Temporal change in inbreeding depression in life-history traits in captive populations of guppy (Poecilia reticulata): evidence for purging?


Line Kristin Larsen, Department of Biology, Centre for Conservation Biology, Norwegian University of Science and Technology (NTNU), N – 7491, Trondheim, Norway. Tel.: +47 735 96090; fax: +47 735 96100; e-mail: linekris@bio.ntnu.no


Inbreeding depression, which generally affects the fitness of small populations, may be diminished by purging recessive deleterious alleles when inbreeding persists over several generations. Evidence of purging remains rare, especially because of the difficulties of separating the effects of various factors affecting fitness in small populations. We compared the expression of life-history traits in inbred populations of guppy (Poecilia reticulata) with contemporary control populations over 10 generations in captivity. We estimated inbreeding depression as the difference between the two types of populations at each generation. After 10 generations, the inbreeding coefficient reached a maximum value of 0.56 and 0.16 in the inbred and control populations, respectively. Analysing changes in the life-history traits across generations showed that inbreeding depression in clutch size and offspring survival increased during the first four to six generations in the populations from the inbred treatment and subsequently decreased as expected if purging occurred. Inbreeding depression in two other traits was weaker but showed similar changes across generations. The loss of six populations in the inbred treatment indicates that removal of deleterious alleles also occurred by extinction of populations that presumably harboured high genetic load.


An increasing number of populations are presently turning into “island” populations as a consequence of habitat fragmentation. The reduction of population size generally provokes a loss of genetic variability as a result of genetic drift and inbreeding (i.e. mating among relatives). Small populations consequently face several challenges. Mildly deleterious alleles can be fixed by genetic drift and the reduced genetic variation hampers the population’s ability to respond to selective challenges induced by environmental changes (Allendorf & Luikart, 2007). Additionally, these populations are more vulnerable to stochastic and demographic factors. Finally, the increase in homozygosity resulting from inbreeding tends to reduce individual and population fitness, a phenomenon known as inbreeding depression. Together, these consequences may substantially increase the risk of extinction in small populations (Mills & Smouse, 1994; Saccheri et al., 1998; Keller & Waller, 2002).

The decline in fitness caused by inbreeding has focused the attention of geneticists and evolutionary biologists ever since Darwin’s early experiments on selfing in plants (Darwin, 1876). Since then, inbreeding depression has been documented for a wide range of organisms, including mammals (Lynch, 1977; Lacy & Ballou, 1998; Hedrick & Kalinowski, 2000; Meagher et al., 2000), birds (Keller, 1998; Marr et al., 2006), fish (Shikano et al., 2001a; Van Oosterhout et al., 2007; Leberg & Firmin, 2008), insects (Saccheri et al., 1998; Frankham et al., 2001; Fox et al., 2006; Swindell & Bouzat, 2006) and plants (Ellstrand & Elam, 1993; Byers & Waller, 1999; Willis, 1999). Two mechanisms can explain inbreeding depression, namely the heterozygote advantage (overdominance) hypothesis and the dominance hypothesis. The overdominance hypothesis posits that the decrease in individual fitness with inbreeding results from alleles that provide superior fitness in the heterozygote state, possibly because of an increase in the physiological range of active enzymes. Under this hypothesis any increase in homozygosity should decrease individual vigour. The dominance hypothesis, on the other hand, posits that the decrease in vigour in inbred individuals results from the unmasking of recessive deleterious alleles, and is widely accepted as the most common mechanism causing inbreeding depression (Charlesworth & Charlesworth, 1999; Roff, 2002; Charlesworth & Willis, 2009).

Decreased fitness following inbreeding, however, may not necessarily be observed, and populations with a history of inbreeding may express unexpectedly high fitness relative to outbred populations, as Darwin (1876) noted. This observation is now explained by the purging of the genetic load, that is, the removal of deleterious recessive alleles exposed to natural selection in inbred populations. Depending on their genetic architecture, purging may restore the normal expression of traits that previously experienced inbreeding depression. Purging can only affect traits where inbreeding depression results from recessive deleterious alleles and not overdominance, because overdominance will generate selection against all homozygous genotypes. Lethal or strongly deleterious alleles under strong selection should be purged at a faster rate than mildly deleterious alleles (Lande & Schemske, 1985; Hedrick, 1994; Wang et al., 1999). Accordingly, selection against deleterious alleles has been found to be strongest for traits closely related to fitness (Derose & Roff, 1999; Keller & Waller, 2002), and we therefore expect purging to be more effective for these traits. As a direct consequence of purging, we also expect that, for similar levels of inbreeding, populations with long history of inbreeding are likely to be less affected by inbreeding depression because they have had more opportunity to purge deleterious alleles, compared to populations with recent inbreeding history (Ehiobu et al., 1989; Day et al., 2003).

