Doubly Uniparental Inheritance Is Associated With High Polymorphism for Rearranged and Recombinant Control Region Haplotypes in Baltic Mytilus trossulus

Abstract

Many bivalve species, including mussels of the genus Mytilus, are unusual in having two mtDNA genomes, one inherited maternally (the F genome) and the other inherited paternally (the M genome). The sequence differences between the genomes are usually great, indicating ancient divergence predating speciation events. However, in Mytilus trossulus from the Baltic, both genomes are similar to the F genome from the closely related M. edulis. This study analyzed the mtDNA control region structure in male and female Baltic M. trossulus mussels. We show that a great diversity of structural rearrangements is present in both sexes. Sperm samples are dominated by recombinant haplotypes with M. edulis M-like control region segments, some having large duplications. By contrast, the rearranged haplotypes that dominate in eggs lack segments from this M genome. The rearrangements can be explained by a combination of tandem duplication, deletion, and intermolecular recombination. An evolutionary pathway leading to the recombinant haplotypes is suggested. The data are also considered in relation to the hypothesis that the M. edulis M-like control region sequence is necessary to confer the paternal role on genomes that are otherwise F-like.

STRICTLY uniparental inheritance of organelle genomes is a rule in nearly all anisogamic organisms. One of the most prominent exceptions is the mitochondrial inheritance system of mussels of the family Mytilidae in which separate maternal (the F genome) and paternal (the M genome) routes of mtDNA inheritance occur (for review, see Skibinski et al.1994a,b; Zouros et al. 1994a,b; Zouros 2000). This system, called doubly uniparental inheritance (DUI) (Zouros et al. 1994a), has also been observed in freshwater mussels of the family Unionidae (Hoeh et al. 1996; Liu et al. 1996) and clams of the family Veneridae (Passamonti and Scali 2001). Phylogenetic analysis indicates that divergence of the F and M genomes can be great, predating speciation events, and that role reversal or masculinization events, whereby the F genome takes on the role of the previous M genome, has occurred repeatedly in the evolution of marine mussels (Hoeh et al. 1996, 1997), but is absent or less frequent in freshwater mussels (Hoeh et al. 2002).

A hybrid zone separates Baltic Mytilus trossulus from North Sea M. edulis populations (see Riginos and Cunningham 2005 for review). Although there is little introgression of mtDNA between American M. trossulus and M. edulis (Saavedra et al. 1996; Comesana et al. 1999) in Baltic M. trossulus, the mtDNA in heteroplasmic male individuals is similar to that in the F genome from M. edulis (Quesada et al. 1995, 2003; Wenne and Skibinski 1995; Zbawicka et al. 2003a,b). It appears that there has been complete asymmetric introgression of M. edulis F mtDNA into Baltic M. trossulus, accompanied by role reversal and masculinization (Rawson and Hilbish 1998; Quesada et al. 1999). These processes might be coupled and associated with cytonuclear incompatibilities that have occurred during hybridization (Quesada et al. 1999). Similar apparent incipient masculinization of the F genome has been reported in M. galloprovincialis from the Black Sea (Ladoukakis et al. 2002). Disruption of DUI, with homoplasmic males for the F genome, and F genome transmission through the sperm, has been observed frequently in crosses (Zouros et al. 1994b; Saavedra et al. 1997). Homoplasmic males for the F genome have also been observed in natural populations (e.g., Fisher and Skibinski 1990; Rawson et al. 1996), again suggesting paternal transmission of the F genome.

There is now growing evidence for widespread recombination in animal mtDNA (Piganeau et al. 2004; Tsaousis et al. 2005). In M. galloprovincialis from the Black Sea, experimental evidence exists for the generation of recombinant molecules within heteroplasmic individuals (Ladoukakis and Zouros 2001). In M. trossulus from the Baltic, paternally inherited recombinant molecules that possessed a small segment of a noncoding region showing high similarity with the M genome of M. edulis were observed, the remainder of the molecule being M. edulis F-like (Burzyński et al. 2003). This prompted the suggestion that part of this segment might confer a paternal role on a molecule that otherwise resembles the M. edulis F genome (Zouros 2000; Burzyński et al. 2003; Cao et al. 2004b). A comparative study of this noncoding region suggests that it is the main control region and thus functionally important and that different parts of it are subject to markedly different levels of selective constraint (Cao et al. 2004b).

Here we demonstrate that the control region of M. trossulus from the Baltic shows a great diversity of structural arrangements, including several different recombinant haplotypes with large duplications. The recombinant haplotypes include a segment of the M. edulis M-like genome and occur at polymorphic frequencies in sperm. By contrast, the rearranged haplotypes that dominate in eggs lack segments from the M genome. We consider the likely evolutionary pathways giving rise to these haplotype structures, the selective forces that might be acting, and the hypothesis that a part of the M control region sequence is necessary to confer the paternal role on masculinized genomes.

