Previous genetic surveys of reef organisms in the tropical Atlantic have produced a mosaic of outcomes in terms of the levels of separation among different provinces, with virtually every conceivable pattern being evident 40. The short-spined sea urchin (genus Tripneustes), the red lip blenny (genus Ophioblennius), and some wrasses of the genera Thalassoma and Halichoeres show deep genetic breaks (d = 2.0% – 12.7% in mitochondrial DNA coding regions) that correspond to the four major tropical Atlantic biogeographic provinces 20313241. In the ocean surgeonfish (Acanthurus bahianus), GC populations are separated from those in Brazil and the central Atlantic by an mtDNA sequence divergence of d = 2.4%, but no difference is observed between populations at Brazil and the mid-Atlantic islands 34. Other species show little differentiation throughout the tropical W Atlantic (doctorfish, A. chirurgus; pygmy angelfish, genus Centropyge; goldspot goby, Gnatholepis thompsoni 343542), or even throughout the entire tropical Atlantic (blackbar soldierfish, Myripristis jacobus 37). In contrast, at the lower end of the spatial scale, some species of wrasses (genus Halichoeres) and cleaner gobies (genus Elacatinus) show deep genetic breaks and reciprocally monophyletic groups on a scale of tens of km within the GC 364344.

In C. multilineata, three levels of genetic diversity were observed: First, the tropical eastern Atlantic populations formed a monophyletic group, separated from all other populations by at least seven diagnostic mutations. Second, there was a significant shift in haplotype frequencies between the GC and the South Atlantic (Brazil + central Atlantic islands) indicating population breaks between the GC and those two provinces (Table 2). Third, there was no detectable genetic difference between the Brazilian and central Atlantic populations. The pattern that emerges from most Atlantic phylogeography studies is that when deep divergences are present, they generally correspond to the major biogeographic provinces, although there are notable instances of strong within-province differentiation (e.g.: gobies and wrasses within both the GC and Brazil 364345).

In the brown chromis and other reef organisms, the divergence between the eastern Atlantic and the remaining populations (Table 2) may be explained by 1,700 km or more (the shorter straight line distances between the eastern Atlantic and the western/central Atlantic populations are ~1,700 km between St. Paul's Rocks and Cape Verde, ~2,300 km between St. Helena and Sao Tome and ~2,500 km between Ascension and Sao Tome) of unsuitable open-ocean distances, combined with the general east to west direction of prevailing currents at low latitudes. While low but significant pairwise comparisons were observed between the two provinces of the South Atlantic (Brazil and the central Atlantic) and the Caribbean, separated by no more than 2,300 km, no difference was observed between Brazil and the central Atlantic Islands, separated by 2,200 km.

Low genetic structure, such as that seen in C multilineata might be expected in organisms with a long pelagic larval stage 3437 and a similar signal has been observed in the long-spined sea urchin, genus Diadema with a ~6 week larval stage 33 and the ocean surgeonfish, Acanthurus bahianus, which has a larval duration of ~60 days 34. C. multilineata, however, has a relatively short pelagic larval stage averaging 27 days (33 days maximum 46), which is substantially less than the average time (48 days based on average current velocity) to cross from Brazil to the mid-Atlantic 47. However, juvenile C. multilineata have been observed in association with floating debris in Belize (L. A. Rocha pers. obs.), and juvenile Chromis atrilobata (C. multilineata's sister species in the eastern Pacific) have been observed in open water under floating objects 4849; D. R. Robertson pers. obs.). Thus, young C. multilineata juveniles survive in the open ocean long after transforming from the larval stage, and relatively strong tropical currents connecting the mid-Atlantic islands and Brazil may transport juvenile Chromis between those locations 5051. Transport from the western and central Atlantic to the eastern Atlantic evidently has been much less frequent, probably due to larger distances involved in direct transport from W to E Atlantic and smaller source populations for dispersal from the tiny central Atlantic islands (Ascension, St. Helena) to the eastern Atlantic.

