Background

Rhodnius prolixus (Triatominae) is a primary vector of Chagas disease across northern South America, where infestation of rural houses by this species is common [1, 2]; it belongs to a group of closely-related taxa with nearly identical morphologies, i.e. the ‘R. prolixus–R. robustus cryptic–species complex’ [3,4,5,6]. Except for R. prolixus, the species in this complex (R. robustus (s.l.) [5, 6], R. montenegrensis [7] and R. marabaensis [8]) do not infest houses and have relatively little medical relevance [4, 9,10,11]. Why just one species within this group of close-kin bugs has the ability to stably infest houses is still unclear [6, 9, 11]. At least in part, this knowledge gap stems from the taxonomic uncertainty inherent to studying cryptic taxa. Thus, even though molecular systematics has contributed substantially to clarify the composition of the R. prolixus–R. robustus complex and the relations among its members, some controversies remain [3,4,5,6]. Using R. montenegrensis as a case-study, here we describe an approach that combines public-domain genetic/transcriptomic data and bioinformatics to address such controversies.

Rhodnius montenegrensis was described in 2012 by researchers of the Universidade Estadual Paulista ‘Júlio de Mesquita Filho’ (UNESP), Brazil, based on bugs resembling R. robustus [7]. The material used in this description came from UNESP colony ‘CTA 88’, which was founded with eight bugs collected in 2008 from Attalea palms in the southwestern Brazilian Amazon [7]. These bugs were compared to benchmark material identified as R. robustus from four laboratory colonies (‘CTA 83’ to ‘CTA 86’) founded with bugs collected in Peru and kept at UNESP since the early 1970s [7]. The original description included a 369-bp DNA sequence of R. montenegrensis’ mitochondrial cytochrome b (cytb) gene (GenBank ID: KR072682.1); a phylogenetic analysis recovered KR072682.1 as very closely related to a sequence of undisclosed origin or GenBank ID but labeled as ‘R. robustus’ (see figure 15 in [7]). In addition, the endonuclease BstUI did not cleave nuclear rDNA 5.8S/ITS2 amplicons from putative R. montenegrensis, but cleaved at one site amplicons from the benchmark R. robustus colony bugs, thus implying that the 5.8S/ITS2 sequences of R. montenegrensis and R. robustus differ by at least one base at the enzyme’s restriction site [7]. The authors concluded that, although closely related, R. montenegrensis and R. robustus are morphologically and genetically distinct [7]. These findings received further support from a comparative-transcriptomics study showing that bugs identified as R. montenegrensis and bugs from UNESP’s R. robustus colony ‘CTA 85’, had “… a substantial quantity of fixed interspecific polymorphisms …”; this was interpreted as “… suggest[ing] a high degree of genetic divergence between the two species [that] likely corroborates the species status of R. montenegrensis” ([12], Abstract; see also [13] for details).

The striking similarity of R. montenegrensis and R. robustus cytb sequences was recently confirmed by a broader analysis [6] showing that R. montenegrensis’ KR072682.1 is nearly identical to cytb sequences from R. robustus genotype II, one of the R. robustus cryptic taxa identified by Monteiro et al. [5] in the early 2000s. While these results clearly suggest that R. montenegrensis is “… part of the variability of R. robustus II …” (caption of figure 2A in [6]), they raise the question of why morphology [7] and transcriptomics [12, 13] both discriminate R. montenegrensis from R. robustus bugs of Peruvian origin. This is even more intriguing when one considers that R. robustus II is the only R. robustus lineage known to occur in western-southwestern Amazonia [5, 6, 14,15,16]; R. robustus material from Peru, then, is expected to belong in genotype II and, hence, to be indistinguishable from R. montenegrensis.

