Persistence of Mhc Heterozygosity in Homozygous Clonal Killifish, Rivulus marmoratus: Implications for the Origin of Hermaphroditism

Corresponding author: Akie Sato, Abteilung Immungenetik Corrensstrasse 42, D-72076 Tübingen, Germany., akie.sato{at}tuebingen.mpg.de (E-mail)

Communicating editor: N. TAKAHATA

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The mangrove killifish Rivulus marmoratus, a neotropical fish in the order Cyprinodontiformes, is the only known obligatorily selfing, synchronous hermaphroditic vertebrate. To shed light on its population structure and the origin of hermaphroditism, major histocompatibility complex (Mhc) class I genes of the killifish from seven different localities in Florida, Belize, and the Bahamas were cloned and sequenced. Thirteen loci and their alleles were identified and classified into eight groups. The loci apparently arose 20 million years ago (MYA) by gene duplications from a single common progenitor in the ancestors of R. marmoratus and its closest relatives. Distinct loci were found to be restricted to different populations and different individuals in the same population. Up to 44% of the fish were heterozygotes at Mhc loci, as compared to near homozygosity at non-Mhc loci. Large genetic distances between some of the Mhc alleles revealed the presence of ancestral allelic lineages. Computer simulation designed to explain these findings indicated that selfing is incomplete in R. marmoratus populations, that Mhc allelic lineages must have diverged before the onset of selfing, and that the hermaphroditism arose in a population containing multiple ancestral Mhc lineages. A model is proposed in which hermaphroditism arose stage-wise by mutations, each of which spread through the entire population and was fixed independently in the emerging clones.

THE mangrove rivulus (killifish), Rivulus marmoratus Poey1880, is a small fish inhabiting temporal pools and land-crab burrows in the coastal mangrove swamps of northeastern Brazil, the Gulf of Mexico, the Caribbean, and the southern part of Florida (HUBER 1992 ). Its main distinction is its hermaphroditism—the ability of single individuals to develop both ovaries and testis (ovotestis) and produce both eggs and sperm (HARRINGTON 1961 ; KALLMAN and HARRINGTON 1964 ; HARRINGTON and KALLMAN 1968 ). The fish reproduce by releasing synchronously ovulated eggs and sperm, fertilizing the eggs internally, and laying the embryos after a short period of development in the body (HARRINGTON 1961 , HARRINGTON 1963 ; ATZ 1965 ). R. marmoratus is the only vertebrate known to reproduce by obligatory self-fertilization or selfing (VRIJENHOEK et al. 1989 ). Although laboratory-reared individuals may pass through an early transient phase of their development in which they possess only the ovarian component of the ovotestis (COLE and NOAKES 1997 ), there is no evidence that these "females" are reproductively active (HARRINGTON 1975 ; SOTO et al. 1992 ).

Male R. marmoratus can arise in one of two principal ways. The so-called "primary males" can be produced at high frequency in the laboratory by incubating developing embryos at temperatures between 18° and 20° (HARRINGTON 1967 ; HARRINGTON and KALLMAN 1968 ; LIN and DUNSON 1995 ). They have no female reproductive organs and so produce only sperm (HARRINGTON 1968 ). "Secondary males" or "false male gonochorists," on the other hand, arise from either immature or mature functional hermaphrodites by the loss of the female function in response to shortened photoperiods (HARRINGTON 1968 ). Males are either absent or exceedingly rare in most natural populations (HARRINGTON and KALLMAN 1968 ; DAVIS et al. 1990 ). The only two known exceptions to this rule are localities in the Netherlands Antilles (KRISTENSEN 1970 ) and in some of the Belize cays, most notably the Twin Caye, in which the frequency of males approaches 25% (DAVIS et al. 1990 ; TURNER et al. 1992A ).

Populations of R. marmoratus are assemblages of clones distinguishable by an exchange of tissue grafts (KALLMAN and HARRINGTON 1964 ; HARRINGTON and KALLMAN 1968 ), electrophoretic enzyme surveys (MASSARO et al. 1975 ; VRIJENHOEK 1985 ), and DNA fingerprinting (TURNER et al. 1990 ; LAUGHLIN et al. 1995 ). Individuals comprising a clonal lineage are indistinguishable by any of these techniques and highly homozygous. The composition of the clonal lineages changes rapidly, often from year to year, presumably because of frequent migrations of individuals between populations (TURNER et al. 1992B ).

In populations in which males occur at high frequencies, progeny testing combined with DNA fingerprinting has revealed the individuals to be heterozygous at multiple mini- and microsatellite loci (LUBINSKI et al. 1995 ). The heterozygosity presumably stems from episodic outcrossing between males and hermaphrodites releasing unfertilized eggs. The circumstances under which the outcrossing occurs are not known, but require a change in behavior since hermaphrodites normally spawn alone and are aggressive to other individuals of the same species (TURNER et al. 1992A ).

The restriction of hermaphroditism to a single species, or at most to a single clade of closely related species, offers an opportunity to inquire into the circumstances under which this mode of reproduction arose. One of the few genetic systems suitable for such an inquiry is the major histocompatibility complex (Mhc), which encodes proteins capable of presenting pathogen-derived peptides to receptors on thymus-derived lymphocytes and so initiating an adaptive form of immune response (KLEIN and HOREJSI 1997 ). Some of the Mhc loci are highly polymorphic (KLEIN and FIGUEROA 1986 ) with a large number of highly divergent alleles (allelic lineages) maintained by balancing selection at sites specifying the peptide-binding residues (PBRs) of the Mhc proteins (HUGHES and NEI 1988 ). Heterozygosity at the Mhc loci is apparently advantageous because it broadens the array of peptides that an individual can bind by its Mhc molecules (ZINKERNAGEL and DOHERTY 1974 ).

In this study we used the polymorphism of the Mhc loci in R. marmoratus to make inferences about the population structure of these fish and about the origin of their hermaphroditism.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Fish and DNA preparation:
Forty-three fish collected at seven different localities were used (Fig 1, Table 1): 23 from Twin Caye, Belize (three different populations designated as Bel, PG, and BEL2K); 2 from Dangriga, Belize (Dan); 2 from Tobacco Range (91-125); 4 from Brevard County, Florida; 5 from Vero Beach, Florida (DS, CCHA); 1 from Marco Island, Florida; and 6 from Norman's Pond Cay, Bahamas (BH). Total genomic DNA was prepared from fresh or ethanol-preserved adult specimens as previously described (SATO et al. 1995 ).

