Introduction

Neotropical diversity is near-legendary, and many hypotheses have been proposed in an effort to understand current distributions of flora and fauna. Three of these, the Miocene/Pliocene marine incursion hypothesis (Räsänen et al., 1995; Webb, 1995), the Pliocene/Pleistocene riverine barrier hypothesis (Wallace, 1852) and the Pleistocene refugia hypothesis (Haffer, 1969), are the focus of this review because they can be examined in the light of phylogenetics and population genetics, and because of their general applicability for vertebrates (Moritz et al., 2000; Bates, 2001; Aleixo, 2004) and several insects, including butterflies (Brown et al., 1974; Brower, 1994; Hall and Harvey, 2002), beetles (Erwin and Pogue, 1988; Erwin, 1998; Scataglini et al., 2006) and bees (Dick et al., 2004).

The earliest of the three hypotheses, the marine incursion hypothesis, includes several aspects of the previous paleogeography hypothesis. The latter focused on early tectonic movements and sea-level changes (Chapman, 1917; Frailey et al., 1988), while the former proposes that sea-level changes during the Tertiary caused Amazonian diversification (Räsänen et al., 1995; Webb, 1995). The marine incursions resulted in extensive flooding, mainly in the Amazonian lowlands, leaving three large land blocks relatively isolated: the Brazilian Shield, the Guiana Shield and the eastern Andean slopes (Figure 1). For Amazonian birds (Bates, 2001), it has been proposed that basal populations in these land blocks began to diverge and eventually speciated, resulting in major areas of endemism (Cracraft, 1985; Hall and Harvey, 2002). Many centers of endemism have been identified by mapping a wide range of Amazonian biota (for example, Prance, 1982; Patton et al., 2000; summarized in Hall and Harvey, 2002).

Figure 1
figure 1

Dotted lines represent geological features and regions that were not flooded during the marine incursion.

Full size image

Wallace (1852) proposed the riverine barrier hypothesis while he was living and working in the Amazon in the 19th century. He postulated that the rivers would act as barriers to gene flow between populations on opposite banks, eventually resulting in speciation. The time frame for this hypothesis is around the Pliocene/Pleistocene boundary when the major Amazon river developed (Gascon et al., 2000; Campbell et al., 2006). In phylogeographic terms, sister intraspecific clades and species should exist across neotropical rivers, rather than within the areas between rivers (Aleixo, 2004). It should also be possible, using phylogenetic and population genetics data, to discriminate between primary divergence across rivers and more recent secondary contact along rivers between nonsister taxa that diverged in another geographic area (Moritz et al., 2000). The geology of neotropical rivers (generally narrower at the source and wider at the mouth) suggests another prediction: differentiation should be lowest among populations on either side near the headwaters and gradually increase to become highest among populations on either side at or near the mouth (Haffer, 1993; Gascon et al., 2000).

The Pleistocene refugia hypothesis posits the most recent time frame, and it was originally promoted for the Neotropics by Haffer (1969) to explain avian diversification. The basic model is that during Pleistocene climatic shifts associated with glaciation, areas of moister vegetation types (such as rainforest) and their populations contracted and became isolated, surrounded by drier vegetation such as savannah. During warmer, wetter periods these vegetation islands (refugia) and their populations expanded. The contractions and expansions occurred cyclically, and are thought to have resulted in cladogenesis. Although significant changes in forest cover during Pleistocene glaciation periods have not been found (Colinvaux et al., 2000), and it is problematic to derive phylogenetic predictions (Patton and da Silva, 1998; Moritz et al., 2000), two population genetics patterns have been shown to be strongly associated with Pleistocene refugia (in North America, Zink, 1997; in Europe, Hewitt, 1999; in the Neotropics, Ruegg and Smith, 2002; Aleixo, 2004): population bottlenecks associated with refugia contraction and isolation and demographic expansions concordant with refugia expansions. Additional expectations include low genetic variability, shallow levels of mtDNA divergence and little phylogeographic structure (Avise et al., 1987). Allopatric speciation is also thought to be a signature of the refugia hypothesis, because small populations could have been confined in pockets of rain forest separated by savannah; then diverged by genetic drift (Monteiro et al., 2003).

