Evolution, 56(3), 2002, pp. 597±616

DISCOVERY AND PHYLOGENETIC ANALYSIS OF A RIVERINE SPECIES FLOCK OF AFRICAN ELECTRIC FISHES (MORMYRIDAE: TELEOSTEI)

JOHN P. SULLIVAN,1,2 SEÂ BASTIEN LAVOUEÂ ,3,4 AND CARL D. HOPKINS1,5 1Department of Neurobiology and Behavior, Cornell University, Ithaca, New York 14853 2E-mail: [email protected] 3Museum National d'Histoire Naturelle, Ichtyologie GeÂneÂrale et AppliqueÂe; 43 rue Cuvier 75005, Paris, France 4E-mail: [email protected] 5E-mail: [email protected]

Abstract. The evolution of species-speci®c mate recognition signals is of particular interest within speciose mono- phyletic groups with restricted distributions (known as ``species ¯ocks''). However, the explosive nature of speciation in these clades makes dif®cult the reconstruction of their phylogenetic history. Here we describe a species ¯ock of riverine mormyrid ®shes from west-central Africa in which electric signals may play a role in the reproductive isolation of sympatric species. In our recent ®eld collections, totaling more than 1400 specimens from many localities, we recognize 38 forms that are distinct in their morphologies and electric organ discharge (EOD) characteristics. Of these 38, only four clearly correspond to described species. Here we treat these forms as operational taxonomic units (OTUs) in a phylogenetic analysis of cytochrome b sequence data from a sample of 86 specimens. We examined support in the molecular data for the monophyly of these 38 OTUs considered together, the monophyly of each phenotypically delimited OTU considered individually, and for relationships among OTUs congruent with those inferred from the distribution of morphological and EOD character states. Trees obtained by both maximum-parsimony and maximum- likelihood analyses, rooted with sequence data from outgroup taxa, provide evidence for the monophyly of these 38 OTUs with respect to other mormyrid ®shes. The small genetic distances between many distinct forms suggest their recent divergence. However, in many instances the cytochrome b tree topology fails to support the monophyly of individual OTUs and close relationships between OTUs that are similar in morphology and EOD characteristics. In other cases, individuals from distinct OTUs share identical or nearly identical haplotypes. Close examination of these cases suggests that unnatural OTU de®nition is not the sole cause of this pattern, and we infer an incongruence between the mitochondrial gene tree and the organismal phylogeny caused by incomplete mitochondrial lineage sorting and/ or introgression across forms. The apparently rapid diversi®cation in this clade of riverine electric ®shes and the problems associated with recovering a meaningful species-level phylogeny from mitochondrial data parallel ®ndings in other species ¯ocks. Selection on EOD waveforms as mate recognition signals may be involved in the radiation of these ®shes. This is the ®rst description of a freshwater ®sh species ¯ock from a riverine, as opposed to a lacustrine, environment.

Key words. Cytochrome b, electric ®sh, phylogeny, speciation, species ¯ock.

Received November 27, 2000. Accepted December 7, 2001.

Species-speci®c signals often mediate prezygotic repro- unrecognized species ¯ock of freshwater electric ®shes from ductive isolation between sympatric, closely related species. Africa in which simple electric signals may play an important The evolution of such signals and their role in the origin and role in the reproductive isolation of sympatric species. maintenance of species boundaries is frequently investigated The African weakly electric ®sh superfamily Mormyroidea in cases of ``species ¯ocks'' in which several to many species (families Mormyridae ϩ Gymnarchidae) is a remarkable have diversi®ed from a single common ancestor in a geo- modern radiation from within the superorder Osteoglosso- graphically restricted area, often over an evolutionarily short morpha, one of the oldest and phylogenetically most basal period of time. Examples include the haplochromine cichlid groups of extant teleosts (Greenwood 1973; Patterson and ®sh species ¯ocks of the African Great Lakes in which visual Rosen 1977; Lauder and Liem 1983; Arratia 1997). Mor- color patterns and displays maintain reproductive isolation myroids have diversi®ed in African freshwater riverine en- among species (Lowe 1953; Fryer and Iles 1972; Seehausen vironments, whereas other osteoglossomorph lineages have et al. 1997; van Oppen et al. 1998; Knight and Turner 1999) undergone a general reduction in their global diversity and and the Galapagos ®nch, Hawaiian picture-wing Drosophila, distribution since the Early Tertiary (Li 1997). Although re- and Hawaiian cricket species ¯ocks in which sounds largely stricted to the African continent, more than 200 living species serve this function (Hoy et al. 1988; Otte 1994; Grant and of mormyroids are recognized (Daget et al. 1984); the re- Grant 1996; Shaw 1999). Species ¯ocks are attractive to stu- maining living osteoglossomorphs, distributed on all conti- dents of signal evolution and speciation because species nents except Europe and Antarctica, comprise just 18 species boundaries are of recent origin (or in some cases still incom- (Nelson 1994). Uniquely among osteoglossomorphs and like- plete) and because rapid diversi®cation has produced a large ly related to their evolutionary success, all mormyroids elec- number of parallel cases from which to observe general pat- trolocate and communicate by means of specialized electric terns. Knowledge of species relationships is critical for signal organs and receptors, enabling them to effectively exploit the evolution studies, and yet the very features that make species nocturnal environment. ¯ocks compelling can complicate the analysis of their internal In this paper we present evidence based on recent collec- phylogenetic structure. In this paper we describe the discov- tions from Gabon in west-central Africa that this mormyroid ery and a preliminary phylogenetic analysis of a previously radiation is continuing explosively within a monophyletic 597 ᭧ 2002 The Society for the Study of Evolution. All rights reserved. 598 JOHN P. SULLIVAN ET AL. group we refer to as the ``Gabon-clade Brienomyrus.'' This genes) with sequence data from two introns in the nuclear group includes Paramormyrops gabonensis Taverne 1971, S7 gene for 38 species belonging to 17 nominal genera within several described species currently placed in the genus Brien- the subfamily Mormyrinae. The single most parsimonious omyrus, and many taxonomically unrecognized forms pos- tree that results (Fig. 1) indicates that Brienomyrus (Brien- sessing distinct electric organ discharge (EOD) waveforms omyrus) brachyistius, the type species of the genus, is not and morphologies. We are interested in the evolutionary his- closely related to any other included Brienomyrus species, tory of species-speci®c EOD and electric organ character- but is instead the sister group to Isichthys henryi and that this istics in these ®shes and the role of electric communication, pair together is the sister group to species of Mormyrus. Quite if any, in the origin and maintenance of species diversity. separate from this clade, Brienomyrus (Brevimyrus) niger is Comparative and experimental investigations into these is- weakly supported as the sister species to Hyperopisus bebe sues require better knowledge of the number of Gabon-clade at the base of a large clade containing species of Marcusenius, Brienomyrus species, their distributions, and phylogenetic re- Hippopotamyrus, Gnathonemus, and Campylomormyrus. lationships. Here we undertake a preliminary study of mor- A third clade containing nominal Brienomyrus species (B. phological and EOD variation to estimate species-level di- hopkinsi, B. longicaudatus) in addition to P. gabonensis and versity and a complimentary study of genetic variation among an undescribed species (VAD in this study) appears as the the identi®ed groups. We assess congruence between groups sister group to Marcusenius ntemensis. These species (in- that we recognize by phenotype and groups diagnosed by cluding M. ntemensis) are endemic to a particular region of DNA sequence characters. Speci®cally, we wish to know if lower Guinea: the OgooueÂ, Ntem, and Woleu/Mbini River sequence data will (1) demonstrate that all phenotypes that basins of Gabon, southern Cameroon, and Equatorial Guinea. we presume to belong to the Gabon-clade identi®ed in our The sister group of these speciesÐBoulengeromyrus knoepf- earlier studies represent a monophyletic group; (2) support ¯eri, Ivindomyrus opdenboschi, and Pollimyrus marcheiÐare the monophyly and thereby the potential species status of all endemic to the same region (Daget et al. 1984; Kamdem each phenotype considered individually (when sequences of Toham 1998). The monophyly of this third clade of Brien- multiple individuals are available); and (3) support relation- omyrus species plus P. gabonensis is supported by high boot- ships among identi®ed groups congruent with the distribution strap and decay index values in this analysis and by a unique of several presumed phylogenetically informative phenotypic 22-bp inversion in the ®rst intron of the S7 gene (S. LavoueÂ, characters. If agreement is good between the phenotypic and J. P. Sullivan, and C. D. Hopkins, unpubl. ms.). genetic data, we will use the molecular phylogeny to look In this study, we are concerned with members of this third, for patterns of EOD and electric organ character evolution. taxonomically unrecognized clade. Despite their demonstrat- ed remote relationship to the type species of Brienomyrus, Taxonomy and Previous Phylogenetic Studies of the Genera the inclusion of P. gabonensis, and their wide distribution in Brienomyrus and Paramormyrops the lower Guinea and Congo ichthyofaunal regions, we in- formally refer to them as the ``Gabon-clade Brienomyrus'' Taverne (1971) established the genus Brienomyrus as part of because they dominate the mormyrid fauna in our collections his osteology-based taxonomic revision of the Mormyridae, but from this country. We have begun morphological studies to without explicit reference to any uniquely shared derived char- search for phenotypic synapomorphies of this group, and spe- acters (synapomorphies). Within the genus, he recognized two cies descriptions and a revision of their taxonomy are un- subgenera: B.(Brienomyrus) and B.(Brevimyrus). Currently, derway (C. D. Hopkins, G. G. Teugels, R. J. Rundell, and J. there are nine valid species within the ®rst subgenus (B. bra- P. Sullivan, unpubl. ms.). chyistius, B. longianalis, B. sphekodes, B. kingsleyae, B. curvif- rons, B. longicaudatus, B. batesii, B. tavernei, and B. hopkinsi) The Gabon-Clade Brienomyrus and a single species (B. niger) in the latter (Alves-Gomes and Hopkins 1997; Teugels and Hopkins 1998). Gabon-clade Brienomyrus species range in adult size from Taverne et al. (1977a) later established the genus Para- about 100 mm to 250 mm standard length. They have relatively mormyrops, describing the type species, P. gabonensis, from elongate, moderately compressed bodies; rounded nontubular the Ivindo River of Gabon and referring to this genus Mar- snouts; and small terminal to somewhat subterminal mouths cusenius jacksoni Poll, a species from the upper Zambesi bearing ®ve to nine (upper jaw) and six to nine (lower jaw) River basin of Angola. Taverne did not include P. gabonensis bicuspid, pincerlike teeth. Dorsal and anal ®ns are set far back within Brienomyrus apparently because of the presence of an on the body and are roughly symmetrical above and below ossi®ed lateral ethmoid in this species, a bone absent in the the midline. They have ¯eshy, somewhat bulbous chins; small Brienomyrus specimens he examined, but which we ®nd is but functional eyes; and are light gray or light brown to near intermittently present as a small ossi®ed or cartilaginous el- black, with little patterning of pigmentation on the body or ement in some Brienomyrus specimens (pers. obs.). ®ns. Like other mormyrids, they are nocturnally active and are Recent studies have demonstrated the polyphyly of the probably benthic foragers of insect larvae (Winemiller and genus Brienomyrus with several molecular datasets (Alves- Adite 1997). General uniformity of jaw structure and dentition Gomes and Hopkins 1997; Lavoue et al. 2000; Sullivan et suggests little trophic divergence among them, although stom- al. 2000). S. LavoueÂ, J. P. Sullivan, and C. D. Hopkins (un- ach contents have not been compared. In the forested regions publ. ms.) performed an unweighted parsimony analysis on of Gabon, they are found in a wide variety of ¯owing water the combined data used in these previous studies (from the habitats, from the shallowest stream headwaters to rocky sub- mitochondrial 12S, 16S, cytochrome b, and nuclear RAG2 strates in the deepest portion of large rivers. Species com- SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 599

