DNA barcoding markers and their performance in plant identification

A group of 50 randomly selected taxa was used to evaluate the intraspecific genetic variability for the three markers. The results of this preliminary survey are reported in Table S2. Amplification and sequencing success were achieved for all samples except Hieracium racemosum for rbcL, Koeleria lobata for matK, and Cistus salvifolius, Hieracium racemosum, and Stellaria media for psbA-trnH. The rbcL sequences showed an averaging complete intraspecific identity for all but six species with a maximum of 0.7% for Inula hirta. The matK and trnH-psbA showed a certain degree of intraspecific variability, but K2P values were consistently lower than 2%, as was expected in the case of a strongly heterogeneous local flora. Based on the reduced intraspecific variability for the three markers, and since this work was not meant to generate alpha-taxonomy, we hereafter conducted our analyses on one sample for each species. This strategy was used to define a local DNA barcoding library to evaluate the discriminating power of the tested markers in the plant identification processes.

High quality and good yield of DNA (from 30 to 50 ng/µl) was obtained from all 347 samples, but 4 species: Inula spiraeifolia, Genista germanica, Trifolium arvense subsp. arvense, and Calamintha nepeta. For the latter, electrophoretic analysis showed partially degraded DNA in the 100–1000 bp range and low yield of DNA extraction (data not shown). As a consequence, marker-specific DNA barcoding libraries were defined on a total of 343 taxa. The rbcL was successfully amplified and sequenced in ca. 98% of the tested samples, and trnH-psbA and matK in ca. 94% of the samples. However, the latter required three different sets of primers (Table 1). Accession numbers for each DNA barcoding sequence are provided in Table S1. On the whole, a total of 323 sequences for matK, 337 sequences for rbcL and 323 sequences for trnH-psbA were submitted to Genbank as ‘standard barcode’. In our DNA library, one sequence for at least one of the three DNA markers was obtained from all samples, while at least two markers were sequenced from 304 samples (88.6%). The amplification and sequencing of all three markers was obtained for 300 samples (87.5%, see Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. List of primer pairs and PCR annealing temperatures used in the present study for the three selected DNA barcoding markers.

https://doi.org/10.1371/journal.pone.0043256.t001

K2P molecular distance (converted into percentage), was used to evaluate the discriminating power of the three DNA barcoding markers in the total flora and in 8 congeneric groups (G1–8). For each group, morphological characters necessary to achieve a correct identification by using the digital key were also reported (Table 2). Between the two core barcode markers the highest genetic variability was observed for matK, with a mean K2P value of 27.9% (34.1% for primer copy A, 21% for B and 28.1% for C, data not shown) computed on 323 samples, which lead to the definition of 313 Molecular Operational Taxonomic Units (MOTUs) (data not shown). In the 8 congeneric groups, K2P values ranged from 0.4% (Gr5) to 5.2% (Gr2). Complete sequence identity was observed for some taxa of Gr5 (Prunus spinosa subsp. spinosa, P. cerasifera var. cerasifera and P. cerasifera var. pissardii), Gr6 (Senecio vulgaris and S. inaequidens) and Gr7 (Solanum villosum subsp. alatum and S. nigrum).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results of DNA barcoding analysis performed for the 8 considered congeneric groups.

https://doi.org/10.1371/journal.pone.0043256.t002

The rbcL sequence showed the lowest genetic variability, with 308 MOTUs on 337 amplified sequences. The average genetic distance on total flora was 10.8%. Complete sequence identity was observed in Gr1 (Acer campestre and A. pseudoplatanus), Gr2 (Euphorbia maculata and E. nicaeensis subsp. nicaeensis), Gr3 (Geranium sanguineum and G. molle), Gr4 (Medicago lupulina and M. minima; M. falcata subsp. falcata and M. sativa), Gr6 (Senecio vulgaris and S. inaequidens), and Gr7 (Solanum villosum subsp. alatum and S. nigrum). K2P distances in Gr5 (Prunus samples) were lower than 1%, and five accessions were indistinguishable (Tab. 2).