The mismatch distribution analysis (Fig. 5), Ramos-Onsins and Rozas's R² test for recent population expansion and Fu's Fs neutrality test (Table 1) indicate that all populations have undergone a recent expansion, except for that at the diminutive St. Paul's Rocks. This expansion was probably caused by the ~10 fold increase in suitable shallow reef habitat associated with the sea-level rise of ~130 m since the last glacial maximum 52. St. Paul's Rocks is an exception: because there is no shelf at that island, which is a pillar rising vertically from deep water 29, available habitat probably hasn't increased with sea-level rise. Thus the lack of an expansion signal may reflect a relatively constant population size due to stability in habitat availability. The lower number of pairwise differences, and haplotype and nucleotide diversities observed at Cape Verde and Sao Tome in the eastern Atlantic, indicate a more recent expansion at those locations than in the western and central Atlantic. In addition to lower sea-levels and corresponding reduced habitat available during the last glacial period in the eastern Atlantic, water temperatures in that area were lowered more than in the western and central Atlantic by enhanced coastal upwelling 53, a stress on tropical species that may have caused strong fluctuations or reductions in population size.

Migration patterns among populations (Fig. 4) are consistent with the general direction of surface current flows (Fig. 1): the highest rates are from the South Atlantic to the Caribbean (coinciding with the flow of the North Brazil current) and from Brazil to the mid-Atlantic islands (coinciding with the flow of the South Atlantic Gyre). In accordance with the phylogenetic results, migration between western and eastern Atlantic (a direction that opposes major surface currents; Fig. 1) was effectively zero.

Our phylogeographic analysis of C. multilineata provides two lines of evidence relevant to the mechanism(s) that produced the GC hotspot of diversity. First, supporting accumulation, the short (i.e. relatively young) branches are mostly in Brazil and the central Atlantic islands (in Fig. 2 ~75% of the short branches are blue, endemic to Brazil and the two mid-Atlantic islands). However, the short branches also include a few Caribbean haplotypes (~25% of short branches in the mostly South Atlantic lineage in Fig. 2), indicating that the lineages diversified in the South Atlantic, and that individuals carrying those haplotypes recently arrived in the Caribbean. This South to North pattern of dispersal is consistent with the pattern of oceanic current flow (Fig. 1) and migration rates (Fig. 4), and indicates that the hotspot of diversity in the Caribbean has acted recently as a CA.

Similarly to what we describe here for C. multilineata, the Panamanian sample of the wrasse Halichoeres bivittatus has one individual (among 23 sampled) with a haplotype that is more similar to those found in Brazil than those in other Caribbean locations (Fig. 6a), indicating the recent arrival of a haplotype of South Atlantic origin 36. Additionally, in the goby genus Gnatholepis, and the angelfish genus Centropyge, the Atlantic species are recently derived from much more diverse Indo-Pacific groups, and apparently only recently arrived in the GC from the Indian Ocean via the South Atlantic 354254, supporting accumulation of species at the GC hotspot.

The phylogeny of parrotfishes (genus Sparisoma) also supports accumulation at the Caribbean (Fig. 6c and 6d). The parrotfish S. axillare is abundant and widely distributed throughout the Brazilian coast, but known only from SE Venezuela in the Caribbean, indicating that the population there is probably a result of recent dispersal from Brazil. Likewise, the parrotfish S. griseorubra is restricted to the southern Caribbean (SE Venezuela) and its sister species is S. frondosum (endemic to Brazil), indicating that S. griseorubra likely originated from ancient dispersal by the ancestor of the S. frondosum/S. griseorubra lineage. The splitting of S. frondosum and S. griseorubra alternatively could be explained by sympatric speciation in the Caribbean (the groups with Brazilian affinities co-occur with their Caribbean counterparts in the southern Caribbean) followed by dispersal towards Brazil. However, speciation with gene flow in reef fishes has only been inferred when there are strong ecological gradients 36, or where the fish are strictly associated with coral hosts 55. As this is not the case for Sparisoma, this alternative is less likely than an allopatric split between Brazil and the Caribbean after dispersal northwards by a Brazilian lineage.

