In contrast to placentals, marsupials are born with forelimbs that are greatly developmentally advanced relative to their hind limbs. Despite significant interest, we still do not know why this is the case, or how this difference is achieved developmentally. Studies of prechondrogenic and chondrogenic limbs have supported the traditional hypothesis that marsupial forelimb development is accelerated in response to the functional requirements of the newborn's crawl to the teat. However, limb ossification studies have concluded that, rather than the forelimb being accelerated, hind limb development is delayed. By increasing the taxonomic coverage and number of prechondrogenic events relative to previous studies, and combining traditional phylogenetic analyses of event sequences with novel analyses of relative developmental rates, this study demonstrates that the timing of limb development in marsupials is more complex than commonly thought. The marsupial phenotype was derived through two independent evolutionary changes in developmental rate: (1) an acceleration of the forelimb's first appearance and (2) a delay of hind limb development from the bud stage onward. Surprisingly, this study also provides some support for an evolutionary acceleration of the marsupial hind limb's first appearance. Further study is needed on the developmental and genetic mechanisms driving these major evolutionary transitions.
Marsupials (e.g., kangaroos, opossums and their kin), in contrast to placentals (e.g., humans, dogs, bats, whales), give birth after extremely short gestation times to immature neonates that must immediately complete a life-or-death crawl to the teat where they attach and continue to develop (Gemmell et al. 2002). Marsupial newborns complete this crawl solely under the power of their massively developed forelimbs, whereas their less-developed hind limbs hang passively from their bodies. Previous research has hypothesized that the reduced morphologic, and potentially as a result, taxonomic diversification of marsupials relative to placental mammals was a result of functional constraints on their limb development caused by their unique mode of reproduction (Sears 2004). As such, understanding how the marsupial mode of reproduction has influenced the development of their limbs is important to our interpretation of the mechanisms shaping mammalian evolutionary history.
Traditionally, it has been assumed that the development of the marsupial forelimb was accelerated relative to the ancestral mammalian condition to perform the crawl to the teat (e.g., Lillegraven 1975; Tyndale-Biscoe and Renfree 1987; Sánchez-Villagra 2002; Bininda-Emonds et al. 2003). Analyses of a limited subset of prechondrogenic and chondrogenic events (e.g., limb bud formation, AER formation, and digital plate crenation) in both New World (Bininda-Emonds et al. 2003) or Old World (Bininda-Emonds et al. 2007) marsupials found support for this hypothesis, and no significant changes in the marsupial hind limb. However, other recent analyses of later developmental events (e.g., ossification) strikingly found no evidence of forelimb acceleration, but rather evidence that the development of the hind limb is delayed (Harrison and Larsson 2008; Weisbecker et al. 2008). In all of these studies, evolutionary changes in the sequence of developmental events were estimated using phylogenetic-reconstruction techniques [using programs such as Parsimov (Jeffery et al. 2005) and Parsimov-based, genetic inference (PGi, Harrison and Larsson 2008)], whereas changes in relative timing (i.e., the proportion of absolute time at which different developmental events occur) were not considered.
Thus, the question remains; how are the large differences in developmental maturity at birth of marsupial forelimb and hind limbs achieved? I hypothesize that the timing of marsupial limb development is more complicated than commonly suggested, and that at least two evolutionary events have occurred: (1) the earliest development (e.g., first appearance) of the forelimb has been accelerated relative to the ancestral condition, (2) at some point later in development, but before birth, the hind limb has been delayed. In their discussion, Bininda-Emonds et al. (2007) acknowledged that their analysis of only the order of events in forelimb and hind limb development could belie its actual complexity (also see Richardson et al., in press).
To test these hypotheses, I take two different approaches to study developmental timing. First, I analyze the evolution of the sequence of early (i.e., prechondrogenic) limb developmental events among amniote species using a phylogenetic approach. In this analysis, I simultaneously include representatives of both New and Old World marsupials, in contrast to previous studies of prechondrogenic development that focused on one group or the other, and use twice as many (six in total) prechondrogenic limb developmental events as previous studies. Although phylogenetic analyses of developmental sequences can identify evolutionary changes in the relative sequence of developmental events, they cannot detect finer scale changes in absolute timing that do not result in changes to the sequence, nor changes that shift all events in the same direction. Therefore, I also take a novel second approach in which I compare the rescaled absolute (i.e., relative) rates of prechondrogenic forelimb and hind limb development in several representative marsupial and placental taxa. Through this novel approach, I am able to pick up previously unrealized complexities in marsupial limb development.
