DNA barcoding and digital taxonomy

1.1. Introduction: the identification crisis in the era of biodiversity

Modern systematics and taxonomy face a fundamental contradiction. On the one hand, Earth's biodiversity is estimated at millions of species, with a significant portion still undescribed (Hebert et al., 2003). On the other hand, the number of qualified taxonomists is steadily declining, and morphological identification of many organismal groups requires specialized knowledge that is becoming increasingly rare (Gaston and O'Neill, 2004). This situation, known as the “taxonomic impediment,” is especially acute in applied fields—plant protection, quarantine, product quality control, and monitoring of agrocenoses.

Traditional morphological identification has a number of limitations. It often requires adult individuals of a particular sex or life stage, is not always applicable to fragmented material, and may fail in the presence of cryptic species—morphologically indistinguishable but genetically distinct taxa (Bickford et al., 2007). In addition, morphological characters may be subject to significant intraspecific variability depending on environmental conditions, which complicates the drawing of clear boundaries between species (Meyer and Paulay, 2005).

In this context, the idea of using short standardized DNA regions for species identification—DNA barcoding—offered a fundamentally new approach to diagnostic problems. This technology does not replace traditional taxonomy, but provides a powerful additional tool that allows rapid and objective identification of organisms based on their genetic information (Hebert et al., 2003; CBOL Plant Working Group, 2009).

1.2. Definition and conceptual basis of DNA barcoding

DNA barcoding is a method of species identification based on the use of a short standardized DNA region (usually 400–800 base pairs) that serves as a molecular “barcode” for a given taxon (Hebert et al., 2003; Kress and Erickson, 2007). The term “barcode” is borrowed from trade, where a universal product code (UPC) uniquely identifies a product. Similarly, the nucleotide sequence of a particular gene is regarded as a unique identifier of a species.

The conceptual basis of the method rests on several key propositions:

First. There are regions of the genome that are conserved enough to be amplified with universal primers across a wide range of organisms, and at the same time variable enough to distinguish closely related species (Hebert et al., 2003).

Second. Interspecific genetic divergence for the chosen marker usually exceeds intraspecific variability. This difference is called the “barcode gap” (Meyer and Paulay, 2005).

Third. It is possible to create a global reference database in which each sequence corresponds to a particular species identified by a taxonomic specialist (Ratnasingham and Hebert, 2007).

It is important to emphasize that DNA barcoding is fundamentally different from whole-genome sequencing. Whereas whole-genome sequencing aims to obtain all the genetic information of an organism, barcoding focuses on a small region selected exclusively for identification purposes. This difference determines all the advantages and limitations of the method.

1.3. Differences from whole-genome sequencing

Whole-genome sequencing is a labor-intensive and expensive process requiring significant computational resources for assembly and annotation. Even with modern technologies, whole-genome sequencing remains inaccessible for routine diagnostics of large samples of organisms. DNA barcoding, by contrast, offers a pragmatic solution: instead of all genomic information, we obtain only the minimum amount of data sufficient for identification (Hebert et al., 2003).

This approach has several important consequences. First, barcoding does not require knowledge of the complete genome sequence—it is sufficient to determine the sequence of one marker region. Second, the method is applicable to organisms whose genomes are poorly studied, which is especially important for understudied taxa. Third, standardization of the marker allows data obtained by different laboratories in different countries to be compared, creating a basis for global cooperation (Ratnasingham and Hebert, 2007).

It should be noted that DNA barcoding is not intended for high-resolution phylogenetic reconstructions. Although barcode sequences can be used in phylogenetic analysis, their main function is identification, not establishing kinship relationships (Hajibabaei et al., 2007). Phylogenetic studies require longer sequences and, as a rule, several independent loci.

1.4. Advantages of the method

DNA barcoding has a number of significant advantages that explain its wide use in modern systematics and applied research.

Speed. Barcode-based identification can be performed within several hours or days, whereas morphological identification of complex groups can take weeks or months. With automated sequencing systems and bioinformatic pipelines, the process becomes even faster (Jain et al., 2016).

Standardization. The use of a single marker for an entire group of organisms (for example, COI for animals) means that results obtained in different laboratories are directly comparable. This eliminates the problem characteristic of morphological identification, where different specialists may use different diagnostic characters (Hebert et al., 2003).

Potential for automation. Barcoding is well suited to automation—from DNA extraction to sequence analysis. This is critically important for processing large samples, for example in biodiversity monitoring or quarantine surveys (Ratnasingham and Hebert, 2007).

Applicability to any life-cycle stage. Unlike morphological keys, which often work only for adult individuals of a particular sex, barcoding allows identification of eggs, larvae, tissue fragments, and other materials that lack diagnostic morphological characters (Floyd et al., 2002).

Objectivity. The identification result is a nucleotide sequence that can be independently verified and reproduced. This reduces the subjectivity inherent in morphological identification (Collins and Cruickshank, 2013).

Ability to work with degraded DNA. Barcoding is often possible even with significant DNA degradation, making it applicable to the analysis of processed products, old herbarium specimens, and museum specimens (Staats et al., 2016).

1.5. Areas of application

DNA barcoding was initially proposed for fundamental taxonomic research, but it soon found wide application in applied fields. For the discipline “Systematics and taxonomy of organisms of the agrosphere,” the following areas are especially important.

Ecology and biodiversity monitoring. Barcoding allows rapid assessment of species richness in various habitats, including agrocenoses. The method is especially useful for inventorying poorly studied groups and identifying cryptic species (Bickford et al., 2007; Deiner et al., 2017).

Quarantine and biosafety. Identification of invasive and quarantine species is one of the most in-demand areas of barcoding application. Rapid and accurate species determination is critical for preventing the spread of harmful organisms (Armstrong and Ball, 2005; Floyd et al., 2010).

Forensics and forensic expertise. Barcoding is used to identify biological materials in forensic investigations, including illegal trade in endangered species (Staats et al., 2016).

Product quality control. Determining the species affiliation of raw materials and finished products is an important task in the food industry. Barcoding makes it possible to detect fraud, for example, the substitution of expensive fish species with cheaper analogues (Galimberti et al., 2013).

Diagnostics of plant diseases and pests. In agronomy, barcoding is used to identify pathogens, nematodes, insect pests, and other organisms important for agriculture (Powers, 2004; Robideau et al., 2011; Choi et al., 2015).

1.6. Historical context and development of the initiative

The idea of using short DNA sequences for identification was not entirely new. As early as the 1980s, attempts were made to use ribosomal genes for diagnosing microorganisms (Janda and Abbott, 2007). However, systematic development of the concept began with the publication of Hebert et al. (2003), where the use of the cytochrome oxidase I (COI) gene was proposed as a universal barcode for animals.

In 2004, the Consortium for the Barcode of Life (CBOL) was created—an international organization uniting research institutes, museums, and universities from different countries. The consortium's goal was to create a global reference library of barcodes for all species of living organisms (Ratnasingham and Hebert, 2007).

In 2010, the International Barcode of Life (iBOL) initiative was launched—a large-scale project aimed at obtaining barcodes for millions of species. In parallel, the information infrastructure developed—the Barcode of Life Data System (BOLD), which became the central platform for storing, analyzing, and publishing barcoding data (Ratnasingham and Hebert, 2007).

For plants, the search for a universal barcode proved more difficult than for animals. The mitochondrial COI gene, effective in animals, evolves too slowly in plants. After extensive studies, in 2009 the CBOL Plant Working Group recommended the use of a combination of two plastid genes—rbcL and matK—as the core barcode for land plants (CBOL Plant Working Group, 2009). Later, the addition of additional markers such as ITS and trnH-psbA was proposed to increase resolving power (China Plant BOL Group, 2011; Hollingsworth et al., 2011).

1.7. The place of DNA barcoding in the system of taxonomic methods

It is important to clearly define the place of DNA barcoding in the methodology of systematics. It is not a replacement for traditional morphological taxonomy and not an independent discipline. Rather, it is a tool that complements and extends the researcher's capabilities.

Barcoding is not intended for:

• describing new species without morphological justification;
• establishing high-resolution phylogenetic relationships;
• replacing taxonomic expertise in solving complex nomenclatural issues.

Barcoding is intended for:

• rapid identification of known species;
• flagging potentially new species;
• verifying the species affiliation of specimens identified morphologically;
• providing a standardized approach to diagnostics.

This distinction is of fundamental importance, especially in the context of criticism of the method. As Collins and Cruickshank (2013) note, conflating the tasks of specimen identification and species discovery leads to methodological errors and unjustified conclusions. A correct understanding of the method's capabilities and limitations is a necessary condition for its effective use.

1.8. Key terms and concepts

For further study of the material, it is necessary to master a number of key terms.

Barcode — a short standardized DNA sequence used for species identification.

Barcode gap — the difference between maximum intraspecific and minimum interspecific genetic divergence for a given marker.

Reference library — a collection of barcode sequences linked to morphologically identified voucher specimens.

Voucher — a physical specimen from which a sequence was obtained, stored in a scientific collection and available for re-examination.

Mini-barcode — a shortened fragment of a standard barcode used for analyzing degraded DNA.

Metabarcoding — a method for analyzing complex mixtures of organisms (for example, from soil or water) using high-throughput sequencing.

MOTU (Molecular Operational Taxonomic Unit) — an operational taxonomic unit distinguished on the basis of genetic similarity, without obligatory linkage to a formal taxonomic name.

2.1. Introduction: why a single marker does not exist

One of the central ideas of DNA barcoding is standardization: the use of a single DNA region for identifying all organisms within a large group. However, the biology of different taxa imposes significant constraints on the choice of such a region. The rate of molecular evolution, genome structure, presence of introns and repeats, and availability of universal primers all vary among the kingdoms of living nature. As a result, each major group of organisms has developed its own standard markers, and in some cases combinations of several loci (Hollingsworth et al., 2011; Schoch et al., 2012).

