The evolution of avian migration

Authors


Abstract

The study of avian migration has reached sophisticated levels in many areas, including ecology, behaviour, and physiology. Traditional discussions of the evolution of migration, however, have been compromised for several reasons. Previous ideas concerning the ancestral home of migrant species, southern or northern, and whether a partially migratory stage always precedes a fully migratory stage, were not expressed as testable hypotheses. Plotting migratory behaviour on phylogenetic trees has become commonplace and allows tests of traditional hypotheses. Some of these studies are reviewed, lending some support for almost all of the previous ideas. Although phylogenetic mapping helps to frame questions about the evolution of migration in a testable framework, there are two serious issues. First, experimental and observational studies reveal that the expression of migratory behaviour can change rapidly within a lineage, which can violate assumptions of character mapping. In addition, a species distribution model is used to show that current conditions for obligate migratory populations of the chipping sparrow were much restricted at the Last Glacial Maximum, and that the species might have been considered a partial migrant at that time. The expression of migratory behaviour in an extant species might be an artefact of the current inter-glacial period. Only if the rate of gains and losses of migratory behaviour can be incorporated into a phylogenetic mapping exercise will the actual evolutionary pattern of migration be revealed. For example, reconstruction of the ancestral area and the evolutionary history of migratory categories in a clade of New World warblers depended on the assumptions of character state transitions. A second concern is that the trait ‘migratory’ is too broad for evolutionary analysis and that, if possible, the expression of hyperphagia, Zugunruhe, and navigation could be mapped individually. Loss or suppression of any of these components can lead to sedentary populations, revealing how migratory behaviour can appear and disappear rapidly. A report of low levels of Zugunruhe in a sedentary bird, Saxicola torquata, is reconstructed as derived in a clade of otherwise migratory populations, suggesting that the loss of migration was a result of suppression (but not elimination) of Zugunruhe. When researchers mention the independent origin of migration in a clade, they are most likely referring to the gain or loss of the expression of the ancestral migratory programme, not the de novo evolution of migration per se. © 2011 The Linnean Society of London, Biological Journal of the Linnean Society, 2011, 104, 237–250.

INTRODUCTION

It is difficult to think of a superlative concerning avian migration that has not already been expressed multiple times (Berthold, 1996, 1999, 2001). Indeed, skies have been described as being filled with migrating birds, with tales of incredible journeys, physiological feats of endurance, and homing abilities that rival our best Global Positioning System devices. These behaviours have fascinated us for centuries and our knowledge of aspects of migration has become impressive. We know how birds fatten up before migration, how they can fly overwater for hours on end, and that they can use the earth's magnetic field or the stars to navigate. However, placing our knowledge of avian migration in an evolutionary context has been less well accomplished. The present review considers past thoughts on the evolution of migration and suggests an agenda for future research.

Historical views on the evolution of migration

The core idea of most historical views on the evolution of migration is that colonists from a sedentary population find and proliferate in a new environment but are then forced by environmental deterioration to return to their ancestral home (Rappole, 1995; Berthold, 2001). It was not clear whether this happened over ecological time or over evolutionary time. Also, it was usually not made clear when the ancestral area was vacated during this process so that, when the migrants returned, there were no sedentary conspecifics remaining with which to interact. That is, when during this protracted process did the ancestral home become unoccupied? Often, a partially migratory population was viewed as a necessary or likely intermediate step. Several other common themes were expressed, including the necessary assumption that the acquisition of migratory behaviour involved a concomitant increase in fitness for migrants relative to sedentary individuals, owing to high costs of migration (Strandberg et al., 2010). Other studies considered whether reliance on a certain food type (e.g. fruit) might predispose populations to be mobile and therefore be pre-adapted to become migratory (Cox, 1968; Bell, 2005). Some studies (Pulido, 2007) frame the discussion of the evolution of migration in the context of proximate and ultimate causes; however, the present review suggests that there is a continuum of causes.

The question of geographical directionality was also debated. Two prominent ideas emerged: the ‘southern home’ and the ‘northern home’ hypotheses (Rappole, 1995; Salewski & Bruderer, 2007). These ideas were derived from examining the current distributions of migratory species and their presumed close relatives. The southern home idea assumed that tropical sedentary ancestors experienced episodes of dispersal (either colonists or range expansion), resulting in some individuals reaching northern areas where a super abundance of summer resources, during interglacial periods at least, gave them enhanced fitness over their sedentary conspecifics, which then presumably went extinct because there was a fitness disadvantage to remaining sedentary (otherwise all species would be continuously distributed with only northern populations migratory). This idea postulates that migrants breeding in north or south temperate regions are returning to their tropical ‘ancestral homes’ during the winter, and that they should therefore have at least some residual adaptations allowing them to fit into environments very different from those in which they now breed. If so, this could ameliorate the cost of evolving adaptations to surviving in often radically different breeding and nonbreeding environments.

The northern home has a similar set of ideas but, instead, the ancestral range becomes too difficult for survival and fitness is enhanced by leaving during the winter to a novel home (the tropics), and then returning. This idea is often accompanied by the view that migrants are able to ‘fit in’ to complex tropical environments by being ‘fugitive species’ and wandering from one opportunistic wintering area to another, biding their time until environmental conditions on the ancestral northern breeding area once again become hospitable, whence they would return ‘home’ to breed. Thus, they could avoid the costs of adaptation to tropical habitats by drifting among habitats with some similarity to those in which they breed.

