The purge of genetic load through restricted panmixia in a Drosophila experiment


Aurora García-Dorado, Departamento de Genética, Facultad de Biología, Universidad Complutense. 28040 Madrid, Spain.
Tel.: 3414973810; fax: 341914974844; e-mail:


Using Drosophila melanogaster, we explore the consequences of restricted panmixia (RP) on the genetic load caused by segregating deleterious recessive alleles in a population where females mate a full sib with probability about ½ and mate randomly otherwise. We find that this breeding structure purges roughly half the load concealed in heterozygous condition. Furthermore, fitness did not increase after panmixia was restored, implying that, during RP, the excess of expressed load induced by inbreeding had also been efficiently purged. We find evidences for adaptation to laboratory conditions and to specific selective pressures imposed by the RP protocol. We discuss some of the consequences of these results, both for the evolution of population breeding structures and for the design of conservation programmes.


The genetic load of a population is of great importance for the evolution of its reproductive structure and other biological features. It is also a main factor determining the population extinction risk and should be considered in the design of conservation programmes for endangered populations. The genetic load owing to segregating deleterious alleles can be split into two fractions. One is the expressed load, which is the reduction in mean fitness below that of a hypothetical population with no segregating deleterious alleles. The other one is the concealed load, which is a main genetic determinant of the evolution of means fitness during periods of reduced population size and is attributed to the deleterious effects that are not expressed in the heterozygous condition. This concealed load can be measured by the fitness inbreeding depression rate δ, i.e., by the rate at which fitness would decline for increasing inbreeding in the absence of selection. For a population that has undergo some inbreeding, we should appreciate the difference between its inbreeding depression rate (i.e., its concealed load) and the inbreeding depression it shows, which is just the magnitude of its reduction in fitness because of the originally concealed load that became expressed. In any case, Inbreeding causes depression primarily because it increases the probability of homozygosity for deleterious (partially) recessive alleles. However, this increased homozygosity also implies an increase in the strength of natural selection against those deleterious alleles, known as purge, which can be critical to determine the actual evolution of both the expressed and the concealed load. Here, we will experimentally analyse the efficiency of enforcing a proportion of inbred mating to purge both the expressed and the concealed load of a Drosophila population.

When the effective size of a population undergoes a stable reduction and its breeding structure remains panmictic, analytical predictions show that purge can substantially reduce the fitness decline (García-Dorado, 2008). This is consistent both with simulation results and with the analysis of experimental evidence (García et al., 1994; Pérez-Figueroa et al., 2009). Nevertheless, populations experiencing bottlenecking followed by expansion in size do not generally recover prebottleneck fitness levels (Leberg & Firmin, 2007), so that bottlenecks should generally be avoided in conservation management, as they increase the extinction risk from several causes (Frankham et al., 2001).

However, purge can occur with just a slight increase in drift if inbred mating is more common than randomly expected from population numbers. This will reduce the concealed load and could also reduce the expressed load, at least after panmixia is restored. This situation may be quite general in plants (Byers & Waller, 1999) and in populations of many animal species structured in breeding groups for which, because of behavioural and geographical factors, only a fraction of the individuals disperse before breeding. The efficiency of such breeding structures to purge the genetic load is of great importance to their evolutionary success. It has been theoretically shown that partially subdivided populations are expected to have smaller genetic load that completely panmictic ones (Whitlock, 2002; Glemin et al., 2003; Roze & Rousset, 2004), but quantitative predictions depend on hardly estimable genetic parameter, so that empirical evaluations are necessary.

The efficiency of inbreeding to reduce δ has been studied by analysing fitness in pedigreed data from cattle breeds and from zoo endangered populations (Boakes et al., 2007; Gulisija & Crow, 2007). This revealed poor purging efficiency but, as the authors note, for nonendangered populations the power of the data is limited by the small purge opportunity, and for endangered populations it is likely that most purge had occurred from inbreeding previous to the available pedigree information. Purging has been found to be efficient in extreme situations where a population is subdivided into a number of small panmictic isolates that, after a number of generations, are finally merged into a single panmictic pool (Swindell & Bouzat, 2006; Fox et al., 2008). In this way, the inbreeding caused by subdivision produces both depression and purge, but the depression will be concealed after panmixia is restored. In practice, however, inbreeding depression and other genetic and demographic factors will increase the extinction risk of the isolates, so that a fraction of them may have been lost by the time they are merged. This would imply between-line selection, which should increase the efficiency of purge, but also would imply that subsequent panmixia would not restore the genetic diversity available before subdivision. Therefore, inbreeding with between-line selection has been discouraged for conservation programmes (Wang, 2000).

