Mock communities were constructed using 48 field-trapped specimens from different countries in Europe (France, Greece, Portugal, Spain), China (Beijing and Zhejiang Province) and the USA (Michigan) (Table 1), as part of a worldwide trapping experiment using multi-funnel traps baited with a generic attractant blend, including eight Cerambycid pheromones (see details of the blend composition and trapping methods in Roques et al. (2023)). Most of the specimens (36/48) were caught using α-cypermethrin insecticide (Storanet®, BASF Pflanzenschutz Deutschland, Germany) in the trap basins, of which the bottom had been replaced with a wire mesh to allow drainage and keep specimens dry (hereafter called the “dry” method). Other specimens (12/48) were captured using a 50:50 ratio of monopropylene glycol (MPG) and water (hereafter called the “wet” method). Cerambycides collected from field-traps were stored in ethanol 95% and kept at -20 °C until further processing. Two individuals were captured by hand (“hand collected” in Table 1) and pinned in collection boxes after capture. The date, country of collection, type of trap and the 48 specimens used in mock communities are detailed in Table 1.

Six mock communities with varying species composition were assembled as follows:

Test 1: Identifying closely-related species

To assess the efficiency of the different sequencing technologies and primers to differentiate between sister species, bulks 1 and 2 were composed of congeneric species (Table 1). Two legs from each individual (one specimen per species) were collected and pooled to constitute the bulks. The whole set of legs was then ground using flame-sterilised metal pestles to limit the risk of contamination. DNA from the ground material was extracted using the DNeasy Blood and Tissue Kit (Qiagen) following the manufacturer’s instructions. Two additional legs were taken from the same specimens to assess the quantity and quality (A260/280 and A260/230 ratios) of DNA for each specimen individually (Fig. 1a).

Figure 1.

Overview of the DNA extraction protocol for tests 1 (identifying closely-related species) and 3 (mimicking field trap content on species composition) (a) and for test 2 (detecting low abundance species) (b).

All bulk samples were amplified with two pairs of primers internal to the commonly used barcode fragment: BF3/BR2 (called hereafter “B”) (CCHGAYATRGCHTTYCCHCG / TCDGGRTGNCCRAARAAYCA (Elbrecht and Leese 2017; Elbrecht et al. 2019), which generates a 458 bp amplicon that was used for all the technologies; and fwhF2/fwhR2n (called hereafter “F”) (GGDACWGGWTGAACWGTWTAYCCHCC / GTRATWGCHCCDGCTARWACWGG), which generates a shorter 254 bp amplicon (Vamos et al. 2017) and was used for Illumina and MinION technologies only. Each PCR comprised 15.3 µl H₂O, 2.5 µl 10X PCR buffer, 2.5 µl dNTPs [1 mM], 1 µl of each primer [0.4 mM], 0.2 µl Dream Taq (Thermo Fisher Scientific), 0.5 µl Betaine [100 mM] and 2 µl DNA for a total of 25 µl per reaction. For both primer pairs, PCR was performed using the same programme: 95 °C for 5 min, 29 cycles of 95 °C for 30 s, 48 °C for 30 s and 72 °C for 50 s, followed by 72 °C for 5 min (Elbrecht et al. 2019). PCR products were then run on a 2% agarose gel stained with ethidium bromide and visualised by UV transilluminator. The PCR products were then purified with the NucleoFast 96 PCR plate purification kit (Macherey-Nagel). Three PCR replicates were performed for the six bulks when using the MinION technology.

A second ligation PCR was performed on the products of the first PCR to add Illumina® tags and adapters, prepared by ligating Nextera XT indices through an eight cycle PCR (with a modified PCR protocol). The second PCR was carried out with the same conditions as for the initial PCR. Reactions (25 μl) contained the following: 5 μl of template DNA (purified products from the first PCR), 1 μl of each primer [10 µM], 5 μl of 5X GoTaq (Promega) reaction buffer, 1 μl of MgCl₂ [25 mM], 1 μl of BSA [1 mg/ml], 0.5 μl of dNTPs [5 mM], 0.125 μl of GoTaq G2 Polymerase (Promega) and 10.375 μl of molecular-grade water to reach 25 µl. The PCR conditions were the same as for the first PCR, with eight cycles. The products of the second PCR were verified on a 2% agarose gel. PCR products were then equimolarly pooled into two different pools (one pool per primer pair used) and purified using the GeneJET Gel Extraction kit from an agarose gel, following the manufacturer’s instructions. This library was sequenced in Illumina MiSeq using V3 chemistry (300 × 300 bp, 600 cycles) in the Sequencing Center within the Biozentrum of the Ludwig-Maximilian University in Munich (Germany).