This brief summary does not do justice to the many other ideas that were expressed in the literature on the evolution of avian migration. However, this summary does illustrate the concern that these views qualify as ‘just-so’ stories, as expressed by Gould & Lewontin (1979). That is, these hypotheses were not expressed in a testable framework (Zink, 2002). For example, many species were suggested as providing support for northern or southern homes, or what caused a species to become migratory, although there were no clear criteria set forth that would allow falsification of either.

Phylogenetic mapping of migration

The emergence of the field of comparative biology offered a means of testing hypotheses on the evolution of migration (Felsenstein, 1985). By plotting migratory behaviour (migratory or sedentary) on a phylogenetic tree, it became possible to determine whether ancestors of migratory species were sedentary and whether northern or southern homes were ancestral. It was also possible to determine whether a consistent transition existed between sedentary, partially migratory, and fully migratory conditions. By using a phylogenetic tree, one could now provide a test of hypotheses, such as the ‘evolutionary precursor hypothesis’ (Levey & Stiles, 1992), which suggests that migration is selected for in species occupying edge or open habitats. In addition, many studies have suggested that speciation can occur when migratory individuals ‘stay behind’ and form a new species on the wintering grounds; this leads to the phylogenetic prediction of southern–northern sister species.

Beginning with studies by Burns (1998), who concluded that seasonal migration evolved once in some tanagers (genus Piranga), it became common to see migration evaluated in a phylogenetic context. The present review evaluates studies of the phylogenetic mapping of migration in birds, and although the intent is not to provide a complete review, several studies are discussed to illustrate typical results. In some cases, novel conclusions have emerged from these mapping studies.

Interspecific studies

Winker & Pruett (2006) constructed morphological and molecular phylogenies of the genus Catharus (Turdidae), a group of New World thrushes, to determine how long-distance seasonal migration evolved. They concluded that migration apparently arose independently four times from an ancestral sedentary ancestor, and they commented on the rapid rate of transitions between migratory and sedentary states over short evolutionary timescales (Berthold et al., 1992; Outlaw et al., 2003).

Amaral et al. (2009) constructed a phylogeny of buteonine hawks and plotted the distribution of migratory, partially migratory and sedentary states. Of the six possible pairwise transitions (e.g. sedentary to partial), it was found that all 15 state changes on their phylogenetic hypothesis involved a partially migratory stage. There were six changes from sedentary to partially migratory, and three reversals from a partially migratory species to a sedentary one. Kondo et al. (2008) studied two oriole (Icterus) sister species and found that, in contrast to the traditional paradigm, a short-distance migrant was derived from a long-distance migrant. At a more inclusive level in oriole phylogeny, Kondo & Omland (2007) used multistate ancestral state reconstruction of migration (coded as long-distance, short distance, partial migrants or sedentary), to examine patterns of migratory evolution in the New World orioles. They mapped migratory distances for each species defined as the distance between the midpoints of breeding and wintering ranges. It was concluded that there were five recent and rapid gains of migration within the genus. Kondo & Omland (2007) concluded that every migratory species' migration type differed from that of its closest relatives, and there were no instances of a sister-species relationship in which one was migratory and the other partially migratory, in contrast to the situation in hawks (Amaral et al., 2009).

Joseph, Wilkie & Alpers (2003) constructed a phylogeny of taxa in the southern hemisphere Myiarchus swainsoni complex (sensuLanyon, 1978) and determined that two groups of long-distance, temperate–tropical migratory populations were not closely related and almost certainly evolved migration independently. In an earlier study on shorebirds, Joseph, Lessa & Christidis (1999) addressed the question of ancestral homes by plotting both the wintering and breeding ranges on a phylogeny. Their reasoning was that, by considering winter and summer ranges independently, it might reveal whether their study species had tropical or temperate ancestors. They concluded that South America was the ancestral home from which migratory species developed.

Outlaw & Voelker (2006) constructed a phylogeny of the wagtails (Motacilla) and tested predictions of the ‘evolutionary precursor’ hypothesis and the ‘stepping-stone’ hypothesis. In the first hypothesis, migration is more likely to evolve in those species occupying edge or open habitats (nonbuffered) as a result of increased temporal resource variability compared to forested habitats (buffered). Alternatively, the stepping-stone hypothesis predicts that seasonal environments drive the evolution of migration and, because of competition, migratory populations winter in areas from which their sedentary sister populations are absent. It was concluded that migration evolved several times and, in most cases, was conditional upon and correlated with a shift to higher latitudes. Although they were not able to support one hypothesis consistently, Outlaw & Voelker (2006) wrote that ‘One aspect of the evolutionary precursor hypothesis, that migratory behavior is correlated with open or nonbuffered habitats, is not supported’. They further suggested that ‘the movement into seasonal environments is probably driving the evolution of migration in the Motacillidae’ (Outlaw & Voelker, 2006: 464). Importantly, regarding methods, they noted that ‘Although our results may seem obvious, the relationship between migration and seasonality has not been explicitly tested in an evolutionary framework such as this’ (Outlaw & Voelker, 2006: 464). Indeed, studies such as Outlaw & Voelker (2006) reveal that the field has entered the hypothesis testing era.