Supporting origin at the Caribbean, almost all basal haplotypes of the C. multilineata tree are observed only in Caribbean individuals (Figs. 2 and 3), which indicates that populations at the three other regions are derived from a Caribbean ancestor. Moreover, the Caribbean samples (either separately or in combination) have the widest mismatch distributions (spanning 13 to 15 mutations; Fig. 5) indicating that this region hosts the oldest and most stable population. We can also infer that, because all eastern Atlantic haplotypes apparently have a single origin (monophyly in Fig. 2), they most likely originated from a single colonization event (Fig. 3) on the order of two hundred thousand years ago (based on our trans-isthmian molecular clock estimate of 4.4%/Myr). The recent finding of Centropyge aurantonotus in the eastern Atlantic (previously known only from the western Atlantic, but recently observed in low numbers in Sao Tome 56) and the general direction of trans-Atlantic range expansion in the gold spot goby 35 support this rare west to east colonization route.

Lending further support to the CO hypothesis, populations of the wrasse H. radiatus in Brazilian oceanic islands are much more closely related to Caribbean H. radiatus than to populations of H. brasiliensis (its sister species) in the adjacent coast line, indicating that these islands were colonized by migrants of Caribbean origin (Fig. 6b). A similar pattern of southward dispersal from the GC to the offshore Brazilian islands has been observed in gobies of the genus Bathygobius 45. Colonization outward from the Caribbean must be a rare event because it goes against prevailing currents, but as our analysis indicates, it has happened and can lead to the establishment of new populations, supporting the CO hypothesis.

Support for origin of diversity in the GC comes from other recent phylogenetic and biogeographic analyses. A survey of seven-spined gobies 57 shows that two peripheral species of Elacatinus (E. figaro from Brazil and E. puncticulatus from the eastern Pacific) are older species whereas the 10 newest (youngest) Elacatinus species are found only in the Caribbean biodiversity hotspot. That is, recent speciation that produced 83% of the species in this genus occurred within the GC. Likewise, the diverse families Chaenopsidae and Labrisomidae are represented in the GC by 45 species each, but elsewhere in the Atlantic there are only four and 11 Atlantic species in each of these families respectively 405859. Hamlets (genus Hypoplectrus) include as many as a dozen or more closely related "species" and are restricted entirely to the GC 60. Thus, speciation leading to considerable faunal enrichment most likely occurred in situ within the GC in those four taxa.

Recent phylogenetic analyses of large reef fish groups also provide useful information to this debate. Among wrasses (Labridae), the Caribbean and Eastern Pacific Halichoeres are for the most part monophyletic, indicating that they diversified in situ and supporting CO in a recent time scale, but at the same time their diverse group of ancestors is composed of Indo-Pacific species 2161 supporting CA deeper in time, on a scale of tens of millions of years. Similarly to the wrasses, the Caribbean groupers of the genera Epinephelus and Mycteroperca are monophyletic groups derived from Indo-Pacific ancestors 62, supporting both ancient CA and more recent CO. The Caribbean butterflyfishes are mostly a paraphyletic assemblage of lineages derived from more diverse Indo-Pacific groups, supporting CA 63. Within the damselfishes, the Caribbean (and Atlantic) species Abudefduf saxatilis seems to be a recent arrival from the Indo-Pacific (it is very closely related to a group containing eight Indo-Pacific and one eastern Pacific species), also supporting CA 64. Even though these large scale phylogenies are useful, most lack peripheral (mostly Brazilian) endemics, making their contribution to this debate limited. However they provide an excellent framework that, with the addition of a few key species, may become an important piece in the tropical biodiversity puzzle.