In this review, we ask when major divergences occurred, and which of these three hypotheses seems most likely to be supported, for each of several insect vectors of human pathogens with broad neotropical distributions for which sufficient prior phylogenetic, population genetic or biogeographic data exist: sandflies (leishmaniasis), triatomines (Chagas disease) and mosquitoes (malaria). Several insect vectors of neotropical human pathogens are notorious for cryptic speciation and geographic patterning and thus could be useful in a better understanding of neotropical insect diversification.

Neotropical biota have been significantly influenced by a long and complex interplay of climatic, tectonic and paleoenvironmental changes (Vrba, 1992; de Fátima Rossetti et al., 2005). Since the late Oligocene (33.7–23.8 MYBP), the major geological events that have been hypothesized to play an influential role in diversification include sea-level changes congruent with a marine transgression during the late Oligocene/early Miocene, the uplifts of the Colombian eastern Cordillera, the Andean Cordillera and Central American mountain ranges, and the emergence of the Isthmus of Panama early in the Pliocene (de Fátima Rossetti, 2001). The prevailing climate was initially tropical (wet) but became dry following the emergence of the Panama Isthmus, reflecting a worldwide trend of drier, more severe climates and expansion of savannahs in place of forests during the Pleistocene–Holocene glaciations (Crowley and North, 1991; Hallam, 1994; Hooghiemstra and Van de Hammen, 1998; de Fátima Rossetti, 2001). In eastern and central Amazonia, paleoecological evidence indicates wet conditions associated with rainforest expansion since the last glacial maximum about 20 000 years ago (Burnham and Graham, 1999).

A current controversy is whether relatively recent climatic instability during the Pleistocene is the primary force driving speciation, or whether diversification followed earlier tectonic events coupled with the marine transgressions. These two scenarios need not be mutually exclusive. Highland and lowland neotropical birds were differentially affected by recent climatic changes, with highland fauna showing a late Pleistocene increase in diversification rates most likely due to habitat fragmentation by Andean glacial cycles; by contrast, diversification rates in lowland fauna were highest during the late Miocene, and decreased toward the present (Weir, 2006). Several geologically based studies suggest that species differentiation can be attributed to environmental stresses (Sheldon, 1996; Renaud and Dam, 2002) resulting from a combination of climate, sea level, sedimentary processes and tectonics (de Fátima Rossetti et al., 2005). This pattern was supported in a study of an Amazonian forest bird superspecies Xiphorhynchus spixii/elegans whose diversification is hypothesized to be the result of an interaction among geology, hydrography and sea level changes (Aleixo, 2004).

One of the main reasons we opted to begin this comparison with mtDNA data is because it is frequently used to infer biogeographical history using the genealogies of organisms that are co-distributed (Avise et al., 1987, 2000; Soltis et al., 2006), and there exist comparable mtDNA data sets among the insect species we chose. We recognize that mtDNA tracks exclusively the maternal history, and that it is prone to introgression (Besansky et al., 2003; Rubinoff and Holland, 2005). We also briefly compare several other markers (allozymes, pheromones, microsatellites) and, in many cases, these data add very relevant components and perspectives, but were not common to all the species of interest or else the sampling was not sufficiently broad geographically across the Neotropics to enable the comparisons we sought.

Materials and methods

In considering our choices of insect vectors we offer several caveats: first, sandflies have very limited dispersal capabilities (usually no more than 1 km, Dye et al., 1991; Morrison et al., 1993) are often abundant in peridomestic environments of rural communities, and are therefore patchily distributed (Soto et al., 2001); second, triatomine bugs are generally either domestic or sylvatic, a characteristic that influences both dispersal and geographic distribution (Forattini, 1980; Abad-Franch and Monteiro, 2005); and third, the number of generations per year for each insect group is not identical which may influence the estimated dates of divergence (Monteiro et al., 2003). Anopheles are estimated to have ten generations per year (Walton et al., 2000); Lutzomyia, 5–7 (Cárdenas et al., 1999), and triatomines range from 3 or 4 per year to one generation every 2 years (Krinsky, 2002). Anopheline mosquitoes are likely to be more broadly naturally dispersed compared with sandflies or triatomines because their life stages are first aquatic and then terrestrial, and flight ranges have been recorded at >7.2 km (Charlwood and Alecrim, 1989 for A. darlingi) and >12 km (Correa et al., 1950 for A. albitarsis s.l.). We also expect differences between A. darlingi and A. albitarsis s.l. based on their preferred aquatic habitat: A. darlingi tends to be found along warm lowland rivers edges (Charlwood, 1996; Roberts et al., 2002); in contrast, A. albitarsis s.l. prefers shallow, sunlit pools (Wilkerson et al., 1995a; Conn et al., 2002; Brochero et al., 2005). We would thus expect these two species could have responded differently to the cyclical contraction and expansion of refugia.

