Pollen and reproductive morphology of Rhigiophyllum and Siphocodon ( Campanulaceae ) : two unique genera of the fynbos vegetation of South Africa

Pollen grains o f Rhigiophyllum squarrosum Hochst., Siphocodon spartioides Turcz. and S. debilis Schltr., are flattened and triangular w ith pores at the angles. This morphology is radically different from known pollen o f the Campanulaceae s.sfr:: the Campanulaceae are treated here as a family separate from the Lobeliaceae, Cyphiaceae, Nemacladaceae, Pentaphragmataceae and Sphenocleaceae (Lammers 1992). As traditionally conceived, the Campanulaceae is very heterogeneous and, in many classifications, these families were treated as subfamilies o f a much-enlarged Campanulaceae. The consistently dif­ ferent floral morphology, biochemistry and pollen structure o f the Lobeliaceae favours the recognition o f this predominantly tropical group as a separate family. The pollen grains o f these species are described in comparison with other members o f the Campanulaceae. Based on surface characteristics o f their pollen grains, we conclude that they represent an early offshoot o f the wahlenbergioid line­ age in southern Africa. We suggest that this unique pollen may also be the result o f a highly selective regime in the lynbos, associated w ith specialized pollinators, and base-poor soils, in addition to possible adaptations for ant dispersal and fire. Rhigiophyllum Hochst. and Siphocodon Turcz. are also unique in having free carpel-like structures within the ovary. These shrink to form seed pockets around the seeds and disperse as units when the capsule matures. Data from molecular studies support the contention that these taxa form a sister group to all other wahlenbergioids and that this should be formally recog­ nized in a classification system. We treat Rhigiophyllum and Siphocodon within the Campanulaceae: Wahlenbergioideae, as a separate tribe, the R higiophylleae tribus nov., the species o f which are distinguishable from other wahlenbergioids by unique angulaperturatc pollen, epipetalous stamens, free carpel-like structures and seed pockets.


INTRO DUCTION
During the course o f a palynological re-investigation of the Cam panulaceae1 s.Mr., a num ber o f pollen samples were obtained from material in the herbarium o f the Royal Botanic Garden Edinburgh and sent to the third author for scanning electron micrograph imaging (SEM ).Subse quently, samples representing Rhigiophyllum squarrosum Hochst.and Siphocodon spartioides Turcz., were found to have flattened, angular (triangular) pollen grains with pores at the angles.This m orphology is radically differ ent from all known pollen o f the Cam panulaceae, although it was reported for both genera in the landmark paper (in Russian) by Avetisian in 1967, which we had inadvertently overlooked.Initially, we suspected that the samples were contaminated, possibly due to alien pollen on the stigmas o f the herbarium material.However, an exam ination of material at the Compton Herbarium at Kirstenbosch by the second author using light microscopy, confirmed that both of these species indeed had radically different pollen mor phology.A third species, S. dehilis Schltr., was also exam ined by the second author and found to have pollen similar to its congener but was not included in the SEM analyses.This report describes the pollen shape and surface fea tures o f the exine o f the tw o principal species and genera involved.We also discuss other features o f these two gen era such as floral morphology and the unique seed pock ets, particularly with respect to their possible ecological significance.Finally, we discuss the systematic usefulness o f these findings for a revised classification of the wahlen bergioid genera and describe a new tribe, the Rhigiophyl leae. to accommodate Rhigiophyllum and Siphocodon.

MATERIALS A ND METHODS
Pollen (Table 1) was examined with a JEOL model 880 scanning electron microscope after cleaning with acetolysis (Erdtman 1960) and made electrically conductive with gold/palladium (Chissoe & Skvarla 1996).For the light microscope (LM) examination, pollen of the respective species was removed from alcohol-preserved flowers.The pollen w as placed on a microscope slide in a drop of water and examined.The gynoecium o f Rhigiophyllum was exposed by a longitudinal free-hand section through the hypanthium wall and the removal o f tissue with forceps.
