spp) + Crown	56.44	0.834	5.96E‐9	−40.785	1.266
Hyp ~ ln(Spp)	93.58	0.808	3.37E‐9	−38.971	3.080
Hyp ~ Crown	58.67	0.724	1.64E‐7	−30.559	11.492

Note: Rows in bold indicate PGLS models that were indistinguishable by comparison of AIC values (ΔAIC < 4). Columns contain the F‐statistic (F), the coefficient of determination (R²), the model p‐value (p), the AIC score for that model (AIC) and the ΔAIC when comparing that model to the best‐fitting model. The top table presents results when fitting the model to all clades, and the bottom table presents results when fitting the models excluding diversification outliers.

Abbreviations: Hyp (transformed hypervolume disparity), Spp (species richness) and Crown (crown age).

FIGURE 3

Visualizing relationships between crown age, species richness and trait disparity in real and simulated clades. In both plots, black points represent data from empirical Western Hemisphere oscine clades. Figure 1b essentially forms the ‘floor’ of these plots, and the vertical axis represents transformed measures of trait disparity. The left plot (a) shows, in red, the regression surface calculated from the PGLS model that best fits the observed kernel density hypervolume (KDH) disparities, whereas the right plot (b) shows, in blue, the regression surface calculated from the same model fit to the pooled distribution of simulated convex hull (CH) disparities from the uniform birth–death simulation scenario.

Comparison of simulated diversification patterns

Targeting observed clade diversities and trait disparities, we used approximate Bayesian computation (ABC) to estimate rates of functional trait evolution in a uniform Brownian motion evolution model across clades. This analysis indicated that optimized rates of trait evolution (σ²) across clades were largely consistent across three diversification scenarios: uniform, decelerating and outlier diversification, with rates of 0.051, 0.049 and 0.059, respectively. All three simulation scenarios yielded distributions of model parameters that were similar to the model fit for the observed oscine clade data (Figure S6). Each model fits the simulated data such that, given complete knowledge of species richness and clade age, variance in simulated trait disparity could be predicted with high accuracy (R² > 0.970; Table S2). One parameter differed significantly between the best‐fitting model for the empirical data and for the simulated distributions across all three scenarios. The empirical coefficient for ln(species richness) was lower in magnitude than the coefficient for the simulated data in every case, meaning that species richness has a weaker effect on morphological disparity in empirical clades than predicted under Brownian motion. The outlier birth–death diversification scenario also differed from the model fit to the real data in its interaction coefficient parameter, which was significantly lower for the observed data. Unsurprisingly, these observations suggest that pure Brownian motion, while providing a surprisingly good fit to empirical data, is likely not the true process underlying them.

On a clade‐by‐clade level, clade‐specific disparity was remarkably similar between the empirical and simulated data (Table S3; Figure 3). The birth–death model provided the best fit, followed closely by the declining diversification model, but, in either case, only eight of the 26 clades had disparities that were significantly different from expectation. By contrast, 17 of the 26 clades had significantly different disparities in the outlier birth–death simulation, showing a substantially poorer fit to the empirical data. Three small clades had greater disparity than expected in all three models (small Corvus, Loxia and Leucosticte), and two had less disparity than expected under all three models (Ptiliogonatidae and Haemorhous), but otherwise the disparity of each clade was within expectations for at least one of the simulation scenarios. Interestingly, the two best‐fitting models (birth death and decelerating speciation) each failed to predict the disparity of one of the two oldest clades of oscines, with the former greatly underestimating disparity of Emberizoidea and the latter overestimating the disparity of wrens and gnatcatchers.

DISCUSSION

In its evaluation of the tempo of avian diversification, our study had four explicit goals. First, we identified independent immigrant oscine clades in the Western Hemisphere and calculated species richness and colonization timing for each. Second, we examined diversification in these replicate continental clades to test whether it was uniform, or whether significant outliers could be identified. Third, we quantified morphological disparity within these clades and evaluated their relative occupation of the total morphospace realized by the assemblage. Lastly, we sought to determine whether morphological disparity among this ecologically diverse assemblage was a straightforward function of species richness and clade age, as expected under a simple homogeneous model of continuous trait evolution. We did find significant instances of heterogeneity, as outlined below. Despite these exceptions, overall patterns of diversity and disparity in Western Hemisphere oscine clades were well‐approximated by a simple uniform process model including only speciation and extinction, a rate of trait evolution and proportionality constants for rates among a few composite functional traits (here, derived from the principal components). Collectively, our findings suggest that, from the perspective of extant diversity, diversification and functional trait evolution appear largely uniform among replicate continental clades, despite varying colonization timing and diverse ecologies.