View larger version (75K): In this window In a new window Download PPT slide   Figure 1. Map of Central America. Localities at which samples were collected are indicated.

Production of cDNA library:
Poly(A⁺) RNA was isolated from hepatopancreases of fish using an mRNA purification kit and cDNA was synthesized with the help of the TimeSaver cDNA synthesis kit (Amersham Biosciences, Freiburg, Germany). The cDNA was then inserted into EcoRI-digested gt10 vector (Stratagene, Heidelberg, Germany), in vitro packaged with the help of the Gigapack cloning kit (Stratagene), and used to transform competent Escherichia coli NM514 bacteria.

Polymerase chain reaction (PCR) amplification:
Standard PCR amplifications were performed in the PTC-200 Programmable Thermal Controller (MJR, Biozym, Hess. Oldendorf, Germany) or in the GeneAmp PCR System 9700 (AB Applied Biosystems, Weiterstadt, Germany). One hundred nanograms of genomic DNA was added to a reaction mixture consisting of 1x PCR buffer (50 mM KCl, 1.5 mM MgCl₂, 10 mM Tris-HCl, pH 8.3, 0.0001% gelatin), 0.2 mM of each of the four deoxynucleoside triphosphates (Amersham Biosciences), 1 mM of each of the sense and antisense primers, 2.5 units of Taq DNA polymerase (Amersham Biosciences), and 0.4 units Pfu DNA polymerase (Stratagene). The amplifications consisted of DNA denaturation for 1 min at 94°, followed by 35 cycles, each cycle consisting of 15 sec denaturation at 94°, 15 sec annealing at the required temperature, depending on the primer combination, and 2 min extension at 72°. The reaction was completed by a final primer extension for 7 min at 72°.

Amplification of the mitochondrial control region:
For the initial PCR the published primers L15995 (MEYER et al. 1994 ) and H00651 (KOCHER et al. 1989 ) were used under standard PCR conditions. On the basis of the sequence thus obtained, the killifish-specific primer pair KR1 (5'-CTCACCCCTAGCTCCCAAAGCTAG-3') and KR2 (5'-TTTAAGCTACACGAGCCCTAAGTTC-3') was designed to amplify a 900-bp fragment of the control region.

DNA sequencing and analysis:
Selected PCR products were isolated from low-melting-point agarose (Life Technologies, Eggenstein, Germany). Bands were stained with ethidium bromide, excised, and eluted with the aid of the QIAEX II gel extraction kit (QIAGEN, Hilden, Germany). The eluted DNA was blunt ended, phosphorylated, ligated to SmaI-digested pUC18 plasmid vector with the help of the SureClone ligation kit (Amersham Biosciences), and used to transform E. coli XL-1 Blue competent bacteria (Stratagene). Double-stranded DNA prepared with the aid of the QIAGEN plasmid kit was resuspended at a concentration of 1 µg/µl and sequenced by the dideoxy chain-termination method (SANGER et al. 1977 ), using the Thermo Sequenase fluorescent-labeled primer cycle sequencing kit (Amersham Biosciences). Sequencing reactions were processed by the LI-COR Long ReadIR 4200 DNA sequencer (MWG Biotech, Ebersberg, Germany). Scattered single-nucleotide changes differentiating a given clone from all other clones of a set obtained from the same DNA sample (individual) were taken for replication or sequencing errors. As such, they were ignored in the final analysis and the nucleotides shared by all the other clones in the set were assigned to the site in the particular gene. They constituted 30% of the sequences.

Southern DNA blotting and hybridization:
Five micrograms of genomic DNA was digested with restriction endonucleases for 18 hr under the conditions recommended by the supplier (Roche Diagnostics, Mannheim, Germany) and fragments were separated by agarose gel electrophoresis and blotted onto Hybond-N+ nylon filters (Amersham Biosciences). Prehybridization, hybridization, and probe labeling were carried out using the AlkPhos Direct kit (Amersham Biosciences). After the overnight hybridization, the filters were washed according to the AlkPhos Direct protocol. Following the application of the chemiluminescent detection reagent CDP-Star of the kit, Hyperfilm ECL (Amersham Biosciences) was exposed to the blot for 2 hr and developed.

Computer simulation:
The initial population consisted of N individuals, each individual carrying two identical alleles at a given locus and each gene containing L sites. The effects of mutation, drift, selection, and selfing were simulated in the process of sampling 2N genes for N individuals to create the next generation. To simulate mutations, a random variable C1 (from 0 to 1) was chosen for each gene and if its value was C1 u (where u was the chosen mutation rate), a mutation was introduced into that gene. Then another random variable C2 (from 0 to 1) was obtained and if its value was C2 r (where r was the chosen selfing rate), each second gene for an individual of the next generation was taken from the same individual of the preceding generation as the first gene; otherwise it was sampled from a different individual. If the newly generated individual was a homozygote, a third random variable C3 (from 0 to 1) was chosen. If C3 > s (where s was the chosen selection intensity), the individual was discarded (to simulate selection against homozygotes) and two new genes were sampled. Every 10/u generations, the values (frequency of homozygotes) and F (expected homozygosity) were recorded in 100 replicate experiments.

