Phenotypic Plasticity and Integration in the Mangrove Rivulus (Kryptolebias marmoratus): A Prospectus

Abstract

The mangrove rivulus (Kryptolebias marmoratus) is a small fish native to mangrove ecosystems in Florida, the Caribbean, Central America, and South America. This species is one of only two self-fertilizing, hermaphroditic vertebrates capable of producing offspring that are genetically identical to both the parent and all siblings. Long bouts of selfing result in individuals with completely homozygous genotypes, effectively allowing for the production of “clones.” Rivulus is also extremely sensitive to environmental change, both during development and adulthood. Life-history traits, behavior, physiology, morphology, and even sexual phenotype are shaped to a large extent by the interaction of genes with the environment, and many of these traits appear to co-vary. True reaction norms can be generated for this species in much the same way as has been done for clonally reproducing invertebrates and plants that have contributed immensely to our understanding of the evolution of phenotypic plasticity. That is, rivulus provides the opportunity to place individuals with identical genotypes in many different environments at any point during ontogeny or adulthood. In addition, rivulus populations are characterized by high genotypic diversity, a luxury not afforded by many clonal vertebrates, which allows us to evaluate variation among genotypes in the shape of reaction norms and in patterns of covariance among traits. We provide background information on phenotypic plasticity and phenotypic integration, coupled with a description of characteristics that we feel qualify rivulus as a potentially powerful model in which to study the evolution of reaction norms and covariance among traits.

Introduction

The mangrove rivulus [Kryptolebias (formerly Rivulus) marmoratus; hereafter “rivulus”], a small fish resident to the mangrove ecosystems of Florida, the Caribbean Islands, Central America and South America, possesses some of the most remarkable adaptations of any vertebrate (Taylor et al. 2004; Taylor 2012). This species is one of only two self-fertilizing hermaphroditic vertebrates, the other being its putative sister species, Kryptolebias hermaphroditus (Tatarenkov et al. 2009; Costa 2011). Many generations of selfing results in individuals that are completely homozygous and capable of producing offspring that are genetically identical to one another and to the parent (based on 36 microsatellite loci) (Mackiewicz et al. 2006a, 2006b, 2006c). Males exist, albeit at low frequencies (e.g., maximum ∼25%), and distributions of heterozygosity strongly indicate that the fish employs a mixed mating strategy—selfing and outcrossing—in most populations (Mackiewicz et al. 2006c). Which selection pressures maintain mixed mating in rivulus is unclear, and should be the subject of intense investigation in the coming years. For instance, Ellison et al. (2011) showed that parasite loads increased as a function of homozygosity in fish derived from four sites in Calabash Caye (Belize), suggesting the possibility of a Red Queen dynamic (Lively and Howard 1994).

What triggers the development of males? When embryos are reared at low temperatures (e.g., 18–20°C) males are numerically more abundant but at higher temperatures (30°C), few if any males develop (Harrington 1967). Harrington (1968) identified the thermolabile period for primary male induction as being shortly before hatching, during the final stages of embryonic development. Both light and salinity seem to interact with temperature to alter the threshold for development of males (Harrington 1967). Although primary males can be produced readily in the laboratory, the temperatures required for induction are lower than what is normally measured in the mangroves that these fish inhabit (Turner et al. 2006; R. L. Earley, personal observation), which are typically >20°C but can vary from 7°C to 38°C (Taylor et al. 1995). This has led to the hypothesis that secondary males, which result from functional hermaphrodites undergoing “sex change” or investing more in testicular tissue than in ovarian tissue (Harrington 1971), might be the more ecologically relevant type of male (Turner et al. 2006). Although the cues that trigger the development of secondary males remain somewhat of a mystery, maintaining adult hermaphrodites at higher temperatures (>26°C) can initiate the process (Turner et al. 2006). Harrington (1971) also demonstrated that sex change was age-dependent and influenced by exposure to artificial short-day seasons. Interestingly, the likelihood of “sex change” scales with the environment during early rearing; animals that experienced low temperatures during development are far less likely to undergo sex change during adulthood (Harrington 1971).