The predictions for each of the three hypotheses in Table 1 are modified from Avise et al. (1987), Hewitt (1999), Gascon et al. (2000) and Aleixo (2004). We examined in some detail three major mtDNA sequence data sets: for Rhodnius prolixus and R. robustus, Monteiro et al. (2003); for Anopheles albitarsis s.l., Lehr et al. (2005); and for A. darlingi, Mirabello and Conn (2006a). The sequences from Arrivillaga et al. (2002) for L. longipalpis were not in GenBank or EMBO and therefore the divergence estimates are not included in Table 2. A. albitarsis s.l. cytochrome oxidase I (COI) mtDNA sequences were obtained from GenBank, accession numbers DQ076204–DQ076238 (Lehr et al., 2005). The A. darlingi COI mtDNA sequence data were from GenBank, accession numbers DQ298209–DQ298244 (Mirabello and Conn, 2006a). The R. prolixus and robustus cytochrome b (cyt b) mtDNA data were from Monteiro et al. (2003), and the divergence time was recalculated to an extra decimal place. The percent sequence divergence estimates were calculated using a Kimura 2-parameter model (Kimura, 1980) with MEGA version 3.1 (Kumar et al., 2004). The divergence time estimates were calculated assuming a mutation rate of 2.3% sequence divergence per million years as in Monteiro et al. (2003), based on the estimates for the complete mitochondrial genome for recently diverged arthropods (2.3% per MY; Brower, 1994).

Table 1 Predictions derived from three common hypotheses of neotropical diversification for three co-distributed groups of insect vectors, Lutzomyia longipalpis s.l., Anopheles albitarsis s.l., A. darlingi, Rhodnius prolixus and R. robustus
Full size table
Table 2 Prediction results for L. longipalpis s.l., A. albitarsis s.l., A. darlingi, R. prolixus and R. robustus. based on mtDNA COIa data
Full size table

Sandflies (Diptera: Psychodidae)

Lutzomyia longipalpis s. l. is the main vector of neotropical visceral leishmaniasis, with a discontinuous distribution from southern Mexico to northern Argentina (Young and Duncan, 1994). It is generally found in dry tropical forest or semiarid habitats, including open savannah; but also in the lower Amazon basin associated with well-drained soils subjected to periodic flooding (Lainson et al., 1983). Habitats within regions are often discontinuous and isolated by both geographic and climatic barriers (Soto et al., 2001). Multiple markers indicate that L. longipalpis is a species complex (reviewed in Lainson and Rangel, 2005), although the exact number of species is still under investigation. Morphological differences were the first clue to potential cryptic species in L. longipalpis within Brazil (Mangabeira, 1969), with two forms from the states of Ceará and Pará hypothesized to be reproductively isolated (Ward et al., 1983), and several forms or chemotypes subsequently proposed to be sympatrically differentiated in Brazil by sex pheromones (Ward et al., 1988; Hamilton et al., 1996a, 1996b, 1996c, 2005). There is now considerable corroborative evidence suggestive of incipient speciation in Brazil, including sequences of the period gene (Bauzer et al., 2002a, 2002b), significant sequence divergence in the cacophony IVS6 intron (Bottecchia et al., 2004), microsatellite loci (Maingon et al., 2003) and courtship song differentiation (Souza et al., 2004). A study that combined morphology and allozymes to examine several Brazilian populations that represent three of the morphological patterns in L. longipalpis s.l., however, did not support the hypothesis of multiple species (de Azevedo et al., 2000), but allozymes evolve relatively slowly and are not the most informative markers for detecting recently diverged species; and Dipteran species complexes are rife with morphological ‘look-alikes’.

