Michael F. Braby, Biodiversity Conservation Division, Department of Natural Resources, Environment and the Arts, PO Box 496, Palmerston NT 0831, Australia. Tel.: +618-894-48488; fax: +618-894-48455; e-mail: email@example.com
Butterflies in the family Pieridae (Lepidoptera: Papilionoidea) feed as larvae on plants belonging primarily to three distantly related angiosperm orders: Fabales (legumes and allied plants), Brassicales (crucifers and related plants containing mustard oil glucosides), and Santalales (‘mistletoes’). However, some utilize plants from 13 other families in a further eight orders. We investigated the evolutionary history of host plant use of the Pieridae in the context of a recent phylogenetic hypothesis of the family, using simple character optimization. Although there is a close association between host plant and butterfly higher classification, we find no evidence for cospeciation but a pattern of repeated colonization and specialization. The ancestral host of the family appears to be Fabaceae or Fabales, with multiple independent shifts to other orders, including three to Santalales. The shift to Brassicales, which contain secondary compounds (glucosinolates), promoted diversification and adaptive radiation within the subfamily Pierinae. Subsequent shifts from crucifers to mistletoes (aerial-stem hemiparasites) facilitated further diversification, and more recent shifts from mistletoes to mistletoe host trees led to exploitation of novel host plants outside the conventional three orders. Possible mechanisms underlying these host shifts are briefly discussed. In the Pierinae, a striking association between host plant, larval and adult behaviour, adult phenotype, and mimicry calls for further research into possible relationships between host specialization, plant chemistry and butterfly palatability.
The Lepidoptera are one of the largest extant groups of phytophagous insects in terms of species richness (Kristensen & Skalski, 1999), but the processes and mechanisms driving this diversity are little understood. The larvae exploit a wide variety of seed plants (gymnosperms and angiosperms) (Powell, 1980; Powell et al., 1999), although the butterflies are associated primarily with flowering plants (Ackery et al., 1999). A key question in insect-plant interactions is whether butterflies and flowering plants evolved and radiated together, or butterflies evolved some time after or even before the evolution and radiation of flowering plants? In the former scenario, a long historical association between butterflies and their host plants would be an example of cospeciation (parallel cladogenesis or parallel diversification) of the two groups, possibly associated with stepwise coevolution (Ehrlich & Raven, 1965). That is, the evolutionary history of butterflies should be broadly congruent with the phylogeny of their host plants in terms of similarities in tree topology (i.e. branching pattern and branch lengths) in both time and space. In the two other scenarios, lack of congruence between the evolutionary histories of butterflies and their host plants would preclude coevolution. If butterflies evolved after the origin of their host plants, a pattern of sequential colonization of distantly related hosts (i.e. shifts to novel hosts) and/or repeated shifts to ancestral or related hosts, would be expected (Jermy, 1984; Janz & Nylin, 1998).
In this study, we investigate evolutionary patterns of larval host plant associations at two systematic levels in the family Pieridae. The Pieridae, like other groups of butterflies, are ideal model organisms for testing theories of origin of ecological traits (Boggs et al., 2003). The family, which consists of more than 1100 species arranged in 83 genera, feed as larvae on the foliage of a range of plants (Ehrlich & Raven, 1965; Vane-Wright, 1978; Courtney, 1986; Ackery, 1991; Janz & Nylin, 1998). However, a substantial impediment to host plant analyses in these otherwize well-studied insects as a model-group has been the lack of a robust phylogeny. This situation is slowly changing with the advent of molecular systematics (Wahlberg et al., 2005) and, in the Pieridae, detailed phylogenetic hypotheses at different systematic levels are now emerging (Braby et al., 2006a,b). These phylogenies allow us to examine variation in ecological and life history traits from an evolutionary perspective.
In this paper, we investigate phylogenetic patterns between pierid butterflies and their larval host plants, and attempt to answer the following questions: (1) what was the ancestral host plant of the Pieridae? (2) Do pierids feed on closely related groups of plants at different systematic levels and, if so, is there evidence of cospeciation between these butterflies and their hosts? (3) To what extent have host shifts to phylogenetically distant plants occurred? That is, is a pattern of host colonization evident? And (4) if such host shifts have occurred, have these promoted diversification and adaptive radiation, and by what potential mechanisms? We also explore a possible relationship between host specialization and plant chemistry in the subfamily Pierinae in which we hypothesize that mistletoe feeding has conferred unpalatability, aposematism and mimicry in the adult stage, whereas feeding on crucifers and mistletoe host trees has generally not conferred these attributes.
Pierid butterfly phylogenies
Larval host plant associations were examined at two systematic levels within the Pieridae: family and subtribe. For the family-level analysis, the recent phylogenetic hypothesis and higher classification of the Pieridae by Braby et al. (2006b) was used. This phylogeny estimate, which included 80 lower taxa (74 genera and six subgenera) or 89% of all genera recognized in the family (Braby, 2005b), was based on an analysis of four genes in combination (EF-1α, Wingless, COI, 28S) using all available data. For the subtribal-level analysis, the phylogenetic hypothesis of the Aporiina within the Pierinae: Pierini was used (Braby et al., 2006a). The phylogeny estimate, which included all 14 genera recognized in the subtribe, was based on a combined analysis of three genes (EF-1α, Wingless, COI).
