A total of 255 specimens of Anopheles oswaldoi s.l. from 33 localities in five countries in the Neotropics (Brazil, Colombia, Ecuador, Peru and Trinidad and Tobago) were used in this study (see Table 1 for georeferenced locality data). Co-ordinates were converted to decimal degrees 26 and the distribution data is available in Mosquito Map (http://www.mosquitomap.org). Samples of Anopheles konderi s.l. were collected near Macapá, Amapá, Brazil and identified on the basis of the male aedeagus by MAMS. To avoid confusion, these are referred to herein as An. konderi of Sallum. All specimens used in this study were collected or provided by the authors of this study.

Sequences generated per country include: Brazil (27 ITS2, 57 COI), Colombia (154 ITS2, 27 COI), Ecuador (15 COI), Peru (1 ITS2, 3 COI) and Trinidad and Tobago (10 ITS2, 8 COI). H: ITS2 haplotype number.

DNA was extracted using the DNeasy® Blood and Tissue Kit (QIAgen®, USA) on the automated BioSprint 96® robotic platform. The ITS2 region was amplified for 192 samples from 21 localities using the published primers of Collins and Paskewitz 27 and following the protocol in Linton et al. 28 . DNA barcodes 29 were amplified from 110 individuals from 24 localities using the universal barcoding primers developed by Folmer et al. 30 and the protocol of the Mosquito Barcoding Initiative, expressly listed in Ruiz et al. 31 .

Sequencing reactions were carried out in both directions using the Big Dye Terminator Kit® on an ABI3770 automated sequencer (PE Applied BioSystems®, Warrington, England). Sequence chromatograms were edited using Sequencher™ v. 4.8 (Genes Codes Corporation®, Ann Arbor, MI). Sequences were aligned automatically in Mafft Pro 5.5.7 (http://www.geneious.com) or using MacClade v. 4.06 32 . Basic Local Alignment Search Tool (BLAST) searches (http://blast.ncbi.nlm.nih.gov) were carried out to correlate our sequences with those publicly available in GenBank.

To assess population-level genealogies, COI sequences were analyzed using TCS v. 1.21 33 . A connection limit of 95 % was adopted to investigate whether An. oswaldoi formed a single “meta-population” (reflected by a single network) or is comprised of separate species in South America (reflected by the formation of two or more independent networks).

A COI data matrix was generated in MEGA v. 5 34 . COI sequences were grouped according to the results of the TCS analysis, and intra- and inter-group genetic distances compared using Kimura’s 2-Parameter distance (K2P) algorithm 35 . Phylogenetic analysis was carried out on the separate and combined ITS2 and COI data sets. MrModeltest v. 2.3 36 was used to choose the best evolutionary model for these regions separately using the Akaike Information Criterion (AIC) search. A partitioned Bayesian analysis (by DNA region) was subsequently performed using MrBayes v. 3.1.2 37 , available online (http://cbsuapps.tc.cornell.edu/mrbayes.aspx).

The analysis in MrBayes ran for 10 million generations with two parallel searches using three heated and one cold Markov chain. The first 5 million generations were discarded as burn-in. Support for this Bayesian tree was conducted by generating a maximum parsimony (MP) tree in PAUP v. 4.0b10 38 for both ITS2 and COI, and bootstrapping 39 , using a heuristic search, simple stepwise addition, TBR branch swapping and 1000 bootstrap replicates.

Unique COI haplotypes were further analysed using MYC 40 41 for delimitation of COI clusters. This method optimizes a threshold age that corresponds to the shift from coalescent to species diversification 42 branching processes and calculates the number of resulting independent entities. The likelihood of the null model that all samples belong to a single species is compared to that of the alternative hypothesis where separate coalescent groups are nested within the species tree. Confidence limits correspond to threshold values ±2 logL units around the ML estimate. This analysis was conducted on Bayesian consensus trees and each tree was converted to ultrametric using penalized likelihood as implemented in r8s v.1.7 43 , with the optimal smoothing parameter selected by cross-validation of values between 0.01 and 1000. FigTree v. 1.2.1 44 was used to edit all trees generated.

