The role of founder effects on the evolution of reproductive isolation


  • D. R. Matute

    Corresponding author
    1. Department of Human Genetics, The University of Chicago, Chicago, IL, USA
    2. The Chicago Fellows Program, The University of Chicago, Chicago, IL, USA
    • Correspondence: Daniel R. Matute, Department of Human Genetics, The University of Chicago, 920 East 58th Street, Chicago, IL 60637, USA.

      Tel.: +1 773 834 8073; fax: +1 773 834 0505; e-mail:

    Search for more papers by this author


Several theories argue that large changes in allele frequencies through genetic drift after a small founding population becomes allopatrically isolated can lead to significant changes in reproductive isolation and thus trigger the origin of new species. For this reason, founder speciation has been proposed as a potent force in the generation of new species. Nonetheless, the relative importance of such ‘founder effects’ remains largely untested. In this report, I used experimental evolution to create one thousand replicates that underwent an extreme bottleneck and to study whether founder effects can lead to an increase in reproductive isolation in Drosophila yakuba. Even though the most common outcome of inbreeding is extinction, founder effects can lead to increased premating reproductive isolation in a very small proportion of cases. Changes in reproductive isolation after a founding population bottleneck are similar to changes in other phenotypic traits, in which inbreeding might displace the mean phenotypic value and substantially increase the phenotypic variance. This increase in phenotypic variance does not confer an increase in the response to selection for reproductive isolation in artificial selection experiments, indicating that the increased phenotypic variance is not caused by increases in additive genetic variance. These results also demonstrate that, similar to morphological and life-history traits, behavioural traits can be affected by inbreeding and genetic drift.


When colonizing new areas, species may be subjected to novel selection pressures and develop new adaptations that differentiate them from their ancestors (Gillespie & Roderick, 2002; Kawecki, 2008; Barker et al., 2009). The genetic composition of a colonizing population may also differ from that of the ancestral population simply due to genetic drift, if colonization occurs through a small number of founders in a population bottleneck. Events in which a small group of individuals colonize a new environment are known as founder events. The loss of genetic variation that occurs during the colonization process is known as a founder effect. In these cases, the amount and the nature of the remaining genetic variation of the founders might constrain the evolution of such populations (Hoffmann et al., 2003; Blows & Hoffmann, 2005; Kellermann et al., 2006; Heerwaarden et al., 2009; Andersson et al., 2009).

If the founding population has a reduced population size, it is also possible that the new population will suffer from the effects of inbreeding. In particular, high degrees of inbreeding can increase the amount of additive genetic variance and lead to decanalization, thereby overcoming genetic constraints (Robertson, 1952; Willis & Orr, 1993; Wang et al., 1998; Barton & Turelli, 2004; Hill et al., 2006; Willi et al., 2007). Inbreeding, for example, can play a role in uncovering and perpetuating combinations of alleles that would be selected against in large populations, thus fostering the evolution of traits by pure genetic drift (Wright, 1931, 1932; Lande, 1976). The relative frequency of changes in the genetic and phenotypic composition of a population due to founder effects, and the importance of these changes, remain a contentious issue on the nature of the evolutionary process (Carson, 1983, 1990; Barton & Mallet, 1996; Hollocher, 1996; Coyne & Orr, 2004). Even more controversial is the nature of founder speciation, the events by which founder effects can contribute to the origin of reproductive isolation and thus of new species (Gavrilets & Boake, 1998; Coyne & Orr, 2004; Templeton, 2008).

Founder speciation (i.e. speciation events driven by the effects of genetic drift on reproductive isolation) has been proposed to explain the origin of island and high altitude endemic species that are thought to originate from small parental populations (Mayr, 1959; Carson, 1968, 1975; Templeton, 1980, 1981, 1999a,b; Barton & Charlesworth, 1984; Carson & Templeton, 1984). If a small population colonizes a new habitat, for example an oceanic island, inbreeding might exacerbate the effects of genetic drift and play an important role in establishing which alleles get fixed (Templeton, 1999a,b). Most of the support for this idea comes from theoretical propositions. Several models state that fast changes in allele frequencies induced by genetic drift can lead to changes in reproductive isolation (Mayr, 1959; Carson, 1968, 1971, 1975; Templeton, 1999a,b; Uyeda et al., 2009; Butlin et al., 2012). All these propositions share the assumption that founder events lead to changes in highly epistatic genetic systems. These changes might modify how the selectable phenotypic variation is allocated across loci, and this in turn can lead to reproductive isolation between the parental population and the founder population (Mayr, 1959; Carson, 1968, 1971, 1975; Templeton, a,b; Nei et al., 1983; Nei, 1988; but see Lewontin, 1965; Barton & Charlesworth, 1984; Charlesworth, 1997). Even though founder effects lead to a loss of genetic variation, almost every model assumes that founder populations carry levels of genetic variation that are high enough to trigger changes in the nature of epistasis among loci (‘genetic revolutions’, Templeton, 1998; Coyne & Orr, 2004).