The amplification of trnH-psbA marker produced 318 MOTUs on 322 amplified samples. The high genetic variability did not permit a complete alignment in all taxa. Genetic differences were the greater, ranging from 1.2% (Gr5) to 20.6% (Gr2). Few taxa in Gr5 (Prunus cerasifera var. pissardii and P. cerasifera var. cerasifera; P. spinosa subsp. spinosa and P. domestica subsp. insititia), Gr6 (Senecio vulgaris and S. inaequidens) and Gr7 (Solanum nigrum and S. villosum subsp. alatum) had identical trnH-psbA sequences.

DNA barcoding support to digital identification keys

On the basis of the three simulated scenarios, the digital key returned three groups of 37, 105 and 41 species respectively, which are undistinguishable due to missing characters in the specimens (see Table S1). Table 3 shows how DNA barcoding data improves the identification success by overcoming the absence of some seasonal morphological features. The use of rbcL alone (i.e. the most universal and less variable among the three DNA barcode markers) can reduce the uncertainty in the identification process, leading to the identification of a minimum of 92.7% (Scenario C), up to 95.0% (Scenario B) of the species. matK identifies from 95.0% (Scenario C) to 98.0% (Scenario B) of the species, while trnH-psbA has the highest values of discrimination (up to 100%). For this marker, MOTUs always correspond to the related taxa in two out of three scenarios.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Discriminating performance of the three tested DNA barcode markers (rbcL, matK and trnH-psbA) in the three scenarios (S) depicted by FRIDA digital key identification processes on the flora of Mt. Valerio.

https://doi.org/10.1371/journal.pone.0043256.t003

The use of the core barcode markers or their combination with trnH-psbA did not improve the discriminating success with respect to the use of the last marker alone (Table 3).

As expected, the use of the DNA barcoding does not discriminate among some congeneric species of groups B, Solanum villosum subsp. alatum and S. nigrum.

DNA barcoding and identification of plants

The variability of rbcL, matK and trnH-psbA sequences can be used to identify most of the plants occurring in the area of Mt. Valerio. Ca. 96% can be distinguished by combining the two core-barcode markers rbcL and matK. Similar results were reported in previous studies conducted in different areas [15]. [16] reported that the use of rbcL+matK permits to identify 92% of the woody species in a plot of 50 ha in the tropical forest of Panama. [17] identified 92.7% of the plants of the Koffler Scientific Reserve (Ontario, Canada), using the same markers. Thus, in relatively restricted areas, where a reduced number of closely related species is present [15], [17], as in our case, the combination of rbcL+matK is effective in identifying plant species.

However, there exist some constraints to the use of the two core-barcode markers. The matK gene is considered a good DNA barcode region because it is rapidly evolving [18], but its amplification requires several combinations of primers (3 in this study). As recently discussed at the Fourth International Barcode of Life Conference (www.dnabarcodes2011.org), the matK amplification system requires some improvements (i.e. the definition of clade-specific primers, or the identification of universal combinations of primers), in order to be effective when applied as a universal DNA barcode region for plants. On the other hand, the rbcL marker, which is easy to amplify, sequence and align, has a limited discrimination power, especially when among closely related species. These results are in agreement with the Fourth International Barcode of Life Conference (www.dnabarcodes2011.org/), during which matK and rbcL coding regions were, in any case, confirmed as universal core-barcodes.

As stated by other investigators [15], we support the use of the trnH-psbA region as an additional marker, especially when DNA barcoding is applied to closely related plant taxa. This region has highly conserved PCR priming sites, and a non-coding region with high numbers of substitutions. Hence, trnH-psbA can be a suitable marker to discriminate among closely related species. Although previous research reported the frequent occurrence of stutter PCR products for trnH-psbA due to mononucleotide repeats [19], recent technical advancements (i.e. appropriate polymerases; ideal PCR conditions see [20]) have suggested that these problems could be easily overcome.

In addition, as pointed out by [15], a complete exploration of plastidial non-coding markers (particularly trnH-psbA) could be useful to decide whether to incorporate them into core-barcode when dealing with plants.

A practical result of our investigation concerns the sampling strategy adopted to develop a DNA barcoding database useful for plant identification. In the local context, the low genetic intraspecific variability suggests that a DNA barcoding profile for only one individual per species is enough “to assist in the process of identifying unknown specimens to known species” [15]. We are aware that this strategy is not suitable for a classical alpha-taxonomy investigation, where a deep sampling coverage is necessary, but it is appropriate in our context where it was essential to characterize the local plant genetic profiles in order to identify unknown specimens, as also suggested by [17].

