Combretaceae is a medium-sized family within Myrtales, comprising about 500 species in 17 to 23 genera. It has long been referred to as a complex phylogenetic and taxonomic group (Tan et al. 2002, Maurin et al. 2010, Stace 2010, Jordaan et al. 2011). Based on morphological characters and phylogenetic analysis, the family Combretaceae has been recovered as monophyletic and sister to the rest of Myrtales (Brown 1810, Dahlgren and Thorne 1984, Tan et al. 2002, Sytsma et al. 2004, Maurin et al. 2010, Stace 2010). Members of Combretaceae are mainly trees, shrubs or lianas, occupying a wide range of habitats from savannas, forests, to woodlands (Maurin et al. 2010) and are distributed in tropical and subtropical regions across the globe. With ca. 350 species, Combretum Loefl., the largest genus in the family has its centre of diversity in Africa, with approximately 63 species described in southern Africa – south of the Zambezi river and includes South Africa, Zimbabwe, Namibia, Botswana, Lesotho, Swaziland, and Mozambique (Maurin et al. 2010, Jordaan et al. 2011).

The major distinguishing feature of the family is the presence of unicellular combretaceous hairs on the abaxial leaf surfaces, a diagnostic trait in many other species of Myrtales and even beyond the group e.g. the Cistaceae Juss. family, tribe Cisteae (Maurin et al. 2010, Stace 2010). However, other morphological features such as presence of trichomes, stalked glands, domatia, inflorescence, fruit shape, leaf and pollen morphology are also important for species delimitation in Combretaceae (Exell and Stace 1966, Stace 2007, 2010, Maurin et al. 2010, Jordaan et al. 2011). Nonetheless, all these characters are not adequate enough to delimit species within the family because none is unique to a specific clade. As a result, the family has experienced several splitting and lumping in the past (El Ghazlai et al. 1998, Tan et al. 2002, Maurin et al. 2010, Stace 2010, Jordaan et al. 2011). Also, the taxonomy is further confounded by the high morphological similarity between members of different sections. For instance, inflorescence and fruit shapes are very similar between species and across clades (Figures 1 and 2). Such homoplasious morphological similarities have also been identified as the root of difficulties in delimiting the genera; for example in the Combretum-Quisqualis clade (Jordaan et al. 2011). Consequently, it becomes necessary to search for an alternative method to augment traditional morphology-based taxonomy of Combretaceae.

Here, we propose that DNA barcoding may provide such a complementary tool to ease species delimitation within the group. DNA barcoding involves the use of a short and standardised DNA sequence that can help assign, even biological specimens devoid of diagnostic features, to species (Hebert et al. 2004, 2010, Hajibabaei et al. 2006, Roy et al. 2010, Van der Bank et al. 2012, Franzini et al. 2013). Two DNA regions defined as ‘core barcodes’, i.e. rbcLa and matK have been standardised as DNA barcodes for land plants (CBOL Plant Working Group 2009). In addition to the core barcodes, two other regions, trnH-psbA and nrITS were suggested as supplementary DNA barcodes for plants (Hollingsworth et al. 2011, Li et al. 2011). The rationale for adopting these two regions (rbcLa and matK) is high levels of recoverability of high-quality sequences and acceptable levels of species discrimination (Burgess et al. 2011). The discriminatory power of the core DNA barcodes for land plants was estimated at 70–80% (CBOL Plant Working Group 2009, Fazekas et al. 2009, Kress and Erickson 2007). However, a recent study suggests that efficacy of core barcodes may have been overestimated, arguing that taxon sampling has been biased towards less-related species (Clement and Donoghue 2012). Furthermore, barcoding efficacy is rarely evaluated in a phylogenetic context (but see Clement and Donoghue 2012), resulting in potentially biased estimates of discriminatory power.

In this study, we evaluated the efficacy of DNA barcoding as a tool to augment morphological species discrimination within Combretaceae. Specifically, we (1) assessed the potential of four markers to discriminate southern African species of the family, and (2) assessed the efficacy of barcodes across major clades including subgenera and sections within the largest genus Combretum.

