S. M. Eppley, Department of Biology, Portland State University, PO Box 751, Portland, OR 97207-0751, USA. Tel.: +1 503 725 8986; fax: +1 503 725 3888; e-mail: email@example.com
Which conditions favour the evolution of hermaphroditism or separate sexes? One classical hypothesis states that an organism’s mode of locomotion (if any) when searching for a mate should influence breeding system evolution. We used published phylogenies to reconstruct evolutionary changes in adult mate-search efficiency and breeding systems among multicellular organisms. Employing maximum-likelihood analyses, we found that changes in adult mate-search efficiency are significantly correlated with changes in breeding system, and this result is robust to uncertainties in the phylogenies. These data provide the first statistical support, across a broad range of taxa, for the hypothesis that breeding systems and mate-search efficiency did not evolve independently. We discuss our results in context with other causal factors, such as inbreeding avoidance and sexual specialization, likely to affect breeding system evolution.
If an organism is sessile or nearly sessile, finding a mate can be challenging. Even if an organism can actively search for a mate, finding a partner may be difficult if the energetic costs of mate searching are high. Some models assume a trade-off between time spent searching for a mate and time spent feeding. As mate-search efficiency decreases, this trade-off reduces the likelihood that an individual can continue to search for a mate until it is successful (Puurtinen & Kaitala, 2002). Under these conditions, hermaphroditism is expected to be advantageous for several reasons. Many hermaphrodites have the ability to self-fertilize, allowing for reproductive success even when no mates are found (Darwin, 1876; Jarne & Charlesworth, 1993). While the advantage of self-fertilization for hermaphrodites has been widely discussed in the literature, particularly with respect to plants, hermaphroditism is also expected to be advantageous in self-incompatible hermaphroditic organisms because an individual may be able to mate with any conspecific individual it meets, whereas an organism with separate sexes must wait to encounter a member of the opposite sex (Tomlinson, 1966; Ghiselin, 1969). Thus, hermaphroditism may provide reproductive assurance for organisms with poor mate-searching efficiency. Additionally, hermaphrodites can adjust the allocation of resources to favour female function when opportunities for reproductive success through male function are limited by mating opportunities, as may be the case in species with copulation and internal fertilization in which the cost of movement is high (Charnov, 1979; Puurtinen & Kaitala, 2002). By contrast, when searching for mates is energy efficient, hermaphroditism is no longer a stable strategy (Puurtinen & Kaitala, 2002). Under these conditions, a male or female mutant is expected to be able to invade a hermaphroditic population because of selective pressure for the sexes to specialize in different mate-search strategies (Puurtinen & Kaitala, 2002).
Here, we assess whether changes in one trait, adult mate-search efficiency, result in changes in breeding system across plant, fungal and animal phyla. We tested this hypothesis using three class-level, putative phylogenies of eukaryotic organisms which differ in the basal relationships among the metazoa. We tested this hypothesis over the eukaryotic phylogeny, rather than within a phylum, because, while breeding system has evolved repeatedly within particular phyla and even lower taxonomic orders, evolutionary transitions in adult mate-search efficiency at the scale that is likely to affect breeding system have occurred only a few times across the entire multicellular eukaryotic phylogeny (10 times, given our criteria; Fig. 1). We coded adult mate-search efficiency for each taxon as a binary trait according to two predetermined criteria based on the energy efficiency of locomotion, and determined the predominant breeding system of the terminal taxa. We considered adult mate-search efficiency based on the energy efficiency of locomotion of the adult individual searching for a mate, not the ability of gametes to find other gametes (i.e. through such techniques as broadcast spawning in marine invertebrates and insect pollination in plants). While we expect taxa with low adult mate-search efficiency to be under strong selection to increase the ability of gametes to disperse, we tested the hypothesis that adult mate-search efficiency, regardless of any variation in gamete dispersal ability, changes with breeding system. However, we include an analysis of the correlation between gamete mate-search ability and breeding system to determine the effect of this correlation on our analysis. Mate-search ability of gametes was determined based on whether or not gametes are involved in mate search and not on any measure of efficiency, as little data are available on gamete mate-search efficiency across phyla. We used maximum-likelihood analysis to test whether the evolution of adult mate-search efficiency or gamete mate-search ability correlate with the evolution of breeding system and to test theoretical predictions about the conditional evolution of the two traits.
