Genetic relationships and variation in meristic counts, body shape and colour were examined in a large sample of Symphysodon collected from several locations in floodplain habitats along the length of the Amazon River. Surprisingly, mitochondrial DNA indicates no difference between the two historically described species, Symphysodon discus and Symphysodon aequifasciatus, but shows that non-clinal variation exists with a distinct lineage found in the western Amazon. This lineage is consistent with a colour form that is distinct from other Symphysodon lineages. This form has a parapatric distribution and is recognized as a distinct species, Symphysodon tarzoo. Adaptation to floodwater habitats supports genetic cohesion across a large range preventing fine scale regional diversification of the genus. Possible explanations for the unusual set of distributions for genetic and colour characters relate to the history of the Amazon basin and the probable division of lowland species when submerged geologic arches influence surface topology.

The discus fishes, genus Symphysodon Heckel, 1840, are one of the most intriguing and distinctive groups among the South American Cichlidae. At present, there are two recognized species within the genus, Symphysodon discus Heckel, 1840, and Symphysodon aequifasciatus, Pellegrin, 1904, which differ in the pattern of melanic bars and in some meristic characters. The first species is found in the central Amazon in the Rio Negro, Abacaxis and Trombetas drainages, while the latter is found along almost the entire length of the Amazon from 49° to 70° west longitude (Fig. 1). Species of Symphysodon are restricted to areas where seasonal flooding occurs and are therefore found only near the mainstem of the Amazon River itself and in the lower reaches of tributaries on the Amazonian floodplain (Kullander, 1996). The combined distribution of the genus forms one large continuous range in which many environmental and historical factors could have been involved in isolating populations and driving speciation.

Symphysodon are part of the Heroini, a Neotropical clade of cichlid fishes (Kullander, 1998; Farias et al., 2001). The exact relationship of Symphysodon within the Heroini is partially contentious, though morphological analysis apparently provides stronger support than molecular evidence, placing Symphysodon as sister taxon to a clade containing Heros, Uaru, Mesonauta and Pterophyllum (Kullander, 1998). These are all high-bodied, lowland fishes known mainly from the Amazon basin.

Colour patterns on living fish vary considerably in Symphysodon. There are normally nine dark bars on the sides. These are of equal width and intensity in S. aequifasciatus, but bars 1, 5 and 9 are intensified and wider in S. discus. Although the general colour of S. aequifasciatus is brown with some blue and red markings on the forehead and near the anal fin, some individuals in all populations have brighter colours, often including greater coverage or strength of bright red and blue stripes, spots or reticulations (pers. obs.). This colour variation enhances the value of these species in the ornamental pet trade where they are highly regarded and priced.

Studies of colour variation have resulted in the descriptions of several sub-species based on the purported prevalence of variants in different geographic regions and include Symphysodon a. aequifasciatus, Symphysodon a. haraldi Schultz, 1960, Symphysodon a. axelrodi Schultz, 1960, and S. discus willischwartzi Burgess, 1981. Unfortunately, descriptions are based on only a few samples, often of uncertain origin (Schultz, 1960; Burgess, 1981), and Kullander (1986; 1996; 2003) recognized only two diagnosable species, S. aequifasciatus and S. discus.

The present study includes a new larger set of specimens, with both colour photos and tissue samples for DNA analysis, to determine whether variation in colour, morphology or genetics is correlated with geography or environmental variation. Despite the confusion over the precise origins of colour forms obtained from ornamental fish importers, some populations are relatively easy to identify from preliminary inspection of photographs of recently collected material. For example, fish from Tefé, in the Western Amazon, can be recognised by very light coloured anterior sides with red spots on the anal fin and sides of the body.

Symphysodon aequifasciatus is one of very few cichlid species with an almost linear distribution along the entire lowland course of the Amazon River mainstream.

