Assessing Clivia taxonomy using the core DNA barcode regions , matK and rbcLa

Clivia Lindl., a shade-loving member of the family Amaryllidaceae J.St.-Hil., is endemic to South Africa and Swaziland and consists of six species, C. mirabilis Rourke, C. nobilis Lindl., C. caulescens R.A. Dyer, C. miniata (Lindl.) Regel, C. gardenii Hook., C. robusta B.G. Murray et al., and a natural hybrid C. ×nimbicola Z.H. Swanevelder et al. Clivia miniata is the only species with trumpet-like flowers, while the other species have pendulous flowers.


Introduction
There is a need for a reliable method to identify Clivia species based on any part of the plant. DNA barcoding has been successfully applied in identification of plant species (e.g. Bruni et al. 2015) as well as in re-evaluating taxonomic status and circumscriptions (e.g. Rastegar-Pouyani et al. 2014).
Two DNA barcoding regions have been selected as universal barcodes in plants, namely rbcLa and matK (CBOL Plant Working Group 2009). The rbcLa region is easy to amplify and sequenced over a broad spectrum of taxa (CBOL Plant Working Group 2009), and it has been suggested as a core barcoding region (Hollingsworth et al. 2009). Although matK is not useful in all plants (Lahaye et al. 2008), it is one of the most rapidly evolving coding sections of the plastid genome, and it is considered to be the equivalent of the animal barcode gene, CO1 (Hollingsworth, Graham & Little 2011).
This study assesses the current Clivia taxonomy and delimitations of the Clivia species with the core DNA barcodes (matK and rbcLa). In addition, the efficacy of these barcodes for Clivia identification is evaluated.

Materials and methods
A total of 110 specimens, representing six Clivia species and some atypical specimens, were obtained from legal collectors around the country.
DNA was extracted using the modified CTAB method of Rogstad (1992). The matK primer sets of Ford et al. (2009) and rbcLa primers of Levin et al. (2003) and Kress and Erickson (2007) were used for PCR amplification and sequencing amplification. DNA sequences were obtained using an Applied Biosystems 3130 Genetic Analyser. Sequence data are stored in the Barcode of Life Data System (BOLD: Ratnasingham & Hebert 2007) (Table 1).
The matK and rbcLa regions were separately analysed using the online tools in BOLD (Ratnasingham & Hebert 2007) and Geneious R6 software (http://www.geneious. com, Kearse et al. 2012). Sequences were aligned with the aid of Muscle (Edgar 2004). Only sequences with no missing data were analysed. The barcode gaps' (nearest neighbour, NN) intraspecific distances were determined for both regions separately using the Kimura-2-parameter distance model in BOLD Systems (Ratnasingham & Hebert 2007).  The matK and rbcLa sequences were combined using SequenceMatrix software (Vaidya, Lohman & Meier 2011). The MrBayes plugin Version 2.0.9 in Geneious R6 was used to construct a Bayesian Inference (BI) cladogram for phylogenetic analysis. The barcoding gap and species partitioning were determined with the online tool Automatic Barcode Gap Discovery (ABGD: Puillandre et al. 2011) with the default settings. A pairwise summary, best match and best close match were determined with the program SpeciesIdentifier (Meier et al. 2006). A threshold value of 0.349 was calculated from the pairwise summary (Meier et al. 2006). A median joining network was constructed in Network 4.6.1.1 to depict the number of mutational steps separating haplotypes of the six Clivia species.

