Leaf anatomy of the genus Ehrharta (Poaceae) in southern Africa: the Setacea group

The leaf blade anatomy of the taxa of the Setacea group of species o f the genus Ehrharta is described and illustrated. This group includes E. rupestris Nees ex Trin. subsp. rupestris, subsp. tricostata (Stapf) Gibbs Russell and subsp. dodii (Stapf) Gibbs Russell, as well as E. setacea Nees subsp. setacea, subsp. scabra (Stapf) Gibbs Russell, subsp. uniflora (Burch, ex Stapf) Gibbs Russell and subsp. disticha Gibbs Russell. All these taxa share a very characteristic leaf anatomy with inrolled or infolded leaves without keels and have adaxial ribs with interlocking prickles. The chlorenchyma is dense and compact with inwardly projecting invaginations visible in all taxa except E. setacea subsp. setacea. In £ . setacea subsp. scabra typical arm cells are present. Abaxial costal and intercostal zones are not differentiated and stomata are absent. The long cells are hexagonal or inflated with sinuous walls. Silica bodies are single or paired and rounded in shape. Small hook-like prickles with short barbs are common. Microhairs with a short, truncated distal cell occur. This leaf anatomical structure differs considerably from that o f the other species groups recognized in African Ehrharta and the Setacea group appears to be more distinct from the other groups than they are from each other.


INTRODUCTION
The African species of Ehrharta fall into two natural groups: those with the first sterile lemma similar to the second, and the fertile lemma differing from both; and those with the first sterile lemma reduced and scale-like, and the fertile lemma similar to the second sterile lemma (Gibbs Russell 1984). About 20 species belong to the first group, but the second group, to which the Setacea group belongs, contains only two species: E. rupestris Nees ex Trin. and E. setacea Nees, together with seven subspecific taxa. These two species, therefore, differ considerably in spikelet structure from the other species groups in the genus and, morphologically, the Setacea group appears to be rather isolated from the rest of the genus. This paper will examine whether the leaf anatomy of the representatives of this group confirms this obser vation, and will discuss the affinities of this group by reference to the anatomical evidence. Chippindall (1955) recognized five species with a short sterile lemma but Gibbs Russell (1984Russell ( , 1987 has combined these into only two species, each with several subspecies. Within each species there are parallel trends for reduced size and spikelet number, for rolled, folded or flat, erect or spreading leaf blades and for suffrutescence. This apparently reticulate pattern of relationships, which generally applies to the genus Ehrharta, results in parallel or convergent trends in vegetative parts occur ring in different taxa recognized on the basis of spikelet differences. It is of interest to determine whether the leaf anatomical variation parallels that exhibited by the spike lets and whether it substantiates the morphological clas sification based on relative glume length. This paper will copiously illustrate, describe briefly and evalute the leaf anatomy of each of the seven subspecies of the two spe cies of this group with reference to the systematic posi tion of these taxa and of the group as a whole relative to the genus and the higher classification of the Poaceae.

E. rupestris Nees ex Trin. subsp. rupestris Transverse section
The median vascular bundle is indistinguishable struc turally from the lateral first order bundles (Figure 1.1), as is the case throughout this group. Smaller third order bundles occur on either side of the median bundle and near the margin (Figure 1.1; 1.3) and, perhaps, second order bundles should be distinguished in tnis taxon. However, if these irregular and ill-defined additional bundles are excluded then only two smaller bundles sep arate consecutive first order bundles, as is the case in all other taxa in this group. The presence of these additional bundles is not considered to be taxonomically significant and probably only represents individual variation in the specimen studied.
The chlorenchyma tissue is compact, consisting of small round cells with dense chloroplasts (Figure 1.2). These cells do not have straight, isodiametric sides but do have slight, but clearly discernible wall invaginations (Figure 1.2; 1.3). According to the generally accepted definition of arm cells, these must be considered as such and, in fact, they do closely resemble the arm cells of the Oryzeae and some of the other bambusoid grasses. These invaginations are better developed in subsp. rupestris than in either subsp. tricostata or subsp. dodii and a definite reduction series is evident in these three subspe cies.