Boyle & Conway (2007) examined the relationship between migratory behaviour and foraging group size, membership in mixed-species flocks, elevational range, body mass, and habitat in a large group of New World flycatchers, the Tyranni. Similar to Outlaw & Voelker (2006), they considered the evolutionary precursor hypothesis, noting that, although previous studies had implicated edge habitats and fruit-eating as precursors to the evolution of migration, there had been no phylogenetic controls. They concluded that short-distance migratory movements are not a necessary precursor to the evolution of long-distance migration. Increasing degree of frugivory led to an increasing likelihood of migration but insectivory, not frugivory, was associated with increases in migratory distance. An unexpected conclusion was that the most pervasive correlate of migration was foraging group size. In particular, they concluded that ‘birds foraging solitarily are more likely to migrate and migrate farther than birds foraging either in pairs or in groups (fig. 3) suggesting that migration could impede the maintenance of pair and family group foraging bonds’ (Boyle & Conway, 2007: 354). This intriguing conclusion came from phylogenetic mapping and appears to be a novel suggestion about factors that might influence whether species become migratory.

Figure 3.

Species distribution models for chipping sparrow at present (A), Last Glacial Maximum (LGM) (B) and Last Interglacial (LIG) (C), using only breeding records from north of 40.5° N. Although there is extensive predicted occurrence at LIG and present for obligatory migratory populations, there is a restricted distribution predicted at the LGM, suggesting that, at this time, chipping sparrows were not migratory, or that obligate migrants were restricted to the northwestern (current) USA.

Friedman et al. (2009) mapped migratory/sedentary and monomorphic/dimorphic on a phylogeny of New World orioles. These birds exhibit elaborate male plumage, and it was inferred that sexually dimorphic species evolve as a result of the loss of elaborate female plumage (they assume a duller and less patterned plumage). Interestingly, most sexually dichromatic species belong to migratory or temperate-breeding clades. Friedman et al. (2009) concluded that gains of sexual dichromatism were 23-fold more likely to occur in migratory taxa. This leads to the intriguing suggestion that elaborate female plumage may be selected against in migratory species. The reasons for this are uncertain, although this correlation would not have been found without phylogenetic mapping.

Escalante et al. (2009) constructed a tree for warblers in the genera Geothlypis and Oporonis, which included several migratory and sedentary species, and distinct groups within currently recognized single species. They concluded that migratory behaviour evolved a minimum of three times and that a tropical ancestor was most likely. However, their tree also was consistent with a northern origin for a proto-Geothlypis ancestor that evolved from migratory stock. Importantly, it was concluded that several of the sedentary species of Geothlypis evolved from migrants that established breeding populations on the wintering grounds, a phenomenon referred to as ‘migratory drop-off,’ which they suggested to be an underappreciated form of avian speciation.

Examination of the distribution of migratory species on phylogenetic trees for the Old World warbler genera Phylloscopus and Sylvia suggested to Helbig (2003) that long-distance migrants were rarely related to each other, instead originating from a sedentary common ancestor. Helbig (2003) argued that once a species became a long-distance migrant, some individuals would inevitably go off course during migration and connect otherwise incipient species, with the concomitant gene flow preventing speciation. Winker (2000) argued, in contrast, that migration provided a means by which species could explore novel environments and would lead to species ‘flocks’ of migrants. Helbig (2003) did not observe this pattern in the warblers that he studied, although he noted that an ideal group for testing his ideas was the New World warblers (Parulidae). However, at the time, there was no comprehensive tree for the family.

Lovette et al. (2010) constructed a multigene phylogenetic hypothesis that permits the examination of the evolution of migration in the Parulidae. A full analysis is beyond the scope of the present review, although several observations can be made. First, there are ‘flocks’ of sedentary species, including species in the genera Basileuterus and Myioborus. The species of Dendroica clearly constitute a ‘flock’ of migratory species, in contrast to the pattern observed in Phylloscopus and Sylvia (Helbig, 2003). To explore the proposal of Helbig (2003), the present review examined sister-species and tallied whether they were both migratory, one was migratory, or both were sedentary. Under Helbig's view, sister species should include a sedentary and a migratory species. However, Helbig's hypothesis receives little support from these New World warblers because there were 11 sister species pairs that were both migratory, 13 that were sedentary, and only five containing one of each. This is typical of many studies in which evolutionary patterns of migration appear to be idiosyncratic within and among lineages.

