Genetic characterization of the honeybee (Apis mellifera) population of Rodrigues Island, based on microsatellite and mitochondrial DNA
Apis mellifera is present in Rodrigues, an island in the South-West Indian Ocean. The history of the established honeybee population is poorly known, and its biodiversity has never been studied. In this study, maternal origins of A. mellifera in Rodrigues have been assessed with the DraI test and sequencing of the mitochondrial COI-COII region. Nuclear genetic diversity was investigated with 18 microsatellite markers. A total of 524 colonies were sampled from 16 beekeepers in 20 sites. The Rodrigues population was characterized by the absence of the African lineage and presence of three European haplotypes from the C lineage. C1 haplotype was the most frequent (81.3 %), followed by C2 (18.0 %) (characteristics of A. m. ligustica and carnica, respectively) and a new haplotype, C1-Rod (0.7 %). No genetic structure was detected, and genetic diversity was comparable to continental populations. According to approximate Bayesian computation (ABC) analyses, bottleneck scenarios are most likely to have occurred.
Keywordstropical insular molecular analyses ABC method beekeeping
During the past decade, honeybee populations have experienced declines and colony losses are continually reported in Europe and the USA (van Engelsdorp and Meixner 2010; Chauzat et al. 2013). The reduction of genetic diversity due to beekeeping practices is one of the several causes suggested (Oldroyd 2007) and has had an important impact on managed colonies (De la Rúa et al. 2013). Studies have demonstrated that reduced levels of intra-colonial genetic diversity affect colony fitness (Tarpy et al. 2013).
In its endemic area, Apis mellifera has evolved into at least 28 described subspecies (Ruttner 1988; Sheppard and Meixner 2003; Meixner et al. 2011, 2013) which have been grouped into four evolutionary branches using morphometric analysis (Ruttner 1988): the European North-Mediterranean (C) and the West-Mediterranean (M) lineages, the Oriental O lineage and the African A lineage. The breakdown of the different evolutionary lineages is supported by molecular analyses (Garnery et al. 1992; Whitfield et al. 2006; Wallberg et al. 2014). The European subspecies Apis mellifera ligustica and Apis mellifera carnica (C lineage) have been introduced worldwide because of their commercially desirable traits and have succeeded in adapting to new habitats (Chapman et al. 2008; Delaney et al. 2009).
In the South-West Indian Ocean, Apis mellifera unicolor (Latreille 1804) has been described as an endemic subspecies from Madagascar and belongs to the African lineage (Ruttner 1988). Honeybees are also established in the Mascarene archipelago (East of Madagascar) comprising La Réunion, Mauritius, and Rodrigues (Ruttner 1988). Rodrigues was discovered early in the sixteenth century, on the route to India (North-Coombes 1971). It is not known whether A. mellifera spp. were already present on the island before colonization. Controlled imports of honeybee queens are thought to have started in the 1930s, and around 1500 traditional hives were identified in the 1940s and 1950s (Bappoo and Ramanah 1989). In 1981, 11 European queens were reported to have been introduced to Rodrigues from the USA (Bappoo and Ramanah 1989). In 2002, the Rodrigues government restricted honeybee importation (Rodrigues apicultural services, unpublished data). In 2010, 2251 managed colonies from 173 beekeepers were identified (Belmin 2010). Since human colonization of the island, extensive habitat degradation has occurred with the total loss of the original forest (Thébaud et al. 2009).
In the Mascarene, mitochondrial diversity of honeybees had been investigated in La Réunion (n = 20) and Mauritius (n = 10) and AI was the sole sublineage detected (Franck et al. 2001). This study aims to describe the genetic diversity of the honeybee population in Rodrigues using molecular tools. First, the evolutionary lineages and subspecies were investigated using the mitochondrial COI-COII intergenic region. Then, the level of nuclear genetic diversity and its structure was investigated with microsatellite markers. Finally, we attempted to detect a bottleneck event, date it, and estimate effective population size before, during, and after the event.
