Correlations between fitness and genome-wide heterozygosity (heterozygosity-fitness correlations, HFCs) have been reported across a wide range of taxa. The genetic basis of these correlations is controversial: do they arise from genome-wide inbreeding (“general effects”) or the “local effects” of overdominant loci acting in linkage disequilibrium with neutral loci? In an asexual thelytokous lineage of the Cape honey bee (Apis mellifera capensis), the effects of inbreeding have been homogenized across the population, making this an ideal system in which to detect overdominant loci, and to make inferences about the importance of overdominance on HFCs in general. Here we investigate the pattern of zygosity along two chromosomes in 42 workers from the clonal Cape honey bee population. On chromosome III (which contains the sex-locus, a gene that is homozygous-lethal) and chromosome IV we show that the pattern of zygosity is characterized by loss of heterozygosity in short regions followed by the telomeric restoration of heterozygosity. We infer that at least four selectively overdominant genes maintain heterozygosity on chromosome III and three on chromosome IV via local effects acting on neutral markers in linkage disequilibrium. We conclude that heterozygote advantage and local effects may be more common and evolutionarily significant than is generally appreciated.

Population-level correlations between fitness and heterozygosity (hereafter heterozygosity-fitness correlations, HFCs) are reported across a wide range of taxa (Britten 1996; Chapman et al. 2009). Heterozygosity-fitness correlations most likely arise via two nonexclusive processes. First, HFCs might arise from “general effects” (David et al. 1995). Under a general effects model it is assumed that individuals vary in their genome-wide heterozygosity at neutral markers as a result of variation in levels of inbreeding across a population. Variation in levels of inbreeding is correlated with heterozygosity at nonneutral genes so that individuals that are more heterozygous at neutral markers have higher fitness. Second, HFCs might arise from “local effects” (David et al. 1995). Under a local effects model, it is assumed that some neutral markers are in linkage disequilibrium (LD) with genes that have a strong heterozygote advantage. Thus, the contribution of local effects to HFCs is dependent on the frequency of loci that show heterozygote advantage.

Local effects have fallen out of favor as the most likely explanation for HFCs (Chapman et al. 2009; Szulkin et al. 2010). Szulkin et al. (2010) clarify that the question is not so much if local effects exist, but whether there are circumstances in which local effects are sufficiently strong and frequent to explain HFCs. Szulkin et al. (2010) argue that in most systems, local effects are unlikely to be detected, as the correlation between markers in LD with genes under selection for heterozygosity is diluted by the much larger proportion of markers that are not in disequilibrium with such loci. However, Szulkin et al. (2010) note that the contribution of local effects to HFCs may currently be underappreciated because of the difficulty in empirically untangling the relative importance of general and local effects (but see, e.g., Harrison et al. 2011; Voegeli et al. 2012; Wetzel et al. 2012). Here we will argue that a thelytokous clonal lineage of honey bee, Apis mellifera, provides an ideal “natural laboratory” for documenting the contribution of local effects of genes that show heterozygote advantage, with a minimum of interference from the general effects of inbreeding.

Apis mellifera capensis (hereafter Capensis) is a subspecies of honey bee from the Western Cape of South Africa. Capensis is unique in being the only bee in which thelytokous reproduction is common. An asexual thelytokous lineage of Capensis workers presently parasitizes the commercial population of the African honey bee (A. m. scutellata) that is found throughout northern South Africa (reviewed in Beekman et al. 2008). This parasitic lineage, hereafter the “Clone”, arose in 1990 from a single Capensis worker derived from the sexual Capensis population (Neumann et al. 2010; Oldroyd et al. 2011; Goudie et al. 2012) after commercial Capensis colonies were transported across the hybrid zone that exists between the two subspecies (Beekman et al. 2008). The contemporary Clone population has flourished via asexual reproduction for over 20 years and more than 100 generations as a transmissible “social cancer” (Oldroyd 2002), destroying thousands of commercial A. m. scutellata colonies each year.

Worker reproduction in Capensis is characterized by thelytokous parthenogenesis with central fusion (Verma and Ruttner 1983). In this form of parthenogenesis meiosis ends normally with four haploid pronuclei. Zygosity is then restored, not by fertilization, but by the fusion of the two central pronuclei (Suomalainen et al. 1987). Because the fusion is central, the pronuclei involved are descended from the two alternate products of meiosis I (see Fig. 1 in Rabeling and Kronauer 2013). Thus, in the absence of meiotic recombination, the thelytokous offspring of a Capensis female are diploid clones of their mother. However, wherever recombination exchanges genetic material between chromosomes, there is a one third chance that a locus that is heterozygous in the mother will become homozygous in offspring (Pearcy et al. 2006; Oldroyd et al. 2008; Engelstadter et al. 2010). Therefore, ongoing generations of thelytoky should result in population-wide homozygosity at all loci that are free to recombine (Goudie et al. 2012).

