Multiple sources of reproductive isolation in a bimodal butterfly hybrid zone


Astrid G. Muñoz, Instituto de Genética, Universidad de los Andes, Carrera 1 No 18A – 70, P.O.Box 4976. Bogotá D.C., Colombia.
Tel.: +57 1339 4949 ext. 2750; fax: +57 1332 4069; e-mail:


An important evolutionary question concerns whether one or many barriers are involved in the early stages of speciation. We examine pre- and post-zygotic reproductive barriers between two species of butterflies (Heliconius erato chestertonii and H. e. venus) separated by a bimodal hybrid zone in the Cauca Valley, Colombia. We show that there is both strong pre- and post-mating reproductive isolation, together leading to a 98% reduction in gene flow between the species. Pre-mating isolation plays a primary role, contributing strongly to this isolation (87%), similar to previous examples in Heliconius. Post-mating isolation was also strong, with absence of Haldane’s rule, but an asymmetric reduction in fertility (< 11%) in inter-specific crosses depending on maternal genotype. In summary, this is one of the first examples of post-zygotic reproductive isolation playing a significant role in early stages of parapatric speciation in Heliconius and demonstrates the importance of multiple barriers to gene flow in the speciation process.


The order in which different forms of reproductive isolation evolve in the early stages of speciation can be difficult to determine, as multiple mechanisms often act simultaneously (Coyne & Orr, 2004; Doebeli, 2005; Gray, 2005; Ortiz-Barrientos & Noor, 2005). Hybrid zones offer an opportunity to study the evolution of reproductive isolation in different stages of speciation (Veen et al., 2001; Bridle & Butlin, 2002; Widmer et al., 2009). Studies in these zones have revealed that factors such as linkage of divergent traits and multiple pleiotropic effects of loci controlling species differences can facilitate speciation (Qvarnström & Bailey, 2009). Thus, comparative analysis of hybrid zones in different stages of divergence might help to reconstruct the sequence of events leading to speciation (Jiggins & Mallet, 2000; Cianchi et al., 2003; Putnam et al., 2007; Carling & Brumfield, 2008; Kawakami et al., 2009; Wiley et al., 2009).

In particular, bimodal hybrid zones are characterized by a high number of parental genotypes and few intermediate hybrids and thus represent a stage of divergence close to complete speciation (Jiggins & Mallet, 2000; Redenbach & Taylor, 2003; Peterson et al., 2005; Arntzen et al., 2009). For example, the bimodal hybrid zone between two incipient species of butterflies, Heliconius himera and H. erato in southern Ecuador, is maintained by strong prezygotic isolation because of assortative mating without intrinsic incompatibilities (Jiggins et al., 1996; McMillan et al., 1997). In addition, frequency-dependent predation against hybrids must also contribute to reproductive isolation (Jiggins et al., 1996). The lack of hybrid incompatibility between Heliconius himera and H. erato led to the suggestion that post-mating isolation is less important in the early stages of ecological speciation. Nonetheless, this may not be the case in other bimodal hybrid zones, or in other butterfly species (Presgraves, 2002; Arntzen et al., 2009). Here, we study another bimodal hybrid zone involving H. erato, offering the opportunity to investigate the relative importance of post-mating isolation in this group.

We have recently described the bimodal hybrid zone between H. erato chestertonii and H. erato venus, which meet and form wing pattern hybrids at a frequency of approximately 25% in the Calima river valley, Cauca Valley, Colombia (Arias et al., 2008). This is higher than the 10% hybrids found in the H. erato × H. himera hybrid zone, suggesting that H. e. chestertonii is at an earlier stage of speciation compared to H. himera. Microsatellite markers showed that all chestertonii × venus wing pattern hybrids are either backcrosses or indistinguishable from pure types, such that F1 genotypes are extremely rare. The strong deviation from Hardy–Weinberg frequencies between these parapatric species in this zone must therefore be mediated by strong pre- and post-mating reproductive isolation. As in H. himera, the divergence between H. e. chestertonii and H. e. venus is also associated with ecology. The former is found in higher altitudes, dryer habitat and mimics H. cydno rather than H. melpomene (Arias et al., 2008). Furthermore, the narrow hybrid zone (≈ 4 Km) is located on an ecotone from dry to wet forest, consistent with the idea that adaptation to different habitats has contributed to speciation, although there is no evidence for ecological segregation within the narrow zone (Arias et al., 2008). Both species collect pollen from Psiguria sp. and Gurania sp. to obtain amino acids that are used in egg production. They lay eggs on Passiflora suberosa and P. rubra host plants (Muñoz, Pers. Obs). Also these species belong to the Heliconius pupal mating clade, in which males detect females at this immature stage and guard them prior to emergence. Here, we quantify the relative contribution of reproductive barriers in the H. e. chestertonii and H. e. venus hybrid zone and contrast these with H. himera/H. erato.

