Pollination ecology of Aloe divaricata, Berger (Liliaceae): an endemic plant species of south-west Madagascar

The reproductive biology and pollination ecology of an endemic plant species (Afoe d;varicata) of Madagascar was studied in order to understand the re lative significance of various Hower visitors to fruit set success. The flowers were hermaphrodite and protandrous, and were primarily visited by souimanga sunbirds (Nectarinia souimanga) and stingless bees (Trigona sp.). Flower development and phenology were examined. Fruit set success at flowers from which sunbirds were excluded was significantly lower than that of flowers visited by both bees and sunbirds. Sunbirds (Nectarinia souimanga) appeared to be the most effective pOllinators ot Aloe divaricata, although numerous stingless bees (Trigona sp.) also visited the flowers. The latter visitors did nat seem to play any role in Aloe divaricata pallination.

Informatio n on interactions between flowers and the vectors promoting their pollination is essential (0 an understanding of plant reproductive success (Willson 1983;Dafni 1992;Kearns & Inouye 1993). Flower attractiveness and the effectiveness of vis itors play key roles in these interactions. Flower responses to visitors differ from flower to flower and depend on floral developmental phenology (Dafni 1992;Kearns & Inouye 1993;Guition el al. 1994).
Madagascar, with its high degree of plant diversity and endemism (Guillaumet 1984;Jenkins 1987). is of particular interest for such studies, not only because of the liltle-s tudied. unique flora, bu t also because of the widespread destruction of the natural commu nities (Minenneier 1988;Green & Sussman 1990). Many species have already been classified as rare, threatened, or endangered (see Jenkins 1987), including Aloe divaricala Berger (Uliaceae). which is one of the 46 Aloe species of Madagascar (Reynolds 1966). An understanding of flower biology and the role of flower visitors in pollination may be necessary in order to design conservation and management strategies for these species (Bawa & Krugman 1991). The purposes oflhis study were (I) to describe flower development and phenology, and (2) detennine the relative significance of pollinators to fruit set success in A.

divaricala.
A. divaricata is widely distributed in south-western and southern Madagascar (Reynolds 1966), and has important ornamental and medicinal values (see Jenkins 1987). It is found mostly in arid sandy shrublands, it propagates vegetatively by rhizomes, usually fOnTIS clumps and call reach 3 m in height ( Figure 1).

Study s ite
The study was undertaken between January and April 1994 in a pastured forest, adjacent to the protected Reserve of Beza Mahafaly (Figure 2, ca. 23"30'5 and 44"40'E), 35 km north-east of Betioky-Sud in south-western Madagascar (see Richard el ai. 1987;Andriamampianina 1992) . The area encompasses a gradient of vegetation: a gallery forest dominated by Tamarindus indica along the Sakamcna River gives way to xerophytic, spiny vegetation wilh increasing distance from the water course (Sussman & Rakotozafy 1994). A. divaricata is densely populated in the study area. Mean annual rainfall is 550 mm, almosl all of it Figure 1 The study organism: Aloe divaricata.