The choice of marker is determined by a compromise between two opposing requirements. On the one hand, the region must be conserved enough to be amplified with universal primers across a wide range of species. On the other hand, it must possess sufficient variability to distinguish closely related species. This compromise cannot always be found in a single gene, which leads to the use of multi-locus approaches (CBOL Plant Working Group, 2009; China Plant BOL Group, 2011).

In this chapter we will consider the main markers adopted for different groups of organisms and discuss the criteria for their selection.

2.2. Animals: the mitochondrial COI gene

For animals, a region of the mitochondrial cytochrome oxidase I (COI, cox1) gene was proposed as the standard barcode marker (Hebert et al., 2003). This choice was not accidental. The mitochondrial genome of animals has a number of properties that make it convenient for identification.

Absence of introns. Mitochondrial genes in animals usually do not contain introns, which simplifies amplification and sequence alignment (Hebert et al., 2003).

High rate of evolution. Mitochondrial genes evolve faster than nuclear genes, which makes it possible to distinguish even recently diverged species (Hebert et al., 2003; Cywinska et al., 2006).

Universal primers. Primers have been developed for COI that work in most animal groups. The classic pair—LCO1490 and HCO2198—amplifies a fragment about 658 base pairs long (Folmer et al., 1994; Hebert et al., 2003).

Haploidy and absence of recombination. Mitochondrial DNA is inherited maternally and is not subject to recombination, which simplifies data interpretation (Hebert et al., 2003).

The effectiveness of COI has been demonstrated across a wide range of taxa. For example, in a study of Canadian mosquitoes, intraspecific divergence in COI averaged 0.5%, whereas interspecific divergence was 10.4% (Cywinska et al., 2006). For mollusks, it has been shown that COI allows species to be distinguished, but in some groups (for example, gastropods) resolving power may be lower (Meyer and Paulay, 2005). Nevertheless, COI remains the main marker for animals, and the largest reference database, BOLD, contains millions of sequences of this gene (Ratnasingham and Hebert, 2007).

It is important to note that for some animal groups COI may be ineffective. For example, in cnidarians (Cnidaria) the rate of COI evolution is lower, and this marker does not always distinguish closely related species (Hebert et al., 2003). In such cases, alternative mitochondrial genes (for example, 16S rRNA) or nuclear markers such as ITS are used (Schoch et al., 2012).

2.3. Plants: the search for a compromise and the multi-locus approach

The search for a universal barcode for plants proved considerably more difficult than for animals. The mitochondrial COI gene in plants evolves too slowly to distinguish closely related species, and the structure of the plant mitochondrial genome is unstable (Kress and Erickson, 2007). Therefore, researchers' attention shifted to the plastid and nuclear genomes.

After extensive comparative studies, the CBOL Plant Working Group (2009) recommended a combination of two plastid genes, rbcL and matK, as the core barcode for land plants.

rbcL (ribulose-1,5-bisphosphate carboxylase/oxygenase, large subunit) is one of the most conserved plastid genes. It is easily amplified and sequenced in most plants, but its resolving power at the species level is limited. Nevertheless, rbcL provides reliable assignment of a specimen to family or genus and serves as an “anchor” for comparison (CBOL Plant Working Group, 2009).

matK (maturase kinase) is one of the most rapidly evolving plastid genes. It provides higher resolution at the species level, but its amplification may be difficult due to the lack of universal primers, especially in gymnosperms and non-flowering plants (CBOL Plant Working Group, 2009; Hollingsworth et al., 2011).

The combination rbcL + matK was adopted as the standard, but subsequent studies showed that it is insufficient for many plant groups. China Plant BOL Group (2011), using a large dataset (6,286 specimens, 1,757 species), demonstrated that the addition of the nuclear marker ITS (internal transcribed spacer) significantly increases resolving power. The combination ITS + any plastid marker allowed 69.9–79.1% of species to be distinguished, whereas rbcL + matK distinguished only 49.7%.

ITS (internal transcribed spacer) is a nuclear ribosomal marker that possesses high variability and is often used in plant phylogenetics. However, its use is limited by possible problems with paralogues, polymorphism, and contamination (China Plant BOL Group, 2011). As a more convenient alternative, ITS2 was proposed—a smaller ITS fragment that is easier to amplify and sequence (Chen et al., 2010).

An additional marker for plants is trnH-psbA—an intergenic spacer of the plastid genome. It is easily amplified and possesses high variability, but may have problems with length and mononucleotide repeats (Kress and Erickson, 2007; Hollingsworth et al., 2011).

Thus, for plants, it is currently recommended to use the core barcode rbcL + matK with the possible addition of ITS/ITS2 and/or trnH-psbA depending on the taxonomic group and research objectives (Hollingsworth et al., 2011; China Plant BOL Group, 2011).

2.4. Fungi: ITS as a universal marker

For fungi, the search for a standard barcode was also not simple. Mitochondrial COI proved unsuitable: in fungi it often contains large introns, amplifies poorly, and is insufficiently variable (Schoch et al., 2012). As a result, the international Fungal Barcoding Consortium conducted a large-scale comparison of several markers: ITS, LSU, SSU, and RPB1.

Based on analysis of 742 strains representing 17 major fungal lineages, it was established that ITS (internal transcribed spacer) possesses the best combination of universality, sequence quality, and resolving power (Schoch et al., 2012). ITS showed a probability of correct identification of 0.73, comparable to the two-locus barcode for plants (0.70). Importantly, ITS is successfully amplified in most fungi using universal primers, although for some groups (for example, Glomeromycota) additional efforts may be required.

Protein markers such as RPB1 (largest subunit of RNA polymerase II), RPB2, and MCM7 showed higher resolution, but their amplification is less reliable and not universal (Schoch et al., 2012). Therefore, ITS was officially recommended as the primary barcode for fungi, with the possibility of using additional markers for particular groups.

2.5. Bacteria: 16S rRNA and its limitations

For bacteria and archaea, the standard marker is the 16S rRNA gene (small ribosomal subunit RNA). This choice is due to several reasons: the gene is present in all bacteria, contains conserved regions for universal primers and variable regions for distinguishing taxa (Janda and Abbott, 2007).

However, 16S rRNA has significant limitations at the species level. A similarity threshold of 97% is often used to define a species, but it is not universal: some species may have more than 97% similarity and still be reproductively isolated, while others may have less than 97% but belong to the same species (Stackebrandt and Goebel, 1994). Janda and Abbott (2007) note that 16S rRNA provides reliable identification to genus in >90% of cases, but to species in only 65–83%.

To increase resolution at the species level, additional markers such as rpoB, gyrB, recA, and other housekeeping genes are used. In some cases, whole-genome sequencing or MLST (multi-locus sequence typing) is applied. Nevertheless, 16S rRNA remains the main tool for rapid identification and phylogenetic studies of bacteria.

2.6. Oomycetes: ITS, COX1, and COX2

Oomycetes are a group of organisms phylogenetically distant from true fungi, but often studied together with them because of similar lifestyle and morphology. Many oomycetes are dangerous plant pathogens (for example, Phytophthora infestans—the causal agent of potato late blight). Various markers have been proposed for their identification.

ITS (internal transcribed spacer) is historically the first and most widely used marker for oomycetes. However, in some genera (for example, Phytophthora) ITS may be insufficiently variable to distinguish closely related species, while in others (for example, Plasmopara) it may contain long insertions that complicate amplification (Robideau et al., 2011).

COI (cytochrome c oxidase subunit I) is a mitochondrial marker that was proposed as a supplement to ITS. Robideau et al. (2011) showed that COI provides good resolution for many oomycetes and may be more effective than ITS for some groups. For example, COI distinguishes Phytophthora infestans, P. mirabilis, and P. andina, which have identical ITS sequences.

COX2 (cytochrome c oxidase subunit II) is another mitochondrial marker widely used for phylogenetic studies of oomycetes. Choi et al. (2015) compared cox1 and cox2 for 31 genera of oomycetes and concluded that cox2 possesses better amplifiability, higher nucleotide divergence, and better resolving power at the species level. They recommended cox2 as the primary barcode for oomycetes, with the possibility of using ITS as a supplement.

Thus, for oomycetes, a combination of ITS and COX2 (or COX1) is currently recommended for reliable identification.

2.7. Nematodes: SSU, ITS, and COI

Nematodes are one of the most difficult groups for morphological identification. Their small size, limited set of diagnostic characters, and high cryptic variability make molecular methods especially in demand (Floyd et al., 2002; Powers, 2004).

Floyd et al. (2002) proposed using SSU (small subunit, 18S rRNA) to create MOTU (Molecular Operational Taxonomic Units) in soil nematodes. They showed that SSU sequences about 450 base pairs long allow species to be distinguished with a similarity threshold >99.5%. This approach works well for assessing diversity, but does not always provide species identification due to the absence of complete reference databases.

ITS (internal transcribed spacer) is a more variable marker widely used for diagnosing nematodes, especially in the genera Meloidogyne, Heterodera, and Globodera. For example, ITS allows distinguishing M. chitwoodi and M. fallax, which are morphologically indistinguishable (Powers, 2004).

COI is a mitochondrial marker also used for nematode identification. However, in some groups (for example, Meloidogyne) the mitochondrial genome has an unusual structure, which limits the use of COI (Powers, 2004). Nevertheless, for many nematodes COI provides good resolution and is used in combination with ITS.

2.8. Criteria for marker selection: conserved vs variable regions

The choice of marker for DNA barcoding is always a compromise among several criteria (Hollingsworth et al., 2011; Schoch et al., 2012):

1. Primer universality. The marker should amplify across the widest possible range of taxa using standard primers. This is especially important for metabarcoding and analysis of mixed samples.