Abaxial epidermis
The epidermis conforms very closely to the type diag nostic for this group, with the costal and intercostal zones being structurally undifferentiated and with the long cells being very short and inflated to hexagonal with sinuous walls (Figure 1 Silica bodies are structurally identical throughout the epidermis and appear as single, rounded bodies located between the long cells (Figures 1.4;1.5). Under the SEM these are clearly seen to be raised, rounded struc tures covered with epicuticular wax (Figure 2.3; 2.4). As such they are indistinguishable from papillae and do not resemble the usual silica bodies with their concave faces devoid of wax platelets. Their refractive nature is visible with the light microscope (Figure 1.5) and they can also be seen to be raised (Figure 1.6). They are undoubtedly silica bodies but they are highly unusual and their exact nature and structure remains somewhat uncertain. Com parable structures have not been described elsewhere in the literature and they deserve further study. Similar structures also occur in E. setacea and they appear to be a unique characteristic of this group of Ehrharta species.
The entire abaxial epidermis is densely covered in shortly barbed hooks occupying the same locations be tween the long cells as the silica bodies (Figure 1.6). These hooks can occupy up to 50% of the available loca lities (Figure 1

Discussion
A small and inadequate sample was studied anatomi cally and, consequently, little information on variation in this subspecies can be deduced. However, the specimen from the Van Rhynsdorp District in Succulent Karoo comes from an atypical habitat and is particularly robust (Gibbs Russell 1984) and yet closely resembles the spec imen from the Mountain Fynbos in leaf anatomy. It, therefore appears as if the leaf anatomy of this taxon is stable and consistent despite differing environmental conditions.
E. rupestris subsp. rupestris is obviously very closely related to subsp. tricostata and the two can hardly be separated anatomically and, with a larger study sample, can be expected to grade into one another. The crosssectional anatomy tends to support the diagnostic differ ence of folded (  3). This morphological difference is, therefore, not very convincing when it is analysed struc turally. Further anatomical studies are required to con firm whether these two taxa are anatomically distinct; all indications are that only infraspecific separation is justi fied. This study, therefore, supports the decision of Gibbs Russell (1984Russell ( , 1987 to reclassify E. tricostata as a subspecies of E. rupestris. It must also be noted that these two subspecies show greater similarities with each other than either does to subsp. dodii. E. rupestris Nees ex Trin. subsp. tricostata (Stapf) Gibbs Russell

Transverse section
No keels or midribs are present (Figure 3.1; 3.4) and successive first order vascular bundles are separated by two third order bundles (Figure 3.2; 3.5). The outline of the lamina is very similar to that of subsp. rupestris and these two taxa differ only in subsp. rupestris having considerably wider leaf blades which are infolded but with the arms of the lamina curved outwards. In subsp. tricostata the leaf blades are narrower and setaceous with no angle being present at the median bundle (Figure 3.1; 3.4) and these leaves can be considered as being inrolled although the margins do not overlap as in subsp. dodii. This distinction between subsp. tricostata and subsp. rupestris is only minor and intermediates can be ex pected when a larger sample is studied. Both these subspecies also possess a ribbed adaxial surface with interlocking prickles overlying the furrows (Figure 3.3).
The mesophyll is compact, of small regular cells. The cells are rounded and very tightly packed (Figure 3.3) and closer inspection reveals very slight inward projec tions on only a few cell walls (Figure 3.3; 3.6). These cells resemble arm cells but cannot be considered to be typical and may result from reduction of the type of invagination described in subsp. rupestris.

Abaxial epidermis
The epidermal structure is virtually identical to that of subsp. rupestris as illustrated in

Discussion
Anatomically subsp. tricostata is very similar indeed to subsp. rupestris and these two taxa cannot be satisfac torily distinguished on anatomical criteria. This is sup portive of their separation at only the subspecies level (Gibbs Russell 1984) and the leaf anatomy confirms the extremely close relationship of these two taxa. The re cognition of two separate species is not corroborated by the anatomical evidence.

Transverse section
No keel is present ( Figure 6.1; 6.3) and the adaxial ribs and furrows are not well developed and interlocking prickles are absent ( Figure 6.2). Of particular interest is the convolute inrolling of the leaf blade (Figure 6.1; 6.3). This type of inrolling does not occur in either subsp. rupestris or subsp. tricostata and appears to be a significant anatomical difference between these two subspecies and subsp. dodii.
The mesophyll tissue is compact, of small rounded and regular cells ( Figure 6.2). These cells cannot be regarded as being typical arm cells as they possess only very slight invaginations.