Intraspecific studies

The previous studies involved plotting migration on phylogenetic trees representing relationships of species within and among genera. The evolution of migration has also been addressed with phylogeographical studies. For example, Milá, Smith & Wayne (2006) studied mitochondrial DNA variation in a widespread North American species, the chipping sparrow (Spizella passerina). Chipping sparrow populations in Mexico and the southern USA are sedentary (or at least the range is occupied year around) and populations in the northern USA and Canada are migratory. Southern populations have higher genetic diversity than northern populations, which Miláet al. (2006) interpreted as a consequence of recent range expansion into the northern part of the range. Thus, populations would have become migratory when range expansion reached seasonal north temperate areas that required migration to the ancestral area for survival, which is consistent with an early stage of the ‘southern home hypothesis’ in which southern ancestral populations are still present. They concluded that ‘migratory behaviour was not a prerequisite for the northward expansion in chipping sparrows’ (Miláet al., 2006: 2407), suggesting instead that it developed because recent post-glacial range expansion extended the species distribution into an area that was not habitable year-around. Although this interpretation is consistent with the data, it overlooks the fact that with only two areas, temperate and tropical, it is not possible to tell which one is ancestral because they are sister areas. It is possible that chipping sparrows existed in the north temperate regions south of the ice sheet in small populations and expanded as the glacier retreated, whereas tropical populations were relatively unaffected. This would leave the observed signature of lowered genetic variation in temperate populations owing to an in situ bottleneck, which would be difficult to distinguish from the range-expansion process suggested by Miláet al. (2006).

The northward expansion hypothesis of Miláet al. (2006) implies that chipping sparrows occurred in Mexico at the last glacial maximum (LGM) and that populations in this area served as the source for northward expansion. The present review constructed a species distribution model (SDM; Peterson, Soberon & Sanchez-Cordero, 1999; Elith et al., 2011) to predict where populations of chipping sparrows resided at the LGM and whether they were absent from North America. Breeding records from Mexico (N = 713; A. Navarro, pers. comm.) were used together with the breeding bird census (N = 3876; Breeding Bird Survey, http://www.pwrc.usgs.gov/bbs), which were input into MAXENT, version 3.2.2. (Phillips, Anderson & Schapire, 2006) to infer a SDM. Climatic data (19 layers) were obtained from the Worldclim bioclimatic database (Hijmans et al., 2005), and Maxent was set to use 30% of values for training. The SDM was based on the mean of five Maxent runs and plotted using DIVA-GIS, version 7.1.7.2 (Hijmans et al., 2004). The SDM predicted approximately 80% of the occurrence points. The model predicted the current distribution well (Fig. 1A), although over prediction (albeit at low suitability) occurred in some regions (e.g. south central USA, and southern Florida). Using wintering records from Audubon Christmas Bird Counts, a SDM was also constructed for wintering conditions (Fig. 1B), which corresponds well to the known wintering/permanent residency range (Middleton, 1998); the LGM winter range (not shown) was displaced to the south of the current range. The SDM (Fig. 1C) predicted an extensive LGM breeding distribution in the southern and western USA, with relatively low amounts of suitable habitat in Mexico (and a wintering distribution displaced south of the current winter range; Fig. 1D). The SDM is a prediction of where conditions for the species existed, and not whether it was present. Nevertheless, the SDM (Fig. 1C) suggests that chipping sparrows that currently breed in North America are recent colonists from the western and southern parts of the USA and not from Mexico. The current populations in Mexico might have been derived from populations distributed farther south in Central America at the LGM, which could also enhance genetic variability.

Figure 1.

Species distribution models for present breeding populations (A), present wintering distribution (B), breeding populations at LGM (C), and wintering distribution at the last glacial maximum (LGM) (D), computed from current breeding locality records in Breeding Bird Survey and from Mexico (A, C), and wintering records from Christmas Birds counts (for which there were at least 25 individuals recorded; B, D). Black indicates the highest predicted probability of occurrence. There is no clear evidence of the LGM breeding refuge predicted by Miláet al. (2006) in Mexico.

To explore distributions earlier in time, the distribution of chipping sparrows at the Last Interglacial (LIG) was also reconstructed. The predicted distribution (not shown) suggests a widespread range similar to that observed at present (Fig. 1A). Hence, as expected, chipping sparrows have undergone major range contraction (LIG to LGM) and expansion (LGM to present) during the past 120 000 years. The consequences of this cyclical history are explored below. Nonetheless, the SDM is inconsistent with the speculations of Miláet al. (2006) that chipping sparrows extended their range from a tropical sedentary ancestor into a region that at present precipitates migratory behaviour. This is not to say that migration could not develop this way in other species.

Summary

Most traditional inferences on the evolution of migration fall into the trap outlined by Gould & Lewontin (1979) in their famous critique of the adaptationist programme. That is, discussions of southern versus northern homes, or the hypothesized progression from sedentary to partially migratory to migratory, were ‘just so’ stories and not tested in a rigorous way (Zink, 2002). The framework provided by a phylogenetic tree yields tests of the evolutionary hypotheses (Felsenstein, 1985) involving migration.

As the above few examples illustrate, it is possible to find examples of nearly every transition from one migratory state to another, and to trace ancestral areas to southern and northern regions. It is not possible in my opinion to conclude that any previous notions about the location of ancestral distributions or the ordering of sedentary–partial migratory–migratory have received unambiguous support; it depends on which group of birds is considered. Of course, there is no reason to presume that any one pattern ought to be consistent across different avian lineages.

There is a deeper issue, however, that clouds these phylogenetic tests of migration hypotheses. This issue involves the assumptions involved in phylogenetic character mapping. At present, it is suggested we do not know whether various groups of birds evolved migration differently, as the phylogenetic mapping studies suggest, or whether there is a common pattern across avian lineages that is obscured by violations of assumptions concerning mapping of ancestral states.