MATERIALS AND METHODS

Samples and DNA extraction:

Baltic M. trossulus samples were collected from the Gulf of Gdańsk at locations 54° 35′ 3″ N 18° and 33′ 51″ E, 10–24 m in depth. Mussels were sexed by microscopic examination of gonadal tissue. Fifty-two females and 68 males were analyzed by Southern blotting and PCR. Of these, 39 were chosen for sequencing to clarify haplotype structure. DNA was isolated from somatic tissues and gametes as described previously (Burzyński et al. 2003). Gametes were obtained from mature animals by vigorous washing of the mantle cavity with sterile seawater. The seawater was subsequently centrifuged at 500 × g to sediment gametes. After microscopic examination, verified gametes were suspended in STE100 buffer (0.1 m NaCl, 1 m EDTA, 0.05 m Tris–HCl, pH 8.0). For DNA isolation from somatic tissues, small pieces of gill were chopped on a microscope slide in a drop of STE100 buffer. Both gametes and somatic tissue samples were lysed with 10% SDS (600 μl of STE100 with 20 μl 10% SDS). Samples were phenol extracted and DNA precipitated with ethanol. DNA was suspended in TE buffer (1 mm EDTA, 0.01 m Tris–HCl, pH 8.0) at ∼1 mg/ml. DNA quality was checked on ethidium-bromide-stained agarose gels.

PCR amplification strategy:

PCR primers suitable for amplification of various parts of the major noncoding region/putative control region were designed using GenBank sequence records. Their location on comparative summary diagrams for this region obtained in earlier studies (Burzyński et al. 2003; Cao et al. 2004b) are shown in Figure 1. Universal primers flanking the control region (CBM1 and CBM2) (Figure 1) were used for the first amplification of DNA from gametes. A number of length variants were identified in accordance with earlier work (Zbawicka et al. 2003a). The amplifications were repeated with somatic DNA from the same individuals. Additional PCR products of different length were frequently observed in males, but not females, consistent with size heteroplasmy. If the results of PCR and Southern hybridization (see below) were compatible, then the individual haplotypes were named according to the size of the PCR product. In some instances, the CBM1–CBM2 amplification was poor or inconsistent with Southern hybridization. Such problems were resolved by examining long-range PCR products obtained with the flanking primers AB23–AB22 for the F genome and AB23–AB21 for the M genome. Further long-range PCR with other sets of F-specific (which did amplify) or M-specific (which did not amplify) primers were used to establish that the remainder of the mtDNA genome was F-like in recombinant haplotypes. To further elucidate the structure of length variants, the possibility of repeat structure was checked for by means of PCR with the AB32–AB16 primer pair. These primers are pointing in opposite directions (Figure 1), which allows detection of multiplication of the region in which they are situated. Further characterization was carried out by sequencing three shorter PCR fragments, reamplified from gel-purified long-range PCR product. First, the AB32–AB16 product was sequenced. When the region encompassing these primer sites is multiplicated, sequence information is gained for regions both 5′ and 3′ of the primer sites. From this sequence information, the new primers AB25, AB20, and AB35 were designed. Then, sequencing of the AB25–CBM2 PCR product provided information on the 3′ part of the control region and sequencing of the AB15–AB20 or AB15–AB35 PCR product provided information on the 5′ part of the control region.

Figure 1.—

Comparison of diagrams of the control region from the studies of Burzyński et al. (2003) (for M. trossulus) and Cao et al. (2004b) (for M. edulis and M. galloprovincialis) for M and F genomes. The positions of VD1, VD2, CD, lrRNA (large ribosomal RNA), cytB (cytochrome B), and trnY (tyrosine tRNA) genes are shown. The locations of primers, probes, and restriction enzyme cut sites (E, EcoRI; X, XbaI; H, HindIII) used in this study are shown below these diagrams. The universal probe (open box) hybridizes with both M and F genomes; the M-specific probe (solid box) does not cross-hybridize with the F genome. Hybridization probes were obtained as follows: the M-specific probe by reamplification from the M-specific AB15–AB16 PCR product with the AB25–AB26 primer pair, and the universal probe by reamplification from recombinant variant type 1 (see Figure 2) AB15–AB16 PCR product with primers AB15–AB35. Only the approximate position on the M and F genomes of AB35 (shown in parentheses) can be shown as it is specific to variant 1 (Figure 2).

The position, sequence, and annealing temperature (optimized in gradient PCR, T-gradient cycler from Biometra, Tampa, FL) for all primers are summarized in Table 1. Approximately 20 ng of total DNA was used for PCR amplifications, which were carried out in a volume of 20 μl with primers at 0.4 μm, nucleotides at 200 μm, magnesium chloride at 1.5 mm, and high-fidelity DyNAzymeEXT2 DNA polymerase (0.5 unit) and appropriate reaction buffer from Finnzymes. After an initial 3-min denaturation at 94°, 33 cycles were used with denaturation at 94° for 1 min, annealing for 30 sec (Table 1), and extension at 72° for 1.5 min with a final 5-min extension at 72°. PCR product (5 μl of each amplification) was visualized on 1% agarose gel stained with ethidium bromide. For long-range PCR, AccuPrime (Invitrogen, San Diego) polymerase was used according to the protocol of the manufacturer.

TABLE 1.