In addition to a mating system that is unique among vertebrates (androdioecy) (Mackiewicz et al. 2006c) and environmental sex determination, rivulus also is incredibly tolerant of the harsh, variable conditions, characteristic of upland mangrove habitats (e.g., hypoxia, hydrogen sulfide, broad temperature ranges) (reviewed by Taylor 2012). When conditions become extreme, rivulus employ an arsenal of behavioral and physiological responses to escape their current environment, or adjust resource allocation strategies to cope with new environments (e.g., Dunson and Dunson 1999). For example, when exposed to hypoxic conditions, oxygen-sensing neuroepithelial cells in the gills and skin enlarge (Regan et al. 2011), perhaps prompting the fish to jettison themselves from the water and breathe cutaneously when oxygen levels drop below a threshold (Grizzle and Thiyagarajah 1987). While emersed from the water, often in the moist galleries of decaying logs (Taylor et al. 2008), rivulus remodel their gills in ways that might reduce branchial water loss or provide structural support for gill lamellae (e.g., Ong et al. 2007; Turko et al. 2011). Rivulus also are quite plastic in response to their social environment, and have become a model for investigating dominance interactions and experience-dependent changes in aggressive behavior (Hsu and Wolf 1999, 2001; Earley et al. 2000; Hsu et al. 2008; Huang et al. 2011; Lan and Hsu 2011; Molloy et al. 2011; Edenbrow and Croft 2012a). Both, kinship (e.g., similar versus different genotypes) and familiarity cause rivulus to modulate their aggression towards conspecifics (Edenbrow and Croft 2012a). Previous experiences with winning and losing contests affect the rivulus’ perception of their own fighting ability and their motivation to initiate and win future aggressive encounters (Hsu and Wolf 1999, 2001). Although the effects of winning and losing are thought to be transient, recent evidence points to outcome of encounters having unexpectedly prolonged impacts on future performance during contests (e.g., at least 1 month; Lan and Hsu 2011).

A burgeoning literature on the flexibility of adult rivulus’ phenotype has, to some degree, overshadowed pioneering work on developmental plasticity in this species (Harrington 1967; Lin and Dunson 1995, 1999; see Rivulus as an Emerging Model). Elucidating how genetic variance, environmental variance, and their interaction (hereafter G × E) influence phenotypic variance is essential for understanding the evolution of individual traits, suites of covarying traits, and the ability of animals to respond adaptively to environmental conditions during development and during adulthood. In organisms that reproduce primarily by outcrossing, partitioning components of variance often requires the implementation of controlled breeding designs, such as the nested full-sib, half-sib design wherein males (sires) are mated with a number of unrelated females (dams), which results in families of full-sibs nested within half-sibs (Falconer and Mackay 1996; Lynch and Walsh 1998). Phenotypic traits are quantified in the offspring and ANOVA-based models help to resolve the sources of phenotypic variance (e.g., among sires, among dams, environmental). These types of breeding designs can be prohibitive in species with relatively long generation times (e.g., many vertebrates). However, some species such as water fleas (Daphnia sp.), nematodes (Caenorhabiditis elegans), evening primroses (Oenothera biennis), and thale cress (Arabidopsis thaliana) are uniquely suited to address questions of this sort without intricate breeding strategies; each of a diverse set of genotypes can reproduce clonally, thereby allowing researchers to place replicate individuals from each genotype in a variety of developmental environments or multiple genotypes under common garden conditions (e.g., Schlichting 1989a; Scheiner and Callahan 1999; Pigliucci 2002; see also Abzhanov et al. 2008; Braendle and Félix 2009; Miner et al. 2012).

In this prospectus, we propose the mangrove rivulus as a promising vertebrate model in which to investigate the evolution of developmental plasticity and patterns of covariance among traits (phenotypic integration). Homozygous individuals generated through selfing can effectively “clone” themselves and generations can turn over quite quickly, at least relative to most vertebrates (∼100 days for functional hermaphrodites) (Cole and Noakes 1997). Rivulus therefore, provides the same advantages mentioned above for invertebrates and plants and the power to resolve genetic, environmental, and G × E effects on morphology, behavior, physiology, life history, and disease (e.g., cancer; Lee et al. 2008) in an organism with closer phylogenetic affinity with higher vertebrates, including humans. We begin with a brief overview of phenotypic plasticity and phenotypic integration. In recent years, this literature has exploded; although we cannot provide an exhaustive review, we hope to supply enough background to excite readers about the future of research in this area. We then describe some hallmark characteristics of rivulus populations in the field and laboratory that sparked early studies on phenotypic plasticity in this species (Harrington 1967; Lin and Dunson 1995, 1999), and that we hope will inspire a new wave of empirical studies targeting both the proximate and ultimate causes/consequences of plasticity and integration.