Materials and Methods
Relative and absolute timing data for several prechondrogenic developmental events were gathered from the literature for several species, and from a combination of the literature and personal observation of embryos for three additional species: Carollia perscpicillata (56 specimens for nine stages personally observed), Sus scrofa (288 specimens for 14 stages personally observed), and Monodelphis domestica (127 specimens for 13 stages personally observed) (see Table 1 for literature sources and Table 2 for a list of events). Selected events include important transitions in both limb (e.g., formation of the limb buds) and nonlimb (e.g., closure of the anterior neuropore) development. Limb events were selected to capture the complete range of prechondrogenic limb development—from first appearance of the limb to the final emergence of the digits. Nonlimb events that have previously been demonstrated to have similar relative developmental timing in marsupials and placentals (e.g., Bininda-Emonds et al. 2003) were included to provide a reference against which limb event timing could be compared.
Table 1. Included taxa. Species indicated with * were included in the phylogenetic analyses, and species indicated with # were excluded from the rate analyses. Pers. obs. = personal observation.
Table 2. Developmental events. While all events are included in phylogenetic analyses of event sequences, only limb events are included in analyses of relative timing.
Limb thickening (ridge) appears
First thickening of presumptive limb tissue to form a ridge (limb wider than long)
Limb bud forms
First appearance of a limb bud (limb longer than wide)
First appearance of club-like enlargement of distal part of limb
First appearance of mesenchymal digit condensations
Interdigital tissue begins to recede
Initial recession of the interdigital tissues
Digits are completely separate
Complete recession of the interdigital tissues, digits completely separate
Closure of anterior neuropore
Anterior neuropore completely closed
Formation of 10 somites
10 somites present
Formation of 15 somites
15 somites present
Formation of 25 somites
25 somites present
Formation of tail bud
First appearance of tail bud
Information on the timing of the selected developmental events was compiled for marsupials, placentals, and nonmammalian taxa (see Table 1). All mammals for which data could be found in the literature were included. Six orders of placentals are represented in this sample, as well as one order each of both New and Old world marsupials (two marsupial orders total). The nonmammalian amniotes were used as an outgroup in this study. With the exception of the two Old World marsupials (Antechinus and Sminthopsis), information for every developmental event was included for all taxa (see Appendix S1 for data included in this study). Data on somite formation were unavailable for the Old World marsupials, and hence were excluded from this study.
COMPARISONS OF THE SEQUENCE OF DEVELOPMENTAL EVENTS: PHYLOGENETIC ANALYSES
Parsimov-based, genetic inference (PGi) was used to reconstruct evolutionary changes in the relative timing of developmental events (Harrison and Larsson 2008). The developmental events described above were ordered into a temporal sequence from first to last for each taxon. These sequences were then compiled and loaded into the PGi program along with a well-supported phylogenetic tree (for phylogenetic hypothesis see Cao et al. 2000; Iwabe et al. 2004; Springer et al. 2004; Hugall et al. 2007) (see Fig. 1). Using these data, the PGi program inferred ancestral temporal sequences, quantified sequence heterochronies (i.e., relative accelerations and decelerations), and mapped them to specific branches on the phylogeny. PGi works by determining the minimum number of sequence heterochronies needed to explain the differences between an ancestral and descendent developmental sequence, using a slightly modified version of the Parsimov algorithm (Jeffery et al. 2005). However, PGi has several advantages over Parsimov (see Harrison and Larsson 2008 for a complete description of PGi). Notably, PGi treats the temporal sequence directly as a single complex character, rather than relying on the “event-pairing” method used by Parsimov, which is problematic as it treats dependent events as if they are independent (see Schulmeister and Wheeler 2004).
