Ethics statement

No specific permits were required for fieldwork, as the sampled locality is not protected. The sampled area is owned by the Židlochovice Forest Enterprise, a division of the Czech National Forests, Inc., and the data were collected with their approval. No specific permissions were required to collect insect specimens, because the collected taxa are not protected in the Czech Republic.

Insect sampling and rearing

Lepidoptera and Hymenoptera (sawflies) larvae were sampled from all plants with a diameter at breast height (DBH) >5 cm on a 0.2 ha plot of broadleaf deciduous forest in Lanžhot, Czech Republic (48.689685 N, 16.944742 E), using an elevated truck-mounted work platform (cherry-picker). Sampling was conducted from May to August 2013 and 2014. Each larva was morphotyped and photographed. Larvae were then transferred to plastic containers (one larva per container) where they were reared on fresh leaf material from the plant species it was collected from (Acer campestre, Carpinus betulus, Fraxinus angustifolia, F. excelsior, Quercus cerris, Q. robur, Tilia cordata, or Ulmus laevis) in the laboratory, until either adults or parasitoids emerged [23]. In total, 6473 hosts were sampled and 1700 parasitoids were reared (from 1032 rearing events).

Morphological, standard barcoding, and metabarcoding identifications

We prepared five host–parasitoid, multispecies, mock samples consisting of 22 individual insects (five hosts and 17 parasitoids) representing 14 BINs (S1 Table). Every sample was prepared by mixing the remnants of the host’s body with parasitoids reared or dissected from it. To increase the complexity of samples before DNA extraction, we added body parts from parasitoids reared from other host individuals into mock samples 1–4 (from the same host species, except of sample 4).

Standard barcoding vs morphotyping

Morphological, standard barcoding, and metabarcoding identifications

Based on the mock samples, there were no significant differences between the three methods regarding identification success into individual taxonomic levels (χ2 = 2.03, df = 2, P = 0.362; χ2 = 5.12, df = 2, P = 0.078; and χ2 = 0.50, df = 2, P = 0.780, for family, genus and species level, respectively; Table 1, S1 Table). The metabarcoding successfully generated 19,963 forward and 20,155 reverse reads after quality trimming. On average, we retained 3992 forward and 4031 reverse reads per sample with median read length of 254 and 252 bp, respectively. After performing all quality processing and clustering procedures and discarding the generated singletons, 19,822 reads were retained (i.e. ~50% of all reads). Their clustering resulted in 69 different MOTUs (S3 Table) including eight (11.6%) host clusters (82.7% of reads; n = 16,386) and 15 (21.7%) parasitoid clusters (4.88% of reads; n = 966). Interestingly, 46 MOTUs (66.7%) had no match with standard COI barcoding reference sequences; 27 of these clusters (39.1%) were further identified as putative symbionts (5.39% of reads; n = 1068) and 19 (27.5%) were assumed as environmental or laboratory contaminants (6.63% of reads; n = 1315). Eighty-seven reads (0.44%) were not assigned to a taxonomically specific cluster and might, therefore, represent chimeric sequences (with no similar hit in the reference database and/or with various parts of the same sequence showing very different taxonomic identifications) that were not excluded by the de novo chimera identification procedure (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Success of each method in the identification of the 22 individuals present in the mock samples.

https://doi.org/10.1371/journal.pone.0187803.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Detailed composition of the individual mock samples (S1–S5) recovered by DNA metabarcoding (Illumina MiSeq).