Summary of information on PCR primers

Name	Region	Strand	Lineage	Annealing temperature	Sequence	GenBank reference
CBM1a	lrRNA	F	F, M	61°	AGAACGGCGTGAGCTAGTTC	NC_006161 1445
CBM2a	cytB	R	F, M	61°	ACCTTCACCAGGCGTTTAAG	NC_006161 2946
AB15	lrRNA	F	F, M	66°	TTGCGACCTCGATGTTGG	NC_006161 1330
AB16	CD	R	F, M	66°	CAGGCTATAGAGCATAATCTAAAACG	AY115479 759
AB32	CD	F	F, M	66°	TGTCAGAGTCATGTGAGACTTAACC	NC_006161 2320
AB22	COIII	R	F	70°	GAAGACCTGTCTCAAACAGACTAGTAGATGA	NC_006161 7981
AB20	trnY	R	M	65°	GCCTTTTCCTCAGCCATCT	AY115482 1034
AB21	COIII	R	M	70°	CTGTCTCGAACAGCCTTGTGGAC	U50217 367
AB23	lrRNA	F	F, M	70°	AAGATTGCGACCTCGATGTTGG	NC_006161 1346
AB25	VD1	F	M	61°	CGCTTAACTTCCCTGCCA	AY115482 385
AB26	VD1	R	M	61°	TCTAAAACGAGGTATGG	AY115482 588
AB35	VD1	R	1/1c	65°	GCCTTTTCCTCCGCCATC	AY115481 236

^aFrom Burzyński et al. (2003).

Restriction enzyme digestion and Southern hybridization:

Southern blots with three different restriction enzymes and two probes of different specificity were used to confirm and clarify the variant lengths determined by PCR and to minimize the chance of erroneous interpretation due to PCR reaction artifacts or amplification of nuclear inserts of mtDNA. The location of the restriction sites and probes used are shown in Figure 1. Total DNA was digested with restriction endonucleases (EcoRI, XbaI, HindIII) according to the protocol from Promega (Madison, WI). Digests were separated on 0.7–1.5% agarose gels followed by denaturation in 0.4 m NaOH and capillary transfer to Zeta-probe nylon membranes (Bio-Rad, Hercules, CA). Hybridization probes were synthesized by PCR with DIG-dUTP (Roche). Labeling was carried out in a volume of 20 μl with primers at 0.4 μm; dATP, dCTP, and dGTP at 200 μm; dTTP at 175 μm; DIG-dUTP at 25 μm; magnesium chloride at 1.5 mm; 1 unit TaqI DNA polymerase (Promega); and 20 ng of DNA in appropriate reaction buffer. After an initial 3-min denaturation at 94°, 33 cycles were used with denaturation at 94° for 1 min, annealing for 30 sec, and extension at 72° for 1.5 min with a final 5-min extension at 72°. Hybridization and detection were carried out with the DIG-starter kit from Roche according to the protocol of the manufacturer. For hybridization, an overnight incubation at 42° was used. A high-stringency wash at 68° in 0.1% SDS, 0.1× SSC was used to ensure high specificity of signal.

Sequencing and alignment:

Selected PCR products were purified by alkaline phosphatase and exonuclease I treatment and sequenced directly with BigDye kit from PE-Biosystems using appropriate primers. Sequence assembly was facilitated by Phred (Ewing et al. 1998) and Staden (Bonfield et al. 1995) computer programs. Sequences were aligned using the ClustalW algorithm (Higgins and Sharp 1989; Thompson et al. 1994).

Dot-plot and recombination detection analysis:

Sequences must be aligned before recombination detection programs can be applied. In this study the sequences have different lengths and are not collinear, and thus alignment is difficult. Dot plots provide a means of guiding the alignment of the sequences and making a preliminary assignment of regions in the haplotypes to reference F and M sequences. The recombination detection programs can then be used to pinpoint the location of recombination breakpoints more precisely. Local alignments of relevant fragments from the sequences were constructed for each potential recombination breakpoint for variants 1, 1a, 1c, and 11a and compared with the reference F and M sequences. Each alignment was analyzed by a series of recombination detection programs: Geneconv (Padidam et al. 1999), Chimaera, MaxChi (Maynard Smith 1992, Posada and Crandall 2001), SiScan (Gibbs et al. 2000), Bootscan (Salminen et al. 1995), and RDP, all as implemented in RDP v2 beta8 (Martin and Rybicki 2000). The default set of parameters was used for each program but sequences were considered linear (the default is that sequences are considered circular). The Bonferroni correction was applied to the set of P-values associated with recombination breakpoints generated duing the analysis of the data set. The analyses were run in automatic mode by RDP. Only recombination events detected by two or more methods were considered. The most conservative measures for probability of false positives were used: global KA p-value for Geneconv, binominal probability for Bootscan, and region probability for MaxChi and Chimaera.

Phylogenetic tree construction:

Lateral transfer occurring through intermolecular recombination or other mechanisms such as transposition complicates the interpretation of evolutionary trees. Comparison of trees built from segments involved in such transfer with those not involved, such as sequence flanking the control region, could provide information about the history of transfer. In this study, sequence flanking the control region is F-like and insufficiently variable for tree construction. Within the control region (see Figure 1), the best candidate for tree construction is the variable domain 1 (VD1). Another potential candidate is the conserved domain (CD), but this displays little variation and is little diverged between the F and M genomes as noted previously (Burzyński et al. 2003; Cao et al. 2004b). Hence phylogenetic trees were constructed using VD1, which has a 473-bp segment, apparently unbroken by recombination, which could be aligned for the observed haplotypes. Phylogenetic relationships and genetic distances (Kimura 2 parameter) were calculated using MEGA3 (Kumar et al. 2001), Phyml (Guindon and Gascuel 2003), and MrBayes (Ronquist and Huelsenbeck 2003) software packages. The best model of sequence evolution was chosen on the basis of Modeltest (Posada and Crandall 1998). The TRN + G model was used, with gamma shape parameter at 0.1729. All trees were bootstrapped whenever possible, with at least 100 iterations. The Markov Chain was run with nst = 2 and gamma distributed rates for 1,000,000 generations. Data for the first nonstationary phase of the run were discarded; 50% majority rule trees with branch lengths are presented.

