Intrinsic reproductive isolation between Trinidadian populations of the guppy, Poecilia reticulata

Abstract

Although Trinidadian populations of the guppy, Poecilia reticulata, show considerable adaptive genetic differentiation, they have been assumed to show little or no reproductive isolation. We tested this assumption by crossing Caroni (Tacarigua River) and Oropuche (Oropuche R.) drainage populations from Trinidad's Northern Range, and by examining multiple aspects of reproductive compatibility in the F₁, F₂ and BC₁ generations. In open-aquarium experiments, F₁ males performed fewer numbers of mating behaviours relative to parental population controls. This is the first documentation of hybrid behavioural sterility within a species, and it suggests that such sterility may feasibly be involved in causing speciation. The crosses also uncovered hybrid breakdown for embryo viability, brood size and sperm counts. In contrast, no reductions in female fertility were detected, indicating that guppies obey Haldane's rule for sterility. Intrinsic isolation currently presents a much stronger obstacle to gene flow than behavioural isolation, and our results indicate that Trinidadian populations constitute a useful model for investigating incipient speciation.

Introduction

A burgeoning literature has accumulated over the past approximately 80 years that is dedicated to the cataloguing and description of barriers causing reproductive isolation, and it has provided many important insights into the mechanics of speciation (see Coyne & Orr, 2004 for review). For instance, we now know that multiple barriers typically isolate any given species pair (e.g. Ramsey et al., 2003), and that certain patterns characterize the evolution of these barriers. Notable patterns include Haldane's rule (Laurie, 1997) and the more rapid appearance of sterility over inviability problems in hybrids (e.g. Price & Bouvier, 2002). However, many fundamental issues remain unresolved.

One concerns the relative importance of intrinsic and prezygotic barriers in causing speciation. This issue is difficult to address because of the typically long timescale of speciation (e.g. see Coyne & Orr, 1989, 1997), which often obliges researchers to study taxa that have already speciated. Unfortunately, the resulting inferences may poorly reflect what actually happens during speciation (Presgraves, 2003; Reed & Markow, 2004). A particular problem with inter-specific studies is that they may suffer from an ascertainment bias that favours the detection of prezygotic over intrinsic barriers. This ascertainment bias becomes increasingly severe with time following speciation, as prezygotic isolation effects a greater proportional barrier to gene flow (Coyne & Orr, 2004).

Uncertainty over the relative importance of different barriers has been compounded by shortcomings in the treatment of intrinsic isolation. Crossing experiments have generally concentrated on a few groups (e.g. Lepidoptera and Drosophila), even though the tempo and mode of speciation may differ between groups. For example, it is possible that the genetic architecture of intrinsic isolation varies in parallel with the phylogenetic distribution of life-history traits (Rieseberg et al., 2000) or the importance of genomic imprinting (Reik & Walter, 1998; Vrana et al., 2000), meaning that the evolution of intrinsic isolation could show lineage-specific properties. In addition, crossing experiments have frequently failed to assay hybrid generations other than the F₁ (Edmands, 2002; but see, e.g. Lu & Bernatchez, 1998), or components of intrinsic isolation other than physiological sterility and inviability. One component that is poorly understood is hybrid behavioural sterility, where hybrids fail to mate or reproduce because of a defect (neurological, pheromonal, etc.) that renders them incapable of effective courtship (Coyne & Orr, 2004). These problems exist despite the commonness of hybrid breakdown (Edmands, 2002) and an awareness that behavioural sterility can be important, at least between species. To our knowledge, behavioural sterility has not yet been assayed in intra-specific crosses.

Resolving such uncertainties is critical for a holistic understanding of speciation and requires further consolidation of the literature describing isolating traits. To this end, Trinidadian populations of Poecilia reticulata (the guppy; family Poeciliidae) may be of particular value as they show considerable adaptive genetic differentiation (Reznick et al., 1990), suggesting that they may serve as a model for the investigation of isolating barriers prior to speciation. Moreover, the large body of literature on Trinidadian guppies (reviewed in Houde, 1997 and Magurran, 2005) means that the evolution of reproductive barriers can be productively interpreted within specific ecological and genetic contexts.

