A Microsatellite-Based Linkage Map of the Honeybee, Apis mellifera L.

Biological material and DNAs

To maximize the number of informative loci, we have analyzed the meioses of queens that were hybrids between different subspecies of A. mellifera. These were the progeny of two F₁ hybrid queens named B and V (Figure 1). Virgin A. m. ligustica queens (the grandmothers) were imported from Italy (Bologna and Perugia) and instrumentally inseminated with the mixed sperm from 12 drones (the grandfathers) from four A. m. mellifera colonies from France (Charente). The 12 inseminated queens were allowed to found as many colonies. Adult hybrid queens were obtained in several of these colonies and we retained one colony that provided 9 queens. Every one of these 9 F₁ hybrid queens (the mothers) was instrumentally inseminated with the mixed sperm from two males (the fathers), one belonging to the subspecies A. m. ligustica and the other one to the subspecies A. m. mellifera (i.e., to both parental subspecies). The pair of males was different for each inseminated queen. The female progeny (workers) of these backcrosses were collected for two of the nine families (queens B and V). The bodies of the males used for insemination (grandfathers and fathers) were preserved in alcohol.

The male (haploid) progeny of a third hybrid queen (M) were used, mainly to map the MDH locus. The queen was a “triple hybrid”: her mother was the daughter of an A. m. ligustica queen artificially inseminated by an A. m. caucasica male and this hybrid was naturally fecundated by A. m. mellifera drones. Laying of unfertilized eggs was obtained after carbon dioxide treatment of the queen.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

—Crosses performed to obtain workers for map construction. The grandmother was Apis mellifera ligustica (white) and it was instrumentally inseminated with the sperm admixture from 12 drones (grandfathers) of A. m. mellifera (black). Two hybrid queens (mothers, black/white), B and V, were backcrossed to two drones, one for each parental subspecies (fathers). The two families (workers) were composed of an admixture of two backcross subfamilies. The two grandfathers among the 12 drones were determined on the basis of their genetic profile (see text); their genotypes were used to establish the allelic phase of the B and V queens.

For each of the three families, DNA extracts from 96 workers (B and V) or drones (M) were prepared from the heads and later from the thoraces. DNAs were diluted to 1/40 and conserved at –20° in 96-well microtitration plates. We used only 92 workers for the B family. The remaining 4 wells were used to test DNA from three other Apis species and to genotype the B grandfather for establishing the allelic phase of the B hybrid queen. A total of 95 workers for the V family (the remaining sample being the V grandfather) and 96 drone progeny were analyzed. For the two families in which workers were studied (B and V), the two patrilines corresponding to the two different fathers are hereafter called subfamilies.

Markers

We used 552 markers described in Solignac et al. (2003), comprising mainly microsatellites and a few long repeated regions (ITS of the rDNA locus, royal jelly gene), and adopted the nomenclature proposed in this article: Am followed by a four-digit number.

A total of 82 markers were not mapped because the mothers were homozygous. Conversely, 67 additional bands (not corresponding to microsatellite loci) were mapped. In Figure 2, the numbers appearing in loci names correspond to bands showing the size of the sequenced allele. Additional bands that were associated with a specified microsatellite are labeled with an “s” (for “supplementary” bands) plus a number if several bands were generated. A total of 31 supplementary bands for 21 primer pairs were detected.

In addition, one RAPD marker (P13, Operon), the locus Q (Hunt and Page 1995) using Sau3A restriction of the PCR products, and the locus FB, all three known to be linked to the sex locus, were mapped. The malate dehydrogenase (MDH) locus was genotyped as described in Cornuet (1979) and was positioned on the linkage map only with the male progeny.

PCR and genotyping

Radioactive PCRs were performed following Solignac et al. (2003). PCRs were generally performed in multiplexes and the remaining loci were amplified separately but two or three PCR products were loaded on the same gel.

Initially, genotyping of the backcross progeny of queens B and V was established without preliminary tests of heterozygosity of the queens. Later, heterozygosity of the mothers was tested before analyzing the whole progeny: the tests were performed either on the queen itself for family V or on eight worker descendants for family B in which the queen was lost.

The grandfathers B and V were run in parallel with the workers to determine allelic phases of the B and V queens. Because the ligustica queens had been inseminated by a mixture of sperm from 12 mellifera males, it was first necessary to determine which ones fathered the hybrid queens. When a reasonable number of mapping gels had been completed, the grandfathers were easily determined on the basis of their genetic profiles: for each locus they must share their unique allele with their daughter. The fathers were used only to assist genotype determination and to confirm null alleles.

Statistics

Map construction: A series of preliminary checks and computations were performed. In two female progenies (B and V), the two patrilines had to be separated to determine the maternal allele. This was determined in the first 12 loci. Because of male haploidy, all workers of the same patriline share the same paternal alleles. The assignment of the workers to their respective patriline allowed designation of individual genotypes and tests for segregation distortion. The following checks were performed on every locus. Within a patriline, there are two possible diploid genotypes in the progeny of a heterozygous queen. Additional genotypes are due to either a mutation or a mistyping error. Two patrilines arising from the same queen must provide the same maternal genotype. Within each patriline, the two maternal alleles should be distributed equally among workers in the absence of segregation distortion or mistyping errors.

Once all detectable mistyping errors had been ruled out at individual loci, linkage of a locus was tested against previously mapped loci. The phase of the alleles was determined as each locus was entered. Eventually, the input file was edited by distinguishing the two maternal alleles (e.g., A, grandpaternal; H, grandmaternal). This setting corresponds to the usual “F₂ backcross” data type analyzed by genetic mapping software.

