Gene Flow in a Facultative Apomictic Poacea, the Savanna Grass Hyparrhenia diplandra

Corresponding author: Michel Veuille, CNRS UMR 7625, CC 237, Laboratoire d'Ecologie, Université Pierre-et-Marie Curie, 75252 Paris Cedex 05, France., mveuille{at}snv.jussieu.fr (E-mail)

Communicating editor: M. K. UYENOYAMA

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 APPENDIX 2 LITERATURE CITED

The genetics of the poacea Hyparrhenia diplandra was studied in four natural populations from an ecological station in West Africa, where it makes up 80% of grasses from wet savanna and constitutes a dense continuum of randomly distributed individuals. DNA content and cytogenetical observations suggest it is an allotetraploid. Using two highly variable microsatellites (heterozygosity H = 0.615–0.616), we show that this species is an apomict with rare sexual reproduction events that account for 0.5% of seeds pollinated in the wild. Hexaploid individuals were also produced, corroborating the observation of aberrant genotypes in the wild. The spatial extent of asexual clones in the field was low in comparison with the predominance of apomixis, thus indicating a low dispersal of seeds from their parent. Heterozygosity and departure from Hardy-Weinberg predictions were similar in the four populations, revealing a high apparent selfing rate (s = 0.599) among sexually produced seeds. This is an overestimate since we could not distinguish true selfing from reciprocal outcrosses between neighboring individuals from the same apomictic clone. Gene flow by pollen could be substantial, possibly explaining the absence of isolation by distance in the studied area.

DESPITE the universality of sexual reproduction among angiosperms, a number of species have independently evolved apomixis, a reproductive system that can be defined as asexual reproduction through the production of seeds that are genetically identical to their parent (NOGLER 1984 ). Interest in apomixis has changed dramatically over the last few years. Until recently, the prevailing opinion was summarized by a frequently cited quote from DARLINGTON 1939 : "... with the loss of sexual recombination, the apomict, like the permanent hybrid, is cut off from the ultimate survival. Apomixis is an escape from sterility, but it is an escape into a blind alley of evolution" (cited, e.g., by ASKER and JERLING 1992 ). However, the use of genetic markers in the field has shown that many apomicts have retained substantial levels of genetic variation (ELLSTRAND and ROOSE 1987 ). In fact, apomixis is generally only partial, thus allowing genes to recombine and escape the long-term accumulation of deleterious mutations in surviving genotypes ("Muller's ratchet," reviewed, e.g., by MAYNARD-SMITH 1978 ). In the short term, apomixis seems to be a way of preserving heterozygotes produced by occasional outcrosses, while obligate sexual reproduction would generate a recurrent gamble with selfing and its associated fitness-depressing effects. Apomixis may be adaptive in several ways. It can be a way of retaining heterozygosity in species exposed to high rates of autogamy, of combining vegetative reproduction and seed dispersal, and of preserving locally adapted genotypes from recombination. A possible reason that apomictic plants are only facultatively so is that outcrossing periodically generates new genotypes.

Most apomictic species show no meiosis in ovules, whereas male meiosis occurs normally. Pollen is also produced because the fusion of a nucleus from the pollen with the polar nuclei of the embryo is required for the formation of the endosperm, the seed storage organ. Many apomictic species are polyploid, for reasons that are still unclear (ASKER and JERLING 1992 ).

Apomixis is known in 350 species belonging mainly to three families: asteraceae, rosaceae, and poaceae. Most published work on poaceae is devoted to plant breeding issues, since apomixis—which allows hybrids to breed true—is as valuable a trait in plant breeding as in population biology (VIELLE CALZADA et al. 1996 ). Population genetics data are more limited, and most pioneering studies, using allozymes and cytogenetical data, were general surveys over whole taxa with an emphasis on interspecific hybridization (DE WET 1967 ; PERNES et al. 1975 ; SAVIDAN and PERNES 1982 ). Further studies showed that most apomicts displayed substantial levels of variation (ELLSTRAND and ROOSE 1987 ; WIDEN et al. 1994 ). This diversity could be ascribed to interspecific gene flow for such obligate apomicts as the dandelion (MENKEN et al. 1995 ), but may be due to residual sexuality in facultative apomicts. Since their effective population size is not necessarily reduced (YONEZAWA 1997 ), their mode of reproduction may allow for substantial levels of variation.