Libraries were prepared according to the Oxford Nanopore Technologies ® protocol: “PCR barcoding (96) amplicons (SQK-LSK110) (version: PBAC96_9114_v110_revF_10Nov2020)” with the following specifications. After the first PCR described above, the Nanopore PCR barcoding expansion Pack 1-96 (EXP-PBC096) was used to perform the second PCR to incorporate the Oxford Nanopore Technologies ® barcode sequences on the amplicons generated in the first PCR.

Reactions (50 µl) contained the following: 2 μl of template DNA (purified products from the first PCR), 0.5 μl of each primer [10 µM], 10 μl of 5X GoTaq (Promega) reaction buffer, 2 μl of MgCl₂ [25 mM], 2 μl of BSA [1 mg/ml], 2 μl of Q solution, 1 μl of dNTPs [5 mM], 0.3 μl of GoTaq G2 Polymerase (Promega) and 29.7 μl of molecular-grade water to reach 25 µl. The thermocycling conditions followed the manufacturer recommendations: 95 °C for 3 min, followed by 15 cycles of 95 °C for 15 s, 62 °C for 15 s and 65 °C for 30 s and 65 °C for 7 min.

Final PCR products were then quantified using Qubit and equimolarly pooled before being purified with Agencourt AMPure XP beads (Beckam Coutler). The final pool was then sequenced on the MinION sequencer (Mk1c; Oxford Nanopore Technologies ®, UK) using a R10.3 flowcell (MIN111) with 1331 pores available and the LSK110 ligation sequencing kit. The two replicates of bulk 6 using the MinION technology were of insufficient quality (Nanodrop) and were, therefore, removed from the analysis.

For the production of the libraries, we started with 5 ng of DNA extract (Qubit measurement). The Nextflex Cellfree DNAseq kit (PerkinElmer) was used for the process. The quality of the libraries was assessed using Qubit (for quantification) and Bioanalyzer (using the HighSensitivity kit from Agilent, for size verification). After quality control, each library was amplified by emulsion PCR on the Ion One Touch 2 instrument, with a concentration of 15 pg/µl. Subsequently, the libraries were sequenced on an Ion GeneStudio S5 system using a single-end sequencing protocol with a 300 bp read length. Sequencing was performed on an Ion 520 Chip by the GeT-BioPuces platform (Toulouse, France).

False positive detections are considered to be the detection of a species within a bulk that was not initially present when the bulks were constructed. In order to estimate the representativeness of false positives within true positives in the bulks, the total number of reads assigned to false positive OTUs was reconciled and compared to the number of reads assigned to non-false positive detections. The number of false positives detected according to the different tested combinations is indicated in Suppl. material 1: table S2.

A dataset was built using all the public sequences of all Cerambycidae species available in BOLD systems v.4 (Ratnasingham and Hebert 2007). It was then verified whether all 33 species present in the bulk samples were represented by at least one sequence in the database. Three species not previously included in the database were barcoded through Sanger sequencing on an ABI 3500 genetic analyser (Applied Biosystems) using the Big-Dye Terminator V3.1 sequencing kit (Applied Biosystems) and BF3/BR2 primer pair. They were subsequently added to our local database to ensure that they were represented by at least one barcode sequence. The number of sequences in the database for each species is shown in Suppl. material 1: table S1.

The final reference dataset is available from BOLD in the dataset DS-MINION (dx.doi.org/10.5883/DS-MINION) and includes one barcode per species together with the three newly-generated barcodes. Lab Sheet from the DS-MinION database is shown in Suppl. material 1: table S3.