Materials and methods

Specimen sampling and experimental populations

Samples of H. e. venus and H. e. chestertonii were collected in Cauca Valley and Risaralda (Colombia) in the following localities: Ladrilleros (03°57′08′′N, 77°22′03′′W), Montañitas (3°41′03′′N, 76°31′33′′W), Calima river valley − Km 15 (3°53′60′′N, 76°37′57′′W) and Marsella (4°55′50′′N, 75°44′43′′W). Stocks of these populations were established in outdoor insectaries of 2 × 3 × 2 m3 in la Vega (Cundinamarca) with access to enough pollen sources (Lantana and Psiguria) and artificial nectar (10% sugar solution). Plants were provided for oviposition, mainly Passiflora rubra and P. suberosa. Individuals of each species and their hybrids were reared as previously described (McMillan et al., 1999).

Analysis of pre-mating isolation

Two designs of mating experiments were performed to investigate the strength of prezygotic reproductive isolation in the Cauca valley hybrid zone. First, the probability of males and females to mate inter-specifically was determined in experiments without choice, recreating a natural situation in which males encounter females singly (at adult or pupal stage). Second, we carried out choice experiments to test the importance of wing pattern as a mating cue, using wing colour pattern models.

No-choice mating experiments

A single virgin female was exposed to a group of 6–10 mature males of the same species (H. e. chestertonii or H. e. venus) in an insectary for a period of 1 day. Males were at least 5 days old before each experiment to ensure reproductive maturity. H. erato shows pupal mating in the wild; males patrol the forest searching for female pupae, so they can mate with uneclosed or freshly emerged females (Gilbert, 1976; Sourakov, 2008). We tested both adult and pupal females in our trials (Deinert et al., 1994; McMillan et al., 1997). In these experiments, we used intra-specific, inter-specific and F1 individuals. The success or failure of mating was recorded throughout by direct observation of coupling and/or the presence of a spermatophore in the female abdomen and post-mating female odour (Gilbert, 1976). Mating pairs were removed and not re-used. The remaining males were re-used for the subsequent experiment, with replacement of the mated male (Number of males in Table 1). Likelihood was used to calculate the probability of mating (Pixj) for each cross between female type i and male type j, using the function: m loge Pixj + n loge (1 − Pixj), where m and n are the number of experiments with success or failure to mate, respectively. Subsequently, we evaluated the minimal number of mating categories (parameters) that could explain our data using likelihood. This process involves the comparison of models with different numbers of parameters, grouping mate categories when their probabilities were statistically similar. For example, a model with two parameters (× b and c × d) represented two mating categories: intra-specific and inter-specific including four types of crosses (for more details see Appendix S1). We compared hierarchical models of mating with different number of parameters using a likelihood ratio test (McMillan et al., 1997).

Table 1.   Summary of no-choice experiments with adult females and pupae.
Cross typeVirgin adult femaleFemale pupa
SuccessfulFailureTotalMean of males in each experimentSuccessfulFailureTotalMean of males in each experiment
Intra-specific or Pure8723110101701710
Inter-specific (♀H. e. chestertonii × ♂H. e. venus)21214101001010
Inter-specific (♀H. e. venus × ♂H. e. chestertonii)0661040410
Backcrosses (♀ F1 × ♂ Pure)225271060610
Backcrosses (♀ Pure × ♂ F1)381173037
Total  174   40 