Despite the acknowledged importance that purging can have on inbreeding depression, the occurrence and efficiency of purging remain actively debated (Ballou, 1997; Byers & Waller, 1999; Crnokrak & Barrett, 2002; Boakes et al., 2007). One recurrent problem in the study of purging is that a large majority of experimental studies on inbreeding depression have focused on the effects of inbreeding after just a few generations of full sib-mating. Although full-sib mating provides insight in the consequences of a rapid increase in inbreeding, it represents an unlikely situation in natural populations and even in most captive populations. Few studies, however, have experimentally investigated the effect of a slow accumulation of inbreeding (but see Ehiobu et al., 1989; Latter et al., 1995; Day et al., 2003; Reed et al., 2003). Contrary to full-sib mating or selfing, a slow accumulation of inbreeding over several generations should promote purging, and provide a more realistic view of the possible combined effects of inbreeding and purging on individuals and population fitness.

A variety of methods have been used to analyse purging, but many suffer from confounding effects such as genetic drift, epitasis or adaptation to laboratory conditions (Willis, 1999; Crnokrak & Barrett, 2002). The comparison of the expression of traits in populations exposed to increased inbreeding with control populations (i.e. contemporary outbred populations where the level of inbreeding is maintained at the lowest possible level) is regarded as the most powerful method to detect purging (Crnokrak & Barrett, 2002). In this study, we compared the expression of four life-history traits in captive populations of guppy (Poecilia reticulata) where inbreeding accumulated over 10 generations with control populations where inbreeding was maintained at low levels. Evidence of inbreeding depression in life-history and morphological traits in guppies has been reported (Fujio & Nakajima, 1992; Sheridan & Pomiankowski, 1997; Shikano et al., 2000, 2001b; Nakadate et al., 2003; Van Oosterhout et al., 2003, 2007; Pitcher et al., 2008), making this species particularly suitable for studying the combined effects of inbreeding and purging.

By strict control of mating, we were able to estimate the inbreeding coefficient (F) for each individual at each generation and analysed changes in clutch size, offspring survival, time to produce the first clutch, and sterility, in relation to the increasing level of inbreeding. We predicted that, if purging occurred, an initial deterioration in the expression of the different traits would be observed in the inbred populations, followed by a rebound in trait value.

Materials and methods

Founding population and general holding conditions

The guppy is a tropical freshwater fish native to Trinidad and Tobago and the northern part of South America. They are live bearing with a relatively short generation time and considerable reproductive potential. The fish used in this study were descendants from 500 individuals captured in 1998 in the Quare River, Trinidad (10°39′N, 61°12′W). They were collected from a high predation site, where the main threat was the pike cichlid (Crenicichla alta; Reznick & Endler, 1982; Houde, 1997). Once the collected fish were safely located in laboratory conditions in Norway, 106 female guppies sampled from the river were placed individually in separate 6 L aquaria to give birth. Offspring were used as founders of the experimental populations, and all females contributed to this first experimental generation. All populations were reared in constant conditions for the duration of the experiment, with a 12 : 12 h light: dark cycle and temperature of 24 °C (± 2 °C). Fish were fed every day, alternating dried flakes and live newly hatched brine shrimp every other day. Each generation took ca. 12 months to complete. This experiment started in June 1998 and finished in March 2008.

Breeding regimes

We compared changes in life-history traits across 10 generations between two different breeding regimes. The study was initiated with 20 populations, 10 replicated populations in each regime. These two regimes were a Design mating treatment, where inbreeding was maintained at the lowest possible level by mating the least related individuals, and a Random mating treatment, where random assignment of breeding pairs resulted in an increased rate of inbreeding over generations. Populations from the design mating treatment functioned as controls. They are referred to as such in the following, and the populations from the random mating treatment are referred to as inbred populations. Populations from the two regimes were synchronized, that is, each new generation was started simultaneously for the two breeding regimes.

In each control population, 10 mating pairs, one female and one male, were formed at the start of each generation by pairing individuals who were least related based on their pedigree. Each pair was placed in an isolated compartment (two 38 L breeding aquaria, each divided into five equal parts) for 21 days. After this period, the males were removed and the females left alone to give birth. Each female stayed in her compartment until she had given birth to a minimum of five offspring. Breeding aquaria were checked daily for newborns. Just before maturation, male offspring, recognizable by the development of their gonopodium (Houde, 1997), were removed and kept separate from female offspring. To produce the breeding pairs of the next generation, we used one male and one female offspring from each family, thereby equalizing family size.

In the inbred populations, each new generation started with five mating pairs that were formed by a random assignment of five males to five females, disregarding their pedigree. Because of the small population size, this random pairing of individuals sometimes involved full-sib mating, therefore increasing the inbreeding level and the variance in reproductive success. Consequently, these populations accumulated inbreeding at a faster rate than the control populations. All other procedures were identical to those of the control populations, ensuring that rearing conditions and density from birth to adulthood were similar in both treatments.

At each generation, offspring were given plenty of time to mature and females were also given a sufficiently long time to produce offspring. Therefore, selection on maturation rate should have been minimal. Similarly, the selection of the breeding individuals in the inbred populations was largely independent of the clutch size, thus minimizing direct selection on fecundity. In the case of sterility, randomly chosen offspring from another female within the population replaced the sterile female’s planned contribution to the next generation.