Here, we consider a population that is not completely subdivided, but has a breeding structure inducing restricted panmixia (RP) and causing a bounded increase in the average inbreeding coefficient. It has been theoretically shown that partial inbreeding can reduce the average equilibrium frequency of deleterious alleles in natural finite populations (Glemin, 2003) and can increase the rate of fixation of recessive favourable alleles under artificial selection (Caballero & Hill, 1992; Caballero & Santiago, 1995), and a simulated system where full sibs were mated whenever possible proved to be particularly efficient in the later respect (Caballero et al., 1991). However, a reduction in the average frequency q of unfavourable alleles can concur with a reduction in mean fitness, because of an increase in the probability of homozygosity caused by inbreeding. Therefore, for many purposes, to determine the extinction risk of natural populations and the viability of conservation programs, the behaviour of the expressed and concealed load induced by RP is more relevant than the change in mean deleterious frequencies. Although RP has resulted in a slight improvement in the response to artificial selection from new mutation after relaxed selection (Merchante et al., 1995), empirical estimates of the consequences of RP for fitness components in segregating populations have not been obtained, and the interpretation of results for populations maintained under subdivision is often obscured by other genetic changes, as is the case of synchronous adaptation to captive conditions (Crnokrak & Barrett, 2002).

Our aim is to enquire whether RP can lead to a net reduction of load of not too small populations. For this purpose, we tested the efficiency of a RP scheme inducing a large proportion of full-sib mating (about 50%) to reduce the expressed and concealed load in a laboratory population of Drosophila melanogaster that could be assumed to be at the mutation-selection-drift (MSD) balance (García-Dorado et al., 2007).

Materials and methods

Base population and culture conditions

As a base, we used the population C2 of Drosophila melanogaster which had been maintained in the laboratory for 124 generations by circular mating in 25 bottles under high crowding levels until generation 108 and in 35–40 bottles thereafter (see Ávila et al., 2006 for details on the design). Hundred and eight generations after being founded, this population was near the MSD balance regarding the genetic variability for viability and for two bristle traits, with an effective population size about 500 (García-Dorado et al., 2007). A balancer stock marked with the Cy (Curly wings) and L2 (Lobe) genes was used to estimate competitive fitness.

Flies were reared in the standard medium formula of this laboratory (brewer’s yeast-agar-sucrose) and were handled at room temperature under CO2 anaesthesia. All cultures were incubated at 25 ± 1 °C, 45 ± 5% relative humidity and maintained under continuous lighting.

Experimental design

Males and virgin females were sampled from the C2 population to derive a restricted panmixia (RP) and a panmictic control population (Fig. 1).

Figure 1.

 Experimental design for the maintenance of the restricted panmixia and the control populations.

The control population was started with 55 vials, each with two males and two virgin females as parents. Each generation, the order i of the vials was randomly assigned, two male and two virgin female offspring were sampled from each vial during the first 5 emergence days, and the males from vial i were mated to females from vial + 1. The mating scheme was circular, i.e. males from the last vial where mated to females from the first one.

In the RP population, the procedure was similar and synchronous to that followed in the control, but only two females were sampled from each vial after 5 emergence days, and they were chosen at random between those whose pigmentation indicated they were old enough as to have been fertilized. Each generation, a female from vial i was placed into a new vial together with a female from vial + 1, again following a circular scheme. This breeding structure mimics that of a population where mating occurs within groups that are partially shuffled each generation because of the dispersal of a fraction of the breeding individuals. Similar structures could also be implemented in conservation programmes to favour purging.

In occasions, some vials failed to provide two male or two female offspring, so that the number of vials in which the population was maintained (initially 55 per population) slowly decreased. By generation 43, only 43 control and 33 RP vials were left. From generation 44 onwards, spare individuals were sampled (a spare couple per vial in the control population and a spare inseminated female per vial in the RP population) and were randomly used to replace missing individuals (and, in generation 44, to restore 43 vials in the RP population) so that 43 control and 43 RP vials were maintained up to the end of the experiment.