As mentioned previously, information on somite development was not available for Old World marsupials. Although the PGi program does not explicitly allow missing data, it is possible to enter the missing data as “unscorable.” PGi then treats missing data as if they were lost evolutionarily. Because this simplification is not biologically accurate in this case (somites are conserved across vertebrates), two separate analyses were run. The first included Old World marsupials (e.g., Antechinus and Sminthopsis), with their somite events coded as “unscorable,” and the second excluded Old World marsupials. See Table 1 for a complete list of taxa included in the PGi analyses. For both analyses PGi was run for 100 cycles with 100 iterations per cycle, with a heuristic genetic algorithm. Also, ancestral sequences were allowed to have simultaneity, and were evaluated using the edit cost function of the Parsimov model with a greedy heuristic. If the development of the marsupial forelimb has been evolutionarily accelerated, or the hind limb delayed, then the phylogenetic analysis should recover these transitions. All statistical analyses were performed in JMP 7.0 (SAS; Cary, NC).
COMPARISONS OF RELATIVE TIMING: RATE ANALYSES
To detect finer scale changes in relative timing that do not result in changes to the developmental sequence, and provide another method of examining differences in timing, the rescaled absolute timing of limb developmental events was compared between marsupial and placental mammals (see below for details). The species excluded from these absolute rate analyses are indicated in Table 1.
Gestation length varies greatly among examined mammals (ranging from ∼10.5 days in Sminthopsis through ∼270 days in Homo), so differences in absolute timing of development are not directly comparable. To facilitate comparison, the absolute time (in days) separating the first developmental event (e.g., the first thickening) and the last (e.g., beginning of the recession of the interdigital tissue) was scaled to 1.0 for each species individually. Then, the absolute time at which the rest of the limb developmental events occurred could be expressed as the proportion of this total limb development time. The sixth developmental event, complete separation of the digits, was excluded from this analysis as data for this event were unavailable for most marsupial species. The scaled timing results for each limb developmental event were averaged separately for each limb, and for each clade (i.e., marsupials and placentals). These averages were plotted graphically to allow a visual comparison of the relative rates of forelimb and hind limb development within marsupials and placentals. The ranges of values for a few developmental events for marsupials and placentals were also compared statistically using the Mann–Whitney test. If the initial development of the marsupial forelimb is accelerated relative to the hind limb, or the initial development of the hind limb is delayed, then the expected result is that the time between the first forelimb and hind limb events will be significantly greater in marsupials than in placentals. However, it is also possible that the observable differences in marsupial limb development at birth are due to differences in the rates of forelimb and hind limb development. If this is the case, the expected result is that the developmental rates will be different in marsupial forelimb and hind limbs. The relative timing of placental events is predicted to stay relatively constant between limbs (i.e., the difference in timing between the forelimb and hind limb will stay the same through development).
In a second analysis of relative developmental rates, duration of forelimb development (in days, from initial formation of the limb ridge [Event 1] to complete separation of the digits [Event 6]) was divided by duration of hind limb development. A resulting value of 1.0 would indicate that the limbs develop in the same amount of time, and at the same rate. Values greater than or less than 1.0 indicate that the forelimb develops in more or less time than the hind limb, respectively. Values for marsupials, placentals, and the outgroup taxa were statistically compared using the Kruskal–Wallis test. Note that this method cannot distinguish whether the time it takes to build the forelimb is relatively reduced, or that of the hind limb increased relatively. Therefore, if either the development of the marsupial forelimbs is accelerated, or the development of the hind limbs delayed, I expect to find marsupial values to be less than 1. If the developmental durations of the marsupial limbs have been evolutionarily modified, I also expect the marsupial values to be statistically lower than those of placentals and the outgroup. In contrast to marsupials, I expect that placentals and the outgroup taxa will have values of around 1.0 in accordance with the relative synchronicity of their limb development.
Results and Discussion
COMPARISONS OF THE SEQUENCE OF DEVELOPMENTAL EVENTS: PHYLOGENETIC ANALYSES
Results of the two phylogenetic analyses (the first with Old World marsupials included and the second with them excluded) are largely congruent (Fig. 1). Within mammals, the branch leading to marsupials is characterized by the most heterochronic events: four when all taxa are included, and five when Old World marsupials are excluded. Both analyses indicate that the formation of the forelimb ridge and bud is accelerated along the branch leading to marsupials, whereas the condensation of the hind limb digits and the recession of the hind limb interdigital tissue are delayed. These findings are consistent with previous phylogenetic studies focusing on prechondrogenic and chondrogenic limb development (e.g., Bininda-Emonds et al. 2003; Bininda-Emonds et al. 2007) and later limb ossification (e.g., Harrison and Larsson 2008; Weisbecker et al. 2008). Taken together, these results reaffirm that the evolutionary changes that have led to the divergent pattern of limb development in marsupials are more complex than a simple acceleration or delay of one entire limb.