Six populations went extinct during the course of the study, five in the inbred treatment and one in the control treatment. Additionally, one population from the inbred treatment was excluded because of a manipulation mistake. We considered a population as extinct when the total number of young produced in the population was insufficient to start a new generation (i.e. less than five males or five females in the inbred populations, or 10 males or 10 females in the control populations).

Recorded variables

At each generation, we recorded the following variables.

Female length

Standard length was obtained by immobilizing the female in cold water (8–10 °C) and filming (for 10 s) or photographing her each time she produced a clutch. The fish rapidly regained vigour after being returned to the experimental tanks (24 °C). For generations 1 to 6, filming was performed using a digital video camera (Canon MVX3i) and for generations 7 to 10, a digital SLR camera (Canon E 300D) with two mounted lights and a white background was used to photograph the fish. Standard length, from the tip of the upper jaw to the end of the body, excluding the tail, was calculated using Image-Pro® Plus, version 4.5, for the video images and Image-J version 1.32j (http://rsbweb.nih.gov/ij/) for the photographs.

Time to first clutch

This corresponds to the number of days it took for a female to give birth to her first clutch after the male was removed from her compartment.

Clutch size

Clutch size corresponds to the total number of offspring in the first clutch produced by the female. After mating, female guppies store sperm and can produce a clutch ca. every month (Larsen, Kleven, Pélabon and Rosenqvist unpublished data). Each female was kept in the experiment until she had produced a minimum of five offspring to ensure a high probability of obtaining at least one offspring of each sex. Females that laid a first clutch of less than five offspring were therefore allowed to produce a second and sometimes a third clutch. To avoid the possible confounding effect of a change in clutch size with clutch succession (Larsen, Kleven, Pélabon and Rosenqvist unpublished data), we only considered the size of the first clutch.

Offspring survival

This corresponds to the proportion of offspring within each family that survived from birth until 1 year of age.


A female was considered sterile if she failed to produce a clutch within 1 year after removal of the male. Note that because males were mated with only one female, we were unable to separate male sterility from female sterility.

Inbreeding coefficient F

We calculated the Wright’s inbreeding coefficient (F) from the pedigree using the program FSpeed 2, version 2.04a. The coefficient F is defined as the probability that two alleles are identical by decent.


As the number of offspring in the rearing compartment may affect their survival, growth and subsequent reproductive performance, offspring density was recorded.

Statistical analyses

As a result of the design of the experiment (i.e. replicated populations within each treatment), all analyses were performed using Generalized Linear Mixed-effects models (GLMM, lmer function in R) with population identity as random factor.

Purging should be characterized by an initial period where inbreeding depression increases in the inbred populations, followed by a period where inbreeding depression decreases and the expression of the life-history traits cease to differ between the inbred and the control populations. Using only the persisting populations, we tested for changes in inbreeding depression in the four life-history traits using models, where generation and the interaction between generation and treatment were entered as fixed factors, female length or rearing density as covariates. To determine the significance of the interaction term, we tested whether the explanatory power of the model differed significantly when the interaction term was excluded. We compared models with and without the interaction term using likelihood ratio tests (LRT) (Crawley, 2007, p. 498). In these models that tested for inbreeding depression, we did not include treatment as a fixed effect in order to obtain the contrast between treatments at each generation (Crawley, 2007, pp. 330–331). The value of this contrast, denoting the difference between the control and the inbred treatments was the quantity of inbreeding depression (δ). This difference was then back-transformed to the original scale (Tables 2–5). Preliminary analyses revealed that female length, corrected for female age, negatively affected the reproductive delay (parameter estimates ± SE of the regression: β = −1.27 ± 0.64, = 0.036, β2 = 0.27 ± 0.14, = 0.039) and the probability of sterility (β = −3.37 ± 0.49, < 0.0001), and positively affected the number of offspring produced (β = 0.92 ± 0.059, < 0.0001). Furthermore, female length decreased with density in the rearing aquaria and with the inbreeding coefficient F. Controlling for age, female length differed among treatments (= 875, F1, 96 = 19.32, < 0.0001) and generations (= 875, F8, 96 = 18.73, < 0.0001), without interaction (= 875, F8, 96 = 1.046, = 0.41). Rearing density, on the other hand, did not differ significantly among treatments (= 1076, F1, 119 = 2.07, = 0.15), but differed among generations (= 1076, F9, 119 = 15.41, < 0.0001), as a result of variation in clutch size (see Results). We therefore included these covariates when necessary (see Tables 2–5) centred on the generation mean to present parameter estimates as mean (±SE) trait values for each treatment. Clutch size was Poisson distributed and analysed with a log link function. Time to first clutch was log-transformed to improve the normality of residuals. Sterility and survival had binomial distributions and were analysed with a logit link function. As R does not provide P-values for variables with normal errors, P-values were estimated by Markov Chain Monte Carlo (MCMC) sampling of the parameter estimates in the model testing the significant difference among treatments in each generation for the time to first clutch (Crawley, 2007).