Nomenclature of genomes:

Throughout, “F” stands for the maternally inherited genome of M. edulis and the similar maternally and role-reversed paternally inherited genomes of M. trossulus, which are thought to derive from it: “M” stands for the paternally inherited genome similar to that of M. edulis which, it is argued, has contributed to the control region sequence of the recombinant haplotypes observed in M. trossulus.

RESULTS

Structure of the control region revealed by PCR, Southern hybridization, and sequencing:

PCR with primers flanking the control region suggested considerable length variation among mussels in this region. This was confirmed by the Southern hybridization analysis. This latter technique also indicated that many of the haplotypes had both M-like and F-like sequences. For example, for haplotype 11a, a single band of 4220 bp was obtained with the M probe and XbaI digestion, but with the universal probe (which encompasses the XbaI site shown in Figure 2) an additional band was observed (Figure 3). Other haplotypes, for example, 2 and 190, gave hybridization signals only with the universal probe, suggesting the presence of only F-like sequence. PCR with primers AB32–AB16 confirmed that all the longer variants contained multiplication of at least part of the control region. In some instances, the AB32–AB16 sequence was heterogeneous as a result of the presence of nonidentical repeats. This was accompanied by the presence of additional band(s) at ∼1 kb in the XbaI blots (Figure 3, for haplotype 15). Because the central XbaI site is located in the M-specific VD1 region, these results are consistent with the presence of multiple copies of VD1–CD differing slightly in length in some of these variants. Such variants (with multiple copies of M-like VD1–CD, regardless of the number or completeness of copies) were assigned as haplotype 15. For four representatives of haplotypes 1a, 1b, 11a, and 1, and at least one representative of haplotypes 2, 190, 120, 80, and 3a, a full sequence was obtained from within the lrRNA gene, 5′ to the control region, to within the cytochrome B gene, 3′ to the control region. For other haplotypes, less complete sequence data were obtained. For example, the available sequenced PCR fragments for haplotypes 15 and 1c could not be combined into a full sequence because of the large repeat encompassing VD1 and CD. Nevertheless, the experimental strategy as a whole allows the characterization and positioning of segments within the control region.

Figure 2.—

Diagrams of M. trossulus haplotypes. Darkly shaded and open boxes represent M-related and F-related regions, respectively, identified without ambiguity by at least two recombination detection programs; lightly shaded boxes represent regions that could not be similarly assigned without ambiguity. All variants are aligned at the VD1/CD boundary. Back-to-back arrows indicate the position of the AB16 and AB32 PCR primers used to detect repeated units. “X” marks the XbaI site important for detecting length variation in Southern blot analysis (see Figure 3). The larger boxes represent units that are tandemly repeated at least twice in representatives of that haplotype (see text) although only one of each repeated unit (the RH one) is shown for simplicity. Arrowheads on boxes denote the ends of lrRNA and trnY genes. Haplotypes were named following the system used in previous studies. Haplotypes 1 and 1a were reported by Burzyński et al. (2003). Haplotype 11a is a variant of haplotype 11 reported by Zbawicka et al. (2003a) where haplotypes were characterized only on the basis of PCR product length. Haplotype 80 is the pool of variants 3b, 5, and 7 (with 80-bp repeats of 2, 4, and 5 copies, respectively); haplotype 120 is the pool of variants 4, 6, 7b, 9, 10a, and 10b (with 120-bp repeats of 2, 3, 4, 5, 6, and 7 copies); and haplotype 190 is the pool of variants 5, 11, 12, 13, 14 (with 190-bp repeats of 2, 6, 7, 8, and 9 copies) of that same study. Two new long variants with multiplicated regions, haplotypes 15 and 16, were discovered in this study. Potential breakpoints identified by the RDP suite of programs are indicated as bp1, bp2, and bp3 for the relevant haplotypes.

Figure 3.—

An example of Southern hybridization analysis. DNA from gametes from M. trossulus was digested with XbaI and hybridized with the M-specific probe (top). A second hybridization with the universal probe was performed without stripping off the first probe (bottom). Marker fragment sizes (in base pairs) are given at the left. The arrow indicates the position of additional ∼1-kb bands (variable in size) for haplotype 15, resulting from VD1M multiplication.

The resulting haplotype structures shown in Figure 2 indicate a wide variety of structures with many haplotypes containing multiplicated regions. Haplotypes 2 and 1b are, respectively, the normal F and M genomes characteristic of M. edulis. Haplotypes 1a, 11a, 15, 1, and 1c have both F-like and M-like sequence. Haplotype 15 has a large M-like multiplicated region encompassing part of VD2, trnY, the terminal part of lrRNA, VD1 and CD. A similar multiplicated region occurs in 1c but lacks the lrRNA segment and has a shorter VD1. The remaining haplotypes are entirely F-like in the studied region. Haplotypes 80, 120, and 190 have repeats at the 3′-end of the control region that are, respectively, 80, 120, and 190 bp long. The number of repeats for these haplotypes varies from two to nine, but these variants are not considered separately here. Haplotype 3a lacks these 3′ repeats but has a small tandem duplication of the central part of VD1. Haplotype 16 has a large F-like multiplicated region that encompasses the 3′ part of VD1, the whole CD, VD2, trnY, and the beginning of the cytB gene.

