Using the Evolutionarily Significant Unit (ESU) to Define Breeding Units

Identifying population structure relevant to conservation and management has centered on the concept of evolutionarily significant units (ESU) (reviewed in Crandall et al. 2000; Fraser and Bernatchez 2001). ESUs are reproductively or historically isolated groups (i.e., lineages) that represent the adaptive distinctiveness found within a species (Moritz 1995; Crandall et al. 2000). Ideally, these distinct evolutionary lineages contain a “reservoir upon which future evolutionary potential depends” (Waples 1995).

There is general consensus regarding the existence of ESUs in nature, but there has been debate in the scientific literature regarding how they should be defined (reviewed in Crandall et al. 2000; Fraser and Bernatchez 2001). To fully capture a species' evolutionary potential, ESUs are best defined on the basis of ecological, life history, morphological, and genetic similarity. Ecological and life history data are invaluable when managing a species of concern, but often the collection of these data does not become a priority until a species is already in decline. Thus, comparative range-wide data of these types are rarely available for defining and managing ESUs. A diagnostic morphological character or suite of characters would provide a fast and efficient guide for defining units, yet determining which characters are appropriate unambiguous indicators of distinct evolutionary units is more problematic (Hey et al. 2003). A number of turtles have concordant phenotypic and genetic structure suggesting morphology could be used as a proxy to define ESUs (e.g., Walker et al. 1997, 1998b; Stuart and Parham 2004). However, several morphologically defined subspecies of turtles and tortoises show a marked lack of genetic differentiation and serve as counter examples (e.g., Phillips et al. 1996; Caccone et al. 1999). For instance, the 12 distinct morphologically defined species of Graptemys in the United States comprise only 3 distinguishable genetic groups (Lamb et al. 1994). In contrast, alligator snapping turtles (Macroclemys temminckii) exhibit little morphological variation across their range, and yet genetic methods have uncovered a high degree of divergence among populations (Roman et al. 1999). As a whole, these studies suggest the sole use of phenotypic characters may often be too coarse a guide for identifying ESUs a priori (Hey et al. 2003).

Because highly restricted gene flow is central to all ESU definitions, the coordinated use of neutral genetic markers is often the most straightforward approach for detecting population structure and defining ESUs (Moritz 1999; Fraser and Bernatchez 2001). Highly restricted gene flow increases the likelihood of local adaptation and differentiation by drift (Lenormand 2002), and so managing these units separately should help to conserve both unique sets of alleles and local adaptations (Moritz 1999; Fraser and Bernatchez 2001). A review of phylogeographic studies suggests that definable genetic units are common among tortoises and freshwater turtles and typically subdivide a species' range into several large regional units that could be considered putative ESUs (Table 1). For example, the gopher tortoise (Gopherus polyphemus) is found across much of the southeastern United States and analyses of mitochondrial DNA bisect the species' range at the Applalachicola river basin into 2 large units with a minor genetic unit found in western Florida (Osentoski and Lamb 1995). Differentiation among distinct units was typically attributed to historic isolation resulting from glacially induced sea level changes or the presence of significant biogeographic barriers (Table 1). Because these phenomena rarely influence only a single species, concordant patterns among sympatric species can be used to identify biogeographical provinces applicable to unstudied species facing critical management decisions (Avise 1996; Walker and Avise 1998).

Table 1. Summary of published tortoise and freshwater turtle studies of geographic genetic variation.

View Fullscreen

Table 1. Extended.

View Fullscreen

We suggest that ESUs based on molecular markers will be the only way to objectively sort turtles and tortoises into breeding units whose geographic origins are unknown, especially for those species with large or fragmented geographic ranges. This is not to say that genetic methods are a panacea, since local adaptation can take place in the presence of gene flow (i.e., low differentiation between populations at molecular markers) and different phylogenetic methods may yield different groupings (Crandall et al. 2000; Fraser and Bernatchez 2001; McKay and Latta 2002; Sanderson and Shaffer 2002). Most of the world's turtles and tortoises that are of conservation concern, however, are from subtropical and tropical regions for which there is little or no information on phylogeographic structure, much less variation in adaptive differences with which to construct ESUs. For instance, only 6 (19%) of the 32 turtle and tortoise species (assuming 3 Graptemys species) listed in Table 1 are found outside temperate regions, highlighting the paucity of basic data for species in these ecosystems.