Phenotypic diversity: Polymorphisms and plasticity

Phenotypic diversity abounds in almost every species and in almost every population, and this has led to intense interest in understanding the mechanisms that govern variation in morphology, physiology, and behavior and the maintenance of such variation through evolutionary time (e.g., Gross 1996; Kassen 2002; Roulin 2004; Rueffler et al. 2006; Schwander and Leimar 2011). Genetic polymorphisms underlie discrete variation in body size, reproductive tactics, and color in many vertebrate systems (e.g., Borowsky 1987; Kallman 1989; Sinervo and Zamudio 2001; McKinnon and Pierotti 2010), and this variation is probably maintained by frequency-dependent selection. For instance, in the side-blotched lizard (Uta stansburiana), three heritable color morphs exist in males and each morph adopts a different mating strategy (Sinervo and Lively 1996). The frequencies of the three morphs cycle in a dynamic, negative frequency-dependent fashion reminiscent of the “rock-paper-scissors” game (Bleay et al. 2007).

Phenotypic diversity also results from interactions between the genome and the environment such that a single genotype can produce different phenotypes depending on the conditions experienced during development (West-Eberhard 2003). The full gamut of phenotypes that a single genotype might produce along an environmental gradient is referred to as a “reaction norm.” Reaction norms can be continuous or discontinuous (Fig. 1), and the form they adopt is likely to depend on the mechanisms underlying phenotypic expression. Inducible polyphenisms are observed when threshold environmental conditions (e.g., above a certain temperature) govern the expression of one discrete phenotypic state over another (Roff 1996; Oostra et al. 2011; reviewed by Moczek et al. 2011; Gilbert 2012; Fig. 1a). Examples are the emergence of “supersoldier” castes, worker castes, and queens in the ant genus Phelidole as a function of nutritional and hormonal inputs during development (Rajakumar et al. 2012); helmeted and spined morphs of Daphnia sp. that develop in response to kairomones released by hemipteran and dipteran predators (e.g., Parejko and Dodson 1991; Lüning 1992); temperature-dependent sex determination in fishes and reptiles (e.g., Harrington 1967; Quinn et al. 2011); and the incredible morphological and behavioral shifts accompanying sex change in response to social cues in various fish species (e.g., Munday et al. 2006). In contrast to polyphenisms, many phenotypic characters including behavior and the life-history traits of both plants and animals, scale more continuously with environmental gradients experienced during development (e.g., Pigliucci and Schlichting 1998; Huber et al. 2004; Ghalambor et al. 2007; Stamps and Groothuis 2010; Champagne 2011; Fig. 1b).

Fig. 1.

Discrete and continuous reaction norms for individual traits (a, b) and suites of integrated traits (c, d). The ball-and-stick diagrams in (c) and (d) represent covariances (lines) among different phenotypic characteristics (circles) where integration, not values of specific traits, increases along the y-axis and changes as a function of the environment. The degree of integration may change among a certain fixed set of traits (connectivity) or that different traits may be recruited into, or lost from the integration network (composition). Panels (a) and (c) show the induction of a polyphenism in response to threshold environmental conditions (dashed line). Different genotypes might exhibit more, or less, sensitivity to induction at a given point along the environmental gradient; thus, the dashed line can move along the environmental gradient in a genotype-specific fashion. Panel (c) depicts a situation in which the discrete polyphenisms are characterized by different degrees of integration of a trait. Panels (b) and (d) show continuous phenotypic responses to an environmental gradient that differ between two hypothetical genotypes. In panel (b), one genotype (solid line) shows a steep response to the environmental gradient, whereas the other (dashed line) is relatively insensitive to environmental input; these reaction norms could very well be nonlinear. In panel (d), the two genotypes (gray versus white circles) exhibit different patterns of integration (or disintegration) along an environmental gradient.