Integrated taxonomic identification system

Some authors suggested a superiority of molecular tools in comparison with “classic” identification keys based on morphological data [21]–[22]. In our opinion, however, molecular and morphological data should not be seen as colliding worlds, but as different solutions to a common problem. In some cases DNA barcoding is not successful in discriminating among species, which are closely related phylogenetically, as shown here and in several other studies [23]–[25]. Problems in achieving species identification by using DNA barcoding alone were reported within several angiosperm families, such as Orchidaceae [26], Ericaceae [27] and Lamiaceae [25]. DNA barcoding markers require a certain “evolutionary distance” among the taxa, in order to be used in their identification [14].

Morphology can be useful to discriminate among closely related taxa, as shown in Table 2. Morphological differences could derive from one or few DNA mutations [28]–[29] or from epigenetic variation [30], which could be not detected by DNA barcoding. However, identification keys based upon morphology could be difficult to use when some features are not visible, as happens when specimens are not well developed, or outside specific life stages (e.g. flowering period). For this reason, matrix-based digital keys are usually equipped with multi-entry or multi-access query interfaces [31], which do not force users to follow a fixed sequence of characters in the identification process. Also in these cases, however, the absence of some morphological features could make the identification impossible. In this case, the use of DNA barcoding could complete the identification process by compensating the limits of the morphological approach as clearly showed in our simulations.

While identification keys based on morphology discriminate among morphospecies, DNA barcoding discriminates among MOTUs, and sometimes these two “entities” could not match [32]. In this study, some specimens which showed the same DNA barcode sequences (e.g. Prunus spinosa subsp. spinosa and P. domestica subsp. insititia) were readily distinguished by morphological features, while species belonging to the genera Mentha (i.e. M. longifolia and M. pulegium subsp. pulegium), and Inula (i.e. I. hirta and I. spiraeifolia) showed similar morphological features, but were successfully distinguished by DNA barcoding. The integration of the two approaches, with the development of a multi-parametric identification system, may enhance the overall effectiveness, and represent a real advancement in plant identification. Such a system could be used in very different scenarios, from the identification of plant parts [33]–[35], to floristic researches and the discovery of new taxa [24].

Software such as FRIDA has the potential to integrate different data sources, including the capacity of using ‘filters’ to automatically reduce a large key by retaining only subsets of species which share a certain character. The inclusion of molecular characters as ‘filters’ does not present any technical problem. In an integrated system, DNA barcoding data could be used as a ‘filter’ by the software which produces and manages the interactive morphological keys. When molecular data are not sufficient to identify a plant at species level, the system will automatically produce a morpho-anatomical key only to the species which are not distinguished by barcode data. This would be an innovative approach to digital identification, which combines morphological and molecular data, overcoming the limits of both approaches, with the potential of becoming the core of a standardised protocol useful in biodiversity surveys, as a new integrated plant identification system, as already proposed by [22], in the so-called Automated Identification Technology (AIT). Moreover, thanks to the available platforms including laptops and mobile phones, these resources could be easily accessible to society-at-large to identify plants, as shown by the KeyToNature initiative (http://www.keytonature.eu/wiki/).

Experimental design

The area of Mt. Valerio (Trieste, NE Italy), which hosts a well-known flora, was selected to create a digital identification key and a DNA barcoding library. To investigate how molecular-based data could improve the effectiveness of a digital key, a two-step pipeline was followed. In the first step, the universality of three DNA barcode markers (i.e. matK, rbcL and trnH-psbA), the extent of their intraspecific variability, and their discriminating power on a subset of congenerics were investigated. In the second step, the performance of each marker (or of different combinations of them) was quantitatively estimated in three simulated scenarios in which the digital key could fail.

Survey area

Mt. Valerio is a low hill (215 m) in the north-east suburban area of Trieste. It is characterized by a submediterrean climate, strongly influenced by the Adriatic Sea, with a dry summer, rainy autumn and spring, and occasional frosts in winter. Average annual precipitation is 1016,9 mm, and average annual temperature is 14,1°C. Prevailing winds are the cold and dry Bora, blowing in winter from east-north-east, and the southern, mild and humid Scirocco. The substratum is Flysch, a base-rich formation of sandstones and marls. The whole survey area has a surface of ca. 0,25 km², and includes both highly and semi-natural sites with different types of vegetation. Small oak stands with Quercus petraea subsp. petraea and Q. pubescens subsp. pubescens cover the western and southern sides of the hill. A stand of Carpinus orientalis subsp. orientalis lies on the more humid western side, while an artificial pine wood of Pinus nigra subsp. nigra covers the northern and eastern slopes. Shrublands with Spartium junceum and more or less close formations of Robinia pseudoacacia are located mostly on the south-eastern slopes.