In studies that focused on a broader geographical range of L. longipalpis s.l., crossing experiments detected male hybrid sterility among populations from Costa Rica, Colombia and Brazil (Lanzaro et al., 1993). Several allozyme studies (Lanzaro et al., 1993; Dujardin et al., 1997; Mukhopadhyay et al., 1998; Mutebi et al., 2002), a chromosome study (Yin et al., 1998), and mitochondrial DNA sequences (Soto et al., 2001; Arrivillaga et al., 2002, 2003; Hodgkinson et al., 2003) provide evidence for cryptic species except in Brazil, and thus are in disagreement with the pheromone, courtship song, period and cacophony sequence and microsatellite data discussed above.

Sympatric speciation in Lara state, western Venezuela, using allozymes (Lampo et al., 1999), parapatric and allopatric speciation followed by secondary contact in Ceará state, northwestern Brazil, based on crossing experiments, pheromones, copulation songs, period gene data and microsatellites (reviewed in Watts et al., 2005), and allopatric speciation (throughout its neotropical distribution, Soto et al., 2001; Arrivillaga et al., 2002) have been invoked, and each mode of speciation could play a significant local or regional role across the broad geographic range of this species complex.

Testing the marine incursion hypothesis is complicated because L. longipalpis s.l. appears to be absent from the eastern part of the Guiana Shield (no records from Guyana, Suriname or French Guiana; Arrivillaga et al., 2003). The Brazilian Shield populations form a monophyletic clade in both the NADH dehydrogenase subunit 4 (ND4) and COI studies, following prediction 1. No samples from the Guiana Shield were tested in the ND4 study, and the single sample from the Guiana Shield in the COI study belongs to a clade with three localities in the eastern Andean Cordillera (cis-Andean; Figure 2). Both studies show the Brazilian as the most derived of the four clades, but disagree on which clade is basal. In the ND4 study, the Central American localities are basal and strongly supported (95%, six synapomorphies; Soto et al., 2001); in the COI study these populations are included with northern Colombian and western Venezuelan samples in the trans-Andean clade (moderately well supported – 74%; Arrivillaga et al., 2002). No western Amazonian lowlands samples were included in either study to enable us to test whether these populations would be more derived. In summary, prediction 1 of the marine incursion hypothesis is partially supported, prediction 2 is not supported and prediction 3 could not be evaluated (Table 2).

Figure 2
figure 2

Lutzomyia longipalpis geographic locations of the populations and clades (specified in legend) based on the mtDNA COI phylogenetic analysis of Arrivillaga et al. (2002).

Full size image

Limited geographic sampling in the Amazon Basin prohibits discussing the riverine hypothesis. Soto et al. (2001) detected four clades across the distribution of L. longipalpis s.l. using sequences of the mtDNA ND4 gene: northern South America, Brazil, Central America and an isolated population in Colombia. Using additional samples from Brazil and Venezuela, and sequences of the mtDNA COI gene, Arrivillaga et al. (2002) also detected four clades (A–D): Laran (Venezuela); cis-Andean (Venezuela; Colombia; northern Brazil); trans-Andean (Venezuela; Colombia; Central America); and Brazilian (Figure 2). Although high within population genetic diversity was found using the ND4 gene (Soto et al., 2001) and low variation using COI (Arrivillaga et al., 2002), both studies demonstrated deep levels of mtDNA divergence (and cladogenesis), and substantial phylogeographic structure. Additionally, no sign of a demographic expansion was detected using ND4 sequences (this was not tested using the COI sequence data). These results together suggest there is very little support for the Pleistocene refugia hypothesis for L. longipalpis s.l. based on mtDNA (Table 2). Furthermore, the moderate genetic variability, substantial phylogeographic structure and considerable divergence among populations from Venezuela and Brazil detected using microsatellite markers (Watts et al., 2005) do not seem to support the refugia hypothesis either, but tests were not conducted to detect bottlenecks or expansions. Estimations of divergence time in this species complex, especially among sympatric Brazilian clades, would be a valuable contribution.

The COI sequence data support speciation events that began during the Pliocene, and the large geographic gap separating the Brazilian and Laran clades (Figure 2) suggests their most recent ancestor was part of a widespread sub-Andean/Amazonian gene pool (Arrivillaga et al., 2002). Populations are hypothesized to have subsequently dispersed across Brazil, then to the Andes, and lastly into Central America, after the final East Andean uplift in the early Pleistocene. The large genetic distances and the predominantly allopatric ranges among the Laran, Andean and Brazilian clades are attributed to vicariant geologic events as the most probable significant evolutionary force (Arrivillaga et al., 2002); however, this is disputed by the microsatellite data provided in Watts et al. (2005). Interestingly, the samples from Curarigua, Venezuela, which together comprise the Laran clade, show quite low nucleotide divergence and are considered to be relictual (Arrivillaga et al., 2002). The sympatric Andean clades could be the result of initial allopatric speciation followed by a more recent secondary contact via dispersal. This could be tested by assessing populations in this region for demographic expansions, frequently associated with recent range expansions resulting in secondary contact.

