The genus Rubus ( Rosaceae ) in South Africa . IV . Natural hybridiza tion

The genus Rubus L. is represented in southern Africa by the subgenera Eubatus Focke and Idaeobatus Focke. A combination o f morphological data, data on the reproductive systems of some collections and meiotic chromosome beha­ viour indicates that a hybrid swarm in the eastern Transvaal was formed subsequent to the hybridization between R. cuneifolius Pursh. taxon B (subgenus Eubatus) and R. longepedicellatus (C. E. Gust.) C. H. Stirton (subgenus Idaeoba­ tus). Other examples of intraand intersubgeneric hybridization were found during this study of the South African material. These instances, with examples found in the literature, indicate that the subgeneric subdivisions of Rubus are artificial. Three different methods were used to analyse the meiotic chromosome configurations. The genome relationship system of Alonso & Kimber (1981) and Kimber & Alonso (1981) and the modification of the binomial system of Jackson & Casey (1980) by Spies (1984) proved to be the most sensitive for distinguishing between alio-, segmental alioand autoploids.


INTRODUCTION
The genus Rubus is somewhat enigmatic in South Africa.It forms part of our indigenous flora but natura lized species also occur.Most taxa are considered weedy and yet they are included in a breeding programme to improve their agricultural production.The genus also contains agamic species as well as sexual species.In short, it is a taxonomist's nightmare.
The genus Rubus comprises 12 subgenera of which two are represented in South Africa: Eubatus Focke and Idaeobatus Focke.In South Africa the subgenus Euba tus, or true brambles or blackberries, includes only exotics, whereas the subgenus Idaeobatus, or rasp berries, contains a few exotics and a number of indige nous species (Spies & Du Plessis 1985).
It has been proposed (Stirton 1981a & b;Spies & Du Plessis 1985) that the problems with Rubus taxonomy in South Africa are caused by the occurrence of apomixis, hybridization among indigenous species and between in digenous and exotic species, the variation produced by a breeding program with subsequent escape from cultiva tion and inadequately collected herbarium material.
Each paper in this series has dealt with a different aspect of the cytogenetics of Rubus in South Africa.The aim of this paper is to determine whether natural hybridi zation occurs in the South African Rubus complex and whether this hybridization, if it does occur, is restricted to intrasubgeneric taxa.Stirton 9797,9798,9799,9801,9855,9860,9862,9863,9864,9865,9866,9867,9869.This cytotaxonomic study concentrated upon a possi ble hybrid swarm in the area between Graskop and Sabie in the eastern Transvaal Lowveld (2430DD) (Stirton 1984).The cytogenetical methods and results were re ported by Spies & Du Plessis (1985& 1986) and Spies, Du Plessis & Liebenberg (1985).These investigations included meiotic analyses of aceto-carmine anther squashes and embryo sac studies.
In an attempt to determine cytogenetically whether hybridization has occurred, three different methods were used to compare the observed chromosome configura tions of polyploids with the expected values for autoploids.These methods included the genomic relationship system developed by Kimber and others (Kimber & Hulse 1978;Driscoll 1979;Driscoll, Bielig & Darvey 1979;Alonso & Kimber 1981;Espinasse & Kimber 1981;Kimber & Alonso 1981;Kimber, Alonso & Sallee 1981;Alonso & Kimber 1984), the binomial system de veloped by Jackson et al. (Jackson & Casey 1980& 1982;Jackson & Hauber 1982) and the modification of this binomial system by Spies (1984).Computer pro grammes were used to calculate these values.The model with the smallest average sum of squares between the expected and observed frequencies, was considered as being the most appropriate model.