The Brassicales, formerly known as the Capparales, contains more than 4300 species (Magallón & Sanderson, 2001) and includes many families (c. 15) that traditionally were classified in several distantly related orders. These plants contain mustard oil glucosides and they comprise a monophyletic group (Rodman, 1991; Rodman et al., 1993, 1998). The family Capparaceae comprises a polyphyletic group (Rodman et al., 1993, 1998) and was not recognized by APG II (2003), with most of its taxa now subsumed under the Brassicaceae.
The Santalales comprise a well-supported monophyletic group (Nickrent & Franchina, 1990; Nickrent & Soltis, 1995; Nickrent et al., 1998) of over 2100 species (Magallón & Sanderson, 2001) in the families Olacaceae, Misodendraceae, Loranthaceae, Opiliaceae, Santalaceae and Viscaceae. These plants are either aerial-stem or root hemiparasites, both of which we have designated as ‘mistletoes’ in the broad sense. The family Eremolepidaceae, which includes the genus Antidaphne, an important host plant for pierids in the Catasticta group (Braby & Nishida, Unpublished data), is generally now considered to belong in the Santalaceae (Wiens & Barlow, 1971; Nickrent et al., 1998; APG II, 2003). In Nickrent et al. (1998) phylogeny, monophyly of most traditional families was well supported (bootstrap 100%), including the Loranthaceae, Opiliaceae and Viscaceae. The latter family, however, was nested within the Santalaceae, rendering the Santalaceae nonmonophyletic. APG II (2003) therefore proposed that the Viscaceae be synonymized with the Santalaceae s.l., although supporting evidence for paraphyly of the Santalaceae s.l. was rather weak (bootstrap 66%).
The Fabales, in the broad sense, are monophyletic (Doyle et al., 2000; Kajita et al., 2001; Wojciechowski et al., 2004), and include about 19 000 species (Magallón & Sanderson, 2001) arranged in only four families, three of which are relatively small, with the Mimosaceae and Caesalpiniaceae now subsumed under the Fabaceae s.l. (APG II, 2003). The monophyletic Mimosaceae are now regarded as a subfamily of mimosoid legumes (i.e. Mimosoideae), whereas the Caesalpiniaceae, a paraphyletic entity, are no longer recognized.
Butterfly-host plant data
Larval host plant data were recorded at the level of plant family. As pointed out by Wahlberg (2001) and others, when considering larval host plant data as a character it is important to recognize the taxonomic level of the plant, especially given that historical associations between plants and butterflies probably date before the Tertiary (Janz & Nylin, 1998). The level of plant species or even genus is too recent and therefore too labile for most butterfly genera and species, let alone higher taxa (subfamilies, tribes, etc).
Patterns of larval host plant association were mapped on the phylogenetic trees of pierid butterflies by simple character optimization. Character states were optimized using macclade version 3.08a (Maddison & Maddison, 1999) in order to reconstruct ancestral nodes and establish the most parsimonious sequence of butterfly-host plant associations. In all analyses, multiple butterfly host plant associations were treated as polymorphic states.
For the family-level analysis, the host plant data (‘major’ and ‘minor’ records only) were coded as an unordered multistate character at the level of plant order, with four states: Brassicales, Santalales, Fabales, and ‘Others’. In our analysis, state ‘Brassicales’ included the families Bataceae, Brassicaceae, Bretschneideraceae, Resedaceae, Salvadoraceae and Tropaeoleaceae; state ‘Santalales’ included the families Olacaceae, Loranthaceae, Opiliaceae, Santalaceae and Viscaceae; while state ‘Fabales’ included just the single large cosmopolitan family Fabaceae. ‘Others’ included several distantly related families (Asteraceae, Berberidaceae, Ericaceae, Rhamnaceae, Zygophyllaceae and Pinaceae) for which several butterfly genera were monomorphic, that is, the larvae fed exclusively on one family of plants. For the subtribal-level of analysis, larval host plant records were also considered at the level of order, and coded as a multistate character with six states: Brassicales, Santalales, Ranunculales, Ericales and Pinaceae (Gymnospermae) and ‘Others’. The state ‘Ranunculales’ included only the family Berberidaceae, while state ‘Ericales’ included the family Ericaceae (APG II, 2003); ‘Others’ included the distantly related families Polygonaceae, Rosaceae and Elaeagnaceae.