Abaxial epidermis
The costal and intercostal zones are structurally undif ferentiated although their relative positions can be deter mined due to the staining of the sclerenchyma girders underlying the costal zones ( Figure 6.4). The long cells are considerably longer and less inflated than in either of the other two subspecies of E. rupestris. All three subspecies do have sinuous long cell walls and stomata are either very rare (subsp. rupestris) or absent from the abaxial surface. The silica bodies are also rounded but they are closely associated with a cork cell. Although these structures were not examined ultrastructurally, there can be little doubt that they are typical graminoid silica bodies and not the unusual type described in subsp. tricostata and subsp. rupestris. Prickles with prominent oval bases and very short barbs are common ( Figure  6.4). Small microhairs, typical of those of the Setacea group, are also present ( Figure 6.4).

Discussion
In its leaf anatomy E. rupestris subsp. dodii differs more from subsp. rupestris and subsp. tricostata than these two subspecies differ from one another. The find ings of this study, based on only a very small sample, appear to indicate that subsp. dodii is sufficiently distinct from E. rupestris to warrant separate specific status. Additional anatomical material may reveal anatomical intermediates with subsp. tricostata but this appears rather unlikely in view of the entirely different nature of the involution of the blades of these two taxa. The con volute type of inrolling in subsp. dodii may indicate a different type of bud vernation to that of subsp. trico stata where inrolling is of the involute type. If this proves to be true then the case for separation at the spec ies level will be strengthened. Although subsp. dodii differs somewhat from subsp. rupestris and subsp. tricostata in leaf anatomy, their anatomy is still typical of the Setacea group and different from that of all the others of Ehrharta (Gibbs Russell & Ellis 1987). Subsp. dodii, therefore, unquestionably be longs to this very distinct group.
The less acicular type of leaf transection and the elongation of the epidermal long cells seems to be paral lelled in E. setacea in subsp. disticha and subsp. uni flora. This trend may be related to the milder climate of the Cape Peninsula and Caledon District to which these three taxa are restricted. These three subspecies do not occur at higher altitudes in the mountain ranges further inland as do E. rupestris subsp. rupestris and subsp. tricostata and E. setacea subsp. setacea and subsp. sca bra. This trend, associated with a more mesophytic type of anatomy and elongated epidermal cells with costal and intercostal zones and with abaxial stomata, may repre sent the convergent trends so evident in this group. Alternatively this pattern may reflect the reticulate nature of relationships which also characterizes this group. It is therefore an academic decision whether subsp. dodii should be recognized as a separate species or not. What ever basis is used to define species in this group, other characters will not necessarily follow the same pattern. The decision is also dependant on a similar decision being made for subsp. disticha and subsp. uniflora in E. setacea.

E. setacea Nees subsp. setacea Transverse section
This taxon has the leaf outline which typifies this group: the blades are setaceous, without midribs or keels ( This chlorenchyma is identical in appearance to the type so common in the Danthonieae in genera such as Merxmuellera, particularly M. disticha (Schrad.) Nees (Ellis 1980), Pseudopentameris (Ellis 1985a) and some species of Pentameris (Ellis 1985b,c).

Abaxial epidermis
Costal and intercostal zones are not differentiated (Figure 7.5; 7.7), the long cells are short and fusiform