1 The Campanulaceae is treated here as a family separate from the Lo beliaceae.Cyphiaceae, Nemacladaceae, Pentaphragmataceae and Sphe nocleaceae (Lammers 1992).As traditionally conceived, the Campanulaceac is very heterogeneous and, in many classifications, these families were treated as subfamilies o f a much-enlarged Campanulaceae.The consistently different floral morphology, biochemistry and pollen struc ture o f the Lobeliaceae favours the recognition o f this predominantly tropical group as a separate family.Welman (2000) treats the Campanu laceae separately from Lobeliaceae, within which she included the ge nus Cvphia P.J.Bergius.1).
Rhigiophyllum squarrosum is a rigid, sparsely branched subshrub, ± 0.30-0.45m tall with the habit of species o f Roella L. Its broadly ovate, coriaceous leaves are imbricate, squarrose, entire and in four ranks.Lan ceolate bract-like leaves subtend the azure-violet flow ers, which are aggregated in a terminal head.The corolla is elongated and consists o f a long narrow tube termi nated by five spreading obtuse lobes.The style is fili form.exserted, and terminates into three short stigmatic lobes (Figure 2A.B).
The two species o f Siphocodon are radically differ ent in appearance from Rhigiophyllum.They are gla brous, wiry subshrubs (S. spartioides is 0.3-0.6 m tall.S. debilis somewhat smaller) w ith sparse, minute, scale like, appressed leaves.The flowers are solitary, termi nal and axillary, mostly tow ards the apices o f the stems in a loose, few-flowered inflorescence.The flowers of Siphocodon spartioides are bluish purple, whereas those o f S. debilis are violet or whitish w ith pinkish brown honey-guides on the corolla tube.The corolla is nar rowly tubular-campanulate with five spreading obtuse lobes incised to about one-third the length o f the tube.The style is filiform, included and terminates into three short stigmatic lobes in S. spartioides, whereas in S. debilis the stigma is capitate (Figure 2C-E).
On closer inspection o f the corolla and capsule o f both genera, a number o f common features are found.For example, both have rather long, tubular corollas with the stamens adnate via the filaments to the corolla tube, the latter feature being unique among the Cam panulaceae.The stamens o f Rhigiophyllum squarro sum , which are almost exserted.are attached below the apex o f the corolla tube but the filaments are decurrent nearly to the base.In Siphocodon debilis, the stamens are included and are attached at the middle o f the corolla tube, whereas in S. spartioides, they are attached in the upper part o f the tube.
The capsule dehiscence is by means of a plug in Rhigio phyllum (Figure 3C) or circumscissile by means o f an operculum in Siphocodon (Figure 3E).In both genera, these structures comprise the upper part o f the ovary and the style, surrounded by the persistent corolla.In Rhigio phyllum, the line of dehiscence is above the calyx lobes and the seeds (within seed pockets) are dispersed through a narrow hole, whereas in Siphocodon it is below, so that, when the operculum detaches, the remaining lower  part o f the capsule is a neat, open, cup-like structure.
In Rhigiophyllum , the rem ainder o f the capsule easily detaches from the pedicel and disperses, probably with some seeds rem aining inside.Since the line o f dehiscence in the capsule o f Rhigiophyllum is above the calyx lobes, it resem bles that o f R oella and therefore differs in posi tional hom ology from the m echanism in Siphocodon.
Unlike other w ahlenbergioids, these two genera have (2)3 free carpel-like structures within the inferior ovary, each o f which has tw o to several pendulous ovules attached near the top (Figure 4B7).It is difficult to decide if the seed pockets separate from the wall o f the infe

4E11
).The walls o f these carpel-like structures shrink to enclose the seed at maturity, forming a carunculated pocket (Figure 4D10), which is released entire from the mature capsule.This structure was apparently overlooked by Botting Hemsley in Hooker's leonesplantarum (1897) w here he described the ovary simply as: 'Ovarium 3-loculare, loculis pluriovulatis, ovulis pendulis '. In Sonder (1865: 596), this seed pocket is apparently misidentified as a 'very loose, rugose testa'.The protuberances on the surface o f the seed pocket are similar in both genera although in Siphocodon they are more round and regular.