The Amazon River has a dramatic history, being diverted from a Pacific outlet to a northern outlet about 67 million years ago (MYA) and finally to an Atlantic outlet about 8 MYA (Lundberg et al., 1998). Symphysodon aequifasciatus was selected to test for clinal variation in morphology and genetic characteristics along the Amazon River, as a possible explanation of phenotypic variation, and at the same time for possible breaks in continuity that could be correlated with hypotheses about the historical development of the Amazon River.

Samples were collected at several localities (Fig. 1) in 1998. Specimens were deposited in the collections of INPA, Manaus, Brazil (INPA 14334–14345, INPA 14386 and INPA 25960). Samples from further east in the overall distribution could not be obtained as the area is less frequently visited by discus fishermen whose assistance was required to obtain samples. Additional material examined included the type series of S. aequifasciatus from the Muséum National d’Histoire Naturelle, Paris, France (MNHN 1902–130, MNHN 1902–134 and MNHN 1902–135). Analysis of variation includes molecular data (from 23 individuals), characters of colour pattern and body shape (assessed from 263 photographs of live fish) and morphological characteristics (meristics and rhomboid measures for 94 individuals).

Tissue samples were stored in ethanol before preservation of whole fish in formalin (except for Rio Bauana specimens). Genomic DNA was extracted using DNeasy extraction protocols and Kits (QIAGEN, through VWR International AB, Stockholm, Sweden). The sequences were then amplified using puReTaq Ready-To-Go PCR beads (Amersham Biosciences Europe, Uppsala, Sweden) and polymerase chain reaction cycles as follows: 1 denaturation cycle at 94° C for 4 min, 30 cycles with a denaturation temperature of 94° C for 30 s, an annealing temperature of 54° C (for all cytochrome b primers) or 62° C (for rhodopsin primers), and extension at 72° C for 30 s, all followed by 1 extension cycle at 72° C for 7 min. For cytochrome b, preliminary amplification used the primers FISCYTB-F (5' ACC ACC GTT GTT ATT CAA CTA CAA GAA C 3') and TRUCCYTB-R (5' CCG ACT TCC GGA TTA CAA GAC CG 3'), and a secondary nested amplification used these with the internal primers CYTBi3-R (5' GGG GTA AAG TTG TCT GGG TCT CC 3') and CYTB1-F (5' CGA TTC TTC GCA TTC CAC TTC CT 3'), respectively, to obtain sequences of two overlapping fragments. Amplification of rhodopsin used the primers RodF2w (5' AGC AAC TTC CGC TTC GGT GAG AA 3') and RodR4n (5' GGA ACT GCT TGT TCA TGC AGA TGT AGA T 3'). All primers are from the European Union project FishTrace (M. Norén, pers. comm.). Reactions included negative controls. Sequencing was performed using the BigDye Terminator cycle sequencing ready reaction kit (Applied Biosystems Inc., through Applera Sweden, Stockholm, Sweden) on an automated DNA sequencer (Applied Biosystems 377) following manufacturer’s instructions. Nucleotide sequences were deposited in GenBank under accession numbers DQ533587–DQ533606.

Sequences were aligned by eye using BioEdit v. 5·0·9 software (Hall, 1999). Cytochrome b gene sequences for several genera, which are potentially close relatives of Symphysodon (Kullander, 1998), were obtained from GenBank to root the phylogenies produced. These included Acaronia sp. Myers, 1940 (AF370666) and all available sequences for the genera of the Heroini: Pterophyllum scalare (Lichtenstein, 1823) (AF370676), Hoplarchus psittacus Kaup, 1860 (AF370673), Hypselecara coryphaenoides (Heckel, 1840) and Hypselecara temporalis (Günther, 1862) (AF370674 and AY050612, respectively), Caquetaia kraussi (Steindachner, 1878) and Caquetaia spectabilis (Steindachner, 1875) (AF009938 and AF370671, respectively), Theraps maculicauda (Regan, 1905)—now Vieja maculicauda (Regan, 1905) (U97165), Mesonauta insignis (Heckel, 1840) (AF370675), Uaru amphiacanthoides Heckel, 1840 (AF370678 and AY050622) and Heros sp. Heckel, 1840 and H. appendiculatus (Castelnau, 1855)—now Heros efasciatus Heckel, 1840 (AF370672 and AF009951). Also included was the cytochrome b sequence submitted for Symphysodon aequifasciatus (AF370677).