Molecular analyses
Bidirectional sequences for both matK and rbcLa were obtained for 74 specimens. Only specimens with both gene regions and no missing data were included in this study. The aligned rbcLa region was 552 base pairs and the aligned matK region was 783 base pairs, resulting in a combined aligned length of 1335 base pairs.
The genetic distances of the two core barcoding regions were separately analysed in BOLD ( Table 2). The mean intraspecific variation of 0.02 in rbcLa is much lower than the 0.21 in matK.
The ABGD algorithm makes use of pairwise distances to group sequences according to proposed species (Puillandre et al. 2011). Depending on the maximal distance (P) tested (P = 0.00100, 0.001668 or 0.002783), the six species are divided into one to three groups, respectively.
In the pairwise SpeciesIdentifier-based analysis (Table 3), the total overlap is 0.42% (from 0.0% to 0.42%, covering 80.03% of all intra-and interspecific sequences) at a 0.34% intraspecific cut-off. The majority (71.61%) of the intraspecific distances are between 0.0% and 5.0% with 28.38% of the species having an intraspecific distance of 0.0%. The interspecific distances in the genus are mainly (95.87%) between 0.0% and 5.0%. Only a small percentage (4.12%) of the species has an interspecific distance equal to 0.0% and has, therefore, no genetic differences between them.
According to the 'best match' and 'all species barcodes' option in SpeciesIdentifier, correct species identifications are 37.33% and 12.0%, respectively (Table 3); incorrect Clivia identification can be as high as 10.66% when using the 'best match' option. The non-variable sites were removed from the sequences, and a network was constructed in Network 4.6.1.1 (Figure 1) with 67 specimens clearly belonging to a species (in other words excluding specimens classified as Clivia 'aff.' specimens) in the data set resulting in 17 haplotypes.

Identification of Clivia species
Clivia mirabilis is the only species from the Western Cape Province, with several unique characteristics distinguishing it from the other species. Their seeds mature in 4-6 months compared to the 12-24 months of the other species. The root system of C. mirabilis is much thicker and more succulent than the other species. Their pedicels turn from red or orange during the flowering stage to green when bearing fruit. Clivia mirabilis is a sought-after plant, and seedlings of other species can be sold to inexperienced buyers under the name of C. mirabilis.   The flowers of C. nobilis are similar to C. mirabilis, but are mainly distinguished by their compact umbel and the pedicels are shorter than in C. mirabilis.
The nearest neighbours and barcoding gaps from BOLD are presented in Table 4. Clivia nobilis and C. mirabilis have distances greater than zero to their NNs. The remaining Clivia species do not have a barcoding gap or any significant genetic distances between them and their closest relatives.
Clivia nobilis and C. mirabilis each have a single haplotype and demonstrated no gene flow with any other Clivia species. The haplotype network supports the monophyletic grouping of C. nobilis and C. mirabilis with five mutations separating each of these haplotypes from a common ancestor. Further support for separation is provided by a posterior probability of 1 for C. nobilis and 0.958 for C. mirabilis ( Figure 2).
Clivia caulescens is distributed as isolated populations in the northern Limpopo and Mpumalanga provinces. The southern C. caulescens populations on Bearded Man Mountain are sympatric with and hybridise freely with   C. miniata resulting in the natural hybrid, C. ×nimbicola (Swanevelder et al. 2006). Clivia caulescens was divided into three haplotypes (Figure 1). The first haplotype is unique to Mariepskop Forest Reserve, but is not represented by all the plants from this population (three specimens from Mariepskop share their haplotype with the second haplogroup). The second haplogroup is widespread from the most northern (Wolkberg Mountains) to the most southern locality (Swaziland). The third haplogroup is mainly limited to the most southern distribution area (Bearded Man Mountains and Swaziland), but a single specimen from the central distribution area (God's Window) shares this haplotype. Clivia caulescens has three to five mutations corresponding to three haplogroups.
Phylogenetic results from the combined matK and rbcLa BI cladogram are presented in Figures 2 and 3. Three species (C. mirabilis, C. nobilis and C. caulescens) can be clearly identified based on monophyletic groupings in the BI analysis. The other species (C. miniata, C. gardenii, C. robusta and the affinis specimens) are paraphyletic. These results imply there is very little or no correlation between the phylogenetic cladogram and geographical distribution in Clivia (Figure 3).
Clivia miniata has a widespread distribution covering areas over three provinces and two countries, Eastern Cape, KwaZulu-Natal and Mpumalanga in South Africa and Swaziland. Clivia miniata has eight haplogroups, of which five are unique to the species. Clivia mirabilis is the only species in the western part of South Africa on the border between the Western and Northern Cape and has a unique haplogroup.
Clivia gardenii is a pendulous species from KwaZulu-Natal, of which the most southern distribution is Durban (Swanevelder & Fisher 2009). Felbert (2003) indicates that the distribution is as far south as Port Edward near the border between KwaZulu-Natal and the Eastern Cape. Clivia gardenii has six different haplogroups, of which three are unique to C. gardenii. The other three remaining haplogroups are shared with C. aff. miniata, C. aff. robusta, C. miniata and C. robusta (haplogroup 1); C. robusta (haplogroup 2) and C. miniata (haplogroup 3).
Clivia robusta and C. nobilis are species with pendulous flowers from the Eastern Cape Province and are separated from an area of the species C. miniata known for its uprightstanding flowers. Clivia nobilis has a unique haplogroup, whereas C. robusta has two haplogroups that are not unique to the species.
A rare yellow flowering form has been observed in nature in all six Clivia species. Watson (1899)  After the discovery and recognition of the yellow C. miniata var. citrina Watson, it was predicted in 1899 that more atypical colour variations (excluding the more common orange and yellow varieties observed in nurseries) will be found beyond the borders of KwaZulu-Natal (Watson 1899