Figure 1.

Correlation between cumulative map distances measured in sexually produced Capensis drones and those previously reported by Solingac et al. (2007) in a progeny of obligately sexual Apis mellifera ligustica × Apis mellifera carnica, on chromosomes III and IV.

Despite the predicted loss of heterozygosity arising from ongoing thelytoky, empirical studies have revealed remarkably high levels of heterozygosity in the Clone (Baudry et al. 2004; Neumann et al. 2010; Oldroyd et al. 2011). High levels of heterozygosity in the Clone have often been attributed to a reduction in meiotic recombination (Moritz and Haberl 1994; Baudry et al. 2004). However, Goudie et al. (2012) demonstrate that a reduction in recombination is insufficient to explain current levels of heterozygosity. This is because loss of heterozygosity in an asexual lineage is ratcheted. Heterozygous mothers produce homozygous daughters at one third the rate of recombination (r), whereas homozygous mothers produce homozygous offspring exclusively (Engelstadter et al. 2010; Goudie et al. 2012). Thus, for any value of r > 0, homozygosity will inevitably accumulate in a thelytokous lineage. After over 20 years of exclusively thelytokous reproduction, reduction in recombination cannot explain the maintenance of heterozygosity in the Clone at any but the most centromeric loci where recombination is exceedingly rare (Goudie et al. 2012).

Maintenance of heterozygosity in the Clone can instead be explained by selection against homozygous recombinants at genes that are subject to heterozygote advantage (Oldroyd et al. 2011; Goudie et al. 2012). In particular, loss of heterozygosity at the complementary sex-determining locus (csd) in honey bees is lethal because homozygosity at this locus results in an inviable diploid male (Woyke 1963; Beye et al. 2003). In support of this hypothesis, Goudie et al. (2012) showed that indeed, recombination produces diploid males that are homozygous at csd, but that these recombinants are rapidly removed from the population, permanently retaining heterozygosity at csd. Goudie et al. (2012) further found that homozygosity at neutral microsatellite markers across the Clone genome was correlated with reduced survival between egg and larval/pupal life stages; a strong, if unconventional, HFC. Thus, it is selection that maintains heterozygosity at the csd, and selection combined with local effects that explain maintenance of heterozygosity at neutral markers in proximity to the csd (Oldroyd et al. 2011).

It is clear that selection maintains heterozygosity in perpetuity at the homozygous-lethal csd. The question now becomes: how is heterozygosity maintained at neutral microsatellite markers throughout the Clone genome? Goudie et al. (2012) suggested that microsatellite markers might be linked to unidentified overdominant loci, much like the csd, on which selection acts to maintain heterozygosity. If so, local effects may play a major role in maintaining heterozygosity in the Clone genome, and provide a prime example of the evolutionary significance of local effects and overdominance in the generation of HFCs.

Using contemporary Clone workers we performed chromosome walks on chromosomes III and IV and mapped the pattern of zygosity along these chromosomes at microsatellite loci that we assumed to be selectively neutral. In doing so we identified the location of loci that have heterozygote advantage and which maintain heterozygosity at flanking neutral markers via local effects. The csd, a locus under well-characterized selection for heterozygosity, is located on chromosome III. Chromosome IV is a chromosome of similar length to III, but lacks the csd. Thus, chromosome IV acts as a control chromosome, allowing us to identify whether patterns of zygosity similar to those observed around csd are also seen on a chromosome that does not contain the csd.

Materials and Methods

We studied 42 Clone workers, a subset of those identified by Oldroyd et al. (2011). We further studied a sample of 96 Capensis drones, the progeny of a queen from the sexual population at Cape Point, well within the natural Capensis range (Beekman et al. 2008), which we used to determine rates of recombination in normal sexual meiosis of Capensis. Finally, we genotyped a sample of 96 Capensis workers, each collected from a different colony across the natural Capensis range (Holmes et al. 2010) which we used to determine allele frequencies in the contemporary Capensis population.

DNA was extracted from all samples using a high salt extraction method (Aljanabi and Martinez 1997; Holmes et al. 2010). On chromosome III, we genotyped bees at csd Exon 7 (Oldroyd et al. 2011) and at 68 microsatellite loci in Clones, 59 microsatellite loci in drones and 58 microsatellite loci in workers. On chromosome IV, we genotyped bees at 29 loci in Clones and 26 loci in drones and workers (Table S1). DNA was amplified using polymerase chain reaction (PCR) as described in Oxley et al. (2008, 2010).