Morphology
The two probable species participating in the forma tion of the apparent hybrid swarm were identified as R. longepedicellatus (C.E. Gust.) C. H. Stirton of the subgenus Idaeobatus Focke and R. cuneifolius Pursh taxon B belonging to the subgenus Eubatus Focke.It was assumed that these species formed morphologically distinct hybrids, referred to here collectively as, R. X proteus C. H. Stirton.The morphology of the different plants is summarized in Table 1.
In order to determine whether the R. x proteus speci mens are intermediate between the putative parental species or fall within the normal infraspecific variation of these species, all R. cuneifolius B and R. longepedi cellatus specimens in the National Herbarium (PRE) were scored for the selected characters listed in Materials and methods.These results are also summarized in Table 1 and clearly indicate that both these species are morpho logically variable.
Nevertheless, several distinct morphological differ ences between R. longepedicellatus and R. cuneifolius B were observed.For example, the average petal length in R. cuneifolius B was 17,1 mm, compared to the average of 6,4 mm for R. longepedicellatus.R. cuneifolius B is separated from R. longepedicellatus mainly on flower colour, petal and rachis lengths, ratio between the lengths of the petal and the sepal and whether the primo cane leaves are pinnate or pinnate/palmate.Characters that did not contribute to the separation of these species were double or single flowers, petiole length, straight or recurved thorns, con-or discolourous leaf surfaces, number of leaflets per leaf in the floricane and the termi nal leaf length.It was therefore decided to use only those characters which contributed to the separation of the species, to determine a hybrid index (Figure 1) according to the method developed by Anderson (1949).A scatter diagram (Figure 2) was constructed using the rachis and petal lengths on the X-and Y-axes respect ively.Other morphological characters used in the scatter diagram were flower colour, the ratio between the lengths of the petals and sepals and whether the primocane leaves were pinnate or pinnate/palmate.

Reproductive system
The presence of both reduced (sexual) and unreduced (aposporic) embryo sacs was described in the triploid R. cuneifolius B specimens, Henderson & Gaum 18 and Stirton 9800 (Spies & Du Plessis 1986).However, all the reduced embryo sacs were observed to degenerate at maturity.The one tetraploid specimen, Stirton 9861, was 100 % sexual, whereas the other one, Stirton 9868, was only 35 % sexual.In addition to this sexual and asexual reproduction through seeds, all specimens repro duced vegetatively through stemtip-rooting.
In contrast to the apospory described in the R. cuneifo lius B specimens, no apospory was observed at any ploidy level in the R. longepedicellatus sample studied, except that in the pentaploid R. longepedicellatus speci men, Henderson & Gaum 36, all the reduced embryo sacs degenerated at maturity and the plant was, there fore, sterile (Spies & Du Plessis 1986).Vegetative re production occurs through rhizomes.

Chromosome behaviour
Both putative parental species contain specimens on different polyploid levels.R. cuneifolius B has somatic chromosome numbers of 21 and 28 and R. longepedicel latus 14, 28 and 35, whereas their presumed hybrid, R.
The meiotic chromosome behaviour observed in R. cuneifolius B differs in some respects from that of the comparable ploidy level of R. longepedicellatus (Spies et al. 1985).The diploid R. longepedicellatus (Henderson & Gaum 22) specimen has a chiasma frequency of 1,12 per bivalent and that of the putative hybrid diploid speci men is similar, namely 1,1 (Table 2).Both diploid speci mens usually formed bivalents, with the exception of two univalents in one R. longepedicellatus cell.The meiotic chromosome configurations in the triploid R. cuneifolius B and R .X proteus specimens were very similar (Table 2).No triploid R. longepedicellatus speci men has yet been found.