Guppies are freshwater fish that are naturally distributed across the north-eastern corner of South America, including Trinidad and Tobago. They are notable for a number of reproductive traits, including internal fertilization, live-birth of young and the absence of any maternal provisioning of nutrients to eggs following fertilization (i.e. ovoviviparity) (Houde, 1997).

Phylogenetic studies using the mitochondrial control region have shown that populations from Trinidad's Oropuche Drainage form a monophyletic clade, which has diverged substantially from populations from Trinidad's Caroni Drainage (mean pairwise sequence divergence: 0.042) and from mainland South America (mean divergence: 0.034) (Alexander & Breden, 2004). So, although the Caroni and Oropuche drainages, located in Trindad's Northern Range (Fig. 1), are separated by as little as 70 m during the wet season (Magurran, 2005; p. 118), their populations are among the most genetically divergent currently known from throughout the guppy's natural distribution. This incongruity is difficult to explain because the region's geological history is highly complex (Donovan & Jackson, 1994). Carvalho et al. (1991; see also Fajen & Breden, 1992) have suggested that independent colonization of the Caroni and Oropuche drainages by different lineages from the Venezuelan mainland may be responsible. Unfortunately, no independent biogeographical or fossil evidence are available to aid calibration of the guppy molecular clock, and the divergence of the Caroni and Oropuche populations cannot be reliably dated. Complementing the phylogeographic data, allozyme studies have also found that, within Trinidad, most genetic diversity exists between the Caroni and Oropuche drainages (Carvalho et al., 1991).

Figure 1.  The Caroni and Oropuche Drainages in Trinidad's Northern Range.

Research on the Northern Range populations has dealt with the adaptive geographic differentiation seen in many morphological, behavioural and life-history traits. For example, male colour patterns (Endler, 1980), brood size (Reznick et al., 1990) and schooling propensity (Seghers, 1974) often show consistent and large differences between the upstream and downstream compartments of individual rivers, or between equivalent compartments of different rivers (e.g. Endler, 1980). Such differentiation often has a large hereditary basis (e.g. Reznick et al., 1990; Magurran & Seghers, 1994), and has been attributed, in part, to differences in predation regime and resource availability (reviewed in Endler, 1995; Arendt & Reznick, 2005; Magurran, 2005). However, the considerable phenotypic differences between populations are not paralleled by large changes in female mating preferences. Instead, behavioural isolation is uniformly slight and often statistically undetectable, even between populations from different drainages (Endler & Houde, 1995; Magurran et al., 1996). This discrepancy has led some authors to consider that ‘the divergence of guppies has not proceeded beyond a preliminary stage in speciation (Houde, 1997, p. 152)’, and to speculate whether certain factors may actually be impeding speciation (e.g. Magurran, 1998, 2001; see also Discussion).

Yet, in the absence of methodical studies of alternative isolating barriers, any conclusions about the nature of reproductive divergence in guppies are provisional. Little is known, for example, about intrinsic isolation. Several preliminary surveys have suggested that F₁ hybrids between Caroni and Oropuche drainage populations are viable, but these studies assayed only obvious, easily scored traits in the F₁ generation. These studies remain unpublished and are alluded to in Endler (1995), Endler & Houde (1995) and Houde (1997).

We crossed the Caroni and Oropuche drainage populations to systematically test for reductions in hybrid fertility and viability. We paid particular attention to behavioural sterility to establish whether it can evolve prior to speciation. We considered post-F₁ generations in case hybrid breakdown existed, and to enable preliminary biometrical analyses that might suggest the existence of large-X effects.