Phylogenetic analysis:
Sequences were aligned using the SeqPup 0.6f software for Macintosh (GILBERT 1996 ). Variability of the sequences was assessed with the help of the MEGA 2.0 program (KUMAR et al. 2001 ), using synonymous and nonsynonymous substitutions estimated by the NEI and GOJOBORI 1986 method for exons and Kimura's two-parameter distances (KIMURA 1980 ) for introns. Phylogenetic trees drawn by the neighbor-joining method (SAITOU and NEI 1987 ) using the PAUP*4.0b8a program (SWOFFORD 2001 ) were based on p-distances for amino acid sequences and on Kimura's two-parameter distances for nucleotide sequences. Trees were rooted at midpoint. Maximum-parsimony trees were drawn by the same program using the heuristic search algorithm. Gaps were treated as missing data. The topological stability of the trees was assessed by 500 bootstrap replications.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Cloning of killifish Mhc class I genes:
The first R. marmoratus (Rima) class I sequences were obtained by PCR amplification of genomic killifish DNA using the primer pair KFF1 and KFR1 (Table 2). The sequence of the oligonucleotides corresponded to the regions of exons 2 and 3 conserved between the guppy (SATO et al. 1995 ) and haplochromine cichlid fish (SATO et al. 1997 ). The PCR therefore amplified a region of the class I genes encompassing exon 2 from codon 36, the entire intron 2, and exon 3 up to codon 169 (exons 2 and 3 being the most variable parts of the class I genes). Subsequently, a cDNA library prepared from killifish RNA was used in anchor PCR to obtain the 5' and 3' parts of the cDNA clones. The combined sequence of the clones covered the entire coding region of the Rima class I genes (Fig 2). These initial sequences were then used to design additional primers specific for the different groups and different regions of the class I genes present in 25 individuals representing seven different killifish populations (individuals 1–25 in Table 1). Additional sequences were obtained by PCR amplification of fragments eluted from agarose gels in a Southern blotting experiment (see below), as well as of other samples. Altogether, 165 sequences encompassing different lengths of Mhc class I exon 2, intron 2, and exon 3 were obtained. Among these were 30 unique exon 2 sequences and 87 unique exon 3 sequences (Fig 2 Fig 3 Fig 4 Fig 5).

View larger version (68K): In this window In a new window Download PPT slide   Figure 2. R. marmoratus Mhc class I locus G1*01 cDNA sequence. The clones on which this sequence is based were obtained from a cDNA library. The translated amino acid sequence in the IUPAC-IUB three-letter code is given underneath the coding nucleotide sequence. The positions of exons as deduced in this study are indicated.

View larger version (18K): In this window In a new window Download PPT slide   Figure 3. Nucleotide sequence alignment of the R. marmoratus Mhc class I exon 2. The alignment encompasses 54 codons of the 3' part of exon 2. Dashes indicate identity with the consensus sequence at the top and dots indicate missing information. The numbering starts with the first nucleotide of the G1*01 cDNA sequence (Fig 2) in an alignment of all available genomic and cDNA sequences. Only unique sequences are shown.

View larger version (78K): In this window In a new window Download PPT slide   Figure 4. Nucleotide sequence alignment of R. marmoratus Mhc class I exon 3. Asterisks indicate alignment indels. For other explanations, see Fig 3.

View larger version (45K): In this window In a new window Download PPT slide   Figure 5. Alignment of variable sites of representative R. marmoratus Mhc class I intron 2 sequences. Only variable sites of the exon-flanking regions are shown; they are separated by a gap of unalignable sequence of 613–237 bp. For other explanations, see Fig 3 and Fig 4.

Southern blot analysis:
To facilitate the assignment of individual sequences to loci and alleles, four fish caught in Florida (two at a locality in Brevard County—nos. 32 and 33—and two in the Vero Beach area—nos. 35 and 36; see Fig 1) were chosen for exhaustive analysis. The genomic DNA isolated from these fish was digested with the HindIII endonuclease, the digest was separated by gel electrophoresis and blotted, and the filters were hybridized with a probe covering exon 3 of the killifish class I locus. Four different sizes of bands were revealed: 7.5 kb (band I), 5.5 kb (band II), 4.0 kb (band III), and 2.5 kb (band IV; Fig 6A). Bands I and IV were found to be present in all four individuals, band II in three individuals (nos. 33, 35, and 36), and band III in two individuals (nos. 32 and 36). The DNA restriction fragments from the areas of the gel corresponding to the position of the individual bands on the filter were then eluted and amplified by PCR using the exon 3-specific primer pair KE31 and KE32 (Table 2), the amplification products were cloned, and the clones sequenced. Altogether 21, 8, 7, and 20 clones were sequenced from individuals 32, 33, 35, and 36, respectively (Table 3). Disregarding differences apparently representing replication errors, 6 different sequences could be distinguished in the collection of 56.

View larger version (74K): In this window In a new window Download PPT slide   Figure 6. Southern blot hybridization of killifish genomic DNA with a 300-bp-long fragment of a class I Mhc gene (exon 3). (A) DNA from the four indicated individuals was digested with HindIII. (B) DNA from two individuals was digested with the indicated restriction endonucleases.

Identification and characterization of class I loci and alleles:
The starting point of the identification was a phylogenetic tree based on the collection of exon 3 sequences. Both the maximum-parsimony (Fig 7) and neighbor-joining (not shown) trees identified the same major clades; they differed, however, in the arrangement of the clades and the branching patterns within them. Distinct, well-separated, and statistically strongly supported clades were taken to represent separate groups of class I loci, provisionally designated by letters A–H. Some of the clades consist of a single locus, and others are composed of multiple loci designated by numbers. Alleles at a given locus are also designated by numbers, separated from the locus designation by an asterisk (KLEIN et al. 1990 ). The assignments of loci and alleles within clades were based primarily on the number of distinct sequences obtained per individual. Other criteria included the relationship among the sequences of exon 2 and intron 2, distribution of diagnostic substitutions (i.e., substitutions characterizing a given group or locus), and the presence of other markers such as specific indels (data not shown). Intron 2 sequences (Fig 5) were particularly helpful in the assignment of sequences to loci and alleles. The introns of the different clades differ in their length, their sequence, and the presence of indels at specific positions (data not shown). The longest intron 2 sequence (that of clade E) consists of >521 sites; the other introns are shortened by indels or truncations at their 3' parts. The 5' parts of intron 2 of the different clades can be aligned unambiguously. The middle and the 3' parts (where present) are largely clade specific and alignable only partially or not at all. The middle part contains a (CA)_n repeat, which varies in length even between individuals bearing otherwise identical or nearly identical sequences of a given gene (not shown). Clade E1 contains a (GATA)_n repeat in this part of the gene.

View larger version (17K): In this window In a new window Download PPT slide   Figure 7. Maximum-parsimony tree of R. marmoratus Mhc class I exon 3 nucleotide sequences based on the heuristic search algorithm. The majority consensus tree using the midpoint rooting option is shown. Major clades are indicated.

Haplotype polymorphism:
PCR-amplification experiments suggested that not all the loci and all groups of loci were represented in all the killifish populations. The exhaustive typing of the four fish from Florida (nos. 32, 33, 35, and 36) was particularly informative in this regard (Fig 6A, Table 3). It revealed the presence of the A1, B1, C1, D1, D2, and D3 loci in the Florida populations, but no evidence for the presence of any of the other class I loci found in other populations. Furthermore, differences were also apparent among populations from the same area and even among individuals from the same population (Table 1). Thus, for example, individuals 33 (Brevard County) and 35 (Vero Beach) showed no signs of the presence of any loci other than A1, D1, and D3. Similarly, no evidence of the B1 locus could be obtained for the Belize samples, which, by contrast, possessed the E, F, G, and H loci apparently absent in the Florida samples.