Assay of competitive fitness w

General protocol

Panmictic assay: one male and one virgin female offspring were sampled from each vial, and a circular mating scheme was applied where a male offspring from vial i was mated to virgin females offspring from vial + 1 (generation 1 of the assay protocol). Order numbers were randomly reassigned to the new vials, three virgin female offspring and six male offspring were sampled from each vial and, after 5 days, the six males from vial i were placed in a single vial together with three females from vial + 1 and three virgin females from the Cy/L2 stock (generation 2). Although mating to individuals from the same vial was avoided, this should cause negligible departure from panmixia given the number of vials involved. The number of wild and nonwild offspring per vial (generation 3) was recorded after emergence. Competitive fitness w was computed from the ratio of wild to nonwild numbers in the progeny of generation two females, which was log-scaled to normalize the variable. Therefore, we measure = Loge (number of wild offspring/number of nonwild offspring).

Inbred assay: a similar procedure was synchronously applied, but males were mated to females from the same vial both at generations 1 and 2, so that fitness was assayed for females whose parents were full sibs (i.e. generation two females with inbreeding coefficient 0.25).

Specific evaluations

Using the protocol described above, inbred and/or panmictic fitness were assayed in several instances:

  • • Assay of the initial δ (generation t = 0): Just before the beginning of the experiment, using 50 vials for the panmictic assay and 50 vials for the inbred assay with males and virgin females sampled from the C2 population.
  • • Assay of the expressed load (= 9 and t = 22): Prior to the assay protocol, we sampled a male and a virgin female from each vial in the RP and control populations. For each population, these were randomly mated in pairs so that inbreeding caused by departure from panmixia was destroyed in the RP population. The panmictic assay protocol (explained above) was synchronously applied to the control (314 vials at generation 9 and 253 at generation 22, respectively), the RP population after restored panmixia (247 and 198 vials at generations 9 and 22, respectively) and the RP population without previous destruction of its inbreeding because of the lack of panmixia (197 vials, each with three Cy/L2 females and with three females and three males directly sampled from the same vial of the RP population at generation 22).
  • • Assay of the concealed load (t = 34): Prior to the assay protocol, we sampled 3–4 males and 3–4 females from each vial. In each population, these individuals were pooled and randomly mated in pairs, so that inbreeding attributed to RP was destroyed. Starting from the progeny of these panmictic vials, we used the general protocol stated above to synchronously assay panmictic and inbred fitness for both populations. Fitness was assayed in 109 panmictic vials and 115 inbred vials for the RP population and in 116 panmictic vials and 111 inbred vials for the control population.

Evaluation of fitness components at generation 48

This assay was intended to compare mating precocity and fitness at different emergence times for RP and control individuals. At generation 48, two male and two virgin female offspring were sampled from each vial of the control and RP populations. For each of these four sets (control males, control females, RP males and RP females), individuals were mixed in a bottle to have them randomized. When 4 days old, each male was placed in a single vial together with one of these females and with a Cy/L2 4 days old virgin female. Fifty vials were obtained for each possible case according to the male and female origin: 50 where both male and female came from the control population, 50 with a control male and a RP female, 50 with a RP male and a control female, and 50 with a RP male and a RP female. These vials were observed for five hours, and the time to first mating (τ) was recorded. The phenotype of the first mating female was also observed, so that the proportion of vials where the tested female mated before than the reference Cy/L2 female could be computed (π). After five days, parents were removed from the vials. The numbers of wild and nonwild offspring were daily recorded during 9 emergence days and were used to compute the female’s fitness for early and late emergence periods. Thus, early fitness was computed as:


using offspring born during the emergence period indicated by the subscript (first four emergence days). Similarly, late fitness was


Estimates were similarly obtained for each single emergence day (wi).

Viability evaluation

To assay the effect of RP on the load concealed for viability, this trait was assayed at generations 49 and 60 for the progeny of panmictic females mated to random males and for the progeny of inbred females (F = 0.25) mated to their brothers, both for the control and for the RP populations. In each population, prior to the assay protocol, we sampled 3–4 males and 3–4 females from each vial, which were randomly mated in single pair vials, so that inbreeding attributed to RP was destroyed in the RP population. Starting from the progeny of these panmictic pairs, we obtained inbred or panmictic individuals following the same general protocol as for the fitness assays. To evaluate viability, individuals of generation 2 were mated in pairs for four days and then transferred to a plate with coloured medium. After 24 h, 30 eggs (or the number of eggs available in case this was smaller than 30) were scored and transferred to a new vial with standard culture medium.