The method described by Spies (1984) for analysing the meiotic configurations in the pollen mother cells, indicates that the tetraploid R. cuneifolius B specimen is a segmental alloploid tending towards autoploidy, whereas the tetraploid R. longepedicellatus specimen is a segmental alloploid tending strongly towards alloploidy.Some of the tetraploid R. x proteus specimens appear to be segmental alloploids tending towards autoploidy (.Henderson & Gaum 27 & 52), whereas one is probably an alloploid (Stirton 9798).The tetraploid R. rigidus x R. cuneifolius A specimen (Henderson & Gaum 51) seems to be a segmental alloploid tending towards auto ploidy.
In a tetraploid R. x proteus specimen, Stirton 9798, asynapsis occurred in many pollen mother cells.In this specimen only 44,6 % bivalents were formed, whereas the remaining chromosomes were univalents (Table 2).The pentaploid R. longepedicellatus specimen tended to form less bivalents than the R. x proteus specimen.No higher ploidy levels than pentaploid were found in the parental species and comparison with the hexaploid R. x proteus specimens was, therefore, not possible.How ever, a surprisingly high frequency of multivalents (14,05 %) was observed in the higher ploidy levels of R. X proteus (Table 2).
The genome analysis indicated that there is no differ ence between the 2:1 and 3:0 models of Alonso & Kimber (1981), because the x-values in the 2:1 model were 0,5 for each triploid specimen, indicating that the two more closely related genomes are also closely re lated to the third genome.The model with 0-2 chiasmata of Jackson & Casey (1982) produced the same expected values as those obtained by using Kimber's models (Ta ble 3).In all the specimens studied the average sum of squares increased from the 0-2 chiasmata model of Jack son & Casey (1982) to the 0-4 chiasmata model, indica ting that the specimens studied have two or less chias mata per chromosome pair (Table 4).
The genome analysis further indicated that the 2:2 model of Kimber & Alonso (1981) shows the best corre spondence with the observed frequencies of chromosome associations in all the tetraploid R. cuneifolius B, R. longepedicellatus and R. X proteus specimens studied (Table 5).In each case the value of x was 1, indicating that two genomes are much more closely related to one another than to one of the other two genomes.The only exceptions were Henderson & Gaum 93 (R .cuneifolius A) and Stirton 9798 (R. x proteus) in which the 3: 1 model fitted with x-values respectively of 0,5 and 0,501, indicating that the three closely related genomes have also a great affinity for the other genome.In both these cases the average sum of squares of the expected and observed frequencies of the accepted model varied very little from that of the 4:0 models.
In contrast to this phenomenon the model described by Jackson & Casey (1982) indicates that all the specimens TABLE 3.__Comparison between observed chromosome configurations and the expected chromosome configurations in triploids according to the methods described by Alonso & Kimber (1981), Jackson & Casey (1980,1982) and Jackson & H auber( 1982).
Only the model with the lowest average sum of squares is given in this table  studied are autotetraploids with 0-2 chiasmata per chro mosome pair and with partly random chromosome asso ciations.The model of Spies (1984) indicates that all the specimens are segmental alloploids but they vary from almost autoploid (Henderson & Gaum 21,51,93 and Stirton 9868) to almost alloploid (Henderson & Gaum 14,32,Stirton 9798,9861 & 9862).