Minimally, assurance colonies should be structured to reflect a species' genetic ESUs and in some cases it may be possible to further subdivide ESUs with high resolution markers (see below) to create more fine-scale breeding units that should be even more closely matched to a given geographic area. We recommend that individuals should only be interbred if they are from the same ESU, but if the number of breeding individuals is so low as to make this strategy unfeasible, breeding decisions should be based on the relatedness of different ESUs (see section II).

The Utility of Mitochondrial and Microsatellite Loci

Mitochondrial DNA (mtDNA) haplotypes are useful for discerning historic patterns of gene flow and major genetic lineages (i.e., define ESUs), while highly polymorphic nuclear markers (e.g., microsatellites, amplified fragment length polymorphisms) are required to understand more recent patterns of interpopulation gene flow (i.e., reproductive subunits) (reviewed in Avise 2004). The difference in resolution is partially explained by characteristics of the 2 genetic markers. MtDNA is passed to offspring via the mother's egg and therefore exhibits a uniparental mode of inheritance free from the complications of chromosomal recombination. Estimates of the mutation rate in the turtle mitochondrial genome are generally slow, ranging almost an order of magnitude from ca. 0.15% to ca. 1% sequence divergence per million years (Avise et al. 1992). In contrast, microsatellite loci are short tandemly repeated segments of nuclear DNA that are biparentally inherited and mutate at a rate 10 to 100 times faster than mtDNA. These characteristics make microsatellite loci ideal for assessing parentage and fine-scale geographic patterns because they mutate rapidly enough to retain a record of mating success and recent gene flow, but not so rapidly that they destroy the underlying signal from generation to generation.

MtDNA work in turtles has been extensive over the past 10 years (e.g., Walker and Avise 1998) and has been very successful at detecting broad-scale genetic structure (Table 1). Only a few studies to date have examined geographic patterns of microsatellite variability in freshwater and terrestrial turtle species (Table 1). Microsatellite markers have detected a significant drainage effect in the Amazonian river turtle (Podocnemis expansa) (Sites et al. 1999) and regional population differentiation in the Galapagos tortoise (Geochelone nigra) (Beheregaray et al. 2003). Microsatellite genetic structure has also been observed in globally ranging sea turtles (FitzSimmons et al. 1997).

Using Genetic Markers to Define ESUs and Reproductive Subunits

Before we can define ESUs or reproductive subunits within ESUs, we need genetic markers for the target species. Most of the mtDNA genes, including the more rapidly evolving control region, have been sequenced for various turtle species (for a partial list see Table 1). Sequencing the control region is most appropriate when a species is well defined, but when species relationships are unclear or the control region sequence proves to be highly variable within a species, more slowly evolving mitochondrial genes should also be analyzed. It should be noted that phenomena such as hybridization and incomplete sorting of ancestral polymorphisms can cause analyses based on a single gene to be misleading with regard to lineage relationships (Maddison 1991; Rognon and Guyomard 2003). Comparing sequence analysis at multiple, unlinked loci (e.g., nuclear and mitochondrial genes) or the results of mtDNA and microsatellite analyses can help uncover such issues if they are confounding. Over 120 microsatellite loci have been developed for turtles and many of these markers have shown useful levels of polymorphism in other species (e.g., FitzSimmons et al. 1995; Osentoski 2001; Osentoski et al. 2002; 2005 Turtle Genetics Workshop). A variety of recent techniques have increased the speed and efficiency of isolating numerous polymorphic microsatellite loci (e.g., Zane et al. 2002), and the need to develop species-specific markers will decrease as the catalogue of turtle-specific markers grows.

Once loci have been selected, DNA is extracted from blood or tissue samples of known (when available) and unknown individuals for sequencing (mtDNA) and genotyping at microsatellite loci (nuclear DNA) (Fig. 1, part I). Because both methods are based on the polymerase chain reaction (PCR), only a small amount of blood or tissue is required to obtain sufficient DNA for both analyses. From this point forward, the process depends upon the number of individuals involved in colony formation (Fig. 1, part I). Because there are limits to the minimum size of an interbreeding unit within a colony (see Section II) and the power of resolution for microsatellite analyses (see below), we propose that breeding groups be based on ESUs defined solely from mtDNA data if there are fewer than ca. 50 individuals of a species. This does not mean microsatellite genotypes are not useful for these species; loci could still be required to monitor breeding and estimate relatedness within colonies (Fig. 1, part II).