In the assay corresponding to generation 49, the progeny from the vials was scored after emergence and egg-to-adult viability was computed as the proportion of eggs producing adults (66 panmictic and 56 inbred vials scored for the control population, and 59 panmictic and 65 inbred vials scored for the RP population).

In the assay corresponding to generation 60, the procedure followed was similar, but the temperature accidentally raised up to 30 °C during one night, just at the end of the pupation period in the last generation of the assay. Therefore, pupae numbers instead of adult numbers were scored, and egg to pupa viability is reported (64 panmictic and 60 inbred vials scored for the control population, and 64 panmictic and 63 inbred vials scored for the RP population). This accident produced delayed emergence and very high pupa-to-adult mortality. The experiment was discontinued at this point.

Data analysis

At each assay, the differences between the groups were tested using one-way anova. For each evaluation, panmictic means for the RP and control populations were compared using standard t tests.

We estimated the inbreeding depression rate δ using synchronous estimates of the panmictic and inbred means. Both fitness and viability were considered as maternal traits, so that inbred assays always corresponded to an inbreeding coefficient F = 0.25.

As our measure of fitness is log-scaled (see above), its δ was directly estimated as


where inline image and inline image stand for the panmictic and inbred fitness average, respectively.

Viability, however, is estimated through a proportion (instead of a ratio), and the natural scale is statistically appropriated in this case. Therefore, δ can be computed from the viability reduction relative to the panmictic viability mean, as inline image where inline image and inline image stand for the panmictic and inbred mean viability, respectively. The rate of inbreeding depression for viability was also estimated as inline image but this estimate cannot be associated with a direct empirical estimate of the standard error. Given the large sample sizes, one-tiled Z contrasts were used in each assay to test whether δ was smaller under RP than in the control population (alternative hypothesis). To test whether δ undergoes an overall reduction, the three P-values were pooled using the standard Fisher’s method.

Regarding the evaluation of fitness components performed at generation 48, time to first mating (τ) and fitness estimated from the pooled emergence for the first 4 or last 4 emergence days (w1–4 and w6–9, respectively) were analysed by two-way anova, with ‘male origin’ and ‘female origin’ as main fixed factors. For the proportion of vials in which the female from the assayed population mated first (π), the significance of ‘male origin’, ‘female origin’ and the corresponding interaction was tested using logit analyses. A sequential Bonferroni correction for multiple testing is applied to the set of 12 P-values to conservatively control overall type-I error for these four analysis.

Neutral predictions for the inbreeding coefficients and fitness averages

In the RP population, assuming that all the eggs laid by any female were fertilized by a single male, and that each of the two females in the same vial contributed one-half of the female and male offspring, a randomly sampled female offspring would have a 0.5 probability of mating its full-brother and a 0.5 probability of mating a nonrelated male before being sampled as a breeding female for the next generation. Therefore, if inbreeding was just because of the RP mating scheme, the average inbreeding coefficient expected at generation t (FIt) would be half that expected from full-sib mating, i.e.


This inbreeding coefficient increases with generation number t, rapidly approaching 0.2 (e.g., F8 = 0.199). As a result of the fitness inbreeding depression, females mated to their brothers are expected to contribute fewer offspring than those mated to unrelated males, but the consequences of this regarding deleterious alleles can be ascribed to purge and is not considered in the present neutral prediction.

In any case, eqn (2) refers just to the inbreeding caused by departure from panmixia. The expected overall inbreeding coefficient can be computed as


where FNt is the inbreeding coefficient expected at generation t from the effective population number Ne, that must be computed from the number of potential breeders per generation (= four times the number of vials) taking into account the reduction in effective population size caused by nonrandom mating.

To obtain an approximate estimate of the effective population size Ne, we use the expression inline image where inline image is the variance of family contributions (Caballero & Hill, 1992). Given the breeding structure of our experimental design, the number of breeding individuals (k) contributed per family to the next generation is constrained to be, at most, four individuals. Therefore, in the case of no differences in fertility, family size is binomial with parameter n = 4, P = 1/2. This gives inline image and Net = 4N/(3 + FIt)), where N is the number of potentially breeding individuals, so that Net is slightly larger in the control population than under RP. Note that, for inline image RP would lead to Ne larger than panmixia. However, inline image is a lower bound, because of differences in fertility from environmental and genetic origin, so that we use the conservatively high value inline image, giving


which can be expected to give a lower bound for the actual effective population size.