Morphology
Different methods can be used to ascertain whether a given specimen represents a true species or a hybrid.During this study several of these methods were used to determine the degree of hybridization in the genus Ru bus.The first method used was based on morphological characters and in this process a hybrid index was deter mined and a scatter diagram constructed.
A study of morphological characters revealed that R. cuneifolius B has a short to medium length inflorescence (Table 1:1) with white flowers, whereas R. longepedicellatus has a medium to long inflorescence with pink flowers (2 & 3).R. x proteus has a short to long inflorescence with pink, pale pink or white flowers.The petal length (4) varied from 13 to 20 mm in R. cuneifolius B, from 4 to 10 mm in R. longepedicellatus and from 4 to 15 mm in R. x proteus.The petal width (5) varies from 7 to 15 mm in R. cuneifolius B, from 3 to 6 mm in R. longepedicellatus and from 3 to 11 mm in R. X proteus.The same intermediate arrangement position is observed when the ratio between the lengths of the petals and sepals (7) is compared; in R. cuneifolius B the petal is always longer than the sepal, whereas in R. long epedicellatus the petal is as long or shorter than the sepal and R. x proteus has the whole range of ratios.R. cunei folius B, has acute petal apices compared to the acumi nate apices with an occasional acute apex in R. longepe dicellatus and both acute and acuminate apices found in R. x proteus.The leaf apex ( 12) is always acute in R. cuneifolius B and the leaf margin ( 13 The hybrid index diagram (Figure 1) indicates that only one specimen had all the characters associated with R. longepedicellatus, whereas four specimens had all the characters associated with R. cuneifolius B. The hybrid index also indicates that R. longepedicellatus and R. cuneifolius B are clearly separated morphologically.However, a continuous bridge of morphological charac ters spans the gap between them in the form of the very variable hybrid species, R. x proteus ( Figures 1, 2, 3 &  TABLE 5.-Comparison between observed chromosome configurations and the expected chromosome configurations in tetraploids according to the methods described by Kimber & Alonso (1981), Jackson & Casey (1980,1982)   TABLE 5 .-Comparisonbetween observed chromosome configurations and the expected chromosome configurations in tetraploids according to the methods described by Kimber & Alonso (1981), Jackson & Casey (1980,1982) and Jackson& H auber (1982).
Only the two models with the lowest average sum of squares of each method are shown in the  4).It is also indicated that the hybrid species overlaps morphologically with both parental species.The five major characters described above (i.e.flower colour, petal and rachis lengths, ratio between length of petal and sepal and whether the primocane leaves are pinnate or pinnate/palmate) are, therefore, essential for dis tinguishing between the true species and the different hybrids.
The pictorialized scatter diagram (Figure 2) indicates that more hybrid specimens overlap with R. longepedi cellatus than with R. cuneifolius B. R. longepedicellatus is completely surrounded by R. x proteus specimens in this diagram.Distinguishing between them will, there fore, be more difficult than between R. cuneifolius B and R. X proteus.
It is evident from these two diagrams that R. cuneifo lius B and R. longepedicellatus represent the two ex tremes of a very variable population of plants (Figure 3).It is further evident that R. x proteus, which constitutes the morphologically intermediate population (Figure 4), resulted from hybridization between R. cuneifolius B and R. longepedicellatus and subsequent backcrosses and in tercrosses to produce a continously variable hybrid swarm.The above hybridization hypothesis is also supported by the geographical distribution of the species con cerned.R. cuneifolius B is restricted to the Transvaal, whereas R. longepedicellatus specimens were collected in the Transvaal and Natal, with the majority of them collected in the Transvaal.The hybrids are restricted to the Transvaal.The low frequency of R. x proteus and R. longepedicellatus specimens from Natal in the collection may be attributed to an insufficient number of Rubus collections from Natal.The absence of R. cuneifolius B specimens from Natal in the National Herbarium may be due to inadequate collecting or to its non-occurrence in this province.If the latter is true, the paucity of R. x proteus specimens from Natal is explained.The speci men resembling R. x proteus (Henderson & Gaum 51) from Natal rather represents a hybrid between R. cunei folius A and R. rigidus than R. x proteus itself.The morphological differences between/?, cuneifolius B and R. cuneifolius A are very slight and hybrids between any one of these taxa and R .longepedicellatus will result in morphologically similar hybrids.The only differences observed between these taxa are small differences in the leaf texture and leaf margin, as well as the frequent oc currence of double flowers in R. cuneifolius B. No R. cuneifolius A specimen with double flowers was ob served.Since all R. x proteus specimens have single flowers, it is possible that R. cuneifolius A and B are interchangeable as parents with R. longepedicellatus.