2 Materials and methods
2.1 Sampling and DNA extraction
2.2 Mitochondrial DNA analysis
Mitochondrial analysis was carried out following Garnery et al. (1993). This consisted of PCR amplification of the intergenic COI-COII region using E2 and H2 primers, followed by a restriction of the amplified product with the DraI enzyme (Promega). PCR reaction was performed in a volume of 15 μL containing 2 mM MgCl2, 2 mM dNTPs, buffer 5× (Promega), 20 pmol of each primer, 30 units of GoTaq Flexi® Promega, 0.5 μL of DNA at 5 ng/μL. Samples underwent initial denaturation for 5 min at 92 °C, 35 cycles of 30 s at 92 °C, 45 s at 48 °C, 2 min at 62 °C, and a final elongation step of 7 min at 92 °C. DraI digestion was performed according to the manufacturer’s recommendations (Promega). Restriction fragments were separated in 5 % agarose MetaPhor gels.
The mitochondrial DNA (mtDNA) intergenic COI-COII region was sequenced in at least 30 % randomly selected individuals within each sample site, using the same E2-H2 primers. Individuals exhibiting rare restriction profiles were systematically sequenced. Sequence checking and alignment was done using Mega 5.1 software (Tamura et al. 2011) and then blasted on NCBI GenBank. All newly detected sequences were submitted to the GenBank database.
2.3 Microsatellite DNA analysis
A total of 18 polymorphic microsatellite loci (Estoup et al. 1994, 1995; Franck et al. 1998; Solignac et al. 2003) were scored A113, A24, AC306, AP55, AP81, A107, A29, A88, AP273, A28, AP289, B124, A35, A8, AP33, AP43, AP66, and A43 (Table S1). Primers were distributed into four multiplex PCR reactions (Table S1). PCR reactions were performed in 10-μL volumes containing 5 μL of Master Mix Type-it 2× Qiagen, 0.2 μL of each primer at 20 pmol/μL, and 1 μL of DNA at 5 ng/μL. All programs started with a denaturing cycle at 94° for 5 min, followed by 35 cycles of 30 s at 94 °C, 30 s at 52 °C, 45 s at 72 °C, and a final elongation at 72 °C for 20 min, except for Mix 1 where annealing was done at 55 °C. The samples were run through DNA sequencer ABI Prism 3130×L, and alleles were scored using Genemapper 4.0 (Applied Biosystems).
As a preliminary analysis, Micro-Checker 2.2.3 (Van Oosterhout et al. 2004) was used to identify genotyping errors. The frequency of null alleles per locus was estimated with FreeNa (Chapuis and Estoup 2007) based on 1000 bootstrap replicates. For each site with N ≥ 10 colonies, microsatellite diversity was estimated through mean number of alleles (Na), observed (Hobs), and unbiased expected heterozygosity (He) (Nei 1978) with GENETIX 4.05 (Belkhir et al. 1996). Allelic richness (Ar) per site was calculated with FSTAT 22.214.171.124 (Goudet 2001) using the rarefaction method. Heterozygosity excess or deficit was estimated by calculating FIS according to Weir and Cockerham (1984). Deviations from the Hardy-Weinberg equilibrium (HWE) and linkage disequilibrium were tested using GENEPOP 3.4 (Raymond and Rousset 2004). False discovery rate (FDR) estimators (q-value) were calculated for probability values of repeated comparisons, using the R package QVALUE (Dabney et al. 2004). Pairwise FST values were calculated for all combinations of the 18 sites (with N ≥ 10 colonies) and tested using permutations with FSTAT (Goudet 2001). Population structure and assignment of individuals to clusters were inferred using a Bayesian model-based clustering method with STRUCTURE 2.3.3 (Pritchard et al. 2000). Analyses were based on the admixture model with correlated allele frequencies with simulations of 105 burn-in steps and 106 iterations of MCMC algorithm for each run. Ten runs for each K (number of genetic clusters) value [1–20] were computed to estimate the most likely value of K. Optimal number of clusters was determined using Evanno’s estimator ∆K (Evanno et al. 2005) implemented in STRUCTURE HARVESTER (Earl and vonHoldt 2011). Principal components analysis (PCA) was used to visually identify and describe clusters of genetically related individuals with adegenet 1.3-9.2 package (Jombart 2008).
2.4 Testing for a bottleneck event using ABC
Parameter names of all four scenarios used in DIYABC software prior values minimum and maximum, posterior parameter estimate values (mode as point of estimation), and 95 % confidence interval, relative bias, and relative square root error.
