A DNA barcoding framework for taxonomic verification in the Darwin Tree of Life Project

1The views expressed in this article are those of the authors. Publication in Wellcome Open Research does not imply endorsement by Wellcome.

2Species identification is central to biodiversity research but is hampered by a multitude of challenges, from biological issues such as cryptic diversity that prevent the unambiguous assignment of names in complex species groups, to technical issues associated with degraded or incomplete specimens. Species identification becomes ever more important in the era of biodiversity genomics, where a large investment is made to collect, sequence and assemble complete genomes, and where mistaken identification could cause pervasive issues downstream. DNA barcoding - the sequencing of standard DNA regions to differentiate species - is one approach that may provide a valuable independent confirmation of species identity in large biodiversity genomic projects. DNA barcoding has been widely used to inform species identification and for estimating species diversity in a wide range of studies, ranging from ecological forensics, vegetation surveys, community phylogenetics and environmental monitoring, but is not routinely implemented as an identification tool in most studies generating complete genomes. However, DNA barcoding has great promise for rapidly confirming species identities before a sample enters the genome sequencing pipeline.

3Here, we develop a framework for DNA barcoding for the Darwin Tree of Life (DToL) project, and consider how this approach may be adopted by other large-scale biodiversity genome initiatives. DToL has the aim of sequencing and assembling high quality reference genomes for all eukaryotic species present in Britain and Ireland. As a large-scale project sampling the full regional diversity of eukaryotic life, DToL faces many potential issues with species identification. Initially, all specimens are identified in the field at the point of collection by taxonomic experts. In most cases, the initial taxonomic identification will be correct. However, taxonomic complexes and cryptic species in particular pose problems, where species misidentification or uncertainty must be anticipated. This challenge is especially acute in certain groups of organisms. This includes fungi, where morphological characters alone often fail to tell species apart, as well as many arthropods, where identification is often impossible in the field as it requires examination of internal or concealed structures. The original species identifications are therefore verified by DNA barcoding using taxon appropriate loci (mitochondrial, plastid or ribosomal RNA). Moreover, despite centuries of intensive study of the natural history of Britain and Ireland, the integration of new morphological observations with DNA barcoding and genomic sequencing will inevitably lead to species discovery and taxonomic change. This is particularly pertinent as new species colonise Britain and Ireland due to range expansion driven by factors such as anthropogenic introductions and changes in climate. Finally, the scale of sampling and the extensive set of downstream laboratory processes in DToL and other large biodiversity genomic initiatives make the potential for sample tracking an important benefit of generating barcodes prior to sequence submission. Here, a comparison between the DNA barcode generated after specimen collection, and the barcode sequence recovered from the genome assembly, can provide a valuable check for such mistakes.

4DNA barcoding for DToL is aided by the extensive DNA barcoding reference datasets that have been generated for diverse British and Irish species, although there are notable limitations of the currently available data. Natural England commissioned a comprehensive assessment of the DNA sequences available in the major DNA reference libraries for approximately 76,000 eukaryotic species in the UK. The results of this formal gap analysis revealed that 52% of UK species had some publicly accessible DNA sequence data, varying widely across different databases (e.g. the Barcoding of Life Data System, BOLD; Genbank; European Nucleotide Archive, ENA). Within BOLD, for example, at least one barcode sequence was available for 42.5% of UK species, however in many cases this was not from a specimen collected from the UK. Strikingly, imposing stringent quality standards on the barcoding data accessible in BOLD resulted in a reduction of UK species coverage to just 4%. Although around half of UK species are represented in reference databases such as BOLD, that figure is significantly taxonomically biased toward, for example, invasive species, or those with a conservation status. Unsurprisingly many species that are challenging to identify (e.g. species that are small or show cryptic species differences), or rarely encountered, are poorly represented in reference libraries, or may be misidentified. DNA sequence quality, metadata completeness, and the accuracy of species identification varied markedly both within and between reference libraries. The paucity of funding, expertise, and capacity across various domains, including taxonomy, molecular laboratory techniques, bioinformatics, as well as quality control and assurance have been identified as the main challenges in building up more accurate and complete barcoding reference libraries. To this end, DToL offers a unique opportunity to leverage its expertise in these areas to expand the coverage of UK species with high-quality, rigorously verified data.