Recently, Monteiro and colleagues [6] suggested a possible explanation for these apparently contradictory findings. They observed (i) that the members of the R. prolixus–R. robustus species complex all have virtually identical phenotypes [1,2,3,4]; (ii) that several species-pairs within the complex are inter-fertile [17]; (iii) that there is evidence that many laboratory colonies of bugs identified as either R. prolixus or R. robustus (s.l.) are mixed/contaminated or wrongly labeled (see SI Appendix of [18]); and (iv) that cytb sequences of bugs from colonies labeled as ‘R. robustus’ from ‘Lima, Peru’ match R. prolixus sequences [19]. These observations suggest that the Peruvian R. robustus colonies kept at UNESP may have become contaminated with non-R. robustus material, with the main suspect being R. prolixus [6]. This hypothesis predicts that bugs drawn from the Peruvian ‘R. robustus’ colonies at UNESP will have R. prolixus genetic material, perhaps mixed to some degree with R. robustus (likely genotype II). In contrast, bugs from the younger (and hence less likely to have become contaminated) R. montenegrensis colonies should be genetically indistinguishable from R. robustus II. Here, we use publicly-available transcriptome data derived from UNESP Rhodnius spp. colonies to test these two predictions. More generally, we present a methodological approach (Fig. 1) that leverages public-domain information from traditional and next-generation sequencing projects to investigate the molecular systematics of cryptic species in the face of taxonomic confusion.

Fig. 1
figure 1

Transcriptome-based molecular systematics: a diagrammatic representation of our methodological approach. We determined 10 ITS2 query sequences and deposited them in NCBI’s (National Center for Biotechnology Information) GenBank (broken grey arrow). Other query and outgroup sequences were retrieved from GenBank, and transcriptome read archives (SRAs) from NCBI’s Sequence Read Archive and the European Nucleotide Archive (ENA). The right panel (broken-line box) illustrates some details of the process of generating consensus sequences from raw transcriptome reads plus query sequences and then computing breadth- and depth-coverage (pale-blue histograms behind the consensus sequence). The dotted-line vertical boxes highlight two variable sites: one in which the consensus sequence differs from the query (T/A), even though a minority of reads have T as in the query; and another in which a rare C variant appears in a minority of aligned reads and the consensus sequence, therefore, retains G as in the query

Full size image

Methods

Transcriptome queries

We used selected DNA sequences from members of the R. prolixus–R. robustus cryptic–species complex to query publicly-available transcriptome sequence read archives (SRAs) derived from UNESP colonies, namely the R. montenegrensis colony and the Peruvian ‘R. robustus’ colonies used as the taxonomic benchmark in the description of R. montenegrensis [7] and in later transcriptome-based comparisons [12, 13] (Table 1). We chose three loci that have been widely used to study the systematics and evolution of Rhodnius spp. and other triatomines: the mitochondrial cytb [4,5,6, 11, 14, 16, 19] plus the nuclear rDNA ITS2 [6, 20] and D2-28S [5]. Cytb and D2-28S sequences were retrieved from GenBank, and our ten ITS2 query sequences (R. prolixus and R. robustus I to IV; GenBank: MK411269–MK411278) were determined using previously described primers and protocols [20]. Figure 1 presents a summary of our methods, and Additional file 1: Tables S1 and S2 provide details on query sequences including GenBank ID codes.

Table 1 Transcriptome raw data (sequence read archives) used in this study
Full size table

Figure 1 shows an outline of our methodological approach. We downloaded six public-domain transcriptome SRAs determined by two independent groups at UNESP and the French Centre National de la Recherche Scientifique (CNRS), respectively (see Table 1 and [12, 13, 21]). We aligned transcriptome paired-end reads to our query sequences using the Burrows-Wheeler aligner [22] with the BWA-MEM algorithm and default parameter values except for − t = 10, − B = 5, − O = 7.7, − L = 6.6, and − U = 18. We stored aligned reads in Binary Alignment Map (BAM) format, removed PCR duplicates with SAMtools markdup [23], and used SAMtools mpileup and BCFtools [23] to generate a consensus sequence for each alignment. We then exported consensus sequences in fasta format using UGENE [24], and computed sequence breadth-coverage and site-specific read depth-coverage using BEDTools [25]. We retained for further analyses consensus sequences with full-breadth-coverage of the query sequence (Fig. 1).

As a quality check for our focal results, we queried SRAs with (i) cytb and D2-28S sequences from R. neglectus and R. nasutus, which are relatively close kin to the members of the R. prolixus–R. robustus species complex [6]; and (ii) sequences of a putative nuclear intron (AmpG) from R. prolixus and R. robustus I–IV [15]. Our expectation was that these complementary queries would not generate any full-breadth-coverage consensus sequence.