The taxonomic assignment of recovered host and parasitoid taxa is emphasized together with the proportion of putative symbionts. The width of each sector corresponds to the relative proportion of its reads (n_r = total number of reads). Taxonomic levels are displayed hierarchically from order (the innermost layer) to species level (the outermost layer). The inset table shows organisms (H = host, P = parasitoid) put in the mock samples, and their identification to the lowest possible taxonomic level based on consensus of the three methods (morphological identification, standard barcoding and metabarcoding). See S1 and S3 Tables for details.

https://doi.org/10.1371/journal.pone.0187803.g001

The overall metabarcoding diversity recovery was high, reaching 92.8% (13/14) of the host and parasitoid taxa (unique BINs), and 95.5% (21/22) of the individuals employed in this study had successful blast hits in the local reference library consisting of representative standard barcoding sequences from all specimens included in the mock samples (S1 Fig). Only one individual belonging to Alsophila aescularia (lepidopteran host) was not recovered and thus could not be assigned.

Standard barcoding vs morphotyping

Standard barcode (i.e. COI) sequences were recovered from 865 of the 1072 analyzed specimens resulting in 795 (465 parasitoid, 330 host) barcode compliant records (>500 bp; 74.2%). In total, we obtained 104 different parasitoid and 108 host BINs. The 795 DNA barcodes obtained increased the putative diversity of assigned parasitoid morphospecies in our studied area from 91 to 104 BINs (MOTUs), with subfamilies Eulophinae and Microgastrinae presenting the highest increase (from six to 14, and from 19 to 26, respectively). Similarly, barcodes recovered from hosts increased their putative diversity from 97 morphospecies to 108 BINs, with families Geometridae and Tortricidae presenting the highest increase (from 29 to 34, and from eight to 12, respectively). Regarding the completeness and usability of these local reference libraries, the identification success of herbivore hosts was much higher than that of parasitoids. In hosts, we successfully identified 90.7% of BINs to the species level (98/108), whereas in parasitoids, identification success was limited to higher taxonomic levels with only 39.4% (41/104) of BINs successfully identified as putative species (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Host–parasitoid diversity at the study site (Lanžhot, Czech Republic) sampled in 2013–2014.

https://doi.org/10.1371/journal.pone.0187803.t002

Metabarcoding, standard barcoding, and morphological identifications performance

Overall, recovered diversity and identification precision using metabarcoding agreed with that obtained using standard barcoding and accurately reflected morphological identifications. Thus, our data suggest that employing both paired-end reads in a parallel manner in the classification procedure used in metabarcoding produced assignments as accurate as in the standard barcoding approach (except in one case; see specimen “Hyposoter” in sample S1 in S1 Table). Despite this accuracy, we recorded an inferior precision (although non-significant) of metabarcoding identification when compared to standard barcoding, resulting from the overall shorter sequences produced by the MiSeq platform (658 bp vs. ~2*300 bp). This was tested in silico for above-mentioned specimen by cutting out the missing part from the middle of its full barcode sequence: in such case, RTAX confirmed the same taxonomic assignment as with MiSeq representative read. Nevertheless, the metabarcoding approach, using a combination of MiSeq sequencing, amplification of almost full COI barcodes by means of updated wide-coverage degenerate primers, and bioinformatic procedures including RTAX classification, seems to effectively recover host–parasitoid diversity. A similar conclusion was reached in a plant diversity study using metabarcoding, showing that full-length barcode markers have a potential to outperform shorter barcode fragments [39]. An alternative to the RTAX method for processing the classification of paired-end reads would be to search concatenated reads with N’s filling the gap of non-overlapping parts of the markers against a reference library; however, current search algorithms, as that implemented in BLAST or USEARCH, do not produce ideal results with this type of query [32] and, therefore, RTAX is the most effective solution. Compared to other currently available classifiers, RTAX seems to suffer from over classification and higher error rates, particularly at the species and genus levels, related to the coverage of the reference database [40]. However, its main indisputable advantage remains: the ability to use both non-overlapping reads producing doubled information, which has been proven to result in better classifications compared to partial information from single read and thus shorter overall sequence [39]. In our study, we used a conservative (3%) pairwise sequence divergence as the threshold to distinguish between two MOTUs, although the arbitrary nature of this clustering dissimilarity threshold has been criticized and alternative methods of MOTU delimitations are available (see [11] and references therein).