There are also slight differences in seed shape.Siphoco don seeds are slightly diamond-shaped in comparison w ith the oval seeds o f Rhigiophyllum.In both genera the seeds have a strong electrostatic charge and 'jum p' to about 0.1 m when manually extracted from the pockets.The function o f the seed pocket is unknow n, but it may perform a role in dispersal, for example by ants.The seeds o f these three taxa are large and few in number and this may be correlated w ith the establishment o f the seedling in nutrient-poor environments (Eddie & Cupido 2001).The shiny testa o f the seed would suggest that dormancy and nutrient release by fire may be important components in their ecology.Shiny testae are a feature o f many annual species o f the Campanulaceae where seed dormancy is the norm (Eddie 1997).Pollen grains disperse as monads and they are superfi cially like A1 nus MiUer/Betula L./Corylus L. (Betulaceae) or Rhamnus L. (Rhamnaceae).but very unlike the pollen o f Pentaphragma Wall, ex G.Don (Pentaphragmataceae), which was formerly considered to be close to the Cam panulaceae.and which has trilobate pollen with the pores between the lobes (Dunbar 1978(Dunbar , 1979(Dunbar , 1981)).Their shape in polar view is reminiscent o f species o f Acarpha Griseb.(Calyceraceae) (De Vore et al. 2007)

Palynological investigations o f the Campanulaceae
Studies o f the pollen o f the Campanulaceae are exten sive and the family is comparatively well known palynologically, but there are gaps in our know ledge o f the wahlenbergioid taxa o f the southern hemisphere, and o f many endemic campanuloid taxa o f central Asia.A brief, if diverse, survey o f Campanulaceae pollen was pro vided by Erdtman (1952), followed by a similar survey o f 21 genera by Chapman (1967).Avetisian (1967) pro vided a firm foundation for a systematic re-appraisal o f the family using palynological characters, but the most thorough examination o f the family using scanning elec tron microscopy was conducted by Dunbar (1973a-c, 1975a , b, 1978, 1979, 1981, 1984), w ho also studied ontogeny, and by D unbar & W allentinus (1976) using phenetic methods.The pollen o f the C am panulaceae can be divided into two broad groups as follows: 1.
The platycodonoid taxa o f Asia and Africa (e.g.Platvcodon A .D C., C yananthus Wall, ex Benth., Codonopsis Wall., C yclocodon, C am panutnoea, and Canarina L.) have pollen that is either 6 -10-colpate, 3-colporate, or 5-or 6-colporate (Figure 6).They have in com m on an oblate-spheroidal shape, a relatively high num ber o f colpi and an exine sculpturing that consists o f spinules, verruca-like spinules, or verrucae, between which are small pits o f uniform diam eter, or a reticulum in low relief with very small lumina.The ektexine structure consists o f a tectum perforated by m ostly narrow chan nels, medium to high bacula that are closely adpressed in som e species, and a reduced or absent foot layer.The endexine is alm ost undivided.

2.
The cam panuloid and wahlenbergioid taxa have pollen that is 3-or 4-porate, 6-porate, or 12-porate (Fig ure 5).The porate taxa are mostly zonotrem e or rarely pantoporate.The pantoporate condition is approached in those species that have an increased num ber o f pores and where the position o f the pores becomes irregular and not strictly zonotreme.The shape o f the pollen is sphe roidal or oblate-spheroidal, rarely suboblate or prolatespheroidal.The exine sculpture consists o f spinules o f different number, shape and size.Between the spinules there are ridges, protrusions or a low -relief reticulum , finger-like structures, or ridges with the top end bent upwards (D unbar 1975a).The ektexine structure (and sculpture) varies from sim ple to complex.Complex ektexine consists o f a surface covered by spinules, dis tinctly divided basally, short ridges/protrusions between spinules, a sponge-like tectum , stubble-like bacula con tinuous with an undivided foot layer, and connections to the tectum may be thin.Less complex ektexine con- sists o f a surface covered by spinules, which, in some species, divide basally, together with protrusions, lowrelief reticula, a thin, distinctly outlined tectum perfor ated by channels, high bacula that are continuous with the tectum and an undivided foot layer.Simple ektexine consists o f a surface covered by discrete spinules, less distinctly divided than the complex type, and low pro trusions.The uniformly outlined tectum is perforated by narrow channels and has the same thickness as the undi vided foot layer.The bacula are medium/high, and are continuous with the tectum and foot layer.The endexine, which varies in thickness, is lamellated, except in the simple ektexine type.