Table 2.   Mean clutch size in inbred and control populations at each generation. Means (inline image) were obtained by back-transforming the parameter estimates (β) from the model controlling for mother length (see Materials and Methods). Parameter estimates and standard errors are reported on log scale. Sample sizes are given at two levels; number of families, nfam, and number of populations (npop). Inbreeding depression δ ± SE corresponds to the parameter estimates given in the model testing the difference among treatments within generation (see Materials and Methods). The P-values of these estimates are given in the right-most column. Percentage difference between treatments was calculated from the log difference between treatments (see Materials and Methods).
GenerationControlInbred% difference (δ ± SE)p (δ)
nfam (npop)*inline image (β ± SE)nfam (npop)*inline image (β ± SE)
  1. *Sample sizes vary because of female sterility and missing data in the covariate; female length.

170 (9)4.37 (1.47 ± 0.08)19 (4)5.05 (1.62 ± 0.13)13.5 (−0.15 ± 0.158)0.359
286 (9)7.17 (1.97 ± 0.07)18 (4)6.96 (1.94 ± 0.13)2.27 (0.023 ± 0.149)0.880
385 (9)5.48 (1.70 ± 0.08)20 (4)4.00 (1.39 ± 0.15)36.95 (0.314 ± 0.172)0.067
482 (9)4.67 (1.54 ± 0.08)18 (4)4.22 (1.44 ± 0.15)10.70 (0.102 ± 0.167)0.543
569 (9)4.26 (1.45 ± 0.09)18 (4)3.04 (1.11 ± 0.19)40.02 (0.337 ± 0.212)0.111
673 (8)4.04 (1.40 ± 0.09)13 (3)3.32 (1.20 ± 0.19)21.64 (0.196 ± 0.213)0.358
783 (9)4.41 (1.48 ± 0.08)18 (4)5.11 (1.63 ± 0.14)13.81 (−0.149 ± 0.160)0.354
883 (9)4.37 (1.47 ± 0.08)18 (4)3.74 (1.32 ± 0.15)16.96 (0.156 ± 0.169)0.357
985 (9)4.29 (1.46 ± 0.08)21 (4)4.99 (1.61 ± 0.13)14.17 (−0.153 ± 0.147)0.299
1088 (9)4.70 (1.55 ± 0.08)19 (4)4.14 (1.42 ± 0.14)13.54 (0.127 ± 0.167)0.448
Table 3.   Average proportion of offspring surviving during their first year in inbred and control populations at each generation. Means (inline image) were obtained by back-transformation of the parameter estimates β from the model controlling for rearing density (see Materials and methods). Estimates are reported as log-odds. Further details on sample size and δ are given in Table 2.
GenerationControlInbred% difference (δ ± SE)p (δ)
nfam (npop)*inline image (β ± SE)nfam (npop)*inline image (β ± SE)
  1. *Sample sizes vary because of female sterility and missing data in number of young that survived.

170 (9)0.97 (3.34 ± 0.31)19 (4)0.98 (3.97 ± 0.62)1 (−0.633 ± 0.693)0.361
287 (9)0.97 (3.47 ± 0.29)18 (4)0.95 (3.08 ± 0.55)2 (0.394 ± 0.614)0.521
385 (9)0.84 (1.65 ± 0.23)20 (4)0.83 (1.63 ± 0.36)1 (0.024 ± 0.429)0.955
482 (9)0.95 (3.01 ± 0.32)18 (4)0.77 (1.21 ± 0.34)18 (1.797 ± 0.467)< 0.001
579 (9)0.98 (3.81 ± 0.36)18 (4)0.93 (2.64 ± 0.44)6 (1.168 ± 0.569)0.04
683 (9)0.96 (3.33 ± 0.29)18 (4)0.96 (3.14 ± 0.56)0 (0.193 ± 0.634)0.76
783 (9)0.89 (2.09 ± 0.22)18 (4)0.89 (2.15 ± 0.36)0 (−0.054 ± 0.421)0.898
884 (9)0.90 (2.25 ± 0.23)19 (4)0.91 (2.26 ± 0.38)1 (−0.011 ± 0.443)0.981
984 (9)0.92 (2.43 ± 0.27)20 (4)0.92 (2.42 ± 0.51)0 (0.005 ± 0.572)0.993
1088 (9)0.87 (1.88 ± 0.22)19 (4)0.89 (2.16 ± 0.41)2 (−0.275 ± 0.471)0.559
Table 4.   Mean time to produce the first clutch, in days, in inbred and control populations at each generation. Means (inline image) were obtained by back-transformation of the parameter estimates β (on log scale) from the model correcting for mother length. Parameter estimates are reported on log scale. Further details on sample size and δ are given in Table 2.
GenerationControlInbred% difference (δ ± SE)p (δ)†
nfam (npop)*inline image (β ± SE)nfam (npop)*inline image (β ± SE)
  1. *Sample sizes vary because of female sterility and missing data in the covariate; female length.