A variety of phylogenetic methods are available to group animals based on their mtDNA sequences and thereby establish the presence or absence of major genetic groups (i.e., ESUs) (Hillis and Moritz 1996; Sanderson and Shaffer 2002). There has been debate regarding the requirement that genetically defined ESUs exhibit reciprocal monophyly sensu the Moritz (1995) definition (Hey et al. 2003; Kizirian and Donnelly 2004), but we feel in such a triage situation and in the absence of other data that this may be too restrictive a criterion and that all well-supported genetic clades be considered as separate breeding units when establishing assurance colonies. Due to the vagaries of phylogenetic analysis, less familiar workers are encouraged to collaborate with investigators experienced with such analysis.

If a large number of animals are available, microsatellite genotype data can be used to 1) independently validate the mtDNA based ESUs, and 2) determine if there are further population subdivisions within the ESUs which could potentially be managed separately (Fig. 1). New Bayesian statistical methods have been developed that can group individuals of unknown origin into population groups using multilocus genotype data making them particularly useful for groups of rescued animals (Pritchard et al. 2000; Dawson and Belkhir 2001). These and other related “assignment” methods can also be used to assign individuals to known populations and identify hybrids and subspecies groupings (reviewed in Manel et al. 2005). Both simulation and empirical data have indicated that these approaches will be most powerful for detecting the presence of population structure when using 10 or more polymorphic loci, populations are well-differentiated (F_ST ≥ 0.1), and are represented by a relatively large number of individuals (e.g., 20 or more individuals from each putative population) (Cornuet et al. 1999; Rosenberg et al. 2001; Manel et al. 2002; Evanno et al. 2005). Several methods currently exist that can be used to select loci with the most discriminatory power to use in assignment tests (Banks et al. 2003; Topchy et al. 2004).

These methods will allow turtles and tortoises to be grouped into appropriate breeding programs within assurance colonies. As new animals are obtained, they could be assigned to their respective breeding groups using either mtDNA sequence data or multilocus microsatellite genotypes. The geographic origin of major genetic groups in assurance colonies could also be identified post hoc by sampling and including known-locality specimens from the wild, museums, or zoos in additional analyses (e.g., Burns et al. 2003) (Fig. 1, part I). Because of its relatively slower rate of change relative to nuclear markers, mitochondrial markers are more likely to exhibit regional diagnostic differences and therefore will be most useful for this task. Multilocus assignment methods can also be used to assign individuals to their geographic origin, but relatively large numbers of individuals would need to be sampled across their geographic range to accurately estimate population specific allele frequencies.

Maintaining Genetic Variability in a Breeding Program

The effective population size (N_e) represents the actual number of breeding individuals in the population, adjusting for the sex ratio of breeders, variation in number of offspring produced, and variation in the number of breeders over time (generations). A population is expected to lose 1/(2N_e) of its genetic variation every generation; therefore as N_e is increased, less variation is lost (e.g., Lande and Barrowclough 1987). In general, factors that act to skew or increase the variance in the number of breeders, sex ratio, offspring produced, and population size will lower N_e and accelerate the loss of genetic diversity. Because variation in these factors is common, N_e is typically only 10%–30% of the census size (N_c) in both wild and captive populations (Frankham et al. 2002).

A goal often cited by captive conservation programs is the maintenance of 90% of the original founders' genetic variation and the avoidance of inbreeding depression for 100–200 years (Frankham et al. 2002). The target population size required to preserve a given amount of genetic variation depends on the number of available founders, generation length, and the rate of population growth (Frankham et al. 2002). Animal breeders have determined that inbreeding problems can be avoided if the rate of inbreeding per generation is below 2%. Assuming a conservative 1% loss of genetic variation per generation, N_e would have to be at least 50 (i.e., 1/2N_e = rate of inbreeding) to avoid inbreeding depression (Soule et al. 1986). Lande (1995a) has modeled the target population size required for a captive breeding program to maintain 90% of its initial level of genetic variation for 200 years. For instance, assuming random genetic drift and realistic mutation rates, a breeding program with a founding population of N_e = 30, a population growth rate (measured as r, the intrinsic rate of natural increase) of 0.2 (many zoo populations have growth rates between 0.05 and 0.2), and a generation length of 10 years, it will be necessary to reach and maintain a target population size (N_e) of a little over 100 individuals. If generation length is increased to 20 years, the target N_e required to meet this goal drops to ca. 50 (Lande 1995a).

