Rike Stelkens, Department of Ecology and Evolution, University of Lausanne, Biophore, CH-1015 Lausanne, Switzerland. Tel.: +41 21 692 4249; fax: +41 21 692 4165; e-mail: firstname.lastname@example.org
The hypothesis of ecological divergence giving rise to premating isolation in the face of gene flow is controversial. However, this may be an important mechanism to explain the rapid multiplication of species during adaptive radiation following the colonization of a new environment when geographical barriers to gene flow are largely absent but underutilized niche space is abundant. Using cichlid fish, we tested the prediction of ecological speciation that the strength of premating isolation among species is predicted by phenotypic rather than genetic distance. We conducted mate choice experiments between three closely related, sympatric species of a recent radiation in Lake Mweru (Zambia/DRC) that differ in habitat use and phenotype, and a distantly related population from Lake Bangweulu that resembles one of the species in Lake Mweru. We found significant assortative mating among all closely related, sympatric species that differed phenotypically, but none between the distantly related allopatric populations of more similar phenotype. Phenotypic distance between species was a good predictor of the strength of premating isolation, suggesting that assortative mating can evolve rapidly in association with ecological divergence during adaptive radiation. Our data also reveals that distantly related allopatric populations that have not diverged phenotypically, may hybridize when coming into secondary contact, e.g. upon river capture because of diversion of drainage systems.
As long as divergent selection outweighs gene flow, ecological adaptation can cause rapid changes in the mating system (Schluter, 2001). One might hence predict that selection-driven evolution of mating preferences is prominent and dominates over other mechanisms during adaptive radiations, when multiple available niches are colonized in short succession by the same ancestral population. The ideal study system to test this prediction is one that comprises populations having evolved in geographical isolation but under similar ecological selection regimes, and also populations having evolved in an environment without geographical barriers but with different selection pressures between alternative niches. We recently discovered such a system in the upper Congo (Zambia/DRC). The cichlid fish Pseudocrenilabrus philander is geographically widely distributed in rivers, swamps and lakes of South and East Africa. Most populations are ecologically unspecialized, and phenotypically similar. Such populations occupy the upper Luapula/Congo River system and Lake Bangweulu. Yet, in nearby Lake Mweru (Zambia/DRC) a population coexists with a diverse and phylogenetically young adaptive radiation that emerged from the same species.
Pseudocrenilabrus takes a basal position in the phylogeny of haplochromine cichlids, a group that comprises all large African cichlid fish radiations except some of those in Lake Tanganyika (Joyce et al., 2005; Katongo et al., 2005; Koblmueller et al., 2008). Pseudocrenilabrus has been described as a species-poor genus that, in contrast to some other haplochromines, did not radiate into sympatric species (Katongo et al., 2005). It currently comprises three nominal species (P. philander, P. nicholsi and P. multicolor). Of these, P. philander has the widest geographic distribution throughout South-Central, South-East and South Africa. It contains several subspecies and shows strong biogeographic sub-structuring (Skelton, 1991). This species and both others are morphologically rather unspecialized, ecologically generalized cichlids of small body size (adult males are 5–8 cm, females are 4–6 cm long) that typically inhabit slow flowing parts of rivers, swamps and sink holes (Greenwood, 1989). They are generally thought of as poor dispersers. In lakes, they are typically restricted to the marginal habitats, i.e. shallow shores with sandy or muddy bottoms and dense vegetation where they feed on insect larvae, detritus and algal material. Apart from Koblmueller et al. (2008), who described a population from a small tributary of Lake Tanganyika containing two closely related but genetically distinct colour morphs, which they suggest are currently undergoing incipient speciation, there was no evidence for sympatric or parapatric species diversity anywhere within the genus.
During fieldwork on Lake Mweru, we discovered at least 13 distinct phenotypes of Pseudocrenilabrus within the lake, differing in size, body shape, jaw morphology and male colouration, most exhibiting novel phenotypes not reported from anywhere else in the large distribution range of the genus. Strikingly, most of them occur in sympatry. Comparative analysis of morphometric distances reflecting variation in ecologically relevant shape elements, revealed that this Pseudocrenilabrus species flock diversified in phenotypic dimensions resembling those of the classical cichlid adaptive radiations (R.B. Stelkens et al., unpublished data). Yet, phylogenetic reconstruction revealed that there are only two major mitochondrial haplotype lineages in P. philander, one of which is endemic to Lake Mweru while the other one is widely distributed in lakes and rivers of East and Southern Africa, including Lake Mweru and Lake Bangweulu in the same drainage system (Stelkens et al., in preperation). Lake Mweru was apparently colonized by both of these deeply divergent lineages, but while one diversified rapidly and strongly dominates in all the endemic species, the other lineage went almost extinct in the lake.
The different geographic populations of P. philander, together with the adaptive radiation in Lake Mweru, provide the rare opportunity to compare effects of strong genetic but weak ecological and phenotypic divergence in geographical isolation with effects of strong ecological and phenotypic despite weak genetic divergence on reproductive compatibility.
We conducted mate choice experiments in the laboratory using four different groups: (i) an allopatric P. philander population from Lake Bangweulu, representing one of the two mitochondrial lineages that seeded the Lake Mweru radiation and presumably resembles the ancestral phenotype (Fig. 1a); (ii) a sympatric P. philander population from Lake Mweru, representing the other mitochondrial lineage that seeded the radiation and also resembles the ancestral phenotype (Fig. 1b); (iii) a sympatric species (Pseudocrenilabrus sp. ‘grey back’) from Lake Mweru, representing an ecologically derived phenotype (Fig. 1c); (iv) another sympatric species (Pseudocrenilabrus sp. ‘weed picker’) from Lake Mweru (Fig. 1d), representing another ecologically derived phenotype with a different ecological specialization than (c). The three members of the adaptive radiation (ii, iii and iv) are genetically closely related to each other while the allopatric P. philander population (i) represents a genetically distant lineage. The radiation members differ distinctly in male breeding colouration, whereas the Lake Bangweulu P. philander resembles the Lake Mweru P. philander.