A critique of phylogenetic mapping of migration

Mapping a trait on a phylogenetic hypothesis and subsequent interpretation of the evolution of the trait involves assumptions. One assumption of ancestral character state reconstruction using parsimony is that the trait does not evolve more rapidly than the pace of speciation events (Schluter et al., 1997; Cunningham, Omland & Oakley, 1998; Blomberg, Garland & Ives, 2003). That is, the observed state of the feature in an extant species has either changed one (autapomorphic) or zero times since the species last shared a common ancestor with its sister species. If a feature changes too rapidly, reconstruction or transitions will be an artefact of whatever phenotype occurs in the modern population. Another assumption is that the trait qualifies as a character in a phylogenetic sense. That is, the feature being mapped should not be overly broad in what it encompasses. If a character being mapped consists of many independent components, then the same character state could be ‘evolved’ in different ways, making the mapping exercise of limited value because the characters are not homologous. For example, eating fruit has evolved independently in many lineages and, although its phylogenetic distribution is of interest for ecological reasons, frugivory per se is not homologous and an understanding of its independent origins requires mapping morphological, physiological, and behavioural components independently.

The influence of rates of gain/loss for mapping migration on a tree

If a species has lost and regained ‘migration’ multiple times since it last shared a common ancestor with its sister species, estimating the frequency or phylogenetic pattern of the transition from sedentary to migratory will be difficult (Fig. 2). The question is whether rapid gains or losses typify avian lineages (Pulido & Berthold, 2010). The expression of migratory behaviour can be altered experimentally in aviaries in less than ten generations (Berthold et al., 1990a, b; 1992; Pulido, 2007). Not all evidence comes from aviary experiments. We know from recent observations that blackcaps (Sylvia atricapilla) have added a new migratory direction and wintering grounds in the past few decades (Terrill & Berthold, 1990; Helbig, 1991). A flock of fieldfares (Turdus pilaris) blown off course in autumn migration made landfall in Greenland and became sedentary in essentially one generation (Salomonsen, 1950), and introduced house finches (Carpodacus mexicanus) in eastern north America have exhibited an increasing frequency of migratory individuals in less than 50 years (Able & Belthoff, 1998). Rohwer, Hobson & Rohwer (2009) discovered north temperate migratory species undergoing a second annual breeding effort on their ‘wintering’ areas, which raises the question of how offspring from the same parents obtain migratory information. These observations suggest that mapping migration per se on a phylogeny might be akin to mapping transitions at third-base codon positions on a phylogeny which is a flawed process owing to multiple hits obscuring the sequence of evolutionary transitions that have occurred in the character.

Figure 2.

A, example of a parsimony reconstruction of migration. The state at the terminals (extant taxa) is assumed to be inherited unchanged from a common ancestor. B, hypothetical example of multiple rapid gains and losses of migration, showing that the state in the extant species is not necessarily representative of the state of the common ancestor.

At a deeper evolutionary level, it is possible that migration has been expressed episodically in species throughout the Holocene glacial cycles. That is, during glacial maxima, species that are today migratory were obviously displaced southwards and might have been sedentary. With recolonization of recently glaciated northern areas, species might have regained their migratory habits, as suggested by Miláet al. (2006). If this process is repeated cyclically, then the current migratory state of a species might simply be a reflection of whether the earth is in a glacial or an inter-glacial period. To explore this, the chipping sparrow is again considered. The chipping sparrow is the sister species to all other congeners and is approximately 5% divergent from them at the mitochondrial cytochrome b locus (Canales-del-Castillo et al., 2010). If a rate of 2%/Myr (Weir & Schluter, 2008) is assumed, then chipping sparrows have been evolutionarily independent from their congeners for much of the Pleistocene and they have persisted through multiple cycles of glacial advance and retreat. Thus, although many of their northern populations are now migratory, it might be incorrect to assume that the state ‘migration’ was inherited from a distant common ancestor and has remained unchanged since its lineage split from its most recent common ancestor.

To evaluate the potential for gains and losses of migratory behaviour at the scale of glacial cycles in the chipping sparrow, a different SDM was constructed. From the data set used above, breeding localities south of 40.5° N were excluded, and Maxent was used to build a SDM for the present, the LGM and the last interglacial period (LIG). It was assumed that individuals in populations north of 40.5° N are obligate migrants (Middleton, 1998) and the goal was to determine whether the climatic conditions that these populations currently use today were present at the LGM.

The SDM predicted over 83% of the training points, and recovered the breeding distribution of current obligate migrants, as well as more southerly breeding sites (Fig. 3A). The projected LGM distribution for obligate migratory populations (Fig. 3B) is limited to relatively small areas in western and southeastern North America (less than 10% of the area currently occupied by obligate migrants) and overlaps with those for all chipping sparrows (Fig. 1A). This suggests that the climatic conditions that are associated with modern obligatory migrant chipping sparrows were not extensive at the LGM, and that it is possible that the chipping sparrow was non-migratory, or only partially so at that time. It is suggested that this provides at least indirect evidence for the switching on and off of migratory behaviour at the temporal scale of glacial cycles. The SDM predicted a widespread occurrence of migratory populations at the LIG (Fig. 3C). Thus, the expression of migratory behaviour is extremely labile in laboratory experiments, over recorded history, and across evolutionary time.