Colour pattern models

The role of colour pattern in maintaining reproductive isolation between H. e. chestertonii and H. e. venus was tested in experiments carried out with wing colour pattern models. Briefly, twenty-eight males of H. e. chestertonii and nineteen males of H. e. venus were isolated after emergence, during a period no < 5 days to ensure sexual maturation. These males were tested individually against colour pattern models made with dissected female wings. Previous experiments have demonstrated the importance of cuticular hydrocarbons in H. erato courtship thus, for comparison, colour pattern wing models were used both without any treatment or after being washed with 96% hexane for 12 h to remove pheromones (Estrada & Jiggins, 2008). Two models were presented simultaneously, one being a ‘control’ with the same pattern as the males being tested and the other being an ‘experimental’ colour pattern. Both species were tested against the other species wing pattern and against the F1 hybrid phenotype. The models were hung from nylon lines in the centre of a spherical area of 60 cm as described previously (Estrada & Jiggins, 2008; Melo et al., 2009). Males observed entering the sphere were scored as an ‘approach’ whereas sustained fluttering towards the model was recorded as ‘courtship’. Male behaviour was recorded in 5 min trials, and males were not reused between trials (Estrada & Jiggins, 2008; Melo et al., 2009).

We evaluated the male response to the models using a general lineal model with binomial error distribution, or quasibinomial in the case of overdispersion. We conducted this analysis with R statistical software version 2.9.0 ( For these estimations, we set the male phenotype as the response variable (H. e. chestertonii or H. e. venus) and the model type in each experiment (control or experimental) as the explanatory variable. We evaluated approaches and courtships for each treatment (hexane and without hexane) independently, and we constructed a total of eight different input files for the analysis (data for each input are in Appendix S2).

Additionally, we estimated the probabilities Qixj that males of type j approached or courted models of type i relative to that of their own type j, using likelihood. Thus, for H. e. chestertonii males (EC) with EC versus H. e. venus (EV) models, the actual probabilities are QA EC × EV/(QA EC × EV + 1) that males approach EV and 1/(QA EC × EV + 1) that they approach EC. The loge likelihood for this experiment is therefore [XA EC × EV loge{QA EC × EV/(QA EC × EV + 1)}+XA EC × EC loge{1/(QA EC × EV + 1)}], where XA EC × EV is the number of EC males approaching EV, and XA EC × EC is the number approaching to EC. The probability for courtships was obtained with the same model. The summed loge likelihood was maximized over all experiments by varying the Qi × j parameters. Maximization was carried out with the Solver algorithm implemented in Excel. Support limits were obtained by determining the parameter values that decrease the maximum likelihood by two units, equivalent to a 95% confidence interval. Hypothesis testing was carried out using likelihood ratio tests (LRT), where the significance was established using the χ2 distribution (Melo et al., 2009).

Analysis of post-mating isolation

We took advantage of the successful matings obtained in the no-choice experiments to estimate hybrid sterility and inviability. After mating, each female was separated into an individual insectary with enough food resources and larval host plants. Eggs were collected once a week and after eclosion, larvae were reared individually in small plastic pots. New shoots of P. rubra and P. suberosa in each pot were replaced daily to ensure larval survival. Both intra-specific and inter-specific crosses were reared in the same environmental conditions. The number of eggs laid, their hatching success and the number of individuals successfully reared to adults was recorded each week. The data were analysed with the likelihood approximation implemented in BETABINO 1.1 program ( and explained by Jiggins et al. (2001a). Briefly, a beta-binomial distribution is used with count data (number of eggs) to obtain the maximum probability of observing an event (success of hatching) through different classes (type of cross). This likelihood function takes into account the variability within replicates of the same type and between different types of crosses. The program calculates the maximum log-likelihood under four models considering variation in the hatch rate among classes. As earlier, likelihood ratio tests were used to differentiate among alternative models with dissimilar number of parameters (Jiggins et al., 2001b; Naisbit et al., 2002; Salazar et al., 2005).

Larval survival was quantified as the proportion of first instar larvae reaching the adult stage. To test for variation in this proportion among intra- and inter-specific crosses, we performed a nonparametric analysis of variance (Kruskal–Wallis) in spss 17.0.0. software (SPSS, 2008). We tested for deviation from a 1 : 1 sex ratio using goodness-of-fit G tests in spss 17.0.0. software (SPSS, 2008).