Clarification of recombination breakpoints:

The dot-plot analysis was used to further clarify the control region haplotype structures. It suggested that the presence of fragments originated from different lineages in haplotypes from groups 1, 1c, 1a, 11a, and 15. Two examples of dot plots are shown in Figure 4 for groups 11a and 1c. These examples indicate repeats and several potential breakpoints for both haplotypes. For some DNA segments, solid lines on the dot plot indicate similarity to both F and M sequences. This is mainly a consequence of the F and M genomes having low divergence in these regions (Figure 1 “core” region). Establishing the exact position of recombination breakpoints within these regions is problematical because of this low divergence. When regions of low divergence are flanked by much more divergent fragments, recombination detection programs have difficulties in accurately identifying “parent and daughter” sequences and in producing realistic P-values for potential breakpoints. The recombination signal was strong in all alignments analyzed using the RDP suite of programs. Representative P-values for breakpoints identified with Geneconv, Bootscan, Chimaera, and MaxChi are given in Table 2. A range for the P-value is given where different sequences within an alignment gave different P-values for the relevant breakpoint. The breakpoints are also shown on the haplotype structures of Figure 2. They have also been indicated in the GenBank record annotation of the sequences.

Figure 4.—

Dot-plot analysis for two M. trossulus recombinant haplotypes 11a and 1c compared with the typical reference M (from M. edulis) and F (from M. trossulus) sequences used previously (Burzyński et al. 2003). A 20-bp window with 80% threshold was used.

TABLE 2.

Statistical support for recombination breakpoints

Haplotype	Breakpoint	Geneconv	Bootscan	Chimaera	MaxChi
1a	bp1	1.8 × 10−3–4.9 × 10−18	10−16–2.7 × 10−21	10−15–1.2 × 10−17	5.0 × 10−15–6.0 × 10−16
11a and 15	bp1	2.4 × 10−5–1.2 × 10−8	1.3 × 10−8–1.5 × 10−7	1.1 × 10−9–1.2 × 10−10	3.6 × 10−11–2.6 × 10−12
	bp2	8.5 × 10−12–3.6 × 10−15	5.0 × 10−13–1.8 × 10−13	2.7 × 10−17	2.6 × 10−17–4.0 × 10−17
	bp3	3.6 × 10−15–6.3 × 10−17	1.5 × 10−15–2.3 × 10−17	10−13	2.6 × 10−13–10−13
1 and 1c	bp1	3.9 × 10−6–8.5 × 10−5	3.7 × 10−5	2.8 × 10−8–1.1 × 10−9	8.0 × 10−9–2.2 × 10−9
	bp2	7.5 × 10−5–5.8 × 10−3	7.5 × 10−3–2.3 × 10−6	5.7 × 10−6–1.9 × 10−8	2.0 × 10−5–4.0 × 10−9
	bp3	1.3 × 10−17–3.9 × 10−17	1.8 × 10−6	1.3 × 10−12	3.8 × 10−13–6.2 × 10−13

For haplotype 1a, only one recombination breakpoint (bp1) was identified. Although all programs indicated high probability of this breakpoint, they did not agree on its exact position, which varies from the VD1/CD boundary (Chimaera and MaxChi) to the end of the “core” (Bootscan, for one of the 1a sequences). For haplotypes 11a and 15, there are three recombination breakpoints identified: the first (bp1) marks a change from F to M sequence, the second (bp2) marks a change from trnY to lrRNA, and the third (bp3) marks a change from M to F. The first occurs within the first CD with good agreement across programs and sequences. The second is also very well defined, occurring within an AACTCAT motif present in both lrRNA and trnY of the reference M sequences. As indicated in Figure 2, this breakpoint has more than one potential position in haplotype 15 due to the repeat structure of this haplotype. The third breakpoint was positioned at the “core” boundary with high probability for all programs. For haplotypes 1 and 1c, three breakpoints were identified, the second (bp2) having more than one potential position in haplotype 1c due to the repeat structure of this haplotype. Since the distance between the first and the second breakpoints was small (24 bp), the alignments used for RDP were also shorter than usual (140 and 153 bp), which can reduce the probability of significant results. Nevertheless, the support for each of these two breakpoints is good. The first of these defines the boundary between VD1F and a VD2M fragment, and the second indicates the switch between the fragment of trnY and VD1M. The third is located unambiguously in the same area of VD2 as bp3 in haplotypes 11a/15.

Haplotype frequencies:

The frequencies of haplotypes in males and females are given in Table 3. Using a Monte Carlo association test (Roff and Bentzen 1989), the distribution of haplotypes is significantly different between the sexes (P < 0.0001). Variants 1a (with one exception), 1b, 11a, 15, 1, and 1c occur only in sperm, whereas variants 2, 190, 80, 120 (with one exception), and 3a occur only in eggs. The haplotypes predominating in sperm, excepting 1b, have regions deriving from both the M and F genomes (Figure 2). By contrast, the haplotypes predominating in eggs derive from the F genome. An exception is the F-like haplotype 16, which occurs at a frequency of 10% in sperm but is absent from females.

TABLE 3.