We determined the chromosomal haplotypes of the mother of our drone sample indirectly from her drone progeny. When the mother was determined to be homozygous at a marker locus (i.e., all drone progeny carried just one allele), the locus was considered uninformative and was not included in further analysis. Based on the genetic map of Solignac et al. (2007), we then identified pairs of linked markers along each chromosome.

For each sequential pair of informative linked markers, we determined the two maternal haplotypes as being the two pairs of alleles that co-occurred at the highest frequency in the brother haploid drones. After we had determined the two maternal haplotypes for both chromosomes, we identified the recombinant haplotypes in the drone progeny. The frequency of recombination between each pair of markers was then used to calculate their genetic distance and the cumulative map distance in cM of all markers along the two chromosomes. We then compared these distances with the genetic distances reported by Solignac et al. (2007) in the progeny of a non-Capensis queen.

The genotype of the Clone's ancestor was, when possible, inferred from the contemporary Clone individuals via parsimony (Oldroyd et al. 2011). The Clone ancestor was inferred to have been heterozygous at a marker whenever two alleles were identified across the Clone cohort, either in heterozygous individuals or in two different homozygous individuals. These markers are thus considered informative. When only one Clone allele was identified across the Clone cohort, we were not able to distinguish between two possibilities: (1) the Clone ancestor was heterozygous, and heterozygosity has been lost in the current Clonal population due either to selection or genetic drift; (2) the Clone ancestor was homozygous. For these loci (n = 46), we used allele frequencies in the current sexual Capensis population (n = 96) to calculate the expected heterozygosity in the ancestral clone worker, assuming it had heterozygosity typical of the sexual population at that time.

We classified each locus in each contemporary Clone worker as being either heterozygous (1) or homozygous (0). We then calculated the frequency of heterozygosity as the average of the values across our Clone cohort, such that 0 indicates complete loss of ancestral heterozygosity in all individuals and 1 indicates maintenance of heterozygosity in all individuals studied.

We plotted observed heterozygosity against cumulative map distance (Solignac et al. 2007) for each informative marker on chromosomes III and IV. We additionally plotted heterozygosity in pooled Clones against cumulative genetic distance measured in our drone population. When the map distance for a marker could not be determined in drones (i.e., when a marker was homozygous in the mother and was therefore uninformative), we used the distance of Solingac et al. (2007) adjusted so that the distance between flanking loci equaled that estimated from our drone cohort.

The expected frequency of heterozygosity in a thelytokous population at generation n can be written as follows:

math image(1)

where LoH is the probability that locus A, which is heterozygous in a mother, will become homozygous in her offspring following recombination. LoH was derived from a given frequency of recombination using the Rizet and Engelmann (1949) mapping function DRE = −(2/3)ln(1−3LoH). Following Goudie et al. (2012), we computed the expected heterozygosity for each marker along the chromosomes after 20 (t20) and 100 (t100) generations of thelytoky as a function of cumulative map distance from the two centromeres. Expected heterozygosity was calculated at t20 and t100 based on three assumptions: (1) Solignac et al. (2007) map distances; (2) the empirically derived map distances measured in our sexually produced drone cohort, and (3) assuming a 10-fold reduction in recombination (Baudry et al. 2004) from the Solignac et al. (2007) map distance. This allowed us to assess the expected frequency of heterozygosity at all our marker loci.


Of the 69 markers genotyped in Clones on chromosome III, 37 were confirmed ancestrally heterozygous (i.e., multiallelic in contemporary Clones). Of the 29 markers on chromosome IV, 15 were confirmed ancestrally heterozygous (Table S1). We successfully amplified a total of 2207 (chromosome III) and 976 (chromosome IV) locus/individual combinations. Loss of heterozygosity with respect to the inferred common ancestor was identified in 101 (4.6%) of all locus/individual combinations on chromosome III, and 119 (12.2%) on chromosome IV, indicating that a significantly higher proportion of loci had lost heterozygosity on chromosome IV than on chromosome III (PFisher < 0.001).


Assuming that the Clone's ancestor was drawn from the sexual Cape population, we can calculate its expected heterozygosity for all loci (Table S2). Expected heterozygosity in the ancestral Clone worker averaged 0.80 for the loci on chromosome III and 0.76 on chromosome IV (Fig. 2).

Figure 2.