The design of degenerate primers applied in the present study might yet be suboptimal for the amplification and sequencing of particular species as we were unable to detect the host Alsophila aescularia. Although Sanger sequencing was able to recover the COI sequence of this lepidopteran using standard insect primers, an attempt to generate Sanger sequences using Fol-degen primers repeatedly failed. The unsuitability of degenerate primers for Alsophila species might be further confirmed when the mitochondrial genome of any of the Alsophila species becomes available. Another inconvenience of using highly degenerated primers is their sensitivity to environmental or laboratory contaminations [41]. This aspect was confirmed by our study as our metabarcoding data contained obvious contaminants such as plant (e.g. Malphigiales, Solanum lycopersicum), leaf beetle (family Chrysomelidae, SPH01), or endemic Australian butterfly Leucania cruegeri DNA (S3 Table). In our particular case, laboratory contaminations might have had a greater impact than that expected for the practical application of metabarcoding due to the excessive manipulation of specimens during the rearing, preparation of artificial mock samples, and morphological identification procedures. Rinsing and carefully handling specimens during processing could eliminate a substantial part of environmental and laboratory contaminations.

Another inconvenience of using metabarcoding is the need for a careful interpretation of the recovered diversity as more insect species can be revealed than that involved in the real interaction. In multiple hosts’ detection, this situation is simply resolved because the DNA of the real host is usually present in the largest amount. However, in multiple parasitoids’ detection, deciding which species was indeed present in the sample is much more complicated as one host can be parasitized by multiple species, and the amount of their DNA in the sample is usually an order of magnitude lower than that of their host. In our particular case, in addition to revealing the identity of the parasitoids added to the mock samples, DNA metabarcoding revealed parasitoid taxa of uncertain origin: Carcelia tibialis (0.2%, sample S4), Microplitis demolitor (0.03%, sample S5), and a Tetrastichinae species (2%, sample S3). Since M. demolitor is a non-European species commonly used in insect pest management, and its DNA was present in a very small amount, this was considered a laboratory contamination. However, C. tibialis is common in Europe, Carcelia spp. are known to attack the host in which C. tibialis was detected, and it was represented in a considerable amount of reads; thus, this species might indeed have been present in the host's body but it was overlooked during dissection. Similarly, Tetrastichinae parasitoids are known to attack the host where they were found, and due to the relatively high number of reads we can assume that representatives of this subfamily were present in the samples. Alternatively, they might have been hyperparasitoids of added or dissected parasitoid species [42]. Moreover, parasitoids’ oviposition in the ‘wrong’ host is common [43] and, therefore, detected parasitoid DNA could have originated from such unsuccessful parasitation attempts. However, unpredicted taxa in our samples could as well represent misclassifications based on the misidentified hits in reference database, which can occur more frequently in understudied and taxonomically challenging groups [44]. These aspects emphasize the importance of building accurate and comprehensive reference libraries and knowing the ecology of detected insect species.

Standard DNA barcoding is not affected by most of the above-mentioned issues arising from the nature of mixed samples. However, a common problem of Sanger sequencing in the identification of host–parasitoid interactions are the cross contaminations between host and parasitoid DNA [13]. Metabarcoding not only eliminates such problems, as it can also use them to its own benefit by simultaneously detecting and identifying host and parasitoids without rearing or dissecting procedures, which are common time-consuming and labor intensive procedure linked to the classical morphological approach [13]. Moreover, the morphological identification of host–parasitoid communities requires experienced taxonomists as these are composed of larval and/or other hardly identifiable developmental stages. Cryptic species, common in many parasitoid taxa [12, 14, 45], and unsuccessfully reared developmental stages often represent a substantial proportion of dataset (see Table 2 for example) and further complicate their identification. Molecular methods allow detecting such stages and cryptic species and, by increasing the recovered host–parasitoid diversity, can substantially extend the final dataset. This aspect was confirmed in our study as the morphological approach failed to classify a substantial part of the parasitoid samples to the genus and species levels (eight unclassified individuals in larval and pupal stages), while these samples were identified by both molecular methods. However, molecular methods are unable to distinguish special life strategies (e.g. hyperparasitism) and to prove successful parasitism [46]. For these situations, rearing methods allowing the observation of all life stages and the outcome of such interactions seem to be more appropriate. Moreover, relying exclusively on the molecular methods for the species description is inadequate, mainly because building species hypothesis based on a single locus is inappropriate [47]. Still, combining sequence data and other evidence is becoming a widely accepted practice [48, 49].