ITS2 sequences were generated for 192 specimens of An. oswaldoi s.l. (Table 1). The overall alignment (544 bp, after primer trim) revealed 13 unique ITS2 haplotypes (GenBank: KC970065-77), labelled HI-HXIII (Table 2). The ITS2 sequences ranged in length from 530 bp in Haplotype I (H1) (Espírito Santo and São Paulo) to 540 bp in haplotypes HVI-IX from Mato Grosso in Brazil (Table 2). Thirty-two variable bases were noted (5.9 %), with the variation concentrated toward the second half (3′ end) of the ITS2 fragment between aligned nucleotides 207–494. Indels (insertions/deletions) were observed at bases 337–338, 344–345, 365–369, 408–409, 416–419, and 467–468 (Table 2).

Following primer trim, the ITS2 fragments varied in length from 530–540 bp. In the final 544-bp alignment, 32 sites were found to be variable. H1 includes specimens of the lectotype series of An. oswaldoi s.s 12.

Whereas in general all geographical areas were represented by unique ITS2 haplotypes, the 14 specimens of An. oswaldoi from Mato Grosso (Brazil) showed seven unique haplotypes (HII, HIV - HIX), varying in length between 538 bp and 540 bp (Table 2). Specimens from Putumayo, Colombia (n = 123) and Caquetá, Colombia (n = 1) shared the same ITS2 haplotype (HXII) and one specimen from Amazonas, Colombia shared the same haplotype with the single specimen sequenced from Amazonas, Brazil (HIII) (Table 2). Haplotype H1 comprises specimens from Espírito Santo and São Paulo, including those in the redescription and lectotype series of An. oswaldoi s.s. 12 .

The mtDNA COI barcode sequences (n = 110, 658 bp without primers) revealed 84 unique haplotypes in the 33 localities sampled (Table 1). The amino acid (AA) reading frame starts at the second base of the primer-edited sequences. Protein translations, carried out using the invertebrate mitochondrial code showed no stop codons, indicating that all sequences represent functional protein coding genes.

Meta-population analysis of the COI data set using TCS 33 clearly split the data set into four independent networks (Figure 1): An. oswaldoi s.s., An. oswaldoi A, An. oswaldoi B and An. sp. nr. konderi. This analysis is concordant with the ITS2 haplotypes as follow: An. oswaldoi s.s. 12 (haplotype HI), An. oswaldoi A (named herein) (HII-IX), An. oswaldoi B 18 (HX-XII) and An. sp. nr. konderi (named herein) (HXIII). Although high levels of genetic divergence (0.020) were noted in COI sequences between samples from northern Colombia and Trinidad and Tobago when compared to those from southern Colombia, the TCS analysis showed that these comprised a single species, An. oswaldoi B (Figure 1).

Pairwise COI sequence comparisons between the groups listed above (TCS network) revealed higher levels of COI divergence between them. Intra-specific divergence ranged from 0.007 in An. oswaldoi s.s. (n = 12), to 0.02 in An. oswaldoi B (n = 41). Inter-specific COI sequence divergences ranged from 0.038 in the most closely related species, An. oswaldoi s.s. and An. oswaldoi B, to 0.064 between An. oswaldoi A and An. oswaldoi B (Table 3).

Analyses of COI sequences separated 110 specimens of Anopheles oswaldoi s.l. into four groups using the TCS model as calculated by Kimura’s 2-Parameter distance model (K2P) 35.