Knowledge o f pollen m orphology in the w ahlen bergioid genera is patchy and mostly concentrated on the genus Wahlenbergia (Thulin 1974;Dunbar 1975a, b), although a detailed study o f the pollen o f Hetero-chaenia A.DC. was undertaken by Badre et al. (1972), and by Straka & Simon (1969).W ahlenbergioid gen era such as Prismatocarpus, Roella, and Wahlenber gia are all in D unbar's Group 1, i.e. pollen grains that are mostly radially symmetrical, isopolar, zonotreme, 3-5-porate, spheroidal and tectate.Spinules are evenly distributed over the non-apertural surface o f the pollen grains.These three genera did not show any particular clustering with each other with respect to the other taxa (Dunbar 1975a, b).Apparently Von Brehmer (1915) considered the pollen morphology to be o f no value as a taxonomic character in Wahlenbergia.However, Erdtman (1952) placed Wahlenbergia, Roella and Pris matocarpus in a group o f genera with (2)3-{5)-porate, suboblate to oblate spheroidal pollen with spinuliferous sexine, which is thinner than the nexine.Avetisian (1967) studied five species o f Wahlenbergia (including W. hederacea (L.) Rchb.), seven species o f Lightfoo- Wahlenbergia, C ephalostigm a and L ightfootia could not be distinguished on pollen characters, although W. hir suta has an increased num bers o f pores that could be o f taxonom ic value.A ccording to Thulin, with increas ing num ber o f pores, the position o f the pores becom es irregular and not strictly zonotrem e.Several differences in spinule size and density exist betw een different groups o f the genus Wahlenbergia.For exam ple, the W. undulata (L.f.) A.DC. group has longer spinules than other groups, and the area o f the exine betw een the spinules in the W. undulata and W. m adagascariensis groups is dis tinctly granular or with short ridges.Thulin (1974) also reported that the pollen o f N am acodon Thulin disperses in tetrads, unlike the pollen grains o f all other taxa in the family, which disperse as m onads-tetrads have been recorded in Legousia falcata (Ten.)Fritsch ex Janch.(first author, unpublished data).

M olecular studies
Recent m olecular studies using /at?L-F and ITS gene sequences (Cupido 2008)  M erciera however, forms a weakly supported subclade.Roella, Prism atocarpus and Wahlenbergia were also found to be paraphyletic, the latter massively so.Theilera E.Phillips was found to be closest to several species o f Wahlenbergia, all o f which were formerly treated as Lightfootia, and in a clade com prising Craterocapsa and Wahlenbergia procum bens (L.f.) A.DC., W. huttonii (Sond.)Thulin, and W. stellarioides Cham.
A nother surprising result o f the m olecular studies has shown that Rhigiophyllum and Siphocodon form a sister group to all the other southern hemisphere w ahlenber gioids, including taxa from the Mascarene Islands and St Helena (H aberle et al. 2009).This has profound impli cations, for it suggests that this split in lineages was a very ancient one.Previous m olecular studies (Eddie et al. 2003) using ITS nrDNA found a clear dichotomy between the col pate/coIporate platycodonoid taxa and the porate wahlenbergioid and cam panuloid taxa.This m ajor split in the Cam panulaceae is hypothesized to be a consequence o f the isolation engendered by tec tonic activity in a fragm enting Early Tertiary Gondwana (Eddie et al. 2003).Subsequent evolution o f these two lineages was independent, with the bulk o f the platyco- donoids in eastern Asia, the wahlenbergioids in Africa, and the campanuloids differentiating primarily in north ern Africa and the evolving Mediterranean region.