Methods to mitigate rapidly changing states

Accounting for rapid changes between character states requires more sophisticated mapping methods than simple parsimony reconstructions in which gains and losses of migration are treated equally. To illustrate one of these approaches (Kondo & Omland, 2007), the evolution of migration was evaluated on the clade of Lovette et al.'s (2010) phylogenetic hypothesis containing all currently recognized species of warblers in the genus Dendroica and five species from other genera (Wilsonia, Parula, Setophaga, Catharopeza). Each species was scored as sedentary, a short-distance migrant or a long-distance migrant. Maximum parsimony was used in MESQUITE, version 2.74 (Maddison & Maddison, 2010) and two reconstructions were performed: one in which gains and losses were equally likely and the states were unordered, and a second analysis using stochastic character mapping (Huelsenbeck, Nielsen & Bollback, 2003). The purpose of these two analyses was to illustrate the different types of conclusions that result from differing assumptions of the analysis (see Kondo & Omland, 2007).

The parsimony reconstruction (Fig. 4A) is consistent with the conclusion that a sedentary warbler species evolved from a short-distance migrant on three occasions, long distance migration evolved from a short-distance migrant twice, and all other transitions were ambiguous. The stochastic reconstruction (Fig. 4B) suggests six transitions from short-distance to long-distance, four from short-distance to sedentary, and three from sedentary to short-distance. Stochastic character mapping suggests a more fluid history of transitions along branches, which is likely to be more realistic given the potential for migration to be turned on and off rapidly over ecological and evolutionary time (Winker & Pruett, 2006). Thus, different methods of reconstructing migratory states lead to different interpretations and accounting for variation in branch lengths and the likelihood of each transition would further alter conclusions. It appears that an estimate of the rate of gain/loss of migratory behaviour is necessary to make mapping exercises trustworthy.

Figure 4.

Phylogeny of warblers (from Lovette et al., 2010), with sedentary (white), short-distance (grey), and long-distance (black) migration plotted. A, reconstruction using maximum parsimony with the states unordered. B, reconstruction using a stochastic mapping algorithm in which characters can change multiple times on a branch (Huelsenbeck et al., 2003). These trees show differing histories of ancestral state changes owing to differing assumptions of character (migration) evolution.

To illustrate how a phylogenetic analysis could test the southern-home hypothesis, the breeding areas of each species were plotted on this tree (not shown), suggesting that the ancestral geographical area was the Caribbean. However, under a vicariance hypothesis, there is a split near the base of the tree between Caribbean and continental species. There were no instances of a North American species being the sister species of a South American species, in contrast to the predictions of Rappole (1995). Although the base of this part of the tree could be construed as tropical, it is clear that many long-distance migrants are North American, and there are three potential cases of migratory drop-out (Escalante et al., 2009) involving sister species that are sedentary and migratory, although this history of this phenomenon depends on which reconstruction is used. For example, on the tree using stochastic ancestral state reconstruction, Dendroica pinus (North America) and Dendroica pityophila (Cuba and the Bahamas) are migratory and sedentary, respectively, although the reconstruction suggests that the latter only recently became sedentary. However, if species ranges change rapidly during their history, as we know from SDMs, as well as knowledge of current and glacial habitat distributions, mapping ancestral ranges might also be compromised by rapid changes (Chesser & Zink, 1994; Covas & Blondel, 1998). It is possible that the expression of migration and range size are tightly linked.

In summary, unless a phylogenetic mapping exercise incorporated estimates of the rate of gains and losses, reconstructions of ancestral states are not trustworthy. Unfortunately this probably includes most studies published to date, including those reviewed above. Hence, it is possible that the various conclusions reached by different studies are correct, or that the apparently high variability or inconsistency is an artefact of plotting a trait, migration, that simply evolves too quickly to yield meaningful conclusions. If one accepts the results of the chipping sparrow analysis that migration is gained and lost every glacial cycle, then the rate of gain/loss is sufficient high such that every node might be ambiguous. This would argue that mapping migration would be most likely meaningful only if there were very few transitions from one state to another.

Dissecting the components of migration

Overly broad depictions of a behaviour, such as migration, run the risk of leading to invalid evolutionary conclusions (Grandcolas et al., 2011). Piersma et al. (2005) argue that there is no common migratory syndrome in birds, such as an ancestral ‘bauplan’ (with apologies to morphology), owing to the observation that different migratory species might use different navigation systems. That is, even if sister species are both migratory, if one uses celestial navigation and the other magnetic, they would not be migratory in the same ‘way’. Therefore, mapping ‘migration’ would be an overly general exercise and could mask the real evolutionary phenomena that have occurred. That is, both species would be migratory but, because of different mechanisms, making migration per se not homologous in this case. From a purely ecological perspective, it could be worth reconstructing the phylogenetic distribution of migratory species, irrespective of whether migration per se was homologous. From an evolutionary perspective, character state or ancestral reconstructions require the assumptions noted above.