  2. P-values obtained by MCMC sampling of the parameter estimates of the model.

170 (9)18.76 (2.93 ± 0.09)19 (4)20.41 (3.02 ± 0.15)8.08 (−0.084 ± 0.173)0.5934
286 (9)18.18 (2.90 ± 0.08)18 (4)21.87 (2.98 ± 0.15)16.90 (−0.185 ± 0.173)0.2262
385 (9)19.52 (2.97 ± 0.08)20 (4)24.66 (3.08 ± 0.15)20.86 (−0.234 ± 0.172)0.1352
482 (9)19.67 (2.98 ± 0.08)18 (4)19.96 (3.20 ± 0.15)1.45 (−0.015 ± 0.172)0.9442
569 (9)21.07 (3.04 ± 0.09)18 (4)19.52 (2.99 ± 0.16)8.00 (0.077 ± 0.179)0.6748
673 (8)15.36 (2.73 ± 0.09)13 (3)25.50 (2.97 ± 0.17)39.76 (−0.507 ± 0.198)0.0060
783 (9)17.26 (2.85 ± 0.08)18 (4)16.68 (3.24 ± 0.18)3.47 (0.034 ± 0.172)0.8136
883 (9)16.09 (2.77 ± 0.08)18 (4)19.22 (2.81 ± 0.15)16.24 (−0.177 ± 0.172)0.2610
985 (9)16.82 (2.82 ± 0.08)21 (4)19.10 (2.96 ± 0.15)11.96 (−0.127 ± 0.165)0.3908
1088 (9)19.53 (2.97 ± 0.08)19 (4)19.79 (2.94 ± 0.14)1.33 (−0.013 ± 0.169)0.9408
Table 5.   Average proportion of sterile females in inbred and control populations at each generation. Means (inline image) were obtained by back-transformation of the parameter estimates β (on log scale) from the model controlling for mother length. Parameter estimates are reported on log-odds scale. NA: nonavailable data; the absence of parameter estimates for generations 8 and 9 results from the absence of sterile female at these two generations, we were therefore unable to fit a binomial model. Further details on sample size and δ are given in Table 2.
GenerationControlInbred% difference (δ ± SE)p (δ)
nfam (npop)*inline image (β ± SE)nfam (npop)*inline image (β ± SE)
  1. *Sample sizes vary because of missing data in the covariate; female length.

190 (9)0.14 (−1.84 ± 0.40)20 (4)0.034 (−3.33 ± 1.21)10.6 (1.488 ± 1.272)0.242
290 (9)0.039 (−3.18 ± 0.63)20 (4)0.042 (−3.12 ± 0.92)0.3 (−0.066 ± 1.103)0.952
388 (9)0.019 (−3.93 ± 0.65)20 (4)NA0 
488 (9)0.046 (−3.03 ± 0.53)20 (4)0.059 (−2.76 ± 0.90)1.3 (−0.270 ± 1.038)0.795
580 (9)0.048 (−3.06 ± 0.49)20 (4)0.013 (−4.33 ± 0.97)3.5 (1.2764 ± 1.004)0.204
680 (9)0.058 (−2.78 ± 0.50)15 (3)0.048 (−2.98 ± 1.02)1.0 (0.197 ± 1.125)0.861
790 (9)0.064 (−2.68 ± 0.49)20 (4)0.048 (−2.98 ± 0.94)1.6 (0.302 ± 1.058)0.775
889 (9)0.078 (−2.46 ± 0.52)19 (4)0.041 (−3.15 ± 1.18)3.7 (0.687 ± 1.293)0.595
983 (9)0.057 (−2.83 ± 0.71)20 (4)NA0 
1089 (9)0.008 (−4.79 ± 1.15)20 (4)0.033 (−3.37 ± 1.17)2.5 (−1.421 ± 1.644)0.387

To estimate the temporal effect on trait value, we fitted models without intercepts using a dummy variable combining the generation and treatment effects as a fixed factor. These models provide the mean trait value for each treatment within every generation (Tables 2–5) instead of contrasts from the model intercept (Crawley, 2007, p. 331).

As it was not possible to avoid inbreeding entirely in the control populations (see Results), the decrease in the difference between inbred and control populations after an initial period of increase could result from inbreeding depression affecting the control populations. To assess whether this was the case, we analysed the relationship between F and the expression of the traits. Occurrence of purging should be characterized by a decrease in the strength of the relationship between F and the trait value from generations when inbreeding depression occurred compared to the generations after purging has occurred. We tested for the change in the slope of the regression between F and the trait values using models where the interaction term between F and generation represents the change in the slope with generation. Note that we did not include the first two to three generations in this analysis, depending on the trait, because no inbreeding depression was observed at this early stage of the experiment. To test the significance of the interaction term, we compared models with and without the interaction term using likelihood ratio tests. Inbreeding depression in clutch size, sterility and offspring survival may be related to the level of inbreeding of the mother or to the one of the offspring. Consequently, both the maternal and the offspring F may be entered as predictor variables in the analyses. However, these two inbreeding coefficients are strongly correlated rendering the estimation of their respective contribution to inbreeding depression inaccurate. We therefore decided to use only the offspring coefficient of inbreeding in all analyses.