Materials and methods
We established character states using published sources (see Supplementary material Appendix S1 for a full list of references). We coded taxa as having ‘good’ adult mate-search efficiency if all individuals as adults: (1) swim using bilateral musculature as their primary means of locomotion; or (2) have paired movable appendages for locomotion. All other taxa were coded as ‘poor’. These criteria are based on the energy costs of movement. Studies of energy efficiency of locomotion have consistently shown that across phylogenetic groups the least costly modes of locomotion are: (1) swimming (with or without using paired appendages for propulsion); and (2) flying or running (using paired, movable appendages; Full, 1997; Donovan et al., 1999; Biewener, 2003) for animals specialized for each mode of locomotion (and see Supplementary material available online). Conversely, locomotion has been found to be most costly in animals that crawl without using paired appendages (Casey, 1991; Berrigan & Lighton, 1993; Donovan et al., 1999). For instance, the energetic cost of moving for a terrestrial slug is 100 times more than for a swimming invertebrate and 10 times more than for a running invertebrate of the same mass (Donovan et al., 1999). There is substantially more data on the energetics of locomotion in energy-efficient taxa than in taxa which rely exclusively on modes of locomotion which current research demonstrates as less efficient, such as adhesive and peristaltic crawling. More research across a broader array of adhesive and peristaltic crawler would be informative, and it may reveal an adhesive or peristaltic crawler that is as energy efficient as the least efficient runners. However, because crawlers must overcome substantial friction (Menciassi et al., 2006) and in some cases produce costly mucus to reduce friction (Donovan et al., 2006), crawlers in general are theoretically less likely to be efficient than are runners, fliers and swimmers. We will have presumably miscategorized a few taxa because their particular mode of locomotion is more energy efficient than the norm for their type. For instance, we categorized rotifers as having poor mate-search efficiency, despite the fact that many species can swim in a manner that appears efficient. However, no energetics studies have been performed on rotifers, and their type of swimming does not fall within our category of efficient locomotion. The noise created by such miscategorization will be random, to a large extent.
We stratified organisms according to the predominant mode of locomotion they use as adults (not as gametes or as immature individuals), and thus use for searching for mates and foraging. For example, in the sea hares the ability to swim, at least for brief periods, may have evolved more than once, but swimming as a primary mode of locomotion is rarer (Carefoot & Pennings, 2003). The one exception to the criteria that we are aware of is for terrestrial snakes, which during routine forms of movement expend energy equal (for undulatory locomotion, Walton et al., 1990) or less (for side winding, Secor et al., 1992) than for limbed vertebrates of similar mass. The vertebrate skeleton is thought to be the reason for this exception because it provides a solid element against which muscles can work (Casey, 1991). For this reason, we have coded the terrestrial snakes as having ‘good’ mate-search ability.
We used published studies to determine whether each taxon included simultaneous hermaphrodites. Species were considered simultaneous hermaphrodites if individuals had both sexual functions throughout all stages of their lifetime. When the presence of simultaneous hermaphroditism in a taxon could not be assessed using reviews and texts, we conducted systematic searches of the literature using the ISI Web of Science database with the keywords hermaphrod* + the taxon. Taxa with sequential hermaphroditism and/or separate sexes, but no simultaneous hermaphroditism, were considered as being gonochorisitic. We combined sequential hermaphroditism with gonochorism because an individual with sequential hermaphroditism is either male or female at any given time, and thus faces the same mate-searching challenges as individuals with separate sexes. Of the 124 taxa we used in our analysis, 15 included sequential hermaphrodites. Ten of these cases were categorized as having poor mate-search efficiency, which makes the results more conservative as the alternate categorization is more likely to support our hypothesis. The only case where we recorded sequential hermaphroditism as simultaneous hermaphrodite is when they are functionally equivalent because the sequential hermaphrodite stores sperm (Ghiselin, 1969); this occurred in one taxon.