The idea that founder events could be crucial in the origin of new species was highly influential during the early development of speciation theory (Mayr, 1959; Carson, 1968, 1971, 1975). In its initial formulation, the founder speciation hypothesis argues that founder populations are likely to go extinct, but those that survive will be more prone to display a particular set of traits and genetic make-up, which might differ from the parental population. This evolutionary change might lead to reproductive isolation between the founder populations and the parental group (Mayr, 1959). Whereas founder speciation is a long-standing theory, few direct tests have been made to validate (or to rule out) its prevalence. Some comparative studies have tried to determine the importance of founder effects in the origin of new species using phylogenetic methods (Carson & Kaneshiro, 1976; Carson & Templeton, 1984; Janson, 1987; Simon, 1987; Gittenberger, 1996; Carson, 1990; Harrison, 1991; Eckert et al. 1996; Ueshima & Asami, 2003; Seddon & Tobias, 2007; Balakrishnan & Edwards, 2009). These studies have suggested that small populations can indeed give rise to new species but that the relative occurrence of founder speciation is rare (Chesser & Zink, 1994; Losos & Glor, 2003). Phylogenetic inference though does not disentangle whether inbreeding and founder effects were directly involved in the evolution of reproductive isolation.

Experimental approaches have also been used to address whether founder events can lead to the evolution of reproductive isolation. The possibility of founder speciation is tested in two ways: ‘accidental’ bottlenecks and controlled experiments. An accidental test was performed in Nereis acuminata, a marine polychaete. A single population of this organism underwent two serial bottlenecks following collection and maintenance in the laboratory (Weinberg et al., 1992). Even though the authors observed reproductive isolation between the founder population and the putative parental population, the results might be the result of population stratification that went initially unnoticed (i.e. the founder population might come from a population of N. acuminata that shows reproductive isolation to the putative parental population; Rodríguez-Trelles et al., 1996). The second line of evidence has been collected by direct manipulation of population size. In particular, several experimental approaches have studied the evolution of reproductive isolation in flush and crash experiments where large populations of Drosophila were subjected to extreme levels of inbreeding (e.g. cages with hundreds or thousands of individuals were reduced to a single mated female) and levels of reproductive isolation were measured after several of these crashes (Powell, 1978; Kilias et al., 1980; Ringo et al., 1986; Powell, 1989; Cobb et al., 1990; Galiana et al., 1993; Meffert, 1995; Regan et al., 2003; but see Moya et al., 1986). The rationale behind these experiments is that extreme cases of inbreeding might create ripe conditions for genetic drift and induce the evolution of reproductive isolation in founder populations. If inbreeding plays any role in the origin of new species, and the process occurs rapidly, then it should be possible to study the process in controlled experimental conditions (Carson, 1971; Powell, 1997). The main conclusions from these experiments were that stronger reproductive isolation could evolve after two steps: a strong founder event (crash and inbreeding) and a subsequent population expansion (flush). Inbred populations did not show increased reproductive isolation if they were not flushed, suggesting that the population expansion is a required step in the generation of reproductive isolation (Powell & Richmond, 1974; Powell & Morton, 1979). Second, the increased reproductive isolation detected between the parental population and the inbred/flushed populations was asymmetric, and it was almost always in the form of stronger female choice from the parental population discriminating against males from the inbred/flushed populations (Kaneshiro, 1989; Powell, 1989). All these studies, however, have a common caveat; the number of replicates in these experiments has always been low (< 50 replicates). This replication scheme was enough to determine whether founder events could lead to increase reproductive isolation, but conferred the experiments little power to determine the relative importance of founder events in the origin of reproductive isolation.

Theories of genetic drift propelling speciation became disfavoured, due to their complicated assumptions (especially in terms of the structure of required genetic variance in the founder population) and the lack of evidence for the process (reviewed in Coyne & Orr, 2004). Several authors have demonstrated that the models of founder speciation are unrealistic on two grounds. First, many models have shown that reproductive isolation is more likely to evolve through gradual evolution than through ‘genetic revolutions’ (Coyne & Orr, 2004; Butlin et al., 2012). Second, all the founder speciation models require genetic drift to act for a long time, rather than just in the founding generation, in order to generate new species (Barton & Charlesworth, 1984; Rice & Hostert, 1993; Gavrilets, 2003a,b; Coyne & Orr, 2004; Futuyma, 2006). On the other hand, others have argued that founder speciation might indeed be rare but that it still can have a disproportionate importance in the origin of adaptations and new species (Templeton, 2008), especially given that inbreeding can affect traits that lead to reproductive isolation (e.g. Aspi, 2000; Van Oosterhout et al., 2003; Reid et al., 2005; Mariette et al., 2006; Bolund et al., 2010).