Reproductive system
The embryo sac study indicated that both putative pa rents produce reduced embryo sacs and may, therefore, participate in hybridization.It was further demonstrated that a number of hybrids also produced reduced embryo sacs and so backcrossing to either parent is also possible.In addition to reduced reproduction all hybrid specimens had the potential to reproduce asexually, either through agamospermy or vegetatively.This apomictic reproduc tion provides all plants with the potential to reproduce even when meiotic chromosome pairing fails after inter specific hybridization.Although the embryo sac study cannot prove the occurrence of hybridization, it indicates that hybridization is possible and that interspecific hy brids may either reproduce sexually or perpetuate them selves apomictically.

Chromosome behaviour
The somatic chromosome numbers of 21 and 28 in R. cuneifolius B and 14, 28 and 35 in R. longepedicellatus (Table 2) seem to contradict hybridization because, although a diploid hybrid specimen exists, no diploid R .cuneifolius B specimen has yet been observed.However, the occurrence of triploid R. cuneifolius B specimens with meiotic chromosome behaviour resembling autoploids, suggests that these triploids are formed by polli nation of autotetraploids by diploids, both containing similar genomes.Therefore, it is suggested that diploid R. cuneifolius B specimens exist and that they could have hybridized with diploid R. longepedicellatus speci mens to form diploid hybrids.The occurrence of a di ploid hybrid R. X proteus specimen (Henderson & Gaum 28), with normal chromosome pairing during meiosis (Table 2), indicates that the genomic differences between the parental species are insignificant.The two diploid parents of R. x proteus probably differ only in a few gene loci and as such must be considered varieties of the same species.This homology between the genomes of R. cuneifolius B and R. longepedicellatus is also manifested at higher ploidy levels.However, differences in the meiotic chro mosome behaviour of polyploid R. x proteus specimens was observed.These differences include a variation in chromosome pairing from the multivalent formation ex pected in autoploids to that expected in alloploids.These differences can be attributed to either pre-or post-hybri dization chromosomal evolution.
Pre-hybridization chromosomal evolution would suggest that structural chromosome differences were pre sent in some plants of the parental populations.Hybridi zation between such plants followed by polyploidization would give rise to segmental alloploids with meiotic chromosome pairing resembling that of alloploids.The normal meiosis found in a diploid hybrid specimen (Hen derson & Gaum 28), indicates that only very small struc tural differences exist at the diploid level between the genomes of at least some plants of the parental taxa.
The results of Spies et al. (1985) indicate that the polyploids of R. cuneifolius B may have had an autoploid origin in contrast to the presumed segmental alloploid origin of R. longepedicellatus polyploids.The morphological similarity between the diploid and the segmental alloploids of R. longepedicellatus indicates that the structural chromosome changes in a genome were not accompanied by gene mutations which could produce morphological changes.The differences in chromosome pairing observed in different R. x proteus specimens at higher ploidy levels (Tables 3 & 5), might consequently be attributed to repeated hybridization be tween different R. cuneifolius B and R. longepedicella tus plants which differ in their structural chromosome changes.
Post-hybridization chromosomal evolution is due to structural changes in some chromosomes after hybridiza tion.The occurrence of multivalents tends to increase meiotic instability and to lower fertility.Chromosome changes that will inhibit multivalent formation will, therefore, have a selective advantage due to the in creased number of bivalents and the consequent increase in seed viability.These changes form part of the diploidization process.Different R. x proteus specimens may, therefore, represent different stages of diploidization and their meiotic chromosome pairing may consequently dif fer.However, the post-hybridization hypothesis only provides for autopolyploidization, whereas the pre-hy bridization chromosome evolution hypothesis allows re peated hybridization between different ploidy levels or between plants at the same ploidy level but with different genomic constitutions.The pre-hibridization hypothesis is also supported by the greater morphological variation in R. longepedicellatus when compared with R. cuneifo lius B .This larger morphological variation might be the result of the segmental alloploid origin of the R. longepe dicellatus polyploids.
Other interspecific hybrids and intersubgeneric hy brids have been described in the literature (See dis cussion under hybridization).In addition to the examples cited in the literature, the hybrid origin of certain taxa was inferred from their meiotic chromosome pairing.These taxa include R. cuneifolius A, R. flagellaris, R. apetalus and R. pinnatus.Chromosome pairing indicated that R. flagellaris, R. apetalus (Henderson & Gaum 6 and Wells 5000) and R. pinnatus are true alloploids; the 2:2 model of Kimber & Alonso (1981) was applicable and an x-value of 1 was obtained (Table 5).The tetraploid R. cuneifolius A specimen tends towards autoploidy, because the 3:1 model was applicable and the reduced x-value of 0,5 implied an affinity between the two sets of genomes.The other R. apetalus specimen, G. Hemm s.n., conforms with the 2:1:1 model and has an x-value of 0,86.A specimen that appears to be an amphiploid between R. transvaliensis and R. longepedi cellatus had an x-value of 1 when the 2:2 model was applied.No indication of a hybrid origin could be found for R. affinis, where the 4:0 model of Kimber & Alonso (1981) was applicable.
In contrast to the methods described by Alonso & Kimber (1981) and Kimber & Alonso (1981) and Spies (1984) the method described by Jackson & Casey (1982) and Jackson & Hauber (1982) suggests that all the plants are autoploids with partial random chromosome associa tions and 0-2 chiasmata per chromosome pair (Table 5).The reason why the latter method did not distinguish between different chromosomes in the specimens studied is that the initial assumption of the method, that the formation of chiasmata is random, does not apply in the genus Rubus.From random chiasma formation and a maximum of two chiasmata per chromosome pair, fre quencies of 0,25, 0,5 and 0,25 are expected for chromo some pairs with no chiasmata, one chiasmata and two chiasmata respectively.In the genus Rubus these figures are 0,08, 0,79 and 0,13.This deviation from the ex pected values indicates that this method is not applicable in the genus Rubus.