Sequence-polymorphism and phylogenetic analyses

We used MAFFT v.7 [26] to align, for each locus, query sequences, full-breadth-coverage consensus sequences and selected outgroup GenBank sequences (R. barretti JX273159.1 for cytb, R. stali AJ286890.2 for ITS2 and R. nasutus AF435856.1 for D2-28S); we made some manual adjustments to the ITS2 and D2-28S alignments (Additional file 2: Alignments S1–S3). Locus-specific alignments were first analyzed in terms of sequence polymorphism using MEGA X [27]. We computed a set of basic sequence diversity and similarity metrics (Additional file 1: Table S1), and examined segregating sites in detail to identify those that might be informative in the context of our research question – e.g., species-diagnostic character states and missense or nonsense mutations in the protein-coding cytb sequences. We used the bootstrap (1000 pseudo-replicates) to provide estimates of uncertainty for sequence diversity/similarity metrics. We then evaluated phylogenetic relations among the sequences in each alignment using the Bayesian approach implemented in BEAST v.1.10.4 [28]. For each locus, we completed four independent runs with Yule priors for 1,000,000 generations, sampling every 1000 steps and with a 25% burn-in. We then used FigTree v.1.4.4 (http://tree.bio.ed.ac.uk/software/figtree) to build maximum-credibility trees with the posterior probability cut-off set at 0.5. We assessed clade support based on posterior probabilities. We also estimated maximum-likelihood (ML) and maximum-parsimony (MP) trees in MEGA X [27]; the methods and results of these complementary analyses can be found in Additional file 3: Figure S1, Additional file 4: Figure S2, Additional file 5: Figure S3 and in the captions thereof. We finally examined the results of our sequence-polymorphism and phylogenetic analyses in the light of sequence depth-coverage, measured as the number of reads that aligned at each nucleotide position (see Fig. 1). In particular, full-breath-coverage consensus sequences with mean depth-coverage < 10 reads/position were regarded as unreliable [29], and consensus sequences with mean depth-coverage ≥ 10 reads/position, but with ≥ 15% of positions supported by < 10 reads, as dubious (see Additional file 1: Table S1 and Additional file 6: Figure S4).

Results

General results

We aligned sequence reads from six transcriptomes to query sequences representative of all (cytb) or all but one (ITS2 and D2-28S) known members of the R. prolixus–R. robustus cryptic–species complex; ITS2 and D2-28S sequences of the little-known R. robustus V are so far unavailable. Overall, our query dataset comprised 47 sequences (Additional file 1: Table S1). Using fairly stringent alignment/filter settings, we generated 61 full-breadth-coverage and 198 partial-breadth-coverage consensus sequences; no base aligned to the query sequence in 23 of our 282 queries (Additional file 1: Table S1). Depth-coverage was usually high for queries yielding full-breadth-coverage, but varied substantially across SRAs (Figs. 2, 3, 4, Tables 3, 4; see Additional file 1: Table S1 for breadth- and depth-coverage summary statistics across all queries, and Additional file 6: Figure S4 for full-breadth-coverage consensus sequences in which depth-coverage was < 10 reads at one or more positions.

Fig. 2
figure 2

Alignment of Rhodnius montenegrensis transcriptome reads to Rhodnius spp. query sequences. The graphs present, for each of three loci (mitochondrial cytb plus nuclear rDNA ITS2, and D2-28S), breath-coverage (% of positions; bold vertical lines; left vertical axis) and mean depth-coverage (reads/position; bars; right vertical axis) for queries against two R. montenegrensis transcriptomes (SRX1996481 and SRX1996482; [7, 12, 13]). Red lines/bars highlight queries yielding full-breadth-coverage; queries yielding only partial-breadth-coverage are black/grey. Abbreviation: UNESP, Universidade Estadual Paulista ‘Júlio de Mesquita Filho’, Brazil

Full size image
Fig. 3
figure 3

Alignment of Rhodnius robustus transcriptome reads to Rhodnius spp. cytochrome b sequences. Transcriptome reads were retrieved from the benchmark Rhodnius robustus stocks used in R. montenegrensis’ description and in later transcriptome-based comparisons. The graphs present breath-coverage (% of positions; bold vertical lines; left vertical axis) and mean depth-coverage (reads/position; bars; right vertical axis) for queries against four R. robustus transcriptomes (SRX1996483, SRX1996484, ERX1387159 and ERX1387160; [7, 12, 13, 21]). Red lines/bars highlight queries yielding full-breadth-coverage; queries yielding only partial-breadth-coverage are black/grey. Abbreviations: UNESP, Universidade Estadual Paulista ‘Júlio de Mesquita Filho’, Brazil; CNRS, Centre National de la Recherche Scientifique, France