FIG. 1. Phylogenetic relationships among 36 species belonging to 17 nominal genera of the Mormyrinae as re¯ected by the most parsimonious tree recovered from an unweighted analysis of the combined molecular sequence datasets from 12S, 16S, cytochrome b, RAG2, and S7 introns 1 and 2 (4256 aligned bp, 939 informative characters, CI ϭ 0.51, RI ϭ 0.62). The genus Brienomyrus is shown to be polyphyletic. The clade of species from Gabon is distinct from Brienomyrus (Brienomyrus) brachyistius, the type species of Brienomyrus Taverne, and from Brienomyrus (Brevimyrus) niger. The numbers above nodes are bootstrap values (shown if above 50%); those below nodes are Bremer decay indices (adapted from S. LavoueÂ, J. P. Sullivan, and C. D. Hopkins, unpubl. ms.). The monophyly of the Mormyrinae and the sister group relationship between Myomyrus macrops and remaining Mormyrinae was established by Lavoue et al. (2000) and by Sullivan et al. (2000). 600 JOHN P. SULLIVAN ET AL. position differs among these habitat types and from region to right). The complex anatomy of these electrocytes and the region. Gabon-clade Brienomyrus are frequently the numeri- functional relationship between the stalks and the electrocyte cally most abundant ®sh in stream inventories. In some lo- faces in generating EOD waveforms have been reviewed by calities, ®ve or six distinct forms, presumably different species, Bennett (1971), Bass (1986a), Alves-Gomes and Hopkins are found together with species of other mormyrid genera (1997), and Sullivan et al. (2000). (pers. obs.). Little is known regarding the life history, ecology In a previous study (Sullivan et al. 2000), we generated a and mating system of these ®shes. phylogenetic hypothesis from mitochondrial and nuclear se- quence data for all major groups of mormyrids. Unweighted The Mormyrid Electric Organ and Electric Organ parsimony optimization of this electrocyte character on the Discharge tree, in conjunction with observations of electrocyte ontog- eny, led us to hypothesize that nonpenetrating electrocyte Mormyrids generate electric discharges using four parallel stalks are the primitive condition for the family Mormyridae, columns of electrocytes in the caudal peduncle (Schlichter and that penetrating stalks may have evolved only once in 1906; Lissmann 1951; Bennett and Grundfest 1961; Bennett the common ancestor of the large subfamily Mormyrinae, 1970, 1971). Each electrocyte of the electric organ generates followed by multiple independent reversals to the nonpene- a pulsating waveform identical to the overall EOD waveform trating condition. Whereas both forms of electrocyte are recorded externally to the ®sh. Among species, EOD pulses found within the Gabon-clade Brienomyrus, this analysis in- vary in duration from a fraction of a millisecond to 10 msec dicated that penetrating stalk electrocytes are the primitive and have an amplitude of 100 mV to several hundred mil- condition for the Gabon-clade Brienomyrus as well as for the livolts. These EODs are repeated at irregular intervals of 10 outgroup taxa used in this study. to 100 times per second. The waveform of the discharge is controlled by the structure and physiology of the electrocytes MATERIALS AND METHODS making up the electric organ (Bennett 1971; Bass 1986b), Specimens Used in This Study whereas the rhythm of the discharges is regulated by activity of a pacemaker region in the medulla. We collected and recorded the electric signals of more than Several studies of mormyrid species assemblages have 1400 specimens that we identi®ed as Gabon-clade Brienomyrus during ®eld trips to Gabon and the Central African Republic documented intraspeci®c stereotypy and interspeci®c differ- between 1998 and 2000. Field sites are listed in Table 1 and ences in EOD waveforms (Hopkins 1980, 1981, 1986a; Bass mapped in Figure 3. Methods of capture included funnel traps 1986a,b; Bass et al. 1986). Among species, EODs differ in baited with earthworms lowered to river and stream bottoms, duration, number, and amplitude of peaks, polarity, and wave hook and line, cast netting, rotenone exposure followed by re- shape. In many species, male and female EODs differ during suscitation, and dip netting following localization of ®shes' the breeding season (Hopkins 1980; Hopkins and Bass 1981; EODs with handheld electrode-ampli®er units. Bass and Hopkins 1983; Kramer 1997). Adult male EODs To record a ®sh's EOD, we placed each specimen indi- are typically longer in duration than those of females. These vidually into a plastic container with water from the ®eld EOD differences have behavioral relevance. In a study on site. An electrode was positioned at least 10 cm from either one Brienomyrus species from Gabon, it was demonstrated end of the ®sh and oriented parallel to its body axis. EODs that males discriminate between the EODs of conspeci®c and were ampli®ed (CWE BMA-831/XR Bioampli®er, CWE Inc., heterospeci®c females and between those of conspeci®c Ardmore, PA; DC settings, 0±50 kHz bandwidth), captured males and females based on temporal cues in the EOD, that by a DaqBook (IoTech, Inc., Cleveland, OH) analog to digital is, the duration and shape of the EOD waveform (Hopkins converter (16 bits at 100 kHz) and recorded onto the hard and Bass 1981; Hopkins 1986b). disk of a laptop computer. The relationship of certain species-speci®c EOD waveform We euthanized the ®shes by overdose of the anesthetic characteristics to electrocyte anatomy is well understood. MS222. Tissue samples were subsequently removed from the Electrocytes in mormyrids are ¯attened disk-shaped cells that dorsal musculature with a clean scalpel blade and preserved are electrically active on both anterior and posterior faces. in 90% ethanol or a saturated NaCl solution containing A complex network of electrically active tube-shaped stalks, DMSO and EDTA (Seutin et al. 1991). All specimens were which is an outgrowth of each electrocyte, receives neural ®xed in phosphate-buffered 10% formalin, later transferred innervation and transmits electrical excitation from the syn- to 75% ethanol, and catalogued at the Cornell University apse to the disk faces. Innervation is on the thickest portion Museum of Vertebrates (CU) or the American Museum of of the stalks near the center of the electrocyte disk. The stalk Natural History (AMNH). We attached permanent tags to all system then branches repeatedly to form ever ®ner stalklets specimens bearing a unique number by which specimen, tis- that eventually merge with the posterior face of the electro- sue sample, and EOD recording are linked in our records. cyte. In some species the stalks penetrate through the disk All animal procedures followed National Institutes of Health surface of the electrocyte before fusing with the noninner- guidelines under a protocol approved by the Cornell Uni- vated face on the opposite side of the cell. In other species, versity Institutional Animal Care and Use Committee. stalks are nonpenetrating, and the stalklets fuse on the in- nervated face of the cell. Species with nonpenetrating stalk Diagnosis of Operational Taxonomic Units and Choice of electrocytes produce EODs with only two phases, or peaks Specimens for Sequencing (Fig. 2B, left), whereas species with penetrating stalk elec- Each specimen was assigned to an operational taxonomic trocytes have EODs with an additional initial phase (Fig. 2B, unit (OTU) after examination of its external morphology and SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 601

FIG. 2. Operational taxonomic units (OTUs) are diagnosed using both external morphology and electric organ discharge (EOD) char- acteristics. (A) Tracings of the dorsal pro®les of the heads of four Gabon-clade Brienomyrus OTUs illustrate the difference between blunt snouts (top) and sharp snouts (bottom). The electrocytes in the electric organ have nonpenetrating stalks with posterior innervation (type NPp, left) or penetrating stalks with anterior innervation (type Pa), as illustrated by a schematic of a single electrocyte in sagittal view. (B) Four types of EODs recorded from each of the OTUs in (A). Individuals with type NPp electrocytes produce EODs with only two phases: peak P1 and peak P2. OTUs with penetrating stalks (type Pa) have EODs with an initial head-negative phase (P0) in addition to P1 and P2. For each OTU, multiple EODs have been normalized to the same peak-peak amplitude and superimposed with head positivity upward. (C) Within each category of snout and electrocyte type, we recognize distinct OTUs on the basis of more subtle, but consistent, differences in morphology and EODs. Illustrated are three allopatrically distributed sharp-snouted forms with Pa electrocytes. The ®rst, BEN, from the Ivindo River, has a relatively short EOD compared to the others. SP4 from the LoueÂtsi River has a similar morphology to the ®rst, but slightly longer EOD duration; SP7 from the Ntem River differs from the other two in head shape and in EOD duration.

EOD waveform, recorded in the ®eld. OTUs that did not ristic counts of ®n rays, scales, and teeth on 415 preserved correspond to described species were given unique, three- specimens. These measures were then subjected to separate character alphanumeric codes. We base some of our OTUs PCAs to look for natural clusters of individuals. We also on the natural groups derived from a principal component made quantitative measures of their signals by determining analyses (PCA) of morphometric, meristic, and EOD data. the time of occurrence, voltage, and slope at as many as 10 We measured 29 linear morphometric characters and 12 me- different landmark points on each EOD. For each individual, 602 JOHN P. SULLIVAN ET AL.

TABLE 1. List of all 89 specimens sequenced in this study, organized by operational taxonomic unit (OTU). Included is a brief explanation of the OTU abbreviation, the total number of voucher specimens in the ®eld samples, the number of individuals sequenced, the collection localities, and the GenBank, museum, and individual specimen numbers. NPp, electrocyte with nonpenetrating stalk, posterior innervation; Pa, electrocyte with penetrating stalk, anterior innervation.