The observed frequency of heterozygosity in pooled Clones at each informative marker (solid line) along chromosomes III and IV, relative to the genetic distances reported by Solignac et al. (2007). Observed heterozygosity is compared to average heterozygosity in the sexual Cape population (dashed horizontal line), which is expected to approximate the heterozygosity of the ancestral Clone worker. Observed heterozygosity is also compared to predicted heterozygosity in the absence of selection (Goudie et al. 2012) after 20 (t20) and 100 (t100) generations of thelytoky in both the presence and absence of a 10-fold reduction in recombination (Baudry et al. 2004). Markers that are homozygous for a single allele in all Clones (uninformative and possibly ancestrally homozygous) are indicated by vertical dashed lines. Markers for which two alleles were identified (in either heterozygous Clones or two different homozygotes genotypes) are further indicated by grey marks at the top of the figure.


Our drone cohort was informative (i.e., maternally heterozygous) at 40 of 60 markers on chromosome III and 19 of 26 markers on chromosome IV. The cumulative genetic distance of chromosome III in drones was 275.4 cM, compared to 287.5 cM reported by Solignac et al. (2007). On average, the map distance between informative markers on chromosome III was 0.3 cM (SD ± 5.9) shorter in our drone cohort than was measured by Solignac et al. (2007). Genetic recombination in our drone cohort was lower than expected based on Solignac et al. (2007) along the majority of chromosome III, but increased markedly toward the telomere (Fig. 1). This indicates that overall rates of recombination on chromosome III were not dissimilar in the normal sexual meiosis of our Capensis queen and those observed by Solingac et al. (2007) in a progeny of obligately sexual Apis mellifera ligustica × Apis mellifera carnica cross. In contrast, on chromosome IV, the distance between informative markers was 5.2 cM ± 5.9 less in our drone cohort than previously measured by Solignac et al. (2007). The cumulative genetic distance of chromosome IV in drones was 205.3 cM, compared to 299.3 cM reported by Solignac et al. (2007). This indicates that the frequency of recombination on chromosome IV is lower in the meiosis of Capensis relative to that observed by Solignac et al. (2007).


We hypothesized that heterozygosity is maintained in the Clone population by linkage of neutral markers to genes that have heterozygote advantage. Under this hypothesis we predicted that along a chromosome there would be clusters of heterozygous markers that are linked to an overdominant locus. These heterozygous regions would be interspersed with regions of complete homozygosity where heterozygosity has been lost by recombination without protection by linkage to an overdominant gene. A runs test (Zar 1996) showed no evidence of clusters of heterozygous loci on either chromosome III (Z = 0.42, P = 0.68) or IV (Z = 0.75, P = 0.45). However, inspection of Figure 2 suggests that the expected pattern is obscured by loci that are noninformative as the Clone population is now monoallelic.

For all loci that are informative, the Clone retains remarkably high levels of heterozygosity along the length of chromosomes III and IV (Fig. 2). The average individual heterozygosity at informative markers was 0.91 on chromosome III and 0.78 on chromosome IV, suggesting that the frequency at which heterozygosity is maintained in the Clone lineage is significantly higher on chromosome III than it is on IV (math formula = 75.6, P < 0.0001). If contiguous homozygous markers are regarded as linkage groups (single data points), assumed to have resulted from single loss of heterozygosity event, this difference remains significant (math formula = 27.46, P < 0.0001).

When observed heterozygosity is plotted against the Solignac et al. (2007) map distances (Fig. 2), heterozygosity was consistently higher along both chromosomes than would be predicted under even our most conservative assumptions about rates of recombination (eq. (1)). After 20 generations with a 10-fold reduction in recombination, equation (1) shows that we expect a 50% reduction in heterozygosity for loci located 50 cM from the acrocentric centromere (Fig. 2). Equation (1) shows that toward the telomere, virtually all heterozygosity should be lost by generation 20 (Fig. 2). If there is no reduction in recombination, all heterozygosity should be lost at loci more than 80 cM from the centromere after just 20 generations (Fig. 2). However, we estimate that more than 20 years after the Clone's emergence, and conservatively assuming five generations a year, the Clone has undergone more than 100 generations of thelytokous reproduction since its last sexual ancestor. Under the more realistic assumption of 100 generations, and an assumed 10-fold reduction in recombination (Baudry et al. 2004), we expect virtually all heterozygosity to be lost at loci more than 40 cM from the centromere, and dramatically reduced at loci more than 20 cM from the centromere (Fig. 2). If there is no reduction in recombination, heterozygosity should be entirely lost at loci more than 5 cM from the centromere after 100 generations of thelytoky. Thus, there is strong evidence that in addition to csd, overdominant loci are maintaining heterozygosity on chromosomes III and IV.