Full size image
Fig. 4
figure 4

Alignment of Rhodnius robustus transcriptome reads to Rhodnius spp. ribosomal DNA sequences. Transcriptome reads were retrieved from the benchmark Rhodnius robustus stocks used in R. montenegrensis’ description and in later transcriptome-based comparisons; query sequences include two nuclear rDNA loci: ITS2 (upper row) and D2-28S (lower row). The graphs present breath-coverage (% of positions; bold vertical lines; left vertical axis) and mean depth-coverage (reads/position; bars; right vertical axis) for queries against four R. robustus transcriptomes (SRX1996483, SRX1996484, ERX1387159 and ERX1387160; [7, 12, 13, 21]). Red lines/bars highlight queries yielding full-breadth-coverage; queries yielding only partial-breadth-coverage are black/grey. Abbreviations: UNESP, Universidade Estadual Paulista ‘Júlio de Mesquita Filho’, Brazil; CNRS, Centre National de la Recherche Scientifique, France

Full size image

Mitochondrial cytochrome b

The 369-bp R. montenegrensis cytb sequence reported in the species’ description (KR072682.1) [7] differs at just one to four bases from those of bugs identified as R. robustus II collected in Rondônia, Brazil (including Monte Negro, the type locality of R. montenegrensis), and at five bases from that of a bug collected in Napo, Ecuador [5, 14, 30] (Table 2). We note that all Rhodnius spp. cytb sequences known to us have A at site 280 in our 663-bp alignment; the only exception seems to be EF071583.1 (R. robustus II from Rondônia [30]), which has G, a mutation that yields an Asparagine/Glycine change in the predicted protein (Additional file 2: Alignment S1). We strongly suspect that this first-codon position ‘mutation’ is a base-call error in EF071583.1. In support of this suspicion, we found that depth-coverage was lowest around position 280 when we aligned R. montenegrensis reads (SRAs SRX1996481 and SRX1996482) to query EF071583.1, with values of just 1.4–1.5% of the mean depth values (Table 3).

Table 2 Negligible mitochondrial cytochrome b sequence divergence between Rhodnius montenegrensis and R. robustus genotype II
Full size table
Table 3 Full-breadth-coverage consensus sequences generated from transcriptomes determined using UNESP Rhodnius montenegrensis colonies
Full size table

When we used R. montenegrensis’ KR072682.1 as the query sequence, R. montenegrensis SRAs yielded consensus sequences with full-breadth-coverage and substantial depth-coverage (Table 3, Fig. 2, Additional file 1: Table S1). This was also the case, however, when we aligned the same two SRAs to our three R. robustus II cytb query sequences, with, in addition, consistently improved depth-coverage (Table 3, Fig. 2, Additional file 1: Table S1). Reads from the R. montenegrensis SRAs aligned with full-breadth-coverage and substantial depth-coverage to query EF011724.1 (R. robustus II from Monte Negro [14]) (Table 3, Fig. 2, Additional file 1: Table S1). These two consensus sequences were identical, and both differed at a single, second-codon position (A/T, with depth-coverage > 11,000 reads) from the query sequence. Because all Rhodnius spp. cytb sequences we are aware of, except for R. pictipes, have T at this position (623 in our 663-bp alignment; Additional file 2: Alignment S1), we also suspect a base-call error in EF011724.1. This query sequence, as well as the consensus sequences it generated from the two R. montenegrensis SRAs, differed at three third-codon positions from the KR072682.1 sequence used to support R. montenegrensis as a valid species [7] (see Additional file 2: Alignment S1). Figure 3 shows that transcriptome reads from UNESP ‘R. robustus’ colonies aligned to R. prolixus query sequences with the highest breadth- and depth-coverage.