2. Variability. The region should possess sufficient variability to distinguish closely related species. However, too high variability can lead to intraspecific polymorphism and complicate identification.

3. Sequence quality. The marker should be easy to sequence, without long mononucleotide repeats, introns, and other structures that complicate alignment and analysis.

4. Length. The optimal length is 400–800 base pairs. Too short sequences do not provide sufficient resolution; too long ones are difficult for routine sequencing.

5. Availability of a reference database. Even an ideal marker is useless if there is no reference database for it. BOLD and GenBank contain millions of sequences, but their coverage is uneven across taxa.

6. Unambiguity. The marker should provide unambiguous identification, that is, intraspecific variability should be less than interspecific variability (barcode gap).

Table 2.1 presents the main markers for different groups of organisms.

Table 2.1. Standard DNA barcoding markers for different groups of organisms

3.1. Introduction: from specimen to sequence

DNA barcoding is a multi-stage process that begins with the collection of biological material and ends with comparison of the obtained sequence against a reference database. Each stage has its own methodological features, limitations, and sources of error. Understanding the complete workflow is necessary for critical evaluation of results and correct interpretation of data (Hajibabaei et al., 2007; Collins and Cruickshank, 2013).

In general, the DNA barcoding process includes the following stages:

1. Sample collection and preservation

2. DNA extraction

3. Amplification of the target region by PCR

4. Sequencing of the amplicon

5. Sequence analysis and comparison with a reference database

6. Interpretation of results and assignment of a taxonomic name

Each of these stages can be automated to varying degrees, and modern technological developments allow some operations to be performed in the field (Jain et al., 2016). Let us consider each stage in detail.

3.2. Sample collection and preservation

The first stage—collection of biological material—seems trivial, but it is precisely here that the foundations of future problems are often laid. A wide variety of material can be used for DNA barcoding: tissue fragments, whole small organisms, herbarium specimens, museum specimens, stomach contents, feces, soil, water, and even air (Deiner et al., 2017; Staats et al., 2016).

The key requirement for a sample is DNA preservation. DNA degradation begins immediately after the death of an organism or separation of tissue. Enzymes (nucleases), microorganisms, and physical factors (temperature, humidity, UV radiation) destroy the DNA molecule. Therefore, proper sample preservation is critically important for successful barcoding (Staats et al., 2016).

Main preservation methods:

• Freezing. Storage at −20 °C or −80 °C is the optimal method for most tissues. However, it requires equipment and is not always available in the field.
• Ethanol (95–100%). Widely used for preserving animals, especially insects and nematodes. Ethanol fixes tissues and prevents microbial growth, but over time it can lead to DNA degradation.
• Silica gel. Used for rapid drying of plant tissues (leaves, roots). Silica gel absorbs moisture and creates conditions that prevent DNA decomposition (Kress and Erickson, 2007).
• Lysis buffers. Special solutions (for example, CTAB) allow DNA to be preserved at room temperature for a long time.
• FTA cards. Paper impregnated with reagents that lyse cells and immobilize DNA. Convenient for field research (Staats et al., 2016).

Special cases:

• Herbarium specimens. DNA in herbarium specimens is often fragmented, but with careful handling it can be used for barcoding. It is important to avoid contamination and to use mini-barcodes (Staats et al., 2016).
• Museum specimens. DNA in old collections may be heavily degraded. Successful analysis requires special methods of extraction and amplification of short fragments (Hajibabaei et al., 2007).
• Processed products. In food products, medicines, and traditional medicines, DNA may be fragmented and mixed with other organisms. Mini-barcodes and metabarcoding are used to analyze such samples (Staats et al., 2016; Galimberti et al., 2013).

The problem of contamination. Contamination of a sample with foreign DNA is one of the main problems of barcoding. Sources of contamination include reagents, equipment, the researcher's hands, and other organisms in the sample. To minimize risk, negative controls, disposable materials, and thorough sterilization are used (Collins and Cruickshank, 2013).

3.3. DNA extraction

After sample collection, DNA must be extracted—freed from cellular components, proteins, lipids, and other impurities that can inhibit PCR. Extraction methods vary depending on tissue type and sample condition.

Main approaches:

• Phenol-chloroform extraction. A classic method based on phase separation. Proteins and lipids pass into the organic phase, while DNA remains in the aqueous phase. The method is effective but labor-intensive and requires work with toxic reagents.
• Silica sorption. DNA binds to silica in the presence of chaotropic salts, then is washed and eluted. This principle underlies most commercial DNA extraction kits. The method is fast and convenient for automation (Ivanova et al., 2006).
• CTAB method. Widely used for plants containing polysaccharides and polyphenols that inhibit PCR. CTAB precipitates polysaccharides, leaving DNA in solution (Kress and Erickson, 2007).
• Magnetic particles. DNA binds to magnetic nanoparticles coated with silica or other affinity groups. The method is easily automated and used in high-throughput systems.

Extraction problems:

• PCR inhibitors. Polysaccharides, polyphenols, humic acids, heme, and other compounds can inhibit polymerase. Additional purification steps are used to remove them (for example, PVPP, BSA).
• DNA degradation. When working with damaged samples, DNA may be fragmented. In such cases, mini-barcodes are used (Staats et al., 2016).
• Low DNA content. Some samples (for example, single cells, old museum specimens) contain minimal amounts of DNA. Special amplification methods (whole genome amplification) or sensitive detection systems are required for analysis.

Quality control. After DNA extraction, its concentration and purity are assessed using spectrophotometry (NanoDrop) or fluorimetry (Qubit). The A260/A280 ratio indicates purity from proteins; A260/A230 indicates purity from salts and organic compounds.

3.4. Amplification of the target region by PCR

Polymerase chain reaction (PCR) is the central stage of DNA barcoding. It is at this stage that the target barcode region is amplified using specific primers (Hebert et al., 2003; Hajibabaei et al., 2007).

Universal primers. The key idea of DNA barcoding is the use of primers that work across a wide range of organisms. For animal COI, the classic pair is LCO1490 and HCO2198, amplifying a fragment about 658 base pairs long (Folmer et al., 1994). For plants, primers to rbcL, matK, ITS, and trnH-psbA are used (CBOL Plant Working Group, 2009; China Plant BOL Group, 2011). For fungi—primers to ITS (Schoch et al., 2012). For oomycetes—primers to ITS, COX1, and COX2 (Robideau et al., 2011; Choi et al., 2015).

The problem of universality. There are no primers that work in all organisms without exception. Even for COI in some groups (for example, cnidarians, nematodes), standard primers may not work. In such cases, alternative primers, degenerate primers (mixtures of oligonucleotides with different bases at variable positions), or primer cocktails are used (Ivanova et al., 2006; Cywinska et al., 2006).

PCR stages:

1. Denaturation. Double-stranded DNA separates into single-stranded molecules upon heating to 94–98 °C.

2. Primer annealing. Primers bind to complementary regions on the template at 45–65 °C (depends on primers).

3. Elongation. Polymerase synthesizes a new DNA strand starting from the primer at 72 °C.

4. Repetition. The cycle is repeated 30–40 times, leading to exponential accumulation of the target fragment.

PCR optimization. For successful amplification, it is necessary to optimize:

• MgCl₂ concentration (polymerase cofactor)
• Primer concentration
• Annealing temperature
• Number of cycles
• Buffer composition

PCR problems:

• Nonspecific amplification. Non-target fragments may be amplified, leading to contamination of the sequence.
• No amplification. May be associated with mutations in the primer binding site, PCR inhibitors, or DNA degradation.
• Contamination. Contamination with foreign DNA can lead to false-positive results.
• Heteroplasmy and NUMTs. In animals, mitochondrial genes may exist as nuclear copies (NUMTs—nuclear mitochondrial pseudogenes), which amplify together with mitochondrial ones and distort results (Hebert et al., 2004).

Mini-barcodes. For degraded DNA, shortened fragments (100–300 base pairs) are used, which are easier to amplify. However, mini-barcodes may have lower resolving power (Staats et al., 2016; Meusnier et al., 2008).

Multiplex PCR. For analysis of mixed samples (metabarcoding), multiplex PCR is used, allowing several markers to be amplified simultaneously (Deiner et al., 2017).

3.5. Sequencing of the amplicon

After successful amplification, it is necessary to determine the nucleotide sequence of the obtained fragment. Several sequencing methods can be used for DNA barcoding.

Sanger method. The classic sequencing method developed by Frederick Sanger in 1977. It is based on termination of DNA synthesis by dideoxynucleotides. The Sanger method remains the “gold standard” for DNA barcoding due to its high accuracy and read length (up to 1000 base pairs). However, it is relatively expensive and low-throughput, which limits its use for large samples (Hajibabaei et al., 2007).

High-throughput sequencing (NGS). Second- and third-generation technologies allow millions of sequences to be sequenced simultaneously. Main platforms:

• Illumina (MiSeq, HiSeq, NextSeq). Short reads (150–300 base pairs), high accuracy, low cost per base. Widely used for metabarcoding (Deiner et al., 2017).
• Ion Torrent. Semiconductor sequencing, medium read length (200–400 base pairs), fast runs.
• PacBio (SMRT). Long reads (up to 20 kbp), high error rate (10–15%), but errors are random and can be corrected.
• Oxford Nanopore (MinION). Portable sequencers, long reads (up to 100+ kbp), operate in real time. MinION is especially promising for field research (Jain et al., 2016).

Choice of sequencing method depends on the research objectives. For routine barcoding of individual samples, Sanger is often used. For metabarcoding and large samples—NGS. For field research—portable sequencers such as MinION (Jain et al., 2016).