When observed heterozygosity is modified to account for the observed reduction in rates of recombination in our sexually produced Drone sample relative to Solignac et al. (2007) (Fig. 3), our conclusions about predicted loss of heterozygosity, above, are unchanged: observed heterozygosity is higher than predicted in the absence of selection along the length of both chromosomes. This conclusion is also unchanged if map distances are corrected using Kosambi's (1944) or Haldane's (1919) mapping functions to account for double crossovers (data not shown).

On chromosome III there is no apparent relationship between heterozygosity and distance from the centromere (Fig. 2). High heterozygosity is maintained at even the most distant markers from the centromere. Interestingly, however, there are two markers (6844 and UN125) that have lost ancestral heterozygosity in current Clones, with subsequent restoration of heterozygosity at closely linked telomeric markers (K0314 and SV196). Such an architecture can only arise from a simultaneous double recombination event in which heterozygosity, lost at the first crossover, is restored by the second (Oldroyd et al. 2011).

Chromosome IV shows a stronger association between distance from the centromere and loss of heterozygosity (Fig. 2), but heterozygosity is still higher than predicted in the absence of selection for heterozygotes (eq. (1), Fig. 2). As with chromosome III, complete loss of heterozygosity is observed at marker K0457, followed by its restoration at closely linked marker UN161 which is slightly telomeric. It is particularly interesting to note that K0457 is only 31.8 cM from the centromere. Therefore, heterozygosity has been lost due to recombination in close proximity to the centromere, but restored by a second, simultaneous crossover before UN161. This appears to be the signature of a local effect of a gene under selection for heterozygosity near UN161.


The pattern of zygosity along chromosomes III and IV in the contemporary Clone population shows that heterozygosity is maintained via local effects imposed by selectively overdominant loci. Complete loss of heterozygosity occurs in restricted regions, with subsequent restoration of heterozygosity in telomeric regions. We suggest that there are at least three overdominant genes that are maintaining heterozygosity at linked neutral loci on chromosome IV, and four overdominant genes (including the known csd locus) that are maintaining heterozygosity on chromosome III. This is, however, the most conservative interpretation of our data. It is likely that additional nonneutral genes maintain heterozygosity on both of these chromosomes and indeed on other chromosomes in the genome.

Our Clone cohort was monoallelic at 47% of the markers examined. However, allele frequencies observed in the contemporary sexual Capensis population provide evidence that an individual emerging from this population to found a clonal lineage would be heterozygous at the majority of these markers. We thus conclude that homozygosity at many monoallelic markers in the contemporary Clone population is a result of directional selection acting on additive loci to purge deleterious alleles (Goudie et al. 2012) or genetic drift.

In Figure 4, we present a descriptive model for the maintenance of heterozygosity on chromosomes III and IV. On chromosome III, we propose that heterozygosity at neutral markers in Region 1 is maintained by linkage to the centromere and at least one proposed selectively overdominant gene AIII. In Region 2, we observe a breakdown in gametic disequilibrium between AIII and neutral marker 6844. In Region 3, heterozygosity is restored at marker K0314 by its linkage to a proposed overdominant gene BIII. Heterozygosity is maintained in Region 3 by linkage to BIII and csd. In Region 4, we again observe a dramatic loss of heterozygosity caused by a breakdown of gametic disequilibrium between csd and UN152. Finally, in Region 4, the proposed gene CIII maintains heterozygosity even at the most telomeric loci.

Figure 3.

The observed frequency of heterozygosity in pooled Clones at each informative marker along chromosomes III and IV, plotted against genetic distances reported by Solignac et al. (2007) (solid line) and empirically derived map distances measured in our sexually produced drone cohort (dashed line). Observed heterozygosity is compared to predicted heterozygosity in the absence of selection (Goudie et al. 2012) after 20 (t20) and 100 (t100) generations of thelytoky in both the presence and absence of a 10-fold reduction in recombination (Baudry et al. 2004).

Figure 4.

A descriptive model for the maintenance of heterozygosity along chromosomes III and IV. The observed frequency of heterozygosity in pooled Clones at each informative marker (black line) is plotted against the genetic distances reported by Solignac et al. (2007). The frequency of heterozygosity ranges from complete maintenance (1 on the Y-axis) to complete loss (0 on the Y-axis). Chromosomal regions are designated by numbered and shaded areas. In each region, the frequency of heterozygosity is driven by the interaction of selection and recombination acting at key putative and one known overdominant genes. Microsatellite markers of particular interest are annotated.