Larval host plant data compiled for all higher taxa of Pieridae are summarized in Appendix S1. These data show that pierids feed on plants belonging primarily to three angiosperm orders: (1) Brassicales; (2) Santalales; (3) Fabales. In terms of the higher classification of the Pieridae, there was a close association between butterflies and these three major groups of plants. The Pseudopontiinae feed on root mistletoes, Pentarhopalopilia (Opiliaceae) (Heath, 1977), the Dismorphiinae and Coliadinae feed predominantly on legumes (Fabaceae), while the Pierinae feed mainly on crucifers and allied plants (Brassicales), and to a lesser extent on mistletoes (Santalales).
Figure 1 summarizes the phylogenetic relationships of the Pieridae in terms of the higher taxa recognized by Braby et al. (2006b). Tracing the broad larval host plant character states at the ordinal level, this phylogeny suggests that legumes were the ancestral host plant of the family, whereas crucifers and mistletoes were both derived states. Crucifer feeding evolved at least once from a legume-feeding ancestor relatively early in pierid evolution, whereas mistletoe-feeding evolved independently three times: (1) in Afrotropical Pseudopontia (Pseudopontiinae) from a legume-feeding ancestor; (2) in the Neotropical Hesperocharis group (Pierinae: Anthocharidini) from a crucifer-feeding ancestor; (3) in the clade Mylothris + (Aporia + Delias group + Catasticta group) (Pierini: Aporiina), also from a crucifer-feeding ancestor.
Within the Coliadinae and Aporiina, there was evidence of further host changes, as revealed by the presence of polymorphic states among these taxa (Fig. 1). In the Coliadinae, for instance, there were at least four independent changes from Fabales (Fabaceae) to: (1) Asterales (Asteraceae), in Nathalis; (2) Zygophyllales (Zygophyllaceae), in Kricogonia; (3–4) Rosales (Rhamnaceae), in both Gandaca and Gonepteryx (since these genera do not comprise a monophyletic group two independent changes are required). Each such change was associated with a monomorphic genus, that is, the butterflies were monophagous on these plant families.
These general patterns, however, underscore the full extent of host plant changes in the Pieridae. Inspection of the larval host plant data (Appendix S1) in the context of Fig. 1 indicates these few relatively ancient host changes have been overlaid by numerous more recent changes. For example, in the Coliadinae, the genera Eurema s.l. and Colias are recorded secondarily from Malpighiales (Clusiaceae and Euphorbiaceae) and Rosales (Rhamnaceae), and from Malpighiales (Salicaceae) and Ericales (Ericaceae), respectively. Among the predominantly Brassicales-feeding clades there have been secondary changes to Celastrales (Celastraceae) and Malpighiales (Rhizophoraceae) in the Colotis group (Nepheronia), to Malpighiales (Euphorbiaceae) in the Appiadina (Appias) and even back to Fabales (Fabaceae) in the Pierina (Tatochila and Hypsochila).
The Aporiina feed on a diverse array of larval host plants and tracing the character states revealed several interesting patterns (Fig. 2). As previously observed, Santalales-feeding was a derived state, having evolved once from a Brassicales-feeding ancestor relatively early in the evolution of the crown-group. Three closely related clades of pierids specialize on mistletoes: (1) Mylothris; (2) the Delias group; (3) the Catasticta group. Larvae of Afrotropical Mylothris feed primarily on aerial-stem hemiparasitic shrubs in the Loranthaceae (65–80% of all records, if each butterfly species is treated independently), but also to a lesser extent on related families in the Santalaceae, Viscaceae and Olacaceae (Braby, 2005a). Larvae of the Indo-Australian genus Delias also feed primarily on aerial-stem hemiparasites in the Loranthaceae (78% of all records), and to a lesser extent on the Santalaceae and Viscaceae (Braby, Unpublished data). Most of the genera of Neotropical Catasticta group for which the life histories are known feed on aerial-stem hemiparasites, either in the Santalaceae, Viscaceae or Loranthaceae (Braby & Nishida, unpublished data). Of the other families in the Santalales, there are no Aporiina host plant records from Opiliaceae or Misodendraceae, and Olacaceae-feeding appears to be very rare in this subtribe, being restricted to a single species in Africa, M. agathina (Cramer), which also feeds on Loranthaceae and Santalaceae (Braby, 2005a).
Within the Santalales-feeding clade, there was evidence of further host changes at the lower taxonomic levels, with three butterfly genera specialising on a range of nonmistletoes (Fig. 2). In the Palaearctic Aporia, of the few species for which the life histories have been recorded, larvae of two subgenera feed primarily on Ranunculales (Berberidaceae). In the Catasticta group, two Nearctic genera specialize on woody trees: Neophasia feeds exclusively on a range of conifers (Pinaceae) (Howe, 1975; Scott, 1986; Layberry et al., 1998), while Eucheira feeds specifically on Ericales (Ericaceae) in the genus Arbutus (Kevan & Bye, 1991; Underwood, 1994).
Polymorphic character states in two genera revealed at least two recent secondary changes to additional plant families (Fig. 2). Mylothris showed a change from Santalales to Caryophyllales (Polygonaceae), with two closely related species in the chloris group feeding exclusively on Persicaria (Braby, 2005a), while Aporia showed a change from Ranunculales to Rosales, with several species feeding on plants in the Rosaceae and Elaeagnaceae (Appendix S1).