Methods
Stigma receptivity was estimated by hand crossing 120 Dowers at three diffe rent stigma developmental stages (40 flowers in each stage) and determining fruit set sue'cess (see Thomson & Harett 1981 ;Rertin 1982). Fruit set was defined when toe fruit attained about 1.5 em in length. The anthers in each targeted flower were removed prior 10 their ,mthcslS to prevent autogamy. Tbe inflorcsc.:cnccs were covered with mosquito netting after drawing a small brush loaded with pollen from another plant across the stigmas of the experimental flowers (see Ratsirarson & SHander 1995) . Developmental stages of tbe stigma were divided into three categories depending on the dryness and the colour of the stigma: (1) wet and greenish-yellow, (2) wet and whitish-yellow, and (3) dry and brownish-yellow.
Flower develo pment was examined in 80 flowers from eight different inflorescences (10 flowers in each) in March 1994. Flower buds about to open were marked and then visited every 2 hours. On each visit, the size of the corolla, and length of the filament and style were measured with a caliper. The c haracteristics and the duration of events during flower development (e.g. anther dehiscence, stigma receptivity, nectar availability etc.) we re n!cordt.!d . Nectar volume was quantified with micropipeltes and the sugar concentration with a hand-held refractometer (see Dafni 1992; RalSirarson & Silander 1995). Tbe antber was considered to have dehisced when pollen grains were seen around it; tbe stigma was receptive when its tip had a Whitish-yellow colour and was wet (see below), and wilted when it shortened and became brownish.
Stingless bees (Trigona sp.; Apidae) and the souimanga sunbirds (Neclarinia souimanga; Nectariniidae) were the main flower visitors. An exclusion experiment was used to determine the relative sig-nifIcance of these visitors to pollination. Ten inflorescences were bagged for each of the follow ing experimental conditions: (1) complete exclusion, in which no visitor was allowed access to tbe flower, permitting an assessment of self-compatibility, (2) bird exclusion, by covering the whole plant with I-em -gauge chicken wire, thereby excluding birds , while allowing access to the fl owers by various insects (see Ford 1979; Wright 1994); and (3) control fl owers, allowing botb sunbirds and bees access to the flower. Analyses of variance were used to compare the percentage of fruit set success among treatmen ts.

Results and Discussion
The peak flower production in A . divaricala is bctween February and April (see rigure 4). An individual pl an t produces one inflo rescence in general (rarely two) at the apex of the plant, which bears about 850 tubular red flowers per inflorcscence (mean -= 852. s.d. ::; 375, n ::; 18). rruits with winged seeds mature, begin- ing in March (see Figure 4) and fruits and flowers can be seen simultaneously in one inflorescence ( Figure 3). An individual inflorescence produces mature flowers over a 2-to 3-wcek period (mean;;;; 20 days, S.d. ;;;; 7.14, n = 18). The flowers arc about 2.5 em long (mean = 2.63 ± l.55. n = 80). and mature o ne after the other from the bottom of the inflorescence upward s. each one lasting for less than 16 hours (see Figure 5).
A. divaricafa has a perfect hermaphroditic, prolandrous flower. The whilish-yellow filaments are elongated (Figure 6), and the anthers produce pollen 2 hou rs after the flower opens ( Figure 5b). When the flowcr is open, nectar is already present (Figure 5e). Covered flowers had 15 ± 7.5 ~1 (n ; 30) of neelar 1 hour after opening, prior to anthesis. Pollen shedding lastec abou t 5 hours (Figure 5b), followed by a colour change in the anthers and a shortening of the filament ( Figure 6).
The style, located in the centcr of the flowe r, continued to elongate ( Figure 6) after the anthers had become brown and the filaments had shortened. In Aloe divaricala the stigma wa s receptive for less than 10 hours after flower opening (Figure 5c). In contrast, stigma receptivity has been reported 10 be between 8 and 24 hours in A. ferox (Hoffman 1988) and up to 30 hours in Gasleria (Martinie el al. 1993). Fruit set success in A. divaricala was higher when the stigma was wet and had a whitish tip than 181.86; df, ; 2, df,; 11 7; P < 0.01). This whilish colour stage is most likely the optimal state of stigma receptivity. The green and brown colour of the sti gma appeared to be an earlier and later, subop timal stage of recept ivity, respectively.
Individual flowers last about 16 hours, whereafter wi lting of the pelals and sepals occurs. Pollen shedding and stigma rccep~ livity are not well separated within a nower or within the inflor~ escence. Fruit production of completely covered inflorescences indicated that self poll ination docs occ ur and fl ow ers arc selfcompatible although frui t set occurs a t low rates (see Table 1). This fin ding contrasts with Aloe ferox which has been reported to be self-ineompalible (Hoffman 1988).
Flowers are visited mainly by sunbirds and bees, with nectar a nd pollen as rewards. The bird exclusion experiment showed sign ifican tl y lower fru it sci success tha n the flowers visited by bolh sunbirds and bees (ANOYA, f; 14.80; df,; I, df,; 18; P < 0.01; Table 1). Sunbirds (Nectarinia souimanga; Nectariniidae) therefore appear to be the primary pollinators . The birds inverted the stiff tubular flowers of Aloe divaricala with their beaks and collected the nec tar produced without damaging the flowers. Pollen was deposited at the base of the sunbird's beak and trans~ fered to the next flowers visited. Although it was not examined in this study. it has been suggested that sunbirds prefer Aloe flowers located in direct sunlight regardless of nectar quantity and quality (Goldstein el al. 19~7) . Sunbirds in general are reporled to have a bifid longue and feed o n insects and flow er nectar (Schla-moWilz. el a1. 1976;Langrand 1990). One other bird species, Neomixis lene/Ja (Sylv idae) was occasionally seen visiting A. divaricala fl owers and may playa min or role in its pollination.
Stingless bees (Trigona sp.) were attracted by the large amount of po llen produced by the flowers, but, frui t set success of fl owers from which birds were excluded was not significantly differ-   cnl to that of comple te ly covered Oowers (ANOVA, f = 0.007; dfl = I, df2 = IX; P > 0.05; Table 1). These bees did not seem to play any role in pollinat ion of A. divaricata and were mostl y seen co llec ti ng pollen 011 male fl owers in anthcsis as soo n as the anthers were exerted beyo nd the floral tube, but fewe r bees visited receptive female flowers.