All the sequences in our cytb alignment, including consensus sequences generated from SRAs, comprise an open reading frame (Additional file 2: Alignment S1); this suggests that pseudogene sequences were absent from the dataset. Figure 5 shows the Bayesian cytb phylogenetic tree. This analysis recovered a well-supported clade including (i) R. montenegrensis’ original sequence (red in Fig. 5); (ii) all previously determined R. robustus II sequences; and (iii) full-breadth-coverage consensus sequences generated using R. montenegrensis or R. robustus II query sequences, irrespective of whether the SRAs were determined from UNESP colonies identified as R. montenegrensis or R. robustus (Fig. 5). ML and MP analyses also recovered this clade with moderate to high support (Fig. 5 and Additional file 3: Figure S1). These cytb trees also show that R. prolixus query sequences and the full-breadth-coverage consensus sequences they generated from ‘R. robustus’ transcriptome reads were almost identical; overall nucleotide diversity was π = 0.0020 ± 0.0011 SE, with maximum divergence of just ~ 0.3% (Table 4). Finally, the alignment of ‘R. robustus’ reads to the R. robustus II query AF421341.1 yielded two full-breadth-coverage consensus sequences that were identical to the query (Figs. 3, 5; Table 4).

Fig. 5
figure 5

Mitochondrial cytb Bayesian phylogenetic tree of members of the Rhodnius prolixus–R. robustus cryptic–species complex. The tree was inferred using query sequences (black font) and the consensus sequences generated from R. montenegrensis (black/blue font) and R. robustus (black/green font) transcriptome-read archives (for which taxon labels include NCBI Sequence Read Archive codes). Rhodnius montenegrensis’ original sequence (KR072682.1; [7]) is highlighted in red font. Two full-breadth-coverage consensus sequences had mean depth-coverage < 10 reads/position and were therefore excluded from these analyses (see Table 4 and Additional file 6: Figure S4). Numbers at nodes are posterior probabilities from Bayesian analysis (BA) and bootstrap support values (1000 pseudo-replicates) for maximum-likelihood (ML) and maximum-parsimony (MP) topologies (see Additional file 3: Figure S1). The scale-bar indicates substitutions per site (from BA)

Full size image
Table 4 Full-breadth-coverage consensus sequences generated from transcriptomes determined using UNESP ‘Rhodnius robustus’ colonies
Full size table

From these cytb sequence-data analyses we conclude that R. montenegrensis is, in all likelihood, the R. robustus cryptic taxon dubbed ‘R. robustus II’ by Monteiro et al. [5] in 2003. In addition, we found evidence strongly suggesting that at least some of the UNESP ‘R. robustus’ colonies used as the taxonomic benchmark to infer that R. montenegrensis is distinct from R. robustus [7, 12, 13] contain large amounts of R. prolixus mitochondrial DNA, and perhaps also smaller amounts of mitochondrial DNA from R. robustus II (i.e. R. montenegrensis) (Figs. 3, 5, Table 4).

Nuclear ribosomal ITS2

We generated 14 full-length consensus sequences from five of the six SRAs we queried with our 10 ITS2 Rhodnius spp. sequences (Tables 3, 4; Additional file 1: Tables S1, S2). Depth-coverage values were overall substantially smaller for ITS2 than for cytb transcriptome queries (Figs. 2, 3, 4 and Tables 3, 4; see also Additional file 1: Table S1 and Additional file 6: Figure S4). We generated two full-breadth-coverage consensus sequences when we aligned R. montenegrensis SRAs to ITS2 sequences from R. robustus II collected in Rondônia, Brazil (Fig. 2, Table 3). We identified six variable sites (π = 0.005 ± 0.002 SE) and three indels when we compared these four R. robustus II/R. montenegrensis sequences (Additional file 2: Alignment S2). The MK411275 (R. robustus II from Rondônia) vs SRX1996482 query generated a 99.86% breadth-coverage (732 out of 733 bp) consensus sequence with mean depth-coverage of 28.06 reads (Additional file 1: Table S1). In addition, queries performed with our four R. prolixus ITS2 sequences generated full-breadth-coverage consensus sequences in three of the four SRAs derived from UNESP ‘R. robustus’ colonies, although depth-coverage was low for those generated from SRX1996483 (Fig. 4, Table 4, Additional file 6: Figure S4). We found just two variable sites (π = 0.0007 ± 0.0006 SE) and no indels in the comparison of this subset of R. prolixus and ‘R. robustus’ query and consensus sequences (Additional file 2: Alignment S2).