Sequence quality control. After sequencing, it is necessary to assess data quality:

• Phred score. Assessment of the probability of error at each position. High quality—Phred >30 (error probability <0.1%).
• Chromatograms. Visual analysis of peaks to identify ambiguous bases.
• Sequence length. For standard COI barcode—about 658 base pairs; for mini-barcodes—100–300.
• Presence of stop codons. For protein-coding genes (COI, rbcL, matK), the presence of stop codons may indicate NUMTs or sequencing errors.

3.6. Comparison with a reference database

After obtaining a high-quality sequence, it must be compared with a reference database to determine the species affiliation of the sample. There are several main databases used in DNA barcoding.

BOLD (Barcode of Life Data System). The central platform for DNA barcoding, developed at the University of Guelph (Canada) (Ratnasingham and Hebert, 2007). BOLD contains:

• Millions of barcode sequences
• Metadata on vouchers (morphological identification, collection site, collection)
• Images of specimens
• Analysis tools (Identification Engine, Taxon ID Tree, BIN system)

BOLD uses the BIN (Barcode Index Number) system—clusters of sequences that correspond to species with a high degree of concordance. BIN allows samples to be identified even in the absence of an exact match with a reference sequence (Ratnasingham and Hebert, 2013).

GenBank. The largest public database of nucleotide sequences, maintained by NCBI (USA). GenBank contains millions of sequences, including barcode ones. However, data quality in GenBank varies: many sequences are not linked to vouchers, may contain identification errors, or may result from contamination (Janda and Abbott, 2007). Therefore, for DNA barcoding it is preferable to use BOLD or carefully curated databases.

Other databases:

• UNITE—for fungi (ITS)
• SILVA—for ribosomal genes
• PR2—for protists
• MMDBD—for medicinal plants

Comparison algorithms:

BLAST (Basic Local Alignment Search Tool). The most popular comparison method. BLAST finds local alignments between the query and the database, calculates a similarity score and E-value. The result is a list of the most similar sequences (Altschul et al., 1990). Limitation of BLAST: it does not take sequence length into account when assessing similarity, which can lead to false results.

Neighbour-Joining (NJ). A phylogenetic method used to visualize barcoding results. Sequences are clustered into a tree, and the sample is assigned to the cluster with which it groups. NJ trees are simple to interpret but can give erroneous results with an incomplete reference database (Collins and Cruickshank, 2013).

ABGD (Automatic Barcode Gap Discovery). A method for automatically determining the barcode gap and dividing sequences into hypothetical species. ABGD does not require phylogenetic analysis and works quickly even on large samples (Puillandre et al., 2012).

Best Match / Best Close Match. Alternative identification methods based on genetic distances. Best Match assigns the sample to the species of the closest reference sequence. Best Close Match takes a similarity threshold into account and can give a “no match” result when no close sequences are present (Meier et al., 2006).

3.7. Similarity thresholds and the “barcode gap” problem

A key question in interpreting barcoding results is what level of similarity is considered sufficient for species identification. Traditionally, a threshold of 97–98% is used (Hebert et al., 2003), but this threshold is not universal.

Barcode gap. The concept of the barcode gap assumes that interspecific genetic divergence exceeds intraspecific divergence. If this is so, there is a threshold above which sequences belong to one species and below which they belong to different species. However, Meyer and Paulay (2005) showed that in mollusks intraspecific variability can overlap with interspecific variability, making the threshold approach unreliable.

Problems with thresholds:

• Overlap of distributions. Intraspecific variability may be higher than interspecific variability for some species pairs.
• Differences among taxa. The optimal threshold varies among groups. For example, for nematodes the threshold may be 99.5% (Floyd et al., 2002), for birds—98% (Hebert et al., 2004).
• Young species. Recently diverged species may have low interspecific divergence.
• Hybridization. With mitochondrial DNA introgression, species may have identical or similar sequences (Meyer and Paulay, 2005).

Alternatives to thresholds:

• BOLD BIN system. Clustering of sequences based on genetic distances rather than a fixed threshold (Ratnasingham and Hebert, 2013).
• ABGD. Automatic determination of the barcode gap for each specific sample (Puillandre et al., 2012).
• Statistical methods. Bayesian delimitation, GMYC (General Mixed Yule Coalescent), and others (Puillandre et al., 2012).

Practical recommendations. Collins and Cruickshank (2013) recommend:

• Do not use fixed thresholds without verification
• Use Best Close Match instead of a simple threshold
• Take into account the incompleteness of the reference database
• Verify results with independent data (morphology, ecology)

3.8. Automation and miniaturization

Modern technological developments are aimed at automating and miniaturizing all stages of DNA barcoding, which is especially important for field research and high-throughput screening.

Laboratory automation. Robotic systems for DNA extraction, PCR setup, and sequencing allow thousands of samples to be processed per day with minimal human involvement. An example is the Canadian Centre for DNA Barcoding (CCDB), which processes more than 100,000 samples per year (Ivanova et al., 2006).

Microfluidic chips. Lab-on-a-chip devices allow DNA extraction, PCR, and even sequencing in a miniature format. This reduces reagent consumption and analysis time (Deiner et al., 2017).

Portable sequencers. Oxford Nanopore MinION is a portable sequencer the size of a USB flash drive that operates in real time. MinION can be used in field conditions, including remote expeditions and even the International Space Station (Jain et al., 2016).

Field laboratories. Portable DNA barcoding kits are being developed that include everything necessary for DNA extraction, PCR, and sequencing. This opens possibilities for rapid diagnostics in places where there is no access to stationary laboratories (Jain et al., 2016).

CRISPR diagnostics. CRISPR-Cas technology can be used to detect specific DNA sequences without sequencing. This is a promising direction for rapid field diagnostics (Jain et al., 2016).

4.1. Introduction: limits of applicability of the method

DNA barcoding has demonstrated high effectiveness for species identification in a wide variety of organismal groups. However, like any method, it has limits of applicability. In this chapter we will consider situations in which barcoding fails, leads to erroneous conclusions, or simply does not work. Understanding these limitations is critically important for correct interpretation of results and for choosing adequate diagnostic methods (Meyer and Paulay, 2005; Collins and Cruickshank, 2013; Taylor and Harris, 2012).

The key problem is that DNA barcoding is based on several assumptions that are not always fulfilled:

1. Interspecific genetic divergence exceeds intraspecific divergence (a “barcode gap” exists)

2. The chosen marker evolves at a constant rate in different lineages

3. The reference database contains sequences for all possible species

4. Morphological species used to create the reference database correspond to biological species

When these assumptions are violated, barcoding can give false-positive, false-negative, or ambiguous results. Let us consider the main problem situations.

4.2. Closely related species and cryptic complexes

One of the most frequent problems is the presence of closely related species that have low interspecific divergence for the barcode marker. This problem is especially acute for cryptic species—morphologically indistinguishable but genetically distinct taxa (Bickford et al., 2007).

Cryptic species are widespread in all groups of organisms. Bickford et al. (2007) note that cryptic complexes have been found in deep-sea mollusks, freshwater fish, tropical butterflies, Arctic plants, fungi, nematodes, and many other groups. The causes of crypticity are diverse: nonvisual mating signals (pheromones, acoustic signals), morphological stasis under stabilizing selection, and recent divergence (Bickford et al., 2007).

For DNA barcoding, cryptic species pose particular difficulty. If two species diverged recently, their barcode sequences may be identical or differ by 1–2 nucleotides. At a standard similarity threshold of 97–98%, such species will be indistinguishable (Meyer and Paulay, 2005). Moreover, if one of the species is not represented in the reference database, the sample may be erroneously assigned to another species.

Examples of cryptic complexes:

• Mosquitoes. Cywinska et al. (2006) found that Aedes fitchii and Aedes grossbecki have low interspecific divergence in COI and morphological signs of hybridization. Intraspecific divergence in Ae. fitchii reached 3.6–3.9%, overlapping with interspecific differences.
• Nematodes. Floyd et al. (2002) showed that some morphologically distinct nematode species have identical SSU sequences. For example, Cephalobus and Acrobeloides—two genera that are morphologically difficult to distinguish and have similar SSU sequences.
• Fungi. Schoch et al. (2012) note that ITS sequences are identical in several species complexes of Aspergillus, Penicillium, and Fusarium, which are economically important and medically significant.

The problem of recent divergence. Puillandre et al. (2012), in simulation experiments, showed that ABGD (Automatic Barcode Gap Discovery) does not work when the rate of speciation is comparable to the rate of genetic drift. In this case, interspecific and intraspecific divergence overlap, and the barcode gap disappears. This is especially characteristic of groups with recent radiation or with large effective population sizes.

Practical consequences. For groups with cryptic species, DNA barcoding should be used with caution. It is recommended to:

• Use several independent markers (multi-locus approach)
• Apply statistical methods of species delimitation (ABGD, GMYC, Bayesian delimitation)
• Combine molecular data with ecological, behavioral, and geographical data

4.3. Hybridization and introgression

Hybridization—crossing between species—and introgression—the penetration of genes of one species into the gene pool of another—create serious problems for DNA barcoding. During hybridization, mitochondrial DNA (or plastid DNA in plants) of one species may penetrate into the gene pool of another, leading to a mismatch between morphology and barcode sequence (Meyer and Paulay, 2005; China Plant BOL Group, 2011).

Mechanisms:

• Mitochondrial DNA introgression. In animals, mitochondrial DNA is inherited maternally. During hybridization, offspring receive mitochondria from the mother. If a hybrid crosses with a representative of the paternal species, the maternal species' mitochondrial DNA may become fixed in the paternal species' population. As a result, individuals morphologically belonging to species A may have mitochondrial DNA of species B (Meyer and Paulay, 2005).
• Chloroplast capture in plants. In plants, plastid DNA is also inherited maternally (in most angiosperms). During hybridization and introgression, the plastid genome of one species may replace that of another. This phenomenon, known as “chloroplast capture,” is widespread in plants (China Plant BOL Group, 2011; Hollingsworth et al., 2011).