Methods

Stock maintenance and crossing design

We crossed stocks descended from wild populations in the Tacarigua and Oropuche Rivers, which occur in the Caroni and Oropuche drainages respectively (Fig. 1). Both populations occupied downstream high-predation-risk habitats, co-occurring with the pike cichlid Crenicichla alta, an important guppy predator. An advantage of using such populations in crosses is their high fecundity (c.f. Reznick et al., 1990). The laboratory stocks had intermittently received supplements of wild fish to prevent inbreeding; the last supplement occurred at least two generations (10 months) before the start of the experiment.

Stocks were fed ad libitum daily with commercially prepared flake food and Artemia brine shrimp. All tanks contained charcoal filters, gravel and Java moss (Vesicularia dubyana). Illumination was provided by 40 W overhead fluorescent strips set on a 12 : 12 h light : dark cycle, and temperature was maintained at 24–26 °C. Stocks were periodically culled to prevent aged individuals being used in the experiment. The fish were all killed using an overdose of benzocaine anaesthetic (ethyl p-amino benzoate), in accordance with United Kingdom Home Office regulations.

In each cross, single males were left with single virgin females in small 6 L maternity tanks for 7 days – the allocation of individuals to particular crosses and tanks was randomized. Males were then killed and females isolated until they gave birth. Because the birth of fry from single broods can be staggered over 24 h, and because females may cannibalize their young (S. T. Russell & A. E. Magurran pers. obs.), tanks were checked at least daily for the presence of new-born fry, which were then separated from their mothers. Twenty-four hours following the appearance of the first-born, females were also killed. Hence, no individual was used in more than one cross. Although females can store sperm for up to 6 months (Constanz, 1984), females that had failed to produce broods within 3 months were discarded from analysis. All parents were photographed with a digital camera for determination of size.

Broods were reared in 4 L, plastic and transparent bottles. A maximum of six offspring were reared per bottle to minimize competition effects on growth. Broods from different females were reared separately and, upon sexual maturation, the numbers and sex ratios of surviving offspring were noted. Adolescent males were identified by the presence of modifications to the anal fin rays, which eventually form the gonopodium (Clark & Aronson, 1951), and adolescent females were identified by the presence of abdominal spots (Houde, 1997). Progeny from different broods were then housed together in larger, 30 L stock tanks, with individuals being segregated according to sex and cross type.

In addition to the parental controls, eight hybrid lines were established: two reciprocal lines each for the F₁ and F₂ generations, and four different reciprocal backcrosses (Table 1). By convention, the initial sex reported in any cross is female (i.e. ♀ × ♂).

Table 1.   Experimental crosses and their reproductive success.

Cross			No. crosses attempted	No. successful crosses (% success)
T = Taracigua, O = Oropuche.
♀	♂	Abbreviation
T	T	P1	32	24 (75)
O	O	P2	26	21 (81)
T	O	F1	29	24 (83)
O	T	F1R	33	27 (82)
(O × T)	(O × T)	F2	24	20 (83)
(T × O)	(T × O)	F2R	25	20 (80)
(T × O)	T	BC1	27	21 (78)
(O × T)	T	BC2	25	18 (72)
T	(T × O)	BC3	30	22 (73)
T	(O × T)	BC4	26	19 (73)
Total			277	216 (78)

Male behavioural sterility

Male guppies perform two different mating behaviours on a facultative basis. ‘Sneaky mating’, which is prevalent in natural populations (Magurran et al., 1995), entails the attempted insertion of the male's intromittant organ, the gonopodium, into the female, regardless of whether the female is receptive. In contrast, ‘sigmoid’ displays are performed during courtship, and in these a male bends its body into a characteristic S shape (Baerends et al., 1955; Liley, 1966).