In Region 1 of chromosome IV (Fig. 4), heterozygosity is maintained by linkage to the centromere. In Region 2, heterozygosity is lost in 95% of individuals at marker K0457 demonstrating a breakdown in gametic disequilibrium less than 32 cM from the centromere. Heterozygosity is then restored at marker UN161 in Region 3 by linkage to proposed selectively overdominant gene AIV. Heterozygosity is retained throughout Region 3 by linkage to gene AIV and a second gene BIV. In Region 4, we observe a gradual decline in heterozygosity, resulting in moderate levels at the most telomeric of informative markers (UN303, where 45% of Clones sampled maintain heterozygosity).

Why, when heterozygosity is maintained, is it often maintained at a frequency less than 100%? (see Fig. 1 where many markers that retain heterozygosity do so at a frequency less than 1, for example marker K0353 on chromosome III and K0422 on chromosome IV). Heterozygous mothers produce homozygous offspring at one third the rate of recombination (Pearcy et al. 2006; Oldroyd et al. 2008; Engelstadter et al. 2010) and homozygous mothers go on to produce homozygous offspring exclusively. Therefore, in the absence of selection, we predict that heterozygosity would be either completely lost (in the case of loci that are free to recombine) or maintained at 100% (in the case of non-recombinant loci), with no intermediate frequency at equilibrium (Engelstadter et al. 2010; Goudie et al. 2012). In contrast, an equilibrium where there are both heterozygous and homozygous individuals in the population can be reached by the interplay of selection with recombination (Goudie et al. 2012). When the number of homozygous offspring removed from the population by selection at each generation is equal to or greater than the number produced by recombination, heterozygosity will reach an equilibrium greater than 0 but less than 1.

Marker loci that are tightly linked to a gene under selection for heterozygosity will behave identically to the gene under selection. That is, heterozygosity will be maintained at the marker at the same frequency as the selected gene. When linkage is relaxed, so that recombination occurs between marker and gene, marker loci can become homozygous while selected genes maintain heterozygosity, that is a breakdown in gametic disequilibrium between the marker and gene can occur. However, we show below that this process is expected to be slow. Consider a neutral heterozygous marker m, linked to a gene A that must be heterozygous A1A2. Given an individual with the genotype: math formula with a map distance of d cM between A and m. The proportion of offspring in which heterozygosity is lost (L) between A and m (i.e., in which an offspring becomes homozygous at m while maintaining heterozygosity at A) is determined by the equation:

display math

Thus, if m is located 10 cM from A, heterozygosity will be lost between A and m in 4.6% of offspring each generation. If we further consider that A is freely recombining with the centromere, then one third of all offspring will lose heterozygosity at A, regardless of recombination between A and m. Thus, the combined probability of individuals being produced that have lost heterozygosity at m while escaping selection at A is 3.1% each generation. However, in future generations heterozygosity will still be lost at A in one third of these individuals, and so they will continue to be subject to selection that will not be detected at marker m. Therefore, the population-wide frequency of homozygosity at neutral markers will increase over time. For any marker closely linked to a gene under selection, the proportion of homozygous individuals produced each generation is sufficiently small for heterozygosity to have been maintained at neutral markers over the short evolutionary history of the Clone.

The probability of a recombination event between a locus and its centromere increases with distance from the centromere. Loss of heterozygosity at a given locus will affect all loci in a telomeric direction unless heterozygosity is restored by a second simultaneous “rescue” recombination event (Oldroyd et al. 2011). It is important to note that the likelihood of a second recombination event decreases as genome-wide rates of recombination are decreased. Therefore, although reduced rates of recombination decrease the likelihood of a loss-of-heterozygosity event occurring, it increases the number of loci affected when such an event occurs. If heterozygosity were solely determined by rates of recombination, there would be a strong correlation between distance from the centromere and the frequency of heterozygosity. This effect is not observed (Fig. 4).

If recombination is rare, so that double recombinants are extremely rare, then we would observe a binary pattern in the genotypes of individual Clones, where a switch in zygosity occurs once along each chromosome at the site of an ancestral recombination event, with complete heterozygosity toward the centromere and complete homozygosity toward the telomere. The probability of this switch occurring increases with distance from the centromere. However, as Clones evolve, rare centromeric recombination events would drive this switch point closer to the centromere. Of two chromosomes in each of 42 Clones studied, not one displays this predicted single switch to homozygosity extending to the telomere (Fig. S1). Twenty-one Clones show a switch to homozygosity on chromosome IV that extends to the telomere. However, each of these individuals has also lost heterozygosity at more centromeric loci, with subsequent restoration of heterozygosity that can only be explained by a double recombination event (Oldroyd et al. 2011). Thus, we conclude that the Clone genome has been shaped by frequent double recombination events. For any reasonable level of recombination in the Clone (reduced or otherwise), the likelihood of observing such a clear signature of double recombination is low, unless selection has led to an overrepresentation of double recombinant genotypes. We suggest that selection acts strongly against any individual in which a single recombination event results in loss of heterozygosity extending to the end of the chromosome. Thus, the individuals that lose heterozygosity at some point along the chromosome, but survive selection to be represented in the current population, will be those that carry doubly recombinant genotypes.