Hybridization
Hybridization in the genus Rubus is a topic as contro versial as the taxonomy of the genus.Taxonomists usually adhere to one of two extremes.Either every en tity not fitting the species description exactly is as a hybrid, or the occurrence of hybrids in the genus is totally ignored.Bailey (1941Bailey ( -1945) ) described over 500 different spe cies of Rubus for North America without the recognition of hybrids.He considered three points as essential for hybridization: (1) both parents must be in the vicinity of the hybrid; (2) hybrids occur in small numbers as incidental or as exceptions to the main population and (3) characters appear to belong to the parents in various degrees of combinations.
We support Bailey in his plea that all unidentifiable examples should not be regarded as hybrids.However, the validity of his three criteria for hybridization must be discussed before any conclusions can be made.His claim that both parents must be in the vicinity of the hybrid was usually fulfilled in the present study as the hybrids and the parental taxa often occurred together.However, hy brids were sometimes found with no parental form in the vicinity.This phenomenon may be attributed to one or more of several factors.Pollination by insects over large distances might occur and in such cases only the mater nal parent need be in the vicinity.Seed could also have been transported from the mother plant by birds or man, dropping it far from the parental forms.This may be a common means of dispersal in southern Africa as the fruits of Rubus are relished by birds and man.One or both parents may die and only the hybrid may survive, especially in a weedy taxon like Rubus where hybrids might be very aggressive.Only one or neither parent need therefore be in the vicinity of the hybrid.The first of Bailey's criteria for hybridization is therefore invalid.
The second criterion claims that hybrids occur in small numbers as incidental or exceptions to the main popula tion.This will be valid only for newly formed hybrids or weakly developed hybrids or species which have good barriers against hybridization.Rubus hybrids are often aggressive (Bammi 1964) and, due to hybrid vigour, they may exceed their parents and could become more abundant than either parental taxon.This is definitely the case with R. x proteus in the Graskop and Sabie areas of the Transvaal where the hybrids are exceptionally vigo rous and are more abundant than the putative parents.
Characters do not have to be intermediate in the hy brids.They may exceed the ranges of both parents, new traits may be present in the hybrid or the traits of one parent may be absent in the hybrids due to dominance or epistasis.An example of the hybrid's trait exceeding that of its parents is found in the R. trifidus x R. hirsutus hybrid which has a larger flower diameter than either parent (Jinno 1957).In the present study it was observed that some hybrid specimens had longer rachises than either parent.
The three criteria for the determination of hybridity described by Bailey are, consequently, not always valid.These criteria are all based on morphological characters.Therefore, cytogenetic studies seem to be the only posi tive way of identifying hybrids.However, even this field is beset with problems and must be handled with extreme care to obtain meaningful results.This is illustrated by the different results obtained when using the different methods described for analysing genome homology.
Morphological, reproductive and cytogenetic evi dence indicates that hybridization does occur in the South African Rubus complex.Futhermore hybridization appears to take place on both the present taxonomic in trasubgeneric and intersubgeneric levels.The progeny derived from certain intersubgeneric hybridizations are fertile (Jinno 1958;Newton 1975).