MODELTEST software (Posada & Crandall, 1998) was used to determine the best model of molecular evolution for the cytochrome b sequences. The general time reversible model with proportion of invariant sites = 0·5491 and gamma shape of 1·4368 (GTR + I + G) was chosen under both heirarchical likelihood test criteria and Akaike Information Criteria (AIC) and used for subsequent analysis. Phylogenetic analysis using parsimony (PAUP*) (Swofford, 1998) was used to construct trees using the sequence of Acaronia as an outgroup. Sequences were then analysed with 1000 bootstrap replicates under parsimony (options: ACCTRAN, TBR, MULTREES, Gaps = missing, random addition, transition to transversion (ti/tv) weighting as estimated in MODELTEST), 100 heuristic bootstrap replicates using the maximum likelihood setting described under MODELTEST and 1000 bootstrap replicates for neighbour joining (weighted least squares and parameters described by MODELTEST). Bayesian support was obtained using MrModeltest software (Nylander, 2002) to determine the model (also GTR + I + G) and MrBayes (Huelsenbeck & Ronquist, 2001). The support was obtained using 2 000 000 MCMC iterations, sampling every 100 iterations with a burn in where the first 10% of trees sampled were discarded (likelihood values were stable after this point). A consensus of the remaining trees was then obtained in PAUP* (Swofford, 1998).

Six colour pattern characters and two body shape characters were classed for each photograph using the characters and character states outlined in Table I. The eight characters were each classed into at most five states, with the first state being the most common in the outgroup (Table I).

Morphological distance measurements were made using digital callipers accurate to 0·01 mm. Methods for taking meristic counts follow Kullander (1983; 1986). Measurements and counts were made on the left side of the specimen when possible. Specimen lengths are standard length (SL). Vertebrae, pterygiophore, fin spine and soft ray counts and the measurement of SL were taken from radiographs made on Kodak X-Omat V plates, with a Philips MGC 30 low voltage X-ray unit. Distances between landmark points on specimens were determined from radiographs using a double rhomboid system (Fig. 2).

Statistical analyses and graphs were made using the SYSTAT package (Wilkinson et al., 1992). Log transformations of body distances relative to SL, square root transformed meristic counts and raw data for classed characters were used in correlation and principal component analyses (PCA).

The aligned mitochondrial cytochome b gene sequence was 1134 bp long, and the nuclear rhodopsin partial sequence obtained was 514 bp long. Rhodopsin sequences showed no variation within Symphysodon. A blast search of sequences in GenBank indicated that Symphysodon sequences were most similar to rhodopsin sequences of African cichlids. No Neotropical cichlid rhodopsin sequences were found in GenBank. Unambiguous alignment of the deduced amino acid sequence in Symphysodon with sequences from GenBank confirmed the single open reading frame.

Analysis of mitochondrial DNA (mt DNA) (cytochrome b gene) indicates distinct lineages of Symphysodon that do not conform to previous taxonomic classifications (Fig. 3).

One lineage comprises western populations that under the existing classification would have been identified as S. aequifasciatus. However, a second large lineage included not only sequences from central and eastern populations conforming to S. aequifasciatus but also sequences of S. discus from Manaus and the Rio Abacaxis. There is one notable exception. An individual from the Madeira River, which is central in the geographical distribution of samples, possesses both the phenotype and the mtDNA haplotype typical of western populations. Additionally, a far eastern genetic lineage may also exist from which only two samples were collected, Boím samples 2 and 3. However, their divergence from the central and the eastern lineage of Symphysodon aequifasciatus is small.