Our interpretation of the COI data differs slightly in the most probable order of dispersal, from a possible Laran ancestor (basal in the Maximum Parsimony analysis and supported by 100%; Arrivillaga et al., 2002, 2003), to the Andes, then into Central America, and lastly across Brazil. Based on the ND4 study, the order of dispersal would differ, beginning with a likely Central American ancestor, and dispersing through the Andes and lastly across Brazil (Soto et al., 2001). Overall, the timing for L. longipalpis s.l. divergence is Pliocene and Pleistocene, and the primary mode of divergence could be an interaction between the marine incursion and refuge hypotheses, although neither one is well supported (Table 2).

Triatomines (Hemiptera; Reduviidae)

Triatomines are a group of hematophagous insects that transmit Trypanosoma cruzi, the causative agent of Chagas disease, one of the most serious parasitic diseases in the neotropics (Miles et al., 2003). Rhodnius prolixus is predominantly domestic, and is considered the primary Chagas disease vector in Central America, Colombia and Venezuela (Lent and Wygodzinsky, 1979). R. robustus is sylvatic, although its taxonomic status has been uncertain (Schofield, 2000). The two species overlap broadly in South America, and the distribution of R. prolixus extends as far north as southern Mexico (Lent and Wygodzinsky, 1979). No diagnostic allozyme loci have been detected in these species (Harry et al., 1992), and overlapping morphological characters make field identification problematic (Harry, 1993). They were found to be distinctive evolutionary lineages using two mtDNA fragments, cytochrome b and large subunit ribosomal RNA (Lyman et al., 1999). R. robustus was determined to be heterogeneous based on mtDNA and nuclear fragments combined, and hypothesized to consist of multiple cryptic species (Monteiro et al., 2000).

The two Orinoco region Rhodnius (R. prolixus and R. robustus clade I) are estimated to have diverged from the three R. robustus Amazonian clades (II–IV; Figure 3) during the Pliocene (Monteiro et al., 2003; Table 3). Amazonian clade III consists of localities from the Brazilian Shield and clade IV comprises four localities from the Guiana Shield plus one from the Brazilian Shield, and these clades are reciprocally monophyletic (R. robustus II, the most derived of the three R. robustus Amazonian clades, is primarily from the western Amazonian lowlands (Figure 3)). Thus prediction 1 of the marine incursion hypothesis is strongly supported (Table 2). The Brazilian and Guiana Shields are ancestral for R. robustus, because the two most basal R. robustus clades, IV and III, are both strongly supported (80 and 99%, respectively) in the MP tree (Monteiro et al., 2003). Together these data provide some support for prediction 2 of the marine incursion hypothesis (Table 2). Furthermore, the most derived of the Amazon region R. robustus clades (II; Figure 3) spans the central and western Amazonian lowlands, giving some support to prediction 3 of the incursion hypothesis (Table 2). The geographic sampling was not designed to examine the riverine hypothesis.

Figure 3
figure 3

Rhodnius robustus I–IV and R. prolixus geographic locations of the five clades (specified in legend) based on the mtDNA cyt b phylogenetic analysis of Monteiro et al. (2003).

Full size image
Table 3 Mean mitochondrial sequence divergence levels (Kimura 2-parameter model) and the estimated time of divergence between the clades of R. prolixus and R. robustus, A. albitarsis s.l., and A. darlingi genotypes
Full size table

Monteiro et al. (2003) detected a molecular signature consistent with the refugia hypothesis for R. prolixus that included an extremely low level of nucleotide diversity, shallow levels of mtDNA divergence and little phylogeographic structure (Table 2). A recent population bottleneck with subsequent human-aided dispersal was hypothesized, which would fit prediction 1 (Table 1). Tests were not conducted for an expansion. The estimated divergence time between R. prolixus and R. robustus I (in the Orinoco region), and among R. robustus II, III and IV within both the Amazon and Orinoco regions is less than two MYA (Table 3), providing support for diversification during the Pleistocene (Monteiro et al., 2003), despite deep levels of divergence, and considerable phylogeographic structure. Monteiro et al. (2003) suggest that R. prolixus originated 1.4 MYA in the Orinoco lowland forests from an ancestral prolixus/robustus I stock that was separated in distinctive refugia, consistent with our interpretation (Table 2). In summary, R. prolixus was both monophyletic and homogeneous, and strongly supported the refuge hypothesis, in striking contrast to R. robustus, which consists of four monophyletic allopatric clades and supported the marine incursion hypothesis (Table 2).

