Morphological and biochemical genetic evidence for hybridization in the genus CenteUa (Apiaceae), with notes on phylogenetic and taxonomic implica tions

The main aim of this paper is to explore the occurrence of hybridization in the genus Centella. Morphological as well as genetic characters are investigated to confirm the identity of a putative hybrid between C. triloba and C. macrocarpa. These two independent data sets, one from enzyme electrophoresis and one from morphology, are compared and interpreted by means of cladistic analysis. Centella glauca and C. virgata were included in the analysis and the effect of hybridization on cladistics is demonstrated. Hybridization gives a new perspective on infrageneric relationships within the genus Centella, as it may have obscured discontinuities between previously discrete infrageneric groups.


INTRODUCTION
The subject of hybrids and hybridization has been cov ered extensively by a number of authors (e.g. Stace 1989; Nason et al. 1992), and the influence of hybridization on cladistics has also previously been the subject of detailed analyses (McDade 1990(McDade , 1992(McDade , 1995. These authors in dicated that hybridization significantly affects infrageneric classification because the limits between taxa may become obscured by reticulate evolution. The following facts pointed to hybridization in the genus Centella: 1, lack of correlation between characters of species; 2, reported oc currence of putative hybrids in the herbarium record; and 3, the discovery of a putative hybrid initially thought to be a new species. Apart from the two populations of parent species (C. triloba and C. macrocarpa) and their putative hybrid, one population each of C. virgata and C. glauca (and an additional population of C. macrocarpa) were in cluded in a cladistic study of morphological characters and enzyme data. A second C. macrocarpa population was included to compare relationships at the population and species level. Centella virgata and C. glauca were chosen because of their obvious close relationship to C. macro carpa. Centella macrocarpa from the Swartberg Pass and C. virgata are reseeders, i.e. plants that are killed by fire and which can only regenerate from seed after fire, whereas all the other populations and species are resprout ers, i.e. plants that survive fire by coppicing, and have a slow rate of seed germination. As a result, only a relatively small number of seedlings are added to the populations after each fire. The inclusion of reseeders and resprouters gives the results broader applications in terms of the ef fects of fire-survival strategy on genetic variation, and also on the circumscription of species in the C. virgata group. Acid phosphatase ACP-1* ACP-2 3.1.3.2 Histidine-citrate (Kephart 1990) Morpholine-citrate (Clayton & Tretiak 1972) 6.5 6.1 Cytosol aminopeptidase CAP-1 3.4.11.1 Tris-EDTA-borate (Goncharenko e ta l. 1992

Data analysis
We used DISPAN (Ota 1993) for phylogenetic analysis of allozyme data by using neighbour-joining and bootstrap methods (1000 replications) and N ei's (1978) genetic dis tance values. The analysis of allozyme data was executed using BIOSYS-1 (Swofford & Selander 1981). The mor phological data as well as the allele frequencies in the different populations were polarized using C. triloba as outgroup. It is probably the least derived of all the species included in the study, judged by the shrubby habit and the broad, dentate, dorsiventral leaves, which we consider to be plesiomorphic within the genus. Of these, the dor siventral leaves are perhaps the most convincing plesiomorphy, as this character state occurs in related genera and in all the basal species of Centella (C. asiatica and the C. eriantha group). All character states of the outgroup were polarised as plesiomorphic. Alleles that were absent from one or more populations, while exhibiting both a low and a high frequency of occurrence in other popula tions, were treated as binary or multistate characters. Al lele frequencies without any obvious discontinuities (such as AAT-1 and ACP-2) were not included in the analysis. Only those frequencies of which the values were obvi ously low or high, with no intermediate values (low taken as less than 0.3, high taken as 0.5 or more) were polarised. Again, C. triloba was used as outgroup. The intermediate states of the GPI A and PGM-1 B alleles were present in the outgroup, so we were unable to polarize these char acters (only the one polarity is shown in Table 3). How ever, reversing this polarity had no effect, neither on the topology nor on the tree lengths or consistency indices, even when the genetic data were analysed separately (Fig  ure 3D, E). Table 2 lists the allele frequencies used to polarize the enzyme characters, as shown in Table 3. These data were analysed using HENNIG86 (Farris 1988). Five to ten individuals were studied in each population except for the hybrid where only two individuals were present at the locality sampled. Despite the small sample size on which allele frequencies for the hybrid were based, we believed that useful results could be obtained because the observed allele frequencies in the other populations were generally either very high or very low.  Characters and polarization of character states using C. triloba as out group: 1. Leaf type: broad (6-36 mm wide) = 0; narrow (3-5 mm wide) = 1; acicular (approximately 1 mm wide) = 2. 2. Number of teeth on leaf margin: 3(5-11) = 0; 1(-3) = 1; 1 = 2. 3. Presence of petiole: distinguishable from lamina = 0; not distinguish able from lamina = 1. 4. Tissue arrangement in leaf: dorsiventral = 0; isobilateral = 1. 5. Surface sculpturing in fruit: smooth = 0; ribbed = 1; prominently ribbed = 2. 6. Size of fruit: 3-4 x 3 mm = 0; 4.0-5.5 x 4.0 mm = 1. 7. Indumentum of petals: glabrous or villous = 0; glabrous = 1. 8. CAP-1 A: allele present = 0; allele absent = 1. 9. CAP-1 B: allele absent = 2; allele present at low frequencies = 1; allele present at high frequencies = 0. 10. CAP-1 C: allele absent = 0; allele present = 1. 11. GPI A: allele absent = 0; allele present at low frequencies = 1; allele present at high frequencies = 2. 12. GPI C: allele present at high frequencies = 0; allele present at low frequencies = 1; allele absent = 2. 13. PER B: allele absent = 0; allele present = 1. 14 PER C: allele present at high frequencies = 0; allele absent = 1; allele present at high frequencies = 2. 15. PGM-1 A: allele present = 0; allele absent = 1. 16. PGM-1 B: allele absent = 0 ; allele present at low frequencies = 1; allele present at high frequencies = 2. 17. PGM-1 C: allele present at high frequencies = 0; allele present at low frequencies = 1; allele absent = 2. 18. PGM-2 A: allele present at low frequencies = 0; allele present at high frequencies = 1. 19. PGM-2 B: allele absent = 1; allele present = 0.