When compared with those of published Maximum Likelihood-based trees of mitochondrial sequences including African rift lake cichlids (Farias et al., 2001; Klett & Meyer, 2002) the branch lengths between Symphysodon species (0·02 substitutions/site, Fig. 3) are comparable in some cases to branch lengths between genera of rift lake cichlids.

The resolution of the position of Symphysodon within the Heroini is weak. This is not surprising as many of the sequences used are those obtained from previous investigations (Farias et al., 2001).

Principal Component Analysis (PCA) of colour pattern and body shape places specimens from the same population in the same cluster, but with many populations the overall pattern becomes ambiguous. Groups of specimens defined by environmental differences between sites, or by sex, did not lead to any distinct clusters in the PCA. However, when specimens were classified by region (western v. central/eastern groups), two weakly overlapping groups appeared in the PCA (Fig. 4). A re-examination of photographs of four individuals from the central/eastern group that lie within or close to the western group show that these individual fish possess the red spots typical for fish in the western group, even though these four fish were collected from a central population dominated by fish lacking red spots. The main western and central/eastern groupings in Fig. 4 were largely defined by Factor 1 that had high loadings for the presence or the absence of red spots and for anal fin red colouration. Even under very close inspection, darker bodied fish from central and eastern populations always lack red-pigmented spots on the anal fin and body and remains distinct from western fish. Two sub-groups appear to be present in the central/eastern cluster, though the current data do not support the recognition of a far eastern Boím genetic lineage.

PCA of morphology (meristic counts and rhomboid measurements) on all populations indicated no clear population groupings. Within-population variation is greater than between-population variation.

The analysis of mitochondrial DNA haplotypes, colour pattern and morphology indicates that the western Amazonian Symphysodon are a distinct species. This is supported by the distinct mitochondrial lineage and by the diagnostic phenotypic character of red spots on the sides of the body.

Schultz (1960) described several sub-species of S. aequifasciatus based on the colour slides of individuals obtained through the ornamental fish industry. Among those, S. a. haraldi, is described from Benjamin Constant in the Western Amazon. However, this locality is unlikely to be correct as the specimen illustrated and described has a colour pattern conforming to eastern populations.

The type series of S. aequifasciatus includes specimens from Tefé (MNHN 1902–134 and 1902–135) and Santarém (MNHN 1902–130) that would represent the distribution of the western and eastern species, respectively. No attempts were made to extract DNA from these specimens, and the live colour patterns of preserved specimens could not be determined. Therefore, MNHN 1902–130 from Santarém is selected as lectotype of S. aequifasciatus, thus restricting the name to the central-eastern species. The specimen is 107 mm SL and has the following meristic data: Dorsal fin, IX.31; anal fin, XIII.30; vertebrae, 13 abdominal, 18 caudal; scales, 56 on row E1, 20 in anterior lateral line, 14 in posterior lateral line. Symphysodon aequifasciatus is distinguished from S. discus by having a melanic colour pattern of lateral bars of equal intensity and width. Symphysodon aequifasciatus does not possess distinct red spots on the anal fin and body, although it may have red reticulated pigmentation primarily on the anal fin. Symphysodon haraldi and S. axelrodi are synonyms of this species.

The western species is S. tarzoo Lyons, 1959, that was described as having distinct red spots on its anal fin and body. The description was based on live aquarium specimens from Leticia, Colombia, and no type specimens were apparently preserved (Kullander, 1996; 2003). The availability of the name tarzoo was rejected by Schultz (1960) based on the supposed absence of a diagnosis in Lyons (1959). However, several statements in the latter study clearly describe the new species. Kullander (1996) discusses the availability of names for Symphysodon species and recognized that the epithet tarzoo was available under the International Code of Zoological Nomeclature. Because of the nomenclatural and taxonomic problems in Symphysodon, a neotype for S. tarzoo was selected. INPA 25960 from Brazil, Est. Amazonas, Rio Jutaí. 1998, S.O. Kullander and E.F.J.G. Ferreira. It is an adult male of 132·4 mm SL and has the following meristic data: Dorsal fin, X.30; anal fin, XIII.31; vertebrae, 14 abdominal, 17 caudal; scales, 58 on row E1, 20 in anterior lateral line, 14 in posterior lateral line. The red spots on the anal fin and in the body distinguishes S. tarzoo from all other Symphysodon species that have reticulations.