Discussion
The leaf anatomy conforms closely to that typical of the Setacea group, and subsp. setacea resembles E. ru pestris subsp. rupestris and subsp. tricostata very clo sely in leaf anatomy. It differs mainly in the absence of abaxial prickles. Other slight differences, such as the chlorenchyma cell invaginations and the epidermal cells in tranverse section which are regular in size, due to the absence of prickle hair bases, are only differences of degree. Anatomically, therefore, subsp. setacea clearly belongs to the Setacea group. Whether it should be sepa rated from E. rupestris is debatable according to the anatomical evidence and E. setacea subsp. setacea has more in common with E. rupestris subsp. tricostata than it has with any of the other subspecies of E. setacea. The leaf anatomy appears to indicate that these two taxa are conspecific and the relative glume length, which is used to separate E. setacea from E. rupestris (Gibbs Russell 1984), results in an apparently artificial anatomical grouping. This observation is particularly relevant when the anatomy of subsp. setacea is compared with that of subsp. scabra, as these taxa exhibit significant anatomi cal differences which would justify at least specific sepa ration in other grass groups.
Once again the classification of taxa within the Seta cea group presents problems with contradictory anatomi cal and morphological evidence. A possible explanation for this situation lies in different rates of evolution in different organs combined with a reticulate evolutionary pattern. Present knowledge indicates that, whatever characters are used to distinguish taxa in this group, heterogeneity of other character sets will result. This makes a generally acceptable classification almost im possible to achieve.
The anatomical resemblance between subsp. setacea (and E. rupestris subsp. tricostata as well) and many of the winter rainfall danthonoid grasses deserves further mention. Both in section and epidermis subsp. setacea closely resembles species of Pseudopentameris (Ellis 1985a), and Pentameris (Ellis 1985b,c) in particular. These similarities may only be coincidental but it is more probable that they result from convergent evolution. Convergent and parallel patterns appear to be common in Cape fynbos grass species. Most species of the mountain regions, where nutrients are severely limiting, have evolved a structurally complex, long-lived leaf, in con trast to the lowland species in which the leaves are decid uous (or the plants annual) with a much simpler internal structure, with thinner cell walls, less sclerenchyma and larger air spaces. This general pattern cuts across phylo genetic lines at the generic level and does not appear to be taxonomically significant. What is of importance, however, is that here taxa from different subfamilies have responded identically to similar environmental con ditions.

E. setacea Nees subsp. scabra (Stapf) Gibbs Russell Transverse section
This subspecies has a wide, expanded type of lamina (Figure 8.1; 8.3) which is not of the setaceous type of subsp. setacea. In all specimens, even those in which the blade is almost flat, there are indications of convolute infolding (Figures 8.1; 8.3). This is reminiscent of the condition in E. rupestris subsp. dodii. Slight adaxial ribs and wide, open furrows (Figures 8.2; 8.4; 8.6) resemble the subsp. dodii condition more than subsp. setacea.
The single most significant anatomical feature of subsp. scabra is the possession of definite arm cells in the mesophyll (Figures 8.2; 8.4; 8.5; 8.6). These are typical arm cells with distinct invaginations of the cell walls visible in all chlorenchyma cells (Figures 8.2; 8.4) and the chloroplast pattern is also indented, indicating the presence of these inward projections of the cell walls (Figures 8.5; 8.6). Typical arm cells are, however, sometimes lacking in subsp. scabra (Ellis 4653). Arm cells do not occur in subsp. setacea but indications of their existence are evident in all other taxa of the Setacea group.

Abaxial epidermis
The structure of the epidermis is very similar to that of E. rupestris. Absence of zonation, short fusiform long cells, no stomata, rounded silica bodies and hook-like prickles (Figure 9.1-9.4) occur in subsp. scabra and characterize the taxa of the Setacea group. Ultrastructurally this similarity is confirmed by the raised papilla like silica bodies (Figure 12.3) and short microhairs with rounded distal cells (Figure 12.4). This epidermal struc ture differs somewhat from that of subsp. setacea and subsp. uniflora but is virtually identical to that of E. rupestris subsp. rupestris and subsp. tricostata.