Clearly, a tractable experimental system that allows for direct observation of founder effects is needed to properly assess any theory on the effects of founder effects on reproductive isolation. No study has provided direct evidence of the relative contribution of inbreeding (and subsequent genetic drift) to the evolution of reproductive isolation. I aimed to test how often founder effects lead to increases in reproductive isolation. Using Drosophila, I experimentally tested the importance of founder effects on the origin of reproductive isolation. In this study, I explored the consequences of inbreeding, one of the immediate consequences of founder effects, on the evolution of reproductive isolation. I took an experimental evolution approach to determine whether extreme population size reductions and the effects of inbreeding can lead to an increase in reproductive isolation towards the parental population. Using a diverse population of Drosophila yakuba, I generated a large number of parallel populations that underwent several generations of inbreeding and measured their reproductive isolation levels from the population that gave rise them. The inbreeding scheme resembles a biogeographical scenario in which a small founder population has colonized a new habitat and thus undergoes founder effects led by inbreeding and genetic drift. The majority of replicates went extinct, but a small percentage of them (~12%) survived after 30 generations of inbreeding. I measured premating isolation between these lines and the parental line that gave rise to them. These mating experiments mimic the situation in which the inbred lines and the progenitors come into secondary contact and have the chance to interbreed. I show that founder effects can lead to a transient increase in reproductive isolation in a very small proportion of cases but that such changes are unstable and wane over time.

Materials and methods


All stocks and populations were reared on standard cornmeal/Karo/agar medium at 24 °C under a 12 h light/dark cycle. I collected 250 D. yakuba isofemale lines in São Tomé (July 2009) along the north-eastern transect of the island (Llopart et al., 2005). These isofemale lines were used to generate the admixed/parental population by mixing ten virgin females and ten virgin males from each isofemale line. All stocks, except the treatments that went through experimental bottlenecks, were kept in large numbers after they were created and in a light cycle incubator (a 14 h/10 h cycle of 24 °C/22 °C) and constant relative humidity (60%).

Experimental populations and bottlenecks

I kept two kinds of experimental populations: a large population that represented a population with high levels of diversity and low levels of inbreeding (herein referred to as parental line), and inbred populations (described below). The parental population was created by combining ten females and ten males of each of 250 D. yakuba isofemale lines in a large population in an experimental cage (~5000 flies per generation). I let admixture proceed for ten generations. I kept three replicates of this population and always maintained the population size in high numbers (> 5000 flies).

To generate inbred populations, I collected one virgin female and one male from the parental population and housed the pair in an eight-dram cornmeal food vial. I repeated this procedure for 1000 replicates. Every generation, the hatching progeny from each vial was lightly anaesthetized with CO2 and one virgin female and one male were collected and transferred to a new vial. Once the pair started producing progeny (i.e. I observed larvae in the vial), the parents were discarded and the vial was tended by damping the media with propionic acid and adding tissue paper (Kimwipes; Kimtech Science, Roswell, GA, USA) to the media for larvae to crawl up and pupate. This regime was applied for thirty generations. If in a given generation a population did not produce enough progeny to set up the next generation because either no progeny were produced, or flies from only one sex hatched, the population was categorized as extinct.

PCR and DNA polymorphism analysis

I measured genetic diversity (as determined by nucleotide diversity, π) for the parental population and the experimental lines that survived inbreeding. To assess the genetic variation in different experimental treatments, I selected the seven most diverse loci reported in D. yakuba (Llopart et al., 2005) and sequenced ten individuals per treatment. I extracted DNA from single female flies using the Qiagen DNeasy Blood and Tissue Kit and performed PCR amplifications using 25 ng of genomic DNA (Qiagen, Inc., Valencia, CA, USA). PCR products were sequenced directly with an ABI PRISM 3100 genetic analyzer (Applied Biosystems, Foster City, CA, USA) after I performed cycle sequencing reactions using Big Dye 3.0 (Applied Biosystems). Both strands were sequenced. Sequences were edited with the software 4Peaks 1.7.1 ( and MacClade (Madisson & Madisson, 2007) and aligned using the ClustalX program (Larkin et al., 2007). All the individual alleles resulting from this study were deposited in GenBank under the accession numbers: JX944234JX944303. Coding sequences were assigned by visual inspection of each gene region. To calculate nucleotidic diversity (π), I used only synonymous polymorphism and the Nei–Gojobori algorithm with MEGA version 5 (Tamura et al., 2011).

Reproductive isolation

All measurements of reproductive isolation were taken between the focal population (each of the assayed inbred populations) and the parental population during the first 6 h of the light cycle (which is when flies are most active). For each population (either from the parental lines or from the surviving inbred lines), I collected 50 virgin flies of each sex over the time span of 5–6 days. Once I had collected 52 flies of each sex, I housed 50 of them in sex-specific vials with coloured food for 3 days, so their abdomens were marked with one of three different colours (red, blue, green; Ting et al., 2001). This meant that flies could be anywhere from 4 to 10 days old. In parallel, I collected the same number of flies each day from the parental population, so I would have the same age structure as in the focal populations. On day 4, I aspirated 50 females and 50 males from the inbred line to be tested, and 50 females and 50 males from the parental population (which had been fed a different coloured food) into a fresh food-containing vial. (The two extra flies per vial were collected to mitigate the impact of escapees and death; if no flies escaped or died, the extra flies were removed from the vial with an aspirator.) Each trial was run for 60 min. At the end of that time, the whole vial was lightly gassed with CO2 and the number of copulations of each type was recorded. (If the vial is lightly gassed, copulating pairs are not separated.) Even though it has been observed that dying the abdomens of flies has no effect on mating choice on Drosophila melanogaster (Ting et al., 2001), I assessed the possible effect of different dyes in mating choice by measuring mating choice in a similar experimental setting between D. yakuba individuals from the parental population that had been labelled with different colours. I found that the colour of the dye had a marginally significant effect on mating choice in D. yakuba (female × male interaction term F9,144 = 1.883, P = 0.059; Table S1). To minimize any effect on mate choice because of abdominal female and/or male colour, I used different and random dyes in each trial.