First, we tested for assortative mating among the three phenotypically and ecologically divergent putative species from Lake Mweru (ii, iii and iv). Then we tested the two allopatric P. philander against each other (i vs. ii). If loss of premating compatibility resulted from evolution in geographical isolation, we would predict stronger assortative mating between the genetically divergent populations from different lakes. On the other hand, if loss of premating compatibility resulted from divergent adaptation, we would predict elevated levels of assortative mating among the sympatric incipient species.
Materials and methods
The four populations were collected in Lakes Mweru and Bangweulu in September 2005. Sampling occurred on several days using beach seine nets covering a large area (approximately 100 m of lake shore). Both lakes are part of the Upper Congo River system and connected by the Luapula River that drains Lake Bangweulu in the south, then bends northwards and feeds into Lake Mweru through large swampy areas (Fig. 2) (Jackson, 1961). Although the linear distance between the lakes is only 250 km, the length of the connecting river is approximately 550 km.
From each lake, we collected one population of P. philander (Skelton, 1991). The two populations are phenotypically similar to each other and match the typical appearance of P. philander across much of South and East Africa (Fig. 1a,b). Both have bright yellow–green flanks and ventra with a greyish to brownish dorsum. The anal fins of territorial males are intensely blue with rows of transparent-white or orange–red speckles on the membranes. Dorsal and caudal fins are, to varying degrees, yellow, red and orange and also show speckles. The pelvic fins turn black when males are territorial. Males are further characterized by a bright sky blue line extending from the lower lip across the cheeks to the operculum, a trait widely distributed across the geographical range of the species. Females share the overall colour of the males but are less intense making them appear pale brownish or grey. The only conspicuous difference between the two populations is that males from Lake Mweru display a bright red spot on the trailing edge of the anal fin, while this spot is bright orange in males from Lake Bangweulu.
Two new types of Pseudocrenilabrus
Pseudocrenilabrus sp. ‘grey back’ and Pseudocrenilabrus sp. ‘weed picker’ occur in sympatry with P. philander in Lake Mweru, but they differ strikingly in morphology and colouration (Fig. 1c,d). Pseudocrenilabrus sp. ‘weed picker’ was caught in a reed belt in the littoral zone. Weed picker has a different head morphology than any of the other types studied here. While philander has a curved head profile, weed picker has a pointy snout. In comparison to cichlids in the larger African radiations, this indicates that it may pick insect larvae from between and on the vegetation. Weed picker males display bright orange–yellow flanks and only a thin red spot on the trailing edge of the anal fin.
Pseudocrenilabrus sp. ‘grey back’ was collected further offshore in deeper waters with no emergent vegetation. It is easily and consistently identified by its streamlined body shape, a thinner, elongated caudal peduncle, a conspicuously shaped ‘horse-head’ with a deep dorsal head profile, and strikingly different breeding colouration. Nuptial males have bright yellow flanks without the greenish hue of philander and a characteristically metallic-grey dorsum, sharply defined against the yellow flanks. The lower anterior half of the caudal fin is conspicuously orange. Grey backs lack the characteristic blue anal fin and blue lower lip of philander. Their anal fin is mainly transparent with a yellow hue and only a very thin line of red on the trailing edge. The body shape together with the habitat where grey backs were caught, suggest that it is an offshore demersal feeder. Females of both ‘new’ types resemble the males morphologically. In the case of grey back, females show the same distinctly metallic-grey dorsal colouration but it extends over all of the body.
To query the different types for their mating preferences in the laboratory, live fish of all four groups (philander Bangweulu, philander Mweru, grey back Mweru and weed picker Mweru) were shipped to our laboratory at EAWAG, Switzerland.
Experiments were carried out in 196 × 40 × 40 cm large aquaria, divided into three equally sized compartments (Fig. 3). To measure female mating preferences based on actual spawning events, we used the ‘egg catcher’ experimental design described in Kidd et al. (2006) and previously applied to assess cichlid mate choice (Nelson, 1995; Plenderleith et al., 2005). The design allows for full completion of spawning in two-way choice experiments through clear, ultraviolet-transparent Plexiglas walls separating the female compartment in the middle from the two adjacent male compartments to the left and right. The walls were perforated with holes of 0.5 cm diameter every 2 cm2, allowing females to assess the full sensory repertoire of male mating cues, including chemical (Plenderleith et al., 2005) and visual cues (Seehausen & Van Alphen, 1998) as well as sound (Amorim et al., 2004). Directly in front of the two walls, inside of the female compartment, egg collectors were installed as rectangular boxes running across the entire depth of the tank. The collectors were covered with plastic mesh with holes large enough to allow eggs to fall through during spawning. These egg repositories allowed the observer to assign egg counts to the male behind the transparent wall. The male that obtained the larger number of eggs was considered the preferred sire. Male to male aggressive behaviour was physically constrained, thus not interfering with female mate choice decisions (Morris et al., 1992; Kodric-Brown, 1993). We constructed spawning shelters in both male compartments using four bricks; two serving as walls and the other two as floor and roof leaving two sides open, one side for the male to enter the cave and the other side facing the female compartment and the egg collector. The bottom of the male shelter was at level with the plastic mesh covering the egg collectors on the other side of the perspex wall, forming a platform for the courtship display preceding each spawning event.