Discussion
Arm cells are considered to be a very important cha racter in the systematics of the Poaceae and the presence of these cells in subsp. scabra may be taxonomically meaningful. Arm cells are a diagnostic characteristic of   et al. 1985). This occurrence of arm cells in these apparently unrelated groups is im possible to explain but indicates that the presence of arm cells in the Setacea group of Ehrharta does not necessar ily imply bambusoid relationships for the genus. How ever, their presence can undoubtedly be used to substan tiate the arguments of those workers (Renvoize 1986) in favour of bambusoid affinities for Ehrharta. However, the occurrence of fully developed arm cells in only one taxon of the genus Ehrharta urges caution in attaching too much importance to their presence.
A certain degree of anatomical variation was observed in this subspecies. This variation applies particularly to the presence of stomatal bands in the abaxial epidermis ( Figure 10.3) with an associated development of clear structural differentiation between the costal and intercos tal zones. The specimen displaying this type of structure (Ellis 4653) was collected growing together with Ellis 4652, which has the anatomy typical of subsp. scabra, whereas Ellis 4653 is a much more robust plant with an extremely dense underground rhizome system. The ana tomy of this specimen is illustrated in Figure 10 and it can be seen to differ from typical subsp. scabra in the presence of abaxial stomata, the epidermal zonation and by having elongated intercostal long cells (Figure 10.3) and by the absence of typical arm cells, with the invagi nations being poorly developed (Figure 10.2). The mi crohairs are also more elongated (Figure 12.6) and there are fewer prickles (Figure 12.5). The anatomy of this specimen is, therefore, intermediate between that of subsp. scabra and subsp. uniflora which has abaxial stomata, no prickles and elongated, tapering microhairs (Figure 12.7;12.8). A few isolated stomata were also observed on the epidermis of Ellis 4689, a rather robust specimen with wide leaves (Figure 8.1). However, in this specimen, the epidermal zones are not structurally differentiated (Figure 9.1; 9.2) and the arm cells are clearly developed (Figure 8.2). It is possible that a whole range of intermediates will be found between E. setacea subsp. scabra and subsp. uniflora and disticha. Unfortu nately the latter two subspecies are poorly represented in this study and it was not possible to determine the exact pattern of variation between these taxa.

Transverse section
This subspecies has a narrow but flat and expanded leaf outline (Figure 11.1) with poorly developed ribs and furrows (Figure 11.3). This is not the setaceous type of leaf outline as in subsp. setacea, and abaxial stomata are clearly visible even in transection (Figure 11.3; 11.4). There is also no indication of convolute inrolling as in subsp. scabra.
The mesophyll of subsp. uniflora is also not as com pact as in all the other taxa of the Setacea group ( Figure  11.2; 11.4), a difference which appears to be associated with the development of an air-space system linked to the abaxial stomata. The mesophyll cells themselves cannot be considered to be classical arm cells ( Figure  11.2; 11.4) although very slight invaginations are visible in some of the chlorenchyma cell walls.

Abaxial epidermis
The presence of a single file of stomata on either side of each costal zone (Figure 11.5; 11.6) distinguishes the epidermis of this subspecies from all the other taxa of the Setacea group (except subsp. disticha). There is also a definite tendency for the costal zones to be structurally different from the intercostal zones (Figure 11.6) with the silica bodies being different. The intercostal long cells are not separated by raised silica bodies as in the other subspecies of E. setacea (Figure 12.1;12.3;12.5;12.7). In this respect subsp. uniflora differs from all other members of the Setacea group. Other differences are the tapering microhairs (Figure 12.8) and the pre sence of epicuticular wax (Figure 12.7).

Discussion
The leaf anatomy of subsp. uniflora appears to be intermediate between that of the Setacea group and some of the other groups of the genus Ehrharta. This conten tion is supported by the epidermal zonation with abaxial stomata, the presence of epicuticular wax, the absence of raised silica bodies, the tapering microhairs and the less compact mesophyll tissue. These features are all charac teristic of the Setacea group and subsp. uniflora differs from the other taxa in this group in all these attributes which tend toward the condition found in several of the other infrageneric groupings in Ehrharta.
The leaf anatomy of subsp. uniflora is very similar indeed to that of Ellis 4653, an atypical subsp. scabra specimen (Figure 10). Ellis 4653 is a much more robust plant with wider leaf blades than any subsp. uniflora specimen. This is not reflected in the leaf anatomy, how ever, and leaf thickness and epidermal cell size are com parable. Only the width of the lamina is different. It appears that the loss of the setaceous, acicular type of leaf results in a very similar anatomy, irrespective of which subspecies this occurs in. The development of abaxial stomata appears to be the most significant change and it is associated with several other anatomical features such as the zonation of the epidermis and the presence of an extensive air-space system accompanied by less com pact mesophyll. The loss of the Setacea type of silica body and the possession of tapering microhairs is not explained by the development of abaxial stomata, how ever, and they appear to be additional differences not linked to the presence of these stomata.
Subsp. uniflora grows at very low altitudes, from 10-500m (Gibbs Russell 1984) and this is a possible reason for the development of these abaxial stomata. Temperatures are less extreme at these lower altitudes with an equable maritime climate and the nutritional sta tus of the soil is considerably better than the leached soils at higher altitudes. More material of subsp. uniflora is required to gain a better understanding of the inter relationships of these taxa.