To quantify the magnitude of premating isolation (i.e. assortative mating), I used the IPSI statistic (Rolán-Alvarez & Caballero, 2000), which ranges from −1 (complete negative assortative mating) to 1 (complete positive assortative mating). This index incorporates the four possible estimators of pair premating isolation (PSI) for each mating combination (the two homotypic combinations aa and bb, and the two heterotypic combinations ab and ba). IPSI is calculated as:

display math

where PSIij are estimators of paired premating isolation for each mating combination between types of mates i and j. PSIab for example is calculated as:

display math

where ab is the number of matings between ♀a and ♂b, ba is the number of matings between ♀b and ♂a, aa is the number of matings between ♀a and ♂a, bb is the number of matings between ♀b and ♂b, and t is the sum of the total number of matings. (The other PSI indexes are calculated similarly).

I compared the IPSI values from each inbred line–parental population combination to the distribution of parental–parental IPSI values with a two-sided t-test. (The parental–parental IPSI values represent the initial distribution of within-population reproductive isolation values with no inbreeding.) The comparison between each inbred replicate and the parental population aimed to establish whether each replicate subjected to inbreeding had increased (or decreased) its levels of IPSI when compared with the closest extant population to the parental population that gave rise to each replicate. Since the variance of the parental line and that of each inbred line were not equal, I adjusted the number of degrees of freedom of each t-test with a Welch comparison (Welch, 1947) and used a Sidak's correction for multiple comparisons (Sidak, 1967).

I also calculated an alternative index, Is (Coyne & Orr, 1989), to assess whether founder effects lead to asymmetric levels of reproductive isolation between the founder populations and the parental population. Is is calculated for each direction of the cross (i.e. there were two Is indexes for each hybridization). Isab, the index of premating isolation in ♀a × ♂b crosses, was calculated following the equation:

display math

and the index of premating isolation between ♀b × ♂a was calculated following the equation:

display math

I assessed whether there was asymmetry in reproductive isolation between the two directions of the cross by comparing Isab and Isba for the three inbred lines that showed strengthened premating isolation with a Wilcoxon signed rank test with continuity correction (Wilcoxon, 1945; Sokal and Rohlf, 1994).

Artificial selection

To assess whether increased phenotypic variance conferred the inbred populations more evolutionary potential to evolve increased reproductive isolation (i.e. a faster or more marked response to selection), I did artificial selection for mating choice for a subset (n = 15 lines) of the inbred populations (those that showed the highest phenotypic variance in their levels of premating isolation from the parental population) and 15 replicates from parental population (control). I selected the inbred fifteen lines that showed the highest variance on IPSI from the parental population on generation 30. On generation 31, I collected 100 virgin females and 100 virgin males from both inbred and parental populations (same collection scheme as described above but instead of over 5 days, virgins were collected over 10 days) and set up mating experiments (i.e. the 400 virgin individuals were mixed in the same vial for 30 min and were then lightly anaesthetized with CO2). I selected the flies from the inbred population that did not mate with individuals from the parental population and instead engaged in homotypic matings (i.e. copulations of inbred females with inbred males). These flies were used to start the next generation. I followed this collection scheme for 20 generations. The average number of females and males that did not mate with the parental population and thus were used to start the next generation are shown in Table S2.

I followed a similar selection scheme in fifteen populations that were started from the parental population and thus had not been through extreme inbreeding. I started each of these subpopulations with 500 flies each. I followed a virgin collection scheme similar to the one carried out for the inbred lines to have a virgin population with the same age structure (i.e. flies were between 4 and 14 days old upon mating). During twenty generations, I set up matings between these populations and the original parental population, and selected against individuals that mated with the original parental population and not with their own type (i.e. I started the next generation only with individuals that engage into homotypic matings). After generation 20, I measured premating isolation between each of these fifteen replicates and the original parental population (which was always kept with at least 5000 individuals). To assess whether the selection scheme had been effective, I used a nested anova (replicate nested within generation) to compare the level of premating isolation at generation 31 (before selection) and at generation 51 (after selection) for the inbred lines and the noninbred populations (i.e. I fitted two independent anovas).


By combining 250 isofemales lines, I created an admixed population (the ‘parental’ population), which showed high levels of mean heterozygosity (π = 2.561%, SD: 0.782%, Table 1). All the isofemale lines that were used in this experiment have been housed in laboratory conditions for the same period of time and have been submitted to identical laboratory conditions (all isofemale lines were collected in July 2009).