Additional evidence for the variation in the number of loci among individuals was obtained by Southern blot analysis of DNA from two fish, no. 32 from Brevard County, Florida and no. 39 from Tobacco Range, Belize (Fig 6B). The DNA was digested with five enzymes (EcoRI, HindIII, BamHI, MspI, and TaqI) and the blot was hybridized with an exon 3 probe. The differences in the number of hybridizing bands between the two individuals suggest that the two individuals differ in the organization of their class I regions. Assuming that all or nearly all class I loci are located in the same chromosomal region, as they are in other fish taxa (SATO et al. 2000 ) and indeed in all vertebrates tested (KLEIN 1986 ; TROWSDALE 1995 ), it follows that different class I haplotypes exist in the killifish populations. The haplotypes differ in the number and identity of loci they bear, some of them consisting of only one locus and others bearing multiple genes. This situation is analogous to that well documented for the HLA class II (DR) haplotypes (KLEIN 1986 ) and is presumably the consequence of unequal crossing over in a chromosomal region populated with closely related loci.

Heterozygosity of class I loci:
Heterozygotes at Mhc loci were found in two geographically separated regions in the area of R. marmoratus distribution, Belize and Florida, at frequencies of 4 and 44%, respectively (Table 1). The one heterozygous individual in the Belize population was found at Twin Caye, a locality at which the presence of males has been reported (DAVIS et al. 1990 ; TURNER et al. 1992A ). The individual is therefore most likely a product of a mating between a male and a hermaphrodite belonging to a different clone. By contrast, the presence of males in the Florida populations has not been observed and so the four heterozygotes found in these populations might be the result of a mutational divergence of clones that were originally homozygous at the Mhc loci. The smallness of the genetic distance between the alleles in five of the six heterozygous combinations (0.006, 0.006, 0.012, 0.012, 0.012 amounting to one or two substitutions) is in line with this proposition. Alternatively, males might also be present in the Floridian populations but at such low frequencies that their existence has thus far gone undetected, or they may arise episodically. The overall frequency of Mhc heterozygotes in the sample of 40 individuals tested is 12%.

Genetic distance analysis:
In pairwise comparisons, the genetic distances between Mhc alleles found in the sampled R. marmoratus specimens range from 0.003 to 0.076 at nonsynonymous sites, from 0.014 to 0.240 at synonymous sites of exons 2 and 3, and from 0.001 to 0.026 for intron 2 sites (data not shown). The lower parts of the ranges can be explained as being the result of divergences since the onset of selfing in R. marmoratus. The upper parts are characteristic of allelic lineages whose divergence predates the separation of the species involved. Hence the genetic distance analysis reveals the presence of polymorphisms presumably generated both before and after the onset of selfing (see DISCUSSION).

Dispersal time estimate based on mtDNA analysis:
We sequenced 900 bp of mtDNA control region from 10 R. marmoratus individuals (nos. 4, 9, 14, 17, 22, 24, 25, 31, 36, and 37) and found two lineages distinguished by substitutions at four sites (not shown, but sequences are deposited in the GenBank database). One lineage appeared to be restricted to Belize populations, while the other was found to be present in Belize, Bahamas, and Florida populations. Pairwise comparisons of sequences revealed the identities of three sequences from Belize individuals (nos. 14, 22, and 37) and two sequences from Florida individuals (nos. 9 and 22). The largest difference of eight nucleotide substitutions was observed between individuals 25 (Belize) and 36 (Florida), giving a nucleotide diversity of 0.009 ± 0.003. Using the haplochromine cichlid substitution rate of 5.6% substitutions per site per 10⁶ years (NAGL et al. 2000 ), we estimate that the lineages diverged 80,000 ± 30,000 years ago. The identity or near identity of sequences within each lineage suggests that R. marmoratus reached Central America and Florida relatively recently. WEIBEL et al. 1999 distinguished three mtDNA lineages in R. marmoratus by the analysis of restriction fragment length polymorphism: one restricted to South America, another to the Bahamas, and the third shared by the Florida and Belize populations. From the estimated divergence times of these alleles, the authors concluded that the dispersal to Belize, Bahamas, and Florida occurred 18,000 years ago, at the time of the last glacial, when the sea level in the Caribbean was 75 m lower than present (CLARK et al. 1978 ).

Computer simulation:
Earlier studies revealed a high frequency (close to 100%) of homozygotes at non-Mhc loci in the R. marmoratus populations (MASSARO et al. 1975; VRIJENHOEK 1985 ; TURNER et al. 1990 ; LAUGHLIN et al. 1995 ). By contrast, this study provides evidence that the frequency of homozygotes at Mhc loci is as low as 56% in some populations. Presumably the difference between the Mhc- and non-Mhc-based estimates is due to selection, which is absent at the neutral loci of the earlier studies but demonstrable in the case of the Mhc loci (HUGHES and NEI 1988 ). To determine under what conditions a selfing population could become nearly fully homozygous at neutral loci while retaining up to 44% heterozygotes at loci under balancing selection, we carried out a computer simulation in which we varied the selfing rate r, while keeping other population parameters (population size N, mutation rate u, and selection intensity s) constant. The N was estimated from the average nucleotide diversity () at the control region of the mtDNA. Using the relationship = 2N_fv (see NEI and KUMAR 2000 ) and values = 0.0047 ± 0.0023, v = 5.6 x 10^-8 substitutions per site per generation, we obtained N_f = 42,000 ± 21,000 or 5 x 10⁴. Since in a selfing population most individuals can transmit their mtDNA to the next generation, here N is approximately equal to N_f. The mutation rate µ at nonsynonymous sites in the segment of Mhc loci controlling the PBR, the region under balancing selection, has been estimated for cichlid fish to be 3 x 10^-9 per site per generation (or per year since in these fish, one generation equals 1 year; see FIGUEROA et al. 2000 ). Using this estimate and taking the length of the PBR as L = 139 sites (assuming correspondence between fish and human PBR positions), we obtain the estimate of Nu = 0.02, where u is the mutation rate of the PBR. The selection intensity parameter s was estimated from the observed average nucleotide diversity at Mhc introns using the relationship = 4Nµfs, where fs, the scaling factor of balancing selection, is a function of Ns and Nu (TAKAHATA 1990 , TAKAHATA 1993 ). The observed nucleotide diversity of Mhc loci ranged from 0.001 to 0.026. Using 4Nµ = 4 x 5 x 10⁴ x 3 x 10^-9 = 0.0006, fs ranges from 2 to 40 and the Ns becomes of the order of 100. The Ns = 100 value corresponds to N = 5 x 10⁴ and s = 0.002, but since simulation with N = 5 x 10⁴ would have taken a long time, the actual parameters used were N = 200 and s = 0.5.