Contributions to reproductive isolation

To determine the relative contribution of barriers involved in reproductive isolation, we estimated the absolute contribution of the reproductive barriers analysed earlier for the H. e. chestertonii and H. e. venus hybrid zone. First, individual values were estimated as 1 − (heterospecific frequency/homospecific frequency) for both pre- and post-mating barriers. Thus, the individual contribution of premating isolation barriers was estimated as: sexual isolation (No-choice mating experiments) = 1 − (heterospecific mating frequency/homospecific mating frequency). A proportion of this total sexual isolation was because of colour pattern, where preference (colour pattern models/with hexane) = 1 − (frequency of approaches to heterospecific model/frequency of approaches to homospecific model), with the remainder of the sexual isolation assumed to be because of pheromonal communication. In the case of post-mating isolation barriers we have: hybrid viability = 1 − (hybrid mean hatch rate/parental mean hatch rate). Although larval survival and sex ratio could be components of viability, no significant differences were found between crosses (> 0.05, Table 2) so they were not included in the analysis. Mimicry selection against hybrid phenotypes, although not measured directly, was also included in this analysis as a post-mating barrier. The absolute value of each barrier was then calculated as a multiplicative function of the individual components of reproductive isolation of previous barriers during the life history (Ramsey et al., 2003).

Table 2.   Mean hatch rates, larval survival and sex ratio in control, F1, and backcross broods.
Cross genotypeHatch rateLarval survivalMean (SE)Sex ratio (female/total)Mean (SE)
FemaleMaleNo. experimentsNo. eggsMean (SE)Variance (SE)
  1. Hatch rate column correspond to values derived from the beta-binomial model fitted over all broods (< 0.001). C are Heliconius erato chestertonii, V are Heliconius erato venus, both genotypes are named as Pure. Cross types are given with the female genotype first. The standard error (SE) for each parameter is showed in brackets.

CC4915010.30 (0.04)0.0001 (0.01)0.83 (0.8)0.55 (0.06)
VV145820.24 (0.06)0.0003 (0.02)0.82 (0.06)0.57 (0.05)
CV84920.14 (0.04)0.0002 (0.01)0.77 (0.09)0.53 (0.07)
VC2910.10 (0.04)0.001 (0.002)0.79 (0.14)0.55 (0.05)
C × VPure (C or V)1612940.12 (0.02)0.002 (0.002)0.42 (0.08)0.56 (0.06)
V × CPure (C or V)3980.35 (0.06)0.002 (0.007)0.82 (0.18)0.52 (0.06)
Pure (C or V)C × V51060.22 (0.07)0.002 (0.01)0.81 (0.15)0.47 (0.06)

These parameters can have values between −∞ and 1, where one means total isolation, 0 implies no contribution to isolation. Negative values would represent gene flow that is exceeding random expectation (Ramsey et al., 2003). The total reproductive isolation was estimated as a sum of the absolute contribution of all barriers (Ramsey et al., 2003; Nosil et al., 2005).


Analysis of pre-mating isolation

No-choice mating experiments

In total, 214 no-choice experiments were used to evaluate differences in the probability of mating for intra- and inter-specific crosses between H. e. venus, H. e. chestertonii and their hybrids. All the 40 pupae virgin females mated successfully irrespective of the type of males used (Table 1). In contrast, among the 174 adult virgin females, mating success was considerably lower, and there was heterogeneity in the mating probabilities with respect to the genotypes of the males and females tested. The initial twelve parameter model applied over the adult female experiments (encompassing all the types of crosses, Appendix S1) was reduced to a best fit model with five parameters: intra-specific (control), reciprocal inter-specific crosses and two types of backcrosses (L = −82.9070, Fig. 1, Table 1, Appendix S1). This model indicated a significant reduction in the mating probability in the inter-specific trials when compared to the 79% mating rate among controls. When the female was H. e. chestertonii, a mating success of 10% was observed. In the reverse direction, only 6 of 6 trials were completed, with no successful mating. Backcrosses to the parental species involving F1 females showed a similar frequency of successful mating (81%) compared to the controls (79%). However, there was a significant reduction in the mating probability when the F1 males were backcrossed, implying some breakdown of male hybrid mating (27%). Overall, the degree of isolation because of assortative mating is therefore likely to depend on the frequency of pupal mating in the wild. We observed no assortative mating in pupal mating trials, albeit with a small sample size, but strong assortment when mating was between adults.

Figure 1.

 Relative probabilities in mate no-choice experiments with virgin adult females. C is H. e. chestertonii, V is H. e. venus. Cross types are given with the female genotype first. The number on each barr represents the total of experiments in the respective cross type.