Although the developmental environment exerts profound and often irreversible effects, an organism’s phenotypic fate certainly is not sealed early in life. Phenotypic flexibility such as dramatic adjustments in physiology, behavior, and morphology continue through adulthood in response to changing seasons, feeding regimes, social conditions, predation pressures, and demands for energy (e.g., Piersma and Drent 2003). For instance, mangrove rivulus (Kryptolebias marmoratus) remodel their gills in response to prolonged exposure to air (Ong et al. 2007; LeBlanc et al. 2010); Burmese pythons (Python molurus) reorganize intestinal physiology and morphology in response to episodes of feeding and fasting (Secor 2008); many vertebrates exhibit seasonal changes in reproductive and stress physiology, gonadal state, morphology, and behavior in response to photoperiod and the onset of breeding (Wingfield 2005; Goymann et al. 2007; Hahn et al. 2008; Migaud et al. 2010); and the outcome of social interactions—dominance or subordination—transforms individual neural, hormonal, and behavioral states (Hsu et al. 2006; Oliveira 2009; Maruska et al. 2010). Such labile phenotypes also can be modeled as reaction norms (e.g., Fuller et al. 2005; Dingemanse et al. 2010).

Phenotypic plasticity as a quantitative trait

If plasticity is a quantitative trait on which selection can act, we expect to observe genetic variation in the threshold for induction of polyphenisms (Fig. 1a) or in the slope of continuous reaction norms, (Fig. 1b). We also expect to observe differences in fitness among genotypes (genotype-specific reaction norms) when exposed to relevant selection pressures (Stearns 1989; Scheiner 1993; Via et al. 1995; de Jong 2005; Ghalambor et al. 2007). There is mounting evidence for genetic variation underlying both discontinuous (polyphenisms) (Braendle et al. 2005; Buzzato et al. 2012) and continuous (Miller and Fowler 1993; Postma and van Noordwijk 2005; Winterhalter and Mousseau 2007; Hutchings 2011; Rabus et al. 2012) developmental reaction norms, as well as behavioral reaction norms exhibited by adults (Scotti and Foster 2006; Dingemanse et al. 2012a, 2012b). Fewer studies have capitalized on knowledge of genetic variation in plasticity to test predictions about the relative fitness of different reaction norms in alternative environments (e.g., Kingsolver 1995; Scheiner and Callahan 1999; van Kleunen and Fischer 2003; Johnson 2006; Reed et al. 2009).

Phenotypic plasticity should be favored in heterogeneous environments (see also Schlichting and Smith 2002) because it allows organisms to “match” their phenotype to prevailing environmental conditions in ways that maximize fitness. If the environment changes unpredictably or too regularly (e.g, within a generation), organisms whose phenotype was irreversibly matched to developmental conditions might suffer considerable fitness costs as adults if placed in a new environment (e.g., Bateson et al. 2004; Gluckman et al. 2005; Monaghan 2008). This, perhaps, is why selection might favor behavioral and physiological flexibility during adulthood—as a mechanism that buffers the potential costs of a mismatch between phenotype and environment (DeWitt et al. 1998; Callahan et al. 2008; Auld et al. 2010). Additional costs of developmental plasticity include the maintenance of the genetic, hormonal, or sensory machineries that translate environmental cues into phenotypic expression; linkage disequilibrium involving “plasticity genes” and genes that have negative consequences for fitness; and negative pleiotropic effects of “plasticity genes” on other fitness-related traits (DeWitt et al. 1998). It is worth noting that, at present, evidence for the costs of plasticity is equivocal (e.g., Scheiner and Berrigan 1998; DeWitt 1998; Weijschedé et al. 2006; see Auld et al. 2010). However, armed with an understanding of the genetic variation underpinning reaction norms, we can estimate the costs and benefits—ultimately, fitness trajectories—for a given reaction norm in any number of relevant environments (e.g., Kingsolver 1995; Scheiner and Callahan 1999).

Phenotypic integration—multivariate “traits”

Phenotypic traits often are functionally related to each other or are regulated by the same genes or physiological processes (e.g., Schlichting 1989b; Roff and Fairbairn 2007). Thus, we often observe strong correlations and nontrivial genetic covariances among traits (e.g., Berg 1960; Lande and Arnold 1983; Klingenberg et al. 2004; Blows 2007; Øverli et al. 2007; Sih and Bell 2008; Calsbeek et al. 2010; Martin et al. 2011; Pryke et al. 2012). Various disciplines refer to suites of correlated traits in different ways—integrated phenotypes, multivariate phenotypic axes, coping styles, behavioral syndromes (see references in previous sentence). As a scientific community, we are steadily moving away from viewing the phenotype as a set of atomized traits and toward a multivariate approach to the study of phenotypic evolution and plasticity (e.g., Pigliucci 2003; Dochtermann and Roff 2010; McKinnon and Pierotti 2010; Santos and Cannatella 2011). This is critically important because, our understanding of the dynamics and directionality of phenotypic evolution changes considerably when viewed through a multivariate lens (Phillips and Arnold 1989; Arnold 2005; Arnold et al. 2008; Hansen and Houle 2008; McGlothlin and Ketterson 2008; Ketterson et al. 2009).