The results of the simulation are summarized in Fig 8 Fig 9 Fig 10. They reveal that under conditions of neutrality and low Nu value (0.02), the frequency of homozygotes, , is close to 100%, irrespective of the variation in the selfing rate (Fig 8). It is only at Nu values >0.02 that the influence of selfing becomes apparent. When Nu = 0.1 and r = 0.90, the simulated value of becomes 96.6 ± 4.2%, which is close to the observed value of (TAYLOR et al. 2001 ). When balancing selection is introduced in the simulation process under these conditions (i.e., r = 0.90 and Nu = 0.02), the frequency of homozygotes is determined by the value of Ns: as Ns increases, decreases (Fig 9). At Ns = 100, = 58%, which is close to the value observed for the Mhc loci. We conclude therefore that the observed difference in the frequency of homozygotes between neutral loci and loci under balancing selection can be accounted for by assuming a selfing rate of 0.90, Nu of 0.02, and Ns of 100.

View larger version (28K): In this window In a new window Download PPT slide   Figure 8. Parameters affecting the frequency of homozygotes () at different selfing rates (r) as determined by computer simulation. For a description of the simulation, see text. N, effective population size; u, mutation rate; s, selection intensity.

View larger version (131K): In this window In a new window Download PPT slide   Figure 9. The effects of balancing selection on the frequency of homozygotes () at Nu = 0.1 and selfing rate of r = 0.95, as determined by computer simulation. For explanation, see text. N, effective population size; u, mutation rate; s, selection intensity. Standard errors are indicated by segments.

View larger version (31K): In this window In a new window Download PPT slide   Figure 10. Persistence of allelic lineages in a population of selfing individuals under different selfing rates (r) and chosen Ns and Nu parameters, as determined by computer simulation. N, effective population size; u, mutation rate; s, selection intensity. Persistence time, expressed as 2N generations, is measured from the onset of selfing until only one of the ancestral lineages is left. For details, see text.

The difference between the heterozygosity of neutral vs. Mhc loci was one observation in want of an explanation. Another was the large genetic distance (up to 0.026) between some of the Mhc alleles. Such alleles differ by multiple substitutions whose accumulation requires long persistence of an allelic lineage in a population. However, in a population consisting of selfing individuals the odds are stacked heavily against the persistence of allelic lineages and it is therefore necessary to define the conditions that could explain the presence of multiple allelic lineages in the killifish populations. Here, too, we resorted to computer simulation. To generate allelic lineages at an Mhc locus (defining a lineage as a collection of genes sharing substitutions at the sites under balancing selection), we started the simulation with a homozygous gene pool of size 2N and let the pool evolve under random mating with a mutation rate u and selection intensity s, until it reached an equilibrium in which the loss of allelic lineages was balanced out by the emergence of new lineages. (This took 100N to 4000N generations, depending on the mutation rate.) At that point we introduced selfing at a rate r, keeping track of individual lineages and recording the time of their disappearance from the pool. The time from the onset of selfing until the moment when only one of the original ("ancestral") lineages present at the outset remained for the first time was designated T_d. To obtain the average T_d, the simulation was repeated 1000 times for each of the different selfing rates.

As expected, the T_d depended on the selfing rate r: It decreased as the rate increased. At the rate r > 0.90, the T_d dropped rapidly to values obtained for neutral loci (Fig 10). At r = 0.90, the T_d depended on the parameters Nu and Ns: The lower the former and the larger the latter, the longer the persistence time T_d. To simulate conditions corresponding to the observations, we chose one value of the Nu parameter (Nu = 0.02) and one value of the Ns parameter (Ns = 100) to obtain T_d = 23 ± 14 in units of N generations. However, the rapid reduction of T_d to the neutral level (Fig 10) indicates that selfing influences selection intensity: Heterozygotes tend not to be produced in a selfing population and this weakens the intensity. With the selfing rate r, the intensity appears to be approximated by s(1 - r) (N. TAKAHATA, personal communication). Thus when r is as large as 90%, selection operates with the intensity of s/10 in a selfing population. Therefore, by taking Ns = 100 in a random mating population, the effective Ns value in a selfing population becomes 10.

The killifish Mhc data reveal that their Ns value is of the order of 100, suggesting that the Ns in a random mating population is much larger, perhaps of the order of 1000. We therefore ran two replications of simulations to examine the effect of a larger Ns on and T_d under the conditions of Ns = 1000, Nu = 0.02, and r = 0.90. The result shows that Ns = 1000 does not affect much: The simulated value of 56.4% ± 0.3 is similar to the observed -value (and to the simulated value when Ns = 100). However, Ns = 1000 affects T_d strongly. While under the conditions Ns = 100, Nu = 0.02, and r = 0.90, the estimated T_d value is 23 ± 14; when Ns = 1000, T_d becomes 51 ± 19 in units of N generations. By assuming a generation time of 1 year for killifish and N = 50,000, the persistence time of the allelic lineages is 2.5 ± 0.9 million years (MY). The implications of these results for the interpretation of the genetic composition of the killifish populations and for the origin of the species hermaphroditism are described in the DISCUSSION.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

In the population samples analyzed in this study, only four pairs of individuals and one quintet were found to be identical in their class I genes (Table 1). The differences between the individuals were in the presence or absence of loci (haplotype polymorphism) and in the alleles found at shared loci (allelic polymorphism). Genetic distances obtained by pairwise comparisons of Mhc alleles revealed that the Mhc composition of the killifish population resembled that of other fish species, for example, that of the Lake Victoria haplochromine flock (FIGUEROA et al. 2000 ). Hence, at this level of analysis, the Mhc class I heterogeneity of the killifish populations is comparable to that of the haplochromines, even though the former are considered to be a single species (albeit a species to which the definition based on breeding behavior does not apply), while the latter are a collection of several hundred closely related species. These observations are consistent with the interpretation of the killifish populations as agglomerates of highly differentiated clones. The clonal hypothesis was originally formulated on the basis of tissue grafting experiments (HARRINGTON and KALLMAN 1968 ) and was later strengthened by microsatellite typing data (TURNER et al. 1990 , TURNER et al. 1992B ). The Mhc analysis described in the present study provides an explanation for the grafting results since differences at class I Mhc loci are known to be the main source of histoincompatibility between individuals (KLEIN and HOREJSI 1997 ).