Specimen OTU OTU description No. Seq. Locality Genbank no. Museum no. no. BEN manuscript abbr. 20 2 Ivindo basin AF477450 CU78338 2018 Ivindo River AF477455 CU78344 2295 Brienomyrus curvifrons described species 55 1 AF477469 CU81661 2050 B. hopkinsi described species 24 1 AF201575 CU78352 2285 B. longicaudatus described species 5 1 AF201576 CU78355 2289 BN1 blunt, NPp, 1 87 4 LoueÂtsi/Ngounie AF477458 CU84579 2682 AF477459 CU84566 2718 AF477432 CU80521 3124 AF477457 CU84579 2681 BN2 blunt, NPp, 2 53 1 Upper Ogooue basin AF477431 CU80474 3542 Boulengeromyrus knoepf¯eri described species 7 1 Ivindo River AF201573 CU79692 2248 BON manuscript abbr. 44 4 LoueÂtsi/Ngounie AF477470 CU84649 2656 AF477471 CU84649 2659 AF477472 CU80310 2979 AF477473 CU84567 2711 BP1 blunt, Pa, 1 210 9 Coastal/Mayumba AF477478 CU80520 3358 LoueÂtsi/Ngounie AF477460 CU84659 2684 AF477400 CU80341 3195 Lower Ogooue basin AF477403 CU80320 2818 Mouvanga/Ngounie AF477444 CU84583 2530 AF477443 CU80355 3016 AF477445 not catalogued 2704 Okano basin AF477441 CU81308 1638 Woleu basin AF477477 CU80892 3771 BP4 blunt, Pa, 4 1 1 Upper Ogooue basin AF477442 CU80511 3389 BP5 blunt, Pa, 5 46 1 AF477468 CU80862 2322 BP6 blunt, Pa, 6 17 1 AF477430 CU80476 3547 BP7 blunt, Pa, 7 4 1 Okano basin AF477424 CU81309 1634 BX1 blunt, mixed, 1 30 1 Coastal/Pointe Mbini AF477479 CU81264 4018 CAB manuscript abbr. 73 3 Ivindo River AF477466 CU80816 2116 Ntem basin AF477423 CU80586 1789 AF477422 CU80893 3848 IN1 intermediate, NPp, 1 15 1 Lower Ogooue basin AF477439 CU80591 1844 IP1 intermediate, Pa, 1 1 Ntem basin AF477421 CU80586 1692 Ivindomyrus opdenboschi described species 28 1 Ivindo River AF477480 CU81668 2184 LIB from Libreville 15 1 Coastal/Libreville AF477446 CU80867 2427 LIS manuscript abbr. 25 1 Upper Ogooue basin AF477427 CU81090 3366 MAG manuscript abbr. 69 4 Ivindo River AF477452 CU78326 2297 AF477451 CU78323 2168 Ntem basin AF477415 CU80904 3945 Ntem River AF477416 CU80902 3999 Marcusenius ntemensis described species 21 2 Ivindo River AF201593 CU79706 2186 Ntem River AF477418 CU80723 1616 NGO from Ngounie River 11 2 LoueÂtsi/Ngounie AF477463 CU84665 2708 AF477464 CU84665 2710 NZO from Nzoundou 4 2 Ngounie River AF477462 CU84661 2553 AF477461 CU84661 2552 OFF from Offoue River 16 2 LoueÂtsi/Ngounie AF477447 CU84667 2643 Upper Ogooue River AF477402 CU80526 3394 PAR manuscript abbr. 42 1 Upper Ogooue basin AF477419 CU80934 3461 Paramormyrops gabonensis described species 31 2 Ivindo River AF201603 CU79702 2048 Ntem River AF477425 CU80713 3980 SAN from Sangha River 8 1 Sangha River basin AF477467 AMNH231046 2485 SN2 sharp, NPp, 2 52 3 LoueÂtsi/Ngounie AF477437 CU80299 2969 AF477438 CU84664 2595 Upper Ogooue River AF477435 CU80458 3415 SN3 sharp, NPp, 3 11 3 Mouvanga/Ngounie AF477465 CU84603 2606 AF477456 CU84603 2619 AF477436 CU80356 3027 SN4 sharp, NPp, 4 76 2 Upper Ogooue River AF477434 CU80458 3396 AF477433 CU80463 3465 SN7 sharp, NPp, 7 2 1 AF477428 CU80496 3666 SN8 sharp, NPp, 8 1 1 Ntem River AF477420 CU80928 1611 SP2 sharp, Pa, 2 40 6 LoueÂtsi/Ngounie AF477398 CU81312 3203 SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 603

TABLE 1. Continued

Specimen OTU OTU description No. Seq. Locality Genbank no. Museum no. no. AF477399 CU84571 2672 Mouvanga/Ngounie AF477405 CU84582 2638 AF477404 CU80231 3013 AF477406 CU84578 2542 AF477407 CU84602 2585 SP4 sharp, Pa, 4 72 7 LoueÂtsi/Ngounie AF477408 CU80305 2995 AF477449 CU84602 2673 AF477448 CU84602 2671 Mouvanga/Ngounie AF477412 CU84658 2592 AF477409 CU80358 3014 AF477411 CU80357 3026 AF477410 CU80337 3342 SP6 sharp, Pa, 6 19 1 Upper Ogooue River AF477413 CU80485 3658 SP7 sharp, Pa, 7 32 1 Ntem basin AF477414 CU80877 3966 SP8 sharp, Pa, 8 8 1 Upper Ogooue River AF477417 CU80488 3657 SZA manuscript abbr. 80 4 Ivindo River AF477475 CU80848 2008 Ntem basin AF477440 CU80881 3839 Okano basin AF477429 CU80929 1633 Woleu River AF477401 CU80884 3788 TEN manuscript abbr. 61 3 Ivindo River AF477453 CU80809 2011 AF477454 CU80807 2191 Ntem basin AF477426 CU81311 3850 VAD manuscript abbr. 42 3 Ivindo River AF477474 CU79704 2105 AF201578 CU79740 2425 Woleu basin AF477476 CU80888 3814 TOTALS: OTUs: 41 Vouchers: 1457 89 Sequences: 89

FIG. 3. Map of west-central Africa with the collection localities of the specimens of the Gabon-clade Brienomyrus indicated as closed circles. All ®eld sites are in the either Gabon or the Central African Republic. Locality names indicated here identify the collection basin for each specimen listed in Table 1. 604 JOHN P. SULLIVAN ET AL. we subjected the separate measures to PCA. Of the 16 OTUs From a total of 1450 voucher specimens (Table 1), we derived from this analysis, four are described species of selected 85 individuals for sequencing representing all iden- Brienomyrus and 12 are being prepared for species descrip- ti®ed OTUs. When possible, we sequenced two or more rep- tions in separate publications (C. D. Hopkins, G. G. Teugels, resentatives of each OTU from different populations. In some R. J. Rundell, and J. P. Sullivan, unpubl. ms.; G. G. Teugels cases, we sequenced additional specimens of an OTU if the and C. D. Hopkins, unpubl. ms.). ®rst two sequences failed to form a monophyletic group in Additional OTUs were recognized without PCA and with a preliminary parsimony analysis. less formal morphological analysis than would be required for the preparation of species descriptions. This consisted of Choice of Molecular Marker careful side-by-side visual comparison of the external mor- phology of all specimens in the laboratory (including those We chose to sequence the complete mitochondrial cyto- previously treated by PCA). We also examined and compared chrome b gene because two earlier studies using this marker the EOD waveforms of each specimen assigned to an OTU in mormyrids (Lavoue et al. 2000; Sullivan et al. 2000) in- by overlaying them on a computer screen with similar EODs dicated its utility and provided outgroup sequences. in our database. Few of these OTUs are diagnosed by single, In an effort to obtain a complementary but independent unique phenotypic character states. However, within each dataset from nuclear markers, we assayed a number of nuclear broad class of snout shape and electrocyte type (Fig. 2A), intron and spacer regions, most of which have been used covariance of variable phenotypic features among these spec- successfully in species-level and population-level ®sh stud- imens such as relative head width, forehead slope, eye size, ies. We were successful in obtaining sequences of 400±800 degree of lower jaw protrusion, caudal peduncle width, scale bp for a subsample of several Gabon-clade Brienomyrus counts, and EOD waveform characteristics is not random, but OTUs for ®ve such markers: (1) RAG 1, second intron, prim- usually delineates discrete and recognizable forms between ers designed from data in Willet et al. (1997); (2) S7 introns which we do not observe intermediates. New OTUs were 1 and 2, primers from Chow and Hazama (1998); (3) cal- created when we judged specimens to differ consistently in modulin intron, primers from Chow (1998); (4) GABA intron, the combination of these characteristics from all those pre- primers from J. M. Quattro (pers. comm.); and (5) an LDH- viously examined. Because in most cases both male and fe- A intron, primers from T. Dowling (pers. comm.). In each male specimens were available (sex can be determined easily case, variable sites among the ®ve or six taxa sequenced were in adults from the appearance of the anal ®n), expected sexual too few (maximum pairwise P-distance Ͻ 1%) to warrant dimorphisms in morphology and EOD waveforms could be sequencing all individuals used in this study. We likewise accounted for in these comparisons. assayed the internal transcribed spacer 1 (ITS1) region of the We suspect that many, if not most, of these additional nuclear ribosomal RNA genes with a number of different OTUs will merit species status upon more detailed and formal study. For the purposes of this study, however, the avoidance primers from the literature. Although we were successful in of recognizing polyphyletic assemblages as OTUs was of obtaining clean ITS1 sequences from mormyrid species out- more importance than determining placement of the species side the study group, all sequence chromatograms from Ga- level, which would be required in a work of taxonomy. For bon-clade Brienomyrus appeared to contain signal from mul- example, the relatively subtle differences we used to distin- tiple polymorphic ITS1 copies. For this reason, we limited guish the three allopatric OTUs in Figure 2C could be in- our dataset to mitochondrial cytochrome b sequences. terpreted as evidence for three separate species, or alterna- tively as geographic variation among disjunct populations of Choice of Outgroups a single widespread species. When differences were consis- tent, as in this case, we recognized separate OTUs. Our ex- Outgroup cytochrome b sequences were available from our pectation was that if conspeci®c populations were divided previous study (Sullivan et al. 2000) in which we had se- into multiple OTUs, sequences from these would cluster to- quenced the complete cytochrome b gene from 41 species gether in the phylogenetic analysis. In these cases, subsequent belonging to all 19 recognized genera of mormyroid ®shes. taxonomic studies could determine placement of the species For this analysis, we chose outgroup sequences from the two level. We recognized that such potential splitting would be nearest, consecutive outgroups to the clade formed by the less problematic for the phylogenetic analysis than would four Gabon-clade Brienomyrus taxa included in the Sullivan mistaken lumping of nonsister lineages. Our procedure for et al. (2000) and LavoueÂ, J. P. Sullivan, and C. D. Hopkins, recognizing smallest diagnosable clusters conforms in prin- (unpubl. ms.) studies (see Fig. 1): M. ntemensis, B. knoepf¯eri, ciple to the phylogenetic species concept of Cracraft (1983), and I. opdenboschi. Sequences from two individuals of out- although it is not our opinion that species as they are rec- group taxon M. ntemensis were used: one from the Ivindo ognized in taxonomy must always be such. River and another from the Ntem River.

→

FIG. 4. Photographs of live or preserved specimens from all 38 OTUs recognized in this study and their electric organ discharge (EOD) waveforms, organized by general snout shape and electrocyte type. Also shown are photographs of the holotypes of three described species in this group that we could not con®dently associate with those in our collections. All EODs are plotted on the same time base with head positivity upward. For OTUs with pronounced sex differences in EODs, both male and female waveforms are shown. SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 605 606 JOHN P. SULLIVAN ET AL.