By focusing on the P value of statistical tests, we run the risk of masking biological meaningful results (Yoccoz, 1991). Moran (2003) clearly explained how sequential Bonferroni adjustment of multiple statistical tests may negatively affect the interpretation of studies where repeated patterns of small amplitude were observed. This is particularly true in this study where the statistical significance of the effects tend to be weak or marginal while the estimates show patterns of variation in inbreeding depression that follow the patterns expected if purging would occur. Therefore, we preferred to present in our results the exact P values of the different tests, and discussed the results in terms of effect size and concordance between the observed and expected patterns.


Changes in the level of inbreeding across generations

Inbreeding increased faster in the inbred populations than in the control populations (Fig. 1). For the majority of families in the inbred populations (44/50 in generation 2), the inbreeding coefficient remained below 0.10 during the first two generations and started to increase strongly at the third generation. By generation 10, the average inbreeding coefficient for the families ranged from 0.31 to 0.56 (mean = 0.41) in the inbred populations, and from 0.075 to 0.17 (mean = 0.12) in the control populations.

Figure 1.

 Increase in the inbreeding coefficient, F, from generation 1 to 10 in the control and the inbred populations. Each point corresponds to a family. The solid line connects the generation mean for each treatment.

Inbreeding depression in the extinct populations from the inbred treatment

The five inbred populations that went extinct during the experiment had higher levels of inbreeding than the persisting populations in the same treatment, and tended to show inbreeding depression in all traits (Table 1, Fig. 2). Extinction resulted mainly from high level of sterility combined with low clutch size.

Table 1.   Mean ± SE trait values for the extinct and persisting populations from the inbred treatment at each generation when extinction occurred. We report here the estimates from the models correcting for the effect of female length and rearing density.
 Generation 5Generation 6Generation 7
Extinct (= 3)Persisting (= 6)Extinct (= 1)Persisting (= 5)Extinct (= 1)Persisting (= 4)
F coefficient0.14 ± 0.020.08 ± 0.0090.34 ± 0.060.10 ± 0.0090.32 ± 0.010.11 ± 0.008
Log (clutch size)1.32 ± 0.191.11 ± 0.190.27 ± 0.661.20 ± 0.191.70 ± 0.341.63 ± 0.14
Proportion survival [log odds (survival)]0.89
(2.16 ± 0.48)
(2.64 ± 0.44)
(3.63 ± 0.99)
(3.14 ± 0.56)
(1.96 ± 0.93)
(2.15 ± 0.36)
Log [time to first clutch (days)]3.02 ± 0.152.97 ± 0.173.18 ± 0.353.24 ± 0.183.14 ± 0.282.81 ± 0.15
Proportion sterility [log odds (sterile females)]0.082
(−2.35 ± 0.73)
(−4.33 ± 0.96)
(−2.18 ± 1.26)
(−2.97 ± 1.02)
(−0.34 ± 1.25)
(−2.98 ± 0.94)
Figure 2.

 Difference in the expression of life-history traits between the control treatment (black line) and the inbred treatment across generations. The inbred populations that went extinct are plotted separately from the persisting populations. Mean values for each treatment correspond to the parameter estimates ± SE from the model controlling for variation in mother length (clutch size, time to first clutch and sterility) and rearing density (survival) (see Materials and Methods). The lack of parameter estimates for sterility at generation 3 and 9 is due to the absence of sterile females in the inbred treatment.

Variation in inbreeding depression in persisting populations

For two traits, clutch size and offspring survival, we observed a first phase where differences between treatments increased with successive generations, followed by a phase where these differences decreased (Fig. 2, Tables 2 and 3), as expected if purging occurred. Clutch size showed the maximum inbreeding depression at generation 5, with an average decrease of 40% in the number of offspring produced in the first clutch (Table 2). Offspring survival was affected earlier and showed a maximum decrease of 18% in survival at generation 4 (Table 3). Note, however, that the interaction term between generation and treatment, reflecting changes in the differences among treatments across generations, explained a significant part of the variation for survival only (LRT: clutch size: χ2 = 10.14, = 0.33; survival: χ2 = 16.84, = 0.05).

Time to first clutch and the proportion of sterile females did not show clear signs of inbreeding depression, except at generation 6 where time to first clutch was longer in the inbred populations (Fig. 2; Tables 4 and 5). Excluding the interaction between treatment and generation did not change the explanatory power of the models for either of the latter traits (LRT: time to first clutch: χ2 = 7.42, P = 0.59, sterility: χ2 = 5.89, = 0.75).

For all traits except sterility, the relationship between the trait expression and the inbreeding coefficient F was the strongest at the generations where inbreeding depression was at its maximum (Fig. 3), confirming that inbreeding was the most likely factor affecting the difference in trait expression between treatments. Furthermore, the strength of this relationship decreased at later generations (Fig. 3), confirming that inbreeding (i.e. homozygosity) ceased to affect trait expression negatively, as expected in the case of purging. The interaction term denoting the change in slope across generations was statistically significant for survival (LRT: χ2 = 15.26, P = 0.03), but not for the other traits (clutch size: χ2 = 8.022, = 0.33, time to first clutch: χ2 = 7.66, = 0.36, sterility: χ2 = 2.16, = 0.82). However, changes in slope were in the expected direction for clutch size, and time to first clutch after generation 6.