Examples:

• Plants. China Plant BOL Group (2011) found that in 45.2% of genera with several species there is a mismatch between ITS and plastid barcodes. This is explained by hybridization, introgression, or incomplete lineage sorting. For example, in Quercus, Salix, Pinus, and other genera, plastid haplotypes often do not correspond to species boundaries (Hollingsworth et al., 2011).
• Animals. Cywinska et al. (2006) described possible hybridization between Aedes fitchii and Ae. grossbecki: individuals of Ae. fitchii had morphological characters of both species. Hebert et al. (2004) noted that in butterflies Astraptes fulgerator, some individuals had heterozygous COI sequences, indicating possible hybridization or amplification of NUMTs.
• Fish. In some fish species, mitochondrial DNA introgression is observed, leading to a mismatch between morphology and barcode (Taylor and Harris, 2012).

Problem for barcoding. During hybridization and introgression, the barcode sequence may not correspond to the species affiliation of the individual determined by morphology. This leads to:

• False identifications
• Incorrect assessment of biodiversity
• Erroneous conclusions about species distributions

Solutions. For groups with frequent hybridization, it is recommended to:

• Use nuclear markers (ITS, RPB1, RPB2) in addition to mitochondrial or plastid ones
• Analyze several independent loci
• Take into account biogeographic and ecological data

4.4. Polyploidy

Polyploidy—an increase in chromosome number—is widespread in plants, fungi, and some animals. Polyploid species may have complex genomes with several gene copies, which creates problems for DNA barcoding (Hollingsworth et al., 2011).

Problems:

• Multiple ITS copies. In polyploid plants, ITS may exist in several variants, which may result from hybridization or incomplete concerted evolution. This leads to sequence polymorphism and complicates species determination (China Plant BOL Group, 2011).
• Uniparental inheritance of plastids. In polyploids, the plastid genome is inherited from one of the parents, which may not reflect the complete genomic composition. As a result, the plastid barcode may not correspond to the species determined by morphology (Hollingsworth et al., 2011).
• Allopolyploids. In allopolyploidy (hybrid polyploidy), different parental species may have different barcode sequences. If an allopolyploid inherited the plastid genome from one parent, it will group with that parent rather than the other (Hollingsworth et al., 2011).

Examples:

• Ferns. Many ferns are polyploids. Grusz et al. (2009) showed that Cheilanthes yavapensis is an apomictic tetraploid originating from hybridization of C. fendleri and C. lindheimeri. The plastid barcode does not allow such hybrids to be unambiguously identified.
• Flowering plants. In Taraxacum, Hieracium, Rubus, and other genera with apomixis and polyploidy, barcoding often does not work because of complex genomic structure (Hollingsworth et al., 2011).

Solutions. For polyploid groups, it is recommended to:

• Use nuclear markers with multiple copies (ITS, ETS)
• Analyze several independent loci
• Apply cloning to separate ITS variants
• Use morphological and cytological data

4.5. Intraspecific variability

High intraspecific genetic variability is one of the main problems of DNA barcoding. If intraspecific divergence exceeds interspecific divergence, the barcode gap disappears, and identification becomes unreliable (Meyer and Paulay, 2005).

Causes of intraspecific variability:

• Geographic structure. Populations of one species may be genetically differentiated due to limited gene flow. Meyer and Paulay (2005) showed that in marine gastropods intraspecific divergence can reach 3–4%, overlapping with interspecific differences.
• Historical processes. Ancient population structures, secondary contact, and incomplete lineage sorting all create high intraspecific variability.
• Heteroplasmy. The presence of several mitochondrial DNA variants in one cell or organism. In insects, nematodes, and other groups, heteroplasmy can reach 1–2% (Hebert et al., 2004).
• NUMTs. Nuclear copies of mitochondrial genes (nuclear mitochondrial pseudogenes) may amplify together with mitochondrial ones, creating false polymorphism. Hebert et al. (2004) found that 2.8% of COI sequences in Astraptes fulgerator were NUMTs.

Problem for barcoding. High intraspecific variability leads to:

• False-negative results (the sample does not match the reference sequence of its species)
• False-positive results (the sample matches the reference sequence of another species)
• Ambiguous identifications

Examples:

• Mollusks. Meyer and Paulay (2005) showed that in mollusks intraspecific divergence in COI can reach 3–4%, while interspecific divergence ranges from 0–18%. The optimal threshold for minimizing errors was 2.6%, but even then error reached 17%.
• Nematodes. Floyd et al. (2002) found that intraspecific SSU variability in nematodes can reach 0.5%, requiring a threshold of 99.5% to distinguish species.
• Fungi. Schoch et al. (2012) noted that in Glomeromycota intraspecific ITS variability can reach 20%, making barcoding practically useless for this group.

Solutions. For groups with high intraspecific variability, it is recommended to:

• Increase the similarity threshold (for example, to 99% for nematodes)
• Use several markers
• Analyze several samples per species
• Apply statistical methods (ABGD, GMYC)

4.6. Absence of reference sequences

DNA barcoding works only when a reference database is available in which sequences are linked to morphologically identified vouchers. If a species is absent from the database, identification is impossible (Ratnasingham and Hebert, 2007; Collins and Cruickshank, 2013).

Problems:

• Incompleteness of databases. BOLD contains millions of sequences, but coverage is uneven. For many groups (especially tropical invertebrates, fungi, protists), reference data are absent or fragmentary (Taylor and Harris, 2012).
• Geographic bias. Most barcode data have been obtained for North America and Europe. Tropical regions, where biodiversity is highest, are underrepresented (Usman et al., 2025).
• Taxonomic bias. Some groups (birds, fish, butterflies) are well studied; others (nematodes, fungi, protists) are poorly studied. Taylor and Harris (2012) noted that 53% of barcoding studies are devoted to invertebrates, 16% to plants, 11% to fish, but only 4% to fungi and 5% to protists.
• Identification errors. Even curated databases may contain errors. Nilsson et al. (2006) showed that up to 20% of ITS sequences in GenBank may be misidentified.

Consequences. The absence of reference sequences leads to:

• Impossibility of sample identification
• False identifications (the sample is assigned to the closest species in the database, even if this is incorrect)
• Incorrect assessment of biodiversity

Examples:

• Nematodes. Floyd et al. (2002) noted that only a small proportion of described nematode species have SSU sequences in GenBank. For most species, reference data are absent.
• Fungi. Schoch et al. (2012) showed that, despite the presence of 172,000 ITS sequences in GenBank, only 56% of them are linked to Latin names. Many fungal species are not represented in databases.
• Oomycetes. Robideau et al. (2011) noted that for many oomycete genera (for example, Plasmopara, Bremia), reference ITS sequences are absent or fragmentary.

Solutions. To overcome this problem, it is necessary to:

• Expand reference databases, especially for poorly studied groups and regions
• Ensure linkage of sequences to vouchers
• Check data quality in public databases
• Use “proximal” identification methods (for example, the BOLD BIN system)

4.7. Contamination and low DNA quality

DNA quality and contamination are technical problems that can seriously affect DNA barcoding results.

Contamination:

• Sources. Reagents, equipment, the researcher's hands, other organisms in the sample, laboratory environment.
• Consequences. False-positive results (amplification of foreign DNA), false identifications, impossibility of interpretation.
• Examples. In analyzing traditional medicines, Coghlan et al. (2012) found DNA of Asiatic black bear (Ursus thibetanus) and saiga (Saiga tatarica) in samples where they were not declared. This may be either contamination or intentional adulteration.
• Control. Use of negative controls at all stages, disposable materials, sterilization, work in a laminar flow hood.

DNA degradation:

• Causes. Processing (heating, drying, fermentation), storage (temperature, humidity), sample age.
• Consequences. DNA fragmentation, inability to amplify long fragments, sequencing errors.
• Examples. Staats et al. (2016) noted that in processed food products and traditional medicines, DNA is often fragmented to 100–300 base pairs. Mini-barcodes are required for analyzing such samples.
• Solutions. Use of mini-barcodes, special DNA extraction methods, increasing the number of PCR cycles.

PCR inhibitors:

• Sources. Polysaccharides, polyphenols, humic acids, heme, bile salts.
• Consequences. No amplification, reduced PCR efficiency.
• Solutions. Additional DNA purification, addition of BSA, PVPP, dilution of the sample.

Low DNA content:

• Examples. Old museum specimens, herbarium specimens, single cells.
• Solutions. Whole genome amplification, nested PCR, sensitive detection systems.

4.8. Problems with specific groups of organisms

Different groups of organisms have specific problems for DNA barcoding.

Animals:

• Cnidarians. Hebert et al. (2003) noted that COI does not distinguish closely related cnidarian species due to the low rate of evolution.
• Nematodes. Powers (2004) showed that in Meloidogyne the mitochondrial genome has an unusual structure, which limits the use of COI. ITS may also give ambiguous results due to the hybrid origin of some species.
• Mollusks. Meyer and Paulay (2005) showed that in gastropods intraspecific COI variability can overlap with interspecific variability.

Plants:

• Polyploidy and hybridization. China Plant BOL Group (2011) and Hollingsworth et al. (2011) showed that in plants plastid barcodes often do not correspond to species boundaries due to hybridization and introgression.
• Low resolving power. CBOL Plant Working Group (2009) noted that rbcL + matK distinguish only 49.7% of species in some groups.
• Problems with ITS. China Plant BOL Group (2011) found that ITS can be difficult to amplify in gymnosperms and may also contain paralogues and pseudogenes.