Then, FNt was recurrently computed as


and it was used in eqn 3 to compute the overall inbreeding coefficient. The same procedure was followed to compute the average inbreeding coefficient in the control population using FIt = 0. After a single generation of panmixia, inbreeding from RP (FIt) is destroyed, and only inbreeding attributed to finite effective population number (Ne) should remain.

Note that fertility differences will induce a positive covariance between the number of male and female offspring of an individual, so that the average probability that a random female mates a brother in the RP population will be larger than one-half and, therefore, the increase in FI will be larger than predicted by eqn 2. Although such increase will be destroyed by a single generation of panmixia, the corresponding purge will not. For this reason, our experiment is likely to test for the purging capacity of a rate of sib mating larger than ½.

For both populations, we compute neutral predictions for the evolution of the panmictic mean of our log-scaled fitness (i.e., measured after panmixia is restored in the case of the RP population) using the estimate of the initial δ and the inbreeding coefficient expected from to the effective population size, as


We also computed approximated predictions for this fitness decline taking into account the consequences of new mutation and natural selection, including purge, as proposed in García-Dorado (2008) using mutational parameter estimated for the same genetic background. We obtained very small predicted declines. As details are heavy and as even the neutral approach predicts fitness declines much smaller than observed, these predictions are not given.


For the control population, the inbreeding coefficient Ft, computed using the neutral approach (eqn 3), increased roughly linearly on generation number, reaching F60 = 0.14 by the end of the experiment. For the RP population, Ft was always larger than for the control population. The difference increased quickly during the first few generations and remained stable later, amounting 0.18 by generation 10 and 0.22 by generation 60. Table 1 gives the δ estimates and the corresponding panmictic and inbred averages assayed through the experiment for competitive fitness and for viability.

Table 1.   Estimates for δ obtained through the experiment, together with the corresponding panmictic and inbred means.
Generation numberTraitControl Panmictic meanControl inbred mean (= 0.25)Control δRP Panmictic meanRP Inbred mean (= 0.25)RP δ%δ reduction by RP
  1. Estimates for δ, with their corresponding standard errors and with the means for the log-scaled fitness (w) and for viability (v). For viability, δ is given relative to the viability panmictic average, but estimates from log mean viability are also given within brackets. P-values for one-tailed contrasts are given in italics. The Fisher’s pooled P-value for the three δ estimates gives < 0.0155.

0w−0.046 ± 0.082−0.863 ± 0.1083.27 ± 0.54    
34w−0.944 ± 0.044−1.349 ± 0.0471.62 ± 0.26−1.337 ± 0.042−1.563 ± 0.0570.90 ± 0.2844 ± 23 P < 0.031
49v (egg-to-adult)0.653 ± 0.0240.552 ± 0.0320.52 ± 0.18 (0.676)0.685 ± 0.0200.599 ± 0.0250.44 ± 0.18 (0.539)15 ± 50 P <0.382 (20.3%)
60v (egg-to-pupae)0.782 ± 0.0210.650 ± 0.0370.67 ± 0.20 (0.737)0.770 ± 0.0240.739 ± 0.0250.16 ± 0.19 (0.170)76 ± 41 P < 0.033 (77%)

The evolution of the average competitive fitness is shown in Fig. 2 for the control population and for RP after restored panmixia. Given the large δ value initially estimated for competitive fitness (δ = 3.27 ± 0.54 at generation = 0), a substantial decline is predicted using neutral theory (i.e., ignoring the effect of purge). The observed decline is much greater than even the neutral expectation (eqn 6). Furthermore, in the later evaluation, this decline is larger for the RP than for the control population, despite the additional purge in the former. Note that, as shown in the Material and Methods section, for inline image the reduction in effective population size caused by RP will be smaller than our prediction from eqn (4). All this considered, the observed genetic trend for reduced competitive fitness through the experiment, particularly under RP, cannot be explained by inbreeding depression alone, with or without purge.

Figure 2.

 Panmictic average fitness (inline image, log scaled) plotted against generation number. Thick lines are for observed averages, thin lines for the corresponding neutral expectations computed from eqn 6. Solid lines are for the restricted panmixia population and broken lines for the control population.

Mean values for the traits estimated at generation 48 are shown in Table 2. They are given for the different cases according to the origin of the wild-type male and female in the vial. The significance for the effects of male or female origin, or the corresponding interaction, is given in Table 3.