Table 1. The experimental bottleneck causes a severe and long-lasting reduction in genetic variability. Heterozygosity levels (calculated as π and expressed as a percentage) at seven loci before inbreeding, after 30 generations of inbreeding and after inbred lines were allowed to increase their population size (flush) for 30 more generations. Numbers in parenthesis are the standard error of the mean.
Geneπ admixedπ after inbreeding, % (SEM)π: generation 60, % (SEM)
rux 3.977 (1.148)0.595 (0.315)0.513 (0.241)
sog 7.640 (1.776)0.511 (0.344)0.633 (0.406)
yellow 1.220 (0.610)0.076 (0.041)0.116 (0.076)
dhc 0.889 (0.612)0.040 (0.040)0.056 (0.056)
CG17629 0.467 (0.446)0 (0)0 (0)
npg 1.178 (0.695)0.216 (0.113)0.129 (0.088)
salr1 0.467 (0.457)0 (0)0 (0)
SEM across loci0.7820.1040.108

The inbreeding scheme mimics conditions in which a few individuals went through a bottleneck, started a new population, and have been confined to an allopatric location with small population sizes. I started 1000 vials (replicates) representing extreme founder events: each replicate was established with one female and one male from the parental population. In subsequent generations, each replicate was maintained using full-sib matings (one female and one male). I aimed to address two different but intertwined questions. First, how frequently (if ever) does inbreeding generate reproductive isolation? A second related question was to assess whether premating isolation was similar to other phenotypic traits and showed an increase in among-population phenotypic variance under genetic drift.

Mean premating isolation

Inbreeding causes increased reproductive isolation in 0.3% of the treatments

Of the 1000 replicates that I subjected to inbreeding, most cases (~88% of the replicates) went extinct before generation 30. The vast majority of replicates (~80%) went extinct within the first five generations of inbreeding (Fig. S1).

For the replicates that survived the thirty generations of inbreeding (n = 123), I used a behavioural assay to measure premating isolation to determine whether the bottleneck induced changes in reproductive isolation from the parental population. The assay simulated instances in which each bottlenecked population enters in secondary contact with the parental population (see 'Materials and methods' for a full description).

The levels of premating isolation of the surviving lines (n = 123) from the parental population are shown in Fig. 1 and Table S3. Inbreeding led to an increase in premating isolation from the parental population for most of the lines (~100 lines showed an IPSI > 0, whereas ~20 showed an IPSI < 0). A linear mixed model showed that inbreeding had a significant effect on levels of reproductive isolation, and it induced a significant level of heterogeneity in the inbred replicates when measured to the parental population (F122,246 = 2.463 P = 1.172 × 10−9). The average inbred line–parental population IPSI was significantly higher than the within parental population IPSI (One-sided Welch two sample t-test pooling all inbred replicates: t130.61 = 10.71, P-value < 1 × 10−10). From these lines, three inbred replicates evolved significantly increased reproductive isolation from the parental population (in all cases, Welch two sample t-test with Sidak's correction, P < 2 × 10−4, Table S4). This represents an increase in premating isolation in ~3% of the surviving lines but only in 0.3% of the total of started replicates. (The average proportion of heterotypic copulations in these three lines is ~19% and ~43% in the other 120 lines.) These three lines also showed reduced phenotypic variance within lines, suggesting fixation of some of alleles that increase sexual isolation and depletion of genetic variance within these lines (Table S3). No line showed a significant decrease in their premating isolation to the parental line.

Figure 1.

Phenotypic distribution of premating isolation between the inbred lines and the parental population. The histogram shows the distribution of IPSI mean values of each inbred line (n = 123, three replicates per line = 369 assays). Inbred lines show a strengthened average level of premating isolation towards the parental population. The average level of reproductive isolation across lines is shown with a black line and the standard error around the mean with dotted black lines. The distribution of assortative mating levels in the parental line is also shown (parental mean: red solid bar; standard error of the parental mean: red dotted lines).

Is the fact that three lines show increased premating isolation expected by chance when sampling 123 replicates from the parental population? I compared the distribution of premating isolation within the parental population (i.e. mating experiments to characterize the amount of assortative mating within the parental population; n = 750 replicates) with the premating isolation levels of the three lines that showed significantly higher reproductive isolation (three replicates per line) and further established that these nine replicates were not likely to be drawn from the parental distribution (Two-sample two-sided Kolmogorov–Smirnov test, D = 0.972, P-value = 1.01 × 10−7). Additionally, none of the 750 replicates from the parental population showed a level of premating isolation that was equal or higher than the average level of premating isolation observed in the three inbred lines that showed increased premating isolation. These results strongly suggest that the observation of increased premating isolation is not the product of random sampling from the parental population.

I assessed whether the premating isolation was asymmetric in the heterotypic pairs by comparing the premating isolation of inbred females from males of the parental population (IsINBRED–PARENTAL) with the premating isolation of parental females from males of the inbred population (IsIPARENTAL–INBRED). None of the replicates showed a significant difference in the magnitude of premating isolation between the reciprocal crosses (Wilcoxon rank sum test, W < 0, P-value > 0.1 in all three cases). These tests, though, have a very low power to detect asymmetries (n = 3 replicates per line) and might not be able to detect subtle differences in the magnitude of premating isolation between reciprocal crosses.