Pollen morphology o/Rhigiophyllum and Siphocodon
The triangular pollen o f Rhigiophyllum and Siphoco don is so unlike the known pollen o f the Campanulaceae that it throws their relationship with that family into question.Kolakovsky (1987Kolakovsky ( : 1573) ) excluded both genera from the Campanulaceae, yet, from molecular data (Eddie et al. 2002;Haberle et al. 2009) and their possession o f porate pollen, it would appear that these genera are cor rectly placed close to typical wahlenbergioid, porate taxa.From a biogeographical viewpoint, one would favour a relationship with the wahlenbergioid taxa so character istic o f southern Africa, and with which they have tradi tionally been associated.The surface sculpturing o f the pollen is more simplified, lacks the dense spinuliferous condition, and recalls the surface features o f the pollen found in the platycodonoid genera.This suggests that these two genera may represent an older lineage o f the Campanulaceae in southern Africa that is somewhat inter mediate between platycodonoids and wahlenbergioids, or it may be that the pollen morphology is convergent with that o f the platycodonoids (perhaps the result o f paedomorphosis and/or neoteny).However, some porate pol len in the Campanulaceae is simpler in structure than the dense spinuliferous type.Avetisian ( l 967) suggested that tropical colpate/colporate pollen is the most primitive type within the Campanulaceae, whereas porate pollen from temperate zones, including pantoporate pollen, is considered an advanced type (Van Campo 1966;Muller 1970;Punt 1976).Dunbar's (1984) results agree partially with this view with respect to complex exine.Perhaps the unique triangular pollen represents a highly specialized adaptation either to conditions pertaining to their pol linators or to some, yet unknown, biological component  o f the fynbos vegetation.This explanation seem s highly plausible, but no other m em bers o f the C am panulaceae have this type o f pollen, so the functional significance of the triangular pollen rem ains unresolved.

G eneral m orphology a n d ecology o f the fy n b o s w ahlenbergioids
Rhigiophyllum and Siphocodon share a num ber of m orphological features with other fynbos taxa such as M erciera, Theilera, R oella , P rism atocarpus and m any fynbos species o f the group form erly included in the illegitim ate genus L ightfootia.T hey all show radical departures in a w hole suite o f m orphological characters from the tem perate C am panulaceae bauplan, although most can loosely be described as cricoid.All are dwarf undershrubs or shrublets, som ew hat rigid or wiry, and frequently w ith ericoid leaves in fascicles (e.g.M erciera and Theilera), or im bricate, as in Rhigiophyllum .
Many o f them have long tubular flowers (Merciera, Theilera, Rhigiophyllum , and Siphocodon) and indehiscent capsules ( M erciera) or at least an unusual capsule dehiscence m echanism (Prismatocarpus, Roella, Rhigio phyllum , and Siphocodon).Several o f these genera (e.g.M erciera, Theilera, and Prismatocarpus, subgen.Afrotrachelium Adam son) even show a remarkable, if super ficial, resem blance to the Lobeliaceae and Stylidiaceae, and the flowers o f Rhigiophyllum look similar to those o f Calycera Cav.(Calyceraceae).The flowers are either solitary and more or less sessile (Roella, Prismatocarpus, Theilera and M erciera), in loose terminal inflorescences (Prism atocarpus, Siphocodon), or rarely in dense heads (Rhigiophyllum).These features suggest a general con vergence in morphologies that may correlate with similar ecologies.However, as shown by the molecular studies, these subgroups are not part o f the same phylogenetic sublineages and their similarities are probably superficial.They may be best considered as parallel ecotypes.
Rhigiophyllum is so dissimilar morphologically from both species o f Siphocodon that, on first inspection, a close relationship between these two genera is not obvi ous.Furthermore, Siphocodon spartioides is very unlike S. debilis, yet.the infrageneric disparity in morphology between Siphocodon spartioides and 5. debilis offers a possible clue to the evolutionary history o f all three taxa, in addition to that o f the fynbos w ahlenbergioids in gen eral.This disparity suggests that divergent selection pres sure has been intense, driving three closely related spe cies towards radically different morphologies.A similar situation is seen in the two species o f Musschia Dumort.(Campanulaceae) on Madeira (Eddie et al. 2003).