The raw data were analysed using the FROGS v.4.0.1 pipeline, a standardised pipeline containing a set of tools that are used to process amplicon reads that have been produced from Illumina® sequencing (Escudié et al. 2018; Henrie et al. 2022). First, amplicons with a size between 408 and 508 for the BF3/BR2 primer pair and 204 and 304 for the fwhF2/fwhR2n primer pair were retained using the Pre-process tool. For this first step, paired-end reads are merged using VSEARCH (Rognes et al. 2016) with a mismatch rate set at 10%. Cutadapt (Martin 2011) is then used to remove sequences in which both primers are absent and to trim the primers. Sequence clustering was then performed using the SWARM algorithm (Mahé et al. 2014) with a maximum sequence difference set at d = 1 (--distance 1 parameter), as recommended by SWARM. Chimeric sequences were then removed with the Remove chimera tool relying on VSEARCH with the de novo UCHIME method (Edgar et al. 2011; Rognes et al. 2016). Sequences were aligned to the same database used for the MinION and IonTorrent® data analysis. In order to remove all spurious detections, OTU detections with fewer than 10 reads were removed. In barcoding and metabarcoding studies of insects, the sequence similarity level for OTU identification usually ranges from 95% to 99% (e.g. Gibson et al. (2014); Zenker et al. (2016)). We calculated the best threshold value for our dataset by applying the function localMinima from the R package spider v.1.5.0 (Brown et al. 2012). Based on this analysis, we used a threshold of 98% to assign OTUs to species level with blastn+ (Camacho et al. 2009). The resulting OTU tables for Illumina F and Illumina B are provided in Suppl. material 1: tables S4, S5, respectively.

Bioinformatics analyses were performed on the Genotoul Bioinformatics Platform (INRAE, Toulouse, France). Basecalling and demultiplexing were performed for MinION data using Guppy v.6.1.7; ONT; high accuracy base calling mode; parameters: -c dna_r10.3_450bps_hac.cfg --min_qscore 5 --trim_barcodes. Then, for MinION and IonTorrent® data, we used the msi data processing pipeline v.0.3.6 (Egeter et al. 2022) to reduce the error rate of the reads by polishing them after the basecalling step. Reads smaller than 40 bp were removed with cutadapt v.4.0 (Martin 2011). The size range was set between 408 bp and 508 bp for BF3/BR2 and between 204 bp and 304 bp for fwhF2/fwhR2n. The clustering step was carried out with ISONCLUST v.0.0.6.1 (Sahlin and Medvedev 2020; with parameters: --mapped_threshold 0.825 and --aligned_threshold 0.55) and a consensus sequence per cluster was generated using RACON v.1.5.0 (Vaser et al. 2017). The polished reads were then clustered at 97% sequence identity with CD-HIT v.4.8.1 (Fu et al. 2012) and a representative sequence from each cluster (centroid) was selected. The polished reads were then aligned to the local database with BLAST (BLASTn algorithm). The following parameters were used: -word_size 11 -perc_identity 95 -qcov_hsp_perc 98 -gapopen 0 -gapextend 2 -reward 1 -penalty 1 -max_target_seqs 100. Similarly, to the Illumina® data processing, OTU detections with less than 10 reads were removed. Finally, a taxonomic assignment was performed for each query using a Lowest Common Ancestor (LCA) approach with the bioinformatics package metabinkit (Chain et al. 2016; Egeter et al. 2018; Kitson et al. 2019) with the following thresholds: 98% at species level, 97% at genus level, 95% at family level (Alberdi et al. 2018; Egeter et al. 2022). The resulting OTU tables for MinION B, MinION F and IonTorrent B are provided in Suppl. material 1: tables S6–S8, respectively.

A two-sample test of proportions was used to compare and assess the significance of the proportion of reads assigned to the species levels for MinION, Illumina and IonTorrent technologies using the “Social Science Statistics” website (https://www.socscistatistics.com/tests/anova/default2.aspx). The proportion of reads assigned to different taxonomic levels was calculated by summing the total reads from different bulk samples for each condition. To determine if the number of false positives was significantly different amongst the three technologies and the two primer pairs, we calculated the detection mean for each bulk under different conditions. We then performed an ANOVA test followed by a Tukey HSD test using the “Social Science Statistics” website. The Wilcoxon test, Exact Fisher’s test and standard deviation were calculated in R v.4.3.2 (https://www.R-project.org/). Sensibility, which measures the ability to correctly identify true positives, was calculated using the following formula: true positives / (true positives + false negatives) and precision, which measures the ability to measure the proportion of correct detections, was calculated using the following formula: true positives / (true positive + false positives).