Taxonomic implications o f hybridization
In general, F, hybrids and their offspring cannot be considered to be separate species because they are sterile due to the failure of normal chromosome pairing during the meiotic process of sporogenesis.However, when hybridization is associated with, or followed by chromo some doubling, amphiploids are produced with normal chromosome pairing and good fertility.These new selfreproducing entities may be regarded as new species (Davis 1958) because the amphiploids are reproductively isolated from their parents.In the event of hybridization resulting in apomixis, each apomictic hybrid might represent a different genotypic combination of the sexual parents and a multitude of different self-reproducing en tities can be formed.An increase in the degree of hetero zygosity of the parental forms will result in an increase in the number of different recombinant entities.This array of apomictic self-reproducing entities, which are mor phologically different from each other and genetically isolated, may on superficial study be regarded as sepa rate species or microspecies.It is, however, unpractical to consider each of these apomictic hybrids as separate species, even if only obligate apomixis exists.In fact, they belong to an agamic complex without species boun daries which rests on pillars of sexual diploid (and poly ploid) species.Only cytogenetical studies can distin guish between the true sexual species and the array of apomicts forming the agamic complex.
The fact that many Rubus species are restricted to very small geographical areas (Bailey 1941(Bailey -1945;;Davis & Davis 1951) could indicate that they represent either newly formed species or the abovementioned amphiploid apomicts.Apomixis is restricted to a small number of Rubus specimens in South Africa, i.e. the subgenus Eu batus.The tendency to describe a sexual hybrid as a separate species is frequently encountered in this genus.As an example the diploid species R. toyorensis and R. nishimuranus can be cited (Jinno 1957;Naruhashi 1971).The F, hybrid between the diploid species R. trifidus and R. hirsutus is regarded as a separate species, R. toyorensis, and the backcross of R. toyorensis to one of its parents is regarded as R. nishimuranus.In our opinion many of the described Rubus 'species' are only hybrids.This has resulted in a totally artificial classifica tion of the genus where different morphological entities are regarded as separate biological species.
An example of the hybridization between different morphological 'species' is found in the species R. apeta lus Poir., R. exsuccus Steud., R. adolfi-friederici Engl, and R. ecklonii Focke.Although these four 'species' are morphologically distinct, hybridization among them has produced more intermediate fertile specimens than typi cal specimens.In our opinion these four species belong to one biological species.Spontaneous hybridization is less common among the indigenous Rubus species of southern Africa.It occurs between R. rigidus J. E. Sm. and R. pinnatus Willd.wherever these species are sympatric, e.g.G. Hemm s.n. in PRE and was described by Focke (1914).Hybridiza tion between indigenous and introduced Rubus species is observed much more frequently.Such hybridization takes place between R. fruticosus L. agg.and R. pinna tus in disturbed areas of the Cape Peninsula (Adamson & Salter 1950).Other examples are R. cuneifolius A and R. pinnatus in Natal (G.Hemm s.n.) and R. affinis and R. rigidus described by Gustafsson (1933).All these cases involve hybridization between indigenous Idaeobatus and introduced Eubatus species.No hybrid swarms of any of these examples have been recorded to date.

CONCLUSIONS
The combination of morphological, geographical, re productive and cytogenetic evidence revealed that natu ral hybridization occurs in the South African Rubus com plex and also indicated that the hybridization is not re stricted to intrasubgeneric hybridization, but that intersubgeneric hybridization also occurs.The progeny de rived from certain intersubgeneric hybridizations are fer tile.
The application of the genome analysis method of Kimber & Alonso (1981) on the meiotic data indicated that all the tetraploid plants of R. cuneifolius B, R. fla gellaris, R. apetalus, R. longepedicellatus, R. pinnatus and R. x proteus have two genomes that are more close ly related to each other than to the other two genomes which are also related.This model indicates that all the plants are segmental alloploids with a tendency towards alloploidy.The model of Jackson & Casey (1982), on the other hand, indicates that all the plants are autoploids with partly random chromosome association.Totally dif ferent conclusions can, therefore, be drawn from the same meiotic data.Neither of the two models mentioned above distinguishes between any of the specimens stu died.However, the chromosome configurations indicate that chromosome pairing varies between the different plants.These differences are accentuated by the model of Spies (1984).It is, therefore, concluded that the latter model is the most applicable for plants with very short chromosomes which have a low chiasma frequency, as is the case in the genus Rubus (Spies et al. 1985).
Finally, interspecific hybridization in the genus Ru bus, without loss of fertility in the progeny, indicates that several of the morphological 'species' described in the past, belong to the same biological species.Since the difference in fertility levels between 'intersubgeneric' hybrids and 'interspecific' hybrids is negligible, it was concluded that the present classification of the genus Rubus is very artificial and urgently needs a biosystematic revision.
FIGURE 1.-Histogram of hybrid indices for specimens o f R. longe pedicellatus (area with horizontal lines), R. cuneifolius B (solid area) and R. x proteus (dotted area).

FIGURE 2 .
FIGURE 2.-Scatter diagram of R. longepedicellatus A ,R .cuneifolius B • and R. x proteus ^ specimens.The occurrence o f white flowers in a specimen is indicated by a solid character in contrast to the line character used for pink flowers.Specimens in which the petal length exceeds the sepal length are indicated by a A-sign under the character and pinnate leaves are indicated by a A-sign above the character.