Sampling includes one to six accessions of 58 species out of the 63 species representing the six genera of Combretaceae in southern Africa. These genera include Combretum (43 species included in this study), Lumnitzeria Wild. (one species included), Meiostemon Exell and Stace (one species included), and Quisqualis L. (one species included), Pteleopsis Engl. (two species included), and Terminalia (ninespecies included).

Collection details, taxonomy, voucher numbers, GPS coordinates, field pictures, and sequence data (only matK and rbcLa) are archived online on the BOLD system (www.boldsystems.org). Voucher information, name of herbarium, GenBank and BOLD accession numbers are listed in Appendix 1.

The overall characteristics of single and combined DNA regions are presented in Table 1. In general, our results indicate that the ranges and mean intraspecific distances were both lower than those of interspecific distances. Among single regions, rbcLa showed the lowest interspecific distance (mean = 0.009) with nrITS exhibiting the highest genetic variation between species (mean = 0.110). For all marker combinations, the mean interspecific distances varied between 0.011 and 0.014. Assessing the index of sequence divergence K for each region, we found that nrITS showed the highest divergence (K = 21) whereas trnH-psbA exhibited the lowest divergence (K = 3). For the combined regions, K varied between 10 and 13, with an average of 10 substitutions between sequence-pairs (Table 1).

The distribution ranges of inter- versus intraspecific distances for all regions, showed a clear overlap between both distances (Figures 3a, b and 4), indicating the existence of a barcode gap. Comparing the smaller inter- versus the largest intraspecific distances for each region, our results further support the existence of barcode gap in all regions, but the proportion of sequences with barcode gap varied significantly with the regions tested (Table 2). Notably, the combination of all four regions exhibited the highest proportion of sequences with barcode gap (84%) followed by nrITS (73%), then core + nrITS (64%), and core + trnH-psbA (57%), with the lowest proportion found in rbcLa (13%) (Table 2).

Optimised genetic distances used as threshold for species delimitation in Best Close Match method are shown in Table 1. Apart from rbcLa (threshold = 0.04%), core + trnH-psbA (threshold = 0.5%) and core + nrITS (threshold = 0.7%), the thresholds for the remaining single and gene combinations were greater than 1%.

Our results for the discriminatory power analysis varied with the methods applied (Table 3) at family level. Based on the Near Neighbour method, nrITS provided the highest discriminatory power (65%) followed by rbcLa + matK + trnH-psbA + nrITS (64%), rbcLa + matK + trnH-psbA (62%), and rbcLa + matK (61%). The lowest discriminatory power was found for trnH-psbA (28%).

BOLD species delimitation criteria of 1% threshold provided the lowest rate of correct identification among all three methods used. However, we found that nrITS remains the most efficient region with 47% discriminatory power. The second most successful combination of regions were core + trnH-psbA + nrITS (41%) followed by core + nrITS (30%) and trnH-psbA (22%); the core barcodes were identified as the least performing regions (10%) with the highest proportion of ambiguity (86%).

In contrast to the two previous methods, the Best Close Match provided the highest rate of species discrimination for the combined dataset (core + trnH-psbA + nrITS) yielding the best discriminatory power (87%) with no ambiguity. This was followed by core + trnH-psbA (80%), core + nrITS (70%) and nrITS (63%), with the poorest performance for rbcLa (18%) at family level.

The last criterion used to evaluate the potential of DNA region was PCR efficiency. We found that rbcLa (87%) followed by trnH-psbA (85%) and matK (68%) were easy to amplify, with nrITS being the most difficult (47%; Figure 5).

We complemented previous analyses using species monophyly criteria after verifying the monophyly of Combretaceae. Among all regions, core + trnH-psbA isolated the highest proportion of monophyletic species (83%), followed by trnH-psbA (78%), nrITS (76%), and combination of all four regions (65%). Again, rbcLa provided the lowest performance in identifying species as monophyletic (37%; Figure 6).

