Effectiveness of ITS and sub-regions as DNA barcode markers for the identification of Basidiomycota (Fungi).

BACKGROUND: Fungi are among the most abundant and diverse organisms on Earth. However, a substantial amount of the species diversity, relationships, habitats, and life strategies of these microorganisms remain to be discovered and characterized. One important factor hindering progress is the difficulty in correctly identifying fungi. Morphological and molecular characteristics have been applied in such tasks. Later, DNA barcoding has emerged as a new method for the rapid and reliable identification of species. The nrITS region is considered the universal barcode of Fungi, and the ITS1 and ITS2 sub-regions have been applied as metabarcoding markers. In this study, we performed a large-scale analysis of all the available Basidiomycota sequences from GenBank. We carried out a rigorous trimming of the initial dataset based in methodological principals of DNA Barcoding. Two different approaches (PCI and barcode gap) were used to determine the performance of the complete ITS region and sub-regions.
RESULTS: For most of the Basidiomycota genera, the three genomic markers performed similarly, i.e., when one was considered a good marker for the identification of a genus, the others were also; the same results were observed when the performance was insufficient. However, based on barcode gap analyses, we identified genomic markers that had a superior identification performance than the others and genomic markers that were not indicated for the identification of some genera. Notably, neither the complete ITS nor the sub-regions were useful in identifying 11 of the 113 Basidiomycota genera. The complex phylogenetic relationships and the presence of cryptic species in some genera are possible explanations of this limitation and are discussed.
CONCLUSIONS: Knowledge regarding the efficiency and limitations of the barcode markers that are currently used for the identification of organisms is crucial because it benefits research in many areas. Our study provides information that may guide researchers in choosing the most suitable genomic markers for identifying Basidiomycota species.

Background

Fungi are one of the major eukaryotic lineages that are equivalent in species number to animals but exceed that of plants [1]. Fungi are among the most important organisms in the world because of their vital roles in decomposition, nutrient cycling, and obligate mutualistic symbioses with plants, algae, and cyanobacteria [2]. Fungi also have great economic importance for industrial fermentation, pharmaceutical, and biotechnological industries [3]. They may also cause food spoilage and diseases in plants and animals [4]. The diversity of activities is reflected in the high number of taxa, morphologies, habitats, and life strategies used by this group of organisms. Further studies are necessary to better understand their complex interactions with other organisms and environments.

The phylum Basidiomycota is the second largest of the Fungi kingdom and comprises approximately 30% of all described fungal species [5]. This diverse phylum includes primarily macroscopic but also microscopic fungi, such as mushrooms and basidiomycotan yeasts, respectively [6, 7]; saprotrophs, such as wood-decaying fungi [8]; pathogens of plants [3] and animals [9, 10]; and mycorrhizal symbionts [11]. Basidiomycota species are grouped into the following subphyla: Agaricomycotina, Pucciniomycotina, and Ustilaginomycotina. The first is the largest subphylum with approximately one-third of all described fungal species [5, 12]. Thus, a substantial amount of the data that is currently available on diversity, distribution, and sequencing has targeted Agaricomycotina, particularly in the orders of Agaricales, Polyporales, and Boletales. This subphylum is primarily composed of wood decayers, litter decomposers, and ectomycorrhizal fungi, as well as pathogens and poisonous, hallucinogenic, or edible species [13].

The identification of fungi at the species level is critical to many research areas, such as health sciences and agriculture, where the determination of causal agents of diseases is central to the definition of the suitable treatment, elucidation of outbreaks, and transmission mechanisms [14, 15]. Furthermore, the understanding of the specific roles of microorganisms in an ecosystem, their abundance, and their community composition in ecological and biodiversity studies can only be attained through their reliable identification [16]. However, discovering and describing all extant fungal species appears challenging. According to the Dictionary of Fungi, only approximately 100,000 species have been described thus far [12], and the estimated diversity ranges from 1.5 to 5.1 million [1, 17, 18].