Both artificial and natural selection can drive the evolution of multivariate character states (e.g., Brodie 1992; Arnold and Phillips 1999; Waldmann and Andersson 2000; Murren et al. 2002; Roff and Mousseau 2005; Johnson et al. 2009; Lewis et al. 2011), suggesting a strong genetic component to correlations among traits. Genetic pleiotropy and past correlated selection on heritable, functionally related traits are seen as the primary causal agents of phenotypic integration (Schlichting 1989b; Blows 2007; McGlothlin and Ketterson 2008). Apparent integration can result from the correlated responses of two otherwise independent traits to a shared environment and thus, must be ruled out before any conclusions about the evolvability of integration are made (Schlichting 1989b; see Hansen et al. 2011). Some authors have speculated that highly integrated phenotypes (i.e., strong correlations among traits) might constrain phenotypic plasticity and thus, adaptive responses to environmental change (e.g., Schlichting 1989b; Ketterson et al. 2009). Nevertheless, few studies have explicitly addressed the relationship between integration and plasticity. Gianoli and Palacio-López (2009), for example, studied two perennial plant species (Convolvulus chilensis and Lippia alba) and demonstrated a strong negative relationship between the number of significant phenotypic correlations and the “magnitude” of phenotypic plasticity, that is, slope of a reaction norm across light environments (C. chilensis) or percent change in expression of traits following drought (L. alba).

Plasticity of integration

De Jong (2005, p. 103) pointed out that: “selection on phenotypic plasticity is always selection on more than one character.” Therefore, reaction norms are perhaps best viewed as changes in correlations among traits (or the degree of phenotypic integration) along an environmental gradient experienced during development or as an adult (Fig. 1c and d). Were different genotypes to exhibit divergent reaction norms (Fig. 1d), it would be grounds for investigating the evolution of plasticity in correlations among traits. To our knowledge, this has been investigated in only a few studies, mostly on plants (Schlichting 1989a; Pigliucci et al. 1995; Brock and Weinig 2007; Tonsor and Scheiner 2007; Mallitt et al. 2010). For example, Tonsor and Scheiner (2007) examined changes in the relationships among 14 traits in 35 A. thaliana genotypes exposed to four atmospheric CO₂ concentrations during development. They found significant genotypic variation in the response of trait integration to atmospheric CO₂ concentrations and, importantly, showed that patterns of trait correlations changed despite the overall degree of integration remaining relatively constant. Schlichting (1989a), on the other hand, showed dramatic changes in both the pattern and degree of trait correlations in Phlox sp. grown in variable environments, whereas Mallitt et al. (2010) showed constancy of the genetic variance–covariance matrix across gradients of availability of light and water in pepper grass (Lepidium bonariense). Despite the paucity of studies of this sort in vertebrates, Gonzalez et al. (2011) demonstrated significant changes in the covariance structure of skulls of rats that experienced nutritional stress during development (i.e., retardation of intrauterine growth, in which uterine vessels were experimentally obstructed) relative to controls. In addition, Sih (2011) presented a conceptual overview of how stressors experienced early in life might alter the architecture of behavioral syndromes—consistent correlations among behaviors across contexts or times (see Chang et al. 2012).

How can we know whether selection has, or might, act on integrated phenotypes and their plasticity? How can we know whether patterns of integration change across developmental environments? These questions have captured the attention of quantitative geneticists for quite some time (Lande and Arnold 1983; Lynch and Walsh 1998; Blows 2007; Roff et al. 2012), and a number of multivariate methods have been employed to examine how the structure of phenotypic (P-matrix) or genetic variance–covariance matrices (G-matrix) differs among populations of the same species, throughout ontogeny, or in response to different developmental conditions (Roff and Mousseau 2005; reviewed by Roff et al. 2012). Some of these methods also are becoming more accessible to researchers with little background in quantitative genetics or with little computational savvy (e.g., Wilson et al. 2009; see Roff et al. 2012 who provided open access to R functions for comparisons of G-matrix). Because we cannot possibly do justice to these incredibly powerful statistics herein, we refer the interested reader to the references in this section for elegant descriptions of both the analytical tools and the ways in which these tools are deployed to understand how multivariate selection operates.