We also performed two additional analyses in which we included an additional variable: gamete mate-search ability (see Online Table 1). We defined a taxon as having good gamete mate-search ability if gametes travel between adults of a species by water, wind, animal vector or their own locomotion and potentially increasing the efficiency of mate searching. Thus, taxa such as angiosperms with pollen transported by animal and abiotic vectors were included in this category. Also included were animals that broadcast spawn, such as in echinoderms, and animals with sperm casting (where sperm but not eggs are released into the water and fertilization is internal), such as many hydroids. Taxa were classed as having poor gamete mate-search ability if gametes do not travel between adults and aid in mate search. In general, this means that adults come in contact with one another during mating, either resulting in internal fertilization via copulation or external fertilization after contact between mating partners. In amphibians, there are exceptions in which adults do not come in contact with one another, and yet gametes do not aid in mate search because they are transferred as immobile packages which males deposit and females recover (e.g. Blanchard, 1933). Because taxa with low adult mate-search ability might be under strong selection to increase the ability of gametes to disperse, we also tested the hypothesis that good mate-search ability at either the gamete or adult stage is correlated with breeding system. For this analysis, a taxon was coded as having good mate-search ability if we had coded it as having either good adult mate-search efficiency or good gamete mate-search ability; a taxon was coded as having poor mate-search ability if we had coded it as having both poor adult mate-search efficiency and poor gamete mate-search ability.
Table 1. Results from tests of independence of adult mate-search efficiency and breeding system, using Discrete.
Treatment of taxa variable for a trait
Tests were calculated using three possible metazoan backbone phylogenies, and by treating polymorphisms in breeding system in the terminal taxa two ways (see Materials and methods for details). LI4 is the log-likelihood of the four-parameter independent model; LD8 is the log-likelihood estimate of the eight-parameter dependent model; and LR is the likelihood ratio. P was tested against a chi-squared distribution with four degrees of freedom.
Coded as > 75%
Coded as > 75%
Coded as > 75%
We used published phylogenies to construct a composite class-level phylogeny of eukaryotic organisms. To examine how our results may be biased by uncertainties in basal relationships in the metazoan phylogeny, we compared three phylogenies: a combined morphological and molecular phylogeny for the basal relationships among the metazoa (Peterson & Eernisse, 2001), a molecular phylogeny (from Adoutte et al., 2000 with additional resolution from Mallatt & Winchell, 2002) and a recent morphological phylogeny (Nielsen, 2001). Of the 138 morphological traits used to build the Peterson & Eernisse (2001) phylogeny, none relates directly to breeding system, but several, such as the presence of a dorsal/ventral axis, will influence mate-search efficiency. The morphological phylogeny is more problematic with respect to breeding system and mate-search efficiency characters used to build the phylogeny. For instance, in this phylogeny, the trait ‘sessile adults’ was used to differentiate the phyla Porifora, Entoprocta and Ectoprocta from neighbouring phyla (Nielsen, 2001). Because of these inherent problems with the morphological phylogeny, we tested all three phylogenies, but kept all three in our analysis, despite potential flaws, to determine whether variation in the metazoan backbone influenced our results. For all three phylogenies, we used the combined molecular and morphological phylogeny from Stach & Turbeville (2002) for relationships among the Urochordata and the Passamaneck et al. (2004) combined LSU and SSU phylogeny for the relationship among the Mollusca classes. The several possible placements for the Solenogastres did not change the tree with respect to our tests. We also used an individual phylogeny for relationships among the crustaceans and hexapods (Regier et al., 2005). We used the Tree of Life web project to determine the remaining relationships (http://tolweb.org/tree/phylogeny.html).
We mapped every gain or loss of adult mate-search efficiency onto the phylogenies, and we resolved the phylogenies to the level of class, as this was the level at which information was available for the majority of animal groups. In a few instances, taxa were not included as class because: (1) the taxonomic class had been shown to be a polyphyletic entity (this was the case with the classic classes Reptilia and Aves, which we include as the Archosauria, Lepiosauria and Tesudines and the Ascidians within the Urochordata which have been split into several taxa); (2) a higher taxonomic level above class is not traditionally split into classes (this is true in the case of the subphylum Cephalochordata which has only about 25 species and is sometimes included as a class rather than a subphylum and which we included at the subphylum level; the phyla Phoronida and Entoprocta, as there are only a dozen species described in the first instance and roughly 100 in the second); or (3) the class is variable for mate-search ability, in which case we split it into poor and good mate-search efficiency groupings (this occurred in the Secernentea, Polychaeta, Hirundinea and Gastropoda).