Negative interference (an increased probability of a second recombination event in proximity to a first) would increase the probability of double recombination events in the Clone, and so of maintenance of heterozygosity at selected loci that are flanked by markers that have lost heterozygosity. However, the phenomenon negative interference is rarely reported, and is neither necessary nor sufficient to explain the current Clone genotypes. Maintenance of heterozygosity at a flanking locus is not the default outcome of a double recombination event, and so selection is still required to conserve those double recombination events that result in the advantageous outcome. Furthermore, recombination events that result in loss of heterozygosity extending to the telomere of the chromosome are still predicted under a model of negative interference, and it is only when these are selected against that we would predict the pattern of zygosity we observe in the current Clone population.

Although our modeling demonstrates that overdominance is required to maintain heterozygosity under thelytoky, there are other mechanisms that result in effective overdominance at loci that are not subject to an inherent heterozygote advantage. First, effective overdominance can arise when two linked additive genes have deleterious alleles in trans (i.e., on different homologous chromosomes). In this scenario, a recombination event between the loci and the centromere that results in loss of heterozygosity affecting both genes will always result in homozygosity for the deleterious recessive allele at one or other of the two loci. Provided that selection against homozygosity for the deleterious recessive allele is sufficiently strong at both loci, such selection could maintain heterozygosity at both additive loci, without requiring heterozygote advantage at either locus. Thus, the linked pair of genes will act as an overdominant locus, until such time as recombination between them breaks down LD and allows the deleterious recessive alleles to be purged.

Second, and more speculatively, unlike Drosophila the honey bee has a fully-functional DNA methylation system (Schaefer and Lyko 2007; Foret et al. 2009), providing the possibility of genomic imprinting and parent-specific allele expression (Drewell et al. 2012). If parent-specific allele expression exists in honey bee workers then the Clone most likely maintains its ancestral methylation pattern; otherwise all genes would be expressed based on the maternal imprinting pattern, which would result in incorrect gene dosage for any locus that normally shows parent-of-origin gene expression. Recombination could result in homozygosity of either the paternally or maternally imprinted allele of the ancestral ancestor, which could be lethal. Thus, heterozygosity may be maintained as an artifact of genomic imprinting.

We have shown here that on chromosomes III and IV of the Clone there are at least seven overdominant genes (or gene combinations) that retain heterozygosity at linked neutral markers via local effects. In addition, we have tabulated from the literature all known heterozygous microsatellite markers identified thus far in the Clone genome (Table 1). These markers must be in LD with overdominant genes. Despite the fact that there has been no systematic search for heterozygous loci across the Clone genome, heterozygous loci have already been observed on nine of the 16 honey bee chromosomes (Table 1). We conclude that most, and most likely all, honey bee chromosomes have multiple selectively overdominant loci.

Table 1. Heterozygous loci previously observed across the Clone's genome, tabulated from the literature
IA43(Moritz et al. 2008)
IK0161-UN162-SV042-SV219-K0170(Goudie et al. 2012)
IMultiple Am series markers(Baudry et al. 2004)
IIMultiple Am series markers(Baudry et al. 2004)
IIIcsd(Oldroyd et al. 2011)
IIIA023(Oldroyd et al. 2011)
IIIAT089(Oldroyd et al. 2011)
IIIAP059(Oldroyd et al. 2011)
IIIUN157T(Oldroyd et al. 2011)
IIISV151(Oldroyd et al. 2011)
IIIMultiple Am series markers(Baudry et al. 2004)
VIA113(Härtel et al. 2006; Neumann et al. 2010; Goudie et al. 2012)
VIIA107(Moritz et al. 2008; Neumann et al. 2010; Goudie et al. 2012)
VIIA24(Moritz et al. 2008)
VIIIA14(Härtel et al. 2006; Neumann et al. 2010; Goudie et al. 2012)
XIIIB124(Härtel et al. 2006; Goudie et al. 2012)
XIIIThel1-Thel2-Thel3-Thel4(Shaibi et al. 2008; Oldroyd et al. 2011)
XIVA29(Goudie et al. 2012)

