A CLADISTIC ANALYSIS OF lIdhF SEQUENCES FROM REPRESENTATIVE SPECIES OF SAINTPA ULIA AND STREPTOCARPUS SUBGENERA STREPTOCARPUS AND STREPTOCARPELLA ( GESNERIACEAE )

vegetative structures among some species, and are among the few genera in the Gesneriaceac which include species with chromosome counts or 1/ = 15. The similarity of these features has indicated that the two genera are likely to be closely related. This study examines the sequences of the chloroplast gene ndhF among several representative species of SaiJJlpulIlia and representatives of Slreptocarpu.I' subgenera Strcplocarpella and StreptocarplI.l'. The results of this analysis are congruent with those of a previous analysis based on the nuclear ribosomal region, ITS. Sailltpalliia was found to be nested within Streptocarpll.l' subgenus Streprocarpella. The results raise the possibility of the taxonomic revision or these genera, and this is discussed. Comparisons between the data sets are made regarding utility of the two regions, sample size and outgroup.

Early interpretations of relationships among African genera of the Gesneriaceae indicated neither unity nor a strong relationship between Saintpaulia and Streptocarpus.However, as additional species were described and the characters of the African Gesneriaccae examined more carefully, a closer relationship among all of the African genera of Gesneriaceae, in particular Saill/pauUu and Sll'cpfocarplIs, became apparent (Hilliard & Burtt, 1971).
Both Sailltpaulia and Streptocarplls have similar pollen structure.The pollen of all African species examined was found to be eurypalynous, prolate to spheroidal to oblate, triporate or tricolpate to tricolporate, and the colpus margin is not differentiated (Weigcnd & Edwards, 1996).The similarity of the pollen more than likely reflects a monophyletic lineage of all African genera (Weigend & Edwards, 1996).Additionally, some Sll'eplocal'pus species share similar pollen types to Sailltpaulia although exact relationships between these genera cannot be made with pollen data alone (Weigend & Edwards, 1996).
A cladistic analysis of the tribes of the Gesneriaceae based on morphological data included both Sailltpaulia and S/I'eptocarpus (Smith, 1996).A strict consensus of the trees from this analysis placed both genera in a polytomy, although the majority of trees placed Slreptocarpus as the sister to Omit/wboea Parish ex c.B. Clarke on the basis of twisted carpels common to these two genera.In a similar analysis, based on comparison of IIdhF sequences, Smith et a!. (1997a) found that the single species of Sailllpaulia included was the sister to StreptoC(ll'PllS flolslit.Stl'eptocarplis saXOI'UI1l was the sister to both or these two species.Thus, Sabllpaulia was nested within
The present paper presents a cladistic analysis of the cpDNA data and makes comparisons with the results of MilUer & Cronk (1997).This paper examines a much more thorough sampling of Sailltpaulia and Stl'eplOcarpll.l'with the IIdhF gene than has been examined previously (Smith et aI., 1997a).
The source of molecular data for this analysis is the chloroplast DNA encoded gene IIdhF which in tobacco encodes a protein of 740 amino acids presumed to be a subunit of an NAOH dehydrogenase (Sugiura, 1992).The use of IIdhF sequences for systematic studies has provided a far greater number of characters to resolve relationships than studies using rbcL (Olmstead & Sweere, 1994;Clark et al., 1995;Olmstead & Reeves, 1995;Scotland et al., 1995;Smith et al., 1997a, b).In general, the l1dhF gene is sm';l larger than I'bcL and has a nucleotide substitution rate approximately two times higher (Sugiura, 1989).The larger number of variable characters makes ndhF sequences ideal [or taxonomic groups that have not been resolved well using rbcL data, such as members of the Asteridae and Lamiales s.l.(Olmstead et al., 1992(Olmstead et al., , 1993;;Chase et al., 1993).