In summary, all regions provided evidence for barcode gaps (Figure 3a, b and 4), but the strength of evidence varied with approaches used. Furthermore, the Best Close Match method provided the highest identification accuracy among the three distance-based methods used irrespective of genes or combinations tested. Under this method, the two best potential barcodes for southern African Combretaceae were first, core+ trnH-psbA and second, core + trnH-psbA + nrITS. However, based on species monophyly criteria, the single region trnH-psbA and the combination core + trnH-psbA showed high barcode potential, with trnH-psbA being the second best easy-to-amplify region after rbcLa.

We further evaluated the potential of each region as candidate barcode using a phylogeny of southern African Combretaceae (Appendix 2). Our results are congruent to the corresponding subset in the most recent and largest phylogeny assembled for the family (Appendix 3). Our evaluation for the discriminatory power at subgeneric level using the thresholds determined for the family (1.31% for the core and 0.5% for the core + trnH-psbA) revealed that the core barcodes alone were able to correctly identify 78% of species within the subgenus Cacoucia. However, the core barcodes could discriminate only 50% of species within the subgenus Combretum. In particular, the discriminatory power of the core barcodes within both subgenera increased markedly to 100% when we added the trnH-psbA region (Table 4). This trend was consistent even when we applied the thresholds that have been optimised for the subgenera.

At sectional level, we observed similar trends – the addition of trnH-psbA increased the performances of the core barcodes drastically except for Macrostigmatea (Table 5): Angustimarginata (core: 11%; core + trnH-psbA: 86%); Ciliatipetala (core: 55%; core + trnH-psbA: 73%); Conniventia (core: 38%; core + trnH-psbA: 88%); Hypocrateropsis (core: 63%; core + trnH-psbA: 80%). However, Macrostigmatea (core 34%, core + trnH-psbA 44%) showed the least performance, even with the addition of trnH-psbA to the core barcode, with just 10% increment being observed. This trend is not sensitive to the thresholds applied for the family or the sections.

Finally, we compared the mean number of substitutions between any two species within each section. We found that the mean number of substitutions between representatives of Macrostigmatea is lowest (mean = 4) whereas it ranges between 5 and 19 substitutions in other sections of subgenus Combretum.

We evaluated genetic variation for both single and various combinations of rbcLa, matK, trnH-psbA and nrITS. Comparing ranges of intra- versus interspecific distances, our results indicate that all markers show a barcode gap (Meyer and Paulay 2005); and this is also true for the stringent Meier et al.’s (2008) approach, although the proportion of sequences with gap varies greatly with the marker used.

The discriminatory power of the DNA regions in species identification also varies with the distance-based methods applied. From the methods tested, Near Neighbour and Best Close Match yielded high performance, with the latter giving the best results for the possible three and four different gene combinations. The core barcodes were not recognised among the three best options, and its discriminatory power has been questioned in a number of studies (Hollingsworth et al. 2009, Pettengill and Neel 2010, Roy et al. 2010, Wang et al. 2010, Clement and Donoghue 2012). Based on all three distance methods, nrITS emerges as the most suitable single region (as indicated under both Near Neighbour and BOLD; see also Kress et al. 2005, Kress and Erickson 2007, Chen et al. 2010, Gao et al. 2010, Ren et al. 2010, China Plant BOL Group et al. 2011, Muellner et al. 2011, Pang et al. 2011, Wang et al. 2011, Liu et al. 2012, Yang et al. 2012). Among combined regions, core + nrITS + trnH-psbA (under Best Close Match) emerges as most suitable for barcoding Combretaceae.

However, our study indicates some important drawbacks that discount the inclusion of nrITS as a good barcode. For example, based on amplification success criteria, nrITS was the most difficult of all regions tested with rbcLa and trnH-psbA being the easiest regions to amplify. The technical hurdles in PCR amplification and sequencing of nrITS may be linked to the presence of retro-transposons and other repetitive elements within plant nuclear genomes, resulting in paralogous gene copies (Gao et al. 2010, Hollingsworth 2011, Hollingsworth et al. 2011, Li et al. 2011). This is likely the case for nrITS in Combretaceae as we found evidence of multiple copies that may not be identical to each other (see CBOL Plant Working Group 2009, Hollingsworth 2011, Hollingsworth et al. 2011, Yang and Berry 2011). As such, the addition of trnH-psbA to the core barcodes (rbcLa + matK + trnH-psbA) emerge as the best gene combination useful for species discovery and delimitation in Combretaceae (see also Newmaster and Ragupathy 2009, Petit and Excoffier 2009, Ragupathy et al. 2009, Wang et al. 2009, Arca et al. 2012).