Colour pattern models

Likelihood estimates of approach and courtship probabilities were then estimated showing that in all cases males showed stronger preferences in courtships when compared to approaches (Fig. 2). Approaches and courtships of all males to experimental models were significantly different to the controls model with the same colour pattern as the male (< 0.05, Appendix S2). In general, there was no significant effect of hexane washing of the wings, indicating that preferences are likely to be because of colour pattern rather than pheromonal cues. The only case in which we observed that treatment affected male behaviour was when F1 models were exposed to male H. e. chestertonii (Fig. 2, Appendix S2). The F1 models without hexane treatment were more attractive to H. e. chestertonii males. In the assays with H. e. chestertonii males, the probability of approaching the H. e. venus pattern was around one-third of that to the control, and courtship probabilities were less than 10% (> 0.05). When the same H. e. chestertonii males were tested with the F1 female models a striking pattern appeared. They approached and courted models without the hexane treatment almost three times more than those washed in hexane (Gapproaches, 27 = 158.41; Gcourtships, 27 = 205.97; < 0.05). H. e. venus males approached H. e. chestertonii as well as F1 models less than half as often as to their own colour pattern. Courtships to these models were less than 20% in comparison with the controls (< 0.05). In these experiments, the H. e. venus male behaviour was not affected by hexane treatment (> 0.05). Overall these results show that the colour pattern has a role in mate choice, but in the case of H. e. chestertonii, chemical signals are also important.

Figure 2.

 Relative probabilities in wing pattern choice experiments. The males being tested are shown below the x-axis, H. e. chestertonii (left) and H. e. venus males (right). Dashed lines separate the treatments with intact dissected wings on the left and hexane-washed wings on the right. Wings above data points show the colour pattern of the experimental models (H. e. chestertonii, H. e. venus and F1 in each case). The estimated probabilities of approaching (brown) and courting (green) the experimental model relative to a control with the same pattern as the male being tested are shown. The specimens in this figure were taken of Museo de Historia Natural ANDES collection: ANDES-E12320, ANDES-E12321 and ANDES-E11363.

Analysis of post-mating isolation

The beta-binomial likelihood model detected significant differences in the hatch rate among all cross types (G, 8 = 46.40; = 1.98 × 10−07, Table 2). A considerable (nonsignificant) reduction in the hatching success for both reciprocal inter-specific crosses (14% and 10%, Table 2) was observed with respect to the controls (30% and 24%, Table 2). For unknown reasons, the values for control hatch rates are low; however, experimental and controls were treated in the same way and therefore should provide comparable estimates of hatch rate. Fertility in the backcrosses was similar irrespective of the pure phenotype involved in the cross. For example, there is not a statistical difference in fertility if an F1 female mates with an H. e. chestertonii or an H. e. venus male (> 0.05; Table 2). The 24 backcrosses showed an intriguing pattern whereby the hatch rate was dependent on the source of the mother. Thus, crosses of an F1 female (derived from an H. e. chestertonii mother and an H. e. venus father) to a pure male all produced a markedly reduced hatch rate (12%). In contrast, when the F1 female was derived from the reciprocal cross, the hatch rate was similar to controls (35%). In backcrosses involving F1 males, there was also a reduced hatch rate when the hybrid males were obtained from the cross H. e. chestertonii female by H. e. venus male (22%). When the survival of larvae was compared among crosses, no differences were observed (median rate equal to 0.76, Table 2) with the exception of the larvae obtained from the F1 hybrid females between an H. e. chestertonii female mated with an H. e. venus male (survival rate equal to 0.42, Table 2). Finally, none of the crosses showed any significant deviation from the expected 1:1 sex ratio (G, 1 < 3.84; > 0.05). Together, these experiments reveal a significant degree of post-mating isolation.

Contributions to reproductive isolation

A summary of the reproductive barriers evaluated in this study and their contribution to reproductive isolation between H. e. chestertonii and H. e. venus are shown in Table 3. Gene flow was reduced by 97.9% between the two species (Table 3). The pre-mating barriers (colour pattern preference and pheromone preference) strongly contribute to isolation, preventing 87% of inter-specific matings. It should be noted that our experimental design does not allow a separate estimate of mating propensity and mating preference, and ideally both should be taken into account in estimating sexual isolation (Rolan-Alvarez & Caballero, 2000; Rolan-Alvarez, 2004). However, biases because of a failure to estimate mating propensity are likely to be less in controlled insectary conditions when compared to the wild (Perez-Figueroa et al., 2005). Post-mating isolation barriers were also strong, although their absolute contribution was more minor because of their impact later in the life history after pre-mating barriers had taken effect (Table 3).