The MinION technology accurately identified 28 out of 48 specimens at the species level, Illumina® technology allowed specific identification of 27 specimens and IonTorrent® identified 24 specimens. The primer pair F enabled the specific identification of 27 specimens at species level, while the primer pair B enabled the identification of 31 specimens at species level. Illumina® F, Illumina® B and MinION B allowed for 25 species-level identifications across all bulks (sensibility = 0.52) and 24 for MinION F and for IonTorrent® B (sensibility = 0.50). This difference was not significant (Fisher’s Exact Test, p = 1.00) (Fig. 2).

Figure 2.

Upset plot showing the number of individuals detected at species level according to the three technologies (Illumina, MinION and IonTorrent), primer pairs (F=fwhF2/fwhR2n [254 bp] and B=BF3/BR2 [458 bp]) and technology-primer pair combinations tested.

The number of reads assigned at the species level was significantly higher with Illumina® technology (p.value < 0.00001) compared to MinION. Nearly 97% of reads were assigned at the species level for the Illumina® F combination compared to 90% for the MinION F combination (p.value < 0.0001). As for primer pair BF3/BR2, over 87.3% of reads were assigned at the species level for Illumina®, followed by over 79.7% for MinION technology and over 77.2% for IonTorrent® technology (Fig. 3). The primer pair fwhF2/fwhR2n resulted in a significantly higher percentage of reads assigned at the species level (93.6%) (considering both Illumina® and MinION technologies) compared to the couple of primers B (81.4%) (considering all three technologies) (p. value < 0.00001). The summary of the number of reads is shown in Suppl. material 1: table S9a, while the percentage is shown in Suppl. material 1: table S9b. Both tables display data assigned to the taxonomic levels of species, genus, family or higher, for all conditions tested.

Figure 3.

Proportion of reads assigned to each taxonomic level for each combination of sequencing technology and pair of primers (F: fwhF2/fwhR2n; B: BF3/BR2).

False positive detections (i.e. a species detected within a bulk that is not part of the bulk’s initial composition) were observed regardless of the combination of primers and technology (Fig. 4). The precision (the ability to measures the proportion of correct detections) of the Illumina F and MinION F combinations are 0.625 and 0.667, respectively, while for the Illumina B, MinION B and IonTorrent B combinations, the precision is 0.806, 0.926 and 0.8 respectively. An average of 13.5 false positive OTUs were recorded for the primer pair fwhF2/fwhR2n, compared to an average of four false positive OTUs when using the primer pair BF3/BR2, the difference being significant here (p value = 0.00194). According to the technology used, but regardless of the primers, an average of ten, seven and six false positives were recorded for Illumina®, MinION and IonTorrent® technologies, respectively. There are no significant differences amongst the three sequencing technologies in terms of false positives.

Figure 4.

Number of false positive detections at species level for each sequencing platform and primers used (F=fwhF2/fwhR2n [254 bp] and B=BF3/BR2 [458 bp]).

In total, 33 out of 48 individuals (68.8%) were detected at the species level by at least one experimental condition (Fig. 5).

Figure 5.

Heatmap comparing the identification of individuals present in bulk samples at the species level (green square) or the absence of detection at the species level (grey square) according to the sequencing technologies and primer pairs used (F=fwhF2/FwhR2n; B = BF3/BR2). Species names written in blue were collected using the wet method, those in green were collected using the dry method and those in dark red were hand-captured.

Bulks 1 and 2 were assembled to compare the detection rates of closely-related species under different sequencing and primer conditions. Illumina® detected seven species out of 16 (43.75%), MinION also detected seven out of 16 (43.75%) and IonTorrent® detected six species out of 16 (37.5%). No significant differences were observed amongst the different methods used (Krustal-Wallis chi-squared = 2, df = 2, p value = 0.3679).

Metabarcoding of bulks 3 and 4 aimed at comparing the ability of different sequencing technologies to detect low abundance species in the traps. All sequencing technology/primer combinations allowed for the detection of minor species: Phymatodes testaceus with a presence of 3% in bulk 3 (relative amount of DNA in the mock community) and Xylotrechus chinensis with a percentage of 0.5% in bulk 4. However, some species (although not in minority in the bulks) were not detected in one or several tests (Fig. 5). In total, Illumina® was able to detect a higher number of individuals (11/12 individuals detected) compared to MinION (9/12) and IonTorrent® (9/12).