Morphological characteristics are useful for species description; however, they may be limited because many macroscopic structures are produced infrequently and temporarily [19], and many taxa often harbor cryptic species complexes [20]. Molecular tools complementing morphological ones are very promising in identifying species and can be used to rapidly and reliably evaluate biological diversity. These markers have been applied to the identification of fungal species since the 1990s [21, 22]; however, the strategy based on the sequencing of standardized genomic fragments (DNA barcoding) was recognized afterwards [23]. The primary difference between molecular identification tools and the “DNA barcode” approach is that the latter involves the use of a standard DNA region that is specific for a taxonomic group. The use of a segment of the mitochondrial gene encoding the cytochrome c oxidase subunit I (COI) has been proposed for animals [24]. For plants, various loci combinations have been proposed [25]; however, a study conducted by the Consortium for the Barcode of Life (CBOL) Plant Working Group agreed that the combination of sequences of two plastid genes, matK and rbcL, is the most promising plant barcode [26]. In 2012, the study conducted by Schoch and colleagues compared six DNA regions as promising universal barcodes for fungi. Mitochondrial COI and other protein-coding nuclear gene regions were excluded as potential markers for various reasons such as difficulties in amplifying DNA and insufficient variability. The nuclear ribosomal RNA internal transcribed spacer (ITS) region exhibited the highest probability of correct identification (PCI) for a wide number of fungal lineages analyzed and the most clearly defined barcode gap [27]. Since then, the ITS region has been accepted as the standard barcode marker for fungi. However, a thorough study of ITS sequences in the International Nucleotide Sequence Database (INSD: GenBank, EMBL and DDBJ) revealed that this region is not equally variable in all groups of fungi [28]. Notably, for some genera of Ascomycota, including Alternaria [29], Aspergillus [30], Cladosporium [31], Penicillium [30], and Fusarium [32], identification using the ITS barcode has been difficult.

One advantage of using the ITS region as a standard marker is that most fungal species have been identified based on this genomic region. GenBank [33] is the most comprehensive and widely used sequence repository in the field. A database specific for fungal sequences, the UNITE (User-friendly Nordic ITS Ectomycorrhiza Database) has been developed [34]. UNITE aims to unify the fungal taxonomic identification and correct the annotations associated with the taxonomic names to the greatest extent possible. The Barcode of Life Data System - BOLD [35] represents another bioinformatics platform; however, fungi remain underrepresented in it. BOLD supplies tools for the storage, quality warranty, and analysis of specimens and sequences to validate a barcode library. To obtain a barcode status on BOLD, sequences must fulfill some requirements, such as voucher data, collection record, and trace files. In the last few years, the scientific community has observed the rapid improvement of DNA sequencing technologies and the huge volume of data generated. Trimming and identifying this enormous amount of data requires bioinformatics tools, such as automated pipelines and various programs. However, the success of the analysis greatly depends on the correct taxonomic identification of sequences. Specifically, in the case of publicly available fungal ITS sequences, the reliability and technical quality vary significantly [34, 36]. Schoch and colleagues [27] estimate that only approximately 50% of the ITS sequences that are deposited in public databases are annotated at the species level. Moreover, Nilsson and colleagues [37] estimated that more than 10% of these fully identified fungal ITS sequences are incorrectly annotated at the species level. On the other hand, excellent initiatives, such as UNITE and that from NCBI that include a tool which allows flagging a GenBank sequence with type material [38] have emerged to minimize such a problem.

The ITS region comprises two sections (ITS1 and ITS2) that flank the conserved 5.8S region. The identification of multiple species from environmental samples (the DNA metabarcode) requires the use of high-throughput technologies, which may have limitations in sequencing read lengths [39]. For such approaches, only a portion of the ITS region is usually used, the ITS1 or the ITS2. The efficiency of these sub-regions in the identification of species in many fungal lineages has been evaluated, and some authors claim that ITS1 is more variable than ITS2 [28, 40–42]. Others have found opposite results [43] or that both the sub-regions are suitable as metabarcoding markers [44, 45]. In a recent work, Guarnica and colleagues [46] demonstrated that the ITS1 region is not more variable than the ITS2 region for Cortinarius. Furthermore, the complete ITS region is highly effective in discriminating among species in this highly sampled genus of Basidiomycota.

In the present study, an extensive comparative analysis based on the probability of correct identification (PCI) and barcode gap analyses was performed using a trimmed dataset composed of all Basidiomycota sequences deposited in GenBank. We evaluated the most widely used genomic markers for Fungi (the complete ITS region and the ITS1 and ITS2 sub-regions) to determine which is the most suitable for the identification of Basidiomycota species. Issues related to the need of additional molecular barcode markers as well as the taxonomic complexities within the subphyla are discussed.