Table 3.   Contributions of different barriers to reproductive isolation between H. e. chestertonii and H. e. venus.
Species pairchestertonii/venushimera/eratomelpomene/cydno
Isolation barrierICACICACICAC
  1. Individual contribution (IC) was estimate as 1 − (heterospecific frequency/homospecific frequency) in each barrier (details in Methods and Ramsey et al., 2003). The absolute contribution (AC) to reproductive isolation from pre-mating effects (pheromone and colour pattern preference) is estimated from no-choice experiments with live females. This is likely to be reduced depending on the frequency of pupal mating in the wild (see text). The proportion because of colour pattern preference is estimated from model experiments, with the remainder assumed to be because of pheromonal cues. For comparison, values for the parapatric H. himera and H. erato cyrbia (McMillan et al., 1997) and the sympatric hybridizing species H. melpomene and H. cydno are shown (Jiggins, 2008). Colour pattern selection refers to mimicry selection against hybrid phenotypes. ?This has not been estimated directly in any case, but is assumed similar to that in wing pattern hybrid zones (Jiggins, 2008).

Habitat choice0.660.66
Colour pattern preference0.570.870.890.890.750.32
Pheromone preference0.3   0.880.018
Hybrid viability (hatch rate)0.560.0700
Colour pattern selection0.65?0.0390.65?0.0720.65?0.0016
Hybrid sterility000.70.0003
Total reproductive isolation 0.979 0.962 0.999


One approach to understanding speciation is to compare examples in different stages of divergence, thus assembling a broad picture of how reproductive isolation accumulates (Coyne & Orr, 2004). Hybrid zones show such a continuum of stages and are ideal places to study how barriers to gene flow evolve and their relative importance in species (Carling & Brumfield, 2008; Seehausen et al., 2008; Arntzen et al., 2009; Qvarnström & Bailey, 2009). Here, we have studied two taxa that form a bimodal hybrid zone in the western Cordillera of Colombia, H. e. chestertonii and H. e. venus. These are separated by partial pre- and post-mating isolation, showing that multiple sources of reproductive isolation can be important even in the early stages of speciation.

When they encounter one another as adults, H. e. chestertonii and H. e. venus show strong assortative mating. The probability of mating with a heterospecific is less than 10% that of mating with a con-specific. Nonetheless, when males of either species were presented with heterospecific females as pupae, they invariably succeeded in mating in 14 trials. H. erato are known to ‘pupal mate’ by guarding pupae and mating during or soon after female emergence (Gilbert, 1976; Deinert et al., 1994), but it is unknown what proportion of matings in wild populations occur this way. Insectary populations are variable in their propensity to pupal mating (McMillan et al., 1997). Larger sample sizes might reveal assortative mating even during ‘pupal mating’, but it is clear that the strength of premating isolation in the wild will depend on the relative frequency of matings that occur away from the pupae. This, in turn, is likely to be variable between populations depending on the density and distribution of larvae and host plants.

Nonetheless, adult males clearly show strong preferences for conspecific colour patterns as demonstrated in the experiments with wing models, leading to assortative mating preferences. They similarly are less likely to approach F1 wing patterns, implying that hybrids may be selected against in mating. This is consistent with previous experiments across a wide variety of Heliconius taxa that have demonstrated the use of wing patterns in mate selection (Estrada & Jiggins, 2008; Melo et al., 2009). In particular, a previous study of visual mate preferences showed that several races of H. erato were more likely to approach their own pattern relative to other patterns (Estrada & Jiggins, 2008). Furthermore, males of two races of H. erato were able to distinguish wing models of their own species from the phenotypically highly similar forms of their mimic, H. melpomene (Estrada & Jiggins, 2008). However, this effect disappeared when the wings were washed in hexane, suggesting a likely role for chemical cues. This result prompted us to carry out the same manipulation here. Most strikingly, H. e. chestertonii males were around 80% as likely approach to F1 wing models when compared to their own pattern, but after hexane washing this fell to around 30%. This would seem to imply that the cuticular hydrocarbons used as cues by H. e. chestertonii males are largely dominant and therefore expressed in F1 males, but can be removed by hexane washes.