MATERIALS AND METHODS
The gene sequences used in this analysis were generated by thermal cycle sequencing (Innis et al., 1988) of previously amplified IldhF regions.The IldhF gene was amplified in two overlapping sections (positions 1-1350, and 972-2044) from genomic DNA isolated from fresh or frozen material (Smith et al., 1992).Amplification and sequencing procedures followed that of Smith et al. (1997a) used for other members of the Gesneriaceae.The species used in the analysis, voucher information, and Genbank accession numbers are shown in Table I.Outgroups included two members of the Buddlejaceae, selected on the basis of recent molecular analyses that indicate the Buddlejaceae to be the sister to the Gesneriaceae (Smith et al., 1997b).Other members of the Old World subfamily Cyrtandroideae were included in this analysis to verify that neither Sll'eplOcll/,pUS nor Sai11lpalilia had any affinities with other genera.
Phylogenetic trees were reconstructed using PAUP (Phylogenetic Analysis Using Parsimony) version 3.1.1(Swofford, 1993) to implement parsimony (Farris, 1970;Farris et aI., 1970;Swofford & Maddison, 1987).In this analysis, trees were generated using the general heuristic option, saving minimal trees only, with the collapse zerolength branches, and ignore uninformative characters options in etfect.Because of the large number of taxa in this analysis, the branch and bound and exhaustive search options would have consumed an excessive amount of time.Therefore, the trees presented here are best approximations and not exact solutions.The manner in which the program reconstructs phylogenetic sequences is sensitive to the order of taxa presentation in the data matrix, frequently finding islands of equally parsimonious trees depending on the order (Maddison, 1991).Therefore, it is important to repeat the analysis several times, each time with a different starting place in the data matrix.To do this, the search strategy of Olmstead & Palmer (1994) was implemented searching for 1000 trees each using the 'random addition sequence' in five subsequent analyses with the nearest neighbour interchange (NNI) search option in effect and mulpars 'off'.Each of the results from the five NNI searches was used as the starting tree(s) for a search with tree bisection reconnection (TBR) and mulpars 'on'.The mulpars option, when 'on', will save all equally parsimonious trees.
Branch support analysis was performed to examine trees that were three or fewer steps longer than the most-parsimonious tree (Bremer, 1988(Bremer, , 1994;;Donoghue et aI., 1992).This type of analysis provides an indication of the robustness of the data by determining which clades persist in a consensus tree as parsimony is relaxed.This analysis was performed by saving all trees three steps longer than the mostparsimonious trees and then examining subsets of trees one to three steps longer with the filter option of PAUP.Bootstrap analysis (Felsenstein, 1985) was performed using 1000 replicates.An estimate of how well the data fit the tree was determined by calculating the gl statistic (Hillis & Huelsenbeck, 1992) using 10,000 random trees.

RESULTS
A total of 2223 base pairs of ndhF were sequenced for each species in this analysis.Of these base pairs, 408 were variable and phylogenetically infonnative among the species used in this analysis (Table I).An insertion at position 1440 of 13 base pairs was found in all species examined.The sequence of this insertion was not used in the cladistic analysis.Only two base pairs ditfer among all of the species of Sailllpaniia and Slreplocarpus included in this analysis, therefore the choice to exclude this insertion (due to possible lack of homology) is not likely to have altered the results.
The cladistic analysis of I/dhF sequences resulted in 253 trees of 2904 steps each, consistency index (CI) = 0.31, retention index (Rl) = 0.38.The strict consensus of these trees is presented in Fig. 1.Sailllpaulia and Streplocarpus fonn a monophyletic group.However, neither Sabllpaulia nor Streplocarpus are monophyletic.The Sainlpalilia species are in a poly to my with the two species of Slreptocarpus subg.Streptocarpel/a (Fig. I).There is significantly more structure than random data within this analysis as reflected by the gl statistic (-0.392).Extrapolation from the charts generated by Hillis & Huelsenbeck (1992) for the number of characters and taxa used in this analysis indicates that the gl is clearly significant at P<O.OI.