Relationships in Clivia
Three species are molecularly distinct in the BI cladograms and network based on the combined matK and rbcLa data set. These are C. mirabilis, C. nobilis and C. caulescens. This finding supports previous findings of clear lineage formation involving these three species (Spies, Grobler & Spies 2011). An unknown sample falling in the same monophyletic groups as any of these species in a BI cladogram has a posterior probability of 1.0 of belonging to C. mirabilis or C. nobilis and a posterior probability of 0.997 of belonging to C. caulescens.
Although C. caulescens has three haplotypes in the network, the number of steps (3-5) to the separation of these haplogroups into separate species is small and gives therefore  weak support for the monophyly of these populations (haplogroups) as separate species. Furthermore, analyses relying on the intra-and interspecific genetic distances such as those performed by BOLD, ABGD and SpeciesIdentifier fail to recognise C. caulescens as monophyletic. This is mainly because of the degree of intraspecific variation (mean of 0.12).
Clivia mirabilis and C. nobilis are each geographically isolated with very little or no gene flow to other species. Clivia mirabilis grows in a few small isolated populations in the Northern and adjacent Western Cape provinces and has probably speciated because of isolation by distance. The geographic separation between C. mirabilis in the west and the other species in the east was primarily caused by climate change during the Cenozoic period leading to the extinction of tropical flora (Conrad, Reeves & Rourke 2003;Linder, Meadows & Cowling 1992;Meerow & Clayton 2004;Swanevelder & Fisher 2009). This is the only species in the Northern and Western Cape provinces; the geographic distance to the closest Clivia species (C. nobilis) is approximately 650 km east-southeast (ESE) (by direct measurement using Google Earth Pro). The isolation of C. mirabilis resulted in a lack of intraspecific variation. The nearest neighbours are C. robusta in matK and C. gardenii in rbcLa with genetic distances of 0.37 and 0.56, respectively, separating these species from C. mirabilis (Table 4). Clivia gardenii has the largest degree of intraspecific variation with a mean distance of 0.36 (and maximum 0.7) in matK, followed by C. miniata (mean distance of 0.24 and maximum distance of 0.62). Clivia caulescens and C. robusta both have a mean intraspecific distance of 0.12 (maximum of 0.25) in matK. The intraspecific distances in rbcLa are 0-0.19 (Table 4).
The geographic range of C. nobilis covers a 500 km stretch ranging from slightly north of Port Elizabeth to Coffee Bay in the Eastern Cape. Clivia nobilis is adapted to tolerate the salty, sandy and intense light conditions adjacent to the coastal areas of the Eastern Cape (Swanevelder & Fisher 2009). There are a few rare localities where C. nobilis and C. miniata overlap; the gene flow in these areas results in mutating colonies, containing morphologically unusual plants (Haselau 2010 Clivia miniata has the widest distribution of all Clivia species studied, occurring in patches from the most northerly point of the C. nobilis distribution area to the most southerly end of the C. caulescens range. Mutating colonies occur where C. miniata co-exists with other species (Haselau 2010). The first of these populations is located in the Songimvelo Nature Reserve on the Bearded Man Mountain in Mpumalanga. This is where the only described semi-pendulous hybrid, C. ×nimbicola, exists as a result of crosses between C. miniata and C. caulescens. The Ngome forest contains large numbers of yellow-flowered C. gardenii plants and is where the type specimen for C. gardenii var. citrina was originally collected (Swanevelder et al. 2005). These plants have a unique round thickening at the base of the stem, which is not a typical morphological trait of C. gardenii. Based on a distribution map of Felbert (2003), another overlapping area between C. miniata and C. gardenii is a 4946 km 2 area with a 630 km stretch from Port Edward to Eshowe and 73 km inland to Pietermaritzburg. The final overlapping distribution between these two species is a small area around Port St. Johns in the Eastern Cape (Felbert 2003).
Clivia miniata shares only a small fraction of a geographical area with C. nobilis, which is a 65 km stretch between two major rivers, the Great Kei River and the Mbashe River (measured in Google Earth Pro, according to the Clivia distribution map of Felbert [2003]).