Our review of the literature indicates that pierid butterflies are, in general, host specific, exploiting a rather conservative array of plants primarily in the angiosperm orders Fabales (i.e. legumes and allied plants), Brassicales (i.e. crucifers and allied plants containing mustard oil glucosides) and Santalales (i.e. ‘mistletoes’, including both aerial-stem and root hemiparasites). Moreover, closely related pierids at the level of subfamily feed on closely related plants at the level of order. These findings agree well with conclusions drawn from previous studies (Ehrlich & Raven, 1965; Vane-Wright, 1978; Courtney, 1986; Ackery, 1991; Janz & Nylin, 1998), however, these earlier workers did not have adequate phylogenies of the butterflies on which to place the host plant associations in an historical or evolutionary context.
Ancestral host plant
The most recent estimate of the phylogeny of the Pieridae (Braby et al., 2006b) tentatively concluded that the Coliadinae and Pierinae are sister taxa, and this lineage is sister to Pseudopontiinae + Dismorphiinae. Given this topology, the most parsimonious optimization suggests Fabales is the ancestral host plant, with feeding on Brassicales and Santalales both being derived. In this sense, our findings support the general conclusions of Scott (1985) and Janz & Nylin (1998), who hypothesized that the ancestral host plant of butterflies as a whole was likely to be Fabales, or a clade that included Fabales (i.e. ‘eurosids I’, see APG II, 2003). Ackery (1991) argued that Malvales (‘eurosids II’) was an equally likely ancestral host for the Papilionoidea + Hesperioidea + Hedyloidea, however, pierid butterflies do not feed on Malvales or Sapindales, although they do utilize the eurosids II order Brassicales. If Malvales or the ancestor of the eurosids II clade (i.e. Brassicales + (Malvales + Sapindales)) is the ancestral host, then there have been at least three independent shifts to legumes: (1) Dismorphiinae or Pseudopontiinae + Dismorphiinae; (2) Coliadinae; (3) the Tatochila group (Pierina). It would follow that crucifer-feeding, which is almost exclusively limited to the Pieridae (Ehrlich & Raven, 1965; Chew & Robbins, 1984; Courtney, 1986), evolved either by a shift from a Malvales-feeding ancestor, or as a result of cospeciation with the eurosids II clade but with a loss in Malvales + Sapindales. Multiple colonizations of Fabales are plausible, particularly since legumes are characterized by high nitrogen content and thus are potentially desirable to insects, but is less parsimonious.
Conversely, if Fabales are the ancestral host plant, this may explain the occurrence of legume-feeding in the Tatochila group of the Pierina in which at least three species (Tatochila theodice (Boisduval), Tatochila distincta Jörgensen, Hypsochila microdice (Blanchard)) specialize on Fabaceae in an otherwize crucifer-feeding clade (Shapiro, 1990, 1991a, b). In the case of T. distincta, the larvae apparently utilize only legumes in nature, but can be reared in captivity on crucifers, which may indicate that crucifer feeding is a pleisiomorphic trait in this species. The ability to utilize legumes by at least three members of the Tatochila group most likely represents relatively recent and possibly multiple recolonization of the ‘ancestral host’ from a derived crucifer-feeding state, that is, these butterflies have switched their oviposition and larval feeding preferences to modern representatives of the ancestral host lineage of the Pieridae. Janz & Nylin (1998) and Janz et al. (2001) concluded that recolonization of ‘ancestral hosts’ probably occurs frequently in butterflies, but because the process is dynamic the actual number of host changes is obscured at the higher taxonomic levels.
If pierids did originate on Fabales, their absence from the three other families in the order (Polygalaceae, Quillajaceae and Surianaceae), which possibly comprise a monophyletic group sister to Fabaceae, suggests either an extinction event on the ancestor of these families, or a more recent origin on the Fabaceae stem-group. Alternatively, absence of host records may simply be due to incomplete data, or because pierids ‘missed the boat’ on these families. Further field observations are needed to establish conclusively that pierids do not use Polygalaceae, Quillajaceae and Surianaceae – the latter two families comprise small taxa restricted to the Neotropics. An extensive phylogeny of the Fabaceae is now available (Wojciechowski et al., 2004; Lavin et al., 2005), and a more detailed analysis of host plant relationships of the Coliadinae and Dismorphiinae at the subfamily/tribal level would be most interesting. Such an analysis would establish the extent of host use across the major clades and may provide further insights into the degree of host affiliation and patterns of cospeciation/colonization.
Cospeciation or colonization?