Haplotype frequencies in eggs and sperm

	Eggs		Sperm		Total
Haplotype	Frequency	%	Frequency	%	Frequency	%
1a	1	1.9	22	32.4	23	19.2
1b	0	0.0	4	5.9	4	3.3
11a	0	0.0	12	17.6	12	10.0
15	0	0.0	15	22.1	15	12.5
1	0	0.0	5	7.4	5	4.2
1c	0	0.0	2	2.9	2	1.7
2	20	38.5	0	0.0	20	16.7
190	12	23.1	0	0.0	12	10.0
80	2	3.8	0	0.0	2	1.7
120	16	30.8	1	1.5	17	14.2
3a	1	1.9	0	0.0	1	0.8
16	0	0.0	7	10.3	7	5.8
Total	52	100.0	68	100.0	120	100.0

VD1 phylogenetic trees:

The topologies of the phylogenetic trees based on the VD1 alignment were similar for all tree-building methods used. The Bayesian trees are presented in Figure 5. The F and M clades are rooted on each other in Figure 5A. Recombinant haplotypes form three separate clades for the M-like VD1, one for 1a, one for 11a and 15, and one for the 1 and 1c haplotype groups. Similarly, haplotype groups 11a and 15 form a separate clade on the F clade. Haplotypes 11a and 15 have copies of VD1 on both the M and F clades. The mean between-group distance between the M and F clades is 0.60 ± 0.07. Putting such divergent sequences together on one tree results in low resolution and weak statistical support for clades. Moreover, the exact rooting points are not reliable. Thus, to improve the resolution of recombinant F and M clades, separate unrooted trees have been constructed (Figure 5, B and C). Further increase in resolution can be achieved by expanding the length of alignment for variants that have longer unbroken fragments in common. Using this approach, each of the three recombinant clades mentioned above can be recovered with posterior probability close to 1.0. Thus, for the M tree, the 14 1a, 1b, 11a, and 15 haplotypes could be aligned over 514 bp. In the Bayesian tree constructed from this alignment (not shown), the probability of partition for the 11a, 15 clade has a value of 1.00. The 8 1a, 1b clade haplotypes could be aligned over 575 bp. The separation of the 1a and 1b clades on this tree (not shown) has a probability of 0.99.

Figure 5.—

Phylogenetic trees of VD1 sequences constructed using Bayesian analysis. (A) Tree based on a 473-bp alignment of sequences from M. trossulus with the M and F clades rooted on each other. Included in the tree are two VD1 sequences for M. galloprovincialis from the Black Sea (Smietanka et al. 2004) and one from the Mediterranean (Cao et al. 2004b). (B and C) Unrooted trees for F and M clades, respectively. Posterior probability densities of bipartition are given on internal branches. Labels on terminal nodes represent the last digits of GenBank accession numbers. For labels 22–58, accession numbers are DQ198222–58; for labels 79–81, AY115479–81; for labels 62 and 65, AY629162 and AY6291665; for label 92, AY497292. Labels that are repeated and in boldface type indicate control region sequences with both M-like and F-like VD1 regions. Other labels that are repeated indicate that more than one VD1 region has been sequenced, e.g., two M-like VD1 regions for 41 on the lowest clade in C.

The largest between clade distance for the F tree (Figure 5B) is 0.030 ± 0.008 for the comparison of 11a and the group of F haplotypes (2, 190, 120, 80, 3a, and 16). The smallest at 0.002 ± 0.001 is the average pairwise distance among these F haplotypes. The largest distance on the M tree (Figure 5C) is 0.024 ± 0.009 for the comparison of 11a and 1b, which is similar to the largest distance on the F tree, but the within-clade distances are much higher for M, ranging from 0.008 ± 0.004 for the 11a/15 clade to 0.016 ± 0.006 for the 1b clade.

DISCUSSION

Rearrangement mechanisms and evolutionary origin of recombinant haplotypes:

A variety of mechanisms have been proposed to explain mtDNA rearrangements (Stanton et al. 1994; Lee and Kocher 1995; Boore 2000; Mueller and Boore 2005), including tandem duplication via slipped-strand mispairing followed by random loss or degeneration of redundant DNA (e.g., in lizards, Stanton et al. 1994; in ticks, Shao et al. 2004; in salamanders, Mueller and Boore 2005), nonhomologous intramolecular recombination (e.g., in nematodes, Lunt and Hyman 1997), homologous intermolecular recombination (e.g., in the flounder, Hoarau et al. 2002), and nonhomologous intermolecular recombination (e.g., in mites, Shao et al. 2005). The rearrangements observed in this study can be explained in a parsimonious manner by three processes: tandem duplication, deletion, and intermolecular recombination occurring one or more times during the evolution of the haplotypes. The first process has occurred in both maternal and paternal genomes, but the second and third processes occurred only in paternal genomes. In several of the recombinant haplotypes, putative duplications appear to have occurred within the trnY gene. The potential for forming stem and loop structures associated with tRNA genes could play a role in the duplication processes (Stanton et al. 1994) and cause illicit priming of replication (Cantatore et al. 1987).

The precise evolutionary pathway leading to the haplotypes predominating in sperm is unclear. One parsimonious explanation, shown in Figure 6, is that a tandem duplication occurred within the M genome, beginning with the terminal part of the lrRNA gene and including at least VD1 and CD and ending within the trnY gene. This structure (intermediate 1) has not been observed in this study. A large segment of intermediate 1 could then have been transferred by homologous recombination to the F genome to produce haplotype 11a. Subsequent tandem multiplication could then have given rise to haplotype 15. Haplotype 1 could have been derived from 11a by two deletions. The first, involving the terminal part of lrRNA and part of VD1, leads to intermediate 2, again not observed in this study. The second deletion, involving part of the 5′ VD1, CD, and part of VD2, leads to haplotype 1. Finally, 1c could have been derived from 1 by further duplication events. Haplotype 1a does not fit into this evolutionary pathway, for example, because it does not have an M-like CD–VD2. It could have been produced independently by reciprocal homologous recombination between ancestral F and M genomes.