One of the most striking aspects of the hybrid zone studied between H. himera and H. erato was a complete lack of any detectable intrinsic hybrid sterility or inviability (McMillan et al., 1997). In contrast, here, we have demonstrated a marked reduction in hatching rate among the offspring of both inter-specific crosses. However, this reduction was only observed in the backcrosses involving F1 females from the H. c. chestertonii female by H. e. venus male cross. It remains unclear whether this is a sperm–egg incompatibility, perhaps resulting in a failure of fertilization, or alternatively an F1 hybrid breakdown such that development fails in the early stages of embryogenesis prior to hatching. There was also a lack of Haldane’s rule, with no deviation from parity in sex ratios.

Overall, our data provide an explanation for the deviation from Hardy–Weinberg found in the H. e. chestertonii × H. e. venus hybrid zone in the Cauca Valley (Arias et al., 2008). Only around 25% of individuals collected in the zone of contact were hybrids, significantly fewer than would be expected under Hardy–Weinberg equilibrium, implying partial reproductive isolation between the taxa. Here, we have shown that this is likely a result of both pre- and post-mating barriers, representing approximately a 98% reduction in gene flow between parental forms. This is slightly higher than that estimated for H. himera and H. erato cyrbia, which is surprising as hybrids are rarer in the latter hybrid zone. However, these estimates do not include many ecological factors, and also likely include considerable error. In particular, for example, for inter-specific mate choice involving H. e. venus females where sample sizes were small in this study. Overall, the two pairs of taxa are similar in both showing strong pre-mating isolation, if all mating is assumed to occur between adult butterflies (Table 3). Other closely related Heliconius species and races also show similarly strong pre-mating isolation (Estrada & Jiggins, 2008; Melo et al., 2009). In contrast, post-zygotic isolation observed here is much greater than that seen between H. himera and H. erato (McMillan et al., 1997), and more similar to that between the more divergent sympatric species H. cydno chioneus and H. melpomene rosina from Panama (Naisbit et al., 2001). The only other example of post-mating isolation between geographical races of Heliconius involves crosses between H. m. melpomene from French Guiana and H. m. rosina from Panama, in which partial hybrid sterility was observed. However, in this case, these are geographically distant populations that may be separated through isolation by distance, rather than representing incipient parapatric species (Jiggins et al., 2001b). Thus, in the Cauca valley H. erato hybrid zone, post-zygotic isolation has evolved faster than in other Heliconius species.

One of the current areas of interest in speciation research is the question of whether one strong or many weak barriers to gene exchange are important in the early stages of speciation (Coyne & Orr, 2004; Schluter, 2009). Heliconius butterflies are one of a number of cases where the importance of particular ecological traits has been emphasized. Butterfly wing patterns, the beaks of Darwin’s Finches or the body size of lake sticklebacks are all traits involved in ecological speciation, and have been highlighted as major contributors to this kind of speciation (Podos, 2001; Jiggins, 2008; Schluter, 2009). Nonetheless, it is important to consider the contribution of all possible barriers (Coyne & Orr, 2004) to assess the relative importance of these ecological factors. Several other recent studies of incipient species have demonstrated the importance of multiple factors of more minor effect in reproductive isolation (Ramsey et al., 2003; Dopman et al., 2010). For example, between the E and Z strains of the European Corn Borer, seven of 12 potential sources of reproductive isolation were found to make a significant contribution (Dopman et al., 2010).

The difference may relate to the relative importance of geographical isolation. Where speciation is driven by strong selection in the face of gene flow, one or a few major factors that are under strong selection are likely to play a key role, whereas speciation involving geographical separation is more likely to involve accumulation of multiple factors of more minor effect (Coyne & Orr, 2004). In the case of H. e. chestertonii, there are several barriers to gene exchange with H. erato, perhaps consistent with the fact that the Cauca valley is largely isolated from the Pacific slopes apart from a few narrow zones of contact, suggesting a history of limited parapatric contact. In the future, it will be interesting to further investigate the correlation between genetic architecture and the geographical context of speciation.


We thank members of Instituto de Genética at Bogota for their help in maintaining the butterflies stocks at La Vega. AM was partially supported by Facultad de Ciencias, Universidad de los Andes. ML was funded by Banco de la República, and private donations (Continautos S.A., Proficol El Carmen S.A. and Didacol S.A). Both ML and CS received funds from COLCIENCIAS grant number 120 440 520 138. CJ was fund by a Royal Society University Research Fellowship and a Leverhulme Trust grant.