Polymerase Chain Reaction and Sequencing positions. Assumed nucleotide frequencies were those de- termined empirically from the dataset, a molecular clock was We extracted DNA from tissue samples with the QIAamp not enforced, and starting branch lengths were obtained using tissue kit (Quiagen, Inc., Valencia, CA). Sequences of the the Rogers-Swofford approximation method (Rogers and primers used to amplify the mitochondrial cytochrome b gene Swofford 1998). TBR branch swapping began from a starting were taken from Palumbi (1996) and are: 5Ј-TGA TAT GAA tree obtained by neighbor joining. The option to collapse AAA CCA TCG TTG-3Ј (L14724) and 5Ј-CTT CGA TCT branches of insigni®cantly different length was used. Because TCG rTT TAC AAG-3Ј (H15930). Polymerase chain reaction of the impractical amount of computation time required to (PCR) volumes of 50 ␮l consisted of approximately 100±500 complete branch swapping in a parameter-rich analysis of a ng genomic DNA, 1X GeneAmp Gold PCR buffer (PE Ap- dataset this large, we halted the analysis after it had found plied Biosystems, Foster City, CA), 1.25 units of AmpliTaq a best tree that remained unchanged through 2000 subsequent Gold (PE Applied Biosystems), concentrations of 0.2 ␮Mof branch swappings. The substitution rate matrix parameters each primer, 200 ␮M of each dNTP, and 3 mM of MgCl. and codon-speci®c substitution rates estimated for this tree We carried out PCR on a Hybaid TouchDown thermocycler were then entered as ®xed parameters in a subsequent heu- (Hybaid Limited, Teddington, Middlesex, England) using an ristic ML search in which branch swapping was allowed to initial 95ЊC denaturation step for 10 min followed by 35 continue to completion. repetitions of a three-step cycle consisting of 94ЊC for 1 min, Mutational saturation in the dataset was estimated by look- annealing for 1 min at 42ЊC, and extension at 72ЊC for 1.5 ing for nonlinearity in a plot of pairwise observed raw dis- min, followed by a ®nal extension at 72ЊC for 7 min. We tance (PAUP*'s adjusted character distance) against tree-cor- puri®ed our PCR products with the Wizard PCR Preps DNA rected distance (PAUP*'s patristic distance) for each codon puri®cation kit (Promega, Madison, WI). We sequenced the position and each class of nucleotide substitution on one of double-stranded PCR products directly in both directions the most parsimonious trees resulting from the MP analysis, with the primers used for ampli®cation by automated dye- as described by Hassanin et al. (1998). terminator cycle chemistry on a PE Applied Biosystems 377 automated sequencer. We edited the sequences from the gel RESULTS chromatograms with the Sequencher software package (Gene Codes Corp., Ann Arbor, MI). Operational Taxonomic Unit Diagnosis We recognized 38 OTUs among our 1450 collected spec- Phylogenetic Analysis Methods imens (Fig. 4, Table 1). Four of these we identi®ed as Brien- We reconstructed cytochrome b phylogeny using both omyrus curvifrons (Taverne et al. 1977b), B. longicaudatus maximum-parsimony (MP) and maximum-likelihood (ML) (Taverne et al. 1977b), B. hopkinsi (Taverne and Thys Van analyses in PAUP* version 4.0b3 (Sinauer, Inc., Sunderland, Den Audenaerde 1985), and Paramormyrops gabonensis MA) on an Apple PowerMac G4 computer. For the MP anal- (Taverne et al. 1977a) based on comparison to type material ysis, we performed a heuristic search on the cytochrome b and the original descriptions. We could not con®dently as- dataset in which all characters and classes of substitution sociate three other previously described Gabon-clade Brien- were equally weighted. Starting trees for tree bisection-re- omyrus species from this region, B. sphekodes (Sauvage connection (TBR) branch swapping were obtained by 100 1880), B. kingsleyae (GuÈnther 1896), and B. batesii (Boulen- iterations of the random stepwise addition sequence. ger 1906) to any of our remaining 34 OTUs. However, we PAUP*'s default settings for a heuristic MP search were used cannot rule out the possibility that these three species are in in all other cases. Relative support for the internal nodes of fact represented among them. In the case of B. sphekodes and the MP trees was estimated by bootstrap analysis (Felsenstein B. kingsleyae, for which no illustrations were provided in the 1985) consisting of 1000 pseudoreplicates in PAUP* (starting original descriptions, the poor condition of the type speci- trees obtained by single iteration of random stepwise addi- mens made comparisons to our recently collected specimens tion; MAXTREES set to 2000; otherwise parameter settings dif®cult. Although the specimens we assigned to the OTU were identical to MP heuristic search). VAD in this study resemble the types of B. batesii, we hes- For the ML analysis, we performed an initial parameter- itated to make the identi®cation based on a consistent dif- rich heuristic tree search using the general-time-reversible ference in the number of circumpeduncular scales. model with rate heterogeneity (Yang 1994) in which the six- In three unusual cases (MAG and BEN, SP2 and SP4, and way substitution rate matrix was estimated from the dataset BP1 and BN1) we encountered cases of sympatric, morpho- by ML, as were site speci®c rates for each of the three codon logically indistinguishable forms within which we observe

→

FIG. 5. A strict consensus tree of 680 equally parsimonious trees, each of 636 steps, shown as a phylogram using ACCTRAN character optimization, produced from an unweighted parsimony analysis of the complete cytochrome b gene from Gabon-clade Brienomyrus and outgroup taxa. For all trees, CI ϭ 0.49, RI ϭ 0.84, RC ϭ 0.41 with uninformative sites excluded. Table in inset lists the different OTUs, according to the origin of the names. OTUs NGO, NZO, OFF, and SAN all are abbreviations for one of the collecting localities where these ®sh were captured. Some of the OTU names refer to manuscript names of two manuscripts currently in preparation; others refer to the shape of the snout (B, blunt; I, intermediate; S, sharp) and to the type of electrocytes in the electric organ (NPp, nonpenetrating stalk, posterior innervation; Pa, penetrating stalk, anterior innervation). The remaining OTUs are described species. The letters in this and the following ®gure refer to nodes that are discussed in the text. SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 607 608 JOHN P. SULLIVAN ET AL. two different EOD waveforms (taking sexual dimorphisms Mean p-distance between M. ntemensis and the Gabon-clade into account). We chose to recognize them provisionally as Brienomyrus taxa is 8.3%. separate OTUs in the analysis, although we consider it pos- Within the Gabon-clade Brienomyrus, nodes B, C, D, E, sible that each pair may represent a single species with an and F (Figs. 5, 6) are well supported by long internal branches EOD polymorphism. We sequenced more of these forms for and high parsimony bootstrap values. Nodes B and E de®ne the analysis than we did for most other OTUs with the hope the most basal division within the Gabon-clade Brienomyrus. of resolving their relationships. The OTUs in these two clades are separated from each other by 7% p-distance. Clade B includes the OTUs VAD and SZA, Dataset Characteristics both of which belong to separate, well-de®ned clades (D and C, respectively, separated from each other by an average 6.5% Sixty-®ve unique 1140 bp cytochrome b haplotypes were p-distance. The two VAD specimens sequenced, one from recovered from the 85 Gabon-clade Brienomyrus specimens the Woleu River basin and the other from the Ivindo, appear sequenced. These made up the ingroup dataset for the par- as a monophyletic group which is sister to OTU BP7 1634 simony analysis. We observed no indels, nonsense mutations, from the Okano River basin. Sister to these taxa is clade C or ambiguous sites in any of the ABI chromatograms that in which two distinctly different OTUs, IP1 and LIS, both might indicate ampli®cation of extramitochondrial copies of known from only single populations, are nested within the the gene. Maximum uncorrected p-distance between the out- SZA OTU, a widespread form found in the Ivindo, Ntem, group and ingroup sequences is 10%. We observed no evi- Woleu, and Okano Rivers in northern Gabon. All members dence of mutational saturation for transitions or transversions of clade B have penetrating stalk±type electric organs (Fig. in any codon position in our plots of pairwise adjusted char- 7D). acter distance against patristic distance as calculated in The most basal division in the remaining taxa separates PAUP*. the OTU BON, an OTU only known from the LoueÂtsi River Of 287 variable sites in the sequences, 232 were parsi- in southern Gabon, from clade F. Clade F, which is supported mony-informative characters: 188 (81%) in the third-codon by an exceedingly long branch relative to others on the tree, position, 33 (14%) in the ®rst-codon position, and 11 (5%) is in turn divided into a large clade (G) whose sister group in the second-codon position. Average base frequencies consists of two individuals of the OTU BP1: one from the across all sites are C ϭ 32%, A ϭ 29%, T ϭ 25%, G ϭ 14%, Okano River, the other from the Woleu River Basin. The but the informative sites of the third-codon position show sequences of these two BP1s differ from those in clade G by increased high C, low G bias (C ϭ 43.5%, A ϭ 29.5%, T ϭ an average of 3.0% p-distance. 22%, G ϭ 5%) as reported previously for mormyroid ®shes Outside of clade G, all ingroup OTUs have penetrating (Lavoue et al. 2000; Sullivan et al. 2000), for other ®shes stalk type electric organs, the apparent primitive condition (Meyer 1993; Lydeard and Roe 1997), and for other verte- for the entire subfamily Mormyrinae (Sullivan et al. 2000). brate groups (Irwin et al. 1991; Kornegay et al. 1993). A chi- Within clade G, OTUs have a mix of penetrating and non- squared test failed to reveal any signi®cant heterogeneity in penetrating stalk±type electric organs (Fig. 7D). base frequencies among the sequences. Within clade G, the remainder of the Gabon-clade Brien- omyrus (32 OTUs) segregate into three relatively well-sup- Phylogenetic Analysis Results ported cladesÐH, I, and JÐthat are of uncertain relationship An unweighted MP analysis in PAUP* yielded 680 trees, to each other and that are poorly resolved internally. Within each of 636 steps (CI ϭ 0.49, RI ϭ 0.84, RC ϭ 0.41, un- these clades, sequences of different OTUs often differ by informative sites excluded). A strict consensus of these trees, 1.0% p-distance or less, whereas in some cases specimens depicted as a phylogram (using ACCTRAN character opti- assigned to distinct OTUs share identical cytochrome b hap- mization) is shown in Figure 5. A consensus tree of 1000 lotypes (e.g., BP6 3547/BN2 3542; BP1 3016, 2530, 2704/ parsimony bootstrap pseudoreplicates in which nodes re- SN3 2619). ceiving less than 50% bootstrap level support are collapsed Furthermore, only two OTUs represented by sequences is shown in Figure 6. Well-supported nodes de®ning major from multiple populations in this large clade appear as mono- clades are labeled A±J in Figures 5 and 6. phyletic groups on the tree (OFF in Figs. 5, 6; P. gabonensis The ML analysis yielded a tree with an ln-likelihood score in Fig. 5). Other OTUs (SP2, SP4, MAG, TEN, SN2, SN3, of 4958.3. Estimated relative substitution rates were 0.336 CAB, NZO, and BP1) represented by multiple sequences, for position 1, 0.08 for position 2, and 2.58 for position 3. either from the same or from different populations, appear Because branches of insigni®cantly different length were col- nonmonophyletic on the tree. The remaining OTUs above lapsed during the ML search, some nodes are not resolved. node G are represented by single individuals; therefore, their The topology of this ML tree (not shown) is extremely similar monophyly cannot be tested. to that of the MP consensus tree: The only topological in- In summary, the basal clades in the tree (below clade G) congruence between them concerns the pattern of interrela- are characterized by long internal branch lengths and OTU tionships of clades H, I, and J (Figs. 5, 6), none of which sequence monophyly or at least coherence (although SZA have signi®cant character support. sequences are paraphyletic with respect to IP1 and LIS, they Both analyses indicate that M. ntemensis is the sister group are at least restricted to a single, well-de®ned clade C), where- to all ingroup specimens sequenced (node A; Figs. 5, 6), as the large clade G is characterized by short internal branches supporting the monophyly of the additional, putative Gabon- and general OTU sequence polyphyly. This pattern in clade clade Brienomyrus with those included in previous studies. G is not in general a consequence of lack of resolution of SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 609

FIG. 6. Consensus MP bootstrap tree based on 1000 pseudoreplicates on the same dataset used for Figure 5. Nodes supported by bootstrap proportions greater than or equal to 50% are shown with bootstrap values indicated. 610 JOHN P. SULLIVAN ET AL.