Estimated population sizes, like those mentioned above, are not “magic” numbers that once achieved will assure a species' persistence in perpetuity (Lande 1988). For instance, maintaining demographic stability may require many more individuals than predicted on genetic grounds alone. It has also been estimated that effective population sizes on the order of 5000 or more are required to maintain single locus polymorphisms that may be important in disease resistance and to avoid the long-term accumulation of mildly deleterious mutations (Lande and Barrowclough 1987; Lande 1995b; Lynch and Lande 1998). Obviously, it is not realistically possible to manage this number of individuals and a more practical approach will be to maintain as much genetic variation as possible.

One Large Colony or Many Small Colonies?

Both theoretical and experimental studies predict that genetic variation can be maintained at higher levels if captive breeding programs manage species as a series of subpopulations (i.e., a network of assurance colonies) (Lacy 1987; Lande 1995a; Margan et al. 1998). This strategy takes advantage of the random nature of genetic drift. Gene frequencies will likely drift in different directions in different populations, thereby allowing more genetic variation to be maintained in the entire set of populations than if individuals were maintained in a single colony. This strategy can only work, however, if the separate populations or colonies are large enough to avoid rapid inbreeding depression and if individuals are occasionally moved between colonies to counteract inbreeding. Moving animals among colonies or introducing new animals from confiscations can also lessen the effect of artificial selection (Ford 2002).

The decision to move animals among colonies or introduce new animals will have to take into consideration the breeding units as they were previously defined. Ideally the natural level and geographic scale of dispersal would first be determined among wild populations and then these values mimicked among captive colonies. Since this will rarely be possible, it may be best to maintain a network of colonies (assuming sufficient numbers of individuals are present) for each genetically defined breeding group and only move animals within their own breeding group network. Some breeding units may contain very few individuals though and so it may be necessary to interbreed individuals from different groups in order to counteract the effects of inbreeding depression. A few good examples of “genetic restoration” in the wild, including the Florida panther (Felis concolor coryi) and the adder (Vipera berus), have now been documented (reviewed in Hedrick and Kalinowski 2000; Tallmon et al. 2004). Intentional hybridization will have to be considered on a case by case basis with the following stipulations: 1) hybridization should be between the most closely related populations, and 2) the amount of hybridization should be just enough to counter the effects of inbreeding depression (Moritz 1999; Allendorf et al. 2001).

Population Management 2000 (PM2000; Lacy et al. 2000) is a free computer program that has been developed to assist managers in setting target population sizes to maintain demographic stability and genetic diversity and to explore managing a captive breeding program as a set of populations (Fig. 1, part II). For many species, data will be lacking for important life history data (e.g., generation length, population growth rates) and so it will be necessary to substitute values from other similar species.

Breeding Strategies to Decrease the Loss of Genetic Variation

Genetic and social mating patterns have been examined in surprisingly few species of wild freshwater turtles and tortoises (but see Osentoski 2001; Pearse and Avise 2001). Of the 6 species that have been examined, 5 had multiple sires for 40% or more of all clutches (the exception was the painted turtle, Chrysemys picta, with 4%–13%). Female sperm storage (up to 4 years) has been documented in turtles and is thought to be common (Pearse and Avise 2001). Currently there is little evidence to suggest that females are actively choosing among males, but social dominance among males has been shown to affect mating patterns (Gailbraith 1991; Kaufmann 1992). Reproduction that is strongly skewed toward a few males (e.g., strict dominance hierarchies) will reduce N_e substantially, as has been documented in the captive breeding program of the giant Galápagos tortoise (Geochelone nigra hoodensis) (Milinkovitch et al. 2004).

Complexities such as polyandry, sperm storage, and strict dominance hierarchies can make the production of accurate pedigrees difficult. Assurance colonies should therefore monitor genetic paternity closely, in order to detect skewed mating patterns and facilitate the production of pedigrees that can be used to predict losses of genetic variation and more effectively manage the population (see below) (Fig. 1, part II). High probabilities of excluding nonfathers can be obtained by employing even a few highly polymorphic microsatellite loci, especially when all candidate males can be sampled (Marshall et al. 1998).