Water circulation in the tank was manipulated such that water constantly passed through the perforated plastic walls from both male compartments into the middle compartment, from where it was pumped out and into an external Eheim aquarium filter, and back into the tank through two separate inflows, one in each male compartment (Fig. 3). In this way, females were exposed to water enriched with pheromones of both males. Once a week, one-third of the water was exchanged by tap water. Water temperature was kept constant at 24–25 °C and fish were kept on a 12L : 12D light regime using standard aquarium fluorescent light tubes with full spectrum lighting. Fish were fed daily with flakes and twice a week with a blend of fresh food made from shrimps, peas and Spirulina powder. The bottom of all three compartments was covered with a 2-cm thick layer of aquarium sand.
Four different experiments were conducted between May 2006 and January 2008. The first two experiments tested for assortative mating between the Lake Mweru population of ancestral phenotype and the two derived Lake Mweru species, all of them sympatric and members of the endemic radiation: Experiment 1, philander Mweru vs. grey back; Experiment 2, philander Mweru vs. weed picker. The third experiment tested for assortative mating between the two derived members of the endemic radiation: Experiment 3, grey back vs. weed picker. The fourth experiment tested for assortative mating between the allopatric populations of the ancestral phenotype; Experiment 4, philander Mweru vs. philander Bangweulu.
All males and females used in this study were first generation offspring (F1) of wild caught fish (every F1 family was derived from a different male–female combination), except for Experiment 2 where only wild caught fish were used. Prior to the experiments females had never seen the specific males they were offered, and experimental males and females never were siblings. Females had no prior experience with heterospecific males, but were raised in stock tanks with their brothers.
Females were introduced in the middle compartment 2 days before one male was put into each of the adjacent compartments. Only females of one of the two species tested were used per trial. To facilitate behavioural acclimatization with minimum levels of aggression, females were introduced to the experimental tank in groups of 3–14 individuals, all of which were from the same sib group. After a spawning had occurred, all females of that sib group were removed from the experiment and replaced by a different sib group of females. Sib groups were only used once and only spawnings of females from different families were considered genetically independent data points. It is in principle possible that an egg clutch derived from more than one female, i.e. if one spawning was closely followed by another, but it is unlikely that this would have gone unnoticed as we watched spawnings carefully and checked for eggs at least three times a day (morning, noon and evening). Also, female spawners were identified during or immediately after spawning, and in no case did more than one female show the symptoms of a recent spawning event. Female spawners were identified by inspecting the ventral region (convex before, concave after spawning) and the genital pore, which swells and protrudes before and during oviposition. Occasionally, a female was found with some eggs in her mouth. These eggs were not used for analysis as we were not able to determine with which male the female had spawned.
The position of male types in the aquarium was randomized over trials. Across all experiments, eggs were laid on average every 9.3 days. As soon as eggs were discovered in one or both egg collectors and after allowing sufficient time to ensure that the spawning was completed, eggs were removed and both males were exchanged against two new males of the same type, that were not brothers of the preceding males. Only in one case (Experiment 1), not enough grey back males were available and one male was used in three different trials, but paired with a different opponent in each case. Upon removal, males were measured with a ruler to the nearest millimetre to control for size effects, before they were put back into separate stock tanks.
In Experiment 2, only wild caught males and females of unknown pedigree were used. Due to the sampling method we used to catch these fish (several days of beach seining covering a large sampling area), it is unlikely that they were from the same family. Hence, in Experiment 2, every female was allowed to spawn before it was removed from the experiment together with the two males, which were replaced by a new pair of males as described above. Here, all spawnings were treated as genetically independent data points.
Analysis of mate choice data
The data from the mate choice experiments were tested against the null hypothesis of random mate choice. We used paired t-tests on the number of eggs a female spawned with each male. To relax the normality assumption, we also applied Wilcoxon matched-pairs tests assigning males to categories of ‘winners’ and ‘losers’, depending on which of the two obtained more eggs.
Paired t-tests were applied to see if body size differed between the males of the two species used in each experiment. We also performed paired t-test on the body size of winners and losers to test if size had an influence on mate choice, regardless of species identity.
To quantify the degree of premating isolation between any two groups, isolation indices were calculated using Itrial = (Econ − Ehetero)/Etotal where Econ − Ehetero are the number of eggs spawned with con- and heterospecific males in a trial and Etotal is the total number of eggs in that trial. I can range from −1 to 1, with negative values indicating a preference for heterospecific males and positive values indicating a preference for conspecifics. A value of zero would indicate no preference. Results were averaged over trials within experiment, to obtain a measure of isolation for each of the two female types tested (Iexp1–4_female type). This was necessary to detect asymmetry in assortative mating depending on which female type was asked to choose between con- and heterospecific males.
To obtain an overall isolation index for each of the four group comparisons (Iexp1–4), we then averaged over the two female type-specific isolation indices.
Analysis of genetic distance
To compare genetic distances between the four groups used in our mate choice trials, we calculated pairwise genetic distances as uncorrected p-distances from mitochondrial DNA sequences (D-loop) of between 4 and 14 individuals per group (Mendelson, 2003; Chapman & Burke, 2007). All distances were corrected for within-group variation among haplotypes by subtracting mean within-group distances from mean pairwise between-group distances (Nei, 1987; Mendelson, 2003) (Table 1b).
Table 1. (a) Premating isolation indices for all four groups tested in mate choice experiments. (b) Pairwise genetic distance calculated as uncorrected p-distances from DNA sequences of the mitochondrial D-loop region. (c) Phenotypic distance (Pst) calculated from PC1 scores of 13 ecomorphological measurements. (d) Phenotypic distance (simple mismatch coefficients) calculated from male colouration scores at 10 morphological landmarks.
weed picker Mweru
grey back Mweru
The number of individuals per group is shown in brackets.