Four clusters, corresponding to An. oswaldoi s.s. (Brazil), An. oswaldoi A (Brazil and Colombia), An. oswaldoi B (Colombia, Ecuador and Trinidad and Tobago) and An. sp. nr. konderi (Ecuador, Colombia and Peru), were delineated using the MYC model 41 (Figure 2). The MYC model uses only the unique COI haplotypes, which detects the transition between species to within-population branching patterns. This analysis also lends support for the separate species status of An. konderi of Sallum from Macapá, Amapá, Brazil (GenBank: KF305833), and its close relationship with An. oswaldoi s.l.

The combined analysis of both COI and ITS2 sequences using MrBayes 36 , with the HKY+I+G (Hasegawa-Kishino-Yano + Invariant Sites + Gamma) model, again strongly confirmed that An. oswaldoi s.l. is a complex of at least four species (Figure 3). The posterior probability had a value of one, for three clades (An. oswaldoi A, An. oswaldoi B and An. sp. nr. konderi) and 0.6 in An. oswaldoi s.s. however, bootstrap values fully confirm these four species (An. oswaldoi s.s., An. oswaldoi A, An. oswaldoi B and An. sp. nr. konderi) (Figure 3). The COI sequence of An. konderi of Sallum was found to cluster closely with An. sp. nr. konderi, but these two species are consistently different in four of their ITS2 bases as follows: one transition in position 361 (G/A), two indels (364 (A/-) and 365 (T/-)), and one transversion (449 T/C) (data not shown).

Results of the TCS, MYC model and Bayesian analyses using the ITS2 and COI data sets provides further strong support for the following species: An. oswaldoi s.s. from Brazil (Espírito Santo = type locality), Rondônia, Rio de Janeiro and São Paulo); An. oswaldoi A from Brazil (Mato Grosso, Rondônia, Amazonas) and Colombia (Amazonas); An. oswaldoi B from Colombia (Antioquia, Caquetá, Meta, Norte de Santander, Putumayo), Ecuador (Orellana) and Trinidad and Tobago (Valencia); and An. sp. nr. konderi from Colombia (Caquetá), Ecuador (Orellana) and Peru (Madre de Dios, Loreto) (Figures 1, 2, 3 and 4).

Based on mitochondrial COI and nuclear ITS2 sequences, this study presents clear molecular evidence that the Oswaldoi-Konderi Complex comprises at least five species: two previously determined (An. oswaldoi s.s. and An. oswaldoi B), as well as An. oswaldoi A and An. sp. nr. konderi, and An. konderi of Sallum defined herein for the first time. Phylogenetic reconstruction of COI and ITS2 sequences (both independently and concatenated) verified four distinct species, which concur with the MYC and TCS network models. Construction of genealogical networks using TCS are often used to infer evolutionary relationships at the population level; however, identification of independent networks that fall beyond the 95% parsimony connect limit are considered putative biological species 45 . At this threshold, four independent networks were generated corresponding to the four species listed above (Figure 1). In addition samples of An. konderi of Sallum, initially included as an outgroup in this study, not only represented a separate taxon but also fell firmly within the Oswaldoi Complex in phylogenetic treatments. This supports that An. oswaldoi and An. konderi comprise species complexes of very closely related taxa, which would benefit from being assessed concurrently.

Following this robust molecular delimitation, efforts were made to retrospectively identify samples documented in previously published studies, and relate these to geographical distributions and their regional malaria vector status.

Prior to this study, the distribution of “true” An. oswaldoi s.s. had only been determined from Espírito Santo and São Paulo, Brazil 12 . Herein COI data also confirms its presence in the Brazilian States of Rio de Janeiro, Acre (GenBank: DQ784832-35) and Coari (GenBank: DQ784836-38) 15 (Table 4). Ruiz et al. 8 31 suggested a genetic distance threshold of 0.02 for species delimitation in Nyssorhynchus species. Despite the large geographic distance, our An. oswaldoi s.s. from São Paulo and Espírito Santo, Brazil (type locality, south of Brazil) grouped with Scarpassa and Conn’s COI sequences 16 from Rondônia State, in the north of Brazil (overall mean K2P distance of 0.010, range 0.002 to 0.015). Given its current distribution, it seems highly likely that An. oswaldoi s.s. could also be present in regions between these Brazilian states, including Mato Grosso, Goias and Minas Gerais, and perhaps even south into Bolivia or Paraguay.