5The DToL process starts with the collection of samples by specialist collectors at ‘Genome Acquisition Laboratories’ (GALs) from diverse locations across Britain and Ireland, including field sites and from living collections. A specimen is identified to species level at the point of collection, or soon after in the laboratory, based on morphological characteristics by a taxonomic expert. This taxon identification is stored with other key metadata in a standardised sample manifest. Where possible, tissue for DNA barcoding is taken from the specimen that is used for genomic sequencing. For some taxa, additional specimens are collected as morphological and DNA barcode vouchers and to provide additional material for sequencing if required at a later date. In other cases, small samples or organisms need to be cultured to bulk up enough biomass (e.g., some fungi, protists), increasing the amount of material that requires DNA barcoding to ensure unwanted contaminants are not genome sequenced.

6No single set of DNA barcoding procedures would be suitable for the diversity of organisms processed by DToL. For example, the standard processing procedure for adult terrestrial arthropods involves the removal of one (or more) legs which are preserved in 70% ethanol prior to amplification of the CO1 region, whereas for vascular plants above-ground tissue is stored in desiccating silica gel prior to amplification of rbcL and ITS2 (or rbcL and trnL-F for ferns). In fungi, tissue or spores are plated in petri dishes for in vitro culturing and subsequently identified with ITS. After sequence generation, the process of interpreting the data differs depending on the preferred search query approach, the extent of the reference database, and the ability of a given locus to tell species apart. Despite differences in the interpretation process, a general set of rules are necessary across organismal groups to ensure consistent working practices and to allow DNA barcoding to operate efficiently at scale. As such, the development of DToL DNA barcoding aimed to:

7Establish a network of DToL DNA barcoding hubs that can process and generate sequences for diverse samples at scale.

8Put in place a flexible bioinformatic workflow that can be tailored to the analysis and interpretation of diverse samples.

9Establish principles for data sharing and publication that allow rapid dissemination of data in a fair and equitable manner.

10Our development is focused around the strategy outlined in Figure 1, where a general and standardised set of practices are adapted to: (1) organism specific tissue processing and DNA extraction needs, including variation in specimen size, tissue type, and tissue preservation method, (2) PCR amplification requirements and consideration of the most suitable primers, (3) the best suited bioinformatic approach and reference database.

11Finally, as there are a small number of taxa where morphological identification is certain, but barcode sequencing is known to have very low success, we have instituted a barcoding exemption list. This allows samples from these small number of groups to be submitted without barcoding being attempted. The list is curated by the DToL taxon working groups, and kept under review as new approaches may enable currently intractable groups to be barcode sequenced (discussed more, below).

12To establish protocols and deliver rapid DNA barcoding across a wide phylogenetic diversity of samples, we established a network of four core DNA barcoding hubs, each with a taxonomic focus ( Table 1). These core DNA barcoding hubs work with four satellite hubs and analysis partners providing specific additional expertise. Each hub is tasked with adapting the core DNA barcoding workflow ( Figure 1) for their nominated taxon groups, to ensure suitable data standards and sequencing success rates given the specific challenges of a given taxon group. A guiding principle is that for DNA barcoding to be delivered at scale, it must follow clearly defined standard operating procedures defined at a coarse taxonomic level, rather than requiring species-specific optimisation and manual intervention wherever possible. Full protocols for each hub are available here: https://protocols.io/view/dnabarcoding-sops-for-thedarwin-tree-of-life-proc4yeyxte.

13Each hub receives DNA barcoding tissue samples or PCR amplicons from the GALs, along with sample metadata, ready for processing. These samples are usually small and shipped at room temperature as there are relatively low input quantity and quality requirements for barcoding. All hubs perform their own DNA extractions tailored to the broad organismal group, as well as PCR reactions with one of the current catalogue of taxon-specific primer sets ( Table 2).

14Sequencing is subsequently performed using one of two approaches. Sanger sequencing has been adopted at most hubs as it enables rapid turnaround and is flexible for managing small and variable batch sizes encountered with field-collections. More recently, the NHM is in the process of transitioning to Oxford Nanopore Technologies (ONT), with ~250 samples uniquely indexed and multiplexed in a single library, which is sequenced using the Flongle adaptor for the ONT MinION. This approach is especially useful when dealing with large numbers of samples that use the same primers, with added benefits including information gained in samples with a mix of organisms. There are also cost benefits, with this implementation of ONT with suitably high multiplexing costing a third of Sanger sequencing (£2.23 vs £6.60 per sample), in line with the costings previously described by. RBGE and other hubs are also testing the efficacy and scalability of ONT. Following data production common data standards have been adopted by all hubs to ensure consistency and quality. For PCR amplicons that are Sanger sequenced this means having both directions sequenced as standard, with low quality bases trimmed, while for ONT this means filtering low quality bases.