(a) Premating isolation
weed picker Mweru
grey back Mweru
(b) Genetic distance
philander Bangweulu (12)
philander Mweru (4)
weed picker Mweru (10)
grey back Mweru (7)
(c) Phenotypic distance (ecomorphology)
philander Bangweulu (16)
philander Mweru (15)
weed picker Mweru (13)
grey back Mweru (18)
(d) Phenotypic distance (male colouration)
philander Bangweulu (7)
philander Mweru (6)
weed picker Mweru (13)
grey back Mweru (8)
Analysis of ecomorphological distance
To quantify ecomorphological variation within and among the four groups, we calculated phenotypic distances from 13 linear morphological characters measured with digital callipers to the nearest 0.01 mm on between 13 and 18 individuals per groups (Table 1c). All morphological measures were taken on wild fish. The traits measured were: standard length (SL), body depth (BD), head length (HL), head width (HW), snout length (SnL), snout width (SnW), lower jaw length (LjL), lower jaw width (LjW), eye length (EyL), eye depth (EyD), interorbital width (IoW), cheek depth (ChD) and preorbital depth (PoD). Measuring differences in morphological characters is a standard technique to detect differences in resource use and has been shown to reflect adaptation to different habitats in fishes and cichlids in particular (Caldecutt & Adams, 1998; Ruber & Adams, 2001; Hendry et al., 2002; Kassam et al., 2003; Clabaut et al., 2007).
Raw measurements were log-transformed prior to all analyses to homogenize variance. Normal distributions for all traits were confirmed with Shapiro-Wilk tests. There were significant differences in standard length between groups (one-way anova: F3,61 = 2.99, P = 0.038). A Tukey-Kramer test showed that weed picker were on average longer (77.45 cm) than all other species and significantly differed from grey backs (64.65 cm). Philander Mweru (71.58 cm) and philander Bangweulu (68.38 cm) were intermediate in size.
Because of the size differences between types, we applied statistical corrections following methods described in Hendry et al. (2000). First, linear regressions were calculated within groups for each of the traits against either SL (in case of BD, HL) or HL (in case of the remaining nine traits). These group-specific slopes (b) were then used to standardize each measurement (Mstd) by the observed standard length of each fish (Lo) divided by the mean standard or head length (Lx) of all four groups pooled using the expression Mstd = Mo(Lo/Lx)b. The adjusted trait values were entered into a principal component analysis (PCA). One-way anovas were run both on standardized traits (Mstd) separately and on the first three principle components to detect between-groups differences. Tukey-Kramer tests were applied for pairwise population comparisons.
Canonical linear discriminant analysis (Ostbye et al., 2005) was used to explore if individuals could be correctly reassigned to their source group based on the adjusted, morphological traits (Mstd) alone. This method calculates the distance of each individual to the multivariate group mean (Mahalanobis distance), which was visualized by plotting canonical axis 1 and 2 (Fig. 5). It also allowed for ranking the 13 traits based on their discriminating power using a stepwise forward variable selection model where the highest ranking trait was entered first.
The divergence in quantitative phenotypic traits between groups was estimated by calculating Pst values [the phenotypic surrogate for Qst (Spitze, 1993)] for each group comparison. We calculated Pst for each of the first three principal components of the size adjusted morphological data. We then calculated a weighted average over all three components, using the amount of variance explained by each component as weight, to obtain a total measure of phenotypic divergence on the major axes of shape variation.
Analysis of male nuptial colouration
To quantify variation in male nuptial colouration within and among groups, we visually scored the colour of males in breeding dress at 10 different places on body and fins (Fig. 1c). All measurements were taken on wild fish. At every landmark, colour was scored as either translucent, white, grey, black, blue, green, yellow, orange or red. Scoring was done by one individual (OS) without knowledge of the source population of the fish. To convert the qualitative score data into a distance metric, we used the ‘simple matching coefficient’ (SMC) (Digby & Kempton, 1987) dividing the number of matching characters by the total number of characters in individual pairwise comparisons. We collected data from photographs of between 6 and 13 different wild caught males per group (Table 1d). All colour distances were corrected for within-group variation by subtracting the average intrapopulation distance from the average interpopulation distance. Canonical linear discriminant analysis was used to explore if individuals could be correctly reassigned to their source population based on their colour scores alone.
Testing genetic and phenotypic distance as predictors of premating isolation
To test the effects of genetic and phenotypic distance on the degree of premating isolation, we regressed premating isolation (Iexp1-4) against genetic distance (uncorrected p-distance), ecomorphological distance (Pst) and colouration distance (SMC) (Fig. 7) using either linear (morphology and genetic distance) or logarithmic (colouration) regression models. Normality of distribution was tested with Shapiro-Wilk tests and genetic distance data was log-transformed to acquire normality.
We predicted that if phenotypic differentiation promoted premating isolation, then significant assortative mating would be observed in experiments 1–3 (genetically closely related but phenotypically divergent species from same lake). To test this, we performed a two-factorial anova with egg number per male as dependent variable. The factors were ‘female type’ (fixed with two levels), ‘male type’ (fixed with two levels) and the interaction term between them (female type × male type). With the null hypothesis of random mating, our prediction of assortative mating between the three ecologically different members of the Lake Mweru radiation would be confirmed by a significant interaction term between male and female type.
In Experiment 4, we tested males and females representing distantly related populations of similar phenotype but from different lakes (i.e. different ‘origins’). To compare the importance of phenotypic divergence vs. genetic distance, we performed a two-factorial anova on egg number with the factors ‘female origin’ and ‘male origin’ (either Mweru or Bangweulu) and the interaction term between them (female origin × male origin). Here, the null hypothesis assumes assortative mating, hence the nonsignificance of the interaction term would support our prediction, that ‘origin’ has no impact on the strength of premating isolation.
Lastly, we tested on results pooled across all four experiments if the interaction between female and male type term remained significant when genetic distance (uncorrected p-distance) was added to the model (testing ‘female type’, ‘male type’, the interaction term between them and genetic distance on the number of eggs). All analyses were performed in jmp professional 6.0.0 (SAS Institute, Cary, NC, USA).