Chromis multilineata is a common reef fish that is widely distributed on both sides of the tropical Atlantic, as well as at the mid-Atlantic islands 68. They have demersal eggs that develop into pelagic larvae in about three days 6970. Estimates of pelagic larval duration range from 24 to 33 days 46. Adults can be locally very abundant and are usually found in schools from a few to several hundred individuals, swimming and feeding on plankton above the reef 71.

A total of 183 specimens of Chromis multilineata were obtained from 10 locations, which included at least two locations within each of the four provinces (Table 1, Fig. 1). Specimens were collected with polespears while scuba diving or snorkeling, between 1997 and 2002. Tissues (muscle and/or gill) were stored in a saturated salt-DMSO buffer (Amos and Hoelzel 1991). A recent phylogenetic survey of damselfishes indicates that the eastern Pacific species C. atrilobata is the sister of C. multilineata 64. To confirm this relationship we sequenced C. atrilobata from Cocos Island (n = 3) and Panama (n = 3) and other species of the genus in the Atlantic that were not included in the phylogeny 64: C. scotti from Florida (n = 2); C. enchrysura from Florida (n = 2), C. limbata from the Azores (n = 4) and C. lubbocki from Cape Verde (n = 3). The resulting tree (not shown) supports the conclusion that C. multilineata and C. atrilobata are sister species.

Total genomic DNA was extracted using QIAGEN (Valencia CA) Dneasy extraction kits following the manufacturer's protocol. Extracted DNA was frozen in TE buffer and archived at -20°C. Primer names indicate the DNA strand (H = heavy and L = light strand) and the position of the 3' end of the oligonucleotide primer relative to the human mitochondrial DNA sequence. A segment of 802 base pairs of the mtDNA cytochrome b gene was amplified with the primers L14725 (5' GTG ACT TGA AAA ACC ACC GTT G 3') and H15573 (5' AAT AGG AAG TAT CAT TCG GGT TTG ATG 3') 72.

Thermal cycling in polymerase chain reactions (PCR) consisted of an initial denaturing step at 94°C for 1 min 20 sec, then 35 cycles of amplification (40 sec of denaturation at 94°C, 30 seconds of annealing at 52°C, and 55 sec of extension at 72°C), and a final extension of 2 min 30 sec at 72°C. Each tube contained 15 μl of water, 2.5 μl of MgCl2, Mg free buffer and DNTPs, 0.3 μl of each primer, 0.4 μl of Taq DNA polymerase and 1.5 μl of purified DNA. Excess primers were removed through simultaneous incubation of PCR product with exonuclease I and shrimp alkaline phosphatase (USB Corp., Cleveland OH).

Sequencing reactions with fluorescently-labeled dideoxy terminators (BigDye) were performed according to manufacturer's recommendations, and analyzed with an ABI 377 automated sequencer (Applied Biosystems, Inc. Foster City, CA). All samples were sequenced in the forward direction (with L14725 primer), but rare and questionable haplotypes were sequenced in both directions to ensure accuracy of nucleotide designations. Representative haplotype sequences are deposited in GenBank under accession numbers EU431997 – EU432046. Copies of the complete data set are available from LAR.

Sequences were aligned and edited with Sequencher version 3.0 (Gene Codes Corp., Ann Arbor, MI). Population structure and gene flow were assessed with an analysis of molecular variance AMOVA 73 in the program Arlequin version 3.0, which generated Φ_ST values (a molecular analog of F_ST that includes sequence divergence among haplotypes as well as haplotype frequency shifts; 74). Genetic variation is described with nucleotide diversity (π; equation 10.19 75) and haplotype diversity (h; equation 8.5 75) within each location.