Fungi:

• High intraspecific variability. Schoch et al. (2012) noted that in Glomeromycota intraspecific ITS variability reaches 20%.
• Absence of universal primers. For some fungal groups, ITS primers do not work.
• Identical ITS in different species. In Aspergillus, Penicillium, Fusarium, ITS may be identical in several species.

Bacteria:

• Limitations of 16S rRNA. Janda and Abbott (2007) showed that 16S rRNA provides species-level identification in only 65–83% of cases.
• Horizontal gene transfer. Can lead to mismatch between 16S rRNA and other genes.
• Multiple copies. Some bacteria have several copies of 16S rRNA with different sequences.

Oomycetes:

• Problems with ITS. Robideau et al. (2011) noted that in some genera (for example, Plasmopara) ITS contains long insertions, complicating amplification.
• Insufficient variability. For some groups, ITS does not distinguish closely related species.
• Problems with COI. Choi et al. (2015) showed that cox1 does not amplify in some genera (for example, Eurychasma, Sclerophthora, Albugo lepigoni).

4.9. Practical examples

Example 1: Canadian mosquitoes (Cywinska et al., 2006).

Cywinska et al. (2006) barcoded 37 mosquito species from Canada. They found:

• Mean intraspecific COI divergence—0.5%
• Mean interspecific divergence—10.4%
• However, in Aedes fitchii and Ae. grossbecki, interspecific divergence was low (0.2–1.1%), while intraspecific divergence in Ae. fitchii reached 3.6–3.9%
• One sample of Culex restuans contained a stop codon and 15 amino acid substitutions, indicating a NUMT
• Morphological signs of hybridization in Ae. fitchii

Example 2: Mollusks (Meyer and Paulay, 2005).

Meyer and Paulay (2005) analyzed 2,026 individuals of 263 species of marine gastropods (Cypraeidae). They found:

• Intraspecific COI divergence—0–3.9%
• Interspecific divergence—0–18%
• Optimal threshold for minimizing errors—2.6%
• Even at the optimal threshold, error reached 17%
• When traditional morphological species were used, error increased to 33%

Example 3: Fungi (Schoch et al., 2012).

Schoch et al. (2012) analyzed 742 fungal strains. They found:

• ITS showed a probability of correct identification of 0.73
• In Glomeromycota, intraspecific ITS variability reached 20%
• ITS was identical in several species of Aspergillus, Penicillium, Fusarium
• RPB1 showed higher resolution but worse amplifiability

Example 4: Oomycetes (Choi et al., 2015).

Choi et al. (2015) compared cox1 and cox2 for 31 oomycete genera. They found:

• cox1 did not amplify in Eurychasma, Sclerophthora, Albugo lepigoni
• cox2 amplified in all 31 genera
• cox2 showed higher nucleotide divergence and better resolution
• However, even cox2 did not distinguish some closely related species

5.1. Introduction: from a key to an algorithm

Traditional taxonomy has for centuries relied on analogue tools: herbarium collections, morphological keys, descriptions in floras and faunas. Identification required physical access to specimens, expert knowledge, and considerable time. The digital revolution of recent decades has radically changed this situation (Wäldchen and Mäder, 2018; Joly et al., 2016).

Digital taxonomy is a set of methods and tools that use digital technologies for identification, description, and classification of organisms. It includes interactive keys, image databases, mobile applications, machine learning, and automatic identification by DNA (Wäldchen and Mäder, 2018). These tools do not replace traditional taxonomy, but extend its capabilities, making identification faster, more accessible, and more objective.

The key driving force of digital taxonomy has been the accumulation of enormous data arrays: millions of digitized herbarium specimens, millions of photographs of organisms, millions of DNA sequences. At the same time, the development of machine learning, especially deep neural networks, has made it possible to automatically extract patterns from these data (Wäldchen and Mäder, 2018; Shen et al., 2026).

In this chapter we will consider the main directions of digital taxonomy and automatic identification, their capabilities, limitations, and prospects for systematics and diagnostics of organisms of the agrosphere.

5.2. Digital keys: interactive keys

Traditional dichotomous keys described in Chapter 3 of Singh's textbook (2019) have a significant limitation: the user must follow a fixed sequence of characters, and if information on any character is unavailable, identification becomes impossible. Digital technologies have made it possible to create interactive keys free from this drawback (Singh, 2019; Watson et al., 2015).

Multi-access keys (polyclaves). Unlike dichotomous keys, multi-access keys allow the user to choose characters in any order. Each character is used to successively exclude taxa until one or several candidates remain. This is especially convenient when working with incomplete material, when some characters are unavailable (Singh, 2019).

Historically, multi-access keys were implemented using punched cards: body-punched cards (with holes in the body of the card) and edge-punched cards (with holes along the edge). The user selected cards corresponding to the specimen's characters and superimposed them on one another until one taxon remained (Singh, 2019).

Computer interactive keys. Modern digital keys implement the same principle, but using computer programs. The best-known systems are:

• DELTA (DEscription Language for TAxonomy). A system developed by CSIRO (Australia) that allows taxon descriptions to be created in a standardized format and, on their basis, generates dichotomous keys, interactive Intkey keys, and exports data for phylogenetic analysis. DELTA is used for a wide variety of organismal groups, including fungi, plants, and insects (Singh, 2019).
• Lucid. A commercial system for creating interactive keys, developed at the University of Queensland (Australia). It includes Builder (for creating keys) and Player (for using them). It supports multimedia, images, and fact sheets (Singh, 2019).
• NaviKey. A free Java program for interactive identification that works with DELTA files. It allows keys to be used both offline and through a web interface (Singh, 2019).
• ActKey. A web application developed at the Missouri Botanical Garden, supporting data in DELTA, Excel, Access, and Lucid formats. It is used for the floras of China, North America, Madagascar, and Borneo (Singh, 2019).
• MEKA (Multiple-Entry Key Algorithm). A free Windows program that allows specimens to be identified by selecting characters in arbitrary order. It converts to SLIKS format for web use (Singh, 2019).
• IdentifyIt. A commercial Linnaeus II system for creating taxonomic databases and interactive keys (Singh, 2019).

Advantages of interactive keys:

• Ability to use any available characters
• Work with incomplete material
• Ability to include images, sounds, video
• Ease of updating and distribution
• Accessibility via the internet

Limitations:

• Require creation and maintenance of a database
• May be difficult for non-specialists
• Dependence on completeness and quality of data
• Need for computer equipment

5.3. Image databases and digital collections

Digitization of natural history collections is one of the most important trends in modern systematics. Herbaria, zoological museums, and botanical gardens around the world are creating digital images of their specimens and making them available via the internet (Watson et al., 2015; Wäldchen and Mäder, 2018).

Virtual herbaria. A virtual herbarium is a database containing images of herbarium specimens and accompanying information (labels, metadata). Advantages of virtual herbaria:

• Reduced physical wear of specimens
• Fast search by taxonomic, geographic, and other criteria
• Accessibility for researchers worldwide
• Possibility of integration with other databases
• Preservation of information if physical specimens are lost (Watson et al., 2015; Singh, 2019)

Examples of virtual herbaria:

• NYBG Virtual Herbarium (New York Botanical Garden)—850,000 specimens, 120,000 images
• eFTG (Fairchild Tropical Garden)—more than 100,000 specimens, integration with other herbaria
• AVH (Australia's Virtual Herbarium)—more than 6 million specimens from Australian herbaria
• ePIC (Royal Botanic Gardens, Kew)—577,156 digitized specimens, including 170,760 types
• iDigBio—aggregator of US museum data, more than 1.8 million georeferenced images of vascular plants (Watson et al., 2015; Wäldchen and Mäder, 2018)

Digital images of living organisms. In addition to herbarium specimens, huge databases of photographs of living organisms in nature are being created. Examples:

• iNaturalist—more than 6.6 million observations, 127,000 species (Wäldchen and Mäder, 2018)
• Encyclopedia of Life (EoL)—information about all organisms, including images (Schierwater and DeSalle, 2022)
• CalPhotos—more than 640,000 images of plants, animals, fossils (Singh, 2019)
• Plant Systematics website (Cornell University)—more than 20,000 images of vascular plants (Watson et al., 2015)

Problems of digital collections:

• Image quality. Not all images have sufficient resolution for identification.
• Lack of standardization. Different institutions use different formats, metadata, and protocols.
• Identification errors. Not all specimens in databases are identified correctly.
• Incompleteness. Many species, especially rare and tropical ones, are underrepresented.
• Copyright. Not all images are available for free use.

5.4. Mobile applications for identification

The development of smartphones with powerful cameras and internet access has created a new opportunity for identifying organisms—mobile applications using machine learning. These applications make identification accessible to the general public, including farmers, students, gardeners, and citizen scientists (Wäldchen and Mäder, 2018; Joly et al., 2016).

Pl@ntNet. One of the best-known applications for plant identification. Developed by a consortium of French research organizations (Cirad, INRA, Inria, IRD) together with the Tela Botanica network (Joly et al., 2016; Wäldchen and Mäder, 2018).

Pl@ntNet allows users to:

• Upload one or several photographs of a plant (flower, leaf, fruit, bark)
• Obtain a list of the most probable species with confidence estimates
• View images and descriptions from the database
• Share an observation with the community
• Participate in validation and enrichment of the database

A key feature of Pl@ntNet is the use of crowdsourcing for collecting and validating data. Users can upload their observations, which are checked by experts and the community. This creates a self-improving system: the more data, the better the identification (Joly et al., 2016).

Since 2015, Pl@ntNet has used deep learning (CNN pretrained on ImageNet and fine-tuned on Pl@ntNet data). This led to a significant improvement in accuracy: the application's rating in Google Play rose from 3.2 to 4.3 (Wäldchen and Mäder, 2018).