Recent phylogenetic studies of the angiosperms show that although the Fabales, Brassicales and Santalales belong to the ‘eudicots’ clade they are not closely related (Chase et al., 1993; Soltis et al., 1999; Wikström et al., 2001; APG II, 2003). The Fabales and Brassicales fall within the ‘rosids’ clade, but belong to different subclades (eurosids I and II, respectively). Therefore, it is most unlikely that larval host associations of the Pieridae at level of plant order reflect a pattern of cospeciation. If pierids evolved on the common ancestor of Fabales and Brassicales (estimated to have originated 109–100 Myr), at least eight losses (extinction events) are needed to account for the present pattern under a cospeciation model (see Fig. 2 in Wikström et al., 2001; also Fig. 1 in APG II, 2003). Moreover, Santalales-feeding has arisen independently three times in the Pieridae, indicating multiple colonization events of these plants. Within the Aporiina, host plant associations also strongly suggest a pattern of repeated colonization of distantly related plants. For example, the angiosperm orders Caryophyllales, Ericales and Ranunculales are all distantly related to the Santalales [e.g. Ericales in ‘asterids’ clade (Bremer et al., 2004), Ranunculales sister to rest of the eudicots clade (Wikström et al., 2001; APG II, 2003)]. Similarly, host changes at the generic level in the Coliadinae (to Asterales, Zygophyllales and Rosales) are best interpreted as a pattern of multiple colonizations. The Asterales, in particular, are far removed from the Fabales, being within the ‘euasterids II’ clade of the asterids (Wikström et al., 2001; APG II, 2003; Bremer et al., 2004). Zygophyllales and Rosales belong in eurosids I clade, but are not closely related to Fabales, especially the former taxon. Furthermore, 34 plant families within eurosids I and Rosales are not utilized by Pieridae at all, with only the family Rhamnaceae exploited by some Coliadinae. At least four extinction events on the common ancestor of Zygophyllales (Zygophyllaceae) and Fabales (Fabaceae), and a further five extinctions on the common ancestor of Rosales (Rhamnaceae) and Fabales (Fabaceae), would be required under a pattern of cospeciation (see Fig. 8 in Wikström et al., 2001). If cospeciation did occur in ancient stem lineages at the macroevolutionary scale, it has been limited to within plant orders (e.g. Brassicales) or families (e.g. Loranthaceae), but available data suggests that if such a pattern exists it has been obscured by more recent host shifts within these taxa. However, coevolution between butterflies and their host plants may well occur at the microevolutionary scale (see Chew & Robbins, 1984; Chew, 1988 for reviews for pierid butterflies).
Wikström et al. (2001) provided robust estimates of the age of origin of all higher angiosperm taxa. Although the oldest Santalales fossil is Early Eocene (51.9 Myr) (Magallón & Sanderson, 2001), Wikström et al. (2001) estimated the stem- and crown-groups of the order to be considerably older; 118–111 Myr and 97–85 Myr, respectively. Similarly, the oldest Fabales fossils are of Palaeocene age (59.9–56 Myr) (Magallón & Sanderson, 2001; Lavin et al., 2005), but this order is estimated to have arisen and diversified in the Late Cretaceous (94–89 Myr for stem-group, 79–74 Myr for crown-group), with the family Fabaceae diversifying in the early Tertiary (68–56 Myr for crown-group) (Wikström et al., 2001). A recent, more precise, estimate places diversification of the Fabaceae in the Palaeocene (59.0 Myr) (Lavin et al., 2005). The Brassicales are estimated to have arisen and diversified in the late Cretaceous (90–85 Myr for stem-group, 79–71 Myr for crown-group) (Wikström et al., 2001), although the oldest fossil of the order has been dated slightly earlier than this, from the Turonian (91.2 Myr) (Magallón & Sanderson, 2001). Assuming these estimates and dates of fossils are accurate and our interpretations of host colonization within the family are correct, then the Pieridae stem-group cannot be older than 94 Myr (the maximum age of the Fabales stem-group) or else 79 Myr (the maximum age of Fabaceae stem-group). If Brassicales proves to be the ancestral host, rather than Fabales, then the Pieridae stem-group cannot be older than 91 Myr (the maximum age of the Brassicales stem-group fossil). In either case, these two orders of plants suggest similar maximum limits for an age of origin of the Pieridae in the Cenomanian-Turonian of the late Cretaceous (i.e. 94–91 Myr). These upper limits agree well with an approximate, extrapolated age estimation of 95 Myr (99.9% confidence interval 112–82 Myr) for the Pieridae crown-group based on fossil calibrations from the Tertiary (Braby et al., 2006b). If the butterflies are older than this, there has been extinction of lineages feeding on one or more plant clades related to the Fabales within eurosids I as well as extinction on families related to Fabaceae, or else extinction on plants related to Brassicales within eurosids II. Additional fossils are needed to estimate the age of the Pieridae more accurately.
