Borrowing the Pentatomomorpha tome from the DNA barcode library: Scanning the overall performance of cox1 as a tool

2.1 Data acquisition and filtering

Sequences were retrieved from BOLD on August 17, 2020, generating separate datasets for each pentatomomorphan family. We followed several filtering steps to ensure robust analyses (summarized in Figure 2). First, we removed sequences belonging to other markers (or regions) than COI-5P (e.g., COI-3P, COXII) or without species-level identification (e.g., “Euschistus,” “Pentatomidae sp.”). Then, we made preliminary alignments of each dataset using MAFFT 7.0 (Katoh et al., 2019), keeping the parameters in default, and looking for non-sense mutations, insertions, and deletions. These sequences were removed from the dataset since we assumed they resulted from low-quality sequencing, erroneous amplification, or laboratory contamination. Additionally, misaligned sequences were used as a query for comparison with the NCBI database (http://www.ncbi.nlm.nih.gov/) using the Basic Local Alignment Search Tool—Nucleotide (BLASTn) to verify their identity. Non-pentatomomorphan sequences were purged from the dataset.

After these steps, we realigned the sequences (same parameters as above). The software AliView (Larsson, 2014) was used to visualize the alignments, verify the reading frame, and trim the sequences to restrict them to the canonical barcode region of cox1 (Ratnasingham & Hebert, 2007). Sequences shorter than 400 base pairs were removed. As a final filtering step, we double-checked scientific names seeking for misspellings and non-valid names (i.e., synonyms). Then, we ran a final alignment for each dataset.

2.2 Data analyses

For the barcoding gap analysis, we generated separate FASTA files for each genus represented in the datasets. We only analyzed genera featuring at least two species, and with at least one of the species being represented by more than one sequence, to ensure intra- and interspecific comparisons. For each genus, the R package SPIDER (Brown et al., 2012) was used to estimate uncorrected p-distances. Output values were visualized in a boxplot and we followed Badotti et al., (2017) to rank cox1 efficiency based on three categories: good, intermediate, and poor. Efficiency was considered good when boxplots disclosed a gap between intra- and interspecific distances, intermediate whenever the whiskers of intra- and interspecific distances overlapped, and poor when the boxes overlapped. We additionally used the function “localMinima” implemented in SPIDER to determine a threshold value for genus with a good cox1 efficiency. This function optimizes putative threshold values from a density plot of the genetic distances (Brown et al., 2012).

Since the success of DNA barcoding for specimen identification is not necessarily related to a barcoding gap (see Discussion), we calculated the PCI (Hollingsworth et al., 2009) to measure the identification success when using cox1 sequences. For each species, we considered the maximum intraspecific distance and the minimum interspecific distance (i.e., nearest-neighbor distance). If the maximum intraspecific distance of a species was less than its minimum interspecific distance, identification for that species was considered a success (Hollingsworth et al., 2009). PCI was calculated as the percentage of species correctly identified. Species represented by only one sequence were not analyzed, since it was not possible to obtain intraspecific distances in such cases. PCI rates were then graphically represented in a scatter plot, as suggested by Collins and Cruickshank (2012).

Since some family groups and their rankings are not consensual with current classification, we opted to present data for Thaumastellinae and Thyreocorinae, treated as subfamilies of Cydnidae for some authors (e.g., Schuh & Weirauch, 2020) but as independent families in other classifications (e.g., Grazia et al., 2008; Lis et al., 2017; Rider et al., 2018). Here, we adopted the classification of Pentatomomorpha and numbers of species and genera per family presented by Schuh and Weirauch (2020).

4.1 How is the Pentatomomorpha tome sorted in the BOLD library?

The available data on BOLD are biased to families with economic importance. Economic losses leveraged molecular studies through cox1 to refine the knowledge on crop pest species (e.g., Barman et al., 2017; Gariepy et al., 2019; Soares et al., 2018). For instance, established biosafety and quarantine protocols use DNA barcoding to assess pest diversity on a global scale (Boykin et al., 2012; Lee et al., 2019). For pentatomomorphans, of which egg and immature identification is a challenging task (Lis et al., 2013; Matesco et al., 2014), the diagnosis using this molecular tool may promote a fast and accurate species identification, playing a pivotal role in crop pest management.