Symphysodon aequifasciatus and S. tarzoo overlap in distribution slightly as shown by individuals from the Madeira River that are phenotypically and genetically S. tarzoo.

Traditional morphometric characters do not easily distinguish species, although average lateral line scale counts and melanic pigmentation still distinguish S. discus and S. aequifasciatus (Kullander, 1996). However, the mtDNA sequence shared between S. discus and S. aequifasciatus from the eastern and the central part of the distribution of the genus must be explained.

This appears to be the only case of a Neotropical cichlid genus with a distribution that, although marked and extensive, is limited to lowland Amazonian habitats and that includes a geographic boundary thought to restrict an endemic species to the western lowland region.

The Amazon historically flowed in the opposite direction from east to west. Evidence documenting this change is summarized by Lundberg et al. (1998). The current distribution of Symphysodon probably developed due to the changes in drainage pattern caused by tectonic processes. Species of Symphysodon are restricted to areas experiencing floods, and it is likely that their unusual body shape is an adaptation to this biotope. Under the proposed history of Lundberg et al. (1998), large-scale flood habitats [e.g. Lago Santa Lucía (60–42 MYA), Lago Pozo (43–30 MYA) and Lago Pebas (20–11·8 MYA)] occurred first in western regions of present day Amazonia, caused by the initial uplift of the Andes and the diversion of drainages to the north. Such habitats would have extended eastward when the northern drainage route was cut off and when the Amazon began to drain eastwards into the Atlantic as it does now. Given the sample locations, the ranges of S. aequifasciatus and S. tarzoo are roughly divided by the Purus arch (located between Manaus and Rio Bauana in Fig. 1) with only a few S. tarzoo specimens from the Madeira river downstream of the arch, and no S. aequifasciatus specimens from locations upstream of the arch. This arch was formed during one of the last phases of Andean uplift (C. 3 MYA) and was probably breached when the Amazon changed direction (Lundberg et al., 1998). Arguably, these subterranean arches may have affected organisms in the Amazonian floodplain (Lougheed et al., 1999), although the extent of its influence on surface topology at that time cannot be known. Two scenarios have been suggested: 1) the arch may have forced the overlying sediments to form a temporary barrier that isolated ancestral populations of S. tarzoo from other Symphysodon that subsequently have come into secondary contact, 2) alternatively, arch dynamics may have led to a narrowing of the floodplain that reduced gene flow enough to allow divergence between upstream and downstream forms. This latter scenario is embodied in a parapatric model of divergence between S. tarzoo and other Symphysodon species. The latter case is supported by geological analyses of the Amazon River today, which showed that where the river flows over arches, such as the Purus arch, water flow becomes slow and entrenched. The valley narrows to <20 km compared to an average breadth of nearly 45 km, the water-surface gradient decreases, sediment is deposited and water flow through the channel is negligible (Mertes et al., 1996).

The shared mtDNA sequences of S. discus and S. aequifasciatus and the lack of any variation among rhodpsin sequences supports at least two possible explanations for the observed phenotypic variation. 1) Branch lengths in the phylogram (Fig. 3) are consistent with an ancient divergence between upper (above Purus arch) and lower (below Purus arch) Amazonian fishes, and a divergence between the two species in the lower Amazon that is so recent that mtDNA differences between these species could not be found. 2) The genomes of all three species have diverged significantly, but the similarity of the mtDNA of S. aequifasciatus and S. discus results from introgressive hybridization. Under this scenario, the observed mtDNA sequence probably originally belonged to S. aequifasciatus because of the smaller number of sequences obtained for S. discus. However, it also possible that an mtDNA sweep may have occurred if there is any bias in the direction of crosses, e.g., due to Haldane’s rule. As such, the possibility of the sequences having orginated from S. discus should not be ruled out.