Diversification and adaptive radiation
Given a pattern of host colonization as the most parsimonious explanation in the Pieridae, we now turn attention to diversification and possible mechanisms underlying host shifts that may have facilitated adaptive radiation. A generalized colonization model, summarising the major host shifts is depicted in Fig. 3. The total minimum number of host shifts at the level of plant order or family leading to butterfly diversification at the generic level is 11 steps. Six of these host shifts are from the putative ancestral host Fabales. However, four of these shifts (to Asteraceae, Zygophyllaceae and Rhamnaceae twice) have resulted in only four monomorphic genera (total of 12 species) in the Coliadinae, while another shift (to Opliliaceae) has led to the subfamily Pseudopontiinae but with only a single monobasic genus. In contrast, the shift from Fabales to Brassicales has led to the subfamily Pierinae containing 57 genera and approximately 830 species, which represents 69–75% of the lower diversity within the family. Within the Brassicales-feeding pierids, at least 33 genera (and potentially more than 360 species) specialize on crucifers and allied plants (Table 1). From Brassicales, there have been two further independent shifts to Santalales, each resulting in further diversification, with at least nine genera (and potentially more than 440 species, ca. 40% of all pierids) specialising on mistletoes. Table 1 shows that the early associations of pierids with Fabales and Brassicales have resulted in substantial diversification at both the generic and species levels, whereas the more recent association with Santalales has resulted in diversification mainly at the species level. Assuming all things equal, the earlier colonization of Brassicales and the more recent colonizations of Santalales clearly represent striking examples of adaptive radiation on these hosts. We hypothesize that the evolution of biochemical/physiological adaptations to feed on Brassicales and Santalales permitted the Pierinae to speciate more rapidly than their sister clade Coliadinae on Fabales. Similar patterns in which host shifts have promoted adaptive radiation have been reported in the Coleoptera, particularly those from gymnosperms to angiosperms (Farrell, 1998; Sequeira et al., 2000) but also shifts between different host niches (Mitter et al., 1988; Farrell & Sequeira, 2004).
Table 1. Major larval host plants (at level of order or family) exploited by pierid butterflies (at level of genus or above), showing estimated butterfly diversity for each plant-feeding clade.
What mechanism(s) could account for such radical shifts between distantly related host plants by members of the Pieridae? Host shifts between or within closely related plant taxa are not difficult to explain, and in butterflies are phytochemically driven through common ancestry (Ehrlich & Raven, 1965; Chew & Robbins, 1984; Jermy, 1984). These shifts frequently lead to general conservatism of host range (e.g. Mitter et al., 1988; Farrell & Mitter, 1994; Janz & Nylin, 1998). For example, most shifts in the Coliadinae are to orders or families (i.e. Zygophyllaceae, Clusiaceae, Euphorbiaceae, Salicaceae and Rhamnaceae) in the same clade as Fabales (i.e. eurosids I) (APG II, 2003). On the other hand, shifts in the Coliadinae between more distantly related plants, such as from Fabales to Asterales (euasterids II) or from Fabales to Brassicales (eurosids II), are more difficult to explain. In some cases, shifts to taxonomically and phenotypically distinct plants may at times be due to phytochemical convergence (Strong et al., 1984). For example, the genus Drypetes (Malpighiales: Euphorbiaceae) contains glucosinolates, but the taxon is distantly related phylogenetically to the Brassicales (Rodman, 1991; Rodman et al., 1993, 1998) and currently placed in eurosids I (APG II, 2003). The presence of this compound probably accounts for the widespread occurrence of larvae of at least two taxa in the Appiadina [Appias (Catophaga), Appias (Glutophrissa)] feeding on Drypetes in an otherwise Brassicales-feeding lineage; such a shift implies that Appias actively seek host plants with glucosinolates, as contrasted with larvae merely having a tolerance to this chemical. In other cases, however, radical host shifts may have more to do with physical contact of potential hosts, or with oviposition mistakes. Indeed, Chew & Robbins (1984) discussed two general mechanisms in butterflies: one concerns the close physical proximity of established and potential hosts, the other relates to females visiting flowers for nectar of species in which the larvae specialize on flowers.
In the case of Santalales, it seems likely that mistletoe feeding evolved as a result of the hemiparasite growing on the trunks and branches of a Brassicales host tree (Chew & Robbins, 1984; Venables, 1993). For instance, tall shrubs of Capparis in Australia, a common host plant of Elodina, Belenois and Cepora, are occasionally parasitised by mistletoes in both the Loranthaceae and Viscaceae. In North America, Mooney (2003) concluded that shifts between conifer host trees (Pinaceae) and their dwarf mistletoe hemiparasites (Viscaceae) may be common in Lepidoptera because of the close physical proximity of host and parasite. The close physical association between the two plants would increase ecological opportunity, and provide a mechanism for larvae to be exposed to taxonomically and chemically unrelated potential host plants. Contact with the new host could occur as a result of females ovipositing on or near the hemiparasite, or as a consequence of larvae defoliating the host tree and exhausting their food supply. Further shifts might then occur between the mistletoe parasite and other host trees attacked by mistletoes (Chew & Robbins, 1984; Mooney, 2003).