Anopheline mosquitoes (Diptera: Culicidae)

Anopheles darlingi is a primary malaria vector responsible for transmission of Plasmodium falciparum, P. vivax and P. malariae in several endemic regions throughout much of its extensive range from southern Mexico to southern Brazil (Deane, 1947, 1988; Forattini, 1962; Roberts et al., 1997). Its distribution is discontinuous, with no records from Panama, Costa Rica or Nicaragua (Faran, 1980). A. darlingi has been considered a single species, although there has been evidence of heterogeneity in polytene chromosomes (Kreutzer et al., 1972; Tadei et al., 1982), allozymes (Steiner et al., 1982; Rosa-Freitas et al., 1992), mtDNA (Freitas-Sibajev et al., 1995; Conn et al., 1999; Mirabello and Conn, 2006a), rDNA ITS sequences (Malafronte et al., 1999; Mirabello and Conn, unpublished data), nuclear white gene (Mirabello and Conn, unpublished data) and biological variation (Charlwood, 1996; Voorham, 2002).

The geographic sampling for the mtDNA COI gene fragment study of A. darlingi (Mirabello and Conn, 2006a) was not designed specifically to examine the marine incursion hypothesis. However, none of the populations within the Brazilian or Guianan Shields are in distinctive clusters or monophyletic clades (Mirabello and Conn, 2006a), falsifying prediction 1 (Table 2). In addition, Manguin et al. (1999) hypothesized that A. darlingi originated in South America and then established populations in Central America, and since the largest isozyme mean heterozygosity occurred in southern Brazil and Bolivia these populations may be more ancestral. In support of Manguin et al. (1999), the mtDNA data found the southern Brazil populations (located on the Brazilian Shield) to be more ancestral than those on the Guiana Shield, and elsewhere (Figure 4), which is inconsistent with predictions 2 and 3 of the incursion hypothesis (Table 2).

Figure 4
figure 4

Anopheles darlingi geographic locations of the two clades (specified in legend) based on white gene phylogenetic analysis of Mirabello and Conn, unpublished data. *Only 1 of 28 individuals in Iquitos Peru and 1 of 14 individuals in Guayaramerín, Bolivia was genotype 1, as compared to half of the individuals in Fortuna de Albarico, Venezuela. The distribution of A. darlingi based on the COI data set is virtually identical to this except that sympatric sequences were not detected (Mirabello and Conn, 2006a).

Full size image

The riverine barrier hypothesis was not specifically tested, but populations of A. darlingi north of the Amazon River (Amapá State) were differentiated from those south of the Amazon River (Pará State) using mtDNA sequence data, and this was not due to isolation by distance (Mirabello and Conn, 2006a). Perhaps not surprisingly, microsatellite markers also detected significant divergence between populations of A. darlingi on either side of the mouth of the Amazon (approximately 270 km wide; Conn et al., 2006) compared with very limited divergence for A. darlingi on either side of the Negro River, a major tributary of the Amazon in north central Brazil (Scarpassa and Conn, 2007). Because A. darlingi is most common along warm lowland river edges (Charlwood, 1996; Roberts et al., 2002) and rare in rain forest, rivers per se are unlikely to represent areas of primary differentiation.