FIG. 7. The cytochrome b tree fails to recover higher level groups of Gabon-clade Brienomyrus suggested by shared morphological and electric organ character states, all presumably derived within the group based on outgroup comparison. The phylogenetic tree from Figure 5 is reproduced showing the presence (in black) or absence (gray) of each character state for each OTU. (A) OTUs MAG, BEN, SP2, SP4, SP6, SP7, and SP8, B.curvifrons, and B.hopkinsi all share very sharp snouts with terminal mouths and a jutting lower jaw (inset). Yet these taxa do not form a monophyletic group on the cytochrome b haplotype tree. (B) OTUs with 16 instead of 12 circumpeduncular scales (inset) do not form a monophyletic group and OFF and B. longicaudatus that additionally share an elongate caudal peduncle, a distinctive sloping head shape, and large adult size do not appear as sister taxa. (C) OTUs TEN and BN2 that share monophasic EOD waveforms (inset) similarly do not form a monophyletic group. (D) OTUs SP6, SP8, MAG, and SP2 all possess reversed polarity EOD waveforms (inset) in which the initial head negative P0 derived from current ¯owing through penetrating stalks, becomes a major phase of the EOD and P2 is reduced. These OTUs do not form a monophyletic group on the tree. (E) OTUs possessing nonpenetrating stalk electrocytes with posterior innervation (type NPp, inset) do not form a monophyletic group. SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 611 the data. In many cases, the nonmonophyly of same-OTU nome between populations of closely related species is not sequences is supported by high bootstrap values (e.g., for uncommon (Harrison 1989). Cases in our data particularly TEN, SP4, SN2, SN3, BP1, and BN1). Furthermore, haplo- suggestive of introgression are those in which a specimen's types of allopatric OTUs that we presumed to be closely haplotype is a nearer relative to the haplotype of a different, related due to relatively minor differentiation among them but sympatric OTU than it is to haplotypes of the same OTU (e.g., BEN, SP4, SP7 in Fig. 2C; B. longicaudatus and OFF from another population. An example of this is observed on and B. hopkinsi and SN8) do not appear as nearest relatives the tree with the sequence data from the OTU TEN. This on the tree. OTU is among the most recognizable forms in our collections The status of the sympatric OTU pairs that are morpho- due to its elongate body, small eyes, protruding lower jaw, logically indistinguishable but that have different EODs and monophasic EOD. Yet cytochrome b haplotypes of two (MAG and BEN, SP2 and SP4, and BP1 and BN1) remains individuals from the Ivindo River population (2011, 2191) unclear. In no case do the haplotypes of these OTUs (either cluster with those of other Ivindo OTUs (clade H in Figs. 5, singly, or with its pair) appear monophyletic on the tree, 6) rather than with the haplotype of the TEN specimen 3850 although there are cases of haplotype identity (or near iden- from the Ntem River, about 200 km distant. In other cases, tity) within each pair. specimens of distinct, but sympatric OTUs share identical haplotypes (e.g., in clade I: BP6 3547 and BN2 3542 and DISCUSSION SN7 3666 & SN2 3415; in clade J: BP1 3016, 2530, 2704, and SN3 2619. Strangely, throughout much of clade G, geo- Explaining Operational Taxonomic Unit and Haplotype graphical proximity seems to be a more consistent predictor Tree Incongruence of clade membership than is OTU identity. For instance, all The departure of the tree topology in clade G from ex- the haplotypes appearing in the large clade J are from a single pectations of OTU sequence monophyly and of close rela- region of southern Gabon, although other haplotypes of these tionships between sequences of similar OTUs may indicate same OTUs and from the same localities appear again in either real incongruence between the mitochondrial haplotype clades H and I. It is dif®cult to conceive how any process tree and the overall organismal phylogeny or poor corre- other than introgression across multiple OTU boundaries spondence between our OTUs and monophyletic groups of could produce this pattern. these ®shes. Evaluating Possible Problems with Operational Taxonomic Lineage Sorting Unit Diagnosis Gene tree±species tree incongruence caused by incomplete The cytochrome b haplotype tree is consistent with patterns gene lineage sorting has been implicated in a number of that would be produced by incomplete lineage sorting and studies of other ®sh species ¯ocks (Moran and Korn®eld introgression in clade G, although without additional data, 1993; McMillan and Palumbi 1995; Strecker et al. 1996; assessing the relative importance of each is dif®cult. Alter- Korn®eld and Parker 1997; Parker and Korn®eld 1997) and natively, if our OTUs fail to represent natural groups, the in Darwin's ®nches (Sato et al. 1999). This situation results nonmonophyly of their cytochrome b haplotypes would be when the rate of lineage splitting or speciation exceeds the expected. We could have made three kinds of errors in these rate of stochastic sorting of allelic polymorphisms within an diagnoses. If errors of any one of these kinds were made organismal lineage. The phylogeny of alleles sampled in systematically, we should be able to detect them post hoc, these cases may differ from larger organismal phylogeny because each results in different predictions. (Pamilo and Nei 1988; Harrison 1991). In the ®rst case, we may have been deceived by convergent Incomplete mitochondrial lineage sorting is suggested by evolution into inadvertently lumping a number of different the presence of divergent haplotypes within single popula- natural groups into single OTUs (i.e., we designated fewer tions (Moran and Korn®eld 1993). There are several exam- OTUs than there are natural groups or species). We accept ples of this in our dataset. For instance, the haplotypes of that convergence of individual characteristics is possible and SN3 3027 and 2619 are nested within clade J, whereas SN3 perhaps likely in some cases. However, for convergent evo- 2606 from the same population is nested within clade I. Hap- lution to explain the examples outlined above, both mor- lotypes of SP4 2671 and 2673 are nested together in clade phology and EOD features would have had to coevolve iden- H, whereas that of SP4 2995 from the same population is tically and repeatedly in different localities. We know of no nested within clade J. These results suggest that it may be reason to believe that selection should favor the association futile to interpret the pattern of haplotype relationships within of a particular EOD waveform with a particular external mor- clade G in terms of overall organismal relationships. phology. For some particularly distinctive OTUs, such as the OTU TEN mentioned above, we ®nd it impossible to believe Introgression that the similarity between populations is due to convergence The second explanation for gene tree±species tree incon- and not to common ancestry. gruence is introgression by hybridization (Smith 1992). This In the second case, the mistaken use of intraspeci®c poly- is possible among interfertile species in which reproductive morphisms or plastic characters in our OTU diagnoses could isolation is maintained by extrinsic barriers or by behaviors explain OTU/cytochrome b haplotype tree incongruence (i.e., that break down under some conditions (Arnold 1997; Dowl- we designated more OTUs than there are natural groups or ing and Secor 1997). Introgression of the mitochondrial ge- species). If many of our OTUs are indeed unnatural groupings 612 JOHN P. SULLIVAN ET AL. below the species level, one would predict that the distri- strengthens the case for the role of incomplete lineage sorting bution of those derived phenotypic character states common and introgression in producing the general pattern of non- to two or more OTUs should more often correspond to mono- monophyly of same-OTU haplotypes in clade G, as we sug- phyletic groups of cytochrome b haplotypes than those we gest above. used to diagnose single OTUs, and that examples of haplo- This failure of cytochrome b sequences to resolve mono- type identity between individuals placed in different OTUs phyletic OTUs in clade G (which contains 32 of the 38 Ga- may often be due to their conspeci®c status. bon-clade Brienomyrus we recognized) renders the topology We consider the ®rst of these predictions in Figure 7. Ac- of clade G useless as a framework upon which to map electric cepting the monophyly of clade G, the tree topology below organ and EOD characteristics with a view to inferring evo- node G and the outgroup relationships shown in Figure 1, lutionary patterns. However, we do note with interest that we identi®ed ®ve characters states, each of which appears only in this clade, characterized by great phenotypic diversity uniquely within several OTUs of clade G (shown in black in and apparently little concomitant genetic divergence, the ex- Fig. 7A±E). These character states are: (1) the very sharp clusive appearance in some OTUs of presumably derived snout combined with jutting lower jaw morphology of nine NPp-type electrocytes, in addition to the presumably prim- OTUs (Fig. 7A); (2) 16 circumpeduncular scales found in itive Pa-type electrocytes in others. Sullivan et al. (2000) four OTUs versus 12 in all others (Fig. 7B); (3) monophasic noted that species with the derived NPp-type electrocytes EOD waveforms (vs. typical biphasic or triphasic waveforms; usually have longer duration EODs than do those with Pa- Fig. 7C); (4) EOD waveforms with polarity reversed with type. The general characteristics of clade G are consistent respect to typical EODs (Fig. 7D); and (5) nonpenetrating with our hypothesis in Sullivan et al. (2000) that selection stalk electrocytes (Fig. 7E). These character states all appear may favor reversal to nonpenetrating stalk electrocytes in to be derived within the Gabon-clade Brienomyrus based those primitively penetrating stalk lineages in which contact upon outgroup comparisons. However, none of these char- between many recently speciated forms necessitates novel acter states (or their alternate states) correspond to mono- EOD characteristics, such as increased duration, which can phyletic groups of cytochrome b haplotypes. enhance species recognition (Hopkins and Bass 1981). Un- The second prediction of the hypothesis that many of our fortunately, our inability to posit OTU interrelationships in OTUs demarcate unnatural groups below the species level is clade G from the cytochrome b data precludes closer ex- the conspeci®c status of those sympatric specimens from dif- amination of this character displacement hypothesis. ferent OTUs which share cytochrome b haplotypes. As al- ready stated, we recognize that this could be the case for Estimating the Age of the Gabon-Clade Brienomyrus three morphologically cryptic OTU pairs (MAG and BEN, SP2 and SP4, and BP1 and BN1) within which OTUs were Similar terminal branch lengths in Figure 5 and in the ML distinguished solely by EOD characteristics. However, in oth- analysis phylogram (not shown) suggest that base substitu- er cases of haplotype identity or near identity, such as that tions are accumulating at relatively equal rates in different of SN3 2619 and SN3 3027 to haplotypes of three BP1 spec- cytochrome b lineages of Gabon-clade Brienomyrus. Using imens (Fig. 5, clade J), the substantial and consistent mor- the software RRTree (Robinson et al. 1998), we conducted phological and EOD differentiation between the forms coun- relative rates tests on the sequences when grouped into the terindicates their conspeci®city. We anticipate a more thor- seven major clades recovered in the parsimony analysis (la- ough resolution of this issue from microsatellite studies cur- beled clades H, I, J, C, D and unlabelled clades BP1 1638/ rently underway in communities of sympatric Gabon-clade BP1 3771, BON in Figs. 5, 6). They revealed no signi®cant Brienomyrus (M. A. Arnegard, unpubl. ms.). differences in evolutionary rates in any pairwise comparison The third type of error that would have produced a pattern between groups, relative to the outgroup sequences from M. of nonmonophyly of OTU cytochrome b haplotypes is the ntemensis. diagnosis of OTUs by evolutionarily primitive (plesiomorph- Alves-Gomes (1999) has proposed a molecular clock rate ic) characteristics for the group. OTUs diagnosed by prim- in mormyroid ®shes for the mitochondrial 12S and 16S rRNA itive characteristics would appear paraphyletic on the tree genes of 0.23% per million years. Cytochrome b distances with respect to other OTUs and would produce a pattern are on average 3.0 times greater than 12S and 16S distances similar to that observed for the OTU SZA with respect to in corresponding pairwise comparisons of mormyrid se- LIS and IP1 in clade C (Figs. 5, 6). The use of additional, quences in Sullivan et al. (2000). Thus, accepting Alves- independent markers may con®rm that the widespread OTU Gomes's suggested 12S/16S clock rate implies a cytochrome SZA is truly paraphyletic with respect to the allopatric OTUs b clock rate in the neighborhood of 0.7% per million years. LIS and IP1 that are known only from single populations. To calculate this clock, Alves-Gomes (1999) used mean However, as in this case, OTU paraphyly should often be uncorrected 12S and 16S p-distance between 12 mormyroids interpretable as such, because some OTU coherence on the and the taxon Chitala chitala from the sister group to the tree will be maintained. Because this pattern is not clearly Mormyroidea, the notopteroids. He used the 65 million-year- repeated for OTUs in clade G, we think it unlikely that use old fossil Ostariostoma, putatively the sister taxon to these of primitive character states in OTU de®nition explains all two groups (Li and Wilson 1996), as a calibration point. the cases of OTU nonmonophyly. Recently, however, a fossil notopterid from the Middle Cre- Some of the OTUs we recognized may not represent natural taceous of Morocco, Paleonotopterus greenwoodi, has been species. However, these considerations suggest that no single described and phylogenetically placed as the sister group to type of error was repeatedly made in their diagnosis and the extant notopteroids (Forey 1997; Taverne and Maisey SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 613