Reproductive isolation caused by inbreeding wanes over time

Previous theoretical simulations have suggested that premating isolation generated by founder effects is likely to have a frail evolutionary history, in which isolation waxes and wanes. I tested this hypothesis and measured reproductive isolation (to the parental population) after relaxing population size constraints. I allowed the three replicates that showed an increase in reproductive isolation after inbreeding to expand their population size by maintaining them in cages with census sizes > 1000 individuals for 30 generations. This flush stage has been proposed to be crucial in order to generate reproductive isolation by founder effects (reviewed in Powell, 1989). The levels of premating reproductive isolation after the populations were allowed to expand were significantly lower than directly after the inbreeding treatments (Table S2, Fig. 2). In the flushed populations, premating isolation approached but did not diminish to the point of the initial levels of reproductive isolation in the parental population before inbreeding (Welch two sample t-test comparing IPSI after inbreeding and before flushing –generation 30– and after inbreeding and flushing –generation 60–; all three P-values < 0.004). These results suggest that the flush stage required in many speciation-by-drift models (Mayr, 1959; Carson, 1968, Templeton, a,b) is not required to generate reproductive isolation and that in rare instances, inbreeding might be enough to generate reproductive isolation.

Figure 2.

Increased premating isolation induced by founder effects wanes over time. Mean levels of reproductive isolation before inbreeding (generation 0), after inbreeding (generation 30) and after population size expansion (generations 45 and 60) in the three lines with the highest mean reproductive isolation. Error bars are the standard error of the mean within lines (three replicates per line at generation 30 and ten replicates per line and generations 45 and 60, see text for details).

Phenotypic variance

Populations with high levels of additive genetic variance in a trait usually (but not always) respond to selection in a more pronounced manner than populations with low levels of variance (Frankham et al., 1968; Hill & Caballero, 1992; Mackay & Lyman, 2005; but see Walsh & Blows, 2009). The lines that underwent a bottleneck showed a higher phenotypic variance in reproductive isolation was significantly higher among inbred lines than among replicates of the parental line (F122,749 = 4.805, P-value < 2.2 × 10−16). These results suggest that inbreeding can lead to increased premating isolation but also lead to an increase in phenotypic variance among inbred lines.

Does the increased phenotypic variance in premating isolation observed in the inbred populations confer a higher evolutionary potential to develop stronger reproductive isolation? I selected the fifteen lines with the highest variance in premating isolation and measured their response to selection for increasing premating isolation towards the parental population. (None of these fifteen lines showed a significant change in the mean value of reproductive isolation to the parental line, but they did show a substantial increase in the variance of the trait.) During twenty generations, I selected against the flies that mated with the parental line (Table S2). I kept the flies that readily mated with their own type (and did not mate with the parental population) and started a new generation with them. After 20 generations of this selection scheme, I measured premating isolation of each of these fifteen lines (three replicates per line) towards the parental population.

I followed a similar selection scheme in fifteen populations that were started from the parental population (and thus had not been inbred). During twenty generations, I set up matings between these populations and the parental population and selected against individuals that mated with the parental population and not with their own type. I measured premating isolation towards the original parental population (which was always kept with 5000 individuals) after 20 generations. This selection procedure is akin to a selection-breeding scheme in which truncating selection eliminates the individuals more prone to mate with the ancestral population. The expectation of these approaches is that lines with higher phenotypic variance (i.e. the inbred lines) might show increased potential to evolve enhanced premating isolation when there is strong selection for the trait. I compared the IPSI values obtained in generation 31 and the ones obtained in generation 51 for each of the two sets of selected lines with a nested anova (replicate nested within generation). The results for the two kinds of selection experiments are shown in Fig. 3 and suggest two patterns. First, populations that have not been inbred (controls) evolved slightly stronger premating isolation after 20 generations (nested anova with replicate nested within generation F1,284 = 655.316, P < 1 × 10−10) with very little variation among replicates (F1,284 = 0.688, P = 0.786). These results suggest that the selection scheme can induce changes in premating isolation and that premating isolation shows additive genetic variance in the parental population for the trait to respond to selection. Second, the inbred lines showed no change in their levels of premating isolation before and after being subjected to selection (nested anova with replicate nested within generation, F1,284 = 2.536, P = 0.112), suggesting that the increase in phenotypic variance did not lead to more potential but that inbreeding actually hampered the evolution of increased reproductive isolation. Since this test was only done with inbred lines that showed an increased phenotypic variance and not the lines that showed an increased average in their levels of premating isolation, this test reveals that increased phenotypic variance in levels of premating isolation does not lead to more potential for evolution of the trait. This type of selection experiment was not conducted for the three lines that showed increases in premating isolation levels. For that reason, they do not constitute a test of the feasibility of natural selection as a mechanism to stabilize increased premating isolation once reproductive isolation has originated by inbreeding.

Figure 3.