Biogeographic origins of the American oscine avifauna

It has long been known that the unique avifauna of the Americas comprises several distinct and independent groups of passerines. Mayr (1946, 1964) performed an extensive survey of passerine diversity in North and South America and found that this assemblage, specifically the oscines, is derived from many independent dispersals of passerine clades from the Eastern Hemisphere into the Americas. He also argued that these clades likely arrived in an ecosystem harbouring autochthonous passerine (e.g. suboscine) and non‐passerine (e.g. Piciformes) diversity, which may have limited ecological opportunities, perhaps, especially for later dispersers. Decades later, using biogeographic modelling and molecular phylogenetics, our understanding of this assemblage has been further clarified, showing that the oscines of the Americas descend from even more dispersal events from the Eastern Hemisphere than Mayr imagined (Barker et al., 2004; Kennedy et al., 2014; Moyle et al., 2016; Oliveros et al., 2019; Selvatti et al., 2015). We built on this foundation by synthesizing the phylogenetic knowledge of American oscines from the past 20 years, identifying at least 46 such events. Although our timescale is likely too old given our reliance on Jetz et al.'s supertree as a framework for this study (Moyle et al., 2016; Oliveros et al., 2019), given the ages we inferred for these dispersal events and present knowledge of the biogeographic history of oscines in the Western Hemisphere, the majority of these clades likely dispersed from eastern Eurasia into North America through Beringia, which facilitated countless overland dispersal events in many taxa through the latest Miocene and then sporadically through the Pliocene and Pleistocene (Cracraft, 1973; Hopkins, 1967; Marincovich & Gladenkov, 1999; Sanmartín et al., 2001). For some clades, though, there is evidence that transoceanic dispersal into the Americas was a more likely avenue for colonization, in most cases across the Atlantic, as with thrushes and finches (Imfeld et al., 2020; Nylander et al., 2008; Voelker et al., 2009; Voelker & Klicka, 2008; Zuccon et al., 2012), or more rarely across the Pacific, as with Donacobius (Fregin et al., 2012). Although the historical biogeography of oscine dispersal into the Americas could be further clarified with a more complete phylogeny, biogeographic model fitting or future fossil discoveries, the 46 colonizing clades we identified in this study provided us with a robust resource with which to study the relationship of diversification and functional trait evolution.

Diversification is strikingly uniform among continental clades

We currently lack a completely sampled phylogeny from which we can directly calculate clade‐specific rates of diversification and trait evolution. To circumvent this limitation, we extended an approach applied by Bokma (2003) and Ricklefs (2007) to estimate diversification in the American oscine assemblage using clade age and species richness estimates from many independent clades. Perhaps unsurprisingly given the coarse‐grained data used here, we found near‐equal support for several alternative diversification scenarios, consistent with recent work suggesting that phylogenies of extant organisms can equally support a panoply of diversification scenarios (Louca & Pennell, 2020). Specifically, the assemblage of clades examined here best‐supported scenarios with either constant speciation and relatively high extinction (with an extinction fraction of 64%, Table 2), or with decelerating speciation and either low or zero extinction. A decelerating speciation without extinction scenario was marginally better supported in the American oscine assemblage and is consistent with diversification analyses reported in Oliveros et al. (2019), who inferred low rates of extinction within passerines as a whole although estimating extinction in the absence of fossil data poses significant challenges; (Rabosky, 2010).

At the geographic and phylogenetic scales examined here, it has been proposed that diversification will largely adhere to time‐for‐speciation expectations, which simply posit that older clades exist long enough to undergo a greater number of speciation events than younger clades and, thus, have greater species richness (Marin et al., 2018; McPeek & Brown, 2007; Stephens & Wiens, 2003). Although this is true for most oscine clades in the Americas (Figure 1), our study recovered several exceptions to the uniform accumulation of diversity. On the one hand, the lower than expected diversity of some clades given their age may be attributable to exceedingly low speciation rates or the extinction of closely related lineages in either hemisphere, which would skew the association between stem age and colonization timing (Donoghue & Sanderson, 2015). Such an extinction of close relatives likely occurred among Ptiliogonatidae and close relatives (Figure 1, point 2), one of the several species‐poor and geographically disjunct clades within the superfamily Bombycilloidea (Spellman et al., 2008). This specialized, largely frugivorous superfamily's pattern of old, species‐poor clades with geographically disparate ranges across the globe suggests this group was once more widespread and has likely experienced substantial range contraction and extinction, as noted for other (typically much older) avian clades like mousebirds (Coliiformes; Ksepka & Clarke, 2010; Ksepka et al., 2017) and hoatzins (Opisthicomiformes; Mayr et al., 2011).