1999; Taverne 2000), pushing back the minimum age of the ¯eri, I. opdenboschi, and P. marchei) are themselves OgooueÂ/ notopteroid lineage and thus the mormyroid/notopteroid di- Ntem endemics. In fact, all but one of these outgroup taxa vergence to 100 million years ago. By itself, this ®nding (P. marchei) are further restricted to the Ntem and an adjacent implies that the Alves-Gomes clock rate is overestimated by Ogooue tributary, the Ivindo River, the upper portion of at least a factor of 1.5. However, pairwise mormyroid and which is believed to have been captured from the Ntem at notopteroid 12S and 16S p-distances plotted against the cor- some time in the past (Thys van den Audenaerde 1966; Olivry responding p-distances from the much more slowly evolving 1986). Fish dispersal between the adjacent Ntem and the RAG2 nuclear gene (in ®g. 1 of Sullivan et al. 2000) suggest Ivindo headwaters may continue to occur during periods of that 12S/16S p-distance underestimates the actual divergence high water. Assuming an origin in the OgooueÂ/Ntem region, in these genes between mormyroids and notopteroids by at (perhaps in an ancient Ntem/Ivindo), members of the Gabon- least a factor of three, due to signi®cant substitutional sat- clade Brienomyrus would have subsequently dispersed into uration at this phylogenetic distance. Applying these correc- the Congo basin and into neighboring lower Guinea drain- tions doubles the implied (maximum) clock rates to 0.46% ages. per million years for 12S/16S and to about 1.4% per million The diversity and endemism of Gabon-clade Brienomyrus years for cytochrome b. This conjectural cytochrome b clock species in the Ogooue and Ntem basins is paralleled in the rate is roughly in the middle range of mitochondrial coding similarly forest-dependent killi®sh genus Aphyosemion for gene clock rates (0.9±2.5% per million years) estimated for which Wildekamp (1993) listed 39 species and subspecies ®shes or applied to them in other studies (McCune 1997). from this region. The Ntem and part of the Ogooue basin are The con®dence interval on this estimate for a cytochrome positioned within Hamilton's (1982) ``Cameroon/Gabon core b clock rate in mormyroids must be thought to be exceedingly area'' of species richness for forest-associated groups of ter- wide. Nevertheless, applying this rate to the data implies that restrial plants and animals. These contemporary patterns of the Gabon-clade Brienomyrus stem group arose roughly 6 organismal distribution, in conjunction with pollen-based re- million years ago (based on an average pairwise uncorrected constructions of ¯oral change, indicate the persistence of distance of 8.3% between M. ntemensis vs. taxa in clade A several lowland forest refugia in portions of the Ogooue and in Figs. 5, 6). Likewise, this rate implies an age of about 2 Ntem basins throughout the most arid periods of the Pleis- million years for the large clade G (based on the mean pair- tocene (Maley 1987, 1991, 1996). During these periods most wise uncorrected distance of 3.0% between BP1 1638/BP1 of the modern equatorial forest belt of central Africa was 3771 and members of clade G). Thus, it is possible that the dominated by savanna. Combined, the phylogeographic, pa- bulk of the Gabon-clade Brienomyrus diversity originated leoecological, and molecular clock estimations point to an in within the Quaternary, and much of it within only the past situ diversi®cation of Gabon-clade Brienomyrus in the 500,000 years. OgooueÂ/Ntem region during the past 2 million years. Endemism and Phylogeographic Patterns The Gabon-Clade Brienomyrus as a Riverine Species Flock The precise distributional boundaries of the Gabon-clade Brienomyrus in African freshwaters remain poorly known, Within ®shes, the term ``species ¯ock'' has largely been but available data suggest that they may be restricted to river used in intralacustrine contexts; the most famous examples basins of lower Guinea (from the Sanaga River of Cameroon are the haplochromine cichlid radiations of the East African south to the Kouilou-Niari River of the Republic of Congo) Great Lakes (Brooks 1950; Echelle and Korn®eld 1984; and to the Congo River basin (Teugels and Hopkins 1998). Greenwood 1984; Meyer et al. 1990; Goldschmidt 1996; Their center of diversity appears to be within the Ogooue Korn®eld and Smith 2000). The term has also been applied and Ntem River basins of Gabon and Cameroon. To date, to tilapiine cichlids of the crater lakes of Cameroon (Schliew- most of our collection effort has been concentrated in this en et al. 1994); sculpins of Lake Baikal (Taliev 1955; Berg region, although this conclusion is supported by study of the 1965); cyprinids in both Lake Lanao, Philippines (Korn®eld existing literature and all available museum collections. The and Carpenter 1984), and in Lake Tana, Ethiopia (Nagelkerke diversity of this mormyrid clade appears to be lower in the et al. 1994); killi®shes in Lake Titicaca (Parenti 1984; Parker Congo basin. In our collections from forest streams in the and Korn®eld 1995); pup®shes in Lake Chichancanab, Mex- Sangha River basin of the Central African Republic, a Congo ico (Humphries 1984); and to a Mesozoic radiation of se- River tributary, we found only a single species from this clade mionotid ®shes in North American lakes (McCune et al. 1984; (SAN in our analysis). Similar habitats in the Ogooue basin McCune 1996). Arguing that the term need not be restricted typically host four to six distinct forms. This pattern is re- to lacustrine ®shes, Johns and Avise (1998) have applied it versed for the mormyrid genera Marcusenius, Campylomor- to radiations of northeastern Paci®c Sebastes rock®shes and myrus, Stomatorhinus, Mormyrops, and Mormyrus, which are to Antarctic nototheniod ice®shes (Ritchie et al. 1996) for species-rich in most Congo basin localities, but are repre- which there is evidence of explosive speciation in the past. sented in low diversity, or not at all, in the rivers of lower Although authors differ on the de®nition of a ®sh species Guinea. ¯ock (see Greenwood 1984; Ribbink 1984), we use the term In addition to being the center of Gabon-clade Brienomyrus to mean a monophyletic assemblage of species, at least large- diversity, the Ogooue and Ntem Rivers are also the group's ly restricted to the geographical area of their origin (i.e., probable center of origin, because the two nearest, sequential autochthonous), exhibiting a high level of sympatry, and rap- outgroups (M. ntemensis and the clade containing B. knoepf- id, or explosive, speciation relative to their nearest relatives 614 JOHN P. SULLIVAN ET AL. in neighboring regions. We have shown how all of these sionary Alliance in Bongolo, CIRMF in Franceville, J. Beck criteria seem to apply to the Gabon-clade Brienomyrus. and C. Ella from the U.S. Peace Corps, and C. Aveling of Because river basins, like lakes and islands, are habitats ECOFAC. For help in the Central African Republic, we thank circumscribed by a boundary inside of which gene ¯ow is J. B. Kindi-Moungo and the WWF of®ce in Bangui. From possible, but across which dispersal and invasion are rare Cornell, M. Arnegard helped collect ®shes, EODs, and tissues events, diversi®cation of species ¯ocks can take place within in Gabon, as did J. Friel, who additionally oversaw the cur- them. However, unlike many of the lakes hosting ®sh species ation of specimens at the Cornell Museum of Vertebrates. G. ¯ocks that began as biologically depauperate environments, Harned prepared histological slides of electric organs. We the rivers of west-central Africa have undoubtedly always thank M. L. J. Stiassny and J. Cracraft of the American Mu- harbored ®sh communities, although their courses and inter- seum of Natural History for making possible JPS's collection connections have changed dynamically through time. Perhaps trip to the Central African Republic. M. Arnegard, J. G. Lund- as a consequence of long-term occupation of their riverine berg, A. R. McCune, G. Teugels, B. Turner, and K. Zamudio environments by numerous other groups of ®shes, the Gabon- provided helpful comments on an early draft of the manu- clade Brienomyrus ¯ock seems not to represent an adaptive script. Funding for this work came from the following grants radiation in which species have diversi®ed greatly in ecology to C.D. Hopkins: National Science Foundation International and morphology. Program Grant INT-9605176, National Geographic Society 5801-96, and the National Institute of Mental Health Conclusions MH37972. We have identi®ed a species ¯ock of mormyrid ®shes in west-central Africa. This is the ®rst freshwater ®sh species LITERATURE CITED ¯ock described within a group of weakly electric ®shes and Albertson, R. C., J. A. Markert, P. D. Danley, and T. D. Kocher. the ®rst wholly within a riverine, as opposed to a lacustrine, 1999. Phylogeny of a rapidly evolving clade: the cichlid ®shes environment. This species ¯ock may have arisen within the of Lake Malawi, East Africa. Proc. Natl. Acad. Sci. USA 96: 5107±5110. precursor of the modern Ogooue and Ntem River systems, Alves-Gomes, J. A. 1999. Systematic biology of gymnotiform and and much of its diversi®cation may have taken place during mormyriform electric ®shes: phylogenetic relationships, molec- the past 2 million years. The species-speci®city of EOD ular clocks and rates of evolution in the mitochondrial rRNA waveforms within this group and their demonstrated use in genes. J. Exp. Biol. 202:1167±1183. mate recognition suggest the possibility that selection on Alves-Gomes, J., and C. D. Hopkins. 1997. Molecular insights into the phylogeny of mormyriform ®shes and the evolution of their EODs may play an important role in the origin and/or main- electric organs. Brain Behav. Evol. 49:324±351. tenance of species boundaries. Arnold, M. L. 1997. Natural hybridization and evolution. Oxford Study of species ¯ocks holds out the promise of revealing Univ. Press, Oxford, U.K. details of the interrelated processes of speciation and signal Arratia, G. 1997. Basal teleosts and teleostean phylogeny. F. Pfeil, evolution. However, progress in estimating their internal phy- Munich. Bass, A. H. 1986a. Electric organs revisited: evolution of a ver- logenies may depend on the development of methodologies tebrate communication and orientation organ. Pp. 13±70 in T. with the potential to overcome the limitations encountered H. Bullock and W. Heiligenberg, eds. Electroreception. John with mitochondrial sequence data. Our analysis of cyto- Wiley and Sons, New York. chrome b sequences provides evidence for the monophyly of ÐÐÐ. 1986b. Species differences in electric organs of mormyrids: this ¯ock and for a recent origin of much of its diversity, but substrates for species-typical electric organ discharge wave- forms. J. Comp. Neurol. 244:313±330. fails to resolve much of its internal phylogenic structure. We Bass, A. H., and C. D. Hopkins. 1983. Hormonal control of sexual attribute this to the effects of incomplete mitochondrial lin- differentiation changes in electric organ discharge waveform. eage sorting and to introgression. These phenomena have Science 220:971±974. been similarly problematic in the phylogenetic analysis of Bass, A. H., J.-P. Denizot, and M. A. Marchaterre. 1986. Ultra- other species ¯ocks with mitochondrial datasets. Unfortu- structural features and hormone-dependent sex differences of mormyrid electric organs. J. Comp. Neurol. 254:511±528. nately, alternatives are limited in such cases. Apart from the Bennett, M. V. L. 1970. Comparative physiology: electric organs. dif®culty of ®nding a nuclear marker of appropriate evolu- Annu. Rev. Physiol. 32:471±528. tionary rate, the four-times-larger effective population size ÐÐÐ. 1971. Electric organs. Pp. 347±491 in W. S. Hoar and D. of nuclear markers relative to mitochondrial predicts their J. Randall, eds. Fish physiology. Academic Press, New York. greater susceptibility to lineage sorting problems (Pamilo and Bennett, M. V. L., and H. Grundfest. 1961. Studies on the mor- phology and electrophysiology of electric organs. III. Electro- Nei 1988; Palumbi and Cipriano 1998). Phylogenetic analysis physiology of the electric organs in mormyrids. Pp. 113±135 in of species ¯ocks may prove to be more successful with da- C. Chagas and A. P. d. Carvalho, eds. Bioelectrogenesis. El- tasets consisting of many unlinked molecular characters, such sevier, New York. as SINE insertions (Shedlock and Okada 2000) or AFLP Berg, L. S. 1965. Freshwater ®shes of the USSR and adjacent coun- fragments (Albertson et al. 1999). tries. Israel Program for Scienti®c Translation, Jerusalem. Boulenger, G. A. 1906. Description of new mormyrid ®sh from south Cameroon. Ann. Mag. Nat. Hist. 18:36±37. ACKNOWLEDGMENTS Brooks, J. L. 1950. Speciation in ancient lakes. Q. Rev. Biol. 25: For help in Gabon, we thank J. D. Mbega and J. H. Mve 30±176. Chow, S. 1998. Universal PCR primer for calmodulin gene intron of IRAF, P. Posso and B. Bouroubou of IRET, O. Langrand in ®sh. Fish. Sci. 64:999±1000. and A. Kamdem Toham of WWF-CARPO, the Sisters of the Chow, S., and K. Hazama. 1998. Universal PCR primers for S7 Immaculate Conception in LambareÂneÂ, the Christian and Mis- ribosomal protein gene introns in ®sh. Mol. Ecol. 7:1247±1263. SPECIES FLOCK OF AFRICAN ELECTRIC FISHES 615