Igi
Other insccts, such as anls (Forrnicidae). were also seen collecting nectar inside the tubular flowers but they did not seem to be effective poll inators. because they constantly cleaned and groomed. remuving pollen from their bodies. Ants in general secrete substances on their body surfaces that may inhibit pollen germinat ion (13uckley 1982;Pcakall et al. 1991). Primates did not seem to play any role in the pollination of A. divaricala. l.elll ltr calta (Lcm uridae) was observed eating whole innorescc! nccs, but no other lemur species were seen feedi ng on flowers.
I conclude that sun birds (Nectarinia souimanga) playa significa nt role in the po llination of A. divaricQla. Trigona bee visits do 110t have any role in its pollination. The morphology of the nower. includin g the tubular fonn, red colour and the amount of nec tar secreted during the daytime resembles that of genera l hummingbird (Trochilidae) o r sunbird (Neclariniicfae) pollination syndromes (Gill & Wolf 1975;Berti n 198 2;Cruden eI ai. 1983;Dafni 1992;Kearns & Inouye 1993). The dependence of N. souimanga on Aloe divaricala resources is unknown, allhough the poll ina ti o n of this plant species appears to be largely dependent on thi s s un bird. The management and conservati on of the N. souimanga population is as important as the management of A.
divaricata. The fitncss of the plant is indirectly coupled with the presence of surroun ding forests for N. souimanga habitat. Destruction of natural habitat in the south-west of Madagascar wi ll reduce the densi ti es of pote nt ial pollinators, such as N. sou;mallga. and inevitably affect A. divaricafa abundance. Protec ti on of much of the habitat surrounding the protccted Reserve area is the first priority. The procedure for extension of the Beza Mah afaly Special Reserve is under way and will inc lude most of the fo rested area adjacent to the Reserve .

Acknowledgments
I thank the Direction des Eaux ct forets-Ministere des Eaux et Forbes and the School of Agronomy at the University of Antananarivo in Madagascar for granting permission to work in the special reserve of Deza Mahafaly. Special thanks are due to Manjagasy, A. Rajerison , P. Rakotomanga, H.1. Ratsirarson and E. Razanajaonarivaly fo r their invaluable help and assistance in Madagascar. I also thank A .F. Richard and JA Silander, Uun.) not only for the ir useful suggestions and comments on earlier versions of the manuscrip t but also their continuous encouragement and support. I acknowledge th e help in many different ways fro m R. Dewar, S. O'Connor and M. Schwartz . Numerous constructive criticisms and suggestions came from manuscript reviewers. This project was finane-ially supported by Liz C laiborne/Art Ortenberg Foundatio n.