Regarding bulks mimicking the species composition in a field trap, MinION performed better to detect and identify specimens at the species level in Bulk 6 (detecting 8/12 species (66.7%)) compared to Illumina® and IonTorrent® technologies (5/12 species (41.7%)), whereas the same number of species was detected for Bulk 5 (4/6 (66.7%)) regardless of the technology used. Nevertheless, in bulk 5, the non-native species, Cordylomera spinicornis was detected only by Illumina B. For bulk 6, the non-native species Xylotrechus chinensis was detected by all three technologies and Xylotrechus stebbingi by MinION B only.

Our results demonstrate significant differences in the mean number of detections between samples that were collected using the “dry” method (α-cypermethrin insecticide) and the “wet” method (water-diluted propylene glycol) (Wilcoxon rank-sum test, W = 74.5, p value = 0.0006342) (Fig. 6A). Indeed, 75% (9/12) of specimens collected using the “wet” MPG trapping procedure were not detected by any of the sequencing technologies. Conversely, 88.2% (30/34) of those collected using the “dry” trapping procedure (based on α-cypermethrin insecticide) were detected at least once across all technologies.

Figure 6.

Boxplots representing (A) the average number of detections according to the type of preservation used, (B) the natural logarithm scale (base e) of the average DNA concentration according to the type of preservation used, (C) the A260/280 quality ratio according to the type of preservation used and (D) the A260/230 quality ratio according to the type of preservation used. The black dots represent the outlier values (values outside the whiskers). The bold line represents the average value, outlines of the boxes represent the first and third quartiles and the whiskers represent the range of the values outside the quartiles.

Individuals captured using the “dry” method had higher DNA concentration (39 ng/µl on average (SD = 52.79)) than MPG trapped specimens (18.6 ng/µl on average (SD = 21.80)) (Wilcoxon rank-sum test, W = 123.5, p value = 0.04533) (Fig. 6B). The average A260/280 ratio was 1.9 for the “dry” method and 2 for the MPG method (Fig. 6C). However, the average A260/230 ratio of specimens trapped with the “dry” method (0.8) was higher than that of specimens captured with MPG (0.5) (Wilcoxon rank-sum test, W = 146, p value = 0.1502) (Fig. 6D).

A slightly higher number of individuals were detected and identified to species using MinION (28 specimens) compared to Illumina® (27 specimens) or IonTorrent® (24 specimens), although this difference is not significant. However, this result demonstrates that the sequencing error rates long attributed to the MinION did not impact detection rates, while allowing for the elimination of the long delays often required when sequencing is performed on other sequencing technologies (Piper et al. 2019). It must be considered that we worked on a single pair of primers (BF3/BR2) with the IonTorrent® technology, which may have reduced the number of identifications. More specifically, our results showed that the choice of primer pairs and the length of the amplicon generated led to contrasted results regarding taxonomic assignment. For example, only BF3/BR2 allowed the species-level identification of the invasive species Xylotrechus stebbingi. This difference may be due to the longer amplicon generated by this primer pair, which has more informative nucleotide sites to provide a reliable taxonomic assignment. By contrast, fwhF2/fwhR2n generated a significantly higher number of false positives than BF3/BR2 (Fig. 4). This may be because the amplicon generated by fwhF2/fwhR2n is smaller in size compared to BF3/BR2. As a result, any loss of genetic information is more likely to result in misidentification or false positives (Meusnier et al. 2008).

Regardless of the number of identified species, the Illumina® technology produced a higher percentage of reads allowing species-level identification compared to MinION or IonTorrent®. The detection of specimens at a higher taxonomic level (genus or family) can be explained by sequencing errors that produce reads with less than 98% identity to the reference database. These results confirm that Illumina has a lower sequencing error rate than Oxford Nanopore’s MinION sequencer (Piper et al. 2019), although this did not impact the number of individuals identified to the species level.