The identification of an additional 18 ‘depauperate’ monotypic clades in the Western Hemisphere, all of which fall far outside the confidence intervals of our uniform rate diversification analyses, implies a remarkably different evolutionary trajectory than the rest of the clades in our study. In the absence of a fossil record, open questions remain as to whether these are the last remaining lineages of previously more diverse clades or whether these lineages simple have not diversified within the Western Hemisphere. With the exception of Donacobius and Dulus, all of these monotypic clades are endemic to North America, and many of them are restricted to boreal, montane or arid regions of the continent. Despite their biogeographic similarities, these species differ from one another in many other ways. Their sizes span two orders of magnitude (5 g in Regulus satrapa up to 500 g in Corvus cryptoleucus), their stem ages span 35 million years (though, as discussed elsewhere, this is likely an overestimate) and the clades themselves originate from throughout the oscine phylogeny. Thus, the similarity of their evolutionary outcomes following dispersal into the Americas is intriguing. Specifically comparing the ecological (Ricklefs, 2005), behavioural or environmental niches (Cooney et al., 2016; Rodríguez‐Castañeda et al., 2017) of these taxa to their closest relatives in the Eastern Hemisphere may shed light on what restrained the diversification of these lineages in the Americas and also on factors that promote or constrain diversification in songbirds.

Alternatively, others have argued that diversification at these scales is largely allopatric, driven by abiotic factors in the environment, and is independent of the density of lineages in a given region (Cracraft, 1985). In this case, variation in species richness can be driven by how vulnerable individual clades may be to factors driving vicariance. Susceptibility to vicariant diversification has been demonstrated to stem from niche conservatism, like in plethodontid salamanders (Joyce et al., 2019), or from poor dispersal potential, as seems to be the case within the avian family Furnariidae (Claramunt et al., 2012). By contrast, the largest Turdus clade in our study (Figure 1, point 3) displayed the opposite pattern of greater‐than‐expected diversification, possibly due to efficient dispersal over a large geographic area, potentially contradicting the negative relationships between diversification and dispersal ability exhibited by other groups. Turdus is a nearly cosmopolitan genus comprising many migrant species, and the largest of the four Western Hemisphere clades ranges from northern‐most Alaska to southern‐most Argentina, suggesting that this clade's rapid diversification may have come about through continuous range expansion coupled with allopatric speciation (Kennedy et al., 2018), especially in the absence of greater‐than‐expected functional diversity (Table S3).

It has also been proposed that ecological interactions between daughter lineages and their biotic environment may cause a density‐dependent relationship between species richness and age through time (Pybus & Harvey, 2000; Rabosky, 2009; Rabosky & Hurlbert, 2015), and this density‐dependent relationship has been documented in a number of avian groups (Phillimore & Price, 2008; Rabosky & Lovette, 2008). Although we could not rule out alternative models invoking high extinction, we note that our best‐fit model was, in fact, a model of decelerating diversification (Table 2). Interestingly, the most striking diversification outlier in this study, the superfamily Emberizoidea (Figure 1, point 1), has previously been shown to have experienced relatively high diversification rates (Barker et al., 2013, 2015), even when compared to the entire global passerine radiation (Oliveros et al., 2019; Ricklefs, 2003). Its globally high rates of diversification and long history in the Western Hemisphere have both likely contributed to this clade's unexpectedly high species richness, as has been suggested for Andean frogs (Hutter et al., 2017), and this clade's unparalleled morphological disparity further suggests that clade age, species richness and their interaction underly the evolution of functional diversity. Thoroughly evaluating these alternative diversification scenarios requires integrating information about clade age and species richness with information about temporal patterns in the evolution of functional ecological diversity exhibited within these clades. Our analyses of the morphological disparity among oscine clades (discussed below) give credence to a scenario in which diversification possesses an ecological dimension.