Male behavioural sterility was assessed using an open-aquarium design, which helps to replicate natural social environments and so allows behaviours relevant to wild populations to be observed (Houde, 1997). Three 60 L observation tanks were used. Observations occurred randomly with respect to the time of day and to the particular tank used. The behaviour of single focal males was observed in relation to five females from ‘mixed’ stock. Four ‘mixed’ stock males were also present to simulate a sexually competitive environment and a sex ratio of unity that typifies natural populations in Trinidad (Pettersson et al., 2004). The mixed stock individuals were largely advanced backcrosses derived from multiple parental populations, although pure parental individuals and F₁ hybrids may also have occasionally been present. Behavioural observations used the same common pool of mixed stock fish (approximately 75 individuals); the exact composition used in each trial was randomized. Mixed stock fish were allowed to acclimatize to the observation tanks overnight, before focal males were added. Focal males were given 1 h to acclimatize before observations began. To minimize asymmetries in male competitive success, males were of approximately the same size (standard length: mean ± SE = 22.42 ± 0.12 mm, n =40). Females were size-matched too (standard length: mean ± SE = 30.79 ± 0.28 mm, n = 30). All fish were fed to satiation before trials started.

Each focal male was observed for two 15 min periods, and the number of sneaky matings and sigmoid displays was noted, in addition to the length of time it spent following females. If focal males showed no mating behaviour, their trials were discarded from analysis. Data from individual males were averaged over the two trials. Behavioural sterility of females or of Bc and F2 hybrid males was not appraised because of time constraints.

Physiological sterility and inviability

Mature progeny were killed and photographed. Sperm counts and testes weights were used as measures of male fertility. To extract sperm, males were placed on a slide under a low-power dissection microscope. The gonopodium was swung forward and 20 μL of 100 mm NaCl was deposited onto it. Pressure was then applied to the side of the abdomen in a forward stroking motion to induce ejaculate release into the salt solution (Matthews et al., 1997). This process was repeated twice before the discharged sperm bundles were retrieved using a P100 Gilson pipette (Gilson, Middleton, WI, USA)and added to 80 μL of 100 mm NaCl. NaCl stimulates sperm bundle disassociation (Morisawa & Suzuki, 1980). Samples were left for 20 min to dissociate. Sperm counts were then performed under 400× using an ‘improved Neubauer chamber’ haemocytometer. Sperm counts were restricted to the parental, F₁ and F₂ lines.

In contrast, assays of testes weight used all available lines, but not those fish measured for sperm counts. Fish were dissected to remove the guts and testes, and testes and body carcasses were dried at 60 °C for 24 h.

The fecundity of virgin females was measured as the number and total weight of fully developed eggs available for immediate fertilization. Such eggs occupy stage 3 in Haynes's (1995) classification of embryonic development in poeciliids. They are characterized by being fully yolked and translucent golden-yellow in colour, and by having oil droplets evenly dispersed across their yolk surface (Haynes, 1995). These ova were dried at 60 °C for 24 h along with the body carcasses, minus guts.

To examine embryonic inviability, mothers that had produced broods were dissected for the presence of embryos. If present, embryos were recorded as inviable if they occupied a developmental stage in Haynes's (1995) classification that was under 10. This scheme allowed inviable embryos to be clearly distinguished from viable, fully mature embryos that had not been born and which constitute stage 11. This protocol was adopted because fertilization in guppies occurs over an interval of about three days, or one developmental stage (S. T. Russell pers. obs.), causing slight developmental asynchronies within broods (Haynes, 1995). Besides differential survival, an alternative explanation for the presence of embryos at different stages is superfetatation, where females simultaneously carry different broods. But this explanation is likely to be untenable for guppies, which are considered to be non-superfetatating (Houde, 1997).

Statistical analyses

Planned comparisons are a common biometrical approach wherein crosses that differ with regard to a single factor are compared to determine the influence of that factor (de Belle & Sokolowski, 1987; Huttunen & Aspi, 2003). The influence of non-autosomal factors on behavioural sterility was assessed using planned comparisons (Mann–Whitney U-tests) of the two reciprocal F₁ lines. Analyses of mating behaviours across all lines used median tests.

Standard body length differed between lines (e.g. for mothers: anova: F_9,214 = 7.38, P < 0.01); therefore, ancovas were used to test for differences in fertility. ancovas used log-transformed data (Tomkins & Simmons, 2002), and the covariate was standard length measured from digital photographs using ImageJ v.1.30 (Rasband, 2003). ancovas were also used on dry body weight (including gonads) to test for differences in body condition.

