The introduction and establishment of non-native forest insects is considered to be one of the greatest threats to forest health (Rabaglia et al. 2008, Ramsfield et al. 2016). In response to the potential of non-native forest pests being introduced into Alaska, an Early Detection and Rapid Response (EDRR) program has been implemented to detect, delimit and monitor newly introduced non-native bark and wood-boring beetles at selected high-risk forest areas and quickly assess and respond to new infestations.

A methodological bottleneck constraining the EDRR program in Alaska is the time and expertise required to process EDRR trap catches, limiting the number of traps that can be deployed each season. A potential solution to this taxonomic bottleneck is the use of recently developed metabarcoding methods. Biomonitoring by metabarcoding has been advocated for arthropods because these methods have the potential to be much faster and less costly than identifications obtained by morphology (Hajibabaei et al. 2011, Baird and Hajibabaei 2012, Watts et al. 2019). Metabarcoding methods are already being adopted for biomonitoring of invertebrates (Gibson et al. 2015, Hajibabaei et al. 2016, Bush et al. 2019). Identifications obtained through metabarcoding should be of better taxonomic resolution in Alaska, where a deliberate effort has been made to construct a reference library of DNA barcode sequences useful for species identifications of terrestrial arthropods (Sikes et al. 2017), than in regions where such libraries are lacking (see Watts et al. 2019).

We sought to compare EDRR trapping results obtained by conventional means and by metabarcoding to determine if metabarcoding methods would be more appropriate for EDRR monitoring than methods currently used.

We sought to publish all of our data following the guidlines of Penev et al. (2017). Our reference library and scripts from producing it are available at https://github.com/mlbowser/AKTerrInvCOILib. Complete specimen and occurrence data are available via an Arctos (https://arctosdb.org/) archive at https://arctos.database.museum/archive/2017_edrr_ngs_test_records and are also available on GBIF (https://www.gbif.org/) via Arctos. Sequence data have been deposited in the NCBI Sequence Read Archive under BioProject PRJNA542936 and original, raw FASTQ files are provided in Bowser et al. (2019). Complete occurrence and identification data are provided in Suppl. material 1. Sequences of ASVs are provided in Suppl. material 4 and the ASV table is provided in Suppl. material 3.

Assemblages of bark and wood-boring beetles detected by both methods were largely congruent, composed mostly of Curculionidae and Cerambycidae. High Throughput Sequencing consistently yielded a higher diversity of taxa and provided identifications mostly at the species or BIN resolution; in contrast, morphological methods yielded lower diversity, with identifications mostly at the resolution of genera.

A total of 85 unique identifications were obtained from both methods combined (Table 2, Fig. 2), including 52 species or BIN resolution identifications. Morophological methods yielded 30 unique identifications of which 4 were at the species resolution besides the three exotic species that were added and known a priori. High Throughput Sequencing yielded 59 unique identifications including 46 species or BIN resolution identifications (see phylogenetic tree, Fig. 3). None of the three exotic species that had been added to the samples were detected by HTS. Only a single species, the cerambycid Acmaeops proteus (Kirby 1837), was identified by both methods.

In many cases, coarser identifications at genus resolution from the morphology-based dataset corresponded with species identifications from the HTS data. Molecular identifications of Neospondylis upiformis (Mannerheim, 1843), Dolurgus pumilus (Mannerheim, 1843), and Orthotomicus caelatus (Eichhoff, 1868) corresponded exactly with identifcations of these genera based on morphology. Identifications of Thanasimus undatulus (Say, 1835) and Hemicoelus carinatus (Say, 1823) obtained by HTS corresponded to morphological identifications of Cleridae and Ptinidae, respectively. More generally, HTS identfications of Cryphalus ruficollis Hopkins, 1915; Dendroctonus rufipennis Kirby, 1837; Dryocoetes affaber Leconte, 1876; Hylurgops rugipennis (Mannerheim, 1843); and Trypodendron lineatum (Olivier, 1795) mostly corresponded to morphological identifications of these genera.