Many apomicts studied previously, like Tripsacum (BARRE 1995 ) and Taraxacum (RICHARDS 1996 ), are colonizers of disturbed habitats. Our study reports apomixis in a species that constitutes a dominant and permanent population in a stable ecosystem. Hyparrhenia diplandra belongs to the supertribe Andropogoneae of the Poaceae. It is found in the intertropical zone of Africa, with introduced populations in Laos and Madagascar (HUTCHINSON and DALZIEL 1972 ). It makes up 80% of the grass biomass of wet savanna (MENAUT and CESAR 1979 ) and occurs as a continuous herbaceous cover, with scattered trees and bushes. Since the reproductive system of savanna grasses is unknown, our initial purpose was to document it and to study gene flow through pollen and seed dispersal under a standard isolation-by-distance model. However, as reported below, we found that H. diplandra was a facultative apomict, a trait that may be of crucial importance for the evolution of these populations.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 APPENDIX 2 LITERATURE CITED

Life history:
H. diplandra exhibits densely packed tufts of perennial individuals that produce seeds and pollen every year. Annual growth starts soon after the annual bush fire, at the change from the dry to the wet season (late January), and continues until October, at the next seasonal switch (MENAUT and CESAR 1979 ). Flowering lasts 10 days in late October. Seeds are mature in early December, disperse quickly, and germinate after the first rains of February. Our study took place in the Lamto ecological station in Ivory Coast (5°02' western longitude, 6°13' northern latitude), a state-protected reserve mostly comprised of large areas of native wet savanna, fringed by remains of gallery forest along the Bandama River. It has remained free from agricultural and herding activities since its foundation in the 1960s.

Random sampling:
Random sampling (December 1995 and October 1996) involved four populations hereafter referred to as P1–4. The P2 sample is 4 km distant from the other three, and P1 is 0.5 km distant from the remaining two (P3 and P4), which are very close to each other, being separated only by a road. There was no difference in the general aspect of these collection sites. They were chosen merely for being spatially dispersed. In each population, 30 individuals were randomly collected in an 5000 m² area by taking one plant at 15–20-m intervals (total sample, 4 x 30 individuals).

Structured sampling:
Structured sampling (December 1995) involved several groups of 5 plants. In each group, a central plant was associated with 4 other plants located at compass points on a 1-m-radius circle. Three such groups were collected in a row at 15-m intervals. This sampling (3 x 5 plants) was replicated in a parallel row, 15 m away. This sampling design (two rows of 3 groups of 5 plants) was used in populations P1 and P2 (total sample size, 2 x 30 individuals).

Progeny sampling:
Seeds were collected from plants of samples P1 through P4. These seeds originated from free pollination in the wild. They were germinated in the greenhouse. Their genotype was compared with that of their parent, and their DNA content per somatic nucleus was determined.

Density sampling:
We performed density sampling to collect information on distances between individuals. In five randomly chosen locations of populations P3 and P4, all individual tufts were noted in 100-m² grids (10 x 10 m) divided into 1-m² quadrats.

DNA preparation:
Leaf samples (length, 5–6 cm; dry weight, 10–20 mg) were collected at the base of the sheath. They were dried overnight at 65° immediately after sampling and kept for several months before use. Genomic DNA was extracted using DOYLE and DOYLE's (1987) DNA rapid preparation technique.

DNA content per nucleus:
The DNA content of all progenies was estimated by flow cytometry. We followed BROWN et al. 1991 protocol. A nuclei suspension was prepared from 50–100 mg of greenhouse-grown plantlets (1–2 mo old). Samples were chopped using a razor blade in Galbraith buffer containing 0.1% Triton X-100 and 10 mM metabisulfite. For total DNA estimates, the nuclei suspension was incubated for 20 min at 0° with 100 µg/ml RNAse and 50 pg/ml ethidium bromide. Internal standards were provided by leaf blade fragments from petunia and pea, which were chopped together with each Hyparrhenia sample. For ploidy level estimates, the Triton X-100 concentration was increased to 0.5%. The suspension was stained with either Hoechst 33342 or 4',6-diamidino-2-phenylindole (DAPI), and pea nuclei were used as an internal standard. About 2500 nuclei were counted for each measure. Two measures were routinely made for each individual. However, four to six measures involving two independent nuclei preparations were made for individuals with nonmodal DNA content. In this case, results were checked on a third nuclei preparation, using a known H. diplandra sample as an internal standard. Most of the progeny study (101 progeny from 25 parents) was done using 6 µg/ml Hoechst 33342 as a dye. The fluorescence of isolated nuclei was recorded using an EPICS cytometer (Beckmann-Coulter Co., Miami, FL) equipped with an argon laser beam. A smaller series (34 progeny from 4 parents) was stained with DAPI and studied using a Parsec cytometer.