Functional trait evolution is correlated with diversification

Functional trait diversity is widely variable among Western Hemisphere oscine clades, in which the range of morphological disparity spans three orders of magnitude (Table 1, Figure 2). By quantifying the morphological disparity of 28 oscine clades in this multidimensional morphospace and by modelling this disparity in a phylogenetic comparative framework, we found strong support for a relationship between diversification and functional trait evolution among American oscine passerines, regardless of colonization timing. Previous work investigating the macroevolutionary predictors of functional diversity has proposed that it increases as a function of solely clade age (Foote, 1992) or species richness in a cladogenic manner (Ricklefs, 2004, 2012; Triantis et al., 2016). In contrast to these ideas and to early burst expectations under an adaptive radiation scenario (Harmon et al., 2010), but consistent with expectations from simple uniform models of trait evolution (i.e. Brownian motion), we found that a combination of clade age and species richness (and their interaction—which together provide a good proxy for total tree length) is the best predictor of functional diversity among clades.

Although the vast majority of clades occupy quite small regions of morphospace (<10% of the total oscine morphospace), two clades occupy substantially greater regions than any others (>20%): Troglodytidae–Polioptilidae and Emberizoidea. Unsurprisingly, these two clades are also the most species‐rich and among the oldest of the 46 oscine clades in the Americas. It is conceivable that their early colonization of the Western Hemisphere may have granted them ecological opportunity to expand their morphological diversity beyond that of any subsequently colonizing oscine group (Stroud & Losos, 2016), as well as an incumbent advantage that potentially filtered subsequent dispersal events or suppressed their diversification (Betancur et al., 2012; Tanentzap et al., 2015; Múrria et al., 2018; Rowsey et al., 2020). We found no consistent evidence that Emberizoidea or Troglodytidae–Polioptilidae had greater‐than‐expected trait disparity; however, simulations under the two best‐fitting diversification scenarios could not consistently predict the disparity of both groups, either underpredicting diversity of Emberizoidea (birth–death) or overpredicting diversity of Troglodytidae–Polioptilidae (declining diversification). As the birth–death model provided the best overall fit to the data (Table S3), our best evidence is that Emberizoidea has evolved significantly greater disparity than expected given its age and species richness and thus has evolved more rapidly than other lineages in our analysis.

The evolution of superlative diversity in this clade was likely influenced by a number of factors. Emberizoidea's relatively early dispersal into the Americas provided a longer period of time in which diversification could occur, and its current hemisphere‐wide range implies a substantial opportunity for allopatric speciation within and between both continents. However, its status as a significant diversification outlier among American oscine clades implies that additional processes likely played a role. Another plausible factor that may have influenced the functional trait evolution in Emberizoidea is the phenotype of the colonizing population. Price (2011) has observed that two avian adaptive radiations, the Galápagos finches and Hawai'ian honeycreepers (Tokita et al., 2017), and several ‘mini‐radiations’ of songbirds, such as the finches of Tristan da Cunha (Ryan et al., 2013) and possibly Phaenocophilidae of Hispaniola (Barker et al., 2013), descended from finch‐billed ancestral populations. This scenario was likely the case for Emberizoidea as well, given that they are the sister taxon of the true finch family (Fringillidae) and several of the earliest diverging clades within Emberizoidea are the finch‐like buntings and longspurs, the cardinals and allies, and the sparrows of the Western Hemisphere (Barker et al., 2013, 2015; Oliveros et al., 2019). Thus, the extraordinary diversity of Emberizoidea as a whole may be an instance of a generalist, finch‐billed ancestor priming diversification and functional trait evolution. Indeed, this scenario is supported by the largest family within the clade (Thraupidae; Vinciguerra & Burns, 2021). Direct estimates of ancestral states, morphological evolutionary rates and diversification rates for Emberizoidea as well as the remaining American oscines may provide further insight into their relationships once species‐level phylogenies for each of these clades is available for study.

Continuous diversification as the macroevolutionary norm?