Mate choice experiments
In Experiment 1, 10 spawnings were obtained from each group (grey back, philander Mweru; Fig. 4a). Philander Mweru females spawned on average 19.5 ± 21 eggs with philander Mweru males and only 3.2 ± 5.1 eggs with grey back males. Grey back females spawned 16.8 ± 25.7 eggs with grey back males and 7.9 ± 13 eggs with philander Mweru males. Philander Mweru females preferred their own male type (t9 = 2.14, P =0.031; all P-values one-tailed, Fig. 4a). Grey back females on the other hand did not mate significantly assortatively (t9 = −0.86, P = 0.205). Overall, there was significant assortative mating in this species pair (t19 = −2.0, P = 0.030). This pattern was confirmed when ranking the paired males of each trial by the number of eggs they obtained (categorizing males as either ‘winners’ or ‘losers’). With philander females, philander males obtained significantly more often a higher number of eggs than grey back males (Wilcoxon matched-pairs test (all P-values one-tailed): Z = −1.99, P = 0.024). There was no such effect when ranking males in trials with grey back females (Z = −0.97, P = 0.166).
In Experiment 2, 13 spawnings were obtained from each group (weed picker, philander Mweru; Fig. 4b). Philander Mweru females spawned on average 36.8 ± 23.3 eggs with philander Mweru males and only 5.2 ± 9.6 eggs with weed picker males. Weed picker females spawned 16.8 ± 25.4 eggs with weed picker males and 23.4 ± 19.1 eggs with philander Mweru males. While weed picker females did not mate assortatively (t12 = −0.58, P = 0.285; Fig. 4b), philander Mweru females again mated strongly assortatively (t12 = −3.89, P = 0.001). Overall, there was a trend for assortative mating in this species pair (t25 = 1.59, P = 0.061). Again, this pattern was confirmed when ranking the paired males of each trial by the number of eggs they obtained. With philander Mweru females, philander Mweru males obtained significantly more often a higher number of eggs than weed picker males (Z = −2.69, P = 0.004). There was no such effect when ranking males in trials with weed picker females (Z = −0.91, P = 0.176).
In Experiment 3, six spawnings were obtained from grey back females and five spawnings from weed picker females (Fig. 4c). Grey back females spawned on average 17.8 ± 15.4 eggs with grey back males and only 2.3 ± 3.01 eggs with weed picker males. Conversely, weed picker females spawned 23 ± 20.8 eggs with weed picker males and only 4 ± 6.5 eggs with grey back males. Significant assortative mating was found for grey back females (t5 = 2.14, P = 0.043) and a assortative trend was observed for weed picker females (t4 = −1.69, P = 0.055; Fig. 4c). Overall, there was significant assortative mating in this species pair (t10 = −2.79, P = 0.010). Ranking the males by the number of eggs obtained yielded a strong trend for grey back males to be more successful with grey back females (Z = −1.57, P = 0.055) whereas there was no trend for weed picker males to be more successful with weed picker females (Z = −1.22, P = 0.116).
In Experiment 4, 10 spawnings were obtained from each group (philander Bangweulu, philander Mweru; Fig. 4d). Bangweulu females spawned on average 17.3 ± 16.6 eggs with Bangweulu males and 5.5 ± 7.5 eggs with Mweru males. Mweru females spawned 6.5 ± 9.6 eggs with Mweru males but 9.3 ± 8.8 eggs with Bangweulu males. No assortative mating was found for either of the female types (philander Bangweulu females: t9 = 1.7, P = 0.061; philander Mweru females: t9 = 0.58, P = 0.29; Fig. 4d). The two male types were not ranked significantly differently when competing for Bangweulu females (Z = −1.58, P = 0.072) or for Mweru females (Z = −1.01, P = 0.155).
Premating isolation index
Testing for assortative mating between the ancestral philander phenotype from Lake Mweru and the sympatric derived grey back (Experiment 1) yielded similar premating isolation indices for both female types (Iexp1_philander Mweru = 0.377; Iexp1_grey back = 0.43, positive values indicate assortativeness), resulting in the highest overall isolation index observed in this study (Iexp1 = 0.404).
The same test revealed assymmetric isolation between the ancestral phenotype and the other derived sympatric type, weed picker (Experiment 2). While philander Mweru females clearly preferred their own type, weed picker females did not (Iexp2_philander Mweru = 0.228; Iexp2_weed picker = −0.237), resulting in weaker overall premating isolation for this species pair (Iexp2 = 0.227).
Both females of the two sympatric derived types from Lake Mweru (Experiment 3) mated assortatively (Iexp3_grey back = 0.442; Iexp3_weed picker = 0.36), resulting in a high overall isolation index (Iexp3 = 0.401).
When philander females from Lakes Bangweulu and Mweru were choosing between males of their own population vs. the distantly related allopatric population (Experiment 4), only Bangweulu females preferred males of their own population (Iexp4_philander Bangweulu = 0.368), while Lake Mweru females did not (Iexp4_philander Mweru =−0.246), yielding a low overall premating isolation index (Iexp4 = 0.061). All premating isolation indices are summarized in Table 1a.