The evolutionary divergence o f Rhigiophyllum and Siphocodon probably occurred early, in concert with the progressive aridity o f the Cape Region (Cupido 2008).From the similarity o f their floral morphology w ith other fynbos plants, we can infer that these taxa have highly specialized pollination syndromes, probably with longproboscid flies (including horse flies, tangle-wing flies and bee flies) as the principal pollen vectors (Goldblatt et al. 1995).However, until further studies are completed, we simply do not know what adaptive advantages, if any, are conferred by the unique pollen morphology, and ontogenetic studies are required to determine the signifi cance o f the triangular shape before and after tetrad for mation.
It is clear that these three species display a highly integrated complex o f adaptations to the fynbos vegeta tion and that nuances in ecological requirements prob ably account for the differences between them, but the functional aspects o f these adaptations remain unclear.This argument applies also to all the other wahlenber gioid genera in the fynbos, and therefore the merging o f genera such as Theilera and the illegitimate Lightfootia in Wahlenbergia is surely premature (Lammers 1995;Goldblatt & Manning 2000), although Lammers (2007: 382) acknowledged that some species, currently included in Wahlenbergia, could be given separate generic recog nition.From this perspective, the recognition o f Rhigio phyllum and Siphocodon should be upheld.
From the pollen studies, and the work o f Dunbar, it would appear that there are more similarities between the pollen o f the Campanulaceae and the Cyphiaceae s.lat.than the Lobeliaceae.The unique stylar morphol ogy o f the Cyphiaceae suggests that this family may be the most ancient lineage o f Campanulales in Africa and possibly derived from ancestors, which, themselves, eventually diversified in Australia as the Goodeniaceae and Stylidiaceae.This hypothesis requires further inves tigation.The Cyphiaceae have unicellular stylar hairs, which resemble those o f the Campanulaceae more than those o f Lobeliaceae (Leins & Erbar 2005), and it would be interesting to survey this character as well as second ary pollen presentation mechanisms in all South African genera o f the Campanulales.
The major dichotomy between the pollen o f the platycodonoids (represented in Africa and the Canary Islands by the relict Canarina) and the wahlenbergioids/ campanuloids, suggests that this split is an ancient one (Eddie et al. 2003) dating from the early Tertiary.Yet, w e do not know w hat the ancestral morphology o f these ancient African progenitors was like, but from them the wahlenbergioids diversified into several morphological types such as herbs or shrubs.Perhaps the Mascarene genera such as Nesocodon Thulin and Heterochaenia, both o f which recall the platycodonoids in their floral morphology, most resemble the ancient forms.In south ern Africa, the onset o f aridity, beginning in the Oligocene, probably is the ultimate cause o f evolutionary diversification within the wahlenbergioids, with addi tional factors such as geographic and ecological isolation (especially soil types and pollinators).Shrubby types such as Roella are particularly associated with the Medi terranean climate o f the Cape region, whereas herba ceous, rosette types such as Craterocapsa would appear to be restricted to areas w ith a greater moisture regime.The relictual disjunct distribution o f Craterocapsa from eastern South Africa to the Chimanimani Mountains of Zimbabwe may be highly significant.

Taxonomic implications
One can o f course attempt to analyse it, to fit it into this system o f thought or that, hut by its very nature it is bound to cause a diversion in the neatly-fitted jigsaw.In the end the diversion becomes the devia tion that wrecks the system.No wonder those who create system s fe a r it like the devil.

Neil Gunn 1956 ( The Atom o f D elight)
Since their original discovery and description, Rhigio phyllum and Siphocodon were classified by all southern African workers as being close to other wahlenbergioid genera.We now know that the two genera are more divergent from all other wahlenbergioid taxa than was previously thought, yet.from molecular analyses, they are obviously still part o f that nexus o f southern African Campanulaceae.However, they represent a sister lineage separate from other South African taxa.which suggests that they are an old, albeit highly adapted group (> 28 million years, Cupido 2008).The current classification o f South African wahlenbergioids is not adequate for the recognition o f these taxa and should therefore be modi fied accordingly.Kolakovsky (1987Kolakovsky ( , 1994) ) recognized four subfamilies w ithin the Campanulaceae based largely on the nature of carpel dehiscence and the presence or absence o f an axicom: Prismatocarpoideae Kolak.; Canarinoideae Kolak.; Wahlenbergioideae (Endl.)Kolak.; and Campanuloideae.In this treatment, the South African genera were divided between his Prismatocarpoideae (Craterocapsa, Mamacodon, Prismatocarpus, Roella and Treichelia Vatke) and his Wahlenbergioideae (Heterochaenia.Microcodon, Theilera and Wahlenbergia, plus a number o f typically platycodonoid and campanuloid genera).