Although the conceptual model of adaptive radiation and simple models of trait evolution both predict that an association exists between diversification and functional trait evolution, there are many reasons to doubt that these models should perform so well in empirical examples. Indeed, previous work assessing this prediction in a phylogenetic comparative framework has shown the association to hold in some groups (Derryberry et al., 2011; Ford et al., 2016; Rabosky & Adams, 2012; Rabosky et al., 2013; Silva et al., 2016), where a breakdown in this relationship has been documented in other studies (Adams et al., 2009; Crouch & Ricklefs, 2019; Lee et al., 2016; Oliveira et al., 2016; Rowe et al., 2011; Venditti et al., 2011). Reasons for this breakdown are diverse. In the case of non‐adaptive radiations, like in plethodontid salamanders, there can be substantial diversification in the near‐absence of morphological evolution due to climatic niche conservatism within some temperate lineages (Joyce et al., 2019), or climatic niche adaptation in some tropical lineages (Kozak & Wiens, 2010). Alternatively, diversification of behavioural traits, rather than functional morphology, can facilitate extensive niche differentiation and influence community assembly, as suggested for the Australian honeyeaters (Miller et al., 2017). Furthermore, ecological interactions between recently dispersed lineages and incumbent clades may lead to some being constrained and having lower species richness and less functional diversity than would be expected given their age and an expectation of uniformity (Betancur et al., 2012; Tanentzap et al., 2015; Múrria et al., 2018; Rowsey et al., 2020). This scenario in particular seems possible for dispersing oscine clades given that the Western Hemisphere already harboured the incumbent suboscines for millions of years prior to their arrival (Mayr, 1964; Oliveros et al., 2019; Ricklefs, 2002) and that only one of the oldest and most species‐rich oscine clades seems to have achieved extraordinary diversity. The potential effects of incumbent clades on diversification over time are the subject of a parallel set of analyses based on these data to be reported elsewhere. Our current study assesses the predicted positive relationship between diversification and functional trait evolution using many replicate, biogeographically delimited clades and found that, with some exceptions, these processes appear to be correlated and strikingly uniform among the dozens of oscine clades in the Americas. The fact that this relationship and this uniformity are recovered when studying passerines as disparate as crows, swallows, thrushes, wrens and finches suggests that diversification and functional trait evolution are happening not only at short times scales in tightly circumscribed spatial and ecological conditions, but continuously over broader spatiotemporal and ecological scales.

CONCLUSIONS

The goal of this present study was to evaluate the uniformity in diversification among replicate, independently dispersing lineages experiencing ecological opportunity in a novel continental context. We identified 46 instances in which oscines dispersed from the Eastern Hemisphere into the Americas, providing a rare opportunity to examine the relationship between diversification and functional trait evolution in many co‐distributed continental clades. We found near‐equal support for three alternative diversification processes with varying speciation and extinction regimes and found that, with a number of notable exceptions, diversification—when it happens at all—appears largely uniform among clades that dispersed to the Western Hemisphere. Analysing the morphological disparity of these clades, we show that functional trait evolution is correlated with diversification over broad time scales and that, like diversification, it appears largely uniform among the ecologically diverse clades in this study. One exception to this generalization is a substantial number of lineages that successfully established themselves in the Americas over a span of millions of years, but either never subsequently diversified or have experience significant extinction. A second notable exception is a clade (the superfamily Emberizoidea) that is clearly more diverse than expected under uniform diversification processes and occupies the majority of American oscine morphospace, possibly implying more rapid functional evolution than expected. A final clade (Ptiliogonatidae) is clearly less diverse than expected in both species richness and trait disparity. Despite these outliers, our study of dispersal and macroevolution of oscines in the Americas supports the idea that continuous, fairly uniform evolution of functional diversity is a common pattern in macroevolution.

AUTHOR CONTRIBUTIONS

T. S. I. conceived of the research, collected all morphological data, performed the analyses and wrote the manuscript. F. K. B. contributed substantially to the analyses and writing of the manuscript.

ACKNOWLEDGEMENTS

This study would not have been possible without natural history collections and the people who make this research infrastructure accessible. We especially thank the curators and collection managers at the Bell Museum at the University of Minnesota (MMNH), the Field Museum of Natural History (FMNH), the American Museum of Natural History (AMNH) and the Louisiana State University Museum of Natural Sciences (LSUMZ). All specimens measured in this study were collected under permits to corresponding institutions. This research was funded by a Florence Rothman Grant from the University of Minnesota College of Biological Sciences, a Richard & Judie Huempfner Grant from the MMNH, a Collection Study Grant from the AMNH, and a grant from the National Science Foundation (DEB‐1541312). We also give special thanks to Sharon Jansa, Kieran McNulty, Sushma Reddy, Dakota Rowsey, Shanta Hejmadi, Brie Ilarde and three anonymous reviewers whose feedback improved this work substantially.

CONFLICTS OF INTEREST

The authors have no conflicts of interest regarding this work.

DATA AVAILABILITY STATEMENT

Morphometric data, tree files and all R scripts used in this study are available from the Dryad Digital Repository (doi:10.5061/dryad.0gb5mkm3z).