Minimizing the loss of genetic variation during captive propagation can be achieved by steps to increase and stabilize N_e such as: 1) maximizing the number of breeders within assurance colonies and moving individuals among colonies; 2) increasing population size as rapidly as possible to the target size; 3) equalizing family sizes by ensuring the next generation is founded by equal numbers of offspring from each breeder; 4) equalizing the sex ratio of breeders; and 5) reducing fluctuations in population size (reviewed in Rodriguez-Clark 1999). Since reproductive output and the willingness to breed will vary among individuals, implementing these steps can be difficult. In this more realistic situation, one needs to know which individuals should be preferentially bred in order to maximize the remaining diversity of the founders.

Currently, the best method for maintaining genetic diversity within captive populations is by the mean kinship method (Ballou and Lacy 1995; Lacy 1995) (Fig. 1, part II). This method minimizes group coancestry (mean kinship) by using pedigree information to identify under-represented lineages and then favoring these individuals in breeding. It has proven to be more successful at slowing the loss of gene diversity than other comparable management methods (e.g., “maximum avoidance of inbreeding”) in both simulation and experimental work (reviewed in Rodriguez-Clark 1999; Fernandez and Caballero 2001). To use this method, an individual's “mean kinship” (how related it is on average to all other individuals in the population including itself) is first calculated from pedigree data. Individuals are ranked according to this measure and then males and females with the lowest rankings are preferentially paired. If however, the lowest ranking individuals are more related to each other than the population's average mean kinship, they are mated with the next ranked individual (or another until a suitable match is found), thereby avoiding potential inbreeding. Mean kinship scores are then updated continuously to account for births, deaths, and introductions of new breeding animals (Ballou and Lacy 1995). In the early stages of a captive breeding colony, pedigrees are usually unavailable and the relatedness of the founders to each other is unknown. Founders are then considered to be unrelated, an assumption which may or may not be valid, for management purposes. Microsatellite loci can be used to obtain estimates of pair-wise relatedness and improve the estimation of mean kinship scores (Blouin 2003; Russello and Amato 2004).

Eventually, the mean kinship method could also be used to identify “surplus” offspring that could be sold within the pet trade to help finance breeding colonies. For instance, offspring produced by breeders that have high mean kinship scores (representing an over-representation of certain lineages) could be sold, while offspring resulting from low score pairings would be retained in the colony. Similarly, the method could also be used to select a genetically diverse group of individuals for reintroduction. PM2000 (Lacy et al. 2000) can be used to calculate these parameters. Breeding and pedigree data used by PM2000 can be managed using programs such as Microsoft Excel or SPARKS (an electronic stud book; ISIS 1999).

Selection in Captivity

Selection in captivity occurs when there is a shift in the mean value of a trait (e.g., body size) away from what it was in the wild (Frankham et al. 1986; Arnold 1995; Lynch 1996). Maintaining individuals under conditions that do not mimic a species' natural state, such as abnormally high population densities, skewed sex ratios, or small enclosures can impose strong artificial selection or result in a relaxation of the selective pressures that individuals experienced in the wild (e.g., protection from predators, easy access to food) and thereby potentially leading to the loss of adaptations (Snyder et al. 1996; Reed and Bryant 2001). Numerous examples from a variety of taxa have shown that evolutionary change can occur quite rapidly (< 100 years) within captive populations provided selection is strong enough (reviewed in Ashley et al. 2003; Stockwell et al. 2003). In highly fecund organisms that have very large populations, such as rodents and fish, selection can cause significant changes within 1–2 generations (Allendorf 1993; Arnold 1995). Because many turtles and tortoises have low reproductive rates and long generation times, artificial selection will generally be weaker than it is for more fecund species. Nevertheless, some selection is probably unavoidable and steps should be taken to monitor and reduce its effect.

Equalizing family sizes will lessen the effect of selection in the captive environment by shifting selection to within families rather than between (Allendorf 1993), although in practice this can be a difficult task (Snyder et al. 1996). Structuring assurance colonies so that turtles and tortoises are maintained under conditions that mimic their natural habitat as closely as possible could minimize artificial selection. To counter the potential effect of selection, assurance colonies should be monitored to track changes in morphology and behavior over time (Fig. 1, part II). When identifiable, a good strategy may be to pick traits that are thought to be important in a species natural habitat, such as those involved in mating, feeding, and predator avoidance (Arnold 1995; Storfer 1996). The mean of these traits could then be compared between generations to determine if directional selection is occurring in the captive environment.