15Following sequence data production and data quality control, sequences move to bioinformatic analysis. The bioinformatics workflow for DToL DNA barcoding varies between barcoding hubs, and by technology. For example, the NHM Sanger sequencing pipeline starts with processing of the sequencing data (AB1 trace files) with an in-house Nextflow analysis pipeline, including trimming of adapters and poor quality bases with Trimmomatic, and forward and reverse strand merging with Pipebar. Subsequently BOLD and NCBI database queries are made with bold_identification and Blastn. In contrast, ONT sequencing data are processed using the ONTbarcoder software, followed by database queries using BOLDigger.

16The approach used for interpreting sequence queries creates numerous challenges for DNA barcoding. The efficacy of DNA barcoding for confirming species identification is taxon specific and depends on biological attributes of a taxon group as well as the availability of key contextual information. Firstly, there are intrinsic differences in the efficacy of DNA barcoding for identification in different organismal groups. Groups where DNA barcode sequences track species boundaries can be interpreted with greater confidence than groups where DNA barcodes offer poor species discrimination (such as where hybridisation is rampant and/or where speciation is recent, or where the barcode regions show lower levels of sequence divergence, ). Secondly, query resolution is linked to the completeness of the DNA barcode reference library, with different interpretation required for groups with near-complete taxon coverage in the reference database, as opposed to those with extensive missing data.

17Given the complex nature of species differences and the differing availability of reference data, DToL has adopted a suite of bioinformatics approaches that are chosen for different taxonomic groups. These are:

18The ‘closest matches’ approach. A sequence search using BOLD, GenBank, SILVA, UNITE or other database, followed by looking at the closest matches in the results table.

19Use of the Barcode Index Number (BIN) system. A sequence search is made in BOLD, where search sequences falling within a BIN threshold are considered consistent matches. Only currently relevant for animals using the CO1 barcode region.

20The ‘user specified threshold’ approach. A sequence search, where matches below a group-specific percentage divergence threshold are treated as consistent matches. This approach is most relevant for groups with patchy taxon representation in the database.

21Phylogenetic approach. A phylogenetic tree that places the search sequence either with the same species in the reference dataset, or with the same genus, is considered a definitive match or a consistent match, respectively.

22Based on these analyses, each query falls into a different match type, summarised in Table 3.

23Samples with a definitive match proceed to shipping and genome sequencing. The default outcome for samples with a consistent match or no match is to proceed to shipping, however, individual barcoding hubs may choose to further investigate some specimens, for example if: (1) the taxonomic expert wants to revisit the specimen name in light of the barcode data, (2) if a ‘no match’ result is returned, when some level of match may be expected. Samples with an inconsistent match, on the other hand, have a default outcome of manual intervention and sample verification. The specific actions will vary, but will typically include checking for sample switches in the lab, repeating DNA barcoding, checking unmerged forward and reverse reads, and/or checking the name associated with the specimen. In some cases, the interpretation of DNA barcode data can be more challenging, and a simple search within a public library or repository is not sufficient. It often requires systematic expertise and the placement of that DNA barcode in a phylogenetic context for an accurate ID, a task only possible when taxonomic experts are at hand, such as in DToL.

24Each of these outcomes, above, are determined by the barcoding hub (or analysis partner) for a defined taxonomic group and a specific bioinformatic approach, to ensure samples are handled consistently. For example, in flowering plants where two barcoding loci are used, and where the barcoding loci have different levels of resolution, the interpretation categories can be augmented as follows:

25Definitive match. At least one locus has a definitive match, and other locus/loci that do not have a definitive match have at least a consistent match (e.g. genus-level match) and this lower level of resolution is as expected for that barcoding locus.

26Consistent match. At least one locus has a match to a congeneric sample (the other allowed to be missing data/no match).

27Inconsistent match. At least one locus has a clear match to the ‘wrong taxon’.

28No match. No sequences from the same genus/family are available for either locus in the reference database.