A significant division was detected between (1) Central America and Colombia and (2) Amazonia and southern Brazil populations of A. darlingi using sequences of the COI gene (Mirabello and Conn, 2006a), and confirmed by nuclear gene sequences (Mirabello and Conn, unpublished data) that also detected three localities in Venezuela, Peru and Bolivia (Figure 4) where genotypes 1 and 2 were sympatric. However, this division does not specifically correspond to a prediction of any of the main hypotheses proposed to explain neotropical diversity. A demographic expansion was found in Amazonian and southern South America using mtDNA, and the time to expansion was estimated to be 16 237 years ago (95% confidence interval 5488–25 814; Mirabello and Conn, 2006b). The exactness of this estimation is questionable because it was calculated based on a Drosophila mutation rate, although the confidence interval range includes a large margin of error. The timing of this expansion is consistent with the premise of rainforest expansion in eastern and central Amazonia about 20 000 years ago (Burnham and Graham, 1999), and a main prediction of the refugia hypothesis (Table 1). A. darlingi populations in Amazonia have demonstrated a moderate to large amount of genetic variability and considerable phylogeographic structure (mtDNA; Freitas-Sibajev et al., 1995; Conn et al., 1999; Mirabello and Conn, 2006a; microsatellites; Conn et al., 2006), which contradicts the prediction of low genetic diversity and shallow phylogeographic structure of the refugia hypothesis. The amount of COI sequence divergence between A. darlingi genotypes 1 and 2 is 1.3% and the estimated divergence time is 0.57 MYA (Table 3). In summary, A. darlingi diverged into two genotypes during the Pleistocene, and a portion of the divergence may be accounted for by the refuge hypothesis (Table 2). Although we recommend the collection of additional data to formally test the riverine hypothesis, the habitat preference of A. darlingi makes it unlikely that this hypothesis will provide a satisfactory explanation.

Anopheles albitarsis s.l. is a neotropical species complex distributed throughout Latin America and on some Caribbean islands (Linthicum, 1988) that currently consists of five species (A. albitarsis s.s., A. albitarsis B, A. albitarsis F, A. deaneorum and A. marajoara; Wilkerson et al., 1995a, 1995b; Brochero et al., 2007) and one putative species (A. albitarsis E; Lehr et al., 2005). A. marajoara has been implicated as a regional malaria vector in savannah habitats (Rubio-Palis and Zimmerman, 1997) and in lowland rainforest in eastern Amazonian Brazil (Conn et al., 2002; Galardo et al., 2007), and A. albitarsis E is important in malaria transmission in savannah habitats in northern Roraima state, Brazil (Póvoa et al., 2006). A. deaneorum is quite competent in laboratory studies (Klein et al., 1991a, 1991b) and A. albitarsis F is suspected as a local or regional vector in Colombia (Brochero et al., 2007).

The species determinations for members of this complex are based on several markers, and include crossing experiments that support the specific status of A. albitarsis B and A. deaneorum (Klein et al., 1991c), and A. albitarsis s.s. and A. deaneorum (Lima et al., 2004). Studies of A. marajoara using chromosome inversions (Kreutzer et al., 1976) and allozymes and mtDNA restriction fragment length polymorphisms (Narang et al., 1993), both provide data consistent with multiple species. More recently, neither random amplified polymorphic DNA banding patterns nor ITS2 could discriminate between A. marajoara and A. albitarsis E (Wilkerson et al., 1995b; Li and Wilkerson, 2005), but sequences of the complete COI gene (Lehr et al., 2005) and a combination of 767 bp of the COI gene plus 909 bp fragment of the mtDNA NADH dehydrogenase subunit 5 (ND5) gene (Shaw et al., unpublished data) both determine A. marajoara and A. albitarsis E to be distinctive monophyletic clades. Based on Bayesian analysis, microsatellites also discriminate between the two (Conn et al., unpublished data).

The A. deaneorum and A. albitarsis B clades, seemingly restricted to the Brazilian Shield or to a transition area between the Shield and the lowlands (Figure 5), do conform to prediction 1 of the marine incursion hypothesis (Table 2). However, the A. albitarsis s.s. clade is completely to the south of the Brazilian Shield, the A. marajoara clade stretches across the Guiana and Brazilian Shields, and the A. albitarsis E clade appears to be restricted to savannah and Andean foothills habitat in Venezuela and Trinidad. The position of each of the five clades (species), excluding An. albitarsis F (not discovered until after the publication of Lehr et al. (2005) and only identified from one (Colombian) locality to date) is fairly stable among the three analyses undertaken (MP, maximum likelihood (ML), and Bayesian; Lehr et al., 2005) except for the basal clade. The second prediction of the marine incursion hypothesis suggests, for the A. albitarsis complex, that A. deaneorum and A. albitarsis B should be basal, and A. deaneorum is basal in the ML analysis, but A. albitarsis E is basal in the MP and Bayesian analyses. There were no samples collected from the western Amazonian lowlands so the third prediction could not be tested. Based on these data, ancestry for A. albitarsis s.l. is most likely in Venezuela, followed by dispersal south into the Amazon Basin and then onto the Brazilian Shield. Not all the MP, ML and Bayesian analyses (Lehr et al., 2005) agree on this dispersal order, but all found the most southern species, A. albitarsis s.s., to be the most derived clade, and sister to it is A. albitarsis B, the species found on the eastern side of the Brazilian Shield (Figure 5).