Previous studies have shown that core barcodes are very limited in discriminating taxa that are phylogenetically closely related, and suggested that the efficacy of DNA barcodes should be tested within a phylogenetic context (Clement and Donoghue 2012). We tested this using subgenera and sections of the family Combretaceae. Our evaluation of the discriminatory power of the core barcodes at subgeneric level revealed a striking difference in the performance between the two Combretum subgenera, Combretum and Cacoucia. The difference noted for the discriminatory power of the core barcodes between the two subgenera may reflect differences in their evolutionary history. Indeed, the latest dated phylogeny of Combretaceae indicated that members of the subgenus Cacoucia are represented with longer terminal branches than those in subgenus Combretum (Maurin 2009).

While we found poor performance at sectional level, for example, in Angustimarginata, Macrostigmatea and Conniventia, this result is not unexpected due to a very low genetic variation one could expect within clades (see Ennos et al. 2005, Clement and Donoghue 2012). However, the addition of trnH-psbA to the core barcodes results in a drastic increase of identification rate at both subgenus and sectional levels, validating the utility of trnH-psbA to discriminate even closely related species, except for section Macrostigmatea (Newmaster and Ragupathy 2009, Petit and Excoffier 2009, Ragupathy et al. 2009, Wang et al. 2009, but see Zhang et al. 2012, Arca et al. 2012, Clement and Donoghue 2012).

The result for section Macrostigmatea reflects earlier tangle cited in previous studies regarding its composition (Stace 1980, Maurin et al. 2010, Jordaan et al. 2011). In our analysis, we included Spathulipetala within section Macrostigmatea based on suggestions from recent molecular evidence (Maurin et al. 2010). Morphological studies separate these two sections, Spathulipetala and Macrostigmatea (Stace 1980, Jordaan et al. 2011). Section Spathulipetala comprises two members, Combretum zeyheri Sond. and Combretum mkuzense J.D.Carr and Retief, which occur in the same geographical location and show close morphological similarity in their fruits (Jordaan et al. 2011). The inclusion of Combretum mkuzense, in this section has been controversial, with some authors (Exell 1978, Stace 1980) advocating for a tentative placement pending further investigation. However, recent molecular study shows close relationship between these two species (Combretum zeyheri and Combretum mkuzense) (Maurin et al. 2010), which gives support to earlier morphological treatment. On the other hand, the taxonomy of section Macrostigmatea appears to pose fewer challenges as compared to Spathulipetala. A recent molecular study (Maurin et al. 2010) suggests lumping of these two sections, Spathulipetala and Macrostigmatea as members appear embedded in one clade with a high bootstrap support of 100%. Earlier, Exell (1978) had reported that the sections are closely related, as they share similarities in scale size, scale fragmentation into fruit walls and fruit size.

Based on our results, the unclear taxonomy reported for section Macrostigmatea, is reflected, indicating a need for further molecular analyses involving more taxa and gene sequences to correctly determinemembers of this section. Our results also support the proposal of Exell (1978) to lump these two sections. The low performance of the core + trnH-psbA in fully discriminating the different species within this section is a strong indicator of the close phylogenetic similarity of the species. Our results indicate not only the utility of DNA barcoding data for discriminating species, but also to detect species that require further molecular analyses.

Our analysis indicates that the poor performance of the core barcodes at family level could not be generalised to lower levels: the core barcodes perform poorly in some sections but shows strong discriminatory power in others. Such findings may indicate that the success of DNA barcodes in discriminating closely related species at least in plants may correlate with the evolutionary distinctiveness of the group tested and, as recently indicated, (see Clement and Donoghue 2012) it may also possibly reflects different biogeographic history between clades of the taxonomic group Combretaceae. Overall, we propose the core + trnH-psbA as the best barcode for the family Combretaceae.