Rivulus as an emerging model

An ideal vertebrate model in which to study the evolution of phenotypic plasticity and integration might be expected to possess a number of hallmark characteristics. Below, we list those characteristics and supporting evidence from studies on the mangrove rivulus.

Evidence that phenotypic diversity results from ecologically relevant variation in the rearing environment

As described above, combinations of salinity, temperature, photoperiod, and light intensity ultimately dictate sexual phenotype under laboratory conditions (Harrington 1967, 1968, 1971; Turner et al. 2006). In addition, a pair of studies by Lin and Dunson (1995, 1999) revealed that life-history traits such as growth rates, fecundity, egg volume, and age at maturity change as a function, both of salinity and food availability experienced from 5 days posthatching through adulthood. Both studies also indicated strong “maternal” effects of parental food rations on the growth rates and fecundity of their offspring. Salinity, temperature, light intensity, food availability, and a host of other factors (e.g., composition of leaf litter, availability of refuges) vary temporally and spatially both within and among mangrove sites (e.g., Taylor 2000; Richards et al. 2011). Thus, although the aforementioned studies were conducted under controlled laboratory conditions, they demonstrate that rivulus exhibits developmental plasticity in response to ecologically relevant environmental variations.

Evidence for correlations among phenotypic traits.

Lin and Dunson (1995) revealed that body mass, body length, and age at maturity decreased as a function of increased salinity experienced during development, particularly when fish were fed large rations. However, fecundity showed the opposite trend, suggesting a potential trade-off between somatic and reproductive investment (but see Grageda et al. 2005 for a positive correlation between growth and reproductive effort). A number of recent studies have explicitly addressed correlations among behavioral traits in rivulus. Edenbrow and Croft (2011) showed correlations between exploration (e.g., distance traveled in a novel maze) and boldness (e.g., time to recover from simulated predatory attack) beginning at 61 days posthatching. In a second study, Edenbrow and Croft (2012b) demonstrated that animals which recover quickly from a simulated predatory attack also exhibit more aggression toward a mirror-image stimulus. This correlation existed for secondary males but not for hermaphrodites, suggesting that environmental sex determination can shape correlations among traits (Edenbrow and Croft 2012b). Chang et al. (2012) also showed correlations among exploration, boldness, and aggression, but not performance in a learning task. The first three behaviors also were positively correlated with baseline testosterone levels, indicating that steroid hormones, which are highly pleiotropic in their actions (Ketterson et al. 2009), might underpin correlations among traits (Chang et al. 2012).

Evidence for genetic variance underlying phenotypic diversity

Most studies that include genotype as an independent variable have uncovered significant variance among homozygous lineages in behavior, life history, morphology, and physiology. For instance, Earley and Hsu (2008) demonstrated significant variation among genotypes in the reinforcement of dominance (e.g., attacking an opponent after it had retreated) and in the postcontest endocrine responses of eventual winners and losers. There also is evidence for genotypic differences in the ontogenetic trajectories of boldness and exploratory behavior, and for aggressive behavior (Edenbrow and Croft 2011, 2012b). A number of studies have investigated genotype-level differences in life-history traits. Lin and Dunson (1995) showed genotype effects on egg volume and hatchling size, Edenbrow and Croft (2011) revealed genotypic effects on age at maturity and on the total number of eggs laid between 105 and 151 days posthatching, and Grageda et al. (2005) described genotypic effects on fecundity and growth rates (0–100 days posthatching). Conclusions about genotype-level differences in life history, behavior, and physiology seem contingent upon the number and identity of the genotypes compared; for instance, Grageda et al. (2005) showed minimal differences in morphological characters between Panamanian and Belizean lineages, contrasting Taylor’s (2003) evidence for marked differences in a similar set of characters among populations/genotypes derived from various locations in Florida, Belize, Honduras, Bahamas, and Brazil. We should therefore, expand the number of genotypes used in our studies, and perhaps select many genotypes from each of a number of populations that span the fish’s geographic range (e.g., Edenbrow and Croft 2011).