Methods

Data acquisition and filtering

In this study, only sequences with complete nuclear ribosomal ITS from permanent collections whose taxonomic identifications were curated by specialists (voucher specimens) and deposited in GenBank [33] were used. Taxonomic information regarding the specimens was enriched, when available, from the UNITE database [34]. This step was used after downloading sequences from GenBank and before logical and quality filters were applied. For this enrichment, we firstly downloaded the FASTA sequence files from UNITE, and then we generated a tabular file with the UNITE data, keeping only the access numbers that corresponded to our specimens. Then, we retrieved the information related to sampling area and fungal classification from UNITE. Finally, we used the UNITE information to enrich the GenBank information.

Quality filters removed sequences with one or more IUB/IUPAC ambiguous characters, and logic filters ensured that the sequences were suitable for DNA barcode study in accordance with Barcode of Life recommendations (http://www.barcodeoflife.org/). The first logic filter guaranteed that only sequences identified at the species level were maintained in the database. Therefore, species with inconclusive names ('sp.', 'aff.', 'cf.', and 'uncultured') were removed. FungalITSExtractor [47] was used to guarantee that only sequences with complete ITS regions were maintained in the database. More than 99% of fungal complete ITS sequences deposited in GenBank are shorter than 800 or longer than 400 pb; thus all sequences outside of this interval were excluded from the dataset. The low representativeness together with the potential to distort the multiple sequence alignment justified this filter. Only species with specimens collected from at least three different localities were included to guarantee that only distinct and geographically distant specimens were evaluated and to avoid the possibility of working with genetically identical specimens. The list of all species used to perform the analyses of this study is provided (Additional file 1). All filters were performed using custom scripts written in the Perl programming language and are available upon request. The FungalITSExtractor software was used to identify and extract the ITS, ITS1, and ITS2 regions.

Data Analysis

The ITS, ITS1, and ITS2 datasets were partitioned in several sub-datasets, each containing sequences belonging to only one genus. Sequences from each sub-dataset were aligned using MUSCLE (version 3.8.31) with default parameters [48]. Distance matrices were generated using an uncorrected p-distance because it is simple and without any biological assumptions [49]. To evaluate the discriminative power of the three genomic markers, the probability of correct identification (PCI) was calculated as the ratio of species successfully identified per total number of species. A species was considered successfully identified if the minimum interspecific distance was larger than its maximum intraspecific distance [50]. Custom Perl scripts were written to calculate the distance matrices and the PCI values. Boxplots were plotted in R language.

Two statistical analyses were performed to graphically represent the data, a scatter plot and a dot plot. The scatter plot aimed to evaluate the correlations between the PCI values for the genomic regions pairwise combinations (ITS versus ITS1, ITS versus ITS2, and ITS1 versus ITS2), and the Spearman correlation coefficient was determined. The dot plot was used to compare the PCI with the barcode gap analyses. For this purpose, the PCI values for the four groups previously defined from the barcode gap analyses (Groups 1 to 4) were represented for each genomic region. All data and graphics were generated using Minitab (Minitab Statistical Software, version 17.3.1, State College, Pennsylvania: Minitab Inc., 2016).

Results

Our primary dataset was comprised of all complete ITS (ITS1 + 5.8S + ITS2) sequences of Basidiomycota and consisted of 37,699 sequences. The exclusion of sequences without the field ‘specimen_voucher’ in GenBank file reduced the number to 37,342. Removing sequences with ambiguous nucleotides led to 27,459 sequences, and removing sequences with inconclusive species names resulted in 21,238 sequences. After applying FungalITSExtractor, 19,578 sequences remained. ITS sequences with less than 400 bp and more than 800 bp were also excluded from the dataset, as well as ITS1 and ITS2 sequences less than 100 bp, leaving 19,149 sequences. The last filter was used to ensure that only species with at least three sequences collected from different geographic locations were retained in the dataset. Because most of the sequences did not include information regarding their origin, our final dataset had this number reduced to 7,731 sequences from 112 countries from six continents. This dataset was used to perform all subsequent analyses and represented three subphyla, five classes, 25 orders, 73 families, 211 genera, and 936 species (Additional file 2). This dataset has 167 sequences whose DNA were originated from biological specimens considered as type material. Many sequences from type materials were not included in our dataset only because they did not pass in quality and logic filters.