Prior value range
Posterior parameter values
95 % confidence interval
Relative square root error
Population size effective
Time in generation
Genetic parameters (rate)
[1 × 10−4–1 × 10−3]
1 × 10−3
[4.0 × 10−4–1.0 × 10−3]
[1 × 10−1–3 × 10−1]
3 × 10−1
[1.4 × 10−1–3.0 × 10−1]
[1 × 10−8–1 × 10−5]
1 × 10−8
[1.3 × 10−8–5.7 × 10−6]
3.1 Mitochondrial DNA
PCR amplifications of the mtDNA intergenic COI-COII region were successful for 468 out of the 524 individuals sampled. All amplified products presented the same size ∼570 bp, which corresponds to the length of Q-type sequence, characteristic of the East European C lineage (Garnery et al. 1998). After PCR-RFLP analyses were performed, only one restriction fragment pattern was observed (fragment size ∼40/41, 47, 64, and 420 pb). Of the 278 individuals sequenced (Figure 1), three different sequences were detected. Two were previously described as the C1(a) haplotype belonging to populations of the A. m. ligustica subspecies (GenBank accession FJ478010.1 (Franck et al. 2001), JQ977699.1 (Muñoz et al. 2012), EF033655.1 (Collet et al. 2006)) and the C2(j) haplotype reported in A. m. carnica (JF723978.1). The third sequence exhibited 99 % similarity with C1 and presented a new polymorphic site. This new sequence was named C1-Rod (KF955993). C1 was the most frequent (n = 226, 81.3 %, KF955992) and was present in all sites; C2 (KF955994) was observed in 50 colonies from 11 sites (18.0 %), while C1-Rod was only detected in two colonies (0.7 %) of the same site S15 (Figure 1).
3.2 Microsatellite loci
A107 and AP81 (i) were difficult to score, (ii) presented high levels of null alleles (13.2 and 9.2 %, respectively), and (iii) had the highest levels of missing data (3.8 and 1.5 %, respectively). For these reasons, A107 and AP81 were removed from nuclear analysis. All individuals were successfully genotyped with at least ten loci amplified. Only nine loci combinations of 120 were significantly linked and involved different loci.
Sampling information with site code, number of colonies (Ncolonies), mean number of alleles (Na), observed heterozygosity (Hobs), expected unbiased heterozygosity (Hexp), and Weir and Cockerham’s fixation index (FIS).
Genetic differentiation between sites (pairwise FST) was low and ranged from −0.009 (sites S17 and S16) to 0.022 (sites S3 and S6). After Bonferroni corrections, only one pair of 153 pairwise comparisons was significant (FST = 0.014; p value = 0.000260 < p value adjusted for multiple comparisons = 0.000263) for sites S15 and S12, respectively.
Analyses of Bayesian model-based clustering with STRUCTURE did not reveal evidence of population structure and differentiation, using the method of Evanno et al. (2005) (Figure S3). When K was >1, the probability of posterior assignment of each individual was equivalent for each cluster. Those results were supported by complementary PCA analyses which did not show any differentiation between groups of individuals.
3.3 Testing for a bottleneck event using ABC
Scenarios were ranked according to their posterior probabilities as follows: the best scenario was scenario 2 with a posterior probability of 0.558 (95 % confidence interval [0.543, 0.571]). Then, scenario 4 (p = 0.347 [0.333, 0.361]), followed by scenario 3 (p = 0.089 [0.080, 0.099]). The last was scenario 1 (p = 0.006 [0.000, 0.014]). Based on posterior probabilities, scenarios implicating a bottleneck (2 and 4) were more probable (a total posterior probability of 0.904) and supplanted the scenarios without a bottleneck (1 and 3, for a total posterior probability of 0.096). Full results of simulations to test for scenario confidence are given in Table S2. The probability that data sets simulated under scenarios without bottleneck (1 or 3) were assigned to the scenarios with bottleneck (2 or 4) amounted to 0.101 (type II error rate).
Parameter estimates, relative bias, and relative precision indices for scenario 2 are listed in Table I. Genetic parameters and ancestral population size Nanc could not be estimated with precision (data not shown). Despite this, it is remarkable that noteworthy parameters could be estimated with satisfactory precision. The current population effective size N1 was estimated at around 760 individuals (95 % confidence interval [333–1840]). The population experienced a bottleneck event 49 generations ago, ending 19 [7–95] generations ago and lasting approximately 30 [9–47] generations. Estimations suggested that around 60 [20–95] individuals colonized or were introduced to the island from a larger and ancestral population with an effective size of over 7463 [5470–9840].