Cytogenetical observation:
Root tips were cut from greenhouse-grown plants. They were incubated for 4 hr at 4° in a 0.5% saponine solution saturated with -bromo-naphthalene and fixed overnight in ethanol/acetic acid (3 v/1 v) in the dark at room temperature. After rehydration in distilled water, root tips were hydrolyzed for 15 min in 5N HCl at room temperature and rinsed twice in water. Nuclei were stained by incubation in Schiff's solution for between 45 and 60 min in the dark at room temperature. After rinsing in water, root tips were squashed in acetocarmine. Metaphases were observed under a microscope using a x100 immersion objective.

Microsatellites:
The cloning and characterization of microsatellites will be described elsewhere (L. GARNIER, I. DAJOZ and J. DURAND, unpublished results). We used two highly polymorphic loci (Table 1). Amplification of DNA was performed using either a TB1 (Biometra, Tampa, FL) or a PCT 100 (MJ Research, Watertown, MA) thermocycler. Each PCR reaction was conducted in 0.4 µM of each primer, 200 µM dCTP-dGTP-dTTP, 5 µM dATP, 0.050 µCi/µl [³⁵S]dATP, 50 mM KCl, 10 mM Tris-HCl pH 9.0, 1.5 mM MgCl₂, 0.01% Triton X-100, 0.01% gelatin, 1 unit Taq polymerase. The PCR cycle was 94° for 5 min; 35 cycles of 94° for 1 min, 56° or 60° for 1 min, 72° for 1 min; followed by a final extension for 5 min at 72°. Annealing was routinely carried out at 56° for locus Hd1 and at 60° for locus Hd2. However, these conditions were changed to 50° and 55°, respectively, when null alleles were suspected. Amplification products were denatured in 40% formamide for 5 min at 95° and run on a sequencing gel using an M13mp18 plasmid sequence as a molecular weight scale. They were then autoradiographed.

RAPDs:
Randomly amplified polymorphic DNA (RAPD) fragments were used to ascertain the relatedness of two individuals. Three 10-mers from Operon Technologies (Alameda, CA) primed the amplification of polymorphic bands in the 100–500 bp size range in H. diplandra: I₀₂ (GGAGGAGAGG), F₀₃ (CCTGATCACC), and F₀₄ (GGTGATCAGG). Reaction conditions were as for microsatellites, except that primers were 10 µM and [³⁵S]dATP was not used, the concentration of all dNTPs being set at 200 µM. The amplification cycle was as for microsatellites, except that annealing was carried out at 42°. Amplification products were run on 2% agarose minigels and banding was recorded using ethidium bromide under UV illumination.

Statistical analysis of data:
Most statistical analyses were conducted using the GENEPOP v.1.2 software (RAYMOND and ROUSSET 1995 ); F-statistics were computed using the DIPLOIDL program included in this package. Genetic diversity was calculated as H = , where N is the sample size and p_i is the frequency of the ith of n alleles. Genotypic diversity was calculated as G = - ^m_i=1X²_i, where X_i is the frequency of the ith of m genotypes.

Determining the selfing rate under various genetic hypotheses:
In a diploid population, the selfing rate s can be calculated from the deficit of heterozygotes as F = . This formula can be used for allotetraploids, which behave as functional diploids. However, we also considered the possibility that H. diplandra was an autotetraploid. In this case, there are five possible genotypes at a locus: two kinds of homozygotes and three kinds of heterozygotes, of which only three classes can be recorded, since heterozygotes are indistinguishable from each other and must be pooled. HALDANE 1930 showed that in a diallelic system, the frequency of [A]₄ homozygotes reaches the equilibrium value x = p - D, while the frequency of [a]₄ homozygotes reaches the equilibrium value y = q - D, where p and q are the frequencies of alleles A and a, and

We used two loci; therefore, we calculated the selfing rate numerically. First, we pooled all minor alleles and obtained three genotypic classes: homozygotes for the major allele (the [A] class), homozygotes for the minor alleles (the [a] class), and a mixture of both categories of alleles (the [A, a] class). Then, simulations were carried out for one locus. The probability P of observing X homozygotes [A] and Y homozygotes [a] was calculated as

We used P as a maximization criterion to calculate the likelihood of p and s values. These simulations confirmed the analytic result and showed that genotypic frequencies converge toward this equilibrium depending only on p and s. We then extended simulations to two loci by calculating P_A and P_B for each locus A and B and by maximizing the product P_A · P_B.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 APPENDIX 2 LITERATURE CITED

Population density:
Most plants occurred as dense individual tufts separated by cracked soil. The mean population density was very stable over samples (Table 2), with mean = 4.77 plants/m² and standard deviation = 2.91. This distribution was only slightly more dispersed than under a Poisson distribution, meaning that individuals were almost randomly scattered, with little heterogeneity, resulting probably from trees, shrubs, lower plants, and termite mounds, which are the main competitors for space in this area (MENAUT and CESAR 1979 ).