Such a scenario between host and parasite would explain the numerous shifts observed in the Aporiina. This subtribe shows at least three independent host shifts away from mistletoes to distantly related plants – in the eudicots, asterids and gymnosperms. These shifts are interpreted to signify different host trees parasitised by aerial-stem mistletoes (Fig. 3). Closer inspection of the relevant host plants of the butterflies provides circumstantial evidence in favour of this hypothesis. For instance, larvae of Neophasia in North and Central America (south-western Canada, western USA, northern and western Mexico) feed exclusively on conifers (Pinaceae), particularly on Pinus but also on Pseudotsuga, Abies, Tsuga and Picea (Evenden, 1926; Cole, 1961; Howe, 1975; Ferris, 1980; Scott, 1986; Layberry et al., 1998; Robinson et al., 2002). All of these host trees support dwarf mistletoes, Arceuthobium (Viscaceae), which are obligate parasites of conifers (Hawksworth & Wiens, 1972, 1996; Jerome & Ford, 2002; Mooney, 2003; Nickrent et al., 2004) and co-occur in the breeding range of the butterfly. Moreover, Norton & Carpenter (1998) concluded that Arceuthobium probably evolved as a parasite on the ancestor of Pinus and then differentiated by switching to other conifer hosts. Interestingly, Pinus is also the primary host plant of Neophasia. Likewise, the larvae of Eucheira socialis Westwood in the montane areas of western Mexico feed exclusively on Arbutus (Ericaceae), including Arbutus xalapensis Kunth, Arbutus glandulosa M. Martins & Galeotti, A. macrophylla M. Martins & Galeotti and A. arizonica (A. Gray) Sarg. (Kevan & Bye, 1991; Underwood, 1994). At least one of these host plants, A. xalapensis, is commonly parasitised by Phoradendron bolleanum (Seem) Eichler (Viscaceae) and other mistletoes (J. Kuijt, personal communication) within the breeding areas of the butterfly. Larvae of Aporia in the Himalaya and nearby areas feed primarily on shrubs of Berberis (Berberidaceae), but also on many genera of Rosaceae (e.g. Igarashi & Fukuda, 1997; Tolman, 1997; Robinson et al., 2001). There are no reports of mistletoes parasitising Berberis in Nepal or northern India, and this genus may be resistant to mistletoe attack (G. Glatzel & M. P. Devkota, personal communication); however, the Rosaceae hosts of Aporia are frequently parasitised by Viscum (Viscaceae) (Barney et al., 1998; Devkota & Glatzel, 2005). Unfortunately, the systematic relationships of Aporia (Mesapia) relative to Aporia (Aporia) and Aporia (Metaporia) are unknown, and the larval host plant of Mesapia has not been recorded. A better phylogeny of Aporia, together with more complete host plant data, is therefore needed in order to determine whether Berberidaceae (Ranunculales) or Rosaceae (Rosales) is the ancestral host of the butterfly genus. Regardless of the mechanism of host shifts, it is clear that mistletoe feeding in the Aporiina has led to some unusual and novel host plants outside the conventional pierid host range of Fabales and Brassicales.
Host specialization and plant chemistry
Members of the Brassicales contain a class of secondary compounds known as glucosinolates, which act as a defence against generalist herbivores (e.g. Feeny, 1977; Chew, 1988). When plant tissue is damaged, the nontoxic glucosinolates are hydrolysed by the enzyme myrosinase to produce mustard oils, including isothiocyanates, which are highly toxic to most insect herbivores (Rask et al., 2000; Wittstock et al., 2004 and references therein). Recent studies on Pieris, however, have demonstrated that the larvae detoxify and eliminate, rather than sequester, the degradation products of glucosinolates by redirecting the hydrolysis reaction toward the formation of nitriles instead of isothiocyanates (Müller et al., 2003; Wittstock et al., 2004). The extent of this detoxification process within the Pieridae has not been surveyed, although Agerbirk, Müller, Olsen & Chew (Unpublished data) recently discovered a second detoxification mechanism in Pieris and Anthocharis (both Pierinae) in which metabolism of the glucosinolate sinalbin circumvents the formation and acquisition of mustard oils. Presumably, once such novel mechanisms evolved, these biochemical adaptations allowed the Pierinae to successfully exploit, specialise and radiate on plants within the Brassicales.
A striking association between host plant, larval and adult behaviour, adult phenotype, and mimicry is apparent in the subfamily Pierinae. Adult butterflies of several lineages that feed as larvae on crucifers and allied plants (i.e. Colotis group, Anthocharis group of Anthocharidini, Appiadina, Pierina, Leptosia, Elodina, Dixeia, Belenois, Prioneris and Cepora) are apparently nonaposematic, being generally white with black and often yellow markings, but rarely contrasted with bright red markings; the larvae, as a general rule, are solitary feeders, and the adults have a fast, erratic flight. Despite the host plants containing mustard oil glucosides, detailed studies on Pieris and Anthocharis suggest that these butterflies do not sequester the toxic degradation products of glucosinolates but instead metabolise and eliminate the mustard oils in a nontoxic form (Müller et al., 2003; Wittstock et al., 2004; Agerbirk, Müller, Olsen & Chew, Unpublished data). A notable exception, however, occurs in Perrhybris and Pieriballia, two closely related Neotropical genera in the Pierina, in which the larvae feed gregariously and the adults are highly aposematic, particularly Perrhybris (e.g. DeVries, 1987). It is possible that these butterflies sequester mustard oils, and further studies to test this hypothesis would be most interesting.
