Mitochondrial DNA (mtDNA) is widely used for phylogeographic and phylogenetic studies for two principal reasons. First, mtDNA is a compact double-stranded circular DNA molecule (approximately 15–16kb) and second, recombination (the exchange of homologous DNA sequences between different chromosomes) is thought to be absent in animals owing to the clonal inheritance of mtDNA. Numerous phylogenetic studies rely on mtDNA genes (both protein-coding and ribosomal RNA genes) amplified through conserved polymerase chain reaction (PCR) primers (Simon et al. 1994). Most models of genetic divergence and phylogenetic methods assume absence of recombination. Thus, major conclusions drawn from these analyses may be inaccurate since tree-building algorithms are misled by recombination (Avise 1994; Schierup & Hein 2000; Posada & Crandall 2002).

Despite the fact that animal mitochondria undoubtedly have the necessary toolkit of enzymes to recombine (Thyagarajan et al. 1996; Eyre-Walker et al. 1999; Yaffe 1999; Rokas et al. 2003), evidence for recombination in animal mtDNA is scarce (Rokas et al. 2003). Evidence based on different indirect methods was found in humans, mice (Mus musculus), flatfish (Platichthys spp.), frogs (Rana spp.) and bivalve mussels (Mytilus spp.) However, only in the last case is there a direct experimental confirmation for homologous recombination owing to biparental inheritance (Awadalla et al. 1999; Ladoukakis & Zouros 2001). Whether human mtDNA is really recombining has been questioned since Awadalla et al. (1999) used a potentially uncertain measure of linkage disequilibrium (LD) in their test for recombination (Ingman et al. 2000; Kivisild & Villems 2000; Kumar et al. 2000) and there is a possibility of sequencing errors in their data (Macaulay et al. 1999; Wiuf 2001; Innan & Nordborg 2002).

While working on mtDNA phylogeographies of Old World buthid scorpions (genera Buthus and Mesobuthus; Gantenbein & Largiadèr 2002, 2003; Gantenbein et al. 2003), results were repeatedly obtained that led to the suspicion that there might be recombination in the mtDNA of some of these species. These suspicions were further fuelled by cytological evidence, reporting the fusion of mitochondria into a ring (mitofusion) in the Buthidae scorpion family. The unique cytological phenomenon of mitofusion together with the preliminary statistical evidence for recombination prompted the investigation of this issue in more detail.

Here, we present strong statistical evidence for recombination in newly sequenced populations of west and Central Asian scorpions of the genus Mesobuthus and in previously published scorpion populations (Gantenbein & Largiadèr 2002, 2003; Gantenbein et al. 2001, 2003), sequenced at two mitochondrial loci, the large ribosomal rRNA unit (16S) and a fragment of the cytochrome oxidase subunit I (coxI). In order to precisely estimate the physical distance between the two sequenced loci, which is required for the correlation between genetic LD and distance, we sequenced the complete mitochondrial genome of the scorpion Mesobuthus gibbosus, which will be published elsewhere (Gantenbein in progress).

Additionally, LD between polymorphic sites using two different measures of LD, |D′| (Lewontin 1964) and R² (Hill & Robertson 1968) was estimated, the latter being more robust to variation in mutation rates (Awadalla et al. 1999; Kivisild & Villems 2000; Kumar et al. 2000; Meunier & Eyre-Walker 2001; Wiuf 2001; Innan & Nordborg 2002; Piganeau & Eyre-Walker 2004). LD was estimated using the LDhat computer package (available at http://www.stats.ox.ac.uk/~mcvean/LDhat/LDhat.html). |D′| and R² versus physical distance was then correlated. A spacer fragment, was included corresponding to 5038bp in M. gibbosus, thus, the spacer fragment was approximated to ~5kb for all correlations in all four species (16S-5kb-coxI). The correlation was also repeated with maximum power going the longer way around the mitochondrial genome with a ~10kb spacer fragment (coxI-10kb-16S). Significance of the Pearson correlation coefficient was assessed by using the one-tailed Mantel test computed over 10000 permutations.

