Specimens

A total of 640 hoverfly specimens were collected with sweep nets from different localities in Ghana, Togo, Benin and Nigeria, from May 1994 to December 2014 (Fig 1, S1 Table). Activities reported in the study were conducted neither in national parks nor in any protected areas where a special permit was required. Generally hover fly material was obtained from agricultural land and its adjacent environment. Private grounds were never penetrated without the consent of owners and were visited with national recruited staff and as part of the ongoing projects on pest control and biodiversity of the International Institute of Tropical Agriculture (IITA). IITA is a non-profit making international organisation and a member of the Consultative Group on International Agricultural Research Consortium deeply implicated in research-for-development work in collaboration with national partners of 25 countries in the humid and sub-humid zones of sub-Saharan Africa. Research work in Ghana, Benin, and Nigeria, where IITA is headquartered or has regional offices, is based on bilateral agreements in form of memorandums of understanding (MoU) signed by the ministries of agriculture of all respective governments (more information can be found on http://www.iita.org/). In these MoU's research work in the field is an integral part of IITA’s contracted mandate. In Togo, IITA maintains a close partnership with the National Plant Protection Service (SPV) and the university of Lomé through which material was obtained. Therefore, no specific permissions were required for the collected hoverfly material. None of the collected hoverfly species figure in any red list, are endangered, threatened or considered to be endangered in the involved countries. Similarly, no species collected in the present study are ranked in any IUCN list or protected by CITES.

Specimens collected before 2013 were pinned and stored in the dry collection at the International Institute for Tropical Agriculture (ITTA) in Calavi, Benin (N = 303); those collected in 2013–2014 were kept in absolute ethanol and are vouchered at the Royal Museum for Central Africa (RMCA, Tervuren, Belgium) (N = 337) (S1 Table). All specimens were identified based on external morphology by KJ and MDM (a list of the keys used is available in S1 Text), except for the Microdontinae which were identified by Menno Reemer (Naturalis Biodiversity Center, Leiden, The Netherlands). Whenever possible, identifications were checked against voucher specimens at the RMCA, the National Museum Bloemfontein (Bloemfontein, South-Africa), and the KwaZulu Natal Museum (Pietermaritzburg, South-Africa). Some specimens could not be unambiguously identified, but were linked to the morphologically most similar species by ‘cf.’ (confer) or the most similar genus by ‘sp.’. In both cases the specimens were treated as separate nominal species. One male Syritta specimen (voucher 422C03; labelled as Syritta brevis/unicolor, see S1 Table) keyed out at a position where Lyneborg & Barkemeyer [27], in their identification key on Syritta, noted “the unknown males of S. unicolor new species and S. breva new species will probably come out here” (i.e. the males of both species are unknown). As such we identified 101 nominal species from 28 genera. One leg of each specimen was preserved in 98% ethanol at -20°C for DNA sequencing.

DNA extraction, amplification and sequencing

Genomic DNA was extracted using the NucleoSpin Tissue kit (Macherey-Nagel, Düren, Germany). After adding proteinase K, samples were incubated overnight at 56°C. A fragment of the 5’-end of COI was amplified using primer pair LCO1490 and HCO2198 [28]. Each PCR mixture (25 μl) contained 1x PCR buffer, 0.2 mM dNTPs, 0.4 μM of each primer, 2.0 mM MgCl₂, 0.5 U of Taq DNA polymerase (Platinum, Invitrogen), 2 μl DNA extract and enough mQ-H₂O to reach a total volume of 25 μl. The PCR protocol involved an initial denaturation period of 94°C for 4 min, followed by 35 cycles of 30 s at 94°C, 30 s at 45°C and 45 s at 72°C, and a final extension of 7 min at 72°C. Amplicons were cleaned using the NucleoSpin^® protocol (Macherey-Nagel, Düren, Germany) and bidirectionally sequenced (using the PCR primers) on an ABI 3130xl automated capillary DNA sequencer (Life Technologies) using the BigDye Terminator v.3.1. Cycle Sequencing Kit. Sequences were assembled in SeqScape v.2.5 (Applied Biosystems), and aligned, trimmed and translated into amino acid sequences in MEGA v.5.2 [29] to verify that they were free of stop codons and gaps (none were detected). All sequences were submitted to GenBank (Accession nos. KR830807- KR831281, KR632611-KR632615, and KT624201-KT624233). Additionally, we included the ten >550 bp DNA barcodes of nine nominal Syrphidae species from GenBank and BOLD which also occur in the Afrotropics (S2 Table).