[osw s.s.: An. oswaldoi s.s.; osw A: An. oswaldoi A; sp. nr. kon: An. sp. nr. konderi; kon: An. konderi of Sallum from Macapá, Amapá, Brazil].

There are no records to suggest that An. oswaldoi s.s. is a vector in the eastern Brazilian provinces of Espírito Santo, Rio de Janeiro or São Paulo. In the north-western Brazilian state of Rondônia, where An. oswaldoi s.s., An. oswaldoi A and An. sp. nr. konderi are present, de Oliveira-Ferreira et al. 46 and Klein et al. 47 48 reported low levels of Plasmodium infections in what can only be assumed to include a mix of oswaldoi-konderi lineages.

In this study, three ITS2 haplotypes [HX, HXI and HXII (Table 2)] were found to be identical to those reported as An. oswaldoi B from Putumayo, Colombia by Ruiz et al. 17 31 . These haplotypes shared 99% identity to published sequences from Ocamo, Venezuela (GenBank: AF055071) and Amapá, Brazil (GenBank: AF056318) (= Group II of Marrelli et al. 14 ). Our analyses of both individual and combined COI and ITS2 data sets, grouped the above sequences into a single phylogenetic cluster and TCS network, confirming An. oswaldoi B as a separate species in the An. oswaldoi complex. This study increases the known distribution of An. oswaldoi B to include the Departments of Antioquia, Caquetá and Norte de Santander (Colombia), and the Province of Orellana (Ecuador) and Saint Andrew/Saint David (Trinidad and Tobago) (Figure 4).

Plasmodium vivax infection was detected in one molecularly confirmed An. oswaldoi B specimen (of 361 positive mosquitoes) collected in Putumayo, Colombia 13 . However An. oswaldoi s.l. has also been incriminated in Amapá, Brazil 49 , as a secondary vector in Venezuela 50 and has been experimentally infected with P. vivax in Trinidad and Tobago 51 , where only An. oswaldoi B has been identified to date. Correlation of these studies with the known distribution of An. oswaldoi B suggest the potential involvement of this species on malaria transmission is over a much wider region of northern and northeastern South America than originally realised.

Currently, the only valid synonym of An. oswaldoi is An. aquacaelestis Curry, originally described from Panama 52 . The original description differentiates An. aquacaelestis from An. aquasalis, and suggests that this Panamanian species is identical to the species present in Trinidad and Tobago. As An. oswaldoi B is the only species of the An. oswaldoi complex that has been detected in Trinidad and Tobago, it would be prudent to carefully study the type series of An. aquacaelestis, to ensure that our informally designated “An. oswaldoi B” is not in fact An. aquacaelestis, prior to assigning a new name.

Comparison of our An. oswaldoi A ITS2 sequences (as haplotypes HII-IX; Table 2) revealed high similarity with three in Marrelli et al. 4 and two in Sallum et al. 9 . Our Haplotype III from Tefé, Amazonas, Brazil and Amazon, Colombia, shares 99.8% identity with GenBank AF056317 from Amazonas, Brazil (= Group I of Marrelli et al. 14 ) and varies by only a single base indel (−/T) at position 166. Our ITS2 haplotype HII from Mato Grosso, Brazil, is 99.4% identical to GenBank AF055068 from Acre, Brazil (= Group I of Marrelli et al. 14 ) with base substitutions at positions 349 (C/A) and 427 (G/A). Our ITS2 haplotype HVI from Mato Grosso, Brazil, is 98.6% similar to GenBank AF055069 from Rondônia, Brazil (= Group I of Marrelli et al. 14 ), and differs at only one base (position 370; G/A) from EU636802 9 . Sallum et al. 9 described morphological differences in the male genitalia and ITS2 sequences of An. oswaldoi s.l. from Acrelândia, Acre, Brazil, compared to An. oswaldoi s.s. 12 . Our ITS2 haplotype HIV from Mato Grosso is identical to GenBank EU636809 from Acrelândia, Acre. These results clearly show that the unnamed species of An. oswaldoi s.l. described by Marrelli et al. 14 and Sallum et al. 9 corresponds to our An. oswaldoi A.