We treated taxa in two ways: (1) we coded all taxa variable for breeding system as uncertain (giving 122 total taxa for the combined phylogeny or 124 taxa for the morphological and molecular phylogenies); and (2) we coded a taxon as nonhermaphroditic if it had 0–25% hermaphroditic species; variable for breeding system if it had 25–75% hermaphroditic species; and hermaphroditic if it had 75–100% hermaphroditic species. For the second treatment, the 33 terminal taxa that were classed as variable were removed from the analysis, giving a total of 89 taxa for the combined phylogeny and 91 taxa for the morphological and molecular phylogenies. A similar approach using an 80% criterion has been shown to be relatively robust to the incorrect assignment of ancestral states (see Routley et al., 2004), although this approach tested species within families rather than species within class which was used here.
We used Pagel’s (1994) maximum-likelihood-based analysis using the Discrete module in BayesTraits (http://www.evolution.rdg.ac.uk/BayesTraits.html) to test for correlated character evolution between adult mate-search efficiency and breeding system. We ran each analysis 10 times to increase the chance of finding a global rather than a local maximum likelihood. To test for correlated character evolution, we compared an independent model of character evolution, where each trait evolves independently of the state of the alternate traits, to a model of dependent evolution, where the rate at which one trait evolves is potentially dependent on the state of the other trait. The dependent model estimates rate of change between two states (qij), where the subscripts refer to the beginning and end character states for each trait. The log-likelihood (lnL) of the independent and dependent models were obtained from Discrete, and a likelihood ratio test statistic was obtained using LR = 2ΔlnL. This was tested against a chi-squared distribution with four degrees of freedom. Because accurate determination of branch lengths was not feasible, we coded all branch lengths as equal (see Harding et al., 1974). We also examined the results when coding branch lengths using Grafen branch lengths (Newman et al., 1997), but only report qualitatively different results. Polytomies were arbitrarily resolved and set to a branch length of 0.0001, and we examined the results from 100 different arbitrary resolutions of polytomies.
If there was a significant correlation between the two traits, we then conducted tests of conditional evolution and temporal order of the acquisition of traits. For all tests, we estimated the likelihood of a restricted model of evolution where one transition rate parameter was equal to another rate parameter. For example, to test whether the evolution of gonochorism is conditional upon good adult mate-search efficiency, we restricted q12 (transition rates between hermaphrodites with poor mate-search efficiency with gonochoristic organisms with poor mate-search efficiency) to be equal to q34 (transition rates between hermaphrodites with good mate-search efficiency and gonochoristic organisms with good adult mate-search efficiency) and compared it with the maximum likelihood of an unconstrained model. We also tested whether each parameter value was significantly greater than zero by constraining each of the parameters to be zero and evaluating the subsequent LR. For all tests, the likelihood of the reduced model in which one parameter was constrained was then compared with the full, eight-parameter model. We tested the likelihood ratio statistic against a chi-squared distribution with one degree of freedom.
Likelihood ratio tests suggest that the evolution of adult mate-search efficiency and breeding system are significantly correlated in the phylogenies of the multicellular eukaryotes (Table 1; Fig. 1 and see Supplementary material available online). For all three metazoan backbone phylogenies, a model of correlated evolution was significantly more likely using equal branch lengths (Table 1), regardless of whether terminal taxa variable for breeding system were treated as equivocal or whether terminal taxa were only included if > 75% of species within the taxa were all one character state. Different arbitrary resolution of polytomies did not influence significance of any of these analyses (data not shown).
Tests of conditional evolution found little evidence that the evolutionary transitions to hermaphroditism may be conditional on poor efficiency of adult mate search (Table 2, Fig. 2). We found no evidence that transitions between breeding traits are likely to evolve after evolutionary transitions in adult mate-search efficiency or vice versa. An examination of the estimated transition rates suggested that some transitions are extremely unlikely (see Online Table 2). For example, all but one of the estimated transitions rates from gonochorism to hermaphroditism in good adult mate-searching lineages (q43) were not significantly different from zero, and less than the smallest significant estimate rate in each analysis. By contrast, while the estimated transition rates were low for the pathway from good to poor mate-search efficiency in gonochoristic lineages (q42), if unknown taxa were coded using our 75% criteria, four of the six analyses were significantly different from zero. This suggests that the lack of evidence for temporal order in the evolution of the two traits may be because of the low power in the transition estimate from good to poor mate search in gonochoristic lineages (see Kolm et al., 2006; Rolland et al., 1998). While the instantaneous transition rate away from a gonochoristic phenotype with efficient mate searching is low, it is likely that the loss of good mate-search efficiency occurs before the gain of hermaphroditism, as predicted by the mate-search hypothesis.