Hybridization between C. triloba and C. macrocarpa
At Kogel Bay we discovered two morphologically in termediate plants in the transitional zone between a popu lation of C. macrocarpa and C. triloba. Since the plants

were morphologically intermediate between the only two
Centella species present at this site, we concluded that they must be hybrids. Supporting evidence for their hybrid origin is presented below. Different data sets were ana lysed to obtain a better understanding of character state distributions in the study group.

Morphology
Morphological evidence for hybridization was based on characters of the leaves, petioles and mature fruit of the parent species and the hybrid. Figure 1A-C shows that the leaves of the putative hybrid are intermediate be tween the parent species in width and in the number of marginal teeth. The leaves are laminar in C. triloba and acicular in C. macrocarpa. Based on the outgroup method, broad leaves as well as the presence of three or more teeth are polarized as plesiomorphic while narrow and acicular leaves with few or no teeth respectively are considered as the apomorphic character states of these multistate characters (Table 3) (Ota 1993). C, phylogenetic tree, excluding hybrid, constructed with DISPAN (Ota 1993). D, cladogram, including hybrid, based on genetic data. E, cladogram, excluding hybrid, based on genetic data. F, cladogram based on morphological and genetic data. Characters and polarization of characters used in construction of cladograms listed in Table 3. Solid square, apomorphy without homoplasy; open square, apomorphy with reversal higher up; =, convergence; x, reversal.  Figure 1F1). Furthermore, when male and bisex ual flowers are present, the inflorescence will bear only a single functionally female umbellule ( Figure 1F2). The male umbellules have three to five male flowers with glabrous petals. The hybrid is andromono-ecious ( Figure  IE), but only a limited number of specimens is available. The inflorescence is otherwise compar-able to that of C. macrocarpa with villous male flowers (Figures 1F1, 1F2). In Table 3  The morphological data analysis resulted in a partially resolved cladogram with a length of 11 steps and a con sistency index of 0.90 ( Figure 3A). The cladogram is not fully resolved as a polytomy occurs between the two C. macrocarpa populations and C. glauca.

Allozyme data
Genetic variation within species was observed at seven of the 15 enzyme-coding loci. The choice of five of these loci, i.e., CAP-1, GPI, PER, PGM-1 and PGM-2, for further data analysis was confirmed by their relatively high fixation index (F-statistic) values (Table 4, the values are explained in the caption). The alleles that contributed most to population differences at these loci are represented by characters 8 to 19 in  above 0.500). Figure 3B shows the dendrogram produced with DISPAN and the complete set of allozyme data whereas Figure 3C shows the dendrogram without the hy brid. Four cladograms were produced from the characters (8-19) in Table 3, of which the consensus tree is shown in Figure 3D. The length of the consensus tree is 23 steps with a consistency index of 0.73. When the hybrid was excluded from the data set, the resulting cladogram was fully resolved, with a length of 21 steps and a consistency index of 0.80 ( Figure 3E). According to McDade (1995), a higher consistency value is to be expected, since the removal of a hybrid from the analysis should lead to a significant reduction in homoplasy. Table 5 lists the genetic distances between the species and populations studied. From this information it is evi dent that the shortest genetic distance was found between the hybrid and the C. macrocarpa population from Kogel Bay. The genetic distance between the two C. macrocarpa populations is relatively small (0.091) when compared to the average distance between species (0.218). The species with the shortest genetic distance are C. macrocarpa and C. triloba (0.108) and the species with the longest distance between them are C. macrocarpa and C. virgata (0.350).
The shortest genetic distances are those between C. triloba, the hybrid, and the two C. macrocarpa popula tions. This is what one would expect, given the proposed hybrid origin. Centella glauca and C. virgata are only linked into this group at a much greater genetic distance.   1.13, 1.13 respectively).
When the morphological and genetic data sets were combined, the result was a fully resolved cladogram with a length of 40 steps and a consistency index of 0.70 (Fig  ure 3F). When the hybrid was removed from this data set, the tree length reduced to 32 steps and the consistency index improved to 0.87. This reduction in homoplasy is again consistent with our assumption of hybridization, fol lowing McDade (1995).