Because of occasional difficulties in distinguishing early stage nonpigmented embryos from surrounding ovarian connective tissue, records of embryonic inviability may be biased towards later developmental stages. This possibility, together with the fact that embryo readsorption almost certainly occurs in guppies, meant that data were not analysed in relation to brood sizes as embryo mortality rates. Instead, embryo inviability was analysed as count data using the Kruskal–Wallis H-test following √(y + 0.5) transformation. A total of 10 000 Monte Carlo simulations and a confidence level of 99% were used to generate the probability statistic. Because of limited sample sizes, reciprocal lines in this analysis were pooled to form just three hybrid classes (F₁, F₂ and BC₁) to increase the statistical power to detect hybrid defects.

Juvenile mortality data were analysed as count data using the program betabino (Jiggins et al., 2001). The statistical approach implemented in this program is advantageous whenever biological factors underlie differences in survival between replicate broods, a situation that may be likely (Reed & Markow, 2004). Two models fitted by the program were tested: (1) the lines have the same mean but different variances, and (2) the lines have different means and variances. These particular models were used because variances were heterogenous. The log-likelihood values for these models were compared using a likelihood ratio test, and the resultant statistic compared with the chi-squared distribution to test for differences in mean values between lines, with degrees of freedom equal to the number of lines minus one.

Finally, differences in sex ratios were tested for. Sex ratios per brood were calculated as: number of females/total number of surviving adult progeny. Data were arcsine-square root-transformed, and differences between lines were tested with a general linear model, where initial brood size was the weighted least squares (WLS) weight.

Results

Significant differences existed between lines in the number of displays ( = 21.132, P < 0.001), the number of thrusts ( = 15.441, P < 0.005) and the length of time that a male had spent following females ( = 13.291, P < 0.005) (Fig. 2). In each case, this was due to hybrid dysfunction (Fig. 2). Planned comparisons between the reciprocal F₁s were not significant (Mann–Whitney U-tests: for sigmoids: Z = 0.261, P ≫ 0.05; for thrusts: Z = 0.410, P ≫ 0.05; for time: Z = 1.161, P ≫ 0.05).

Figure 2.  Male mating behaviours in the parental and reciprocal F₁ lines. O = Oropuche, T = Tacarigua. Shared numbers indicate significant differences (P < 0.05) following Games–Howell post hoc multiple comparisons that were sequential Bonferroni-corrected (Rice, 1989). Sample sizes (n) for each line are: T = 28, O = 29, T × O = 31, O × T = 27.

Female mortality contributed to roughly one half of the instances of cross failure (data not shown). Approximately 70–80% of attempted crosses yielded progeny, and there were no marked differences in cross success between lines (Table 1). However, brood sizes differed significantly between lines (ancova: F_9,214 = 6.76, P < 0.001) (Fig. 3). Post-hoc multiple comparisons showed that these differences were restricted to comparisons between parental or F₁ and post-F₁ lines. For example, the Tacarigua control (P₁) differed significantly from F₂, BC₂ and BC₄ (P < 0.05). Similarly, F₁R differed from all post-F₁ hybrid lines apart from F₂R (P < 0.05). Planned comparisons were also conducted between each of the (T × O) backcrosses into Tacarigua (BC₁ and BC3) and the Tacarigua control (P₁), but each comparison was significant (P < 0.05), meaning that any maternal effects on brood size could not be detected.

Figure 3.  Brood sizes for all lines. Untransformed data only. See Table 1 for identities of crosses and sample sizes.

In relation to virgin female fecundity, ancovas detected no significant differences between lines in the weight or number of mature eggs (weight: F_8,215 = 0.704, P ≫ 0.05; number: F_8,259 = 0.831, P ≫ 0.05) (Table 2). Similarly, male fertility did not differ between groups when testes weight was assayed (ancova: F_8,242 = 0.58, P > 0.05) (Table 2). To analyse the sperm count data, reciprocal crosses were pooled within hybrid lines to increase replicate sizes. An ancova on the data generated a significant statistic (F_3,96 = 3.59, P < 0.05), and Bonferroni post hoc multiple comparisons showed that the F₂ line had significantly reduced sperm counts relative to both parental lines and to the F₁ population (P < 0.05) (Fig. 4).