Data analysis

First, we described the success rate of obtaining DNA barcodes of >550 bp. This threshold was arbitrarily chosen to retain a high amount of barcodes without the loss of a long fragment of the barcode region in further analyses. Samples for which the PCR failed, or for which the sequencing yielded shorter fragments, were considered as unsuccessful. The success rate, i.e. the percentage of samples with barcodes >550 bp, of the pinned specimens was calculated for periods of five years (i.e., 1993–1997, 1998–2002, 2003–2007, and 2008–2012), and that for the ethanol stored material (2013–2014) was calculated separately. Then, three analytical approaches were employed to analyse the sequences and are briefly described below.

Kimura 2-parameter (K2P) [30] pairwise distances were calculated within and between species, genera, and subfamilies, since in DNA barcoding studies this is the most widely used distance measure, but see [31] for a discussion on the (inappropriate) use of this measure. Frequency distribution histograms of conspecific and heterospecific pairwise distances, for each of the three subfamilies and for each of the genera with >5 morphospecies, were constructed using the APE package 2.7–1 of R [32] to look for (the size of) barcode gaps [33]. If there is no overlap in the cumulative distribution curves of intra- and interspecific K2P distances, this was referred to as a ‘true barcoding gap’.

The proportion of correctly identified specimens was estimated using the Best Match (BM) and Best Close Match (BCM) criteria in the program SpeciesIdentifier [34]. According to BM, each query was assigned the species name of its best-matching sequence regardless of how similar the query and reference sequences were. Identification then was considered correct when both sequences were from the same species (true positive), incorrect if the query species differed from the closest reference species (false positive) or ambiguous if multiple species yielded a BM with the query species. BCM relies on a threshold value of sequence similarity. This threshold was determined as the ‘best compromise threshold’ based on cumulative distribution curves of intra- and interspecific K2P distances following [35]. The proportion of correct, ambiguous or incorrect identifications was calculated as above. Yet, with the latter method, queries that have no BCM below the threshold are discarded from the identifications (i.e. remained unidentified) (as true negative if identification was incorrect, as false negative if identification was correct). Note that species that are only represented by a single sequence in the dataset will generate incorrect identifications under the BM and BCM criteria, because there are no other conspecific reference sequences in the dataset with which they can match [36]. For this, we have eliminated species with only one barcode sequence (N = 32; S1 Table) from the BM and BCM analyses. In the presence of a true barcoding gap, identification success using BM and BCM is 100%.

A Neighbor-Joining (NJ) tree [37] (using K2P distances) and a maximum likelihood (ML) tree [38] were constructed after removing identical sequences with DAMBE v.5 [39]. We focused on whether individuals of the same species clustered together, rather than on the evolutionary relationships between species (i.e. tree-based identification sensu [13,14], see [34]). Clistoabdominalis ancylus (Pipunculidae, voucher JSS1353, GenBank accession no. DQ337639) was used as outgroup. The NJ-tree was constructed in MEGA v.5.2 [29] using 1,000 bootstrap replicates. In addition, the COI dataset was partitioned according to codon position and the Akaike Information criterion in jModelTest v.2 [38,40] was used to select the most appropriate model of evolution. These were the F81+I+G (first position), GTR+I+G (second position), and GTR+G (third position) model, respectively. Then, Garli v.2.01 [41] was used to perform a maximum likelihood (ML) analysis (two replicates; 200 bootstrap pseudoreplicates) taken into account the most appropriate models of evolution for each of the three codon positions.

Restrictions to the current DNA barcode reference dataset

We evaluated the use of DNA barcoding to identify 90 Afrotropical hoverfly species collected in an area that largely corresponds to the Dahomey Gap, a West African region where the Guinean Forest-Savanna Mosaic reaches the coastal area, thereby splitting the coastal forest ecoregions (with the Eastern Guinean Forests ecosystem to the West, and the Nigerian Lowland Forests ecosystem to the East) [42]. It is currently composed of a mosaic of savannahs, plantations, degraded forest and remnants of natural forest. Public DNA sequence libraries (GenBank and BOLD) lack representative reference barcodes for these, and almost all other, Afrotropical Syrphidae even though about 600 species currently have been described [1,7,8,9].