The three technologies showed similar efficiency in detecting and identifying closely-related species. Moreover, the results show that all three sequencing technologies (regardless of the associated primer pairs) enabled the detection and identification of species whose DNA represented a very low percentage in the mock community (Fig. 5). This high resolution would allow for the detection of exotic species that are poorly represented in traps, which might otherwise go unnoticed. Thus, all three technologies appear suitable for detecting and identifying species present in low numbers in field traps, enabling effective monitoring.

Both the conditions of capture (wet versus dry methods) and storage (i.e. time lag between collection and lab processing) have an impact on DNA concentration and quality and subsequently on the rate of species detection (Piper et al. 2019). Thus, the number of species detected is highly variable between bulks 1 and 2, which can be explained by the capture methods used: ‘dry,’ where individuals were captured without preservative fluid (as is the case for the majority of detected individuals comprising bulk 1) and ‘wet,’ where individuals were preserved in 50% MPG until trap retrieval (as is the case for the majority of undetected individuals comprising bulk 2). For instance, the species Phymatodes testaceus was always detected (10 out of 10 assays) when dry specimens were present, even in low concentrations (3% in Bulk 3). On the other hand, wet specimens of P. testaceus were rarely detected (one detection out of five assays). Individuals captured using the MPG method had lower DNA concentration and presented significantly lower detection rates compared to individuals captured using the “dry” method. Ballare et al. (2019) also found that insects collected in propylene glycol traps produced lower quality ddRADseq assemblages than specimens collected by net sampling and directly transferred into 100% ethanol (EtOH) or by passive trapping followed by 100% EtOH storage before pinning. In contrast to this, Ferro and Park (2013) found that propylene glycol is an effective DNA preservative for molecular marker-based studies on Coleoptera species. However, in their study, insects were first killed and preserved in 100% ethanol before being stored in glycol, while in our study insects were killed directly in propylene glycol. The use of 100% ethanol as the initial killing agent may lead to better initial preservation of specimens compared to direct exposure to propylene glycol.

Despite the precautions taken, several false positives were detected in all tested conditions. The number of false positives was significantly higher with the primer pair fwhF2/fwhR2n, which generates a smaller size amplicon compared to BF3/BR2. Even though Illumina technology is known to have a lower sequencing error rate compared to MinION (Silvestre-Ryan and Holmes 2021), our study found 10 false positives generated by Illumina, while MinION produced seven false positives and IonTorrent produced six. The sensitivity of HTS technologies allows for the detection of very small amounts of DNA, thus detecting even the slightest cross-contamination between samples (Liu et al. 2020). These DNA contaminations may have occurred during sample collection in the field or in the laboratory through cross-contamination between samples from the same study.

The false negative detections for some individuals may primarily be explained by the highly heterogeneous DNA quality of the different sequenced individuals (Suppl. material 1: table S1). In fact, DNA quality can be impacted by numerous mainly abiotic factors (pH, UV radiation, temperature), degrading DNA quality in a matter of days or weeks (Strickler et al. 2015; Collins et al. 2018; Harrison et al. 2019). During field trapping using stationary traps, captured insects are sometimes exposed to such conditions (high temperatures in trap containers when exposed to the sun in summer, high humidity in the container during heavy rains etc...), which can greatly accelerate the speed of DNA degradation in captured individuals. Such degraded DNA is more difficult to amplify, thus generating false negatives, especially when attempting to detect insects in low abundance within a trap, such as an invasive species in the process of establishing (Preston et al. 2022). Another possible cause for the high number of false negatives is the bias induced by PCR, such as uneven amplification of the DNA of the different individuals present in one sample (Preston et al. 2022). To avoid potential bias arising from identification errors or missing species in the reference databases, we decided to work on a local and curated BLAST database. However, when target species are partially unknown, as is the case in field conditions, analyses must rely on public reference databases. Yet, out of the 35,000 known species of Cerambycidae to date, only 2,926 species (8.4%) are recorded in BOLD with a barcode fragment (as of 16 November 2023). Furthermore, databases can contain errors such as mis-assignment of a DNA sequence to a wrong species due to morphological identification errors. This was precisely the error encountered for the species Monochamus sutor, which was genetically identified as Monochamus sartor (Suppl. material 1: table S7) or the species Leiopus nebulosus which has been genetically identified as Leiopus linnei (Suppl. material 1: table S7) using our local BOLD database.