Table 2. Comparison of transition rates between efficiency of adult mate-search and breeding system (hermaphroditic or gonochoristic). P values >0.05 are indicated in bold.
Variable taxa treated as uncertain
Variable taxa coded if > 75%
Change to gonochorism dependent upon good mate-search efficiency (q34 > q12)
q12 ≠ q34
q34 > q12
Change to hermaphroditism dependent upon poor mate-search efficiency (q21 > q43)
q21 ≠ q43
Gain of good mate-search efficiency occurs before gain of gonochorism (q13 > q12)
q12 ≠ q13
Loss of good mate-search efficiency occurs before gain of hermaphroditism (q42 > q43)
q42 ≠ q43
There was very little support for correlated evolution between gamete mate-search ability and breeding system. There was a significant association in only two of the 12 tests of correlated evolution (see Online Table 3). However, when motility was coded as either having good mate-search ability as adults or as gametes, there was a significant correlation between motility and breeding system when polymorphic breeding systems were treated as unknown. If polymorphic breeding systems were coded according to the 75% criteria there was a significant association for only two of the eight tests. Using the 75% criteria, there was no consistent evidence for conditional evolution or consistent temporal order. Contrary to the mate-search hypothesis, in three of six tests, evolutionary transitions away from good mate-searching gonochoristic lineages were more likely to occur through changes in breeding system (q43) than through changes in mate-search ability (q42), and there was no significant difference between any of the other tested evolution pathways (see Online Table 4).
According to our results, adult mate-search efficiency and breeding system did not evolve independently in the phylogeny of multicellular eukaryotes, and these results are robust to uncertainties in the phylogenies and statistical methods. This finding is the first statistical test of the hypothesis that the changes in breeding systems and mate-search efficiency are correlated across a broad range of taxa.
Whereas the metazoan backbone phylogeny is not well resolved, changes in adult mate-search efficiency and breeding system are spread throughout the metazoan phylogeny (Fig. 1), rather than clustered in one or two phyla, and thus, the results are not unduly influenced by changes in one or two taxa. Furthermore, strong significance for three quite different putative metazoan backbones suggests that the results are not artefacts of uncertainty in the metazoan backbone. However, because of lack of data for many taxa, the phylogeny could not be resolved to the genus or species level, which may have affected tests of conditional evolution (see below). In addition, correlational studies do not necessarily imply causality, and it is possible that other unknown variables influence both mate-search efficiency and breeding system evolution.
Maximum-likelihood tests of the conditional evolution of adult mate-search and breeding system traits found no support for the hypothesis that changes in adult mate search precede changes in breeding system. This result is probably influenced by a lack of power: character-state reconstructions using both parsimony and maximum likelihood suggest two reversions to poor adult mate-search efficiency probably occurred within multicellular eukaryotes (the Cirripedia within the crustaceans and in the terrestrial nematodes such as Caenorhabditis elegans within the Ecdysozoa; Fig. 1 and results not shown). In these two cases, reversion to poor adult mate-search efficiency occurred in a number of species, and reversion to hermaphroditism occurred in at least one of these taxa in each case, although not in the entire group in either instance. Moreover, in other taxa that are variable for breeding system characters, transitions between hermaphroditism and gonochorism have probably occurred numerous times. This has particularly been documented in flowering plants, where the evolution of dioecy from hermaphroditism has occurred over 25 times (Weiblen et al., 2000). Expanding our phylogeny until no taxa are variable for breeding system characters would increase the power of the likelihood tests, and it might also increase the number of evolutionary transitions from hermaphroditism to gonochorism, especially in sessile taxa, potentially reducing statistical support for this test further. Thus, a greater resolution of breeding systems and phylogenetic information (information which is currently unavailable for many taxa particularly among the marine invertebrates) is needed before a clear statement about the order of evolution of breeding systems and traits that influence mate-search efficiency can be made.