The effect of body size on mate choice
Male body size had no effect on female preference. Winners and losers did not differ in body size in any of the experiments. There was one exception to this: in Experiment 4, philander Mweru females chose significantly more often to lay more eggs with the bigger male [t = 2.33, P (one-tailed) = 0.022]. At the same time, the two different male types (philander Mweru and philander Bangweulu) did not differ significantly in body size. Conversely, in Experiment 2 with weed picker females, significant differences in body size were found between weed picker (average size = 7.17 ± 0.86 cm) and philander Mweru males (average size = 7.75 ± 0.74 cm), but when assigning the males of this experiment to the winner and the loser category, no difference in body size was detected [t = 0.99, P (one-tailed) = 0.17]. Weed picker females hence did not generally prefer to mate with the larger male, regardless of species identity. Likewise, in Experiment 1, philander males (average size = 5.64 ± 1.1 cm) were larger than grey back males (average size = 4.97 ± 1.33 cm), but philander Mweru females did not preferentially mate with the larger male [t = −0.38, P (one-tailed) = 0.37]. In Experiment 3, no size differences were detected between grey back and weed picker males, and female mate choice was not affected by size.
Pairwise genetic distances found between allopatric populations were 30 times greater than the genetic distance between the three sympatric species from Lake Mweru (distances between all four populations are summarized in Table 1b).
Significant between-group differences were detected for each of the body size-adjusted morphological traits in one-way anovas, except for head length, snout length, cheek depth and preorbital depth (all other P < 0.05; morphological raw data for each of the 13 traits with mean and SD per group is provided as Supporting Information (Table S1). Tukey-Kramer tests showed that the allopatric philander Bangweulu and philander Mweru populations did not significantly differ in any trait except for eye depth. This is consistent with the prediction that the two populations indeed have similar morphological adaptations and inhabit similar ecological niches in the two different lakes although more direct tests of niche overlap would be required to confirm this.
Grey backs were significantly more elongate (smaller body depth) and had narrower snouts than all other groups. They also had a shorter but wider lower jaw, and larger eyes than all other groups. This may reflect their putatively demersal foraging mode, feeding on detritus and insect larvae in the offshore demersal zone of the lake.
Weed pickers were best characterized by their head width, which was significantly larger than in any of the other groups. Further, they had the longest lower jaws and snouts of all, consistent with feeding on insect larvae and other benthic animals in the littoral vegetation zone.
When performing a PCA on all standardized traits [without standard length because it was already incorporated into the calculation of (Mstd)], the first three PCs explained 49%, 15% and 11%, respectively, of the variance in the data set (Fig. S1). There were significant differences between groups, not along PC1, but along PC2 and 3 (PC2: F3,61 = 23.71, P < 0.001; PC3: F3,61 = 10.41, P < 0.001). This method separated grey backs from the three other types along PC2. Grey backs occupy an area in morphospace that does not overlap with that of the other two sympatric species, and slightly overlaps with that of the allopatric population. The morphospace of weed picker shows considerable overlap with that of philander Mweru and philander Bangweulu along PC1 and 2, but was different from sympatric philander Mweru along PC3. On the contrary, philander Bangweulu and philander Mweru broadly overlapped in morphospace along all three axes. Philander Bangweulu showed some sub-grouping along PC1.
The canonical discriminant analysis correctly assigned all but seven out of 62 individuals to the four source groups (Pillai’s trace = 7.62; P < 0.001; Fig. 5). The seven mis-assignments all involved philander Bangweulu, but there was no wrong assignment among the three sympatric species from Lake Mweru. After applying stepwise forward selection of variables, five traits significantly contributed to the model (BD, HL, HW, LjL and ChD). The first, second and third canonical axes explained 62%, 26% and 12% of the variance in the data set.
Strongest phenotypic differentiation was found between the two ecologically derived specialists from Lake Mweru, weed picker and grey back (Pst = 0.445, all Pst-test results summarized in Table 1c). This was followed by the phenotypic distances between the sympatric generalist, philander Mweru and both derived specialists (philander Mweru vs. grey back Mweru: Pst = 0.434; philander Mweru vs. weed picker Mweru: Pst = 0.408). The allopatric generalist, philander Bangweulu was moderately different from one of the specialists from Lake Mweru (philander Bangweulu vs. grey back Mweru; Pst = 0.382). The divergence found between philander Mweru and philander Bangweulu was comparatively small (Pst = 0.369). Only the comparison between philander Bangweulu and weed picker Mweru revealed even more phenotypic similarity (Pst = 0.347).
In summary, largest distances in ecomorphology were found among the three sympatric Lake Mweru species, while the allopatric philander Lake Bangweulu was similar to two members (philander and weed picker) of the Lake Mweru radiation.
Divergence in male colouration
Canonical discriminant analysis revealed that the three sympatric species from Lake Mweru have completely nonoverlapping male nuptial colouration (Fig. 6). At the same time, philander Bangweulu and philander Mweru were relatively similar in colouration, and philander Bangweulu was intermediate between the two Lake Mweru species philander and weed picker. All but one of 35 individuals were correctly assigned to the four source groups (Pillai’s trace = 4.49; P < 0.001). The one exception was a philander Bangweulu which was more similar to weed pickers. After applying stepwise forward selection of variables, three traits significantly contributed to the model (colouration of dorsal fin, flanks and anal fin). The first, second and third canonical axes explained 95%, 4% and 0.5% of the variance in the data set.
Using metric distances calculated as SMC, highest dissimilarity in colouration was found between the two sympatric ecological specialists from Lake Mweru, weed picker and grey back (SMC = 0.607; all SMC-test results are summarized in Table 1d). Colour differences between the two ecologically derived species and the two generalists were small to moderately strong (philander Bangweulu vs. grey back: SMC = 0.546; philander Bangweulu vs. weed picker: SMC = 0.236; philander Mweru vs. grey back: SMC = 0.391; philander Mweru vs. weed picker: SMC = 0.346). The two populations of ancestral phenotype philander Mweru and philander Bangweulu were rather similar in colouration (SMC = 0.279). As with ecomorphology, the similarity between philander Bangweulu and weed picker was slightly surprising, given our prediction that the two generalists from the different lakes would show the highest phenotypic match.
However, as expected largest dissimilarities in colour were found among the three sympatric species from Lake Mweru, whereas the allopatric philander populations were rather similar in phenotype. Remarkably, the dissimilarity patterns in male colour between the four groups closely resembled those in ecomorphology.