Figure 6.—

Hypothetical evolutionary pathway leading to recombinant haplotypes. The labeling follows that in Figures 1 and 2. See text for further explanation.

Phylogeny of haplotypes based on VD1 analysis:

The VD1 tree provides some further information relating to the evolutionary origin of the haplotypes. The occurrence in haplotypes 11a and 15 of one copy of VD1 on the M clade and another on the F clade provides strong support for the recombinant nature of these haplotypes in addition to that provided by the dot-plot analysis and analysis with the recombination detection programs. Although the root of the M clade is not well resolved, and thus caution is needed, the tree structure is consistent with haplotype 1b being ancestral, and it, or a derivative, giving rise separately to haplotype 1a and to the complex haplotypes 11a, 15, 1, and 1c. This interpretation is also consistent with the pathway shown in Figure 6. Thus joint consideration of Figures, 2, 5, and 6 lead to the hypothesis that two recombinant haplotypes have evolved independently and that they or their descendants have become established in the male lineage. A long branch leads to the 11a and 15 haplotypes on the F part of the tree but, surprisingly, the F-like VD1 sequence of haplotypes 11a/15 is not derived from any F haplotype currently in the Baltic. Instead, it groups (with high probability) with haplotypes found in Black Sea M. galloprovincialis. As there are no reports of Baltic haplotypes in the Black Sea, the tree structure is consistent with unidirectional introduction of Black Sea haplotypes into the Baltic. This could have been the result of stepwise introgression along the European coast or of transport by, for example, humans or birds.

Heteroplasmy and the generation and detection of recombinants:

The observation of recombinant haplotypes in this study can be related to the high level of heteroplasmy for distinguishable genomes in Mytilus males. Heteroplasmy for different mtDNA genomes would seem to be a prerequisite for the experimental detection of recombinants (Rokas et al. 2003). Generation of recombinants at high frequency was demonstrated in heteroplasmic male individuals of M. galloprovincialis from the Black Sea, although none of these had become established in the populations studied (Ladoukakis and Zouros 2001). Recombination has also been observed in heteroplasmic humans. A rare case of paternal transmission of mtDNA was reported in a man with mitochondrial myopathy (Schwartz and Vissing 2002). Subsequently, recombinant molecules were detected in this individual (Kraytsberg et al. 2004). It was suggested that the recombinants might arise from template switching during replication. Recombination has also been reported in the skeletal muscle of humans with mtDNA D-loop heteroplasmy (Zsurka et al. 2005). In these examples it could be assumed that parental mtDNA molecules are brought together by mitochondrial fusion. This process is known to occur in spermatogenesis in scorpions where evidence for mtDNA recombination has also been reported (Gantenbein et al. 2004). In Mytilus also, sperm contain five large mitochondria that are the result of previous fusion during spermatogenesis (Longo and Dornfeld 1967). In Baltic M. trossulus, there is experimental evidence for paternal cotransmission of different paternal haplotypes within heteroplasmic mitochondria of sperm (Quesada et al. 2003). Heteroplasmic sperm were not observed in this study and, given the evidence of duplication near to and involving the lrRNA gene, the possibility of an artifact due to lrRNA sequence duplication should perhaps be investigated in the future.

Influence of Baltic genetic structure:

The genetic structure of Baltic M. trossulus could have played a role in the generation of masculinized recombinant haplotypes. There is evidence for reduced efficiency of mitochondrial functioning in interspecific cells (e.g., rat mtDNA in mouse mtDNA-less cells; McKenzie and Trounce 2000), suggesting co-evolutionary constraints on the nuclear and mitochondrial gene products. The general importance of the interaction of nuclear and mitochondrial-encoded proteins in hybrid fitness and breakdown is also supported by studies of the electron transport system in the marine copepod Tigriopus californicus (Harrison and Burton 2006; Rawson and Burton 2006). If many Baltic mussels have nuclear genomes that are partly introgressed (Riginos and Cunningham 2005), loss of compatibility between the native Baltic paternal genome and a partial M. edulis genetic background might have rendered the native paternal genome more susceptible to replacement by the invading F genome from M. edulis. The accumulation of slightly deleterious mutations by the native Baltic paternal genome might also have increased its susceptibility to replacement by a fitter invading F genome from M. edulis (Quesada et al. 1999). Acquisition by the F genome of M-genome control region sequence might have been an additional condition for successful invasion of the paternal line.

DUI, sex determination, and the function of masculinized haplotypes:

A model relating DUI and sex determination consistent with available evidence has been proposed (Saavedra et al. 1997; Kenchington et al. 2002). In this model, sex is determined by the nuclear genotype of the mother, and factors responsible for differentially labeling male and female mitochondria and thereby determining their fate in the fertilized egg are postulated. In male embryos, it appears that sperm mitochondria aggregate and may be partitioned into early germ cells, whereas they are dispersed in a stochastic manner in female embryos (Cao et al. 2004a). The model does not demand that the mtDNA itself is required to label sperm mitochondria, nor does it demand that masculinization requires specific signatures in the mtDNA genome. In DUI, the sperm mitochondria would carry and pass on whichever mtDNA molecules happen fortuitously to be there.