The general results of this analysis are found to be completely congruent with those of Moller & Cronk (1997) based on ITS sequences.Because the sampling between the two studies was not identical, this study does not resolve relationships among the different sections of Streptocarplls and in general provides less resolution among the included taxa than does the ITS region.

DISCUSSION
Cladistic analysis of the cpDNA gene ndhF indicates that Sail/tpalilia is closely related to Streptaearplls subg.Streptacarpel/a (Fig. I).Poor resolution of the relationship between Sail/tpalilia and Streptacarplls subg.StreptDearpel/a prevents further speculation on the exact relationship between these two taxa.For example, Sailltpalllia could be a monophyletic group that is a sister to subg.Streptacarpel/a, or be paraphyletic.Regardless, the data provide evidence that Saintpalilia is derived from within Streplocarplls and in particular shows a close affinity to subg.Slreptocarpella (Fig. I).
Further evidence for the relationship of Sail/tpalilia to StreptDearpll.\' sllbg.Streptocarpelia comes from ITS sequence data (Moller & Cronk,(997).The results from the ITS sequencing are congruent with the results of the IldhF sequencing presented here with the benefit of higher resolution among the species examined (Fig. 2; Moller & Cronk, 1997).The ITS data places Sailltl'aulia as a monophyletic group nested within subg.Streptocarpelia (Fig. 2).
Lastly, the close relationship between Sailltpau/ia and subg.Streptocarpelia is supported by data from pollen analysis (Weigend & Edwards, 1996) , 1984).Oespite different sampling strategies, the results presented here based 011 I/dhF sequence data (Fig. I) and ITS sequence data (Fig. 2; Moller & Cronk, 1997) are congruent.The major difference between the two analyses is the lower level of resolution obtained from the cpONA (Fig. I).Both the epONA and nrONA indicated that Sail/lpall/ia is closely related to Sll'eplocwjJlIS subg.Sll'eplocwjJella (Figs 1,2).The greater sampling from subg.Sll'eplocal'pella in the ITS study identifies two clades within this subgenus, one from mainland Africa and the other from Madagascar (Fig. 2 2).The greater resolution obtained from the ITS study indicates, as had been previously noted (Baldwin et aI., 1995), that this region is better suited to resolve phylogenetic questions at the species and generic level.
A taxonomic revision of StreptoClll'PUS and SailltpClulia is likely to be inevitable in order to reflect monophyletic genera, although no revision is recommended at this time.One possibility is to combine Sailltpaulia into Stl'eplocarpus subg.
Sireplocarpella; the other is to apply the name of StreplocwjJella at the generic level to one of the clades and erect a new genus for the second (Fig. 2).This latter system is likely to be preferable in order to reflect the morphological distinctiveness of the species placed in Sailltpaulia from Stl'eptocarpus, and, since Saintpaulia is the most widely cultivated member of the family, would result in less confusion among horticulturists.Furthermore, the division of StreptocarpIIs subg.StreplocaJjJllS into two sections, one the rosulate foml, the other unifoliate, receives support from the ITS data (Fig. 2; Moller & Cronk, 1997), although these sections were abandoned earlier on the basis of not being tenable by Hilliard & Burtt (1971).To more fully resolve the nomenclatural issues within Streptocarplls, additional studies will be necessary to increase sampling among the species of this large and morphologically variable genus.
The major morphological shifts from Sireptocw1JUS to Saillipalliia are the loss of the twisted fruit, the shortening of the corolla tube, and the loss of eaulescence in Saintpaulia (although some species of Sa;,ztpaulia have a creeping stern rather than being rosulate and some Sireptocarplls are rosulate), and the shift from nectar reward to pollen reward for pollinators (Hilliard & Burtt, 1971; Moller & Cronk, 1997).The loss of the twisted fruit may be the result of developmental constraints since the fruits of Sail1lpalllia are more globose in shape.The shift in fruit type is probably the result of selection pressures due to the habitat shift (most species of Saillipalliia occur on steep cliITs).The occurrence of twisted fruit, although long used to define of the Gesneriaceae along with two species of SlreplocwjJlIs (Smith et aI., 1997a) and none showed any close affinity to SlreptoC{f/jJlIs.A morphological analysis that did not include Boca or Paraboca did result in a close affinity of Streptocarplls and Ornil//Oboea but this relationship was only weakly supported (Smith, 1996).