Table 2.   Descriptive statistics for male and female fertility.

Line (♀ × ♂)	Female fecundity				Male fertility
	No. mature eggs		Dry weight of eggs (mg)		Testes weight (mg)
	Mean (n)	SE	Mean (n)	SE	Mean (n)	SE
T = Tacarigua, O = Oropuche. Data for the L × (O × T) line (i.e. BC4) are not reported as its broods were culled following birth.
T	3.20 (20)	0.38	5.16 (16)	0.65	0.59 (31)	0.04
O	5.74 (34)	0.52	6.92 (29)	1.26	0.58 (43)	0.02
T × O	4.77 (34)	0.55	5.29 (32)	0.64	0.55 (24)	0.02
O × T	4.39 (23)	0.71	4.88 (22)	0.59	0.61 (23)	0.04
(O × T) × (O × T)	4.91 (46)	0.43	6.47 (35)	1.04	0.53 (28)	0.03
(T × O) × (T × O)	4.96 (24)	0.66	4.72 (20)	1.30	0.61 (30)	0.02
(T × O) × T	4.59 (27)	0.52	6.80 (20)	1.22	0.55 (21)	0.03
(O × T) × T	5.14 (29)	0.57	5.39 (19)	0.68	0.48 (19)	0.04
T × (T × O)	4.8 (24)	0.58	5.79 (24)	1.26	0.58 (25)	0.03

Figure 4.  Sperm counts for pooled parental and hybrid lines. T = Tacarigua, O = Oropuche. Sample sizes (n) for each class are: T = 36, O = 21, F₁ = 23, F₂ = 18.

Juvenile survival did not differ between lines, as judged by the likelihood ratio test implemented in betabino (Jiggins et al., 2001) (L_9,197 = 11.929, P ≫ 0.05) (Table 3). But embryo viability did show significant differences (Kruskal-Wallis H test:  = 30.10, P < 0.001) (Table 3). Games–Howell post hoc tests showed that the pooled backcross class differed from each of the P₁, P₂, and pooled F₁ and F₂ classes (P < 0.01). Maternal effects on embryo viability were again tested for using planned comparisons that involved both the (T × O) and (O × T) backcrosses into Tacarigua – all contrasts were nonsignificant (P > 0.05), making any such effects undetectable.

Table 3.   Percentiles for juvenile survival and embryonic inviability.

Line (♀ × ♂)	Juvenile survival (proportion of offspring from a brood surviving to maturity)			Embryo inviability (no. dead embryos per mother)
	n	Median	Interquartile range	n	Median	75th + 90th percentiles
Untransformed data only. Reciprocal hybrid lines have been pooled for embryo inviability.
T	24	1.00	0.88–1.00	20	0	0 + 1.0
O	21	1.00	0.91–1.00	20	0	0 + 1.0
T × O	24	1.00	0.93–1.00	40	0	0 + 1.8
O × T	27	0.96	0.84–1.00
(O × T) × (O × T)	20	1.00	0.85–1.00	40	0	0 + 0
(T × O) × (T × O)	20	1.00	1.00–1.00
(T × O) × T	21	0.86	0.65–1.00	121	0	3 + 7.9
(O × T) × T	18	0.94	0.65–1.00
T × (T × O)	22	1.00	0.75–1.00

Finally, neither adult brood sex ratios [generalized linear models (GLM) with brood size as WLS weight: F_8,195 = 1.69, P > 0.05] nor male or female body condition (males: ancova: F_8,242 = 0.74, P ≫ 0.05; females: ancova: F_8,215 = 1.39, P > 0.05) differed between lines.