When all taxa are included, the formation of the footplate of the hind limb is reconstructed as being accelerated along the branch leading to placentals. However, when Old World marsupials are excluded, the formation of the hind limb footplate is reconstructed as being delayed along the branch leading to marsupials. Taken together, these results indicate that there is an evolutionary shift in the developmental timing of footplate formation that leads to its relatively late development in marsupials. However, whether the footplate is shifted later in marsupials, or earlier in placentals remains ambiguous. Determining when and where this evolutionary shift occurred is important because if the formation of the footplate is delayed along the lineage leading to marsupials, then this would suggest that the delay in marsupial limb development begins at this stage, rather than at a later one. Knowing when the delay in marsupial hind limb development begins is crucial to future studies on the developmental and genetic mechanisms controlling these evolutionary shifts in limb timing. Therefore, this should be further investigated with the addition of more taxa as information about their developmental timing becomes available.
Most interestingly, when all taxa are included, the formation of the hind limb ridge is reconstructed as being accelerated along the branches leading to New World marsupials (Didelphis and Monodelphis) and to one of the two Old World marsupials, Antechinus. These results suggest that three of the four marsupials examined in this study display an acceleration of the earliest manifestation of hind limb development. However, when the Old World marsupials are excluded, this heterochrony is no longer apparent. From these results, it is possible to conclude that the inclusion of Old World marsupials does influence the result of the PGi analysis, and perhaps illuminates an additional heterochrony that is missed when these forms are excluded. However, because the results of the analyses are incongruent, this conclusion is best viewed as a hypothesis meriting further testing when information about developmental timing becomes available for additional marsupial species.
Both analyses recover few heterochronies within the placental clade, suggesting that the evolution of developmental sequences within placentals may be more conservative than within marsupials. On the branch leading to Carollia, the initial appearance of the forelimb (i.e., formation of the ridge) is accelerated and the formation of the footplate of the hind limb is delayed. Additionally, the initial appearance of the forelimb is reconstructed as either delayed along the branch leading to Oryctolagus (when all taxa are included) or as accelerated along the branch leading to the clade of Rattus and Mus (when Old World marsupials are excluded).
COMPARISONS OF RELATIVE TIMING: RATE ANALYSES
When scaled developmental time is plotted against the developmental events, it becomes clear that in all mammals the initial development of the forelimb is slightly ahead of that of the hind limb, and that each subsequent forelimb event occurs before the corresponding event in the hind limb (Fig. 2). However, the relative rates of forelimb and hind limb development differ in the two groups. In placentals, the average difference between the scaled timing of equivalent developmental events in the forelimb and hind limb is 0.14 for all events, with a range of 0.11 to 0.17 among events. This limited range indicates that the difference in timing between placental limbs remains almost constant over the course of development (Fig. 2A). This is in stark contrast to the situation in marsupials (Fig. 2B), in which the average difference in timing between the limbs is 0.34—over twice that of placentals. In further contrast to the placental pattern, the average timing difference between the marsupial limbs is not constant. During event 1 (formation of the limb ridge), the difference between the averaged and scaled timing of the marsupial forelimb and hind limb is 0.21. Note that this value falls outside the placental range. Furthermore, the differences in timing of the initial appearance of the forelimb and hind limb are significantly greater in marsupials than placentals (P= 0.014). This indicates that either the earliest development of the marsupial forelimb has been relatively accelerated, or that of the hind limb is relatively delayed. As there is evidence from the phylogenetic analysis that the earliest development of the forelimb has been accelerated, and none for a delay of the earliest development of the hind limb, it is likely that these results are due to the evolutionary acceleration of the earliest development of the forelimb relative to the hind limb in marsupials. During event 2 (formation of the limb bud), the average timing spread remains fairly constant, at 0.17 (range for forelimb values: 0.08–0.22, for hind limb: 0.20–0.48). From events 3 through 5, the average difference in timing increases from 0.40 (range for forelimb values: 0.08–0.22, for hind limb: 0.20–0.48), to 0.41 (range for forelimb values: 0.19–0.26, for hind limb: 0.55–0.75), to 0.49 (range for forelimb values: 0.30–0.45, for hind limb: 0.63–1.00), and for each of these stages the differences in timing between the limbs are significantly greater for marsupials than for placentals (P= 0.014, P= 0.014, P= 0.014, respectively). The large jump in the difference in marsupial limb timing between the formation of the limb bud (event 2) and the formation of the hand/footplate (event 3) suggests that at this point in development an evolutionary event has occurred in which either the development of the forelimb has been accelerated or the hind limb has been decelerated. Again, as the phylogenetic results provide evidence for a delay in later hind limb development, and none for an acceleration of later forelimb development, it is likely that this evolutionary event involved a relative decrease in the rate of hind limb development after the formation of the limb bud.