It is nevertheless possible that possession of part of the M-like control region confers a paternal role on mtDNA molecules, as proposed previously (Burzyński et al. 2003; Cao et al. 2004b). This conjecture was based on the observation that masculinized F-like mtDNA molecules in M. trossulus possess M-like segments in this region. Although there is evidence from comparative sequence analysis that this region has functionally important motifs and is the main control region, there is yet no evidence that particular motifs are implicated in masculinization. The observation of a strong statistical association between gender and presence or absence of recombinant molecules with M-like sequence is not evidence of a functional link. Rather, it would be necessary to compare how often in evolution recombinants become established in the male lineage as compared with nonrecombinants, irrespective of whether the generation de novo of recombinants is a rare event or is common, as in the study of Ladoukakis and Zouros (2001). A test is difficult even with such data, because the null hypothesis, relating a priori probabilities of establishment for recombinant and nonrecombinant molecules, is completely unclear. In this study there is evidence for two independent evolutionary events leading to the haplotype 1a and to the group of haplotypes 11a, 15, 1, and 1c, an insufficient number to support a statistically adequate test. Finally, it is important to note that haplotype 16, which contains a large control region duplication but apparently lacks M-like sequence in this region, has a frequency of 10% in males. This observation essentially refutes the hypothesis that M-like sequence is necessary for masculinization. It could, however, still be consistent with the hypothesis that M-like sequence in the control region increases the likelihood of masculinization. The possibility that M-like sequence might play roles in DUI other than the labeling of sperm mitochondria should also be considered. For example, it could be that M elements are necessary for sperm mtDNA replication following fertilization, without which paternal propagation would be impossible. Genomes with these M elements, however, could still be inherited maternally if they occur in mitochondria that are labeled for maternal transmission. This is speculation, but consistent with recent observations in American M. trossulus where both masculinized and regular F genomes are recombinant and contain M-like elements in the D-loop region (Rawson 2005; Breton et al. 2006). In this study, there are two other exceptional individuals that do not fit the general pattern of association of sex and haplotype: an egg sample with haplotype 1a and a sperm sample with haplotype 120. Such instances are frequently observed in diverse studies on DUI, and theory must be sufficiently unrestrictive to accommodate these. Sperm carrying F-like haplotypes could also be an artifact due to somatic tissue contamination.

Selective forces acting on paternal mtDNA:

An unresolved question is whether masculinization is a sporadic stochastic event or whether new paternal genomes have a selective advantage over the old that they replace (Hoeh et al. 1997), irrespective of the role of M-like sequence in the control region. The question invites some consideration of the possible selective forces acting on paternal mtDNA in general. Several observations favor the view that the paternal genome is fully functional and under selection. Thus both paternal and maternal genomes are expressed in male and female tissues (Garrido-Ramos et al. 1998; Dalziel and Stewart 2002). Analysis of synonymous and nonsynonymous diversity and divergence within the COIII gene suggests purifying selection, but at a lower level for the paternal compared with the maternal genome (Stewart et al. 1996; Quesada et al. 1998). The paternal genome evolves faster than the maternal genome, which could reflect relaxed selection, as there is as yet no evidence for positive selection (Stewart et al. 1996). Unfortunately, because the control region is noncoding, these alternatives cannot be tested by tests of synonymous and nonsynonymous variation. A possible feature of the variation observed here that is relevant to selection relates to observations that some duplicated control regions in mtDNA can be very long lived with little evidence of decay, suggesting concerted evolution via gene conversion and possible selective maintenance of the duplications (in snakes, Kumazawa et al. 1996; in parrots, Eberhard et al. 2001; in the albatross, Abbott et al. 2005). In human patients with rearranged mtDNA it has been observed that replication proceeds from both duplicated origins, which might confer a replicative advantage on such molecules (Umeda et al. 2001). In this study, the most frequent paternal haplotype (1a at 32% in sperm), although recombinant, does not have any evident duplication within the control region. Thus it must have reached this high frequency for some other reason. There is a potentially important difference between DUI and normal uniparental inheritance of mtDNA that is worth noting in relation to selection on the paternal genome. With normal maternal inheritance of mtDNA, mutations with small effect on female fitness could reach high frequency, even if there is a substantial deleterious effect on sperm fitness (Frank and Hurst 1996; Gemmell and Allendorf 2001). The situation with the paternal genome under DUI is likely to be entirely different, as paternal inheritance results in exposure of this genome to strong selection either in males or in sperm. An experimental test for differences in swimming speeds of sperm between standard paternal genomes and recently masculinized genomes has already been made (Everett et al. 2004). Although the genomes diverged by 8.7% in amino acid sequence, giving good potential for functional differences, no significant swimming speed differences were in fact observed. In conclusion, all the evidence obtained so far pertaining to selection is consistent with purifying selection on paternal mtDNA. Certainly, there is so far no evidence pointing to functionally different sperm variants on which frequency-dependent selection might act to cause the observed polymorphism. Thus the most plausible explanation of the high diversity reported here is that the variants are neutral or nearly neutral and that the high diversity reflects a high mutation rate to new variants in the face of some purifying selection.

Future work:

The control region has not been checked for evidence of recombination in almost all studies focusing on differences between “normal” and masculinized genomes; thus the identification of sequence signatures that are actively selected during masculinization might be advanced by further comparative studies in bivalves. In addition, significant progress in understanding DUI and masculinization might now be made by experimental approaches to identifying functional differences between maternal and paternal genomes.