Therefore, the occurrence of the twisted fruit is likely to have several origins within the Cyrtandroideae.
The change in corolla shape and pollen reward probably reflects a shift in pollinator from StreptocarplIs, albeit little is known regarding the natural pollinators of these plants (Hilliard & Burtt, 1971).The change in habitat, for instance, from rock outcrops to steep cliffsides would probably result in a change in the pollen vector pool, and subsequent changes in the morphology of the flowers would be expected.Clarke are likely to be a monophyletic group distinct from Asiatic taxa (Weigend & Edwards, 1996).Since this present study has confirmed previous hypotheses on the close relationship o[ Sail1lpall/ia and Streptocarplls subg.Streptocarpella, the inclusion of these additional genera in a molecular analysis may provide further support for the hypothesis generated from the pollen data.Unfortunately most of these genera are rarely collected and are currently not available for molecular analysis.
A comparison of the trees generated for these taxa based on different DNA sequences provides not only a chance to gain further confidence regarding the relationship of these genera (Donoghue, 1994), but also an opportunity to compare the utility of the two genomes.Sampling within Slreplocarpus was much greater for the ITS region (Moller & Cronk, 1997;Fig. 2) than it was for the IIdhF gene (Table I, Fig. I) and the sampling among the genera of the Cyrtandroideae was greater for the latter.Regardless, the results of the two analyses resolve the relationship of Sailllpalilia to Streptocarplls consistently, indicating that the level of sampling was adequate for both studies in order to resolve this particular question.Further resolution among the different sections of Streptocarplfs and among the species within these clades was obtained with ITS data that was not resolvable with the more conserved sequences o[ the IIdhF gene (Figs 1,2).Furthermore, the outgroups for both studies were different.Two genera [rom a related [amily (Buddlejaceae) were used for the cpDNA study (Fig. I) and two genera within the Cyrtandroideae (Haber/ea Friv.and Chirita) were used [or the ITS study (Moller & Cronk, 1997;Fig. 2).Based on a wider sampling o[ genera within the Cyrtandroideae (Smith et aI., 1997a) Chirita is not closely related to the Streptocmplis/Sailltpall/ia clade.The congruence of the two analyses regarding the relationship of Slreptocarpus and Sailllpall/ia indicates that the particular choice of outgroup was less important in resolving this question.
provided through NSF grant DEB-9317775, a grant from the American Gloxinia and Gesneriad Society, and Boise State University Faculty Research Associates Program to l.F.S.
FIG. I. Strict consenSllS of 153 trees of 2904 steps each (el = 0.31, RI = 0.38) based on cpDNA ndflF gene sequences.Numbers above clades are decay values, numbers below clades are bootstrap values.Subgenera or Streptocarplls are indicated on the figure.
Stl'eptocarplls, is not unique to this genus.Other members of the Cyrtandroideae possess a twisted fruit, such as Boea, Omilhoboea, Pomboea (C.B.Clarke) Ridley, Rhabdolhallll/opsis Hems!. and Trisepa/1II11 C.B. Clarke.Three of these genera (Boca, Paraboea and Ornitlwboea) were included in an analysis of the tribal relationships

close relationship of Saintpaulia and Streptocarplis based on the above data prompted
further investigations using the chloroplast DNA (cpDNA) encoded m/hF gene, the results of which are presented in this paper.MilUer & Cronk (1997) examined the relationship of Sailllpaulia to the subgenera and sections of Streptocarpus using the nuclear ribosomal DNA (nrONA) ITS region.They concluded that Sailltpaulia was nested within Streptoca11Jl1S subg.S(reptocarpella Fritsch and resolved relationships among other sections of SlreplocarplIs (MilUer & Cronk,