Besides being the most speciose family within Pentatomomorpha (Henry, 2017), Pentatomidae encompasses numerous crop pests (Schaefer & Panizzi, 2000). The pentatomids brown marmorated stink bug [Halyomorpha halys (Stål)], green stink bug [Chinavia hilaris (Say)], dusky stink bug [Euschistus tristigmus (Say)], and Neotropical brown stink bug [Euschistus heros (Fabricius)] sum alone almost 1400 sequences in our final datasets. They are aggressive crops in different regions. Chinavia hilaris and E. tristigmus occur in North America, while E. heros occurs mostly in South America (Schaefer & Panizzi, 2000; Smaniotto & Panizzi, 2015). The H. halys is native from Asia and has invaded other continents [i.e., America (Hoebeke & Carter, 2003) and Europe (Wermelinger et al., 2008)] The brown marmorated stink bug shows rapid spreading throughout invaded territories (Zhu et al., 2012) and seems to be the most prominent pentatomid pest in the past two decades.

The second most representative family in our sample was Lygaeidae, with Nysius Dallas comprising more than half of the sequences for the family, and in fact, being the genus with most analyzed sequences (1695 sequences) among our datasets. Although Nysius groups around a hundred valid species (Malipatil, 2010), our dataset presents only 17 species labels. Nysius binotatus (Germar) (as commonly found on BOLD) is currently a subspecies of Nysius senecionis (Schilling) (treated here only by the binomen), and this is the species with the greatest contribution of sequences for our dataset (1116 sequences). These data are from exploratory efforts on Saharo-Arabian region with year-long deployment of Malaise traps and show the high abundance of N. senecionis in the region. Nysius species occur worldwide and feed on the seeds of a broad range of economically important host plants (Ge & Li, 2019; Schaefer & Panizzi, 2000). The birch catkin bug [Kleidocerys resedae (Panzer)] is another lygaeid species prevalent in our datasets (396 sequences). This species has a Holarctic distribution and is a pest of birch and alder trees, feeding and breeding on the seed catkin (Dioli et al., 2019; Schaefer & Panizzi, 2000).

Only some small families lack cox1 data in BOLD (Table 1). Most of them are endemic to one zoogeographical region, are rare, and have little knowledge of their natural histories (e.g., Canopidae, Cryptorhamphidae, Hyocephalidae). Although Termitaphididae is widespread, occurring in Australian, Afrotropical, Neotropical, and Oriental regions (Schuh & Weirauch, 2020), it is also a small family, with nine species without barcoding data. The lack of barcoding data for Idiostoloidea mirrors the scientific knowledge for its six species: Little is known concerning their natural history, systematics, and evolution (Schuh & Weirauch, 2020). Within the Pentatomomorpha tome, many taxa are blank pages waiting to be filled with barcode sequences.

4.2 Using thresholds to discover species by the book

Some DNA barcoding analyses (e.g., gap discovery, threshold values) perform better if based on curated libraries (Meyer & Paulay, 2005) due to their sample dependency. The singular coalescent depths of each lineage (e.g., genus, species, population) produce a practical challenge to cluster/split sequences in operational taxonomic units (Fujita et al., 2012; Zinger & Philippe, 2016). Hence, assuming a fixed threshold value for higher taxonomic levels is unrealistic and broadly accepted that a comprehensive haplotype sampling and species coverage leads to more reliable values (e.g., DeSalle & Goldstein, 2019; Meyer & Paulay, 2005). The use of empirical data is the best first step to look for barcoding gaps and threshold values (e.g., Gonçalves et al., 2021). The amassing of data for a lineage will progressively reduce inaccurate identification, false positives, false negatives (Meyer & Paulay, 2005) and shed light on cryptic species and specimen identification accuracy. It is convenient to review these values whenever new samples are available in databases (see Qing et al., 2020).

Arbitrary fixed threshold values have been used and tested in Pentatomomorpha (e.g., Park et al., 2011; Zhao et al., 2018). Previous research with Cletus Stål species showed that the use of popular fixed threshold values [e.g., 1% BOLD (Ratnasingham & Hebert, 2007) and 2% (Hebert et al., 2003)] would overestimate the genus diversity (Zhang et al., 2017). The authors suggest a threshold value above 7% based on the minimum interspecific distance, which would help to solve the overestimation of diversity. Our result for the minimum interspecific distance was similar to that (6.86%), and the local minimum was 3.30%. Here we provide both minimum interspecific distance and local minima for three out of four tested genera (Appendix 1). Future studies may benefit from these empirical values, since they are more reliable than an arbitrary fixed threshold. As the taxonomic knowledge for the taxa under study is refined and new sequences are generated, reassessing these values will increase their reliability and improve their practical applications,

The local minimum represents the barcoding gap, where a consistent dip in the density of genetic distances split intra- and interspecific distances is observed (Brown et al., 2012). Although the use of minimum interspecific distance is widespread in the literature as a threshold value (e.g., Meier et al., 2008; Song, Lin, Wang, & Wang, 2018), this may result in misidentifications when sequences from different species have distance values falling below this threshold, increasing the chance of false negatives. Since we understand DNA barcoding as a source of evidence and not the only way to make taxonomic decisions, the local minimum provides more rigor to identify molecular entities (see Casiraghi et al., 2010). It would flag more samples to be explored with other evidence, allowing more robust taxonomic decisions (e.g., identification, species description, synonyms).