This reasoning suggests that the components of the migratory system could be mapped independently. If we consider a species that undergoes a twice annual migratory journey of over 15 000 km, several physiological components are involved. The first is the phenomenon of hyperphagia, or ‘eating to excess’ for the purpose of laying down the fat that provides the fuel for the migratory flight. This carries with it the concomitant cost of increasing the bird's mass and increasing energy for flight, although a supply of on-board fuel is vital for many birds, such as those that cross large ocean or desert expanses, where refueling is not an option. Second, birds have evolved an innate, genetically determined period of Zugunruhe (Berthold & Querner, 1981), or migratory restlessness. More colloquially put, they are ‘fidgety’ for a specified period during the migratory season. Captive migrants exhibit this behaviour in the evening hours of the nights in which they would normally be undertaking a migratory flight. The number of days this behaviour is expressed usually correlates strongly and positively with the length of the migratory journey.

Third, an innate navigation system is required for a successful migratory round trip. True navigation requires a map and a compass. Despite years of study, the avian map system is poorly understood. Much better known are the compasses that birds use. Birds can use a variety of navigational cues, including the pattern polarized light at crepuscular periods, the celestial pattern of stars at night, and the earth's magnetic field (Able & Able, 1995). Additional cues have also been implicated including sight and smell.

Additional factors are likely involved in the evolution of a long-distance migrant, such as changes in wing shape and the addition of an additional molt(s). It is also assumed that successful migrants lack neophobia (Mettke-Hofmann & Greenberg, 2005). Lastly, a long migratory journey can require adjustment of the annual cycle, and possibly altering the timing of major life cycle events. For example, a species might need to shorten or move forward breeding so that young have the time to develop and lay down fat before a long migratory journey; late hatched young would be selected against because they could not complete the journey or would arrive too late to find a winter territory (Norris et al., 2004).

These components of the migratory programme are genetically based, integrated and likely genetically correlated. For example, it has been shown that hybrids between blackcaps migrating south-east and south-west migrate to the south (Helbig, 1991). Furthermore, Berthold et al. (1992) were able to change a population of partially migrant blackcaps to either fully migratory or sedentary within seven generations with selective crosses of parentals indicating genetic control. The duration of Zugunruhe in hybrid redstarts is approximately half as long as between migratory and sedentary parental populations, which is a sign of relatively simple genetic control (Berthold, 2003). It is also fairly certain that these components of migration are co-opted from normal avian behaviours. Many female birds eat far in excess of maintenance requirements when preparing to lay eggs, providing a precursor for hyperphagia. Many birds can find their way home even if displaced, or in the case of seabirds, which undertake long feeding forays during the nesting season. Thus, much is known about the individual components of migration and their potentially independent genetic control.

Avian migration as an integrated system of switches?

One way of examining the evolutionary modification of migration is to consider a simple model in which hyperphagia, Zugunruhe, and migratory orientation function as a series of ‘switches’, each responding to a given threshold for some set of environmental characteristics. A sedentary species can differ from a fully migratory species by having one or more switches turned ‘off’. This assumes that existing environmental thresholds are insufficient to trigger the genetically programmed response, such as hyperphagia, that is likely inherent in all bird genomes, allowing the potential for rapid gains and losses of migratory behaviour. Furthermore, it makes a prediction that some sedentary species, those such as in the genus Basileuterus, where multiple species are all sedentary (Lovette et al., 2010), might require considerable tinkering to become migratory if all of the switches are currently off. That is, although it is possible to make any bird migratory (excepting perhaps flightless birds), it might require greater time, or a greater threshold, for some sedentary species to re-express migratory behaviour. Testing this hypothesis is hampered owing to a lack of knowledge about the existence of behaviours such as Zugunruhe in sedentary birds.

An example is provided by evolutionary and ecological studies of stonechats (Saxicola). One of the few reports of the existence of a migratory component in a sedentary bird is the occurrence of Zugunruhe in stonechats (Saxicola torquata) in Africa. Helm & Gwinner (2006) documented persistent periods of Zugunruhe in captive stonechats taken from sedentary populations. By itself, this observation is interesting but, when put in a phylogenetic context, it reveals a way of dissecting the components of migration in an evolutionary framework. Zink et al. (2009) provided a phylogeny of many of the stonechat lineages, including the migratory and sedentary populations studied by Helm & Gwinner (2006). An abbreviated version of this tree (Fig. 5) shows that the basal condition of stonechats is migratory with full expression of Zugunruhe. In this light, the expression of Zugunruhe in sedentary stonechats is an ancestral retention, and a potential explanation for why this group is no longer migratory; Zugunruhe has been suppressed but not eliminated. An implication is that if an environmental threshold is exceeded in the future, migratory behaviour could be re-established by turning the switch for Zugunruhe to ‘on’ (assuming that the programmes for hyperphagia and a compass were still intact). A future goal ought to comprise an in-depth examination of these individual components in migratory and sedentary taxa in a phylogenetic framework. The working hypothesis might be that there is a relationship between the evolutionary distance of a sedentary species to its nearest migratory ancestor and the number of switches that are turned off or suppressed. Nevertheless, description of the components of migration in sedentary species and subsequent phylogenetic mapping of them could greatly advance our understanding of how migration is maintained and modified over evolutionary time. Eventually study of gene regulation will provide a test of the presence of migratory components in sedentary species (e.g. Mueller et al., 2011).