A considerable part of the diversity of our mock samples recovered by metabarcoding was composed of putative endosymbionts (5.39% of all reads and 27 taxonomic clusters; Fig 1, S3 Table) suggesting its potential for simultaneously recovering endosymbiotic microbiota. It is known that endosymbionts may shape the structure and dynamics of insect food webs [50, 51], and examining insect hosts and their associated microbiota has been tried or discussed [18]. However, we were unable to confirm the endosymbiotic origin of most taxa as many of them were facultative endosymbionts (e.g. can be found either as epiphytic or gut microflora [51–53]); this was the case for Pseudomonas, Stenotrophomonas, Acinetobacter, or Acetivibrio. Conversely, we assumed the endosymbiotic origin of Rickettsia, Ralstonia, or Kocuria, which are obligate endosymbiotic genera. In addition, many endosymbionts identified in our study were not host-specific; Rickettsia, for example, is known from the gut microflora of various herbivorous insects but also from the eulophid wasp, Pnigalio soemius [54, 55]. On the other hand, the cellulolytic bacteria Ralstonia were probably associated with lepidopteran hosts instead of parasitoids (e.g. [55]). Moreover, most endosymbionts’ clusters were only identified to high taxonomic levels (S3 Table) and a better recognition of their origin requires more accurate taxonomic identifications, which could be improved by using more specific bacterial and fungal primers with growing reference libraries and knowledge on endosymbiont ecology.

Standard barcoding vs morphotyping

The second part of our study compared the accuracy of molecular (DNA barcoding) and morphological approaches for the potential reconstruction of a local food web, and assessed the completeness of a local reference library. The previously discussed benefits of DNA barcoding in comparison with morphological identification were evidenced in the markedly increased parasitoid diversity, especially in Eulophinae and Microgastrinae. This agrees with previous studies confirming the high level of crypsis and identification issues in both these groups [12, 56]. Similarly, in hosts, the highest increase in diversity was obtained for Geometridae and Tortricidae, probably due to the presence of hardly identifiable juveniles, and to the high level of crypsis of both groups, which is well documented in the literature [57, 58].

In the Czech Republic, the focal community of lepidopteran and hymenopteran hosts and their parasitoids includes approx. 4000 host and 4500 parasitoid species [59–61]. The reference library of hosts is quite extensive and, from our dataset, we successfully identified the vast majority of putative species (BINs) to the species level while the identification success of parasitoids to the species level was much lower (Table 2). These identification success rates are similar to those reported for the tropics, where 87% of the host and 36% of the parasitoid sequences were successfully identified to the species level based on Papua New Guinea’s local reference library [13]. Successful molecular identifications of central European hosts and their parasitoids benefit strongly from the growing number of large-scale barcoding campaigns, such as the Barcoding Fauna Bavarica [62] or the German Barcoding of Life commenced in 2012, which are substantially contributing to build reference libraries. However, the low efficiency of parasitoids’ molecular identification compared to that of their hosts is caused by their huge unrevealed diversity and lack of molecular data [2] and by taxonomic issues [13]. This seems to be particularly evident in some hyperdiverse and/or cryptic groups of hymenopteran parasitoids, which showed the lowest number of BINs assigned to species (e.g. Eulophinae, 14.3%; Microgastrinae, 26.9%; Campopleginae, 33.3%; Table 2). The situation is much better in European tachinids (Diptera), with 88.2% of BINs identified to species, suggesting that they represent a well-studied and well-barcoded group. These results emphasize the need to continue building comprehensive reference libraries focused on parasitoids.