The digital key

The digital key to the flora of Mt. Valerio was generated by FRIDA, on the basis of morpho-anatomical, ecological and distributional data, plus original images and drawings, deriving from a floristic list by Poldini et al. (unpublished data). The key is freely accessible online at the address http://dbiodbs.units.it/carso/chiavi_pub21?sc=77, in Italian and English. It can be used with two query interfaces [7]: 1) single-access (dichotomous), which requires the choice between two states of a character at each step of the identification process, and 2) multi-entry, which allows the use of several characters at the same time, hence strongly reducing the list of organisms in the first step of the identification process, after which the identification continues with a dichotomous interface for the remaining species. At the end of the identification, a taxon page appears, with scientific name, description and images, which are useful as “visual census”, to verify the correctness of the identification.

Plant specimens for DNA analysis

A total of 347 species were sampled in the period 2009–2010. For each individual, young leaves or buds were collected from at least three different individuals, and stored at −20°C. All samples were vouchered as ‘MIB:ZPL’ following the protocol specified by the biorepositories initiative (www.biorepositories.org), and the data standards for BARCODE Records in INSDC (http://barcoding.si.edu/PDF/DWG_data_standards-Final.pdf). All experiments, procedures and ethical issues were conformed to the competent national ethical bodies. No specific permits were required for sampling activities, which were conducted in a non protected area, in accordance with the national and regional laws. The location was not privately owned or in any way protected and field studies did not involve endangered or protected species. Specimens and voucher codes are listed in Table S1.

A total of 100 mg of plant material was used for DNA extraction. Genomic DNA was isolated using the DNeasy Isolation and Purification kit (Qiagen, Hilden, Germany), to obtain high-quality DNA, free of polysaccharides or other metabolites that might interfere with DNA amplification [34].

DNA Barcoding analysis

DNA barcoding analysis was performed with three different DNA markers; the rbcL and matK coding regions and the non-coding trnH-psbA intergenic spacer of plastidial DNA. PCR amplification was performed by using puReTaq Ready-To-Go PCR beads (Amersham Bioscience, Freiburg, Germany) in a 25 µL reaction according to the manufacturer's instructions. PCR cycles consisted of an initial denaturation step for 7 min at 94°C, 35 cycles of denaturation (45 s at 94°C), annealing (30 s at different temperatures; see Table 1), and extension (1 min at 72°C), and a final extension at 72°C for 7 min. One universal primer pair was used for the amplification of rbcL and trnH-psbA, while three different combinations of primers were used for the amplification of matK, as suggested by [36]. Further details on primers and conditions of amplification are provided in Table 1. PCR products were bidirectionally sequenced by using an ABI 155 3730XL automated sequencer at Macrogen Inc., Korea. Manual editing of raw traces and subsequent alignments of forward and reverse sequences allowed to assign sequences to almost all the species. The 3′ and 5′ terminals were clipped to generate consensus sequences for each taxon. In order to avoid the inclusion of inadvertently amplified nuclear pseudogenes of plastidial origin (see [37]), barcode sequences were checked following the guidelines proposed by [38]–[39]. The sequences have been deposited in the EMBL Data Library.

To verify the universality of the three DNA barcode regions, the markers were routinely amplified and sequenced in the highest number. Only the most universal primer combinations for each candidate marker were tested (Table 1). For all taxa and loci, PCR amplifications in a two-stage trial were made. In the first stage, standard PCR conditions described above were used, starting from 10 ng of DNA template. Samples which were not amplified or which produced multiple or nonspecific PCR products were amplified again under less stringent conditions, by reducing the annealing temperature of 5°C, and executing 40 PCR cycles. In case of further failures, PCR products of both stages were amplified again by using 1 and 20 ng of DNA template. Any further negative result was considered a failure, and the corresponding samples were removed from the dataset.