Cracraft, J. 1983. Species concepts and speciation analysis. Curr. 1993. Pathways of lysozyme evolution inferred from the se- Ornithol. 1:159±187. quences of cytochrome b in birds. J. Mol. Evol. 37:367±379. Daget, J., J.-P. Gosse, and D. F. E. Thys van den Audenaerde. 1984. Korn®eld, I., and K. E. Carpenter. 1984. Cyprinids of Lake Lanao, Check-list of the freshwater ®shes of Africa. ORSTOM/MRAC, Philippines: taxonomic validity, evolutionary rates, and speci- Paris. ation scenarios. Pp. 69±84 in A. A. Echelle and I. Korn®eld, Dowling, T. E., and C. L. Secor. 1997. The role of hybridization eds. Evolution of ®sh species ¯ocks. Univ. of Maine Press, and introgression in the diversi®cation of animals. Annu. Rev. Orono, ME. Ecol. Syst. 28:593±619. Korn®eld, I., and A. Parker. 1997. Molecular systematics of a rap- Echelle, A. A., and I. Korn®eld. 1984. Evolution of ®sh species idly evolving species ¯ock: the mbuna of Lake Malawi and the ¯ocks. Univ. of Maine Press, Orono, ME. search for phylogenetic signal. Pp. 25±37 in T. D. Kocher and Felsenstein, J. 1985. Con®dence limits on phylogenies: an approach C. A. Stepien, eds. Molecular systematics of ®shes. Academic using the bootstrap. Evolution 39:783±791. Press, New York. Forey, P. L. 1997. A Cretaceous notopterid (Pisces: Osteoglosso- Korn®eld, I., and P. F. Smith. 2000. African cichlid ®shes: model morpha) from Morocco. S. Afr. J. Sci. 93:564±569. systems for evolutionary biology. Annu. Rev. Ecol. Syst. 31: Fryer, G., and T. D. Iles. 1972. The cichlid ®shes of the Great Lakes 163±196. of Africa. Oliver and Boyd, Edinburgh. Kramer, B. 1997. Electric organ discharges and their relation to sex Goldschmidt, T. 1996. Darwin's dreampond: drama in Lake Vic- in mormyrid ®shes. Naturwissenschaften 84:119±121. toria. MIT Press, Boston. Lauder, G. V., and K. Liem. 1983. The evolution and interrela- Grant, B. R., and P. R. Grant. 1996. Cultural inheritence of song tionships of the actinopterygian ®shes. Bull. Mus. Comp. Zool. and its role in the evolution of Darwin's ®nches. Evolution 50: 150:95±197. 2471±2487. LavoueÂ, S., R. Bigorne, G. Lecointre, and J.-F. AgneÁse. 2000. Phy- Greenwood, P. H. 1973. Interrelationships of osteoglossomorphs. logenetic relationships of mormyrid electric ®shes (Mormyridae: Pp. 307±332 in P. H. Greenwood, R. S. Miles, and C. Patterson, Teleostei) inferred from cytochrome b sequences. Mol. Phylo- eds. Interrelationships of ®shes. Academic Press, London. genet. Evol. 14:1±10. ÐÐÐ. 1984. What is a species ¯ock? Pp. 13±20 in A. A. Echelle Li, G.-Q. 1997. Notes on the historical biogeography of the osteo- and I. Korn®eld, eds. Evolution of ®sh species ¯ocks. Univ. of glossomorpha (Teleostei). Pp. 54±66 in Proceedings of the 30th Maine Press, Orono, ME. international geological congress. VSP International Science GuÈnther, A. 1896. Report on a collection of reptiles and ®shes made Publishers, Zeist, The Netherlands. by Miss M. H. Kingsley during her travels on the Ogowe River Li, G.-Q., and M. V. H. Wilson. 1996. Phylogeny of the Osteo- and in Old Calabar. Ann. Mag. Nat. Hist. 17:261±285. glossomorpha. Pp. 163±174 in M. L. J. Stiassny, L. R. Parenti, Hamilton, A. C. 1982. Environmental history of East Africa: a study and G. D. Johnson, eds. Interrelationships of ®shes. Academic of the Quaternary. Academic Press, London. Press, New York. Harrison, R. G. 1989. Animal mitochondrial DNA as a genetic Lissmann, H. W. 1951. Continuous electric signals from the tail of marker in population and evolutionary biology. Trends Ecol. a ®sh, Gymnarchus niloticus Cuv. Nature 167:201±202. Evol. 4:6±11. Lowe, R. H. 1953. Notes on the ecology and evolution of Nyasa ÐÐÐ. 1991. Molecular changes at speciation. Annu. Rev. Ecol. ®shes of the genus Tilapia with a description of T. saka Lowe. Syst. 22:281±308. Proc. Zool. Soc. Lond. 122:1035±1041. Hassanin, A., G. Lecointre, and S. Tillier. 1998. The `evolutionary Lydeard, C., and K. J. Roe. 1997. The phylogenetic utility of the signal' of homoplasy in protein-coding gene sequences and its mitochondrial cytochrome b gene for inferring relationships consequence for a priori weighting in phylogeny. C. R. Acad. among actinopterygian ®shes. Pp. 285±303 in T. D. Kocher and Sci. Ser. III 321:611±620. C. A. Stepien, eds. Molecular systematics of ®shes. Academic Hopkins, C. D. 1980. Evolution of electric communication channels Press, New York. of mormyrids. Behav. Ecol. Sociobiol. 7:1±13. Maley, J. 1987. Fragmentation de la foreÃt dense humide africaine ÐÐÐ. 1981. On the diversity of electric signals in a community et extension des biotopes montagnards au Quaternaire reÂcent: of mormyrid electric ®sh in West Africa. Am. Zool. 21:211±222. nouvelles donneÂes polliniques et chronologiques. Implications ÐÐÐ. 1986a. Behavior of Mormyridae. Pp. 527±576 in T. H. paleÂoclimatiques et biogeÂographiques. Paleoecol. Afr. 18: Bullock and W. Heiligenberg, eds. Electroreception. John Wiley 307±334. and Sons, New York. ÐÐÐ. 1991. The African rainforest vegetation and paleoenviron- ÐÐÐ. 1986b. Temporal structure of non-propagated electric com- ments during late Quaternary. Clim. Change 19:79±98. munication signals. Brain Behav. Evol. 28:43±59. ÐÐÐ. 1996. The African rainforest: main characteristics of chang- Hopkins, C. D., and A. H. Bass. 1981. Temporal coding of species es in vegetation and climate from the Upper Cretaceous to the recognition signals in an electric ®sh. Science 212:85±87. Quaternary. Proc. R. Soc. Edinb. 104B:31±73. Hoy, R., A. Hoikkala, and K. Kaneshiro. 1988. Hawaiian courtship McCune, A. R. 1996. Biogeographic and stratigraphic evidence for songs: evolutionary innovation in communication signals of rapid speciation in semionotid ®shes. Paleobiology 22:34±48. Drosophila. Science 240:217±219. ÐÐÐ. 1997. How fast is speciation? Molecular, geological, and Humphries, J. M. 1984. Genetics and speciation in pup®shes from phylogenetic evidence from adaptive radiations of ®shes. Pp. Laguna Chichancanab, Mexico. Pp. 129±140 in A. A. Echelle 585±609 in T. J. Givnish and K. J. Sytsma, eds. Molecular evo- and I. Korn®eld, eds. Evolution of ®sh species ¯ocks. Univ. of lution and adaptive radiation. Cambridge Univ. Press, Cam- Maine Press, Orono, ME. bridge, U.K. Irwin, D. M., T. D. Kocher, and A. C. Wilson. 1991. Evolution of McCune, A. R., K. S. Thomson, and P. E. Olsen. 1984. Semionotid the cytochrome b gene of mammals. J. Mol. Evol. 32:128±144. ®shes from the Mesozoic great lakes of North America. Pp. 585± Johns, G. C., and J. C. Avise. 1998. Tests for ancient species ¯ocks 610 in A. A. Echelle and I. Korn®eld, eds. Evolution of ®sh based on molecular phylogenetic appraisals of Sebastes rock- species ¯ocks. Univ. of Maine Press, Orono, ME. ®shes and other marine ®shes. Evolution 52:1135±1146. McMillan, W. O., and S. R. Palumbi. 1995. Concordant evolution- Kamdem Toham, A. 1998. Fish biodiversity of the Ntem River ary patterns among Indo-West Paci®c butter¯y®shes. Proc. R. Basin (Cameroon): taxonomy, ecology and conservation. Ph.D. Soc. Lond. B 260:229±239. diss. Laboratory of Ecology and Aquaculture, Dept. of Biology, Meyer, A. 1993. Evolution of mitochondrial DNA in ®shes. Pp. 1± Katholieke Universiteit Leuven, Leuven, Belgium. 38 in P. W. Hochachka and T. P. Mommsen, eds. Biochemistry Knight, M. E., and G. F. Turner. 1999. Reproductive isolation and molecular biology of ®shes. Elsevier, Amsterdam. among closely related Lake Malawi cichlids: Can males rec- Meyer, A., T. D. Kocher, P. Basasibwaki, and A. C. Wilson. 1990. ognize conspeci®c females by visual cues? Anim. Behav. 58: Monophyletic origin of Lake Victoria cichlid ®shes suggested 761±768. by mitochondrial DNA sequences. Nature 347:550±553. Kornegay, J. R., T. D. Kocher, L. A. Williams, and A. C. Wilson. Moran, P., and I. Korn®eld. 1993. Retention of an ancestral poly- 616 JOHN P. SULLIVAN ET AL.