*
These freq u en cies are representative o f S tirto n 9 7 9 8 and are n o t in clu d ed in the averages because this sp ecim en d eviates su b stan tial ly from the other specim ens.
freq u en cy; 2:1 = K im ber's m odel where 2 gen om es are m ore closely related to one another than to the third gen om e; 0 -2 = Jack son 's m o d el w here 0 to 2 chiasm ata per bivalent are form ed ; I = univalents; IIC = r o d bivalent; IIR = rin g bivalent; III = trivalent; SS = average sum o f squares o f differen ces b etw een observed and ex p ected frequencies; X = value in d icatin g the relative distance b etw een the tw o h o m o lo g o u s gen om es and the third g en om e according to K im ber's m odels; C = ch iasm a freq u en cy per h a lf bivalent.TABLE 4 .-Com parison betw een the average sum o f squares b etw een the observed and exp ected values for chrom osom e configuration in triploids for different numbers o related to one another than to the fourth genom e; 2:2 = K im ber's m odel where 2 genom es are more closely related to one another than to any o f the other tw o genom es, which are also related to one another; 2:1:1 = Kim ber's m odel where 2 genome's are more closely related to one another than to the third genom e and the third and fourth genom es are n ot closely related; 0 -2 = Jackson's m odel where 0 to 2 chiasmata per bivalent are partially random ly form ed; 0 -2 R = Jackson's m odel where 0 to 2 chias mata per bivaJent are randomly formed; I = univalents; 11C = rod bivalent; IIR =ring bivalent; III = trivalen t; 1VC = r o d quadrivalents; IVR =ring quadrivalents; SS = average sum o f squares o f differences betw een observed and ex p ected frequencies; X = value indicating the relative distance between the different genom es according to K im ber's m odels; C = chiasm a frequency per h alf bivalent.
freq u en cy; 4 :0 = K im ber's m odel w here all 4 g en om es are h o m o lo g o u s; 3:1 = K im ber's m od el w here 3 g en om es are m ore clo sely related to on e another than to the fourth gen om e; 2 :2 = K im ber's m o d el where 2 gen om es are m ore closely related to one another than to any o f the other tw o gen om es, w hich are also related to one another; 2 :1 :1 = K im ber's m o d el where 2 g en o m es are m ore clo sely related to one another than to the third g en om e and the third and fourth gen om es are n ot clo sely related; 0 -2 = Jack son 's m o d el where 0 to 2 chiasm ata per bivalent are partially random ly form ed; 0 -2 R = Jack son 's m od el where 0 to 2 chias m ata per bivalent are random ly form ed; I = univalents; IIC = rod bivalent; IIR = rin g bivalent; III = tr iv a len t; IVC = r o d quadrivalents; IVR = rin g quadrivalents; SS = average sum o f squares o f d ifferen ces b etw een observed and ex p ected freq u en cies; X = value indicating the relative distance b etw een the d ifferen t g en om es according to K im ber's m od els; C = ch iasm a freq u en cy per h alf b iv a len t FIGURE 3.-Specimens o f A, Rubus cuneifolius B (Stirton 9861)\ B ,R .longepedicellatus (Stirton 8135).

TABLE 1 . -List of morphological character values allocated to different Rubus specimens (con tin u ed ) Bothalia 17,1 (1987) 109 E o o a. c O to 6986 uouus 1986 uouiiS 9986 uouus P986 uouus £986 uouus Z986 u ouus 0986 uouus $$86 uouus 1086 uouus n o -o O « 66L6 u o u u s S £ | 2 8 6 1 6 u o u u s
) is usually serrate with a double serrate margin in exceptional cases.R. longepedicellatus and R. x proteus have acute or acumi nate leaf apices and serrulate, double serrate or serrate leaf margins.The stipules (14) vary from lanceolate/ triangular/falcate to flabellate in R. cuneifolius B, from needle/linear/filiform to occasionally lanceolate/triangu lar/falcate in R. longepedicellatus, with all these differ ent shapes being represented in R. x proteus.In contrast to the pinnate/palmate leaves on the floricanes of R. cuneifolius B, R. longepedicellatus has pinnate leaves and both forms occur in R. x proteus.These morpho logical data indicate that R. cuneifolius B and R. longe pedicellatus are morphologically separate species, and the intermediate nature of the R. x proteus specimens suggests a hybrid origin.

H em m s.n .) R. apetalus (H enderson & Gaum 6) R. apetalus (Wells 5 0 0 0 )
and Jackson & H auber( 1982).Only the two models with the lowest average sum of squares o f each method are shown in the table

lo n g ep ed icella tu s (H en derson & G aum 10 ) R u b u s sp. (H enderson G aum 2 4 )
table (continued) X p r o te u s (H en derson & Gaum 5 1 )R.transvaliensis X R .