Discussion

Most crossing studies have only considered physiological sterility and inviability in the F₁ generation (Edmands, 2002). Hence, they may have overlooked other aspects of intrinsic isolation important to speciation. This possibility is amply demonstrated by the present study, which has documented sterility and inviability, but only amongst post-F₁ generations, in addition to behavioural sterility in the F₁s. A diverse range of barriers must be assayed across multiple hybrid classes to allow reliable insights to be drawn into speciation.

Male behavioural sterility

The partial behavioural sterility shown by hybrid males is considerable and is the largest component of intrinsic isolation detected. Although it is well established that such sterility can occur in hybrids between well-established species, our study reveals that behavioural sterility can also arise at the intra-specific level.

Behavioural sterility could result from a number of factors. One is the pleiotropic effect of divergence in other traits. For instance, divergence in physiological processes such as muscle performance could have impaired general vigour, leading to a reduction in mating propensity. This hypothesis has been invoked for Anartia amathea × A. fatima butterfly hybrid females (Davies et al., 1997) and, to a lesser extent, for Drosophila pseudoobscura × D. persimilis hybrid males (Noor, 1997), which both show general lethargy and behavioural sterility. Alternatively, behavioural sterility could reflect specific disruption to the genetic processes underlying courtship and mating, which is feasible as selection can rapidly drive allopatric divergence in sexual traits (Questiau, 1999). The possible influence of general lethargy on the behaviour of hybrid guppies could be examined by measuring physiological performance under stress.

Demonstrating whether female guppy hybrids show comparable levels of dysfunction would be interesting because Noor (1997; see also Davies et al., 1997) has suggested that behavioural sterility is another trait that, like physiological sterility and viability, follows Haldane's rule. Haldane's rule states that ‘when, in the offspring of two different animal races, one sex is absent, rare, or sterile, that sex is the heterozygous [heterogametic] sex (Haldane, 1922).’ Its near universality across different taxa (Laurie, 1997) implies that there are fundamental genetic processes underpinning the evolution of intrinsic isolation (Coyne & Orr, 1998, 2004). Although general observations of the behaviour of hybrid females did not reveal any striking dysfunction (S. T. Russell & A. E. Magurran pers. obs.), more detailed characterisation of female behaviour is required.

Only two studies have examined the genetic architecture of behavioural dysfunction. Coyne et al. (2002) showed that dysfunction in Drosophila yakuba × D. santomea hybrid males might result from interactions involving the santomea X and the yakuba genome. Moreover, Noor (1997) and Noor et al. (2001) have shown that dysfunction in D. pseudoobscura × D. persimilus hybrid males is due to an interaction involving the persimilus X and the pseudoobscura Y. In contrast, the absence of significant differences between the reciprocal F₁ guppy lines indicates that sex chromosomes do not contribute disproportionately to their behavioural dysfunction (de Belle & Sokolowski, 1987). This finding is provisional and stronger support for it would come from genetic studies, but it is consistent with the much smaller degree of sex chromosome differentiation in guppies, relative to Drosophila. Although as much as the proximal half of the guppy Y chromosome is homologous with the X (Traut & Winking, 2001), potentially allowing appreciable recombination between the two (Traut & Winking, 2001; Lindholm & Breden, 2002), Drosophila species have highly degenerate Y chromosomes. An alternative explanation for our finding is that the larger number of autosomes in guppies (44), vs. Drosophila (2), may have diminished the proportional impact of X-linked incompatibilities. But this seems unlikely because large Z-effects have been found in Lepidoptera (e.g. Naisbit et al., 2002), in which the Z chromosome constitutes, on average, only 3% of the genome (Prowell, 1998). Moreover, X chromosome effects are also responsible for hybrid dysgenesis in several rodent genera (Zechner et al., 2004).