The PCI and barcoding gap analysis are methods based on genetic distances that work in distinct scopes. While the PCI identifies discrepancies on the cohesion of the data and points out putative misidentification, the barcoding gap analysis shows the distribution of intra- and interspecific distances disclosing the existence (or not) of a gap. Higher PCI values tend to lead to good barcoding performances (Figure 5), but the relationship between both analyses is not obvious. For instance, Leptocorisa Latreille (Alydidae), Catorhintha Stål (Coreidae), Neortholomus Hamilton (Lygaeidae), and Padaeus Stål (Pentatomidae) presented 0% of PCI although a good barcoding performance. It means the pool of samples for each taxa presented a clear gap, although these specimens seem to be wrongly labeled. The opposite scenario is also possible, where DNA barcoding can be used for specimen identification without a consistent barcoding gap (see Collins & Cruickshank, 2012). Thus, these methods are complementary to taxonomic approaches using barcoding.

Most superfamilies analyzed here presented PCI values around 70%, but Aradoidea showed 94.34%. Grebennikov and Heiss (2014) and Grebennikov and Heiss (2018) are responsible for most of Aradidae sequences deposited in BOLD. The authors built a reference barcoding library for flat bugs (Aradidae) species and identified many cryptic new species exploring the Tanzanian fauna. High PCI values reflect rigorous taxonomic treatment. Although DNA barcoding may speed up the taxonomy, this is only possible (or at least should be) when molecular data meet taxonomic experts—but the training of such taxonomists is a long-term commitment that usually takes many years (Coleman & Radulovici, 2020). The slow increase in taxonomists is incompatible with the increase in articles dependent on taxonomic services (Wang et al., 2020). Specimen identification by non-specialists can be risky, and it is a source for many misidentification in databases.

Misidentification is an issue inherent to any public databases, where there is a claim for accurate identifications and rigorous curation (van den Burg et al., 2020). Ideally, all sequences in BOLD should be derived from vouchered specimens identified by taxonomic experts (Lis et al., 2016; Meiklejohn et al., 2019) and obtained through rigorous laboratory protocols. DNA barcoding success may also be affected by non-biological processes, referred by Mutanen et al., (2016) as sources of operational bias. These include inaccurate reference taxonomy, misidentifications, amplification of non-target sequences, alignment errors, erroneous sequencing due to contamination, and other methodological issues. Operational biases impact the reliability of the libraries and complicate the taxonomy and systematics of species (Mutanen et al., 2016).

We emphasize that the threshold values suggested here are not a “magic bullet” for species discrimination, but a source of evidence to this. Species delimitation based on DNA barcodes is hampered, for instance, in scenarios of hybridization or introgression, incomplete lineage sorting, and bacterial endosymbionts changing pathways of mtDNA inheritance (Magoga et al., 2018; Mutanen et al., 2016). Thus, this kind of approach as the single evidence for taxonomic decisions should be avoided. We advocate for the use of integrative approaches, which produce stronger species hypotheses (Padial et al., 2010).

4.3 Empirical examples: How may these data clue future studies?

4.4 Closing the book, for now

Molecular data changed drastically the way science produces and evaluates biological knowledge. Since the description and understanding of biodiversity are remarkably incomplete (Costello et al., 2013; Sheth & Thaker, 2017), the approaches to optimize robust species delimitation and specimen identification are desired. Here, we provide an overview of the available cox1 information for Pentatomomorpha deposited in BOLD. After inspecting more than 12,000 sequences, we verified for the first time the existence of barcoding gaps and provided individual threshold values for 132 genera, also casting doubt on the label on many specimens. Our results may be used to mitigate the Linnean shortfall—the knowledge gap between formally described species and the number of extant species. The Linnean shortfall underpins the biological knowledge, since the underlying recognition of the biological units precedes any study (see Hortal et al., 2015).

The use of cox1 as a taxonomic tool is copious, and our empirical examples reaffirm the efficiency of DNA barcoding to identify taxonomic inconsistencies. A careful handle of the data (i.e., samplings, specimens identification, laboratory protocols, or deposit of sequences in libraries) and interpretation of the results are mandatory. The simple act to double-check for invalid names and misspellings may denoise analyses and improve results. Here, we flag many genera for further investigations based on the state of knowledge of this infraorder. Future studies, including new sequences and species, may refine values and sharpen the accuracy of barcode tools.