Although GenBank is known to be the most complete available public database, the amount of sequences is biased in our trimmed dataset as follows: 93.1% (7,197 sequences) belong to species of Agaricomycotina, whereas only 5.7% (442 sequences) come from Pucciniomycotina and 1.2%. (92 sequences) from Ustilaginomycotina. When other taxonomic ranks were analyzed, a similar distribution was observed with the vast majority of species belonging to Agaricomycotina (Fig. 1). Inside the subphyla, the imbalance in the amount of sequences is also enormous. For example, in Agaricomycotina, we found very well represented taxa (such as Cortinarius, with 829 sequences from 124 species) and others that were poorly represented (such as Auriscalpium, with only one species represented by three sequences). Most of the genera from the Agaricomycotina dataset were underrepresented; 126 of 194 had 20 or fewer sequences, whereas only 16 genera were represented by more than 100 sequences (Fig. 1 and Additional file 2).

Fig. 1

Pie charts represent abundance (number) of sequences (a) and species (b) for the three subphyla represented in the dataset used in this study. The histograms show the number of species and sequences distributed for genera belonging to Agaricomycotina (c), Pucciniomycotina (d) and Ustilagomycotina (e) phylum

The probability of correct identification (PCI) for the three genomic regions under study was estimated using our trimmed dataset (7,028 sequences from 113 genera). The number of genera analyzed decreased compared with the original dataset (211 genera) because we needed at least two species to estimate intraspecific and interspecific distances. Moreover, the sequences identified as type material are distributed in 27 distinct genera (23.9% of total) (Additional File 3), and only 25 sequences with RefSeq accessions interchangeably with GenBank numbers were identified (Additional File 4). This represented approximately only 0.36% of the sequences that comprised the dataset used to estimate PCI and barcode gap indices.

The mean PCI value for the complete ITS region was 63%, those for the sub-regions were slightly smaller as follows: 59% for ITS1 and 58% for ITS2. For the ITS region, 53.1% of the genera had PCI values higher than the mean, whereas for ITS1 and ITS2, these values were 46% and 48%, respectively (Table 1). The pairwise correlation between the three markers (ITS versus ITS1, ITS versus ITS2 and ITS1 versus ITS2) was estimated considering the PCI values of all genera composing the dataset. The comparisons between complete ITS and the sub-regions showed most of the data on or near the regression line, meaning that most of the PCI values were similar for the genera (Spearman correlation factor for ITS versus ITS1 = 0.8825 and for ITS versus ITS2 = 0.9102). When the sub-regions were associated (ITS1 versus ITS2), the distribution of data had a different profile and a lower correlation was observed (0.8158) (Fig. 2). The pairwise correlation between the genomic regions was carried out at the subphylum level; however, there were no observable patterns at this taxonomic level.

Table 1

Probable Correct Identification (PCI) values (%) for all of the Basidiomycota genera from our trimmed dataset. The PCI values were estimated for the three genomic regions studied, the complete ITS region (ITS1 + 5.8S + ITS2) and the sub-regions ITS1 and ITS2