The computer program MODELTEST version 3.06 76 was used to determine the substitution model that best fits the data through a minimal theoretical information criterion (AIC). The model chosen was TRN+Γ 77 with a gamma distribution of 0.97 to estimate sequence divergences (d values) between haplotypes. Equal weighting of all three codon positions was used. Relationships between all haplotypes and closely related species were estimated using a Bayesian phylogenetic analysis performed with MrBayes 3.1 78. Preliminary runs were performed to monitor the fluctuating value of the likelihoods of the Bayesian trees, and all parameters appear to reach stability before 250,000 generations. The Markov chain analysis was run for 20 million generations. A burn-in period, in which the initial 10,000 trees were discarded, was adopted and the remaining tree samples were used to generate a 50% majority rule consensus tree. The posterior probability of each clade was then provided by the percentage of trees identifying the clade 79. In addition, the software PAUP 4.0b10 80 was used to conduct maximum parsimony (MP) and neighbor-joining analyses that were evaluated with 1000 bootstrap replicates implemented with PAUP* version 4.0b10 80. The minimum evolution criterion was used by applying maximum likelihood distances estimated with the model chosen by Modeltest in the neighbor-joining analysis. The resulting Bayesian, parsimony and neighbour joining trees were not significantly different.

Departure from neutrality was tested using Fu's Fs 81 and Ramos-Onsins and Rozas' R2 statistic 82. The R2 measure is based on the difference between the number of singleton mutations and the average number of nucleotide differences among sequences within a population sample. Both Fs and R2 are powerful tests used to detect recent population expansions under assumptions of neutrality 8182. Significance of R2 and Fs were evaluated by comparing the observed value with a null distribution generated by 10,000 replicates, using the empirical population sample size and observed number of segregating sites implemented by DnaSP version 4.10.9 83. Moreover, in order to estimate the demographic parameters of past population expansions we calculated mismatch distributions (or the distribution of the pairwise genetic distances) for all populations using DnaSP. Time and magnitude of the inferred population expansion were determined by calculating θ₀, θ₁ and τ, where θ₀ = 2N_0μ (N₀ = population size before expansion); θ₁ = 2N_1μ (N₁ = population size after expansion); and τ = 2μt (μ = mutation rate per site per generation; t = time in generations).

The mutation rate per lineage per year (λ) was estimated by solving the formula λ = d/2T, where d is the genetic distance between C. multilineata and C. atrilobata, and T is the time since divergence between the two species used. We used a T of 3.5 million years (Myr), as that is the upper limit age for the final closure of the Isthmus of Panama 84. The rate of 2.21%/Myr (within lineages) that we estimated is very similar to that obtained in a recent survey of Chromis using a slightly different portion of the cyt b gene (2.36%/Myr; 85. The closure of the Isthmus of Panama was used in several other studies to estimate mutation rates in reef fishes (e.g.: 35868788).

Migration rates between the major biogeographical provinces (Brazil, the Caribbean, central Atlantic islands and eastern Atlantic islands) Nm (where N is effective female population size and m is migration rate) were calculated with the software MIGRATE version 1.7.6 89, which uses a maximum likelihood approach based on coalescence theory 90. Estimators of migration rates based on coalescence theory can detect asymmetries (directionality) in migration rates and differences in population sizes, a considerable advantage when testing migration rates in the context of evolutionary theory 91. Most of the default settings for MIGRATE were used on the first run, but the number of trees sampled was increased to 10,000 for the short chains and 100,000 for the long chains to avoid local maxima on the likelihood surface. A trial run was used to generate the input parameters; the average of results of 10 runs (number of trees sampled was progressively increased from 10,000 to 300,000 for the short chains and from 100,000 to 3,000,000 for the long chains) is reported here. We also used the software IM to estimate migration between populations of C. multilineata; IM uses a Markov chain Monte Carlo approach 92. The results from the IM analysis are not shown because they were completely compatible with the results from MIGRATE.

In addition, we re-analyzed data from four groups of reef fish species (the wrasse genus Halichoeres 36, and the parrotfish genus Sparisoma 38) to specifically assess predictions from the center of origin and center of accumulation hypotheses. To do so we constructed statistical parsimony networks for those species as well as Chromis multilineata using the computer program TCS version 1.21 93 and analyzed the geographical distribution of mtDNA haplotypes.