Future of metabarcoding identifications

Our study proposes using metabarcoding for effectively recovering host–parasitoid diversity. As indicated above, this approach will be most effective at localities where species diversity is covered by comprehensive reference libraries, which is still far from reality, especially for parasitoids. Moreover, accurate identifications are still needed. From the methodological point of view, we suggest that a further increase in recovered diversity could stem from the use of multiple independent markers, which are being increasingly developed as reference databases for alternative barcoding markers also increase [63–66]. In terms of classification procedures, a wide array of algorithms and methods has been developed for this purpose, but none seems to be ideal for all eventualities [63]. Recently, a statistical algorithm capable of incorporating data from several independent classifiers as covariates in a probabilistic taxon membership assignment process was developed, providing unbiased probabilities of taxonomic placement [67]. Thus, there is a growing number of tools helping with this aspect and providing as precise and accurate classifications as possible.

To efficiently process a multitude of samples in high-throughput mode and still be able to draw significant conclusions and capture sparse interactions, several alternative approaches were designed. One of them, multiplex PCR, was suggested to decrease analyses’ costs, allowing the simultaneous detection of several target species, and, in combination with two-dimensional sample pooling, decreasing the number of PCR reactions up to 90% [68]. This will be particularly useful for the specific screening the limited number of interactions with low parasitism rates (<10%), while metabarcoding is more useful for the more general detection of all present trophic links involving unexpected interactions. Multiparasitism or hyperparasitism are quite common [69], but, in general, only a few parasitoid species share one host. Recent advances in sample multiplexing allowed new indexing schemes, including double-indexing or nested tagging [6], and the expected diversity recovered per sample and read depth needed increased with the continuously increasing output of modern sequencers (e.g. Illumina NovaSeq series); thus, a large number of samples can now be pooled and should easily cover the needs of any ecological study. Based on our results, one MiSeq run could potentially be used to process thousands of samples (~5000).

The primary aim of our study was to test metabarcoding as an alternative to standard barcoding and classical morphological identifications, both time-consuming and labor intensive, for a prospective large-scale study of host–parasitoid interactions. Due to limited number of mock samples used in our study, further testing, including other host and parasitoid taxa to cover more of the real global diversity, would be necessary to draw more general and robust conclusions. Still, the present study might be a starting point for further comparative studies. Although in our study, the performance of metabarcoding was tested on mock samples, each including one host and multiple parasitoids (except of S5), for the purpose of studies of host–parasitoid interactions, each host larva potentially with parasitoid(s) inside would be a separately barcoded inclusion in the sequencing pool. Many different aspects of HTS are known to bring technical issues into the datasets produced by these techniques, including library preparation biases, stochastic effects during sequencing, and bioinformatic analyses biases. These are being extensively discussed and subjected to thorough testing in relevant stand-alone studies, and recent findings have shown that those negative effects could be effectively solved by simply increasing sequencing depth [70]. However, optimal sequencing depth for recovery of full complexity of the sample involving host with all (potential) parasitoids inside is tricky to define as it depends on an array of conditions. The most relevant are size proportions of organisms involved in the sample and amplification efficiency. Nevertheless, given that high-throughput costs are continuing to decline, we believe that with growing accurate and comprehensive reference libraries and increasing knowledge on the ecology of detected insect species, metabarcoding might be the most effective future solution for large-scale studies of host–parasitoid food webs, allowing the simultaneous assessment of associated microbiota, thus saving both time and money. However, molecular methods cannot entirely replace the reconstruction of ecological interactions by direct observation (e.g. rearing), and, for their accurate interpretation, reliable taxonomic backgrounds will still be needed.