Genera	ITS (ITS1 + 5.8S+ ITS2)	ITS1	ITS2
Agaricus	100	100	100
Alnicola	44	33	33
Amanita	24	27	27
Amyloporia	25	25	25
Antherospora	100	100	100
Antrodia	70	70	70
Antrodiella	75	50	75
Armillaria	20	20	20
Auricularia	80	40	60
Boletus	32	26	47
Butyriboletus	67	33	67
Calvatia	0	0	0
Ceriporiopsis	100	100	100
Chlorophyllum	100	100	100
Chroogomphus	50	50	50
Clavaria	100	100	100
Clavulina	33	33	33
Collybia	50	50	50
Coprinopsis	50	50	50
Cortinarius	36	45	37
Crepidotus	50	50	50
Cystoderma	50	50	33
Cystodermella	100	100	100
Datronia	0	50	0
Endoraecium	100	100	100
Entoloma	100	86	93
Entyloma	100	100	100
Exobasidium	100	100	100
Favolus	100	100	100
Fibroporia	100	100	67
Flammulina	100	50	75
Fomitopsis	50	25	50
Fuscoporia	100	100	100
Ganoderma	43	43	43
Geastrum	100	67	67
Gloeophyllum	100	100	100
Gymnopilus	100	100	50
Gymnopus	67	67	60
Hebeloma	42	37	32
Helicobasidium	100	75	100
Hohenbuehelia	0	0	0
Hydnellum	56	56	56
Hydnum	50	50	50
Hygrocybe	0	0	0
Hygrophorus	67	67	33
Hymenopellis	100	100	100
Hyphoderma	60	60	80
Hyphodermella	100	100	100
Hypholoma	0	0	0
Inocybe	30	19	28
Laccaria	0	0	13
Lactarius	37	37	41
Lactifluus	50	50	50
Lentinellus	43	43	29
Lentinus	57	57	71
Lepiota	92	83	75
Lepista	100	100	100
Leucoagaricus	90	100	80
Leucopaxillus	100	100	100
Lycoperdon	100	100	100
Lyomyces	100	100	100
Lyophyllum	100	100	67
Macrolepiota	50	50	33
Megacollybia	83	83	33
Melampsora	40	40	20
Melanoleuca	38	38	25
Microbotryum	70	80	50
Mucidula	0	0	0
Mycena	22	22	33
Neofavolus	100	100	100
Octaviania	100	100	100
Oligoporus	100	100	100
Parasola	100	67	67
Paxillus	0	0	33
Peniophorella	50	50	50
Phaeocollybia	0	0	0
Phanerochaete	67	67	67
Phellinus	100	33	67
Phellodon	86	86	71
Piloderma	50	50	50
Pisolithus	0	0	0
Pleurotus	29	43	14
Pluteus	27	27	23
Polyporus	100	100	86
Porodaedalea	100	100	100
Postia	50	50	50
Psathyrella	100	100	100
Psilocybe	100	100	100
Puccinia	43	43	43
Pycnoporellus	100	100	100
Ramaria	33	33	33
Resinicium	100	100	100
Rhizopogon	27	27	27
Rhodocollybia	100	100	100
Rigidoporus	100	100	50
Russula	38	27	36
Sarcodon	86	86	86
Scleroderma	33	33	33
Sebacina	33	0	33
Stephanospora	80	80	20
Strobilurus	67	67	67
Suillus	83	83	83
Thecaphora	100	100	100
Thelephora	0	0	0
Tilletia	100	33	100
Tomentella	25	50	25
Trametes	42	42	42
Tricholoma	43	57	29
Tricholomopsis	100	100	100
Tuberculina	67	0	67
Vuilleminia	33	33	33
Xerocomus	100	50	100
Xeromphalina	100	100	50

Fig. 2

Pairwise correlations (a, ITS X ITS1, b, ITS X ITS2 and c, ITS1 X ITS2) between PCI values of all genera from our dataset

Based on the analysis of the barcode gaps, we assessed and compared the efficiency of the three genomic markers for the identification of Basidiomycota. Thus, we classified the marker performance into the following three distinct categories: good, intermediate, or poor. When a clear barcode gap was present (e.g., Agaricus, Fig. 3a), we conventionally stated that the identification was good, even if outliers were overlapping. The genomic markers were considered intermediate if the whiskers from an intraspecific distance overlap those from an interspecific distance (e.g., Hebeloma, Fig. 3b), and poor if the boxes overlap or the intraspecific distance values were superior to those of interspecific distance (e.g., Lactarius, Fig. 3c). For most of the genera (91.5%) evaluated, the three genomic regions performed similarly, i.e., when the identification is good for one region, it is also good for the others. The same occurred when the performance was intermediate or poor. However, for some genera, we found some genomic regions with superior identification performance than others. For instance, the complete ITS had a clearer barcode gap for the genera Auricularia, Flammulina, Lentinellus, Microbotryum, Parasola, and Tuberculina compared with the ITS1 or ITS2 sub-regions. ITS1 performed better than the other regions in the identification of species from the genera Hygrophorus and Stephanospora, as well as ITS2 for the species belonging to the genera Amanita, Amyloporia, Fomitopsis, Scleroderma, and Strobilurus (Table 2, Group 2). In some instances, one of the three genetic markers performed worse than the other(s). The ITS1 sub-region is not sufficient to differentiate the species of the genera Collybia and Pleurotus, and the ITS2 is not a good marker for Sebacina, Hydnellum, or Vuilleminia. Finally, it is important to note that for 11 out of the 113 genera evaluated (Botyriboletus, Clavulina, Crepidotus, Hohenbuehelia, Hydnum, Laccaria, Lactarius, Mucidula, Peniophorella, Phaeocollybia, and Pisolithus), none of the complete ITS, ITS1 or ITS2 sub-regions could be used to differentiate the species based on the barcode gap analyses (Table 2, Group 4). For a detailed classification of genera considering their barcodes, see Table 2 and Additional File 5, where the boxplots for all analyzed genera are shown.