Table 2.   Means of the different traits assayed at generation 48.
Origin of wild-type matesaτπw1–4w6–9
  1. Means, according to the parental origins (C, control; RP, restricted panmixia), for the different fitness components: Time to first mating (τ); Proportion of vials where the first mating involves a wild female (π); Fitness from the emergence of days 1–4 (w1–4); Fitness from the emergence of days 6–9 (w6–9).

C Female-C Male3:24 ± 0:200.147 ± 0.061–1.09 ± 0.170.25 ± 0.18
C Female-RP Male2:08 ± 0:150.222 ± 0.080–0.95 ± 0.25–0.13 ± 0.18
RP Female-C Male3:09 ± 0:220.522 ± 0.104–0.54 ± 0.23–0.45 ± 0.26
RP Female-RP Male2:19 ± 0:190.522 ± 0.104–0.57 ± 0.32–0.40 ± 0.23
Table 3. P-values for the significance of the different sources of variation in the generation 48 assay.
Source of variationτπw1–4w6–9
  1. P-values from two-way anova for τ (time to first mating) and for w (w1–4 = Fitness from the emergence of days 1–4; w6–9 = Fitness from the emergence of days 6–9) and from Logit analysis for π (proportion of vials where the first mating involves a wild female). Bold type indicates P-values that are significant after a sequential Bonferroni correction is applied to the 12 P-values given in the table.

Male origin0.00230.61700.83020.4494
Female origin0.91840.00020.05400.0277
Male origin × female origin0.51720.57160.71680.3240

For time to first mating (τ), only male origin has a significant effect, males from the RP population mating first than those from the control. For the proportion of vials where the tested female mates first than the Cy/L2 female (π), only the origin of the female has a significant effect, females from the RP population mating first more often than those from the control.

Regarding early and late fitness, assayed at generation 48, RP females showed a significantly higher early fitness (w1–4) than control ones, and control population females showed higher late fitness (w6–9). The first result become nonsignificant after a sequential Bonferroni correction was applied to control the overall significance of the twelve contrasts in the table. However, this may be a too conservative correction, as it is being applied to P-values that, despite being summarized in a single table, test for different hypothesis performed on different data for different variables. In any case, the pattern of variation of fitness during the 9 days for which emergence was recorded, given in Fig. 3, consistently shows that the fitness of the RP population is higher than that of the control during the first half of the emergence period and becomes lower later on.

Figure 3.

 Difference between mean fitnesses for females from the restricted panmixia and control populations (inline image) computed from the adults emerged at different days (i). Vertical bars represent two standard errors (one up and one down) for each difference.

For both fitness and viability, δ shows a significant reduction after a RP period, which, on the overall, amounted about 40% the synchronous estimate in the control population. This reduction was significant for two of the three estimates as well as for the pooled analysis.

It should be noted that at generation 22 there were no significant differences between the average fitness measured in (i) the RP population showing the inbreeding attributed to the RP system, (ii) the individuals obtained through panmixia from the RP population, or (iii) the panmictic individuals from the control population (means −0.39 ± 0.04, −0.48 ± 0.05, −0.43 ± 0.04, respectively, P-value = 0.40 in a one-way anova). Considering the mean fitness (−0.43) obtained for RP after restoring panmixia, and the corresponding δ estimate (0.90 at generation 34, maybe larger at generation 22), the neutral prediction of mean fitness at generation 22 while panmixia is restricted (inducing an inbreeding coefficient about 0.2), is −0.43 – 0.9 × 0.2 = −0.61, i.e., 0.22 below the obtained value (−0.39).


The most important result of this experiment is the purge of the concealed load under RP, as shown by the corresponding reduction of the inbreeding depression rate (δ) for competitive fitness and for its viability component. For competitive fitness w, the large initial estimate (δ = 3.27 ± 0.54 at generation = 0) was of the order of those reported for natural populations of the same species (Latter & Mulley, 1995). After 34 generations it dropped down to 1.62 in the control population, and the drop was far more important in the RP population, where it was 56% the value synchronously obtained for the control. The same was observed for viability, for which the average δ in the RP population was about half that of the control population. Note that, for the RP population, δ estimates were obtained after panmixia restoration, so that they could be even smaller while panmixia was restricted.