One reason for the lack of DNA barcodes of Afrotropical hoverflies in public databases is that much of the material is pinned and old. Indeed, the success rate of obtaining a barcode sharply dropped for samples of >10 years old (20% success rate), a phenomenon that is common among pinned insects [43,44]. Nested PCR and mini-barcode approaches may overcome the low PCR and DNA sequencing success rate of old, pinned specimens e.g. [45,46], but are more time-consuming and labor intensive, especially if new primers need to be developed. Next-generation DNA barcoding may become a better solution because of its protocol simplicity, reduced cost per barcode read, faster throughout, and added information content [47]. Currently, using Sanger-sequencing our database represents roughly one third of the more common species (estimated from records given by [48]), and does not include any type-specimen.

Obviously, our dataset does not span the entire taxonomic diversity of African hoverflies and two-thirds of the more common species, and representatives from more than 20 genera, are not represented in the dataset (because of the restricted sampling area). Besides the lack of many species in the dataset, we found a highly significant correlation between the number of individuals sampled per species and the number of haplotypes recovered, and between the number of species sampled per genus and the number of known species in Africa, even though the mean, and maximum, K2P distances were not, or less, affected by sample size. Incorrect species assignments in Afrotropical Syrphidae thus may be because of errors in the construction of the barcode database. This problem is inherently associated with barcoding studies of unknown biotas, such as Afrotropical hoverflies for which only approximately 60% of the known species can be identified using the current identification keys [8]. The more speciose genera even pose more of a problem, because keys for these genera cover only 44% of the Afrotropical fauna [8]. For instance, the genus Eumerus is probably the most speciose hoverfly genus in the Afrotropics with more than 70 species [48] but current identification keys have many shortcomings and species descriptions of many species are vague, insufficient, or based on a single sex. Nevertheless, DNA barcode studies such as ours may set a reference, and may improve or boost ongoing taxonomic studies on the Afrotropical Syrphidae. Conducting similar studies for representative collections of other geographical parts or biomes of the continent may rapidly increase our taxon coverage (including generic level), and thereby increasing the value of the DNA barcode database as a useful tool for identification and other research purposes.

Despite of the shortcomings listed above, DNA barcoding works well for most of the nominal species we examined, and various analytical methods give highly congruent results. In general, the mean intraspecific divergence of 0.0078 (range 0–0.0563) is at the lower end of the distribution of intraspecific divergences found in other insect groups, while the mean interspecific divergence of 0.093 (range 0–0.163) is relatively high for congeneric insect species [49]. Yet, the interspecific divergence distribution overlaps the intraspecific divergence distribution in the subfamilies Syrphinae and Eristalinae, resulting in the absence of a perfect DNA barcoding gap [50]. This makes it impossible to define a general distance threshold for hoverflies. However, some methods allow to estimate an ad hoc threshold for a specific reference dataset so that assignment errors can be minimized e.g. [36]. Such threshold was lower for the Eristalinae (0.037) than for the Syrphinae (0.06), and were much higher than for the Microdontinae that showed a true barcoding gap (0.007–0.02) (i.e. there is no overlap in the frequency histograms of congeneric and intraspecific K2P-distances; see Fig 4C). Our results further show that optimal thresholds may differ strongly among genera and thus is seems more appropriate to calculate optimal thresholds at the genus level to discriminate between intra- and interspecific K2P distances. In our dataset, incorrect assignments were caused both by high intraspecific, and low interspecific, sequence divergence, and this suggests that the taxonomy of several taxa deserves further study, or, alternatively, that DNA barcoding may not work for some genera or species groups.