The distribution of Symphysodon is unusual in consisting of a very large transect across equatorial lowland South America. The habitat consists of major river channels, tributaries and the associated flooded forest, across which continuity should be sufficiently high to maintain gene flow between populations. This adaptation to a narrow ecological niche and the absence of competition from sympatric species may explain why morphology is so relatively constant over such a large range. The different colour forms that have evolved may indicate that mate choice plays a role in the evolution of Symphysodon.

The importance of colour vision and the light spectrum in water have been used to explain differentiation through, and subsequent breakdown of, the reproductive isolation between colour forms of African lake cichlids (Seehausen et al., 1997), and colour pattern has been shown to be important for species identity in South American cichlids (Ready et al., 2006). Discus fishes breed in flooded forests, therefore, the different light conditions produced by blackwater, whitewater and clearwater flooded forest habitats (Furch & Junk, 1997) could provide a mechanism for differentiation of populations. Symphysodon tarzoo and S. aequifasciatus distributions are generally consistent with the distribution of whitewater and clearwater tributaries of the Amazon, while the distribution of S. discus is generally consistent with the distribution of blackwater tributaries. Water colour is influenced by the geographical origins of the water, with whitewater originating from the Andes, clearwater from the south bank tributaries (Brazilian shield) and blackwater from the north bank tributaries (Guyanan shield and eastern Andes). The latter two sources are believed to have remained unchanged for many millions of years (Lundberg et al., 1998).

Therefore, the current distributions of the species may reflect ancient speciation if flood habitats were isolated prior to the change in direction of flow of the Amazon (possibly with hybridization after secondary contact), or may represent parapatric speciation within the modern Amazon drainage pattern (possibly with a hybrid zone).

There are reports that the bones of discus fishes from whitewater (five Symphysodon aequifasciata) and clearwater (four Symphysodon discus) habitats differ in chemical composition as a result of the different water chemistry (Geisler & Schneider, 1976). However, the current water conditions in the Amazon are very variable on small spatial and temporal scales. All species of discus fishes can be found in all water types, and this result may merely indicate a plastic ability of these fishes to utilise different chemicals depending on those available in the environment.

In summary, it is likely that the limited species number in the genus Symphysodon is a result of the adaptation to the floodwater habitats of the Amazon. Such adaptation has ensured that populations across a large geographical area share a continous habitat resulting in high enough gene flow to prevent much differentiation. Apparently, even major differences between the water types and the effect of large-scale tectonic processes have only provided enough selection pressure for three species to have evolved in this genus.

Further investigation of the genus, with larger sample sizes for molecular analyses, faster evolving nuclear markers and samples from elsewhere in the distribution of the genus, may further improve our understanding of these fishes and the evolutionary and the geohistorical processes that have led to their unusual distribution. The challenging result of the present study is the absence of genetic differentiation between two morphologically distinct species. It confirms the need for complementing molecular analyses with morphological analyses to distinguish discovery of morphologically and genetically distinct, but very similar, species that are geographically separated across an apparently homogeneous habitat without obvious physical present day barriers.

This research was supported by a grant from the HIGH LAT programme for Access to the Swedish Museum of Natural History (NRM), within the European Community programme ‘Improving the Human Research Potential and the Socio-Economic Knowledge Base’, making it possible for J.R. (then at the University of Hull) to visit the NRM in 2003. We are grateful to A. Silfvergrip, E. Åhlander, M. Norén, P. Eldenäs and B. Delling for assistance in handling and photographing specimens and generating sequence data, and O. Svensson, R. Zardoya and two anonymous reviewers for comments on the manuscript. J.R. acknowledges the support of his PhD supervisor, G. Turner, University of Hull, and the Fisheries Society of the British Isles for their studentship.