In one important aspect, however, the results of the Mhc analysis are at variance with the results of both the Mhc typing of other fish species and the microsatellite typing data—the heterozygosity of the loci. Mhc typings of other fish species, indeed of most other jawed vertebrates, generally indicate very high frequency of heterozygotes at some of the class I and class II loci, usually close to 100%. On the other hand, microsatellite typing of killifish populations reveals high heterozygosity in only one population, that of Twin Caye in Belize, in which males occur in abundance in addition to the hermaphrodites (LUBINSKI et al. 1995 ). The heterozygosity of this population is explained by the assumption of hermaphrodites outcrossing with males. No evidence of heterozygosity at microsatellite or other loci has been reported for other populations in which males are rare. The Mhc analysis of R. marmoratus, by contrast, reveals that only 4–44% of the individuals are heterozygous at these loci (Table 1). The computer simulation results explain this reduction in the level of Mhc heterozygosity by the high degree of selfing in the R. marmoratus populations. That Mhc heterozygotes have not been entirely eliminated is explained by the assumption of a selfing rate of <1.0. Apparently, not only in the Twin Caye population, in which male R. marmoratus have been found at a relatively high frequency (DAVIS et al. 1990 ; TURNER et al. 1992A ), but also in other populations, males occasionally occur and are responsible for the genetic mixing of the otherwise selfing clones.

The wide range of the genetic distances between the Mhc alleles found in pairwise comparisons indicates that the allelic differences stem from two sources: the ancestral allelic lineages that were present in the population at the time it switched from outbreeding to obligatory selfing and that have not been entirely eliminated as yet by the selfing process and mutations that continue to arise and are promoted by balancing selection. The current state of Mhc diversity in the R. marmoratus populations is therefore apparently the result of a complex interplay of several factors among which the rate of selfing, the intensity of balancing selection, and the persistence of ancestral allelic lineages are the most important ones.

The mangrove killifish R. marmoratus and its close relative R. occellatus are the only two hermaphrodites among the >100 defined species of Rivulus, and of the two, R. marmoratus is the only obligatorily selfing, synchronous hermaphrodite (HUBER 1992 ). The two species form a sister group to another species, R. caudomarginatus, which is nonhermaphroditic, and the whole clade is well separated from all the other nonhermaphroditic Rivulus species (MURPHY et al. 1999 ). Hermaphroditism therefore presumably arose in the common ancestor of R. marmoratus and R. occellatus. Taking the rate of 0.25% substitutions per site per 10⁶ years (calibrated on the African rivulid "Roloffia" geryi and on the split of the African and South American continents 100 MYA) and the mtDNA sequences of the cytochrome b, 12S ribosomal RNA, 16S ribosomal RNA, and cytochrome oxidase I genes (MURPHY et al. 1999 ), we estimate that R. marmoratus and R. occellatus diverged from each other 5.1 MYA and that their ancestor diverged from R. caudomarginatus 26 MYA. The hermaphroditism of the two sister species should therefore be >5 MY old. It apparently arose in South America, perhaps in southeastern Brazil (MURPHY et al. 1999 ), after the closure of the Panama Isthmus, which was completed 2.7 MYA (HAUG and TIEDEMANN 1998 ). Obligatory, simultaneous selfing in R. marmoratus, however, must be of a much younger age. This conclusion follows from the presence of ancestral allelic lineages at the Mhc loci in this species. The simulation results indicate that obligatory selfing must be <2.5 MY old; otherwise the ancestral lineages would have been eliminated. The mtDNA data lower this age estimate further. The estimate of the divergence of the mtDNA control region lineages and the low divergence within the lineages suggest that obligatory selfing in R. marmoratus must be <80,000 years old.

The Mhc analysis provides a glimpse into the possible mechanism by which hermaphroditism may have arisen in R. marmoratus. The genetic distances between the killifish class I genes, including those between alleles found in different putative clones, are too great to have been attained in the last 2.5 MY (KLEIN et al. 1993 ). The divergence of the genes must have begun in the nonhermaphroditic ancestors and some of the variability must have then been bequeathed to R. marmoratus. This observation implies that hermaphroditism could not have arisen in a single individual, but rather that the switch from the separate-sexed condition to hermaphroditism and to selfing involved a whole population of individuals and most likely occurred gradually. At least two stages must be postulated in the evolution of hermaphroditism. In the first stage, a mutation (or mutations) produced protogynous hermaphrodites that functioned first as females and then transformed into males. The mutation then spread through the population among individuals bearing highly diverged Mhc class I genes. At this stage, hermaphroditism may have been facultative so that the fish were reproducing by selfing and outcrossing. In the second stage, another mutation (or mutations) transformed successive hermaphrodites into synchronous ones, in which mature male and female gametes were produced at the same time. If the mutation required homozygosity for its full phenotypic manifestation, it could have again spread through the population and thus assured the retention of much of the existing Mhc variability in the population. If selfing were, for whatever reason (WELLS 1979 ), advantageous, homozygotes for the mutation would have been selected, and obligatory hermaphroditism would have been fixed independently in the different emerging clones into which the population was being fragmented. This scenario would account for the large genetic distances between Mhc genes in a species that has presumably been reproducing by selfing for some 2.5 MY. It would also, at the same time, account for the small genetic distances between alleles of class I heterozygotes. The former would have been founded long before the species shifted to obligatory, synchronous hermaphroditism; the latter would have arisen only in the last 2.5 MY.