Physiological viability and sterility

Hybrid breakdown occurred in brood size (F₂s and BC₁s), sperm counts (F₂s) and embryonic inviability (BC₁s). The 1 : 1 sex ratios observed among adult hybrid progeny indicate that embryonic inviability does not obey Haldane's rule. But guppies do obey Haldane's rule for sterility, since only males showed any reduction in hybrid fertility. It is unlikely that this observation stems from methodological artefacts. One reason is that the metrics used to assay female fertility have previously been shown to be quite powerful (e.g. Reznick & Endler, 1982). Another is that, because of difficulties in catching all the sperm bundles released from a stripped male, which sometimes adhered to its body, the experimental error for measuring sperm counts probably exceeded that for measuring female fecundity. If anything, therefore, the sex bias in infertility may have been underestimated.

Hybrid breakdown is thought to result from the recessivity of the alleles underlying hybrid unfitness (reviewed in Edmands, 2002). Our finding that the backcross, but not the F₂, class showed reductions in embryo viability is surprising, and is inconsistent with a purely additive mode of gene action. The genetic architecture of embryo inviability in guppies offers interesting material for future research.

Relative importance of different reproductive barriers

Female and male guppies only occasionally show significant, though small, sexual preferences for individuals from the same population (Endler & Houde, 1995; Magurran & Ramnarine, 2004). For example, differential female choice is not discernible between Tacarigua and Oropuche guppies (Magurran et al., 1996). In contrast, intrinsic isolation between these same populations is substantial. Although it cannot be assumed that this pattern will hold throughout the remaining speciation process, we have at least established that intrinsic isolation could plausibly contribute to the speciation of guppy lineages. Making comparable inferences from inter-specific studies is often more difficult since the relative importance of behavioural barriers to gene flow increases with time following speciation (Coyne & Orr, 2004). Moreover, the future importance of intrinsic isolation is actually suggested by several considerations. One is that complex intrinsic incompatibilities are more likely to persist following population admixture than other barriers such as behavioural isolation, which are based on phenotypic differences (Rieseberg et al., 2003). This point is important because population admixture must occur frequently during the early phases of population divergence (Coyne & Orr, 2004). A subsidiary consideration is that those factors postulated to retard the evolution of behavioural isolation (see below) are not expected to weaken over time (Magurran, 2001).

Our results strongly contrast with the common notion that behavioural isolation evolves more rapidly than other barriers among allopatric populations subject to strong sexual selection (e.g. Mendelson, 2003). In fact, there is little theoretical or empirical justification for this opinion (Turelli et al., 2001; Coyne & Orr, 2004), especially given that sexual selection is expected to drive the accumulation of genetic factors inducing sterility and gametic isolation (Turelli et al., 2001). Nonetheless, there may be reasons, other than stochasticity, that explain the general lack of behavioural isolation among Trinidadian guppy populations. One possibility is that sneaky mating somehow constrains the evolution of female mating preferences (Magurran, 1998, 2001). Another is that populations from equivalent (upstream or downstream) compartments in different rivers often experience comparable environments, which may then select for similar choice criteria (Magurran, 2001). Both processes are thought to constrain the evolution of behavioural isolation between allopatric populations. Currently, it is impossible to distinguish between these processes, although it may be possible to test the potential importance of sneaky mating by examining post-copulatory selection processes – this study is currently underway.

In summary, the guppy has commonly been assumed to show little or no reproductive isolation. But, we have discovered that strong intrinsic barriers separate Trinidadian populations. Despite the strength of reproductive isolation, Caroni and Oropuche populations are clearly members of the same species, as the reproductive barriers between them are unlikely to prevent substantial introgression across most off their genomes following secondary contact. Moreover, there are no obvious phenotypic differences between the two groups that might warrant their classification as distinct species. Rather, Caroni and Oropuche populations are intra-specific, and at an advanced phase along the road to total reproductive isolation. They thus constitute an ideal model for exploring incipient speciation. This potential has most clearly been demonstrated by the discovery that behavioural sterility can evolve within species, and that it may help to cause speciation.

Acknowledgments

A. Ludlow, I. Maynard, A. Ojanguren and R. Wilson assisted with guppy husbandry. I. Matthews provided advice on methodology. J. Tomkins made essential laboratory equipment available. The research was funded by the UK's Natural Environment Research Council.