Increased phenotypic variance does not confer increased potential to evolve premating isolation. Mean levels of reproductive isolation after inbreeding (generation 0–30) and artificial selection (generation 31–51, black bar) on the fifteen lines with the highest phenotypic variance after inbreeding (grey bars) and in replicates of the parental line (white bars). Values at generation 0 are equivalent as the two sets of populations were started from the same parental population. Fifteen lines were measured at generation 31 and at generation 51 for both kinds of treatments (three replicates per line). The error bars represent the standard error of the mean of the 15 lines per treatment. (At generation 0, the standard error was calculated with 750 replicates.) Even though the artificial selection scheme induces increased premating isolation (evidenced by the response in the parental replicates), selection does not induce increased premating isolation in the inbred replicates.

The founder effect causes a dearth of genetic variability

To test whether the increase in phenotypic variance observed above was caused by decanalization (i.e. the uncovering of phenotypically cryptic genetic variation, Flatt, 2005) and not by an increase in genetic variation (which is actually likely to decreased due to inbreeding), I measured the change of heterozygosity after thirty generations in 12 inbred populations (~10% of the total surviving populations) using seven coding and unlinked markers. The levels of heterozygosity decreased in all the lines after inbreeding, suggesting a strong founder effect on genetic diversity across different regions of the genome (Table 1).

I also measured the heterozygosity levels after the 30 generations that the lines were allowed to expand their population sizes and found that the genetic diversity did not increase significantly over this period of time (Welch two sample t-test, P-value > 0.55 for all loci).


Are inbreeding and genetic drift able to generate reproductive isolation? Although it has been shown that inbreeding can lead to significant changes in the phenotypic average and an increase in the additive component of the total phenotypic variance for a variety of morphological, physiological, sexual and life-history traits (Bryant et al., 1986; Lopez-Fanjul & Villaverde, 1989; Garcia et al., 1994; Fernandez et al., 1995; Wade et al., 1996; Pray & Goodnight, 1997; Fowler & Whitlock, 1999; Whitlock & Fowler, 1999; Phillips et al., 2001), it has remained unclear whether traits that lead to reproductive isolation also follow this pattern. Thus far, all experimental approaches aimed to understand whether founder events are a possible driver of speciation have used small numbers of replicates. This in turn has constrained the possibility of establishing the prevalence of founder effects on the evolution of reproductive isolation. Here, I demonstrate that inbreeding can lead to a transient increase in total levels of reproductive isolation. I also show that the relative frequency of these events is very low. The results here described have several implications for the study of speciation. First, they show that strict inbreeding can lead to an increase in the mean value of reproductive isolation in rare occasions: populations that have experienced a strong bottleneck and suffer from subsequent inbreeding might become reproductively isolated from their parental population. Second, they show that, as predicted by theoretical models (Uyeda et al., 2009), the time required for inducing reproductive isolation is relatively short if inbreeding is strong (in the case reported here, 30 generations of strong inbreeding). Finally, they suggest that the increased premating isolation induced by founder effects wanes after 30 generations of population expansion, despite the persistence of founder effects on genetic variation. Increased premating isolation might require other factors (natural selection, genetic assimilation –environmentally induced phenotypes that may become genetically fixed, West-Eberhard, 2003– or both) to stabilize the trait.

Two loci (of seven) showed a complete depletion of genetic variance (π = 0.0%), which is close to expected after a strong inbreeding treatment (expected heterozygosity for neutral loci after 30 generations of brother–sister matings is in the order of 10−10; Crow & Kimura, 1970; Gillespie, 2004). The other five loci showed a reduction in heterozygosity but did not become completely monomorphic. This suggests that these genomic regions might be not amenable to homozygosity, possibly due to the existence of deleterious alleles linked to these loci. The signature of this founder effect, reduced heterozygosity, prevailed even after the lines were allowed to expand their population size and were allowed to have a large population for fifty generations. These results indicate that the inbreeding treatments I imposed in these experimental populations led to founder effects that could be detected as a lost of genetic variation and persisted over 30 generations of population expansion. The results here shown also suggest that premating isolation induced by inbreeding has an unstable evolutionary history that might wane once the population expands. Even though founder effects can be a suitable, but rare, mechanism to generate increased reproductive isolation, natural selection might be required to stabilize any increased reproductive isolation generated by inbreeding.

The fact that reproductive isolation can arise after founder effects also has implications for our understanding of speciation driven by colonization of new habitats. In these situations, there might be a strong selective pressure to adapt to a new environment, and this can lead to reproductive isolation (i.e. reproductive isolation is mediated by adaptive traits to the new environment). The results here shown suggest that founder effects and small population sizes also induce reproductive isolation without the need of invoking adaptation. These two possibilities are not mutually exclusive, and they might even operate at the same time (i.e. natural selection might help to stabilize the frail premating isolation originated by founder effects).