	FOOTNOTES

Sequence data from this article have been deposited with the EMBL/GenBank Data Libraries under accession nos. AF550048, AF550049, AF550050, AF550051, AF550052, AF550053, AF550054, AF550055, AF550056, AF550057, AF550058, AF550059, AF550060, AF550061, AF550062, AF550063, AF550064, AF550065, AF550066, AF550067, AF550068, AF550069, AF550070, AF550071, AF550072, AF550073, AF550074, AF550075, AF550076, AF550077, AF550078, AF550079, AF550080, AF550081, AF550082, AF550083, AF550084, AF550085, AF550086, AF550087, AF550088, AF550089, AF550090, AF550091, AF550092.
² Present address: Osaka University Graduate School of Dentistry, Department of Oral Pathology, Suita, Osaka 565-0871, Japan.

	ACKNOWLEDGMENTS

We thank Dr. Bruce J. Turner (Department of Biology, Virginia Polytechnic Institute and State University, Blacksburg, Virginia) for introducing J.K. and A.S. to this interesting model system and for generously providing killifish specimens for this study. We also thank Prof. Naoyuki Takahata (Department of Biosystems Science, The Graduate University for Advanced Studies, Hayama, Kanagawa, Japan) for many useful suggestions regarding the interpretation of the data; as well as Sabine Rosner for outstanding technical assistance and Jane Kraushaar for no less indispensable editorial assistance.

Manuscript received July 28, 2002; Accepted for publication September 30, 2002.

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

ATZ, J. W., 1965  Hermaphroditic fish. Science 150:789-797.[Free Full Text]

CLARK, J. A., W. E. FARELL, and W. R. PELTIER, 1978  Global changes in postglacial sea level: a numerical calculation. Quat. Res. 9:265-287.

COLE, K. S., and D. L. G. NOAKES, 1997 Gonadal development and sexual allocation in mangrove killifish, Rivulus marmoratus (Pisces: Atherinomorpha). Copeia 1997: 596–600.

DAVIS, W. P., D. S. TAYLOR, and B. J. TURNER, 1990  Field observations of the ecology and habits of mangrove rivulus (Rivulus marmoratus) in Belize and Florida. Ichthyol. Explor. Freshwaters 1:123-134.

FIGUEROA, F., W. E. MAYER, H. SÜLTMANN, C. O'HUIGIN, and H. TICHY et al., 2000  Mhc class II B gene evolution in East African cichlid fishes. Immunogenetics 51:556-575.[Medline]

GILBERT, D. G., 1996 SeqPup Version 0.6f: A Biosequence Editor and Analysis Application (http://iubio.bio.indiana.edu/soft/molbio).

HARRINGTON, R. W., 1961  Oviparous hermaphroditic fish with internal fertilization. Science 134:1749-1750.[Abstract/Free Full Text]

HARRINGTON, R. W., 1963  Twenty-four hour rhythms of internal self-fertilization and oviposition by hermaphrodites of Rivulus marmoratus.. Physiol. Zool. 36:325-341.

HARRINGTON, R. W., 1967  Environmentally controlled induction of primary male gonochorists from eggs of the self-fertilizing hermaphroditic fish Rivulus marmoratus.. Biol. Bull. 132:174-199.[Abstract/Free Full Text]

HARRINGTON, R. W., 1968  Delimitation of the thermolabile phenocritical period of sex determination and differentiation in the ontogeny of the normally hermaphroditic fish Rivulus marmoratus.. Physiol. Zool. 41:447-460.

HARRINGTON, R. W., 1975 Sex determination and differentiation among uniparental homozygotes of the hermaphroditic fish Rivulus marmoratus (Cyprinodontidae Atheriniformes), pp. 249–262 in Intersexuality in the Animal Kingdom, edited by R. REINBOTH. Springer, New York.

HARRINGTON, R. W. and K. D. KALLMAN, 1968  The homozygosity of clones of the self-fertilizing hermaphroditic fish Rivulus marmoratus (Cyprinodontidae Atheriniformes). Am. Nat. 102:337-343.

HAUG, G. H. and R. TIEDEMANN, 1998  Effect of the formation of the Isthmus of Panama on Atlantic Ocean thermohaline circulation. Nature 393:673-676.

HUBER, J. H., 1992 Review of Rivulus—Ecobiogeography, Relationships. Société Française d'Ichtyologie, Paris.

HUGHES, A. L. and M. NEI, 1988  Pattern of nucleotide substitution at major histocompatibility complex class I loci reveals overdominant selection. Nature 355:167-170.

KALLMAN, K. D. and R. W. HARRINGTON, 1964  Evidence for the existence of homozygous clones in the self-fertilizing hermaphroditic teleost Rivulus marmoratus Poey. Biol. Bull. 126:101-114.

KIMURA, M., 1980  A simple method for estimating evolutionary rates of base substitutions through comparative studies of nucleotide sequences. J. Mol. Evol. 16:111-120.[Medline]

KLEIN, J., 1986 Natural History of the Major Histocompatibility Complex. John Wiley, New York.

KLEIN, J. and F. FIGUEROA, 1986  Evolution of the major histocompatibility complex. CRC Crit. Rev. Immunol. 6:295-386.

KLEIN, J., and V. HOREJSÍ, 1997 Immunology. Blackwell Scientific Publishers, Oxford.

KLEIN, J., R. E. BONTROP, R. L. DAWKINS, H. A. ERLICH, and U. B. GYLLENSTEN et al., 1990  Nomenclature for the major histocompatibility complexes of different species: a proposal. Immunogenetics 31:217-219.[Medline]

KLEIN, J., Y. SATTA, C. O'HUIGIN, and N. TAKAHATA, 1993  The molecular descent of the major histocompatibility complex. Annu. Rev. Immunol. 11:269-295.[Medline]

KOCHER, T. D., W. K. THOMAS, A. MEYER, S. V. EDWARDS, and S. PAABO et al., 1989  Dynamics of mitochondrial DNA evolution in animals: amplification and sequencing with conserved primers. Proc. Natl. Acad. Sci. USA 86:6196-6200.[Abstract/Free Full Text]

KRISTENSEN, I., 1970  Competition in three cyprinodont fish species in the Netherlands Antilles. Stud. Fauna Curacao Carib. Isl. 119:82-101.

KUMAR, S., K. TAMURA, I. B. JAKOBSEN and M. NEI, 2001 MEGA 2: Molecular Evolutionary Genetic Analysis Software. Arizona State University, Tempe, AZ.