Taxonomic implications

Meier et al. [34] investigated whether DNA barcodes could be used for species identification in 449 species (1333 barcodes) of Diptera and found a relatively low success rate (< 70%) based on tree-based and other proposed species identification criteria. Yet, DNA barcoding allows the fast detection of shallow interspecific, or deep intraspecific, barcode divergences and may facilitate the selection of taxa for future taxonomic work [15,19]. As such, we can depict problematic taxa, or species groups, within the Afrotropical Syrphidae that warrant further taxonomic study. Low interspecific differentiation was observed in five nominal species pairs, viz. Allobaccha euryptera—A. picta, Allobaccha praeusta—A. cf. praeusta, Asarkina ericetorum—A. gemmata, Eristalinus quinquelineatus—A. surcoufi, Microdon brevicornis–Archmicrodon sp.1, Rhingia caerulescens–R. semicaerulea, and Melanostoma bituberculatum–M. cf. floripeta–M. cf. bituberculatum, whereas high intraspecific differentiation was observed in Allobaccha picta, Asarkina ericetorum, Syritta bulbus, Phytomia natalensis, Eristalodes quinquelineatus, Eristalinus vicarians, Graptomyza triangulifera, and Polybiomyia divisa. Such observations may reflect geographical structuring or evolutionary history. Some of these lineages may represent recently diverged species in which the COI sequences has not yet accumulated many mutations or may represent ancestral polymorphisms that have been retained in the two taxa. This may not be very surprising for a group that has received little taxonomic attention. One obvious genus that illustrates well how DNA barcoding may be an onset for a taxonomic revision is Allobaccha for which we observed both low inter- and high intraspecific sequence divergence, and a number of putative nominal species, which we could not identify using the current literature/keys, and that may represent undescribed taxa. Perhaps this is even better illustrated by the Microdontinae of our dataset where none of the individuals for which we obtained a COI barcode currently could be attributed to a known species (see S1 Table). Two Metadon taxa, viz. Metadon cf. mythes and M. cf. inermis, morphologically closely resemble known species, and five other clades in the NJ-tree seem to represent undescribed species. Also the Paramixogaster specimen for which we could not obtain a COI barcode, seems to be an undescribed taxon (Menno Reemer, personal communication). Obviously, there is a need for more sequence data and the study of more variable DNA markers to improve resolution but also of a re-evaluation of the morphological characters that currently are used to separate the nominal species (see also [51]). Only such an integrative approach will allow to highlight the effects for the lack of correspondence between sequence variants and current nominal species, in a similar way as integrated taxonomy has improved the taxonomy of the Merodon equestris species complex [52].

Identification of introduced species

DNA barcoding allowed to detect an introduced hoverfly species in the Afrotropics, viz. Toxomerus floralis (Fabricius, 1798). Obviously, the specimens could not be identified using the current literature and identification keys on Afrotropical Syrphidae. Using the search engines BOLD Identification System (BOLD-IDS) and GenBank's BLAST the specimens were assigned to T. floralis. A subsequent morphological identification using specific keys for Toxomerus confirmed the species’ identification [53]. This New Word species is only the second New World species that has been introduced into the Old World and it seems well-established in Togo, Benin, Nigeria and Cameroon. Hence, DNA barcoding may not only facilitate the discovery and identification of (recent) introductions but it may also allow a better bio-monitoring of the species and of the potential impact on endemic plant-pollinator communities and ecosystems [54].

Linking sexes and life history stages of species

The larvae of hoverflies show a variety of feeding modes. Information on the feeding mode, and associated feeding morphology, may yield important information to understand the evolutionary and phylogenetic relationships of hoverflies, since morphological innovation is often associated with feeding modes [55,56]. Unfortunately, such information is lacking for most hoverfly species. For instance, food plants (oviposition sites) are known, and descriptions of immature stages have been provided, for <8% of known phytophagous hoverflies [55,56]. Andrić et al. [57] showed that DNA barcoding is a useful tool to identify the larvae of Merodon species. Similarly, our DNA barcoding reference library of Afrotropical hoverflies may enhance our understanding of the (feeding) ecology and morphology for Afrotropical hoverfly species. In the same view, DNA barcoding will be highly valuable in linking sexes of species, especially since perhaps for even more than 30% of the current Afrotropical hoverfly species only one sex is known. For instance, previously supposed females of Mesembrius ingratus (Loew, 1858) were all attributed to Mesembrius tarsatus (Bigot, 1883) [58] so that the female of M. ingratus remained unknown. We here show that males of both species have a very low intraspecific sequence divergence (M. ingratus: 0.0022, M. tarsatus: 0.0015), yet a high mean K2P-sequence divergence of 0.09. Two females (vouchers 106E08 and 417E08) clustered within the males of M. ingratus while seven other females (vouchers 106E02, 106E04, 107A08, 107C03, 425F06, 418B03, 418C03) clustered within the males of M. tarsatus. We thus have, for the first time, individuals that can be unambiguously identified as females of M. ingratus. This result will further allow to examine the females of both species for morphological diagnostic characters, and to improve the morphological identification key for the genus Mesembrius (Jordaens, Goergen, Backeljau & De Meyer, unpublished data). Another example is the one male Syritta specimen (voucher 426A01) that in the identification key of [27] keyed out at a position where the unknown males of Syritta brevis and S. unicolor are supposed to key out. Unfortunately, the reference database does not contain barcodes of females of both species so that the identification of the male specimen will not be unambiguous.