Similarly, we found no evidence that changes to gonochorism are more likely when organisms possess traits that allow efficient adult mate search, and indeed, ancestral state reconstruction using both parsimony and maximum likelihood suggests that many transitions to good adult mate-search efficiency occurred on branches in which separate sexes had previously evolved (Fig. 1 and results not shown). This may have occurred because separate sexes were favoured in those lineages for reasons not concerned with mate-search ability (see below), or by our use of two, stringent criteria to assign adult mate-search efficiency. If the evolutionary gain of efficient locomotion is a gradual process, then the use of a limited number of characters to assign mate-search efficiency, rather than actual measures of the costs of locomotion in each organism, will underestimate the number of cases where good adult mate-search efficiency preceded the evolution of separate sexes. Thus, it is possible that more accurate estimates of mate-search efficiencies in future may lead to additional support for the role of mate-search efficiency in breeding system evolution.
Our data indicate that adult mate-search efficiency is one factor correlated with the evolution of hermaphroditism in multicellular organisms. Theory suggests that there are several potential causal factors for breeding system evolution, particularly with regard to the loss of hermaphroditism. A male or female mutant is expected to be able to invade a hermaphroditic population when self-fertilization causes severe inbreeding depression (Charlesworth & Charlesworth, 1978a). Inbreeding depression, the reduction in fitness of inbred progeny relative to their outbred siblings, is thought to be the major force countering the evolution of inbreeding, particularly self-fertilization in angiosperms (Darwin, 1876; Fisher, 1949; Stebbins, 1957; Lande & Schemske, 1985; Lloyd & Schoen, 1992; Charlesworth & Charlesworth, 1998). Alternatively, the benefits of investing in either male vs. female function (i.e. sexual specialization), rather than inbreeding depression, can theoretically counter the transmission advantage of self-fertilization and favour the evolution and maintenance of separate sexes (Charnov, 1982; Charlesworth & Morgan, 1991; Charlesworth, 1999; Meagher, 2007). Recent data suggest that inbreeding avoidance (Costich & Meagher, 1992; Dorken et al., 2002) and sexual specialization (Costich, 1995; Bram, 2002; Gleiser & Verdu, 2005; Eppley & Pannell, 2007a) are two factors that lead to the evolution of dioecy in angiosperm systems. Similarly, we expect multiple mechanisms to be at work among the larger metazoan phylogeny. Also, in many instances the distinctive aspects of the reproductive biology of a species will favour certain breeding systems. For example, hermaphroditism in some coral reef fishes occurs in situations where both individuals involved in a pair-mating trade eggs – a practice expected to select for combined sexes because it limits opportunities of increasing reproductive success through male function (Charnov, 1982). Any or all of these factors may have played a role in the evolution of the high degree of breeding system variation that occurs within many eukaryotic taxa, such as plants, particularly those with poor adult mate-search efficiency (Fig. 1).
Furthermore, in taxa such as angiosperms and sessile marine invertebrates with extremely low adult mate-search efficiency, we expect strong selection either for increased adult mate-search efficiency or increased gamete dispersal distance and efficiency. Although we found very little evidence for a correlation between gamete mate-search ability and breeding system across all multicellular eukaryotes (Online Table 3), the degree to which organisms with poor adult mate-search efficiency can disperse their gametes (e.g. internal fertilization vs. broadcast spawning in marine invertebrates; fly, bee, bird, mammal or wind pollination in plants) is expected to be central to the evolution of breeding system variation within those taxa, and consequently, those taxa, particularly angiosperms, have become, the model organisms for the study of breeding system evolution (Geber et al., 1999; Charlesworth, 2006). Work comparing closely related species with breeding system variation, particularly in the less well-studied marine invertebrate taxa, will most likely be the key to future insights in understanding breeding system evolution.
We are grateful to K.L. Abbott, M. Crawford, S. Hillman, P. Gillingham and N.E. Phillips, J. Podrabsky for help with published sources. We thank S.C.H. Barrett, G.W. Eppley, J.M. Eppley, R.K. Grosberg, P.D. Lorch, A.M. Shaw, R.J. Toonen and two anonymous reviewers for valuable comments on earlier drafts of this manuscript. Research was supported by The Royal Society of New Zealand Marsden Fund (grant VUW 303 to L.K.J.), Natural Sciences and Engineering Research Council of Canada, and the National Science Foundation (grant INT 0202645 to SME).