Fig. 3

Examples of the barcode gap performance classifications used in this study. a. Clear barcode gap (identification performance classified as good) for the genera Agaricus, b. Intermediate separation between the intra- and interspecific distances for Hebeloma and c. A poor barcode gap for Lactarius

Table 2

Grouping of Basidiomycota genera based on the barcode gap analyses (See Additional file 5 for more details)

Group 1	Group 2	Group 3	Group 4
Genera for which all the three genetic regions are good markers	Genera for which one or two genetic regions showed a clearer barcode gap and are recommended over the other (s)	Genera for which most of the genetic regions showed intermediate barcode gap and their use should be carefully evaluated	Genera for which all three genetic regions are poor markers
Agaricus, Antherospora, Antrodia, Ceriporiopsis, Chroogomphus, Clavaria, Coprinopsis, Cystodermella, Datronia, Edoraecium, Entoloma, Entyloma, Exobasidium, Favolus, Fibroporia, Fuscoporia, Geastrum, Gloeophyllum, Helicobasidium, Hygrocybe, Hymenopellis, Hyphodermella, Lactifluus, Lepiota, Lepista, Leucopaxillus, Lycoperdon, Lyomyces, Lyophyllum, Melampsora, Neofavolus, Octaviania, Oligoporus, Phanerochaete, Phellinus, Phellodon, Piloderma, Polyporus, Porodaedalea, Psathyrella, Psilocybe, Puccinia, Pycnoporellus, Resinicium, Rhodocollybia, Russula, Sarcodon, Suillus, Telephora, Thecaphora, Tilletia, Tricholomopsis, Xerocomus, Xeromphalina	Amanita (ITS2), Amyloporia (ITS2), Antrodiella (ITS and ITS2), Auricularia (ITS), Calvatia (ITS and ITS1), Chlorophyllum (ITS and ITS1), Flammulina (ITS), Fomitopsis (ITS2), Ganoderma (ITS and ITS1), Gymnopilus (ITS and ITS1), Gymnopus (ITS and ITS1), Hygrophorus (ITS1), Lentinellus (ITS), Leucoagaricus (ITS and ITS1), Macrolepiota (ITS and ITS1), Megacollybia (ITS and ITS1), Microbotryum (ITS), Parasola (ITS), Postia (ITS and ITS2), Rigidoporus (ITS and ITS1), Scleroderma (ITS2), Stephanospora (ITS1), Strobilurus (ITS2), Tuberculina (ITS)	Alnicola, Armillaria, Boletus, Collybia, Cortinarius, Cystoderma, Hebeloma, Hydnellum, Hyphoderma, Hypholoma, Inocybe, Lentinus, Melanoleuca, Mycena, Paxillus, Pleurotus, Pluteus, Ramaria, Rhizopogon, Sebacina, Tomentella, Tricholoma, Trametes, Vulleminia	Butyriboletus, Clavulina, Crepidotus, Hohenbuehelia, Hydnum, Laccaria, Lactarius, Mucidula, Peniophorella, Phaeocollybia, Pisolithus