Geographical distribution o f the putative hybrid and the parent species
Both C. triloba and C. macrocarpa occur in the West ern Cape but C. triloba is limited to the Cape Peninsula and coastal areas of the Caledon District, whereas C. macrocarpa is more widespread in the region. The hybrid is very localized in occurrence. Only two plants were found at Kogel Bay and both parent species occur at the same locality (Figure 4). The C. triloba population occurs on a relatively flat area closer to the sea than the C. macro carpa population, which grows on a steep scree slope.
The putative hybrids occurred between the two popula tions at the foot of the scree slope.

DISCUSSION
Centella triloba, the hybrid and C. virgata formed mor phologically discrete clades with only one overlapping morphological character between C. virgata and the hy brid, namely the presence of ribbed fruit. Centella glauca and C. macrocarpa are mainly distinguished by their habit differences and the distinctly glaucous leaves of the for mer species (Schubert & Van Wyk 1996). A polytomy results as these autapomorphic characters were not used in the cladogram ( Figure 3A). The morphological data showed that the hybrid is intermediate between C. triloba and the rest of the study group. Characters, such as the presence of isobilateral leaves in the hybrid, group it with C. macrocarpa, whereas the similarities with C. triloba all appear to be symplesiomorphies. McDade (1990) also found that hybrids do not lead to unresolved cladograms with high levels of homoplasy, but that they emerge as   Figure 3D. This is again in agreement with the propos als of McDade (1995), who suggested that the hybrid would be placed near to the parent that has the most de rived characters. The combined morphological and genetic data also present supporting evidence for the hypothesized hybrid origin. When the hybrid was omitted, a fully re solved topology resulted, with a substantial improvement in both the tree length and consistency index ( Figure 3E). The morphology gives only a partially resolved cladogram ( Figure 3A) but the C. glaucaJC. macrocarpa polytomy becomes fully resolved in the combined analysis, with the genetic data responsible for the improvement. These analyses are consistent with two generalizations: 1, re moval of the hybrid improves the resolution and 2, both data sets (morphological and genetic) contribute to resolv ing different parts of the topology, as is evident in Figure   3A & F. C. virgata is clearly separated from C. macro carpa in all the analyses, which gives useful support for considering it as a distinct species despite the close simi larity to some forms of C. macrocarpa (particularly the reseeding forms). Fire-survival appears to be a homoplasious character which has evolved convergently in different species.
The C. macrocarpa populations were grouped together on the final cladogram ( Figure 3F). This is interesting since the Swartberg population is a reseeder and the Kogel Bay population is a resprouter, and the genetic distance between them (Table 5)  There is a possible correlation in the genetic variation of the reseeder populations and also of the resprouter populations, i.e. that resprouters are genetically more vari able than the reseeder populations. The latter are more prone to genetic bottlenecks, as was proposed by Schutte et al. (1995), because presumably only one generation is present at any given time, while the survival of the resprouters leads to a mixture of numerous generations with in the same population. Further studies, including larger sample sizes, are needed to confirm the relationship be tween genetic variation and fire-survival strategy in Cen tella.
The demonstration that hybridization occurs between two species of Centella that are not sister taxa indicates that a strictly hierarchical infrageneric classification may be an unobtainable goal. Even the cladistic method may become problematic, particularly when nodes are weakly supported [see McDade (1992McDade ( , 1995 for a detailed dis cussion of the impact of hybrids on cladistic analysis]. The assumption of divergent evolution needed for the cladistic method may thus not be an accurate reflection of the actual mechanisms of evolution within the genus. If evolution is indeed reticulate, then the identification of lineages of ancient hybrid origin becomes problematic (McDade 1990). Nevertheless, the use of different data sets in determining phylogenetic relationships between Centella species, resulted in a closer proximation to the true phylogeny than would have been possible with either of the data sets (Shakalee & Whitt 1981;McDade 1995).
The new insight into evolutionary relationships in Cen tella provides a possible explanation for the reticulate pat tern of character expression in the genus. Adamson (1951) was unable to create mutually exclusive infrageneric taxa and coped with the problem by including some species in more than one series. Roux et al. (1978), Winter & Van Wyk (1995) and Allison (1995) have used cladistic methods to study Apiaceae, marking the beginning of a rigorous, empirical approach to classification in this fam ily. However, if a hybrid disrupts the pattern of character expression by inheriting the defining apomorphies of un related parents from different sections or series, then it will not be possible to use cladistics in the normal way because the method relies on divergence. McDade (1995) highlighted several new attempts at resolving this prob lem.