A Clivia population in the area of the Mzamba River in the Eastern Cape grows in a dense and rocky habitat in a thicket biome. The leaves and flowers of these plants are unusual in that the leaves are greyish and hard and stems are present. The last is characteristic of C. caulescens in the most northern distribution range of Clivia. The flowers are upright, drooping or spider-like and have various colours such as pastel, orange or deep red (Forbes-Harding 2008). For the purpose of this article, these specimens are referred to as C. aff. miniata.
Clivia gardenii and C. robusta (previously known as swamp gardenii) are distinguished from each other by differences in their geographical range and morphology. Clivia gardenii grows in well-drained soil. The distribution is mainly in central and north-eastern parts of KwaZulu-Natal (north of Durban) (Swanevelder & Fisher 2009), and as far south as Port Edward (Felbert 2003). The distribution of C. robusta is confined to a small area in the Pondoland centre of endemism from Port Edward as the northern border to the most southerly distribution around Lusikisiki (Dixon 2005). This species prefers swampy conditions. The C. robusta plants in the north of the distribution area resemble C. gardenii. In the southern distribution area, the plants have a greater resemblance to C. nobilis (Dixon 2005). The morphological differences between C. robusta and C. gardenii are that C. robusta is, as its name suggests, more robust with a prominent stem, it has more flowers (15-40) compared to the 14-20 of C. gardenii and the leaves differ slightly (Swanevelder & Fisher 2009 From the network ( Figure 1) and the BI cladogram (Figure 3), there is no clear differentiation between the species in the Eastern Cape (excluding C. nobilis) and KwaZulu-Natal. The species C. miniata, C. gardenii, C. robusta and the C. aff. miniata and C. aff. robusta specimens included in this study hybridise readily in cultivation and in nature. The lack of isolating mechanisms, such as isolation by distance or by geographical features, contributes to incomplete lineage sorting.
This study supports the findings of Ran, Hammett and Murray (2001a) that C. miniata, C. gardenii and C. robusta are closely related and we suggest that these species (including the yellow variants) need taxonomic revision. We also suggest that the term C. gardenii complex be used as a collective term for C. gardenii, C. robusta, C. aff. robusta and C. aff. miniata. These species contain many morphologic and genetic variations, and it may be difficult to delimit its members based on morphology and molecular data. The key reasons for this taxonomic overlap between species may lie in their geographic distributions or close proximity of different species, leading to hybridisation (ancient or recent) or other factors like adaptation to habitat conditions, for example, swampy areas leading to stem formation.

Conclusions
This article is the first report of the use of matK and rbcLa DNA barcodes to support the motivation for a taxonomic revision of Clivia.
Clivia mirabilis and C. nobilis have DNA barcodes that clearly distinguish them from the rest of the species. The two-locus barcode of matK and rbcLa can be used to identify both of these species.
Clivia caulescens consists of three haplogroups. The phylogenetic support for the monophyletic grouping is strong. The low number of mutations (3-5) separating C. caulescens from the C. gardenii complex is an indication that this species is closely related to the complex.
The C. gardenii complex consists of the species in the Eastern Cape (excluding C. nobilis) and KwaZulu-Natal. Although these species are distributed over a large geographic area, ancient and current gene flow results in morphologic and genetic overlap amongst them. This causes confusion with the identification and classification of many specimens.
This study not only provides us with DNA barcodes for C. mirabilis, C. nobilis and C. caulescens but also proves that DNA barcodes have insufficient discriminatory power to distinguish between C. miniata, C. gardenii and C. robusta as they are currently circumscribed morphologically. The latter three species have incomplete lineage sorting, and a taxonomic revision of these species is suggested.