Just like morphological and life-history traits, behavioural traits can be affected by inbreeding and genetic drift. The changes in reproductive isolation here reported are similar to changes observed in other phenotypic traits after inbreeding: a displacement of the mean phenotypic value and a substantial increase in the between-population phenotypic variance (Bryant et al., 1986; Whitlock & Fowler, 1999; Phillips et al., 2001). The precise mechanisms of these changes remain unknown but there are several possibilities. Notably, founder events might lead to the generation of slightly deleterious combination of recessive alleles. Such combinations might not occur and prevail in large populations because of the action of natural selection. The role of founder effects in speciation therefore might be the uncovering of standing variation from the parental population. This hidden variance might only be revealed when populations undergo an extreme reduction in size and alleles get fixed by chance. A second possibility is that founder effects lead to an increase in the phenotypic variance in premating isolation though phenotypic plasticity. In this case, the increases in premating isolation are not caused by genetic differences, but arise as a plastic response to founder effects.

Both possibilities could also account for the reversal in reproductive isolation observed after 30 generations of flushing: in the former scenario, if there is no complete fixation of the all the required recessive alleles, the allelic combinations that lead to increased premating isolation might break up by recombination and lead to the initial levels of premating isolation (i.e. pre-inbreeding). In the latter scenario, if increase in phenotypic variance is caused by phenotypic plasticity, the disappearance of the enhanced premating isolation induced by the founder effect can be due to lack of genetic assimilation (which is required to stabilize the trait). Even though phenotypic plasticity might explain why increased premating isolation wanes after population expansion, it posses another conundrum: why would enhanced premating isolation be a plastic response to small population sizes? I currently cannot discern between these two hypotheses, and both remain likely explanations for the observed changes in reproductive isolation.

The study of founder effects as a cause of reproductive isolation in laboratory conditions is not new and has been criticized multiple times on several grounds (Rundle et al., 1998; Rundle, 2003; reviewed in Rice & Hostert, 1993 and Coyne & Orr, 2004). First, the biological assumptions from crash experiments (including the one reported here) might not be completely realistic. I, for example, crossed one male and one female in a strong inbreeding scheme (continuous brother–sister mating). These conditions are extreme and likely do not reflect the true nature of a founder effect. An interesting question that stems from this caveat is what is the nature of reproductive isolation when the founder effect is not as pronounced as the one described in this report (i.e. instead of using one brother–sister pair to start the new generation, using of more than one pair of flies). Galiana et al. (1993) found that populations with fewer founding individuals tend to be more reproductively isolated from the parentals than populations with more founders. Second, founder-crash experiments usually involve newly collected lines from different geographic localities. This collection scheme might be prone to the detection of adaptation to laboratory conditions and reproductive isolation as a by-product (Rundle et al., 1998; Rundle, 2003). All the lines used in this experiment were collected from a single island (São Tomé), and previous studies have demonstrated that in average, the genomewide population structure is close to zero (Llopart et al., 2005). Additionally, all the isofemale lines (used to generate the parental population) and experimental inbred replicates were kept in identical environmental conditions, so it is unlikely that the levels of reproductive isolation observed after population bottlenecks are the results of further adaptation to different environments.

Crash and flush experiments have also been criticized because after reproductive isolation is attained in crash experiments, its nature is temporarily unstable (Coyne & Orr, 2004; Chapter 11 and references therein). I observed that the stronger reproductive isolation attained after inbreeding is frail in its nature and wanes over time. These results suggest that although founder effects might lead to reproductive isolation, other forces such as sexual selection, natural selection in the form of adaptation to a new environment, reinforcement or genetic assimilation might be responsible for its perpetuation (Lande, 1981; West-Eberhard, 2003; Kolbe et al., 2012). In this report, I demonstrate that increased phenotypic variance does not lead to a larger evolutionary potential of evolving stronger reproductive isolation. I did not address, however, whether selection could lead to the stabilization of increased premating isolation induced by founder effects.

In this study, I report first that after a population undergoes a strong bottleneck, there is a high likelihood of that population going extinct. This result is not novel in the sense that many studies that attempt to create inbred lines show a low rate of success due to a decrease in the average fitness of the population, probably by homozygosity of lethal alleles. It is useful, though, to think about these results from an evolutionary perspective and think about the likelihood of extinction of a founder population once it reaches a new habitat. I found that this likelihood is close to 90% when the parental population is a Drosophila population with high levels of diversity. The second result is that that founder effects rarely increased reproductive isolation in the surviving lines (< 3% of cases). This proportion is even lower if we take into consideration that the initial trial had 1000 replications, which means that reproductive isolation increased in only 0.03% of the lines that were started. In this study, my aim was to determine how frequently founder effects lead to an increase in premating isolation. I suggest that founder effects might rarely generate increased reproductive isolation but that rarity is not synonymous with unimportance (Templeton, 2008).


I would like to thank J. Ayroles, J.A. Coyne, J. Gavin-Smith, K.L.M. Gordon, L. Segurel, M.F. Przeworski and W. Meyer for scientific discussions and comments at every stage of the manuscript. Finally, I would like to thank the Bioko Biodiversity Protection Program, and the Ministry of Environment, Republic of São Tomé and Príncipe for permission to collect and export specimens for study. D.R.M. is funded by a Chicago Fellowship.