Figure 5
figure 5

Geographic locations of the five species of the A. albitarsis complex based on the mtDNA COI gene of Lehr et al. (2005) plus A. albitarsis F (Brochero et al., 2007) and additional mtDNA data on localities of A. albitarsis E (Shaw et al., in manuscript).

Full size image

The only comment relative to the riverine hypothesis is the observation that samples of A. marajoara from Salvaterra on eastern Marajó Island and those from Macapá in Amapá state across the northern arm of the Amazon at its mouth were not significantly differentiated with mtDNA sequences (Lehr et al., 2005) and, using a Bayesian analysis with microsatellite markers, the probability was extremely high that they were in the same population (Conn et al., unpublished data).

The refugia hypothesis can only be discussed in terms of A. marajoara and A. albitarsis E, for which sufficient data have been collected and analyzed. Both COI (Lehr, 2003) and ND5 (Shaw et al., unpublished data) sequences show strong support for a demographic expansion during the Pleistocene only in five localities in eastern Amazonian Brazil (north of the Amazon River), and low genetic variability for A. marajoara, supporting predictions 1 and 2 (Table 2). However, for samples of A. marajoara overall (including those from three Central Amazonian localities) there is considerable phylogeographic structure and deeper than expected mtDNA divergence. A. albitarsis E, in contrast, has the signature of an old stable population in equilibrium, with no compelling evidence for any of the four refugia hypothesis predictions (Shaw et al., unpublished data). The sequence divergence between A. marajoara and A. albitarsis E is also the highest, the only one estimated to be during the Pliocene, compared with the other members of this complex (Table 3). Taken together, data for A. albitarsis s.l. suggest an interaction between the marine incursion and refugia hypotheses, congruent with the estimated divergence time frame of Pliocene/Pleistocene.

Common patterns?

The similarities among L. longipalpis s.l., R. prolixus/R. robustus and A. albitarsis s.l. include ancestral divergence during the Plicoene (despite distinctive generation times), and predominantly reciprocal allopatric clades. The congruent sympatric clades in the east Andean Cordillera (Figures 2 and 3) within each of L. longipalpis s.l. and R. prolixus/R. robustus are possibly the result of a combination of continuous tectonic activity over a long period of geological history and limited dispersal. However, the triatomines strongly support marine incursion and refuge hypotheses (Table 2), whereas support for each of these hypotheses for L. longipalpis s.l. is only partial. The triatomines and sand flies also differ in ancestry: South American or Amazonian for R. robustus/R. prolixus vs western Venezuela for L. longipalpis s.l. A. albitarsis s.l. also consists of several reciprocal allopatric clades, shares likely ancestry in Venezuela with L. longipalpis s.l. and there are two regions of relative geographical congruence with L. longipalpis s.l. One is the Brazilian Shield for A. albitarsis B (Figure 5) and the L. longipalpis Brazilian clade (Figure 2) and the other is northwestern South America: the basal A. albitarsis E clade and the nearly-basal cis-Andean L. longipalpis clade. Despite very distinctive habitat preferences, A. darlingi genotype 2 shares evidence for a Pleistocene expansion in the Amazon with A. marajoara, but other aspects of its phylogeography differ considerably (Table 2). Perhaps in part because of its high gene flow, its divergence is not easily explained by the available hypotheses.

Conclusion

Because samples of vector insects tend to be collected in or near disease foci, a concerted effort needs to be made to collect more broadly in order to rigorously examine the three testable divergence hypotheses discussed, particularly within each of the three insect vector groups. Further investigations into the role of ancestral Venezuelan localities for each of the three insect groups may lay the foundation for new hypotheses, and could be useful in a better understanding of locations of disease foci and their potential expansions. Molecular dating techniques for insect vector groups could provide more insight into the influence of historical events on the phylogeographic patterns of variation.