The DNA sequence datasets for the four species contain considerable polymorphism in terms of the total number of segregating sites (table 1). The significant decline of LD versus distance is regarded as strong evidence for recombination (Awadalla & Charlesworth 1999). When using R² highly significant correlation was obtained in all four species (p<0.0001; table 2). Since R² is dependent on allele frequencies, the analysis was additionally restricted to sites at which both alleles were present at a frequency greater than 10%. It has been argued that this restriction limits the analysis to alleles, which tend to be older and are more likely to show evidence for past recombination events (Awadalla et al. 1999). Using this subset of polymorphisms, significant correlations were still observed for three out of four species, with the correlation in M. gibbosus becoming non-significant (table 2). Performing the correlation with |D′|, the correlations remain highly significant only for B. mardochei (both when considering all polymorphisms or only those above 10%), and marginally so only for polymorphisms in frequencies above 10% in M. gibbosus (table 2). Conversely, the homoplasy test finds strong evidence for recombination only in M. caucasicus (table 3). The best evidence for recombination is thus found in the two species B. mardochei and M. caucasicus (tables 2 and 3), whereas in the two other species (M. eupeus and M. gibbosus) the evidence is somewhat conflicting. In M. eupeus in particular, only the correlation between R² and physical distance is significant.

Table 1 The total number of sites and the number of segregating sites.

Table 2 Correlation between linkage disequilibrium and physical distance in four scorpion species. (Estimates of R2 and |D′| were carried out using the computer package ‘LDhat’. p values were calculated from a one-sided Mantel test with (more ...)

Table 3 Homoplasy test in four scorpion species. (***p values <0.0001, **p values >0.0001 and <0.01, and *p values >0.01 and <0.05.)

Testing for recombination using the general RDP, Geneconv, Maxchi and Chimaera, on a concatenated ‘supergene’ of the 16S and coxI data recovered many putative recombinant ‘daughter’ haplotypes between ‘parental’ haplotypes in the B. mardochei sequence alignment. An example is given in figure 1, where the RDP and the Geneconv test provide strong support for two ‘breakage points’ at sequence positions 319 and 685: The pairwise identity in the regions 1–319 and 686–823bp is relatively high between BomAA1a and BomAA5a (haplotypes sampled from two populations from the Moroccan Anti-Atlas) relative to a third sequence (BomHA5a, a sequence from the Moroccan Haut-Atlas; figure 1a). In the middle piece (320–685bp), however, the two haplotypes BomAA5a and BomHA5a share many polymorphic sites, thus being more similar than the two haplotypes BomAA1a and BomAA5a. ‘Breaking points’ similar to those shown in figure 1 are found for the three haplotypes BomAS1a (As Saghir region, Morocco), BomHA1a and BomHA1b. When only the protein-coding region was tested using the same four recombination tests, no potential recombination events were found using RDP and Geneconv, whereas the algorithms Maxchi and Chimaera found 12 and 8 putative recombination events in B. mardochei, 18 and 1 in M. caucasicus, 2 and 1 in M. eupeus, and finally, no recombination events in M. gibbosus.

Figure 1 Test results for recombination in B. mardochei scorpion mtDNA of a concatenated 16S and coxI ‘supergene’ using (a) the general recombination method (RDP), (b) the Geneconv method, (c) the Maxchi and (d) the Chimaera test statistics. The (more ...)

The choice of the measure of LD (|D′| or R²) had a strong impact on the correlations of the scorpion data: correlations using R² were significant in all four species analysed (table 2), whereas correlations using |D′| were only significant in B. mardochei (table 2). This observation is in agreement with criticisms on the evidence for recombination in human mtDNA (Awadalla et al. 1999; Kivisild & Villems 2000; Kumar et al. 2000), where strong correlations disappeared if |D′| was used instead of R². Following Meunier & Eyre-Walker's (2001) recommendation to limit the analysis to alleles with a frequency greater than 10% weakened the correlation in two cases and turned the correlation to non-significant in one species. All four tests using the intra-subspecific mtDNA data of B. mardochei, however, were significant, independently of the number of alleles considered or on how LD was estimated (table 2).

The homoplasy test was significant only in the species M. caucasicus (table 3). However, the assumptions of this test are less likely to be met since it assumes a specific model of sequence divergence, whereas the LD versus distance test is more robust to such violations (McVean 2001; Piganeau & Eyre-Walker 2004). Furthermore, the mean sequence divergence was greater than 5% in these datasets, which is higher than recommended (Maynard Smith & Smith 1998). Thus, the results of the homoplasy test should be interpreted with caution.

Figure 2 Schematic drawing of buthid versus non-buthid scorpion spermatogenesis (after Wilson 1916; Sato 1940).