The results of barcode gap analyses were compared with the PCI values for each genus using a dot plot (Fig. 4). For the genera for which the three genomic markers were classified as good in barcode gap analyses (Group 1, Table 2), most of the genera exhibited PCI above the mean value (63%); however, some disagreements were found. Some genera within this group had a PCI equal to zero (Datronia, Hygrocybe, Tecaphora, and Telephora) or between 20 and 50% (Chroogomphus, Coprinopsis, Lactifluus, Melampsora, Phellinus, Piloderma, Puccinia, Russula, Tilletia, Xerocomus, and Xeromphalina) (Fig. 4a). When the group for which one or two genomic regions showing a clearer barcode gap (Group 2, Table 2) was compared with the PCI, most of the genera had a PCI below the mean value (Fig. 4b). When the group for which most of the genomic regions showed an intermediate barcode gap (Group 3, Table 2), only Lentinus and Hyphoderma had higher PCI than mean value (both for the ITS2 region, Fig. 4c). When the groups for which all three genomic regions were classified as poor markers considering the barcode gap (Group 4, Table 2), most of the genera also had a PCI below the mean value (Fig. 4b) with the exception of Butyriboletus (with a PCI value above the mean for ITS and ITS2, Fig. 4d).

Fig. 4

PCI values for the genera classified in the Group 1 (a), Group 2 (b), Group 3 (c) and Group 4 (d) are represented for the ITS, ITS1 and ITS2 genomic region

Discussion

The accepted DNA barcode for Fungi is the rDNA ITS region [27]. ITS is recognized as a fungal barcode because it is the most sequenced region of fungi and is routinely used for systematics, phylogenetics, and identification [51, 52]. In this study, we downloaded all complete ITS sequences of species belonging to the phylum Basidiomycota from GenBank. Although this is the most complete repository of available ITS sequences, misidentifications or low-quality sequencing have been encountered in this public database [37]. However, some authors think that it is unrealistic that future databases or even a barcode database could be more reliable than GenBank because misidentified sequences would be as common as they are currently and because vouchers will not be re-identified by taxonomic experts (for a wide discussion, see [53-55]). To overcome this drawback, logical and quality filters were applied to our original dataset to obtain the most reliable results possible. The restrictiveness in the filtering step aimed to create a high-quality dataset (accurate taxonomic annotation and presence of relevant metadata) that would meet the theoretical assumptions of the biological system of identification via DNA barcode and the principles recommended in BOLD Systems [24, 35].

More than 90% of our trimmed dataset belonged to the subphylum Agaricomycotina. This result is not surprising because it reflects the high diversity of this taxon compared with the other subphyla, which is widely mentioned in the literature [5, 12]. Kirk and colleagues [12] estimated that one-fifth of all known fungal species described belong to the Agaricomycete clade; this diversity is considered to be underestimated because new taxa are continually being described [1, 56]. This discrepancy in the amount of species and sequences from the subphyla may reflect a natural event or may occur due to the specific interests of the scientific community in Agaricomycotina species.

Some criteria have been traditionally used to test the DNA barcoding efficacy to classify and/or identify specimens at the species level, such as similarity measures, tree-based techniques, and identification based on direct sequence comparison [57, 58]. However, all of these approaches present several issues (see [55] for a detailed discussion). Similarity measures are generally used to cluster sequences in “molecular operational taxonomic units”; however, the choice of the threshold value for distinguishing intraspecific and interspecific distances is largely arbitrary [58, 59]. An important and acceptable measure of the efficacy of a genetic marker should reflect the probability of correctly identifying a species. This concept has emerged as the probability of correct identification (PCI) [50, 53, 55, 60]. However, there is no consensus for the definition and calculation of PCI, which currently embraces a broad class of measures. In this work, we assume the concept described by Hollingsworth and colleagues [50] in which the authors considered the “discrimination as successful if the minimum uncorrected interspecific p-distance involving a species was larger than its maximum intraspecific distance” to measure the PCI for each genus included in our dataset. Furthermore, the use of genetic distances enables the observation of the ‘barcoding gap’, which is possible by plotting the intraspecific and interspecific distances. Therefore, an ideal barcode marker would reveal intraspecific divergences lower than interspecific divergences [61].

In this study, we aimed to identify the most suitable genomic marker (complete ITS, ITS1 or ITS2) to identify fungal species belonging to Basidiomycota. Our findings, based on PCI and barcode gap analyses, indicated that for most of the genera, the three genomic regions perform similarly, i.e., when one genomic region was considered a good marker (a PCI above the mean value or the presence of a clear barcode gap) the other regions were also; the same was observed when the performance of genomic markers was considered insufficient. When the performance of the genomic markers was individually evaluated, barcode gap analyses provided a more optimistic view than PCI values. Approximately half of the genera exhibited PCI values lower than the mean (63%); however, the three genomic regions were classified as good for most of the genera (Table 2) when the barcode gap is taken into account. Accordingly, the comparison between barcode gap and PCI for each genus showed some disagreements. This was primarily observed for some of the genera that showed good identification performance using the barcode gap but had low PCI values (Fig. 4a). The opposite, i.e., high PCI values and poor identification performance via barcode gap, was observed for only one genus, Botyriboletus (Fig. 4d).