Takhtajan (1997) also divided the Campanulaceae into four subfamilies: (Cyananthoideae nom.inval.? ; Ostrowskioideae (Fed.)Takht.; Canarinoideae Kolak.; and Campanuloideae, giving great w eight to the type of pollen grains.He subdiv ided the Campanuloideae into about thir teen tribes, including four South African tribes: the Wahlenbergieae ( Wahlenbergia, Berenice, Theilera, Gunillaea Thulin.Nesocodon.Heterochaenia, and Microcodon); Prismatocarpeae (Prismatocarpus, Namacodon, Roella, Craterocapsa, and Treichelia); Siphocodoneae (Siphoco don and Rhigiophyllum): and Merciereae (Merciera).The problem with both o f these system s is that there are too many tribes, that each tribe is alm ost the equivalent o f a genus, and that it is difficult to get a perspective o f the major lineages within the subfam ilies.Kolakovsky's sys tem places lar too m uch em phasis on the axicorn, which is probably more useful in delim iting cam panuloid taxa.Takhtajan's treatm ent o f the South African genera comes closest to our thinking but the num ber and boundaries o f his tribes may have to be revised.Sender (1865) included the four tribes Lobelieae, Cam panuleae, C yphieae and G oodenovieae in the C am panulaceae.He subsequently divided the C am panuleae into three subtribes: W ahlenbergieae (capsule opening at the apex; ovules m any); M erciereae (stam ens free; ovary one-celled, with an incom plete septum ; 4 basal ovules); and Siphocodeae (stam ens epipetalous; ovary 3-celled, each cell with 2 ovules).N ote that S onder used the sam e suffix '-eae' for his subtribal nam es instead o f '-in ae'.He considered Rhigiophyllum to be a 'doubtful g en u s' and w e think he sim ply tagged it on at the end o f his account of the Cam panuleae im m ediately after Siphocodon.He probably never intended to include it in his 'Siphoco deae' but there is som e am biguity to his account (p.597) and it w ould have been clearer if he had placed it sequentially after Roella.R higiophyllum is certainly very distinct from Siphocodon and, if he had m eant to include it in his Siphocodeae, he w ould surely not have consid ered it to be o f doubtful status.Takhtajan (1997) placed these two genera in his tribe S iphocodoneae Takht., which he recognized along with 12 o ther tribes in his subfamily C am panuloideae.As far as we can determ ine, Takhtajan's Siphocodoneae w as not form ally validated.Was Takhtajan sw ayed by S o n d e r's rather am biguous account?Lam m ers (2007: 671) lists the Siphocodoninae Sond.as a subtribe o f the W ahlenbergieae Endl.
It would be tem pting, given the highly divergent pol len morphology, to give subfam ily status to R higiophyl lum and Siphocodon.However, there are a num ber o f other taxa that are also som ew hat anom alous and do not fit com fortably into either the W ahlenbergioideae or the Campanuloideae, e.g.W ahlenbergia hederacea, Feeria Buser, Jasione L., M usschia and C am panula L. sect.Pterophyllum Damboldt.W ahlenbergia hederacea, w hich is a unique com ponent o f w estern European Atlantic regions, is rem ote from all other w ahlenbergioids.Feeria is closer in its m orphology to the w ahlenbergioids, whereas Jasione is closer to the cam panuloids (Eddie et al. 2003;Eddie unpubl.).M olecular studies also support the closer association betw een Jasione and the cam panu loids (Cosner et al. 2004).T hese taxa probably stem from ancestral taxa com m on to both groups, w hat Eddie et al. (2003) referred to as 'transitional g ro u p s'.In the world of classification, there are alw ays taxa that do not fit neatly into man-m ade schem es.G iven the uniqueness of Rhigio phyllum and Siphocodon, we hereby include them in a new tribe o f the C am panulaceae as follows: Fruticuli habitu et affinitate Roellae vel Prismatocarpi, a quibus praecipue poll inis granis applanatis et triangularibus, uno poro in quoque angulo praeditis, differunt; corolla longe tubulosa, staminibus inclusis, ad tubi medium vel infra corollae orem insertis; structuris membranaceis liberis intra ovarium carpella simulantibus, in tem pore maturitatis se contrahentibus, seminum marsupia rugosa vel carunculata formantibus et semina continentibus dispersis; capsula dehiscente aut obturamento supra calycis lobos amoto (Rhigiophyllum ) aut operculo infra calycis lobos circumscissili (Siphocodon).