Predictors of premating isolation
Phenotypic distance was a good predictor of premating isolation (Fig. 7). The greater the ecomorphological distance between groups, the stronger was assortative mating between them (linear regression: R2 = 0.96, F1,4 = 79.75, P = 0.012). The same pattern, but with a weaker effect on premating isolation, was observed for male colouration. Assortative mating increased with increasing differences in colouration but the relationship was not significant (logarithmic regression: R2 = 0.69, F1,4 = 4.37, P = 0.172). The slope of the curve of premating isolation against genetic distance, in contrast, was negative such that at larger genetic distances mating tended to be more random (linear regression: R2 = 0.87, F1,4 = 13.52, P = 0.067).
When females were allowed to choose between males of their own species and males of a phenotypically different, but closely related species from the same lake (Experiments 1–3) the interaction between male and female species was significant (R2 = 0.14, F3,114 = 7.33, P (model) < 0.001, P (interaction) < 0.001), confirming our prediction of assortative mating by phenotype. When females were given the choice between males of their own population and males of a phenotypically similar but distantly related population from a different lake (experiment 4) the interaction term between male and female origin remained non-significant (R2 = 0.09, F3,39 = 2.27, P (model) = 0.10, P (interaction) = 0.21) confirming our prediction that population origin (Lake Mweru or Bangweulu) did not play a role in mating decisions.
Genetic distance between groups did not predict the number of eggs laid when ‘species identity’ (i.e. the interaction term between female and male type) was left out of the model (on results of all four experiments: R2 = 0.02, F1,153 = 3.38, P = 0.07), again confirming that the amount of genetic differentiation per se had no effect on mate choice. When using both parameters, genetic distance and species identity, in a combined model on data from all four experiments to compare their respective effect sizes, species identity clearly exceeded the effect of genetic distance on premating isolation (R2 = 0.14, F1,153 = 7.18, P (model) < 0.001, P (identity) < 0.001, P (genetic distance) = 0.10).
One of the outstanding questions in our understanding of adaptive radiations is, which sources of selection initiate the diversification of species and how can assortative mating evolve in the face of gene flow. An idea that emerged early in the literature (Simpson, 1944) is that the access to new ecological resources can promote population divergence in early stages of adaptive radiation. This idea has received considerable empirical and theoretical support (Schluter, 2000; Rundle & Nosil, 2005). Many species in young adaptive radiations are genetically isolated by divergent mating preferences. Yet, whether the emergence of such premating barriers to gene flow typically can occur without episodes of geographical isolation between the diverging populations is still controversial, not least because empirical evidence from the wild is scarce.
The hypothesis of ecological speciation (Schluter, 2001) argues that assortative mating can evolve along with ecological adaptations to different environments, either because traits under divergent natural selection are linked to traits affecting mate choice (e.g. (Rundle et al., 2000; Vines & Schluter, 2006), or because sexual selection regimes differ between different ecological niches, inducing shifts in mate choice criteria between alternative traits or trait states (e.g. Boughman, 2002; Maan et al., 2006; Seehausen et al., 2008). Because the mechanisms of ecological speciation are thought to potentially work, regardless of the geographical scenario, the ecological speciation hypothesis is particularly interesting when trying to explain speciation in young adaptive radiations, i.e. after the colonization of a new habitat without strong a priori barriers to gene flow.
Here, we tested one prediction of the ecological speciation hypothesis (McPeek & Wellborn, 1998) that phenotypic rather than genetic distance explains the strength of assortative mating among members of a young radiation of cichlid fish of the genus Pseudocrenilabrus in Lake Mweru (Zambia/DRC). Three of the species tested in mate choice experiments represent genetically closely related, but ecologically and phenotypically divergent sympatric species. The first is the widely occurring and formally described P. philander, a generalist species that represents the ancestral phenotype in the Lake Mweru radiation. The second, Pseudocrenilabrus sp. weed picker, inhabits the densely vegetated littoral zone and is an ecologically derived specialist, feeding on insect larvae. The third, Pseudocrenilabrus sp. grey back, is an offshore demersal feeder representing another derived, highly specialized ecotype. Further, we tested P. philander from Lake Mweru against an allopatric population of P. philander from Lake Bangweulu (Zambia) which is genetically distant, but phenotypically and ecologically very close to P. philander from Lake Mweru.
We found that (1) females of the ancestral, generalist species from Lake Mweru mated assortatively against both of the sympatric, genetically closely related but phenotypically distinct species; (2) females of the two closely related but phenotypically distinct species from Lake Mweru preferred their conspecific males over males of the other derived species, but had no strong preferences for their own males over males of the ancestral phenotype; (3) females of the allopatric, genetically distant, but phenotypically similar populations of P. philander mated randomly with each other (Fig. 4). We found all three sympatric species highly differentiated in both ecomorphological traits and male breeding colouration (Figs 5 and 6). At the same time, the two allopatric, generalist populations were similar to each other in both colour and morphology, confirming our prediction of niche conservatism, i.e. the tendency that both have retained ancestral ecological characteristics associated with P. philander over much of its geographical range. Interestingly, the allopatric population of P. philander from Lake Bangweulu, though similar to P. philander from Lake Mweru in morphospace and colouration, was intermediate between all three Lake Mweru species, suggesting character displacement associated with speciation in Lake Mweru. The sub-grouping in morphospace observed within Bangweulu P. philander (Fig. S1) suggests that the ancestral population not only consisted of uniform generalist phenotypes but instead showed considerable phenotypic diversity which could be recruited for niche divergence during radiation.