Figure 3.

 Variation across generation in the strength of the relationship between the inbreeding coefficient F and the trait expression for each life-history trait. We report the average slope ± SE of the regression model between F and the trait value (see Materials and Methods). For clutch size and offspring survival, negative values of β indicate inbreeding depression, whereas for time to first clutch and sterility, positive values indicate inbreeding depression. For offspring survival, the slope at generation 4, that is at the onset of inbreeding depression, significantly differed from slopes at later generations, after purging occurred (G4 vs. G6: = 0.02, G4 vs. G7: = 0.003, G4 vs. G8 = 0.004, G4 vs. G9 = 0.003, G4 vs. G10: = 0.005).


Clear evidence for purging remains rare, partly because of the difficulties of ruling out effects of confounding factors such as adaptation to laboratory conditions or drift (Crnokrak & Barrett, 2002). Using a multigenerational approach where the expression of different life-history traits was compared between inbred and contemporary control populations, we showed that two of four life-history traits displayed inbreeding depression during the initial phase of the experiment, when inbreeding increased. Later, we observed a rebound in trait expression back to the level of the control populations, despite further increases in the level of inbreeding. Simultaneously with the rebound in trait expression in the inbred populations, the strength of the relationship between the inbreeding coefficient and the trait value diminished (Fig. 3), confirming that the deleterious effect of homozygosity faded after a few generations of inbreeding. The recovery of trait value was most evident for survival and clutch size, although statistically significant for survival only. The other two traits display a similar but weaker pattern. Because we only analysed data for the inbred populations that persisted, the most likely explanation for the temporal variation in inbreeding depression is that purging of deleterious alleles for offspring survival and clutch size occurred in these four populations. For the two other traits, however, the process leading to an absence of marked inbreeding depression at the tenth generation is probably different. Indeed, female sterility was particularly high in the five populations that went extinct during the course of the study, and fecund females produced their first clutch later than in the persisting populations. This suggests that the removal of deleterious alleles that caused sterility and possibly late clutch initiation occurred in our experiment by the removal of whole populations with presumably high genetic load for these two traits. Although patterns of temporal variation in inbreeding depression tended to be consistent across traits, the statistical power of the analysis remained relatively weak, because population was the statistical unit in this experiment. In the following, we first present possible alternative explanations for the observed temporal variation in inbreeding depression, and explain why purging of deleterious alleles appears as the most likely mechanism generating this variation.

Despite our attempts to maintain populations in constant conditions throughout the experiment, substantial variation among generations occurred in the expression of the different traits, independently of variation in inbreeding level, as illustrated by parallel variations in clutch size and offspring survival in the inbred and control lines across generations (Fig. 2). As we synchronized the onset of each generation across treatments, differences in trait expression at each generation should be independent of environmental factors, and essentially represent the effect of differences in the level of inbreeding.

Crnokrak & Barrett (2002) noticed that the decrease in inbreeding depression in multigenerational studies with control lines does not guarantee the purging of genetic load in inbred lines. Indeed, drift in control lines may lead to the fixation of mildly deleterious alleles, therefore decreasing fitness and may be mistakenly interpreted as a decrease in the level of inbreeding depression. The effect of genetic drift could be important in our study where the control populations, although larger than the inbred ones, were still of a small size (= 20). Furthermore, control populations also accumulated inbreeding, although at a lower rate than the inbred populations (Fig. 1). Therefore, although the coefficient of inbreeding never exceeded 0.17 (average F generation 10 = 0.12) in the control lines, the decrease in inbreeding depression could have potentially resulted from a fitness decrease in the control lines instead of a fitness rebound in the inbred lines. We do not deny the possibility that fitness components in some control lines could have been negatively affected by the fixation of mildly deleterious alleles. However, several observations suggest that this mechanism does not explain the fitness rebound observed in the inbred populations. First, we did not observe any tendency in the control populations for a decrease in trait expression over time (i.e. with an increasing level of inbreeding). Additionally, clear inbreeding depression in the inbred treatment was first noticed at generation four and onward, when the average level of F exceeded 0.13. Therefore, it is unlikely that inbreeding depression significantly affected any of the control populations. Finally, if the decrease in fitness difference between treatments resulted primarily from inbreeding depression in the control populations, we would expect the relationship between F and the expression of traits to remain similar over the entire study. This was clearly not the case, and the decrease in the steepness of the relationship in later generations (Fig. 3) strongly suggests that the fitness rebound of the inbred populations most likely resulted from the purging of recessive deleterious alleles.