Phylogenetic analyses recovered consistent ITS2 tree topologies (Fig. 6, Additional file 4: Figure S2). Full-breadth-coverage consensus sequences from R. montenegrensis SRAs and the R. robustus II query sequences used to generate them clustered together in a moderately- to well-supported clade. Support was higher for the clustering of R. prolixus query sequences with the full-breadth-coverage consensus sequences they generated from three ‘R. robustus’ SRAs (Fig. 6, Additional file 4: Figure S2).

Fig. 6
figure 6

Nuclear ribosomal ITS2 Bayesian phylogenetic tree of members of the Rhodnius prolixus–R. robustus cryptic–species complex. The tree was inferred using query sequences [black font; new GenBank ID codes for ITS2 sequences generated for this study (MK411269–MK411278) are presented in parentheses] and the consensus sequences generated from R. montenegrensis (black/blue font) and R. robustus (black/green font) transcriptome-read archives (for which taxon labels include the query names and the NCBI Sequence Read Archive codes). Numbers at nodes posterior probabilities from Bayesian analysis (BA) and bootstrap support values (1000 pseudo-replicates) for maximum-likelihood (ML) and maximum-parsimony (MP) topologies (see Additional file 4: Figure S2). The scale-bar indicates substitutions per site (from BA)

Full size image

We finally recall that, in the description of R. montenegrensis, da Rosa et al. [7] reported that BstUI cleaved ‘R. robustus’ ITS2 at one site, but did not cleave R. montenegrensis ITS2 amplicons; here, we found that BstUI’s restriction site (CGCG) is absent from the ITS2 sequences of R. montenegrensis and R. robustus II, III and IV, but present (sites 66–69 of our alignment; Additional file 2: Alignment S2) in those of R. robustus I and R. prolixus, including the latter species’ genome [18], with, e.g. query sequence MK411269 (R. prolixus from Guatemala; Additional file 1: Table S2) yielding a 100%-identity BLASTn hit in RproC3 assembly contig ACPB03046858.1 at VectorBase (https://www.vectorbase.org/). Thus, the BstUI digestion results in [7] suggest that bugs from UNESP ‘R. robustus’ colonies have nuclear rDNA similar to that expected in bugs of the R. prolixus–R. robustus I clade.

Taken together, these ITS2 data lend clear support to the cytb-based findings described above. In particular, they (i) confirm that R. montenegrensis is genetically indistinguishable from R. robustus II, and (ii) further suggest that the UNESP ‘R. robustus’ colonies used as the taxonomic benchmark to conclude that R. montenegrensis is distinct from R. robustus [7, 12, 13] are mainly or fully composed of bugs with R. prolixus DNA, with no detectable R. robustus nuclear ITS2 sequences.

Nuclear ribosomal D2-28S

Our D2-28S R. prolixus–R. robustus complex query sequences generated 21 full-breadth-coverage sequences from the six transcriptome SRAs; depth-coverage was overall substantial (typically > 250 reads/position on average) except for queries against SRA ERX1387160 (Figs. 2, 4; Tables 3, 4). Depth-coverage was, however, low over certain stretches of some full-breadth-coverage consensus sequences (Tables 3, 4; Additional file 6: Figure S4). When we used R. robustus IV’s AF435859.1 as the query, for example, mean depth-coverage was high for the consensus sequences generated from R. montenegrensis’ SRAs, yet depth-coverage fell to < 10 reads/position between positions ~ 260 and ~ 470 (Table 3, Additional file 6: Figure S4). Similarly, the AF435859.1 vs ERX1387160 query generated a full-breadth-coverage consensus sequence with 217 positions (34.3%) supported by < 10 reads each (Table 4, Additional file 6: Figure S4). We therefore regard the consensus sequences generated using our R. robustus IV query as dubious. Queries using the R. robustus III sequence generated full-breadth-coverage sequences from the two R. montenegrensis SRAs (with substantial depth-coverage; Fig. 2) and from one ‘R. robustus’ SRA (Fig. 4), for which depth-coverage was < 100 reads/position at 103 positions, including a stretch with < 10 reads/position between positions 539 and 551 (Table 4, Additional file 6: Figure S4). We note that position 540 in our 633-bp alignment (Additional file 2: Alignment S3) is the variable position that separates the clade including R. robustus II–IV (all with T) and that including R. prolixus and R. robustus I (with a derived C) [5]. We generated one full-length sequence using R. robustus II from Ecuador (AF435858.1) as the query (Fig. 2); depth-coverage was high except for a 70-position section with depth-coverage below 100 reads/position, including < 20 reads at positions 351–369 and < 10 reads at nine positions within this 19-bp stretch, which includes three variable positions (Table 3, Additional file 2: Alignment S3 and Additional file 6: Figure S4).