Figure 5.

Condensed phylogenetic tree from Zink et al. (2009) for taxa in the Saxicola torquata complex. The boxes represent (left to right) hyperphagia, Zugunruhe, and navigation system, with a “+” indicating “on” and a “−” indicating “off”. The sedentary African population was shown by Helm & Gwinner (2006) to express low levels of Zugunruhe, although apparently insufficient for migratory behaviour to be expressed. The loss of this one ‘switch’ has most likely led to the suppression of migration in this lineage and, conversely, could be turned ‘on’ if the appropriate environmental threshold was exceeded.

Revisiting theevolutionof migration

The present review has considered studies in which the authors concluded that migration has evolved independently multiple times in a lineage. What exactly does this mean? There are at least two interpretations. One is that migration did indeed evolve de novo in separate lineages. Alternatively, migration might be latent in avian genomes and turned on or off (in different ways) depending on whether a critical environmental threshold is reached (Pulido, Berthold & van Noordwijk, 1996). It is useful to distinguish the evolutionary origin of a feature from its maintenance and modification (Mickevich & Weller, 1990; Brooks & McLennan, 1991). Figure 6 shows a simple tree and a character that has an evolutionary origin (i.e. it arises as a novelty) and its subsequent maintenance and modification (i.e. changes in state, although still recognized as the same novelty). For simple characters, this framework reveals the phylogenetic location of a traits' first appearance that we equate with its evolutionary origin (i.e. the feature evolved in that lineage). In the case of migration, it would appear that because almost all Neognath bird families have at least one migratory lineage, migration is a ubiquitous solution to exploiting seasonal resources. That is, it is likely that ‘migration’per se is an ancestral feature near the base of the avian tree of life. Thus, when studies refer to the evolution of migration in a species or group of species, they clearly do not mean its first occurrence, rather they are referring to its maintenance and modification (Zink, 2002; Rappole, Helm & Ramos, 2003; Bell, 2005). It is suggested that it is important to use terminology that indicates correct meaning and, although it is likely that many studies do not equate ‘migration evolved multiple times’ with the notion of truly independent origins, it is best to describe these as modifications of migration (which can include gain/loss of migration).

Figure 6.

Hypothetical tree showing the origin, maintenance, and modification of a feature such as migration. Some studies confuse these terms and speak of the ‘evolution’ of migration when, in fact, they are referring to its modification.

CONCLUSIONS

The study of the evolution of avian migration has clearly entered a hypothesis testing era. However, the mapping of migration on phylogenetic hypotheses requires several caveats owing to the apparent ease with which migratory behaviour can appear and disappear over extremely short periods of time. It is possible, in many taxa, that the expression of migratory or sedentary behaviour has changed multiple times since the species last shared a common ancestor. If we lived today at a glacial maximum, it is suspected that many currently migratory species would be sedentary. This would change our understanding of the evolutionary transitions between migratory and sedentary behaviour on a phylogenetic hypothesis. Therefore, studies should attempt to estimate the frequency of transitions from one migratory state to another and use this information to map the evolutionary history of migratory behaviour. The use of SDMs can help determine whether environmental conditions that favor migration today were present at the LGM. If transitions between migratory and sedentary states occur multiple times on a given branch, it will likely mean that ancestral state reconstruction is not possible, which in itself would be an interesting conclusion. On the other hand, in lineages with a low rate of changes between migratory and sedentary, there is potential that mapping ancestral states will be more useful in testing hypotheses about the evolution of migration.

Migration per se can be an overly broad umbrella for a complex set of behavioural and physiological traits. To make progress in our understanding of the maintenance and modification of migration, this needs to be examined phylogenetically at the level of its basic components, such as hyperphagia, Zugunruhe, and the navigational system (which themselves might be overly broad). These components can be altered rapidly, which provides an explanation as to why the expression of migration per se is labile over both ecological and evolutionary time. For example, suppression of migratory behaviour could be accomplished by diminished Zugunruhe, as in some African stonechats.

Even given a non-monophyletic pattern of migratory behaviour on a phylogenetic tree, it is best to avoid the inference that migration evolved multiple times. This is because it is unlikely that the components of migration were ‘invented’ multiple times. It is more likely that the expression of migratory behaviour is turned on or off, and this qualifies instead as a modification of the general migratory template that is apparently innate in most birds.

ACKNOWLEDGEMENTS

I thank A. T. Peterson, H. Vazquez-Miranda, and K. Kozak for their assistance with niche modeling. S. Lanyon, B. McKay, C.-M. Hung, and three anonymous reviewers offered useful comments on the manuscript. The BBS staff assisted with downloading locality records. K. Dale from Audubon Science provided Christmas Bird Count data for chipping sparrow. I thank C. Finlayson for prompting me to think about the topic, as well as J. Rappole, R. Greenberg, D. Irwin, and the rest of the participants at the 2010 Calpe Conference ‘Pleistocene Ecological Change and the Evolution of Bird Migration Systems’ for stimulating discussion.

Ancillary