Formation of the fused mitochondrial structure called Nebenkern during late spermatogenesis is known in many invertebrates and is well-studied in Drosophila (Fuller 1993). During flagellum elongation, the Nebenkern unfolds and results in two mitochondrial derivatives, later producing ATP to power the sperm's flagellum (Yaffe 1999). Mitochondrial fusion and subsequent elongation may allow coordinated ATP delivery to all regions of the flagellar axoneme. The Nebenkern in Drosophila forms during spermatid differentiation, i.e. after the meiosis is completed. On the contrary, the scorpion family Buthidae represents a unique case when mitochondria fuse much earlier (at the very beginning of meiosis) into a ring-shaped Nebenkern, which undergoes two regular divisions (Wilson 1916; Sato 1940). To the best of our knowledge, this phenomenon has not been reported in any other animal.

It is very likely that the fusion of mitochondria during the early stages of spermatogenesis provides increased opportunities for the mtDNA molecules to recombine, either through homologous recombination (intermolecular) or non-homologous recombination (intramolecular). The first case of non-homologous recombination of mtDNA has been reported in the nematode Meloidogyne javanica (Lunt & Hyman 1997). It was later shown in vitro that both types of recombination events can occur in human mtDNA (Lunt & Hyman 1997; Kraytsberg et al. 2004). Conditions for homologous recombination are met if at least two types of mtDNA molecules (haplotypes) are found in the same cell (heteroplasmy). In mammals, it is known that there are different mechanisms acting to prevent recombination of mtDNA (Rokas et al. 2003). Up to 100 paternal mitochondria enter the egg but are destroyed during the first few hours following fertilization. Several mechanisms ensure that the zygote contains only maternal mitochondria. However, it has been reported that the maternal inheritance of mitochondria in mammals seems to be merely a quantitative phenomenon, and situations of paternal leakage have been reported in mice (Gyllensten & Wilson 1987; Gyllensten et al. 1991). Failure of the recognition mechanisms for paternal mitochondria seems to be more common in hybrid populations, in which the specificity in the recognition process is relaxed, as was shown in drosophilids (Kondo et al. 1990).

Two recent papers provide strong support for actual recombination in human mtDNA. A recent reevaluation of lately added/corrected data found clear indirect evidence for recombination in human mtDNA using LD versus distance correlations (Piganeau & Eyre-Walker 2004). Moreover, a patient having paternal and maternal mtDNA (=heteroplasmic) in his muscle tissue was recently screened for recombined haplotypes using single template PCR (Kraytsberg et al. 2004). A frequency of approximately 0.7% of recombinant genotypes was reported, providing direct proof that human mtDNA is able to recombine. A recent compilation on published mtDNA sequence data from various vertebrates and invertebrates also showed indirect evidence for recombination in primates, insects and crustaceans (Piganeau et al. 2004).

The finding of strong negative correlation between LD and distance, together with the unique cytological evidence for mitofusion in early spermatogenesis, provides a strong suggestion for recombination in buthid scorpions. A combination of cytological methods to document scorpion spermatogenesis with controlled crossing experiments of populations of known haplotypes with subsequent genotype screening of offspring would provide a powerful direct approach to address recombination in buthid scorpion mtDNA (Ladoukakis & Zouros 2001; Rokas et al. 2003). If mtDNA recombination were directly linked to the phenomenon of mitofusion, non-buthids (e.g. Euscorpiidae, Scorpionidae) should not show any indirect evidence for recombination. Such a major difference between buthid and non-buthid scorpions would not be entirely unexpected as Buthidae represent a separate, ancient lineage within the scorpion phylogeny (Soleglad & Fet 2003) and possess a number of other unique, derived evolutionary mechanisms (such as mammal-specific neurotoxins).

B.G. was supported by an SNF-IHP grant (83-EU-065528), The Royal Society (R-36579) and the ‘Basler Stiftung für experimentelle Zoologie’ (c/o Eidgenössisches Tropeninstitut). Field trips of B.G. were supported by the commission for travel grants of the Swiss Academy of Natural Science (SANW). The field trip to Central Asia in March–May 2002 was supported by the National Geographic Society Research and Exploration grant 7001-01 to V.F. We thank Alexander V. Gromov for his help in the field and additional specimens of Mesobuthus. We also thank Gwenaël Piganeau, Michael Gardner and Adam Eyre-Walker for their helpful correspondence and for granting access to their unpublished manuscript that included evidence for recombination in scorpions based on a subset of M. eupeus. We are particularly grateful to Bill Amos who let us use his sequencer free of charge to finish the project when our finances had run out. Lily Jin ran three gels for us on the ABI377. B.G. would like to acknowledge Matt E. Braunwalder for his skilled guidance on scorpion biology and for leading him into the field of ‘scorpiology’.

One anonymous reviewer and Adam Eyre-Walker significantly improved an earlier version of the manuscript.