The above result is in agreement with that obtained by Swindell & Bouzat (2006), who carried out a purge Drosophila experiment based on the subdivision of populations into five lines, each maintained with 10 mass-mated breeding pairs, which were outcrossed after 19 generations. In that experiment, the inbreeding coefficient attained because of the subdivision is at least three times our asymptotic value from partial induced inbreeding and, accordingly, they found a larger reduction of δ, to about 1/3 when panmixia was restored. Similarly, Fox et al. (2008) performed an experiment where subpopulations were maintained by serial full-sib inbreeding during 3–4 generations, and they found a similar δ drop after outcrossing.

Restoring panmixia caused no increase in the fitness average, despite causing a substantial reduction of the inbreeding coefficient, both compared to those measured while restricted panmixia is going on (generation 22 data). This implies that the expressed load induced by RP (i.e., the inbreeding depression incurred) is negligible and, therefore, that, under RP, purge reduces the probability of homozygosity for deleterious recessive alleles well below the neutral prediction. Therefore, our population maintained under RP should have survival prospects similar to those of the control panmictic one, but will show reduced risk of extinction under future bottlenecking, because of its lower δ value.

After panmixia is restored, the RP population could be expected to show a smaller expressed load than the control population, because of the purge induced by RP. However, even neutral predictions, based on Ft and on the initial estimate of δ, give a relatively small expected fitness decline in the control population, and the decline predicted becomes negligible after taking into account the slow purge caused by the finite population size (results not shown). Thus, no substantial advantage for panmictic fitness can be expected from purge under RP. In agreement with this expectation, similar viability means were obtained for the control population and the RP one after panmixia was restored by the end of the experiment.

It should be noted that a single control and a single RP population were maintained and analysed. For neutral genetic variation, or for quantitative traits that are loosely related to fitness, drift can cause important genetic differences in the long term, even between large populations. For fitness-related traits, however, natural selection is expected to efficiently halt genetic differentiation between populations with effective sizes on the order of those used in this experiment (Pérez-Figueroa et al., 2009). Therefore, our conclusions can be considered reliable regarding our base population, even in the absence of replication. However, the magnitude and genetic basis of δ varies between populations and species and may depend upon environmental conditions, and the generality of our results should be supported by additional experiments using different base populations. For example, in contrast with our results, the mean fitness of the population purged through subdivision in the experiment by Swindell & Bouzat (2006), discussed above, was substantially larger than that of the panmictic control, either under subdivision or after panmixia, which might be partly because of a more efficient purge, as larger inbreeding had been attained. However, the subdivided lines were maintained with 10 pairs of parents per bottle, while the control population was maintained with 25 pairs per bottle, and fitness was assayed in low competitive conditions. Thus, the fitness advantage of the subdivided population might also be partly attributed to adaptation to low competitive conditions. Similarly, Fox et al. (2008) found that, after outcrossing inbred lines, survival was higher than in an outcrossing control although selection between inbred lines seems to have been important in this case because of line extinction.

All this considered, the observed decay of the panmictic mean for competitive fitness, which is dramatic both in the control and in the RP population, cannot be interpreted as a consequence of inbreeding depression (with or without purge). Therefore, the decay should be associated with a different parallel genetic process or to some temporal environmental trend, although the latter explanation cannot account for the observation that, at the later stage of the experiment, the panmictic mean for competitive fitness became substantially lower in the RP population than in the control one. In the RP population, however, the average time to first mating for panmictic males became shorter and the competitive mating ability for panmictic females was increased, compared to the control. Furthermore, the panmictic average of competitive fitness of the RP population was higher than that for the control when measured early in life, but it was lower in later life stages. These observations indicate selection for precocity fitness components under the RP system compared to the control. The reason must be related with the fact that, under the RP breeding protocol, breeding females (females sampled after five emergence days) were chosen between those old enough has to have been inseminated and should have mated during the first four days after the beginning of emergence. Therefore, individuals contributing genes to the next generation are likely to be the offspring of the females that laid eggs sooner, and whose eggs developed faster into males or females that mated earlier. This selection upon precocity is imposed in low competitive conditions, with only two females breeding per vial, as was also the case in the fitness assay at generation 48. Therefore, the lower panmictic competitive fitness in the RP population at generation 34 is most likely due to the fact that the corresponding assay protocol gave smaller weight to precocity components, as individuals were allowed to mate during a long time and the offspring number contributed by wild of Cy/L2 females was recorded after full emergence. This means that the lower mean competitive fitness under RP can be a negatively correlated response to increased selection for precocity and suggests that the genes determining higher precocity cause reduced competitive ability when resources are limited. The control population could have also experienced some selection for precocity fitness components under the experimental protocol (compared to the previous mass breeding system) which, even being weaker than in the RP population, could account for a correlated negative response for competitive fitness. In any case, we are not directly concerned with the causes of the common overall fitness decline observed for both populations, which could also be attributed to some inadvertent environmental trend, but with the differences between them. The high efficiency of the adaptive process responsible for the lower panmictic competitive fitness of RP at generation 34 implies the existence of important additive variance for the corresponding adaptive traits, suggesting that they should be loosely related to fitness in the base population. Inbreeding depression, however, should be mainly ascribed to unconditional (partially) recessive deleterious alleles, so that it is unlikely to be affected by the adaptive process. Therefore, the main cause of its reduction should be the selective purge. Epistatic interactions could also occur between the loci effecting the traits considered, but the consequences upon the evolution of the population’s genetic parameter are expected to be irrelevant under the relatively slow inbreeding of our experiment (Rosa et al., 2005; Pérez-Figueroa et al., 2009; Crow, 2010).