In the three samples where bycatch was included, 14 species or BINs of flies, the acanthosomatid bug Elasmostethus interstinctus (Linnaeus, 1758), and an ichneumonid wasp identified as Ichneumoninae sp. BOLD:AAU8831 were also identified. Detections of the targeted scolytine and cerambycid beetles were not notably reduced in these samples.

High Throughput Sequencing also yielded detections of nematodes including an ASV identified as Bursaphelenchus sp.

From the perspective of work hours required by our team, the 1.3 hours per sample for processing HTS data was 65% of the 2.1 hours per sample required for identification using morphology.

Although HTS methods outperformed morphological identfication methods in this case in terms of taxonomic resolution of identifications, it should be noted that only the initial steps of a typical EDRR processing workflow were followed. If specimens had been sent out to taxonomic specialists, then most bark and wood-boring beetle specimens would have received species-resolution identifications. Obtaining expert identifications would have also substantially increased processing costs and processing time.

In terms of processing time, HTS methods outperformed morphological methods, with HTS methods requiring 65% of the work hours needed for the morphological identifications. However, this does not take into account processing time at the sequencing lab. In this example, results from the morphological dataset were available 4 days after processing was commenced by a team of workeers while HTS methods required 1 day for shipping, 73 days for processing at the sequencing lab, and about 5 work days for metagenomic processing. Both methods returned results before the next growing season, but morphological methods would have allowed for a rapid response in the fall almost 3 months before the results from HTS were available. However, in more recent sampling efforts comparable sequencing has required only 4 weeks from the sequencing lab. Also, HTS processing time per sample should decrease as the sample size is increased because much of the pipeline can be run in parallel.

We did not attempt to quantify rates of false presences and false absences in the HTS dataset, but it was clear that some taxa observed in the morphological dataset and the three known exotic species in particular were not detected by HTS. We were not surprised that the preserved specimens of Ips typographus and Tetropium fuscum were not detected because these specimens may have had degraded DNA and they represented small portions of the samples. We had expected that the live specimen of Halyomorpha halys, a relatively large insect where DNA degredation should not have been a problem, would have been detected. Potential causes for this non-detection include a failure to homogenize the sample completely or failure to amplify sequences of this species due to primer bias.

In any single primer pair, differences in binding to DNA templates lead to amplification biases, affecting both read abundances and detections of species so that any single primer set will lead to detections of a subset of species (Hajibabaei et al. 2019, Elbrecht and Leese 2017). The mlCOIlintF/HCO2198 primer pair we used amplifies well across a broad range of arthropod taxa (Leray et al. 2013, Brandon-Mong et al. 2015), making it an appropriate choice for surveillance monitoring (as defined by Nichols and Williams 2006) of arthropods. In future efforts we would select the mlCOIlintF/jgHCO2198 primer pair, which has a more degenerate reverse primer and amplifies well across a broader range of arthropod taxa than the mlCOIlintF/HCO2198 pair (Elbrecht and Leese 2017). To optimize EDRR efforts for Curculionidae and Cerambycidae, an appropriate next step would be to compare performance of additional primer sets for amplyfying DNA of these target groups.

In future applications of HTS for biomonitoring we would consider using the SCVUC COI metabarcode pipeline (https://github.com/EcoBiomics-Zoobiome/SCVUC_COI_metabarcode_pipeline) used by Hajibabaei et al. (2019) instead of the QIIME 2 pipeline we employed.

Because our HTS methods failed to detect Ips typographus and Tetropium fuscum, two species of concern for our EDRR monitoring program, we do not recommend wholly replacing our current morphological monitoring methods with HTS methods. Even though our morphological identifcations were mostly at the taxonomic resolution of genera, we believe that we would be more likely to detect certain exotic species by using morphology than by using HTS. However, HTS methods would be especially appropriate as a complement to current EDRR methods for survellaince monitoring. Our detections of two new exotic species for Alaska highlight the effectiveness of HTS methods for detecting species that were not being looked for. In addition, because the results of both methods were consistent overall, HTS methods would be appropriate for increasing sample size without greatly increasing the time required to process specimens, especially if bycatch is included, removing the time-consuming manual sorting step.