In contrast to the crucifer-feeders, adult butterflies that feed, or potentially feed, as larvae on mistletoes (i.e. in the Aporiina –Mylothris, Delias, Melete, Pereute, Leodonta, Catasticta, Archonias and Charonias– and in the Anthocharidini – especially Cunizza of the Hesperocharis group) are generally warningly coloured on the wing underside; the larvae are highly gregarious, and the adults advertise their warning patterns with slow deliberate flight. Furthermore, these mistletoe-feeding butterflies frequently form mimicry complexes with many distantly related butterflies as well as diurnal moths (Finn, 1897; Dixey, 1920; Yata, 1985; DeVries, 1987; Morinaka & Yata, 1994; Beccaloni, 1997; Orr, 1999; Yen et al., 2005; Braby, 2005a). The nature of these mimetic associations has rarely been investigated in terms of Batesian or Müellerian mimicry, although it is generally assumed that the mistletoe-feeding pierids are distasteful models and not palatable mimics. Conversely, adult butterflies which feed as larvae on the host trees of mistletoes (i.e. in the Aporiina –Aporia, Neophasia and Eucheira) are predominantly white with black markings or black with white and/or yellow markings, and more closely resemble the crucifer feeders in being nonaposematic, although the larvae are gregarious. An exception to this general rule is the female of Neophasia terlooii Behr, which is bright orange with black markings, resembling adults of Danaus plexippus (Linnaeus) that are poisonous to most predators. Moreover, N. terlooii flies late in the season (October–November) in the montane coniferous forests of western Mexico where adult D. plexippus overwinters in vast numbers, indicating that N. terlooii may be an example of Batesian mimicry.
The inference from these associations is that mistletoe-feeding pierids are highly unpalatable to most predators, aposematic and form mimetic associations, whereas the nonmistletoe-feeding species that specialise on crucifers and mistletoe host trees generally do not show these traits. Experimental studies on Pieris and Anthocharis in the Northern Hemisphere have in part confirmed this inference (Kingsolver, 1987; Ley & Watt, 1989; Lyytinen et al., 2001) in that these two crucifer-feeding genera are palatable, nonaposematic and do not form mimetic associations. A further prediction from this Pierinae host plant-palatability theory is that the larvae of relict genus Leuciacria in West Papua, Indonesia and New Guinea will be found to feed on mistletoe host trees and not on mistletoes as occurs in its sister genus Delias. The larval host plant of Leuciacia is not known, but adults of the two known species are nonaposematic, being predominantly white. If this prediction proves correct, Leuciacria would represent another independent shift from a mistletoe-feeding ancestor.
The presence of secondary compounds in mistletoes, and sequestration of those compounds by mistletoe-feeders, has not been established conclusively. Finn (1897) demonstrated that adults of the warningly coloured Delias eucharis Drury were rejected or avoided by avian predators, and Orr (1999) and S.-H. Yen (personal communication) have similarly noted that Delias adults are seldom attacked and/or eaten by birds, in New Guinea and Asia, respectively. Orr (1999) suggested that the larvae of Delias either sequester toxic compounds or possibly produce these compounds biochemically from one or more nontoxic precursors. More recently, Yen et al. (2005), in discussing the mimetic relationship between the Delias pasithoe group and Cyclosia pieroides Walker (Zygaenidae: Chalcosiinae), stated ‘adults...have volatile compounds sequestered from their host plants (e.g. Loranthaceae).’ (p. 197), and noted the presence of alkaloids in mistletoes (S.-H. Yen, personal communication). However, the chemical nature and distribution of these secondary compounds in the Santalales has not been surveyed. Clearly, further research into the relationships between plant chemistry, larval gregariousness, adult palatability, aposematism and mimicry among the mistletoe-feeders of the Pierinae would be most rewarding.
We are most grateful to F. Chew, N.E. Pierce, S. Nyin, A.M. Shapiro and W.B. Watt for comments on an earlier draft of the manuscript; A.M. Shapiro, J. Kuijt, G. Glatzel and M.P. Devkota for providing unpublished host plant records and/or botanical information; and S.-H. Yen for his observations on biological interactions between Delias and mistletoes in Asia. A.V.L. Freitas kindly provided us with a copy of his unpublished manuscript on the biology of Leucidia; F. Chew, C. Müller and co-workers greatly assisted with literature and drew our attention to their published and unpublished work on metabolic pathways of glucosinolates in the Pieridae. This work was supported by an Australian Research Council Fellowship (No. F19906650) and an Australian-American Fulbright Postdoctoral Fellow Award to MFB.