It is generally held that loci which show heterozygote advantage are rare (Hedrick 2012), although this view is controversial. For example, in Drosophila melanogaster, when outbred individuals were generated by crossing highly inbred lines, about 25% of all genes showed expression patterns suggesting overdominance (Ayroles et al. 2009). In contrast, few loci remained heterozygous after the formation of highly inbred lines from an outbred population of Drosophila (Mackay et al. 2012), suggesting that selection for heterozygote advantage was insufficient to counteract the effects of inbreeding. However, it should be noted that the genetic architecture of the honey bee differs substantially from that of Drosophila. The breeding system of the honey bee promotes outbreeding (Winston 1987), whereas Drosophila populations are subject to frequent genetic bottlenecks (reviewed in Nei et al. 1975) and so overdominant loci cannot be relied upon to contribute to fitness. However, in haplo-diploid honey bees, overdominance can only occur in female-limited loci (Clarke et al. 1992; Hedrick and Parker 1997).

Honey bees have rates of recombination that are approximately four times higher than most other taxa (Beye et al. 2006). Indeed, eusocial Hymenoptera, including the honey bee, have the highest recombination rate of all multicellular animals (Wilfert et al. 2007). Kent et al. (2012) propose that high rates of recombination in the honey bee coevolved with eusociality. They argue that the transition to eusociality was associated with low effective population size, high LD, and high genetic load. Therefore, selection favored high rates of recombination as a means to reduce interference and enhance natural selection, facilitating the optimization of caste phenotypes. Similarly in an asexual linage such as the Clone, high rates of recombination could facilitate the purging of deleterious alleles (Goudie et al. 2012), while allowing heterozygosity to be maintained at overdominant loci via the selection of multiply-recombinant offspring. Thus, a Clone that produces large numbers of recombinant gametes would benefit from the potential to evolutionarily “refine” its genome, particularly in the early stages of its emergence.

The Clone's parasitic life history is characterized by high rates of reproduction, low maternal investment, and concordantly high rates of mortality (Goudie et al. 2012; i.e., many highly expendable offspring). Therefore, the Clone would be well suited to absorbing the genetic load imposed by loss of heterozygosity at overdominant loci, while benefiting from the high rate of recombination. Furthermore, competition between Clones limits the proportion of offspring that can ever hope to become reproductively active. During the course of Clone invasion, Clones will lay dozens of eggs inside brood cells that can only support one. Once Clones are present in a colony, they quickly establish a dominance hierarchy (Martin et al. 2002; Härtel et al. 2011), so that only a small number of Clones become reproductively active “pseudoqueens”. The remaining Clones do not reproduce, neither do they engage in any work such a foraging or brood care (Martin et al. 2002), they simply exist as an ongoing burden to the colony. For the reproductive Clone, their only evolutionary concern is that at least one egg per brood cell (out of often dozens) is viable, and that enough offspring emerge from those brood cells to fill the reproductive ranks necessary to take over the host colony. In fact, a large excess of nonreproductive Clones may only serve hasten the collapse of the colony, an event which is not beneficial to the parasites it contains (Härtel et al. 2011). Thus, during the emergence of the Clone from a background of millions of worker genotypes transported into the Scutellata range, selection would be predicted to favor a linage with rates of recombination that produce sufficient numbers of fit and highly virulent parasites, regardless of whether this also results in high rates of offspring mortality.

Finally, under a scenario of “low” rates of recombination, we would predict the regular occurrence of a single recombination event per chromosome (particularly because we know that recombination occurs in the Clone, and so the rate of recombination must be greater than zero). When only a single recombination event occurs, loss of heterozygosity will affect all telomeric loci in 50% off offspring. In contrast, multiple recombination events randomize alleles to pronuclei, resulting in only a one third chance of loss of heterozygosity, and providing the opportunity for heterozygosity to be “rescued” by additional crossovers.

Although we do not present here empirical evidence any for the rate of recombination in the Clone (before selection against homozygotes), we instead offer a hypothetical explanation for why selection might favor Clonal lineages with rates of recombination that are high, rather than the low which has been previously assumed (Moritz and Haberl 1994; Baudry et al. 2004). This, combined with the data presented here and previously (Goudie et al. 2012) provides a strong rational for why research must move away from the a priori assumption that low rates of recombination would be favored in clonal lineages. Or indeed a priori assumptions in general about selection in clonal lineages.

We conclude that although overdominant loci are indeed infrequent in the honey bee genome relative to the total number of genes (i.e., just a few per chromosome), the effects of these genes is profound. In this natural system where the confounding effects of additive loci and general effects has been stripped away, we can observe how overdominant genes exerting local effects have shaped the evolution of a genome.


This work was funded by Australian Research Council grants to BPO. The authors thank M. Beekman and other members of the Behaviour and Genetics of Social Insects lab for comments on the manuscript.