One also needs to pay attention to synonymy whereby a species appears in the database under multiple names. We encountered this problem in our analysis with Arhopalus ferus (Bulks 1, 4 and 6) which was detected, but under the name of Arhopalus tristis (Suppl. material 1: table S10). Finally, mitochondrial paralogues such as NUMTs (non-functional copies of mitochondrial genes transported into the nuclear genome) present in databases can also bias results, making it impossible to identify specimens correctly at the species level (Bensasson et al. 2001). NUMTs are numerous in many organisms, including some cerambycids such as Monochamus galloprovincialis (Koutroumpa et al. 2009; Haran et al. 2015).

The differences in identification or non-detection between morphologically similar species belonging to the same genus, as observed, for example, with Monochamus spp., Phymatodes spp. or Arhopalus spp. (Fig. 5), can be explained in part by the reasons mentioned above.

Based on the results obtained, it appears that the main biases observed in metabarcoding analyses of trap contents stem from the degradation of DNA from individuals, which generates false negatives. We recommend favouring a “dry” rather than a “wet” trapping method, especially the MPG method and to plan for the collection, transportation and processing of captured individuals as soon as possible after capture. This includes checking the traps as frequently as possible (at least once a week), thus avoiding excessively long exposure of the individuals to unfavourable environmental conditions. Once individuals are brought back to the laboratory and if DNA cannot be extracted straight away, it is important to limit any further degradation by keeping samples at -20 °C and in 95% ethanol. On the other hand, DNA extractions should be stored in the preservation buffer provided with the extraction kits or in molecular-grade water and kept at -20 °C (Preston et al. 2022). We also recommend limiting the use of primer pairs that generate short amplicons, which can favour the amplification of non-target taxa, NUMTs and lead to identification errors. The quality and completeness of the databases are also very important bias factors. To limit this bias, Egeter et al. (2022) recommended restricting the database used to targeted species in order to minimise the risk of false positives due to contamination. Limited taxonomic and geographical coverage of sequence databases is a huge limitation in metabarcoding studies. For example, Dopheide et al. (2019) found no representative sequence in the GenBank database for more than 900 invertebrate OTUs in their study when analysing the community of soil arthropods from a native forest in Ireland. Additionally, species identification errors and cases of synonymy lead to false negatives or cases of multiple affiliations.

The authors have declared that no competing interests exist.

No ethical statement was reported.

This work was supported by the PORTRAP project “Test de l’efficacité de pièges génériques multicomposés pour la détection précoce d’insectes exotiques xylophages dans les sites potentiels d’entrée sur le territoire national” and HOMED project (HOlistic Management of Emerging Forest Pests and Diseases) which received funding from the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No. 771271 (https://homed-projecteu/). We are grateful to the genotoul bioinformatics platform Toulouse Midi-Pyrenees for providing help and computing storage resources. Loïs Veillat was supported by a PhD studentship from HOMED project and doctoral school SSBCV at the University of Orléans.

Loïs Veillat, Géraldine Roux, Carlos Lopez-Vaamonde and Stéphane Boyer conceived the study. Alain Roques collected field samples. Stéphane Boyer, Marina Querejeta, Emmanuelle Magnoux and Loïs Veillat conducted the laboratory sample processing. Loïs Veillat analysed the data and wrote the first draft. All authors contributed to the preparation of the manuscript. Both senior authors, Géraldine Roux and Carlos Lopez-Vaamonde, contributed equally to this study.

Loïs Veillat https://orcid.org/0009-0004-2149-1336

Stéphane Boyer https://orcid.org/0000-0002-0750-4864

Marina Querejeta https://orcid.org/0000-0003-1803-5239

Emmanuelle Magnoux https://orcid.org/0000-0003-0990-5511

Alain Roques https://orcid.org/0000-0002-3734-3918

Carlos Lopez-Vaamonde https://orcid.org/0000-0003-2278-2368

Geraldine Roux https://orcid.org/0000-0002-1116-2799

Barcode data for the 33 species used in the mock community experiment are available from BOLD in the dataset DS-MINION (dx.doi.org/10.5883/DS-MINION). Raw sequence data for this project and analytical script and files are available on figshare (https://figshare.com/projects/DNA_metabarcoding_an_efficient_way_to_detect_non-native_cerambycid_beetles_in_trapping_collections_/171432).