Mitochondrial analyses showed that all sampled colonies had only the Q sequence characteristic of the C lineage. This result is startling knowing that the honeybees of the surrounding islands are of African lineage (Ruttner 1988; Franck et al. 2001). Furthermore, Rodrigues mainly depends on Mauritius for commercial exchanges; honeybees could have easily colonized Rodrigues. European queens introduced in 1981 could have been more competitive and/or more favored by the beekeepers than the African lineage (if present).
The sequencing of the COI-COII intergenic indicated A. m. ligustica (C1, C1-Rod) and A. m. carnica (C2) descendants (Garnery et al. 1993, 1998). A. m. ligustica maternal lineages were four times more prevalent than A. m. carnica over all the sampling. Presence of European C lineage haplotypes is consistent with the reported introduction of queens bought from the US commercial stock (Bappoo and Ramanah 1989). In the USA, haplotypes C1 and C2 are characteristic of the most popular honeybee strains sold (Delaney et al. 2009). These commercially widespread subspecies have adapted well to the environmental conditions of this tropical island and persisted for over 32 years.
Furthermore, results of the ABC scenario choice support the assumption that the introduction of honeybees to the island was associated with a genetic bottleneck. Results from the bottleneck test are entirely credible in such insular ocean environments where colonization is limited by geographical barriers. According to the best scenario (2 with p= 0.558), this bottleneck event occurred approximately 49 generations ago (in an interval of 16 to 142 generations). One generation of A. mellifera is assumed to be approximately 2 years (Estoup et al. 1995; Excoffier et al. 2005). Present estimations are consistent with historical records of introductions of European honeybees in 1981. DIYABC results also suggest that the origin population had a large effective size (more than 7463 individuals). In the USA, the honeybee population was estimated to be 4.5 million colonies in 1980 (van Engelsdorp et al. 2008).
Nuclear genetic analyses indicated that the honeybee colonies on Rodrigues constituted one genetic population with homogenous diversity among sites. This observation was also demonstrated in other oceanic islands such as Menorca (701.8 km2) and Ibiza (572 km2) (De la Rúa et al. 2003). No genetic differentiation appeared along geographical distances indicating that gene flow is sufficient to homogenize nuclear genetic diversity. Rodrigues is a small island (104 km2) where apiary transhumance is not practiced by beekeepers. The distance between sedentary colonies might be too small to create genetic isolation (0.60 to 13.8 km). Exchange of queens between beekeepers, colony swarming, and male flight most likely ensure gene flow. Here, no nuclear genetic differences were found between colonies with the mtDNA sequence characteristics of A. m. ligustica and A. m. carnica whereas Muñoz et al. (2009) were able to detect differences between subspecies. This absence of genetic difference suggests that either hybrids were introduced or hybridization occurred after introduction. The study gives evidence of anthropogenic effects on the honeybee population in Rodrigues such as (i) the presence of exotic subspecies favored by beekeepers, (ii) hybridization phenomena between subspecies, and (iii) absence and/or non-utilization of the A lineage honeybees present in the area.
The Rodrigues honeybee population has levels of genetic diversity comparable to continental European populations of A. m. ligustica and carnica. For four comparable loci (A113, Ap55, B124, and Ap43), indigenous populations of A. m. carnica and/or ligustica in Croatia and Italy showed between 9.25 and 9.75 alleles per locus whereas 8.50 alleles were detected in the Rodrigues population (Muñoz et al. 2009). Another indication of the high levels of genetic diversity was heterozygosity (Hobs = 0.644) which was equivalent to (i) indigenous Italian and Croatian populations (Hobs = 0.638 and Hobs = 0.674, respectively (Muñoz et al. 2009)) and (ii) US commercial populations (Hobs = 0.53 in 2005 (Delaney et al. 2009)). Levels of heterozygosity of the Rodrigues honeybee population were higher than other insular systems in which A. m. ligustica was introduced. As an example, in the Azores, 35 % of colonies had haplotype C1 (cohabiting with 48 % of sublineage AIII and 17 % AII) and levels of observed heterozygosity were two times lower (Hobs = 0.22 ± 0.03) than those in Rodrigues (De la Rúa et al. 2006). The same can be found in the Canary Islands, with only 31.2 % of observed heterozygotes in El Hierro (n = 17, 18 % of haplotype C1 and 82 % haplotype A15) and 41.3 % in Tenerife (n = 76, 35 % of C1 and three other African haplotypes) (De la Rúa et al. 2001). Unlike the rest of the world, the honeybee population in Rodrigues did not experience strong biological pressure caused by parasites and pathogens and this could explain these differences (vanEngelsdorp and Meixner 2010). A. mellifera in Rodrigues Island has a remarkable sanitary situation with no Varroa destructor (Belmin 2010), which was recently reported in Madagascar (Rasolofoarivao et al. 2013). To this day, the only known parasites and pathogens are the bee louse Braula pretoriensis and Nosema apis (Rehm 1988).