E. setacea Nees subsp. disticha Gibbs Russell Transverse section
No fresh material of this taxon was available for ana tomical study and, consequently, the transectional ana tomy was not examined.

Abaxial epidermis
The epidermal structure is very similar to that de scribed for subsp. uniflora, with stomatal files and clearly differentiated costal and intercostal zones (Fig  ures 13.1; 13.2). Intercostal hooks are absent and the intercostal silica bodies appear to be of the raised type typical of the Setacea group but they are associated with cork cells (Figure 13.2). Microhairs are common and appear to be longer than the type usually found in this group (Figure 13.2).

Discussion
The epidermal anatomy of subsp. disticha is virtually identical to that of subsp. uniflora and it is predicted that the leaf anatomy of these two taxa will also be very similar. This supports their classification as subspecies of the same species (Gibbs Russell 1984). Freshly fixed material for anatomical study is required to confirm the prediction.

DISCUSSION
The leaf anatomy of several species of Ehrharta has been studied by various authors (Metcalfe 1960;Jacques Felix 1962;Renvoize 1985) but none of these studies, except those of Tateoka (1963) dealt with any represen tatives of the Setacea group. Tateoka (1963) included four species of this group but did not notice any arm cells. Consequently, the unique anatomy of this most interesting group has received very little attention. If the presence of these arm cells had been reported earlier, it is certain that it would have had a profound influence on the classification of the genus in the Poaceae. On the evidence of spikelet morphology and embryo anatomy (Reeder 1957(Reeder , 1962 Ehrharta has been placed in the Oryzeae (Stebbins & Crampton 1961). However, the leaf blade anatomy of the representatives studied to date showed no affinity with the Bambusoideae (Tateoka 1963). The genus has therefore been retained in the Arundinoideae, with which it conforms in leaf anatomy (Renvoize 1981). Awareness of the arm cells in the Seta cea group would undoubtedly have strengthened the ar gument for retention of the genus in the Bambusoideae.
For these reasons it is regrettable that the detailed and comprehensive study of the genus by Engelbrecht (1956) was never published. In this thesis, 26 species of Ehr harta were studied in detail and their structure inter preted with considerable insight. Unfortunately, this work has remained virtually unknown and has not been considered in the taxonomy of the genus to date. Engelbrecht (1956) observed wall invaginations in all members of the Setacea group except E. setacea subsp. scabra and subsp. disticha which were not examined by him. In addition, he reported the presence of arm cell like invaginations in E. rehmannii, E. ramosa, E. subspicata, E. gigantea and E. villosa. These latter ob servations are not substantiated by the present study (Gibbs Russell 1987) and will be discussed fully in a later paper. Engelbrecht (1956) used the possession of these arm cells to divide the genus into two distinct groups of species. The group with arm cells was further subdivided into those without dumbbell-shaped silica bodies (the Setacea group) and those with dumbbell shaped silica bodies (the Villosa and Ramosa groups). His circumscription of the Setacea group is, therefore, identical to the conclusions of the present study and it is  truly unfortunate that these valuable observations have had to wait over 30 years before being confirmed and published.
Tateoka (1963), on the other hand, also examined the leaf anatomy of several representatives of the Setacea group (E. dodii, E. rupestris, E. setacea and E. trico stata) but reported that 'the mesophyll cells do not seem to have an arm (a projection of cell membranes), as far as the present author can observe'. The fact that no Ehr harta species were seen to possess arm cells or complex midrib vasculature was used by Tateoka (1963) to con vincingly argue the case for the phylogenetic separation of the Ehrharteae and Oryzeae into different tribes and subfamilies. If he had been aware of the arm cells in the Setacea group, this reasoning would not have been nearly so sound and would probably never have been propounded.
The present paper proves beyond doubt, with substan tiating photographic evidence, that arm cells, compara ble to those in most Oryzeae, occur in some of the taxa of the Setacea group, E. setacea subsp. scabra in parti cular. These invaginations, together with chlorenchyma cell size, appear to become progressively reduced in most of the other taxa and are difficult to observe, even in freshly fixed, fully hydrated material. In reconstituted dried material the invaginations are apparently easier to see (Engelbrecht 1956) but may be overlooked as being artefacts due to dehydration and subsequent wall shrink age.