The severity of inbreeding depression in clutch size and offspring survival suggests that the original population sampled for this experiment harboured some genetic load that affected the expression of the traits when inbreeding occurred. Furthermore, assuming that purging is the mechanism by which the fitness traits rebound, purging of deleterious alleles was efficient and allowed complete recovery of the trait expression during the second half of the study (Fig. 2). Although these results may be surprising in the light of some previous work where the efficiency of purging has been questioned (Ballou, 1997; Byers & Waller, 1999; Frankham et al., 2001), they tend to support the view purported by Crnokrak & Barrett (2002) that purging is common, especially in studies examining changes in inbreeding depression with increasing levels of inbreeding. Nevertheless, the recovery in fitness components observed in our study (ca. 100% recovery), is considerable compared with the average of 20% reported by Crnokrak & Barrett (2002). Furthermore, these authors also noticed that, in mammals, life-history traits tend to show less purging than morphological traits, a result at odds with theoretical expectations (Fu et al., 1998). This raises questions about the factors affecting the efficiency of purging and our capacity to predict the occurrence and magnitude of purging in inbred populations.

Opportunity for purging should be affected essentially by the genetic architecture of the traits, the intensity of selection acting against the deleterious alleles, and how rapidly inbreeding increases (Charlesworth & Charlesworth, 1987, 1999). It appears that inbreeding depression in the four traits studied here is due to completely recessive or partly recessive deleterious alleles, as this is a prerequisite for purging to occur (Lynch & Walsh, 1998). This is further supported by the observation that some inbred lines continued to perform better than the control lines despite high levels of inbreeding (Fig. 3); a situation that would not be possible if over-dominance was responsible for the inbreeding depression, as all inbred lines would be expected to perform worse than the average outbred line.

Intensity and efficiency of selection against deleterious alleles depends on both the selection pressure (i.e. how much the allele decreases the fitness of homozygous individuals), and population size. Drift should limit the efficiency of selection in small populations, and it is expected that purging will primarily affect those alleles closely linked to fitness, whereas mildly deleterious alleles may reach fixation via drift (Bryant et al., 1990; Hedrick, 1994; Willis, 1999). Our results thus suggest that inbreeding depression in the life-history traits studied resulted from the expression of few recessive mutations of large effect size. Considering that life-history traits should be affected by a multitude of loci (Houle, 1992, 1998), this further suggests that purging affected only a limited number of loci for which mutations have the largest effect, and that many deleterious alleles of smaller effect size remained hidden in the inbred populations.

Surprisingly, the rebound of trait expression was similar for clutch size and offspring survival, despite presumably different selection pressures on these two traits. As the contribution of a female to the next generation was limited to a single offspring of each sex, selection for large clutches should have been rather limited. However, assuming that offspring who did not survive their first year were the ones harbouring the highest genetic load, genes affecting survival should have been purged by offspring mortality. Considering the more or less simultaneous purging in clutch size and offspring survival, one can suggest that purging indirectly affected clutch size, either by purging deleterious alleles on offspring survival pleiotropically linked with clutch size, or by a selective sweep (Ridley, 2004). Genetic correlation between survival before and after birth would create a correlation between offspring survival and clutch size. In this case, purging affecting offspring survival could indirectly affect clutch size. This further illustrates the difficulties in predicting the occurrence and the magnitude of purging since genetic correlations (i.e. pleiotropy or linkage) may affect several traits simultaneously. Progress in the understanding of purging will require the integration of multivariate quantitative genetic theory within the population genetic framework generally used to study inbreeding depression.

Although this study suggests that purging could be an efficient mechanism to decrease the deleterious effect of inbreeding, it is important to keep in mind that measures of purging are only relevant to the environment in which it was studied. Adaptation to captive environments is an acknowledged problem in conservation biology, where populations are kept in a benign environment for several generations before reintroduction to the wild (Snyder et al., 1996; Araki et al., 2008; Fraser, 2008). For example, behaviours favourable in natural environments, such as predator avoidance, alertness, foraging and courtship, may be under different selection in captivity, where animals are fed regularly, predators excluded, and mating generally controlled. Similarly, it is well documented that the outcome of inbreeding depends on the environment in which the organisms live (Crnokrak & Roff, 1999), and purging in one environment does not necessarily select the best alleles for another environment (Bijlsma et al., 1999; Reed & Bryant, 2000). Frankham (2008) suggested that animals reared in a captive environment experience selection on different alleles than their conspecifics in the natural environment (see also Van Oosterhout et al., 2007). Wang & Hill (1999) also suggested that if inbreeding is due to alleles of large effect, as it seems to be the case in our study, purging of these alleles may cause the loss of variation at other loci and seriously hamper the population’s evolvability.

Our results suggest that purging occurs within inbreed populations and appears as an efficient process to reduce genetic load. However, inducing homozygosity to remove recessive deleterious alleles also appears as a risky conservation practice considering that five of nine populations in our study went extinct as a result of severe inbreeding depression. Furthermore, purging may favour alleles which are deleterious in a natural environment and removal of deleterious alleles will inevitably deplete the genetic variance of the population.


We would like to thank Henriette Vaagland, Jan Sand, Frode Killingberg, Arnt Narve Bordal, Ronny Höglund, Tonje Aronsen, all PhD and MSc. students for help setting up and maintaining experiments, Ivar Herfindal for valuable aid on the statistics, this study was funded by the Research council of Norway, projects: 121089/220, 141085/720 and 166869/V40. We also thank The Norwegian Animal Research Authority for giving consent to our research.