LAUGHLIN, T. F., B. A. LUBINSKI, E.-H. PARK, D. S. TAYLOR, and B. J. TURNER, 1995  Clonal stability and mutation in the self-fertilizing hermaphroditic fish, Rivulus marmoratus.. J. Hered. 86:399-402.[Free Full Text]

LIN, H.-C. and W. A. DUNSON, 1995  An explanation of the high strain diversity of a self-fertilizing hermaphroditic fish. Ecology 76:593-605.

LUBINSKI, B. A., W. P. DAVIS, D. S. TAYLOR, and B. J. TURNER, 1995  Outcrossing in a natural population of a self-fertilizing hermaphroditic fish. J. Hered. 86:469-473.[Free Full Text]

MASSARO, E. J., J. C. MASSARO and R. W. HARRINGTON, 1975 Biochemical comparison of genetically different homozygous clones (isogenic, uniparental lines) of the self-fertilizing fish Rivulus marmoratus Poey, pp. 439–453 in Isozymes, Vol. 3, edited by C. L. MARKERT. Academic Press, New York.

MEYER, A., M. MORRISSEY, and M. SCHARTL, 1994  Recurrent origin of a sexually selected trait in Xiphophorus fishes inferred from a molecular phylogeny. Nature 368:539-542.[Medline]

MURPHY, W. J., J. E. THOMERSON, and G. E. COLLIER, 1999  Phylogeny of the neotropical killifish family Rivulidae (Cyprinodontiformes, Aplocheiloidei) inferred from mitochondrial DNA sequences. Mol. Phylogenet. Evol. 13:289-301.[Medline]

NAGL, S., H. TICHY, W. E. MAYER, N. TAKEZAKI, and N. TAKAHATA et al., 2000  The origin and age of the haplochromine species flock in Lake Victoria, East Africa. Proc. R. Soc. Lond. Ser. B 267:1049-1061.[Medline]

NEI, M. and T. GOJOBORI, 1986  Simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions. Mol. Biol. Evol. 3:418-426.[Abstract]

NEI, M., and S. KUMAR, 2000 Molecular Evolution and Phylogenetics. Oxford University Press, Oxford.

SAITOU, N. and M. NEI, 1987  The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol. Biol. Evol. 4:406-425.[Abstract]

SANGER, F., S. NICKLEN, and A. R. COULSON, 1977  DNA sequencing with chain-terminating inhibitors. Proc. Natl. Acad. Sci. USA 74:5463-5467.[Abstract/Free Full Text]

SATO, A., F. FIGUEROA, C. O'HUIGIN, D. N. REZNICK, and J. KLEIN, 1995  Identification of major histocompatibility complex genes in the guppy, Poecilia reticulata. Immunogenetics 43:38-49.

SATO, A., D. KLEIN, H. SÜLTMANN, F. FIGUEROA, and C. O'HUIGIN et al., 1997  Class I Mhc genes of cichlid fishes: identification, expression, and polymorphism. Immunogenetics 46:63-72.[Medline]

SATO, A., F. FIGUEROA, B. W. MURRAY, E. MÁLAGA-TRILLO, and Z. ZALESKA-RUTCZYNSKA et al., 2000  Nonlinkage of the major histocompatibility complex of class I and class II loci in bony fishes. Immunogenetics 51:108-116.[Medline]

SOTO, C. G., J. F. LEATHERLAND, and D. L. G. NOAKES, 1992  Gonadal histology in the self-fertilizing hermaphroditic fish Rivulus marmoratus.. Can. J. Zool. 70:2338-2347.

SWOFFORD, D. L., 2001 PAUP: Phylogenetic Analysis Using Parsimony (and Other Methods), Version 4.0b8a. Sinauer, Sunderland, MA.

TAKAHATA, N., 1990  A simple genealogical structure of strongly balanced allelic lines and trans-species evolution of polymorphism. Proc. Natl. Acad. Sci. USA 87:2419-2423.[Abstract/Free Full Text]

TAKAHATA, N., 1993  Allelic genealogy and human evolution. Mol. Biol. Evol. 10:2-22.[Abstract]

TAYLOR, D. S., M. T. FISHER, and B. J. TURNER, 2001  Homozygosity and heterozygosity in three populations of Rivulus marmoratus.. Environ. Biol. Fishes 61:455-459.

TROWSDALE, J., 1995  "Both man & bird & beast": comparative organization of MHC genes. Immunogenetics 41:1-17.[Medline]

TURNER, B. J., J. F. ELDER, T. F. LAUGHLIN, and W. P. DAVIS, 1990  Genetic variation in clonal vertebrates detected by simple-sequence DNA fingerprinting. Proc. Natl. Acad. Sci. USA 87:5653-5657.[Abstract/Free Full Text]

TURNER, B. J., W. P. DAVIS, and D. S. TAYLOR, 1992a  Abundant males in populations of a selfing hermaphroditic fish, Rivulus marmoratus, from some Belize cays. J. Fish. Biol. 40:307-310.

TURNER, B. J., J. F. ELDER, T. F. LAUGHLIN, W. P. DAVIS, and D. S. TAYLOR, 1992b  Extreme clonal diversity and divergence in populations of a selfing hermaphroditic fish. Proc. Natl. Acad. Sci. USA 89:10643-10647.[Abstract/Free Full Text]

VRIJENHOEK, R. C., 1985  Homozygosity and interstrain variation in the self-fertilizing hermaphroditic fish, Rivulus marmoratus.. J. Hered. 76:82-84.[Abstract/Free Full Text]

VRIJENHOEK, R. C., R. M. DAWLEY, C. J. COLE and J. P. BOGART, 1989 A list of the known unisexual vertebrates, pp. 19–23 in Evolution and Ecology of Unisexual Vertebrates. Museum Bulletin No. 466, edited by R. M. DAWLEY and J. P. BOGART. The State University of New York, Albany, NY.

WEIBEL, A. C., T. E. DOWLING, and B. J. TURNER, 1999  Evidence that an outcrossing population is a derived lineage in a hermaphroditic fish (Rivulus marmoratus). Evolution 53:1217-1225.

WELLS, H., 1979  Self-fertilization: Advantageous or deleterious? Evolution 33:252-255.

ZINKERNAGEL, R. M. and P. C. DOHERTY, 1974  Restriction of in-vitro T cell-mediated cytotoxicity in lymphocytic choriomeningitis within a syngeneic or semiallogeneic system. Nature 248:701-702.[Medline]