Our R. prolixus (AF435860.1 and AF435862.1) and R. robustus I (AF435861.1) D2-28S query sequences were identical and generated full-breadth-coverage consensus sequences from all ‘R. robustus’ SRAs (Fig. 4); the exception was R. prolixus query AF435862.1 vs ERX1387160, for which no read aligned to the last base (Table 4, Additional file 1: Table S1). These full-breadth-coverage consensus sequences did not differ by a single base from the also identical query sequences. Depth-coverage was overall high (except, as mentioned above, for SRA ERX1387160), with only a few, short stretches having low depth-coverage (Table 4, Additional file 6: Figure S4).

D2-28S sequences produced gene trees with relatively poor resolution; although the clustering of query and consensus sequences overall mirrored the patterns described for cytb and ITS2 (Figs. 5, 6), most tree nodes had low to very low support (Additional file 5: Figure S3). Overall, these D2-28S data further suggest that the benchmark UNESP’s ‘R. robustus’ colonies are mainly or fully composed of bugs with R. prolixus DNA, with only one R. robustus query (genotype IV) generating a full-breadth-coverage, yet low-depth-coverage, consensus sequence. The data were less informative with regard to the relation between R. montenegrensis and R. robustus II, likely because the only R. robustus II D2-28S sequence so far available (AF435858.1) is from an Ecuadorian bug caught ~ 2000 km from R. montenegrensis’ type locality. Our finding that reads from three ‘R. robustus’ SRAs aligned with full-breadth- and high depth-coverage to one R. neglectus query (Fig. 4, Table 4) is discussed in the next sub-section.

Quality check: cytb, D2-28S and AmpG

As expected, no full-breadth-coverage sequences were generated when we aligned our study SRAs to R. nasutus sequences or to AmpG sequences from members of the R. prolixus–R. robustus species complex (Additional file 1: Table S1). Rhodnius neglectus queries, however, generated four full-breadth-coverage consensus sequences (Figs. 3, 4, Table 4). One R. neglectus cytb query (JX273156.1) yielded full-breadth-coverage against one ‘R. robustus’ SRA, yet with mean depth-coverage < 10 reads/position (Fig. 3, Table 4). In contrast, the only available R. neglectus D2-28S query [31] yielded both full-breadth-coverage and substantial mean depth-coverage (albeit with a ~ 45-bp stretch with depth-coverage < 20 reads/position) against three ‘R. robustus’ SRAs (Fig. 4, Table 4). This D2-28S sequence (JQ897670.1) is from a bug identified as R. neglectus but reportedly collected in western Amazonia (‘Orellana, Ecuadorʼ) [31], where R. neglectus does not occur [9, 10]. Therefore, this is most likely a case of misidentification or mislabeling of the specimen (voucher ‘UCR_ENT_00052203’ at the Entomology Collection of the University of California, Riverside) [31] (see Table 4).

Discussion

In this report we have illustrated how publicly available transcriptome data can be used to clarify the systematics of a taxonomically challenging group of cryptic disease-vector species. This transcriptome-based approach to molecular systematics has, to our knowledge, not been used before in vector studies; it is overall analogous to the assembly of mitochondrial genes from transcriptome data used to study, for example, poison frogs [32], catfish [33], true bugs [34] or ants [35] (see also [36]). We found evidence confirming that R. montenegrensis, a species described in 2012 [7], is genetically indistinguishable from R. robustus II, one of the sibling taxa within R. robustus (s.l.) that Monteiro and colleagues discovered in 2003 [5, 6]. To solve the paradox that R. montenegrensis appears to be morphologically and genetically distinct from R. robustus [7, 12, 13], we then showed that the ‘R. robustus’ stocks used as the taxonomic benchmark in R. montenegrensis’ description [7] and in later transcriptome-based comparisons [12, 13] are almost certainly R. prolixus, likely mixed to some degree with R. robustus.