29All data are submitted and made fully accessible in public databases once they have passed initial quality control checks of sequence quality and specimen identity. Barcoding data for animals, plants and fungi are deposited in BOLD, providing wider accessibility of the sequence data. BOLD allows the upload of sequences and trace files, along with metadata, that are then queried in the context of millions of available reference sequences. Our approach is for data to be uploaded at the point of production, with specimen names amended as required after sequence searches and evaluation. DNA barcoding hubs are encouraged to maintain sequence data in BOLD within the “Darwin Tree of Life [DTOL]” container and to tag them with ‘DToL’, improving searchability of DToL data. For taxa not supported by BOLD, such as protists, samples are submitted to INSDC.

30At present, BOLD supports taxonomic searches for CO1 for animals, rbcL and matK for plants, and ITS for fungi. For protists, where the V4 or V9 region of 18S rRNA gene is sequenced, we have chosen to query the databases at SILVA or PR2, while for fungi, we additionally use UNITE given the reduced amount of fungal sequences in BOLD.

31One additional application of the DNA barcoding data is for sample tracking through the genome production pipeline at the Sanger Institute. Currently, the relevant DNA barcode sequences are extracted from the raw Pacific Biosciences (PacBio) High Fidelity (HiFi) reads at the point of production, searched in the relevant database (BOLD in most cases), and the top matches are examined. The expectation is that the top match will be from the same species as the query specimen. These barcode query summaries are reported in the Tree of Life QC interface ( https://tolqc.cog.sanger.ac.uk/). Since manifest version 2.5 (implemented December 2023), BOLD specimen ID has been submitted with other sample metadata by the GALs. In future updates of the data portal, this will allow the specific barcode sequence from the genome specimens to be recovered from BOLD, improving sample tracking. For protists submitted in INSDC that do not have a BOLD ID, DToL sequence identifiers are linked to INSDC sequence identifiers via a simple two column barcode manifest.

32The individual DNA barcode sequences generated by DToL primarily support taxon verification, with each verified species featuring in its own genome note that reports the genome assembly. More generally, a major outcome from DToL DNA barcoding is a large set of standardised sequences associated with a specimen with a validated name provided by a taxonomic expert. This high quality, curated data will form an important reference resource for species in Britain and Ireland. In some cases, these data can be integrated with existing DNA barcode reference libraries (e.g., UKBOL: https://www.ukbol.org/, Barcode UK, ), populating sections with no existing data, or supplementing the dataset with additional samples allowing the analysis of diagnostic sequences and polymorphism. For example, Barcode UK is a DNA barcoding resource for 1,482 UK flowering plant and conifer species, and DToL data is providing sequences for additional native and alien taxa not represented in the current barcode library.

33DNA barcode data have been generated and interpreted for over 12,000 DToL samples as of December 2023. Here, we consider our experiences from two focal groups, animal samples processed by the NHM, and plant samples processed by RBGE, as well as implications for other taxon groups.

34For animals, we have processed over 10,525 samples. The majority of those samples (~85%) were Arthropoda, followed by Annelida (~5%), Mollusca (~3.4%), and Echinodermata (~1.4%). From the remaining samples, there is a notable representation of: Bryozoa, Nemertea, Cnideria, Hemichordata, Cnidaria, Chordata, and Platyhelminthes. The BOLD query produced a match with the expected species in 59.2% of cases, eliminating the need for further checks. Approximately 20% of the samples required additional verification. This could be attributed to various reasons, such as a mismatch between the BOLD hit and the expected species, or when multiple species were matched. In some instances, the barcode was not present in the BOLD database, indicating it was a new entry. Upon verification by taxonomic experts, 3.5% of the samples had their names changed from the identifier ID to the BOLD ID. The remaining samples were failures of various categories: failure to merge the forward and reverse reads for sanger sequence data (~3%), sample contamination (~1.3%), extraction/PCR failure (~2%), BOLD matches were too low to be informative (~0.8%) or unknown/unclassified (~14.1%). The failure rate for non-marine animals is approximately 2%, whereas for marine animals it is significantly higher (~50%) due to the lower success rate of the quick DNA extraction protocol and the use of universal barcoding primers, as well as the higher levels of contaminants and inhibitors present in the samples. Ongoing development work is focused on taxonomic groups with lower success rates, including the utilisation of different extraction protocols and taxon-specific primers.