To further test these differences in developmental timing, I compared the ratio of the total duration (unscaled) of forelimb development to that for the hind limb, among marsupials, placentals, and the outgroup taxa. Placentals have an average ratio of 0.98, which suggests that the developmental rates in placental forelimb and hind limbs are almost identical. The average value for the outgroup, 1.07, is statistically indistinguishable from that of placentals. In contrast, marsupials had an average ratio of 0.61, confirming that the total time it takes for the forelimb to develop from initial formation through digit separation is only slightly more than half of what it takes for the hind limb. Moreover, the marsupial values are statistically distinguishable from both those of the placentals and the outgroup taxa (P= 0.024). It is impossible to determine from these data alone whether the lower value for marsupials is due to the development of the forelimb being accelerated relative to the hind limb, or the hind limb being relatively decelerated, as evidence exists for both.
Although it has traditionally been assumed that the development of marsupial forelimbs was accelerated as a result of the functional requirements of the newborn's crawl to the teat, recent research has emerged suggesting that the major change in the timing of marsupial limb development was actually a delay in the development of the hind limbs. To resolve this discrepancy, I undertook a study of the sequence and relative timing of prechondrogenic limb events in several representative marsupials, placentals, and nonmammalian amniotes.
The resulting study is the first to confirm that the development of both marsupial limbs has likely been influenced by the unique development of the group, and to provide a framework to integrate previous studies of the timing of mammalian limb development. At some point very early in marsupial history, before the divergence of the Old and New World clades, the earliest development of at least the forelimb was likely accelerated relative to the ancestral state, and the later development of the hind limb (likely from the formation of the bud) was delayed. Furthermore, there is some evidence that the earliest development of the marsupial hind limb is also accelerated, although this hypothesis needs further testing with additional information about developmental timing of more marsupial taxa. However, if this evidence is taken at face value, then it is possible to speculate that the unique development of marsupial limbs arose through two major changes in genetic patterning, one which accelerated the earliest development of both limbs, and then another which decelerated the rate of later hind limb development.
The existence of relatively large differences between the developmental timing of marsupial limbs, coupled with the fact that these differences likely evolved through at least two independent events, suggests that the development of marsupial forelimb and hind limbs may be less integrated than in placentals. As serially homologous structures, limbs are expected to be integrated, that is, they are expected to share a common genetic architecture and developmental network (Young and Hallgrimsson 2005). However, differential selective pressures on the limbs could greatly reduce these developmental and genetic links (Wagner 1996). As a result of the integral role of the forelimb in the crawl, marsupial forelimb and hind limbs have highly disparate functional requirements early in development, which result in differential selection on the limbs. I hypothesize that this early differential selection on marsupial limbs resulted in the divergence of their genetic and developmental networks, and, consequently, reduced their integration. Recent studies have confirmed that marsupials display reduced integration between their forelimb and hind limbs relative to placentals (Goswami et al., in press). Reduced integration is generally thought to lead to increased evolvability (Wagner 1996), but in the case of marsupial limbs I propose that the situation is more complicated. I hypothesize that the increased modularity of the marsupial limbs facilitated the specialization of the forelimb for the crawl, but doing so effectively canalized its development, reduced its evolvability, and lead to the previously identified constraint on marsupial evolution (Sears 2004).
Future study on the genetic and developmental mechanisms driving the multiple evolutionary transitions in developmental timing identified by this study will help in testing these outstanding hypotheses, and potentially reveal the causes of the marsupial-placental dichotomy.
Associate Editor: B. Swalla
I thank J. Marcot and A. Goswami for idea discussion, and J. Marcot for reviewing early drafts of this manuscript. I also thank C. Doroba and A. Rockwell for helping in initial data collection.