The Setacea group of species is obviously a closely related unit with very uniform leaf anatomy. The degree of modification of the arm cells which appears to have occurred in the different taxa of the group is, conse quently, most unexpected. Two entirely different types of modification appear to have taken place which deviate from the E. setacea subsp. scabra condition. E. setacea subsp. uniflora from lower altitudes with more equable climates and possibly higher levels of soil nutrients seems to have developed a diffuse mesophyll with an extensive air space system associated with the possession of abaxial stomata. No invaginations are evident in the chlorenchyma cell walls of taxa with this type of meso phyll and all intermediate stages can be distinguished. Taxa from the opposite type of habitat, from high alti tudes with extremes of temperature and a low level of soil nutrients, on the other hand, have a very different type of leaf anatomy which appears to be derived from the basic subsp. scabra anatomy. This type of anatomy is probably linked to the perennial nature of these leaves and is characterized by the mesophyll cells becoming increasingly reduced in size until they are small, straightwalled and isodiametric with a corresponding almost to tal reduction in the air space system. This anatomical type is always associated with a setaceous type of leaf and with the absence of stomata. A clear reduction series is evident in E. rupestris and these two very different mesophyll structures can possibly be interpreted as dif ferent derivations of the E. setacea subsp. scabra type of anatomy. It may be relevant to this argument that E. setacea subsp. scabra itself is limited to the east of the range of most of the taxa of the Setacea group where summer rainfall becomes a significant factor. It can be postulated that the loss of the arm cell invaginations may be an adaptation to winter rainfall.
These two different anatomical types in the Setacea group mirror the situation prevailing in many of the Danthonieae of the Cape Fynbos region. Taxa from high altitudes, such as Merxmuellera, Pseudopentameris and Pentameris (Ellis 1980(Ellis , 1985a, resemble the latter type, whereas species from the lowland Fynbos areas, such as some species of Plagiochloa and Lasiochloa, have a more diffuse type of mesophyll. It appears, there fore, that habitat has been more significant than phyloge netic relationships in the evolution of these anatomical features of the leaf blades in many Cape Fynbos grass taxa. The fact that the typical arm cell type in the mesophyll of E. setacea subsp. scabra appears to be modified so easily suggests that caution must be exercised in attach ing too much taxonomic significance to the presence of this single character. Watson et al. (1985) discuss the occurrence of arm cells in the family. They rightly con sider them to be a feature of high taxonomic value but point out that they can occur sporadically in distant, unrelated groups. No satisfactory explanation can be given for this type of distribution. The presence of arm cells in the Setacea group of Ehrharta does not neces sarily imply bambusoid affinities, particularly as they are not accompanied by fusoids and complex midrib vascu lature. Rather, arundinoid links are suggested by the type of mesophyll seen in E. setacea subsp. setacea (Tateoka 1963) and perhaps the arundinoid type of leaf anatomy could have been derived in a similar manner to that de scribed in the Setacea group of Ehrharta from a prim itive bambusoid-like ancestor possessing arm cells. The Setacea group, therefore, can be viewed as a primitive arundinoid type exhibiting some links to the peripheral groups of the bambusoids.

CONCLUSIONS
In both leaf anatomy and spikelet morphology the Se tacea group differs more from all the other groups of Ehrharta than any of these differ from each other. Apart from arm cells, many other anatomical characters sepa rate the Setacea group from the other groups (Gibbs Rus sell & Ellis 1987). Examples are microhair and silica body shape. Furthermore no anatomical intermediates were found linking this group to any of the other equiva lent groups of Ehrharta. These other groups, on the other hand, are sometimes linked by intermediate speci mens or taxa. This suggests that the Setacea group may deserve separate subgeneric or generic status but consid eration of the generic and infrageneric classification of Ehrharta will be dealt with in the last of this series of papers.