The COI sequences generated by Scarpassa and Conn 16 overlapped our barcode region by 404 bp. Comparison of their trimmed sequences with our data, revealed sequences from their “Group 3” specimens (GenBank DQ784849-51) from Moju, Pará, Brazil were 99% identical to those of our An. oswaldoi A, thus confirming that An. oswaldoi A is also present in Pará State (Table 4).

Drawing from our data and correlation with other published works 9 14 16 , the known distribution of An. oswaldoi A now includes the Brazilian States of Acre, Amazonas, Mato Grosso, Pará and Rondônia, and the Colombian Department of Amazonas (Figure 4). Vector incrimination studies in Acre, Brazil determined An. oswaldoi as a local vector of malaria 53 54 55 56 . As the only confirmed species of the Oswaldoi Complex present in Acre, Brazil, it seems likely that An. oswaldoi A plays a role in malaria transmission in this state, and most probably throughout its distribution.

ITS2 haplotype HXIII was shared by a number of individuals from Madre de Dios and Loreto provinces in Peru, Orellana province in Ecuador, and in the Department of Caquetá, southern Colombia (Figure 4). These sequences showed high similarity (98-100%) to specimens identified as An. konderi s.l. from Amapá, Brazil [GenBank: JF437934-36, JF437926] 23 , differing by only two transitions (G/A, position 361 and T/C, position 449) and a two base indels (at positions 264–265). We recognize that such intragenetic variation in ITS2 is not uncommon, having also been detected in An. oswaldoi A, An. oswaldoi B and other Nyssorhynchus species 9 57 58 59 . However, COI sequences from the same individuals analysed by Bayesian, MYC model and TCS network analyses confirm an independent sister cluster to the An. konderi of Sallum from Macapá, Amapá, Brazil, also included in our analysis. Both these taxa are unique with respect to a third An. konderi s.l. from Acre reported by Sallum et al. 9 (GenBank: EU636801).

The type locality of An. konderi is Solimões River, Coari, Amazonas, Brazil. However, in their recent redescription of An. konderi, Flores-Mendoza et al. 21 were unable to access material from this locality, using several other localities in Brazil instead. Some samples used in our molecular analysis were the same as those examined by Flores-Mendoza et al. 21 , but as the identity of An. konderi is uncertain, we have designated this cluster as An. sp. nr. konderi.

The trimmed 404 bp An. oswaldoi “Group 2” COI sequences of Scarpassa and Conn 16 (GenBank: DQ784827-31 and DQ784843-44) are 98-99% similar to our An. sp. nr. konderi and 100% identical to GenBank: JF437965 16 (Table 4). Correlation of our data with published works confirms the presence of An. sp. nr. konderi in three localities in Brazil (Acrelândia and Sena Madureira (Acre) and São Miguel (Rondônia)), and our data suggest that An. sp. nr. konderi may be allopatric in Madre de Dios and Loreto provinces in Peru, where An. oswaldoi s.l. has been confirmed as an efficient malaria vector 60 61 62 .

The taxonomic status of An. konderi s.l. clearly needs to be reassessed. This and other studies 9 16 23 have provided evidence for at least three species in An. konderi s.l. Concerted efforts are needed to acquire material from the type locality in order to establish the identity of An. konderi s.s. before the systematics of closely related taxa can be properly understood. Given the data presented here, it is likely that the species discussed here will fall into a more comprehensive Oswaldoi-Konderi Complex in future.