This study revealed a surprising lack of African lineage and exclusive presence of two subspecies: A. m. carnica and A. m. ligustica from the European C lineage. The honeybee population in Rodrigues did not show genetic structure but presented high levels of heterozygosity.
We would like to thank J Félicité, G Volbert, and the Commission of Apiculture and Agriculture of Rodrigues for their welcoming and assistance in organizing the collection of samples. We are also grateful to the beekeepers: JJ Clain, JN Daya, L Felicité, D François, G François, J Herman-Casimir, JP Kan-John, JH Marianne, C & M Peermamode, MJ Ravanne, J Rikarl, and the feminine associations Ile Michel and Rodrigues-Farmer. We wish to thank C Simiand for his assistance in the laboratory. We would like to warmly thank V Ravigné for her precious help with DIYABC software; for the understanding of the methods, the analyses, and results; and for her corrections. We thank L Garnery for his advice on molecular methodology. We would like to thank the anonymous reviewers for improving the manuscript. Finally, we thank the network QualiREG and V Porphyre for helpful contacts. M A Techer is recipient of a PhD fellowship of the Sciences and Technology Doctoral School of the University of La Réunion. This study was partially funded by CIRAD, Région Réunion, and e-PRPV.
- Bappoo, D.R., Ramanah, D. (1989) The state of beekeeping in Mauritius and other Mascarene Islands, p. 2. Proceedings Fouth International Conference on Apiculture in Tropical Climates, Cairo, EgyptGoogle Scholar
- Belkhir, K., Borsa, P., Chikhi, L., Raufaste, N., Bonhomme, F. (1996) GENETIX 4.05, logiciel sous Windows TM pour la génétique des populations. Laboratoire génome, populations, interactions. CNRS UMR 5000, 1996–2004Google Scholar
- Belmin, R. (2010) Intérêt et faisabilité d’une Indication Géographique pour le miel de Rodrigues, p. 72. Université de Nancy, NancyGoogle Scholar
- Dabney, A., Storey, J.D., P.R.S. qvalue Tutorial, Install, R., Packages, I.B., et al. (2004) Q-value estimation for false discovery rate control. Medicine 344, 539–548Google Scholar
- Goudet, J. (2001) FSTAT 2.9. 3.2, a program to estimate and test gene diversities and fixation indicesGoogle Scholar
- Latreille, P.A. (1804) Notice des espèces d’abeilles vivant en grande sociétée, ou d’abeilles proprement dites, et description d’espèces nouvelles. Ann. Mus. Natl. Hist. Nat. 5, 161–178Google Scholar
- Muñoz, I., Stevanovic, J., Stanimirovic, Z., De la Rúa, P. (2012) Genetic variation of Apis mellifera from Serbia inferred from mitochondrial analysis. J. Apic. Sci. 56(1), 59–69Google Scholar
- North-Coombes, A. (1971) The island of Rodrigues. Mauritius, Published by the author, with the assistance of the Mauritius Advertising BureauGoogle Scholar
- Rasolofoarivao, H., Clémencet, J., Ravaomanarivo, L.H.R., Razafindrazaka, D., Reynaud, B., et al. (2013) Spread and strain determination of Varroa destructor (Acari: Varroidae) in Madagascar since its first report in 2010. Exp. Appl. Acarol. 1–10Google Scholar
- Raymond, M., Rousset, F. (2004) GENEPOP version 3.4. Population genetics software for exact tests and ecumenicism. Laboratoire de Genetique et Environment, Montpellier, FranceGoogle Scholar
- Rehm, S.M. (1988) Report on the consultancy on bee diseases (Mauritius and Rodrigues), United nations development promgramme, pp. 77Google Scholar
- Ruttner, F. (1988) Biogeography and taxonomy of honeybees. Springer-VerlagGoogle Scholar
- Thébaud, C., Warren, B.H., Strasberg, D., Cheke, A. (2009) Mascarene islands. Biology 127, 1–216Google Scholar
- Weir, B.S., Cockerham, C.C. (1984) Estimating F-statistics for the analysis of population structure. Evolution 1358–1370Google Scholar