Initially, the low PCI values found for some genera (such as Calvatia, Datronia, Hygrocybe, Hohenbuehelia, Hypholoma, Mucidula, and Pisolithus) could be explained by dataset features, such as the low number of species (genera represented by sequences from only two species) and/or by the high number of outliers, which would have distorted the PCI estimates. Additionally, the taxonomy appears very complex for many of the genera for which the identification performance using ITS and sub-regions were insufficient. Taxonomy issues for two genera (Hygrocybe and Thelephora) for which PCI values were low and three genera (Hypholoma, Phaeocollybia, and Pisolithus) for which both PCI and barcode gap analyses proved that ITS, ITS1 and ITS2 are not sufficient markers for the identification of species are discussed below based on pertinent literature.

Hygrocybe species exhibit extremely high variability in the ITS region, with sequences diverging by more than 25%. Thus, the use of additional DNA barcode markers has been proposed to re-evaluate the taxonomy of this genus [62, 63]. Moreover, significant changes in the classification of Hygrocybe, such as its division, are expected [64].

The phylogenetic relationships between and within species of Thelephora are also doubtful with ITS. The existence of cryptic species was described, and the importance of integrating morphological and molecular data, as well as employing a meaningful number of samples for the accurate identification is highlighted [65]. Hypholoma has been poorly studied. However, a recent study based on the morphological and molecular aspects of H. cinnabarinum samples showed that this species is not a member of the genus Hypholoma but belongs instead to Agaricus [66]. The ecological role of Phaeocollybia is uncertain. Smith [67] argues that the genus harbors both saprobes and mycorrhiza formers. Singer [68] considered that members of the genus were not obligatorily ectomycorrhizal, whereas Norvell [69] presented evidence for the consideration of Phaeocollybia as a mycorrhizal genus. At the taxonomic level, the complexity remains, as may be exemplified in Norvell [70]. The author proposed the re-evaluation of the genus Phaeocollybia by revealing four new agaric species morphologically similar to Phaeocollybia kauffmanii. The wide genetic divergence among Pisolithus ITS sequences [71-73] indicates significant evolutionary divergence and suggests that this genus encompasses a species complex. This hypothesis was reinforced by Kope and Fortin [74] who separated three groups of Pisolithus using incompatibility tests and basidiospore spine morphology.

According to Bickford and colleagues [20], cryptic species are two or more distinct species that are erroneously classified under one species name. Large intraspecific genetic distances associated with morphological and geographical discrete differences have revealed a broad range of cryptic species for many organisms and habitats [75, 76]. Although our knowledge of fungal species remains limited, the presence of cryptic species inside the group is well recognized [20] and was subsequently described for many of the genera covered in this study.

The use of molecular techniques, primarily DNA sequences, generates information to re-evaluate classifications and provides more accurate species delimitations [77]. Currently, the utility of DNA barcoding is evident. However, a universal barcode for the clear identification of all fungal species does not appear feasible, and secondary barcodes for Fungi have already been proposed [78]. In addition to the known limitations of ITS barcodes for some genera of Ascomycota, our results indicated that for some genera of Basidiomycota, such as Hygrocybe and Pisolithus, additional barcode markers may contribute to a clear elucidation of the complex relationships between and within species. The failure to correctly identify biological species hampers the efforts of the scientific community to conserve, study, or utilize them. Future research in this field should include discovering characteristics that natural selection acts upon [20].

Conclusions

Progress in many research areas fundamentally depends on the rapid and reliable identification of biological species. Most fungal diversity is unknown, and issues related to the conservation of these organisms are urgent; thus, studies related to species identification are crucial. Knowledge regarding the efficiency and limitations of the barcode markers that are currently used for specific groups of organisms optimize the work of many studies. Therefore, the present study contributes to the rational selection of barcode markers of species belonging to the phylum Basidiomycota.