Shrublets with the appearance of, and affinity with, Roella or Prism atocarpus, principally differing from them by pollen grains that are flattened horizontally and triangular, with one pore at each angle; with corolla longtubular, with stam ens included and inserted at the mid dle o f the tube or below the mouth o f the corolla; with free m em branous structures within the ovary resembling carpels, shrinking at maturity, forming rugose or carunculated seed pockets and dispersed containing the seeds; w ith the capsule dehiscing either by removal o f a plug above the calyx lobes (Rhigiophyllum) or by a circumscissile operculum below the calyx lobes (Siphocodon).
This also necessitates that we clarify the placement of this new tribe within a suggested overall classification system o f the Campanulaceae.To date, the reclassification o f the Cam panulaceae is still fluid and a presentation o f a new' system is inappropriate in this paper.However, we recom m end the recognition o f three subfamilies within the Cam panulaceae to embrace the platycodonoids, wahlenbergioids and campanuloids, based principally on pollen morphology, but also supported by biogeog raphy.Thus, the tribe Rhigiophylleae would be included in subfam ily Wahlenbergioideae Kolak.(1987)-includ ing Prismatocarpoideae Kolak.(1987), comprising Wahlenbergia, Prismatocarpus and allied genera possessing spherical or triangular, porate pollen and capsule dehis cence that is predominantly apical by valves.Their dis tribution primarily in the southern hemisphere, extending marginally into the northern hemisphere in Eurasia but poorly represented in South America.
This study has shown that what at first sight appears as novel, morphological divergences are integrated with many other features o f a plant's morphology, ecology and evolution, and that no single aspect can ultimately be divorced from the plant as a whole.Because these unique plants are so finely tuned to their unique environment, they are highly vulnerable to disturbance, habitat degradation and climate change.We still know very little about them but we hope that their protection is assured and that further studies o f such intriguing plants will be forthcoming.

FIGURE
FIGURE 2.-A, B, Rhigiophyllum squ arrosu m : A, habit; B, details o f inflorescence.C, D, Siphocodon spartioides: C, details o f flower and remains o f capsule; D, slender wiry stem s. S. deh ilis: E, details o f flower showing honey-guides and the twisted, entwined stems.Photographs: A, W.M.M. Eddie; B -E , C.N. Cupido.
rior ovary o f adult flowers or if they are formations sui generis (proliferations o f the placentae) (Erbar & Leins pers.comm.)*.Some ovules appear to abort before matu rity leaving just one or two seeds per structure (Figure * Profs Claudia Erbar and Peter Leins conducted a preliminary investi gation o f the ovary o f Rhigiophyllum from material supplied by the sec ond author.They report that the inferior ovary develops as in all other cases [o f Campanulaceae] due to an intercalary growth in the floral axis and that the seed pocket is a special form o f an endocarp.The epider mis (and eventually a few cells o f deeper layers) o f the ovary locules separates from the wall o f the inferior ovary to form the seed pockets.A com plete ontogenetical study (including histology and SEM -investigation) is planned and the results will be published in due course.
and com bined chloroplast DNA datasets (rbcL, atpB and m atK ) (H aberle et al. 2009) have shown quite conclusively that the strongest m olecular affinities o f Rhigiophyllum are with the two species o f Siphocodon (Figures 7; 8).M erciera, Roella and Prism atocarpus form a w ell-supported clade, but the relationships within this clade are largely unresolved.