Evidence for genetic variance underlying diversity in reaction norms

Lin and Dunson (1995) showed significant G × E effects in their overall multivariate analysis of reproductive traits; the influence of genotype × salinity, genotype × food ration, and genotype × salinity × food ration were all highly significant but the strength of these effects was reduced for univariate analyses on each reproductive trait. Lin and Dunson (1999) also showed significant genotype × temperature and genotype × food interactions on growth rates. Harrington’s (1967) study on temperature-dependent sex determination revealed among-genotype differences in the percentage of males that were produced at low temperatures (18–20°C). Turner et al. (2006) demonstrated that genotypes derived from Twin Cays, Belize—a site with high abundance of males—produce significantly more secondary males (range from 25% to 100% of F1 offspring) than do genotypes derived from Dangriga, Belize (0–18% with one genotype producing 45%). These differences persisted into the second generation, providing evidence that the threshold for inducing secondary males has a strong genetic component and varies markedly among populations but to a lesser extent among genotypes within a population (Turner et al. 2006). Together, these data provide strong support for plasticity as a quantitative trait in the mangrove rivulus—both discrete and continuous reaction norms appear to have a genetic basis and thus, have the potential to be shaped by selection pressures yet to be identified. We now face the exciting challenge of placing such studies in a firm quantitative genetics framework so that we can evaluate, for instance, the strength and directionality of selection on single traits or, preferably, suites of phenotypic characters.

Evidence for population genetic structure on various scales

Strong population structure provides at least some evidence for restricted gene flow and the potential for local adaptation of, for instance, the shape of reaction norms and the extent of phenotypic integration. Two studies by Tatarenkov et al. (2007, 2012) provide convincing evidence of population structure at scales ranging from few to many hundreds of kilometers. Using 33 polymorphic microsatellite loci, Tatarenkov et al. (2007) demonstrated robust segregation among lineages from Belize, Bahamas, and Florida and, even within Florida, clusters formed for lineages derived from southwestern, western, and eastern populations. The analysis by Tatarenkov et al. (2012) of more than 200 specimens originating from 12 populations in the Florida Keys also showed pronounced population structure at the microgeographic scale. In addition to population structure, species with relatively large ranges provide the opportunity to examine broad geographic patterns of selection. The ranges of many parthenogenetic vertebrates are quite restricted, perhaps because all-female populations rely on sperm donors to activate egg development, typically from one of the parental species that hybridized to generate the parthenogen (e.g., Lampert 2009). On the contrary, rivulus’ range is extensive—from Florida and the Caribbean to Central America and South America (Taylor 2012).

Evidence for genotypic diversity within a population

Many clonal vertebrates exist as genetically uniform populations (e.g., Crews and Fitzgerald 1980; Elder and Schlosser 1995). In these species, genotype is ultimately confounded with geographic locale, which prohibits researchers from discriminating between genotype-level and population-level effects in, for instance, a common garden experiment. However, some asexually reproducing species exhibit clonal diversity within a population (e.g., Vrijenhoek 1978, 1979) and the maintenance of such diversity has been the subject of intense empirical and theoretical interest (e.g., frozen niche variation model) (Joekela et al. 1997; Semlitsch et al. 1997; Gray and Weeks 2001; Pagano et al. 2008). Rivulus does not reproduce asexually but homozygous individuals effectively propagate a “clonal” lineage. Within a population, however, there is extreme clonal diversity (e.g., Turner et al. 1992), which is thought to arise via a combination of de novo mutation, immigration, and outcrossing between native and newly arrived genotypes (e.g., Tatarenkov et al. 2007). What maintains such high “clonal” diversity within isolated populations over evolutionary time, however, remains somewhat of a mystery. Phenotypic plasticity, particularly differences among genotypes in the shape of reaction norms, and diversifying selection in highly spatiotemporally variable environments such as mangroves, have been proposed to explain high within-population genetic diversity in the predominantly selfing rivulus (Lin and Dunson 1995; Tatarenkov et al. 2012).

Replication at the genotypic level

Perhaps the most attractive aspect of rivulus’ biology is its ability to produce offspring that are genetically identical to the parent and all siblings. This allows for replication at both the genotypic and population levels when conducting studies that evaluate diversity in the shape of reaction norms. Most studies of developmental plasticity in vertebrates rely on population-level reaction norms, which essentially averages the reaction norms for all genotypes within a population and assumes: (1) that most of the population’s composite genotypes converge on that average; and (2) that alleles are distributed equally among environmental treatments. This is the best we can do for species in which genotypic replicates are not abundant (e.g., beyond monozygotic twins), and for which controlled breeding designs in a laboratory setting are either impractical or impossible. However, to construct “true” reaction norms and to critically evaluate the evolution of phenotypic plasticity requires: (1) that genetically identical animals be placed in different developmental environments; and (2) that there is enough genotypic diversity (e.g., many different homozygous lineages, in the case of rivulus) to assess both variation in the shape of reaction norms and the potential for selection to act on plasticity.