35There have been many cases where DNA barcoding has clinched animal identification, including of species new to Britain. A parasitoid wasp collected in Beinn Eighe was identified in the field as Plectiscus ridibundus, a common and widespread species, but the DNA barcode matched to both P. ridibundus and P. callidulus. The specimen was then compared to collection specimens at the NHM by a taxonomic expert and subsequently renamed as P. callidulus, a first record for Britain. Further UK specimens could then be found which had been misidentified. Another example where barcoding aided ichneumonid wasp identification was with a sample collected on Winterton Dunes at a DNA Bioblitz in 2022. Originally, it was identified as Lissonota lineata, a rarely found coastal species. The DNA barcode matched Swedish barcode records of L. confusa which had been overlooked in much of Europe. It now seems possible that L. lineata does not occur in the UK after all. Of course, DNA barcoding can also bring into question morphology-based identifications. Two noctuid moths were collected in Kent which, on wing pattern, looked exactly like Agrotis catalaunensis, a southern European species never recorded in mainland Britain. Some experts agreed, based on the photos. However, barcodes suggested they were in fact Agrotis puta, a very common British moth, and therefore there is a confusing wing pattern variety.

36For plants, we have processed over 1,250 samples, including 869 seed plants, 343 bryophytes and 41 ferns, representing approximately 50% of the British native flora. Our experience suggests DNA extraction, amplification and sequencing can be implemented effectively at scale, with mostly low levels of sample-specific dropout. For example, for 820 of 869 (95%) seed plants we have successfully recovered sequences for both loci used for barcoding. The most significant systematic dropout we have observed is for ferns and lycophytes, where the ITS2 locus proved problematic; based on our low sequence success and conversations with fern taxonomists, we have instead sequenced the plastid trnL-trnF locus, with over 90% of ferns now having sequence data for at least two barcode markers. For the seed plant data, 58 of 869 (7%) of samples have had their species identification queried on the basis of their barcoding results, of which 14 (2% of all samples) have had their name changed (with 20 queries still marked as open or unresolved).

37An example of how DNA barcoding has informed species identification in plants is in the weedy brassica genus Rorippa. A sample collected at a pond margin in the Pentland Hills, near Edinburgh, was initially identified as Rorippa palustris. However, two fruit traits did not match this species, casting doubt on the original identification. DNA barcoding provided a match to R. islandica, a congener reported from this site and also known to favour damp conditions, and matching based on these morphological traits, allowing us to confidently implement a name change. In another case, two Geranium accessions identified, based on morphology, as G. endressii and G. versicolor were flagged by DNA barcoding as potential hybrids, based on heterozygous positions ("wobbles") in ITS2. Photographs of both species were subsequently verified by a specialist referee, as Geranium x oxonianum, a named hybrid between G. endressii and G. versicolor, with our specimens at either end of the spectrum of variability within the hybrid. Based on this, neither of these accessions will be used for genome sequencing. However, both remain as DToL collections, with valuable metadata, field images, herbarium specimens, tissue samples, genome size data and DNA barcodes.

38Beyond these groups, general issues that have arisen are that some taxa such as marine invertebrates have proven challenging for DNA extraction; some mixed samples have consistently shown non-target amplification; very small individuals have proven difficult where DNA recovery is low. With fungi, the main issue is the high levels of contamination when samples are cultured to increase biomass and pervasive cryptic diversity that challenges even the barcode identification of most common species, often revealing the existence of more than one species. To mitigate these issues GALs continue to perform R&D and protocol development. To minimise delays, taxon groups that are easy to identify based on morphology but persistently difficult to barcode, such as large mucilaginous marine molluscs, have been considered barcoding exempt (see lists in the lab SOPs). The most likely analysis issues are those groups with limited reference data, though the exact impact is hard to quantify.

39DNA barcoding has proved an important tool for verifying taxon identification as part of DToL. Future developments will look to further improve scalability to match the plans for more extensive genome sequencing planned for the future. Barcode focused projects such as BioScan have improved scalability by sequencing highly multiplexed amplicons, including large sample sizes sequenced with PacBio HiFi and over 100,000 samples on the ONT MinION. We will investigate the utility of this approach, and others such as low-coverage genome skimming, for the next phase of DToL, though any move to greater genomic coverage in the barcoding phase must not compromise sample throughput and cost effectiveness. Where targeted sequencing approaches are required, the genome data generated by DToL provide an opportunity for discovering new barcode regions. Similarly, streamlining the bioinformatic analyses is a priority, and could involve the development of a unified bioinformatic engine building on the framework developed here. This next phase of DToL will also bring new challenges beyond scalability, as we move from common taxa that are easy to collect and identify, to rarer taxa, smaller species, cryptic species and those in poorly studied groups. We anticipate DNA barcoding becoming of increasing value in each phase of the project.

40Ethical approval and consent were not required.