Limitations of Pl@ntNet:

• Accuracy varies depending on plant group and image quality
• The database was initially oriented toward European flora, but is expanding
• For some groups (for example, grasses, sedges), accuracy is lower
• Users must treat results critically, especially for rare species (Joly et al., 2016; Kuchařová et al., 2025)

iNaturalist / Seek. iNaturalist is a citizen science platform that allows users to upload observations of organisms. A community of experts and amateurs identifies observations; when 2/3 of identifiers agree, the observation receives “research grade” status (Wäldchen and Mäder, 2018).

Seek by iNaturalist is a mobile application for automatic identification in real time. It uses the smartphone camera to recognize plants, animals, and fungi. Unlike iNaturalist, Seek does not require registration and does not transmit data to the common database (Kuchařová et al., 2025).

Kuchařová et al. (2025) tested Seek on 40 species of conifers in a botanical garden in the Czech Republic. The results showed:

• Species-level accuracy—38.75%
• Genus-level accuracy—92.5%
• Common species with bright characters were identified most successfully (Larix decidua, Picea abies, Thuja occidentalis)
• Rare species and species with complex morphology were identified poorly
• Lighting influenced accuracy: best under light cloudiness (diffuse light)

LeafSnap. An application for identifying trees from photographs of leaves. Developed by Columbia University, University of Maryland, and the Smithsonian Institution. It uses computer vision to analyze leaf shape (Wäldchen and Mäder, 2018).

Merlin Bird ID (Photo ID). An application by the Cornell Laboratory of Ornithology for identifying birds from photographs. It uses TensorFlow and eBird data to generate a list of probable species taking date and location into account (Wäldchen and Mäder, 2018).

BirdNET. An application for automatic identification of birds by sound. Developed by the Cornell Lab of Ornithology and Chemnitz University of Technology. It uses CNN to analyze audio recordings. BirdNET is capable of identifying more than 3,000 bird species (Usman et al., 2025; Shen et al., 2026).

Usman et al. (2025) conducted a systematic review of 25 studies using BirdNET. Main conclusions:

• Precision varies from 72% to 85%
• Recall varies from 33% to 84%
• Accuracy depends on species, recording quality, confidence threshold
• Geographic bias: most studies were conducted in North America (64.2%) and Europe (50%)
• Tropical regions are underrepresented

General problems of mobile applications:

• Geographic and taxonomic bias. Models are trained mainly on data from North America and Europe, which reduces accuracy for other regions (Usman et al., 2025; Kuchařová et al., 2025).
• Lack of training data. For many groups (fungi, invertebrates, tropical plants), data are insufficient.
• “Black box.” The user does not see on what characters the decision was based, which complicates verification.
• Algorithm errors. False-positive and false-negative results can lead to incorrect conclusions.
• Dependence on image quality. Lighting, angle, and background influence accuracy.

5.5. Machine learning for identification

Machine learning is a key technology underlying modern automatic identification systems. Wäldchen and Mäder (2018) provide an overview of the application of machine learning for species identification from images.

Basic concepts:

• Features. Numerical characteristics of an image used for classification. Traditionally, features were designed manually by experts (shape, texture, color). Modern deep learning methods extract features automatically.
• Classifier. An algorithm that, based on features, assigns an object to a class (species). Examples: support vector machine (SVM), Random Forest, artificial neural networks (ANN).
• Training. The process of tuning classifier parameters on a training set.
• Inference. Applying a trained classifier to new data.

Convolutional neural networks (CNN). CNN is a class of deep neural networks specially designed for image analysis. CNN architecture includes:

• Convolutional layers. Extract local features (edges, textures, shapes) using filters.
• Pooling layers. Reduce dimensionality and provide shift invariance.
• Fully connected layers. Classify the extracted features.

The key advantage of CNN is automatic feature extraction. Instead of manual feature design, the network learns by itself which characteristics are important for distinguishing species (Wäldchen and Mäder, 2018).

History of CNN development:

• 1980s: Neocognitron (Fukushima and Miyake, 1982)
• 1998: LeNet (LeCun et al., 1998)—recognition of handwritten digits
• 2012: AlexNet (Krizhevsky et al., 2012)—victory in ImageNet, reducing error from 26% to 15%
• 2014: VGGNet, GoogLeNet
• 2015: ResNet—the first architecture to surpass humans in ImageNet
• Present: Transformer-based models, foundation models

Application of CNN for species identification:

• Plants. PlantCLEF—an international competition on plant identification. From 2011 to 2017, accuracy grew from 0.47 to 0.92 (mAP) due to the transition to deep learning. In 2017, the dataset included 10,000 species and 1.6 million images (Wäldchen and Mäder, 2018).
• Animals. CNN are successfully used for identifying mammals (Norouzzadeh et al., 2018), fish (Qin et al., 2016; Salman et al., 2016), insects (Marques et al., 2018; Cheng et al., 2017), and aquatic invertebrates (Raitoharju et al., 2016).
• Herbarium specimens. Carranza-Rojas et al. (2017) applied CNN to 260,000 scans of herbarium sheets (1,204 species) and achieved accuracy of 80% (top-1) and 90% (top-5). However, transfer learning from herbarium specimens to field photographs proved impossible due to differences in visual characters (Wäldchen and Mäder, 2018).

Sound identification. In addition to images, CNN are used for sound analysis. BirdNET (Kahl et al., 2021) uses CNN to identify birds by calls. Shen et al. (2026) describe the evolution of methods for automatic bird sound recognition from manual features (MFCC) to deep learning and foundation models (Bird-MAE, Perch 2.0).

Identification by DNA sequences. Machine learning is also used for analyzing barcode sequences. For example, ABGD (Automatic Barcode Gap Discovery) uses statistical methods to automatically divide sequences into hypothetical species (Puillandre et al., 2012). Other methods include GMYC (General Mixed Yule Coalescent), Bayesian delimitation, and machine learning for taxonomic classification.

Problems of machine learning:

• Lack of data. For many groups and regions, training data are insufficient. This is especially acute for tropical regions and poorly studied taxa (Usman et al., 2025).
• Class imbalance. Some species are represented by many images, others by only a few. This biases the model toward frequent classes (Van Horn et al., 2018).
• Overfitting. The model may memorize training data and perform poorly on new data.
• “Black box.” Complex models (especially deep networks) do not provide interpretable explanations of their decisions. This complicates verification and causes distrust (Wäldchen and Mäder, 2018; Shen et al., 2026).
• Domain shift. A model trained on one data type (for example, herbarium specimens) may perform poorly on another (field photographs) (Wäldchen and Mäder, 2018).
• Geographic bias. Models trained on data from North America and Europe show worse results in other regions (Usman et al., 2025).

5.6. Automatic identification by DNA

In addition to identification by images and sounds, digital taxonomy includes automatic identification by DNA sequences. This direction includes metabarcoding and mini-barcoding (Deiner et al., 2017; Staats et al., 2016).

Metabarcoding. Metabarcoding is a method for analyzing complex mixtures of organisms using high-throughput sequencing (HTS). Unlike classical DNA barcoding, which analyzes individual samples, metabarcoding allows many species in a sample to be identified simultaneously (Deiner et al., 2017).

Metabarcoding workflow:

1. DNA extraction from a mixed sample (soil, water, stomach contents, food product)

2. Amplification of the barcode region with universal primers

3. High-throughput sequencing of amplicons

4. Bioinformatic analysis: clustering into MOTU (Molecular Operational Taxonomic Units), comparison with a reference database

5. Identification of species and assessment of their relative abundance

Applications of metabarcoding:

• Ecology. Assessment of soil, water, and air biodiversity (Deiner et al., 2017)
• Diet. Analysis of stomach contents and feces (Deiner et al., 2017)
• Food quality control. Detection of adulteration, identification of components of traditional medicines (Staats et al., 2016; Coghlan et al., 2012)
• Biosafety. Detection of invasive species and quarantine organisms (Staats et al., 2016)
• Forensics. Analysis of illegal wildlife trade (Staats et al., 2016)

Mini-barcoding. For analyzing degraded DNA (processed products, old samples), mini-barcodes are used—shortened fragments of standard barcodes (100–300 base pairs). Advantages: better amplifiability. Disadvantages: lower resolving power (Staats et al., 2016).

Metabarcoding in agronomy:

• Diagnostics of plant diseases (Robideau et al., 2011; Choi et al., 2015)
• Identification of pests (Powers, 2004)
• Monitoring of biodiversity of agrocenoses (Deiner et al., 2017)
• Quality control of seeds and planting material (Staats et al., 2016)

Problems of metabarcoding:

• Primer bias. Different species may amplify with different efficiency, distorting estimates of relative abundance (Deiner et al., 2017).
• Incompleteness of reference databases. Many species are absent from databases, leading to impossibility of identification (Staats et al., 2016).
• Sequencing errors. Although modern platforms provide high accuracy, errors are still possible.
• Chimeras. Chimeric sequences may form during PCR and must be removed (Deiner et al., 2017).
• Contamination. Contamination of a sample with foreign DNA can lead to false results (Staats et al., 2016).

5.7. Data integration: morphology + molecules + geography + ecology

Modern digital taxonomy strives to integrate different types of data for more reliable identification and species delimitation. This approach, known as integrative taxonomy, combines morphological, molecular, geographic, and ecological data (Schierwater and DeSalle, 2022; Watson et al., 2015).

Integrative taxonomy. Dayrat (2005) defined integrative taxonomy as an approach using all available data for species delimitation. Instead of relying on one type of data, integrative taxonomy combines:

• Morphology. Traditional diagnostic characters.
• Molecular data. Barcode sequences, phylogenetic markers.
• Geography. Distribution, sympatry/allopatry.
• Ecology. Habitat, hosts, behavior.
• Other data. Karyology, biochemistry, sounds, pheromones.