Generating crosses in the laboratory

If we discover significant genotype-level variance in a host of phenotypic traits, including plasticity, “hybrid” crosses would prove essential for resolving the genetic basis of these traits. Crosses between distinct homozygous lineages of rivulus have been successfully generated by artificial insemination (Harrington and Kallman 1968; Nakamura et al. 2008) and by placing senescent hermaphrodites, who tend to lay unfertilized eggs, with secondary males (Harrington and Kallman 1968; Mackiewicz et al. 2006a). Of particular interest was the finding by Nakamura and colleagues that the growth rate of the hybrid F2 generation was intermediate between the two parental lineages. Generating crosses between homozygous lineages also might facilitate our understanding of the relationships between inbreeding, phenotypic plasticity, and fitness. One of the most curious aspects of rivulus’ biology is that many populations consist almost exclusively of homozygous lineages that have been maintained through self-fertilization (inbreeding) for many generations (Mackiewicz et al. 2006b). Although there is some evidence that homozygotes might suffer fitness costs in the form of increased susceptibility to parasites (Ellison et al. 2011), homozygosity does not appear to have uniformly detrimental consequences. For example, Molloy et al. (2011) demonstrated that homozygous males were smaller than heterozygous males but were equally likely to win aggressive contests. To date, however, no study on rivulus has explored whether homozygotes differ from heterozygotes in their response to environmental variation. With rivulus, we have the unique opportunity to examine the potential for inbreeding depression, a reduction in the fitness of inbred animals relative to outbred counterparts, in both wild-caught animals and under controlled laboratory conditions where particular crosses can be generated. We also have the opportunity to determine whether: (1) homozygotes are less able than heterozygotes to appropriately buffer the phenotypic response to environmental perturbation (Debat and David 2011)—that is, whether homozygotes are more likely to exhibit maladaptive phenotypic plasticity, or (2) homozygotes show a reduction in the capacity to express adaptive phenotypic plasticity when compared with heterozygotes (e.g., Auld and Relyea 2010).

Conclusions

Although the aforementioned list of characteristics is not exhaustive, the mangrove rivulus certainly appears to fit the mold of a species that can make powerful impacts on our understanding of the evolution of phenotypic plasticity and integration. Nevertheless, it should be quite obvious that we have a long road ahead before fully understanding the biology of the mangrove rivulus. Contributions by others in this symposium highlight how much, as well as how little we know, particularly about the habits, physiology, and life history of these animals in their natural environs. Such information is critical if we are to establish a solid empirical foundation for investigating the evolution of phenotypic plasticity and integration. It is therefore essential that we recruit sharp young minds to tackle broadly integrative questions using a fusion of significant field and laboratory components. Furthermore, Kelley et al. (2012) have begun to develop an arsenal of genomics resources that will undoubtedly serve as a springboard for exploring the genomic underpinnings of phenotypic variation within and among genotypes, populations, and broad geographic locales. As a scientific community, we are likely to face challenges about the general applicability of our findings in rivulus to other vertebrates that have more conventional reproductive strategies. We reject the notion that we must study mainstream organisms to drive conceptual and theoretical advances in fields such as evolutionary biology, physiology, and behavior. Indeed, many model organisms are used “because” their unique biology allows us to view established theories/concepts through a new lens or to design more powerful experiments to address fundamental questions (Satterlie et al. 2009). Rivulus is one such organism and is no stranger than, for instance, water fleas, nematodes, or thale cress, all of which have contributed immensely to a variety of biological disciplines.

Funding

Funding for the symposium: NIH Conference Grant (R13HD070622) from the Eunice Kennedy Shriver National Institute of Child Health & Human Development; SICB through the DCE, DCPB, DAB, and the C. Ladd Prosser Fund; the College of Agriculture and Natural Resources, University of Maryland. This work was also supported by a Whitehall Foundation (Grant 2007-12-79), University of Alabama Research Grants Committee; Fisheries Society of the British Isles Research grants (to R.L.E.); the Howard Hughes Medical Institute; National Science Foundation Graduate Research Fellowship (DGE-1058257 to A.F.H.).