The linkage map was built with Carthagene mapping software (Schiex and Gaspin 1997) version 0.5. This program computes genetic maps using a range of different algorithms. At any time, the best maps are recorded in a heap structure and any new map is compared to the previous maps of the heap. The different algorithms can be chained in a single command. Typically, the command that we used included the following:

1. Starting from each locus (on a linkage group), the algorithm gradually builds the best neighborhoods maps using a two-point LOD criterion (command nicemapl).

2. Starting from two loci, the algorithm adds a new marker by testing all possible positions, until all markers are mapped. The criterion also uses the two-point LOD criterion and the best three orders (maps) are kept at each step (command build).

3. A Tabu search (Glover 1989a,b) is performed on the multipoint LOD score criterion. Starting from the best map of the heap, it explores the two-change neighborhood of maps, i.e., the set of maps obtained by inverting all possible subsections of the map (command greedy).

4. A simulated annealing algorithm based on the multipoint LOD score is used. Two and three changes are randomly applied to the current map (command annealing).

5. Maps are next considered as individuals having a selective fitness proportional to their multipoint LOD score. Starting from an initial population, maps evolve along generations through recombination and mutation. Only the best-fit maps are conserved to produce the next generation (command algogen).

6. Starting from the best map of the heap, the algorithm tests systematically all permutation of markers within a window of a given size (command flips).

7. Starting from the best map of the heap, the algorithm takes one locus at a time and tests all its possible locations (command polish).

Linkage groups were established at the two-point LOD score threshold of 3.3.

Checkpoints during map construction: Additional precautions were taken to prevent errors in identification of loci and individuals and in the determination of genotypes.

Locus: The number given to the locus was confirmed by comparison of the mapping gels with the gels for the test of heterozygosity and the gels made for the completion or correction of data.

Individuals: PCR products of half-sisters from the two subfamilies were loaded on the gels' intermingling subfamilies but always in the same order. Once paternal alleles had been identified, this allowed for an easier detection of any shift when reading genotypes. When the two male parents had the same allele, the preceding control was inefficient and one or several series of individuals with characteristic genetic profiles were reamplified. A similar map location between the two patrilines implied that individuals were correctly assigned to the subfamily.

Genotypes: The films were read independently by at least two people. Personal programs were used to check the data for the presence of only two genotypes for each of the two subfamilies and for possible segregation distortion. Ambiguous or missing genotypes were systematically reamplified and run again in the presence of controls. Once the first version of the map had been constructed, every individual showing two recombination events surrounding a single locus was genotyped again for this locus.

Maps: The last control consisted of comparing maps constructed with the progenies of the B and V queens separately before combining them.

Chiasma interference: The occurrence of a crossover in one genetic region frequently decreases the probability of a concomitant crossover in an adjacent region. This phenomenon, called positive interference, has been observed in many species (Zhaoet al. 1995b). Investigations of interference have classically compared observed frequencies of multiple recombination events in adjacent intervals to their expected frequency (Zhaoet al. 1995b). However, due to the rarity of such events, these analyses require a huge number of meioses (i.e., several thousands). We have therefore made use of the approach proposed by Broman et al. (2002) and Broman and Weber (2000) that considers the distribution of the estimated distances between crossovers.

In this approach, it is important to identify all recombination events that occurred along a chromosome. We have therefore considered only the data from the B progeny, which were genotyped for the largest set of markers. All recombination events were identified and their positions were assumed to be the midpoint of the interval between the two flanking markers. We computed the intercrossover distance distribution for all linkage groups showing at least 40 intercrossover distances, i.e., groups I–XI. We then fitted the observed distribution of the distance between crossovers to that expected under the gamma model with an integer parameter, using the correlation coefficient plot technique (Filliben 1975) with the software Dataplot (http://www.itl.nist.gov/div898/software/dataplot/).

The gamma model assumes that distances between chiasmata are independent and follow a gamma distribution with shape and rate parameters of ν and 2ν (see Zhaoet al. 1995a for a review; Broman and Weber 2000). Several studies have found that the gamma model provides an excellent fit to recombination data (Zhaoet al. 1995b; Lin and Speed 1996; Broman and Weber 2000; Linet al. 2001; Bromanet al. 2002). We focused on a special case of the gamma model where the shape parameter is ν= m + 1 with m a nonnegative integer. This model, also called the chi-square model, explains well certain empirical observations concerning recombination and gene conversion (Fosset al. 1993).

Segregation distortion: In the three progenies, we observed the segregation of the maternal genes only. At each locus, the two alleles were expected to be represented in equal frequencies in the progeny. However, alleles at some loci segregated in proportions significantly different from the expected 1:1 ratio. This may be due to random type I errors, but this may also be due to specific segregation distortion loci (SDL). These SDL were detected by examining observed ratios in adjacent loci.

A graphical approach was used in which the segregation distortion measured by the chi-square criterion was plotted against the genetic distance on each linkage group. This was applied to only the first two progenies (of queens B and V) because the third (drone) progeny did not provide a sufficiently dense map to allow this analysis. The two subfamilies in each progeny were analyzed separately because an SDL paternal allele can interfere in different ways with each of the two maternal alleles.

An additional analysis was performed to detect potential interactions between loci exhibiting segregation distortions. This was done to distinguish between loci that have deleterious alleles and loci that have alleles that are deleterious only in association with alleles at another locus. Linkage does induce such an association, so that the analysis was reduced to all pairs of unlinked loci in which both loci shared significant segregation distortion. In these pairs, we tested whether or not the association of the maternal alleles at each locus was random through a homogeneity chi-square analysis.