Schierwater and DeSalle (2022) describe the “taxonomic circle”: a hypothesis about the existence of a species is put forward on the basis of geographical or morphological data, and then tested with independent data. If at least one data source confirms the hypothesis, the species may be described. If all sources reject the hypothesis, the species is not recognized.

Examples of integrative taxonomy:

• Ferns. Use of rbcL, matK, trnH-psbA together with morphology for species identification (Kumar et al., 2022)
• Nematodes. Combination of SSU, ITS, COI with morphology for species delimitation (Powers, 2004)
• Oomycetes. ITS, COX1, COX2 together with morphology and pathogenicity (Robideau et al., 2011; Choi et al., 2015)
• Fungi. ITS, RPB1, RPB2, MCM7 together with morphology and ecology (Schoch et al., 2012)

Digital platforms for data integration:

• BOLD. Integrates sequences, images, metadata, geography (Ratnasingham and Hebert, 2007)
• GBIF. Global Biodiversity Information Facility—an aggregator of data on organism distribution (Watson et al., 2015)
• EoL. Encyclopedia of Life—information about all organisms (Schierwater and DeSalle, 2022)
• iDigBio. Aggregator of US museum data (Wäldchen and Mäder, 2018)

Problems of integration:

• Incompatibility of formats. Different databases use different standards.
• Absence of unique identifiers. Linking data from different sources is difficult.
• Data quality. Not all data are verified and curated.
• Ethical and legal aspects. Data access, copyright, privacy.

5.8. Problems of automatic identification

Automatic identification, despite impressive successes, has a number of significant problems that must be taken into account in practical use.

Algorithm errors. No algorithm gives 100% accuracy. Errors may be:

• False-positive. A sample is assigned to a species to which it does not belong.
• False-negative. A sample is not assigned to the species to which it belongs.
• Ambiguous. A sample is assigned to several species with similar scores.

Usman et al. (2025) showed that BirdNET precision varies from 72% to 85%, and recall from 33% to 84%. Accuracy depends on species, recording quality, and confidence threshold. For example, for Common Raven, precision varied from 0.29 to 0.94 across studies.

Lack of training data. Training models requires a large amount of labeled data. For many groups and regions, such data are insufficient:

• Tropical regions are poorly represented (Usman et al., 2025)
• Rare species are represented by single images (Van Horn et al., 2018)
• Some groups (fungi, invertebrates, protists) are poorly sampled (Wäldchen and Mäder, 2018)

Class imbalance. In nature, some species are common, others rare. In training data, this leads to imbalance: models recognize common species better and rare species worse (Van Horn et al., 2018).

“Black box.” Deep neural networks do not provide interpretable explanations of their decisions. The user does not see on what characters the decision was based. This complicates:

• Verification of identification correctness
• Detection of errors
• User training
• Trust in results (Wäldchen and Mäder, 2018; Shen et al., 2026)

Domain shift. A model trained on one data type may perform poorly on another. For example, a model trained on herbarium specimens performs poorly on field photographs (Wäldchen and Mäder, 2018). A model trained on recordings from North America performs worse in the tropics (Usman et al., 2025).

Geographic bias. Most training data have been obtained in North America and Europe. This creates inequality: models work well in some regions and poorly in others (Usman et al., 2025; Kuchařová et al., 2025).

Dependence on input data quality. Identification accuracy strongly depends on the quality of the image or sound:

• Lighting (Kuchařová et al., 2025)
• Angle, background
• Noise, overlapping signals (Usman et al., 2025)
• Resolution, focus

Lack of standardization. Different applications and databases use different formats, protocols, and criteria. This complicates comparison of results and data integration (Usman et al., 2025).

5.9. Prospects of digital taxonomy

Despite the problems, digital taxonomy is developing rapidly. Main directions of the future:

Improvement of models. Development of architectures (transformers, foundation models) and training methods (self-supervised learning, few-shot learning) makes it possible to increase accuracy and reduce the need for labeled data (Shen et al., 2026).

Expansion of training data. Projects on digitization of collections, citizen science, and crowdsourcing allow more and more data to be accumulated (Wäldchen and Mäder, 2018).

Data integration. Development of standards and platforms for integrating morphological, molecular, geographic, and ecological data (Watson et al., 2015).

Interpretability. Development of explainable AI methods to increase trust and understanding of algorithm decisions (Shen et al., 2026).

Portable devices. Development of field laboratories, portable sequencers (MinION), and mobile applications for real-time identification (Jain et al., 2016).

Ethical aspects. Development of ethical norms for the collection, storage, and use of genetic data, protection of privacy, and prevention of biopiracy (Bond and Scott, 2020).

6.1. Introduction: balance between enthusiasm and critical assessment

DNA barcoding and digital taxonomy have revolutionized the practice of systematics and diagnostics. However, like any technology, they have limits of applicability. In previous chapters we examined how barcoding works, when it is effective, and when it fails. In this chapter we will summarize the limitations of the method and discuss prospects for its development, including technological innovations, integrative approaches, and ethical aspects (Collins and Cruickshank, 2013; Taylor and Harris, 2012).

The key thesis of this chapter: DNA barcoding is a powerful but not universal tool. It does not replace morphology and taxonomic expertise, but supplements them. Digital taxonomy extends the researcher's capabilities, but requires critical thinking and understanding of limitations (Schierwater and DeSalle, 2022; Wäldchen and Mäder, 2018).

6.2. Limitations of DNA barcoding

6.3. Prospects: portable sequencers and field laboratories

Despite limitations, technological innovations open new possibilities for DNA barcoding and digital taxonomy. One of the most promising directions is miniaturization and portability of equipment.

6.4. Future: integrative taxonomy and global networks

6.5. Ethical aspects

The development of DNA barcoding and digital taxonomy raises a number of ethical questions that require attention.

7.1. Introduction: why agronomy needs DNA barcoding

Agronomy is an area where accurate identification of organisms is critically important. Plant diseases, pests, weed species, quarantine organisms, symbionts, and pathogens all require rapid and reliable determination. Traditional morphological diagnostics often faces limitations: larvae and eggs of pests are difficult to distinguish, pathogens in the initial stage of infection have no diagnostic characters, and fragmented material (plant parts, damaged tissues) cannot be identified (Powers, 2004; Armstrong and Ball, 2005).

DNA barcoding and digital taxonomy offer solutions to these problems. They allow:

• Identification of organisms at any stage of the life cycle (eggs, larvae, adults)
• Determination of species from tissue fragments, seeds, damaged samples
• Rapid and standardized diagnostics
• Detection of cryptic species that are morphologically indistinguishable but differ in ecology, pathogenicity, or pesticide resistance (Bickford et al., 2007; Armstrong and Ball, 2005)

In this chapter we will consider the main directions of practical use of DNA barcoding and digital taxonomy in agronomy: diagnostics of diseases and pests, quarantine, seed certification, monitoring of biodiversity of agrocenoses, and product quality control.

7.2. Diagnostics of diseases and pests

7.3. Quarantine and biosafety

7.4. Certification of seeds and planting material

7.5. Monitoring biodiversity of agrocenoses

7.6. Product quality control

7.7. Examples of successful application

7.8. Limitations and problems

Despite successes, the application of DNA barcoding in agronomy faces a number of limitations (Collins and Cruickshank, 2013; Taylor and Harris, 2012; Staats et al., 2016).

Incompleteness of reference databases. For many agricultural pests and pathogens, reference sequences are absent or fragmentary. This is especially acute for tropical regions and poorly studied groups (Taylor and Harris, 2012; Usman et al., 2025).

Geographic bias. Most data have been obtained for North America and Europe. For other regions, identification accuracy is lower (Usman et al., 2025).

Cryptic species. Many pests and pathogens are cryptic complexes that are not distinguished by standard barcodes (Bickford et al., 2007; Meyer and Paulay, 2005).

Hybridization and polyploidy. In plants, hybridization and polyploidy complicate identification (China Plant BOL Group, 2011; Hollingsworth et al., 2011).

DNA degradation. In processed products, old samples, and herbarium specimens, DNA may be fragmented. Mini-barcodes are required for analysis (Staats et al., 2016).

Contamination. Contamination of samples with foreign DNA can lead to false results (Collins and Cruickshank, 2013; Staats et al., 2016).

Cost and equipment. DNA barcoding requires laboratory equipment and trained personnel. This may be a limitation for small farms and developing countries (Jain et al., 2016).

Interpretation of results. DNA barcoding results are not always unambiguous. Correct interpretation requires knowledge of the biology, ecology, and taxonomy of organisms (Collins and Cruickshank, 2013).

8.1. Synthesis of the material covered

We have examined DNA barcoding and digital taxonomy as interrelated directions of modern systematics and diagnostics of organisms. Over seven chapters, we consistently analyzed the theoretical foundations of the method, its technological implementation, areas of application, limitations, and prospects. The time has come to summarize the key conclusions and formulate practical recommendations for specialists in the field of systematics and taxonomy of organisms of the agrosphere.

The central conclusion of the lecture can be formulated as follows: DNA barcoding is a powerful but not universal tool; digital taxonomy extends the researcher's capabilities but requires critical thinking; the optimal approach is a combination of classical and molecular methods (Collins and Cruickshank, 2013; Taylor and Harris, 2012; Schierwater and DeSalle, 2022).

This conclusion is not trivial. It reflects the evolution of views on the role of molecular methods in systematics—from initial enthusiasm bordering on the expectation of a complete replacement of morphology to a more balanced understanding of the capabilities and limitations of each approach (Hebert et al., 2003; Taylor and Harris, 2012).

8.2. What we have learned about DNA barcoding