When assessing different predictors of premating isolation, only phenotypic distance but not genetic distance explained the variance in strength of assortative mating (Fig. 7). Larger ecomorphological and male breeding colour differences were associated with larger premating isolation, while differences in ecomorphology and in male colouration were significantly positively correlated (Spearman’s rank correlation: Rs = 0.83, P = 0.042). These results are consistent with ecological speciation. However, it must be acknowledged at this point that the relationship between morphological differences and the ecological diversification of the populations studied here remains untested. Additional analyses of the functional morphology are necessary to test the adaptive relevance of body form variation of these fishes in different environmental niches. Also, for a conclusive test of our assumption that Lake Mweru’s phenotypic diversity and the differences in mating preferences have evolved because of ecologically divergent selection, the heritability of traits relative to the variation caused by developmentally plastic responses to inhabiting a particular habitat needs to be assessed. Another point to consider is that, unlike the sympatric comparison, the allopatric population comparison was only replicated once here. If other ecologically similar allopatric populations mated assortatively, our conclusion that random mating was due to an absence of ecological speciation between allopatric populations would have to be revaluated. Hence, this study would benefit from further replication of the allopatric comparison.
Although we do not know the phenotypic traits that female Pseudocrenilabrus from Lake Mweru use in mate choice, the correlation between ecomorphological and nuptial colour divergence suggests an important role of ecological selection in the origin of divergent mating preferences. Variation in nuptial colour may become associated with variation in traits relevant to resource utilization either through ecological reinforcement where colour becomes a marker of ecological variation (Gavrilets, 2004; Reynolds & Fitzpatrick, 2007; Jiggins et al., 2008) or through habitat-dependent sexual selection (Boughman, 2002; Seehausen et al., 2008). Although colour divergence in the Pseudocrenilabrus incipient species is significant, it is generally more muted than in other cichlid radiations (e.g. compared with Lake Victoria) and exhibits variations on the same colour theme rather than major shifts in hue.
Our data suggests that mating preferences can diverge rapidly when associated with ecological divergence during adaptive radiation despite the absence of geographical isolation. At the same time our data suggests that allopatric populations, living in similar environments and maintaining ancestral phenotypes, may fail to evolve premating isolation even after very long periods of geographical isolation. Relaxed molecular clocks (Genner et al., 2007) applied to our sequence data suggest that the population of P. philander from Lake Bangweulu has been isolated from the populations in Lake Mweru for between 1 570 000 and 2 780 000 years (Stelkens et al. in prep.). Such genetically ancient but phenotypically conserved populations are likely to hybridize should they experience secondary contact.
One of the derived types (grey back) shows a significantly smaller genetic distance (t = 5.66, P < 0.01) to the sympatric philander Mweru than the other derived type (weed picker) to philander Mweru (Table 1b). Interestingly, the strength of assortative mating we found in the genetically less distant pair (grey back vs. philander Mweru) was almost two-fold stronger than in the genetically more distant pair (weed picker vs. philander Mweru; Table 1a). At the same time, larger phenotypic differences in both morphology and colouration were found between grey back and philander Mweru. This may provide additional support for the hypothesis that divergent ecological selection, and not divergence time, predicts assortative mating between the young species in the radiation.
Hybrids between all three Lake Mweru species were fully viable when raised in a laboratory environment (R. Stelkens, unpublished data) and there is no evidence for strong intrinsic post-zygotic isolation barriers among species in other cichlid radiations with comparably recent divergence times (Crapon de Caprona & Fritzsch, 1984; Seehausen et al., 1997; van der Sluijs et al., 2008). This makes it unlikely that reinforcement through endogenous fitness disadvantages contributed to the evolution of premating incompatibilities in the Pseudocrenilabrus of Lake Mweru. We have not yet tested for ecological reinforcement-like mechanisms in this system and its role in driving the evolution of mate choice, but other work suggests this could be important in cichlid fish speciation (Seehausen et al., 2008). A possible experiment to isolate effects of ecological by-product mechanisms vs. ecological reinforcement would involve mate choice trials between ecologically divergent populations that have no history of sympatry (Hollander et al., 2005; Vines & Schluter, 2006). If for instance, P. philander Bangweulu females mated at random when choosing between males of their own species and males of the phenotypically divergent species from Lake Mweru (e.g. grey back), we could conclude that ecologically based reinforcement was involved in speciation in this system. These experiments are currently in planning.
Pseudocrenilabrus is known as a genus with a panafrican distribution, yet with very low species numbers. Only three formally described species exist, and some ten geographically fully isolated subspecies of P. philander are thought to be the result of classical allopatric divergence between river basins (Katongo et al., 2005). This failure to undergo adaptive radiations despite presence in various large lakes has been puzzling (Seehausen, 2006), but had sometimes been explained by habitat conservatism with the riverine and lake margin habitats typical of Pseudocrenilabrus thought to not provide ecological opportunity for the evolution of divergent adaptations (Schluter, 2000); but see (Koblmueller et al., 2008). Our data here show that if given the ecological opportunities of a heterogeneous lake, Pseudocrenilabrus may radiate quickly into multiple species similar to some other groups of cichlid fish. We now need to explain why Pseudocrenilabrus radiated in Lake Mweru but not in any of the other large lakes in the geographical range of the genus.
In summary, we found that phenotypic differences, including eco-morphology and male nuptial colouration, but not genetic distance, promotes assortative mating in the adaptive radiation of sympatric Pseudocrenilabrus cichlids in Lake Mweru. Our data provide support for ecological speciation in the face of gene flow.
The authors thank Erwin Schäffer and Andreas Taverna for their help with fish maintenance and Domino Joyce, Cyprian Katongo, Lackson Shiridzinodya, Joseph Chishimba, and Happy Tembo for their help collecting fish. The authors also thank JP Danko and three anonymous reviewers for commenting on earlier versions of the manuscript. This research was supported by Swiss National Science Foundation grant to OS.