morphism in mbuna species ¯ock (Pisces: Cichlidae) of Lake blood and tissue samples for DNA analyses. Can. J. Zool. 69: Malawi. Mol. Biol. Evol. 10:1015±1029. 82±90. Nagelkerke, L. A. J., F. A. Sibbing, J. G. M. Van-Den-Boogaart, Shaw, K. L. 1999. A nested analysis of song groups and species E. H. Lammens, and J. W. M. Osse. 1994. The barbs (Barbus boundaries in the Hawaiian cricket genus Laupala. Mol. Phy- spp.) of Lake Tana: a forgotten species ¯ock? Environ. Biol. logenet. Evol. 11:332±341. Fishes 39:1±22. Shedlock, A. M., and N. Okada. 2000. SINE insertions: powerful Nelson, J. S. 1994. Fishes of the world. John Wiley and Sons, New tools for molecular systematics. BioEssays 22:148±160. York. Smith, G. 1992. Introgression in ®shes: signi®cance for paleontol- Olivry, J. C. 1986. Fleuves et RivieÁres du Cameroun. MESRES- ogy, cladistics, and evolutionary rates. Syst. Biol. 41:41±57. ORSTOM, Paris. Strecker, U., C. G. Meyer, C. Sturmbauer, and H. Wilkens. 1996. Otte, D. 1994. The crickets of Hawaii: origin, systematics, and Genetic divergence and speciation in an extremely young species evolution. The Orthopterist's Society, Philadelphia, PA. ¯ock in Mexico formed by the genus Cyprinodon (Cyprinodon- Palumbi, S. R. 1996. Nucleic acids. II. The polymerase chain re- tidae, Teleostei). Mol. Phylogenet. Evol. 6:143±149. action. Pp. 205±248 in D. M. Hillis, C. Moritz, and B. K. Mable, Sullivan, J. P., S. LavoueÂ, and C. D. Hopkins. 2000. Molecular eds. Molecular systematics. Sinauer Associates, Sunderland, systematics of the African electric ®shes (Mormyroidea: Te- leostei) and a model for the evolution of their electric organs. MA. J. Exp. Biol. 203:665±683. Palumbi, S. R., and F. Cipriano. 1998. Species identi®cation using Taliev, D. N. 1955. Sculpins of Baikal (Cottidae). Academy of genetic tools: the value of nuclear and mitochondrial gene se- Science USSR, Moscow. quences in whale conservation. Heredity 89:459±464. Taverne, L. 1971. Note sur la systeÂmatique des poissons Mormy- Pamilo, P., and M. Nei. 1988. Relationships between gene trees riformes. Le probleÁme des genres Gnathonemus, Marcusenius, and species trees. Mol. Biol. Evol. 5:568±583. Hippopotamyrus, Cyphomyrus, et les nouveaux genres Pollimy- Parenti, L. R. 1984. Biogeography of the Andean killi®sh genus rus et Brienomyrus. Rev. Zool. Bot. Afr. 84:99±110. Orestias with comments on the species ¯ock concept. Pp. 85± ÐÐÐ. 2000. Nouvelles donneÂes osteÂologiques et phylogeÂneÂtiques 92 in A. A. Echelle and I. Korn®eld, eds. Evolution of ®sh sur Paleonotopterus greenwoodi, notopteÂride (Teleostei, Osteo- species ¯ocks. Univ. of Maine Press, Orono, ME. glossomorpha) du CeÂnomanien infeÂrieur continental (CreÂtaceÂ) Parker, A., and I. Korn®eld. 1995. A molecular perspective on du Maroc. Stuttg. Beitr. Naturkde. Ser. B 293:24. evaluation and zoogeography of cyprinodontid killi®shes. Cop- Taverne, L., and J. G. Maisey. 1999. A notopterid skull (Teleostei, eia 1995:8±21. Osteoglossomorpha) from the continental Early Cretaceous of ÐÐÐ. 1997. Evolution of the mitochondrial DNA control region Southern Morocco. Am. Mus. Novit. 3260:1±12. of the mbuna (Cichlidae) species ¯ock of Lake Malawi, East Taverne, L. P., and D. F. E. Thys van den Audenaerde. 1985. Brien- Africa. J. Mol. Evol. 45:70±83. omyrus hopkinsi new species from north Gabon (Pisces: Te- Patterson, C., and D. E. Rosen. 1977. Review of the ichthyodec- leostei: Mormoridae). Rev. Zool. Afr. 99:49±52. tiform and other mesozoic teleost ®shes and the theory and prac- Taverne, L., D. F. E. Thys van den Audenaerde, and A. Heymer. tice of classifying fossils. Bull. Am. Mus. Nat. Hist. 158:81±172. 1977a. Paramormyrops gabonensis nov. gen., nov. sp. du nord Ribbink, A. J. 1984. Is the species ¯ock concept tenable? Pp. 21± du Gabon. Rev. Zool. Afr. 91:634±640. 25 in A. A. Echelle and I. Korn®eld, eds. Evolution of ®sh Taverne, L., D. F. E. Thys van den Audenaerde, A. Heymer, and species ¯ocks. Univ. of Maine Press, Orono, ME. J. Gery. 1977b. Brienomyrus longicaudatus new species and Ritchie, P. A., L. Bargelloni, A. Meyer, J. A. Taylor, J. A. Mac- Brienomyrus curvifrons new species from northern Gabon (Pi- Donald, and D. M. Lambert. 1996. Mitochondrial phylogeny of sces: Mormyridae). Rev. Zool. Afr. 91:200±208. trematomid ®shes (Nototheniidae, Perciformes) and the evolu- Teugels, G. G., and C. D. Hopkins. 1998. Morphological and os- teological evidence for the generic position of Mormyrus king- tion of Antarctic ®sh. Mol. Phylogenet. Evol. 5:383±390. sleyae in the genus Brienomyrus (Teleostei: Mormyridae). Cop- Robinson, M., M. Gouy, C. Gautier, and D. Mouchiroud. 1998. eia 1998:199±204. Sensitivity of the relative rate test to taxonomic sampling. Mol. Thys van den Audenaerde, D. F. E. 1966. Les Tilapia (Pisces, Cich- Biol. Evol. 15:1091±1098. lidae) de Sud-Cameroun et du Gabon eÂtude systematique. Mus. Rogers, J. S., and Swofford, D. L. 1998. A fast method for ap- R. Afr. Cent. Tervuren Belg. Ann. Ser. Octavo Sci. Zool. 153: proximating maximum likelihoods of phylogenetic trees from 123. nucleotide sequences. Syst. Biol. 47:77±89. van Oppen, M. J. H., G. F. Turner, C. Rico, R. L. Robinson, J. C. Sato, A., C. O'Huigin, F. Figueroa, P. Grant, B. R. Grant, H. Tichy, Deutsch, and G. M. Hewitt. 1998. Assortative mating among and J. Klein. 1999. Phylogeny of Darwin's ®nches as revealed rock-dwelling cichlid ®shes supports high estimates of species by mtDNA sequences. Proc. Natl. Acad. Sci USA 96:5101±5106. richness from Lake Malawi. Mol. Ecol. 7:991±1001. Sauvage, H. E. 1880. E tude sur la faune ichthyologique de Wildekamp, R. H. 1993. A world of killies. American Killi®sh l'OgoÃoueÂ. Arch. Mus. Natl. Hist. Nat. 2:5±56. Association, Mishawaka, IN. Schlichter, H. 1906. UÈ ber den feineren Bau des schwach-elektrisch- Willet, C. E., J. J. Cherry, and L. A. Steiner. 1997. Characterization en Organs von Mormyrus oxyrhynchus Geoff. Z. Wiss. Zool. 84: and expression of the recombination activating genes (rag1 and 479±525. rag2) of zebra®sh. Immunogenetics 45:394±404. Schliewen, U. K., D. Tautz, and S. PaÈaÈbo. 1994. Sympatric spe- Winemiller, K. O., and A. Adite. 1997. Convergent evolution of ciation suggested by monophyly of crater lake cichlids. Nature weakly electric ®shes from ¯oodplain habitats in Africa and 368:629±632. South America. Environ. Biol. Fishes 49:175±186. Seehausen, O., J. J. M. van Alpen, and F. Witte. 1997. Cichlid ®sh Yang, Z. 1994. Maximum likelihood phylogenetic estimation from DNA sequences with variable rates over sites: approximate diversity threatened by eutrophication that curbs sexual selec- methods. J. Mol. Evol. 39:306±314. tion. Science 277:1808±1811. Seutin, G., B. N. White, and P. T. Boag. 1991. Preservation of avian Corresponding Editor: L. Bernatchez