For viability, however, we did not observe differences between the panmictic mean of the control and the RP populations. This is to be expected, as there seems to be no direct cause determining different selective pressures between both populations regarding this fitness trait.

In any case, our results uncover complex and to some extent unpredictable properties of the genetic architecture of fitness components. These properties imply that any breeding structure is likely to be associated with specific selective pressures for different fitness components. The same can be concluded regarding to captive breeding, which can induce dramatic changes in selective pressures, particularly when the breeding system is to be controlled. In this respect, attention should be paid to avoid inadvertent selection associated with controlled breeding structures, favouring precocity or prolificacy and leading to negatively correlated responses for competitive fitness in the wild.

Our results indicate that RP strategies can be useful in conservation, in particular when a population with reasonable effective number and reproductive potential is likely to undergo future bottlenecks because of the deterioration of the quality or the continuity of its habitat, or to the recurrence of epidemic events. In this case, the use of a RP strategy can be considered safe, because even an important proportion of inbred mating causes only a limited increase in the inbreeding coefficient and because purge largely reduces the expressed depression, even before panmixia is restored. However, for very small populations, purge can be inefficient to reduce the expressed inbreeding depression, and can induce a relevant reduction of the effective population size. Therefore, when few breeding individuals are available, additional inbreeding from RP might be associated with increased extinction rates, so that RP should not be recommended as a purging strategy for the conservation of critically endangered populations (Hedrick & Kalinowski, 2000).

From an evolutionary point of view, the fact that survival after bottlenecking should be larger for populations with a RP history implies that this breeding system could be more common for populations that have experienced bottlenecking in the past. In principle, if δ is small because of a period of size reduction, but the population recovers a large size, δ is expected to slowly increase up to the value corresponding to the new MSD balance, which is larger for larger populations (Bataillon & Kirkpatrick, 2000; García-Dorado, 2003, 2007; Amador et al., 2010). However, while δ is small, females that tolerate some proportion of closely related mates can be favoured, as the reduction (because of inbreeding depression) of their number of offspring can be outweighed by the larger proportion of copies of the female’s genes in those offspring. Thus, following this classical argument by Fisher (1941), mating under RP can in some circumstances (Kokko & Ots, 2006) confer an advantage for inclusive fitness, particularly after a long-lasting bottleneck. This would lead to the evolution of RP which, in turn, would purge the concealed load with no appreciable increase in the expressed load. Therefore, even if a large population number is recovered, the δ value expected at the new equilibrium could be smaller than that for the MSD balance of the panmictic population, which would reduce the extinction rate during future bottlenecks.

Our main conclusion is that, in a moderate size population, RP (i.e., partial inbreeding achieved without reducing the number of breeding individuals but imposing a relevant proportion of inbred mating) reduces the concealed load (measured as δ), while the expressed load undergoes no appreciable increase, and no relevant reduction after panmixia is restored. As a consequence, populations with a breeding structure implying RP can undergo a lower extinction risk after bottlenecking, which has important consequences both from the conservation and the evolutionary viewpoint.


We are grateful to Carlos López-Fanjul by helpful comments. This work was supported by grant CGL2008-02343/BOS from the Ministerio de Ciencia e Innovación (Spain).