To evaluate the intraspecific genetic variability of the markers, a total of three individuals for 50 randomly selected species were analysed (see Table S2). The performance of each marker was also evaluated by taking into account its resolution power on the total flora, and on 8 congeneric groups of strictly related taxa:

1. Gr1: Specimens of four species of Acer: A. campestre, A. monspessulanum subsp. monspessulanum, A. negundo and A. pseudoplatanus.
2. Gr2: Specimens of five species of Euphorbia: E. characias subsp. wulfenii, E. cyparissias, E. helioscopia, E. maculata, E. nicaeensis subsp. nicaeensis and E. peplus.
3. Gr3: Specimens of four species of Geranium: G. columbinum, G. molle, G. purpureum and G. sanguineum.
4. Gr4: Specimens of four species of Medicago: M. falcata subsp. falcata, M. lupulina, M. minima and M. sativa.
5. Gr5: Specimens of eight species and subspecies of Prunus: P. avium subsp. avium, P. cerasifera, P. cerasifera var. pissardii, P. domestica subsp. insititia, P. laurocerasus, P. mahaleb, P. persica and P. spinosa subsp. spinosa.
6. Gr6: Specimens of four species of Senecio: S. gibbosus, S. inaequidens, S. jacobaea and S. vulgaris.
7. Gr7: Specimens of four species of Solanum: S. dulcamara, S. lycopersicum, S. nigrum and S. villosum subsp. alatum.
8. Gr8: Specimens of six species of Trifolium: T. arvense, T. campestre, T. montanum subsp. montanum, T. pretense subsp. pratense, T. repens and T. rubens.

For each group and for each tested marker, DNA barcode sequences were aligned using MUSCLE – default options [40]. According to the guidelines provided by the Consortium for the Barcoding of Life (http://www.barcoding.si.edu/protocols.html), levels of genetic variation were analyzed by using MEGA 4.0 [41], to generate Kimura 2-parameter (K2P) distance matrices for each locus. Taxa which showed complete identity of DNA barcode sequence were considered non-identifiable with the marker under examination.

Three hypothetical scenarios from the digital key

To simulate situations in which a digital identification key based on morphological features could fail in achieving correct species identification, three different scenarios were defined. A description is given for each scenario, together with the list of characters used in the digital identification key (multi-entry query interface). In all cases, the identification, which should continue with the dichotomous query interface, fails because of missing features on the specimen. The simulations use a special version of the digital key containing only those taxa for which at least a sequence was successfully amplified.

Scenario A – The simulation takes into account a specimen of a non laticiferous, terrestrial herbaceous plant with opposite, non-whorled, entire leaves, collected out of its flowering period. Characters used in the multi-entry query interface of the digital key are: 1) herbaceous plant or a small shrub (<50 cm), 2) green, with chlorophyll, 3) without spines, 4) terrestrial, 5) leaves opposite, 6) leaves simple, 7) leaves not whorled.

The result from the multi-entry query interface is a list of 37 taxa, the identification of which can continue by using the dichotomous query interface, which asks whether the plant is laticiferous or not (answer: no) and then whether flowers have petals or not. The identification process, at this point, cannot be continued, because the specimen does not have flowers.

Scenario B - The simulation takes into account a specimen of a non laticiferous, terrestrial herbaceous plant with alternate, simple leaves, collected out of its flowering period. Characters used in the multi-entry query interface of the digital key are: 1) herbaceous plant or a small shrub (<50 cm), 2) green, with chlorophyll, 3) without spines, 4) terrestrial, 5) with leaves, 6) leaves not opposite, 7) leaves entire, 8) leaves not whorled

The result from the multi-entry query interface is a list of 105 taxa. The dichotomous query interface asks whether the flowers have petals or not. The identification process, at this point, cannot be continued, because the specimen does not have flowers.

Scenario C - The simulation takes into account a specimen of a tree collected in late Autumn, with dry fruits but missing leaves. Characters used in the multi-entry query interface are: 1) tree, woody climber or shrub >50 cm tall, 2) not a woody climber, 3) deciduous, 4) terrestrial, 5) fruit dry.

The result from the multi-entry query interface is a list of 41 taxa. The identification continues with the dichotomous interface, which asks whether the leaves are opposite or not. The identification process, at this point, cannot be continued, because the specimen does not have leaves.