We note that our confirmation that R. montenegrensis and R. robustus II are almost identical genetically does not invalidate the former as a separate species – it just shows that ‘Rhodnius montenegrensis’ is the binomial for what we informally called ‘Rhodnius robustus II’ [5, 6]. Our results, in any case, provide an example of how triatomine-bug taxonomic research can be confounded when sequence-data analyses are loosely interpreted [6]. Thus, the cytb data presented in the description of R. montenegrensis [7] already showed that R. montenegrensis and R. robustus II are all but indistinguishable at that locus (Table 2). However, instead of pointing out the virtual identity of R. montenegrensis and R. robustus II sequences, da Rosa et al. [7] underscored the (effectively negligible) differences – a stance later mirrored in a study involving cytb-based identification of R. montenegrensis specimens [37]. As noted by Monteiro et al. [6], this is probably also the case for R. marabaensis and R. robustus III from southeastern Amazonia [5]; R. marabaensis sequences, however, have not been made available in public databases [8]. Along the same lines, the hypothesis that R. milesi [38] and R. taquarussuensis [39] are R. neglectus variants (see [6, 10]) recently received empirical support from molecular systematics [6, 40]. A further example of taxonomic uncertainty is R. zeledoni [41], whose holotype (the only known specimen) is strikingly similar to the sympatric R. domesticus [1, 6, 42]; however, the data needed to address this uncertainty are so far unavailable.

Our results also show how the use of mixed or misidentified Rhodnius spp. colonies can confound taxonomic research even further – and how we can take advantage of public-domain molecular data to clarify cryptic–species systematics in the face of such confusion. Perhaps more importantly, our finding that some putative ‘R. robustus’ colonies are fully or almost fully composed of bugs with R. prolixus DNA contributes to casting doubts over the conclusions of the many studies that made use of non-genotyped ‘R. prolixus’ or ‘R. robustus’ laboratory stocks. Mesquita et al. [18] noted this problem in their quest for a pure R. prolixus stock to be used in genome sequencing. Of the 15 putative R. prolixus colonies they genotyped, just four were pure: one had both mitochondrial and nuclear R. robustus II DNA and ten had introgressed R. robustus IV mitochondrial DNA (see p. 28 of Appendix SI of [18]). As transcriptome (and genome) data from other putative R. prolixus and R. robustus colonies progressively accrue, approaches analogous to the one we illustrate here may help elucidate their taxonomic identity and genetic integrity. This would be particularly interesting in the case of putative R. prolixus colonies derived from the legendary stock used by Sir Vincent B Wigglesworth in his seminal studies on insect physiology [43], but would also be valuable for assessing the taxonomic status of bugs used in more recent research on Rhodnius spp. morphology, genetics, physiology, behavior, bionomics or interactions with microorganisms including Trypanosoma cruzi (e.g. [44,45,46,47,48]). Similarly, our results suggest that the value of several approaches put forward to investigate the systematics of cryptic or near-cryptic Rhodnius taxa (including, for example, the use of quantitative phenotypic traits [7, 49] or ctyogenetics [39, 50]) will have to be reappraised after careful consideration of the specific status of the bugs, whether field-collected or laboratory-reared, used in comparative analyses.

Conclusions

Here, we have illustrated how public-domain transcriptome reads and locus-specific sequences can be combined to address challenging issues in vector systematics. Using query sequences from mitochondrial and nuclear loci, six publicly-available raw transcriptome datasets, and a straightforward bioinformatics approach, we (i) confirmed that R. montenegrensis and R. robustus II are in all likelihood the same species, and (ii) showed that the UNESP ‘R. robustus’ stocks used as the taxonomic benchmark in R. montenegrensis’ description and in later comparative studies are most likely a mixture of (mainly) R. prolixus and (partly) R. robustus (probably genotype II). In this particular instance of taxonomic confusion, misinterpretation of sequence-data analyses was compounded by the misidentification of taxonomic-benchmark laboratory stocks. More generally, and together with previous reports of mixed and/or misidentified Rhodnius spp. colonies, our results call into question the conclusions of many studies based on non-genotyped ‘R. prolixus’ or ‘R. robustus’ stocks. Rhodnius prolixus and R. robustus (s.l.) are similar in many respects, but differ in a fundamental way: the former is a primary domestic vector of Chagas disease, whereas the latter comprises a suite of sylvatic species (including R. montenegrensis) of limited medical relevance. Correct species identification will be key to any attempt at understanding what physiological, behavioral or ecological factors may underlie this crucial difference.