Ploidy status:
We counted 2n = 40 mitotic chromosomes in individuals having 3.6 pg DNA per nucleus. This is consistent with previous observations of 2n = 40 in H. diplandra from Ivory Coast (KAMMACHER et al. 1973 ) and Zaïre (DUJARDIN 1979 ). Dujardin reports several univalents and multivalents at meiotic diakinesis, noting that "meiosis is quite irregular in nearly all microsporocytes" and that "these meiotic irregularities are probably maintained owing to an apomictic mode of reproduction." Accessions with 2n = 20 have been reported in several Hyparrhenia species (GOLDBLATT and JOHNSON 1981 ). H. diplandra therefore appears as a tetraploid species. The occurrence of a majority of bivalents along with some multivalents and univalents at diakinesis suggests that it is an allotetraploid. In allotet raploids, chromosomes tend to segregate in a Mendelian fashion, while in autotetraploids they can pair with any of three homologs.

DNA content per nucleus:
This study involved 124 siblings from 27 families using pea nuclei as an internal standard. The ratio Hyparrhenia/pea provided a DNA content within the 3.4–3.8 pg range (median 3.6 pg per somatic nucleus) for 120 siblings. However, 4 siblings were in the range 4.7–5.1 pg. If we consider that modal H. diplandra individuals are tetraploid with 2n = 40, then these 4 individuals would be hexaploid with 3n = 60. This ratio was confirmed by analyzing these individuals again, using a modal H. diplandra as an internal standard. The genotype of the seeds was identical to that of their parent in terms of presence/absence of bands. Therefore these 4 individuals were probably derived from selfing, through the fertilization of an unreduced ovule by a reduced pollen.

Microsatellite variants:
Amplified DNA was obtained from all sampled individuals but one. We found six alleles at each locus (TABLE A1). They always differed by an even number of bases. Successive alleles differed by the basic 2-bp motif in two cases only. Most alleles were present in each population (TABLE A2), suggesting that mutation events are rare and would not interfere with the comparison of populations. For each locus, one allele (of size 115 for Hd1 and 152 for Hd2) was clearly predominant. Three bands were obtained four times in the population survey (twice for Hd1 and twice for Hd2). They may have been hexaploid, as observed for some individuals from the progeny survey (see above). Another possibility is that they happened to be trisomic for the chromosomes bearing the microsatellites, as suggested by the occurrence of atypical meioses in this species (DUJARDIN 1979 ). These hypotheses could not be checked, since their fresh tissues were no longer available for study.

Short distance population structuring:
In structured samples (TABLE A2), the genotype of each central individual was compared to that of 4 individuals evenly spaced at a 1-m radius. The same genotype occurred with mean frequency 0.264 (SD = 0.279), that is, in roughly 1 of the 4 surrounding individuals. In these groups of 5 individuals, one or two genotypes belonged to a clone (that is, to a genotype present in 2 or more individuals from the same group). All of the 12 groups involved at least one clone, 7 of them showing one clone, and 5 showing two clones. In all, 44 of the 59 surveyed individuals were identical to at least 1 other individual from their group (mean 3.66, SD = 0.850). This is an underestimate, given the small size of each group. An exhaustive sampling at a microgeographical scale would probably reveal that most genotypes are clonal. On average, the probability of two genotypes being similar among the 5 individuals was 0.316 (SD = 0.185), whereas the proportion of similar genotypes in the pooled random samples was 0.084 (the complement to 1 of genotype diversity). The next group of 5 individuals was 15 m distant. The proportion of similar genotypes between two adjacent groups was 0.040 (SD = 0.196) for 350 comparisons and was thus not different from the average value for random samples (0.084, SE = 0.049). This means that we can estimate allele and genotypic frequencies, and eventually inbreeding, from individuals sampled at 15 m from each other.

Genetic diversity and inbreeding:
Genetic diversity data under a diploid model are shown in Table 3. The four random samples showed similar genetic diversities (average H = 0.616, SE = 0.049 for locus Hd1, average H = 0.615, SE = 0.097 for locus Hd2). The number of genotypes thus generated was very high, with an average genotypic diversity G = 0.916 (SE = 0.049). As already noted, the four random samples departed from Hardy-Weinberg expectations under a panmictic diploid model. The average inbreeding coefficient was F_IS = 0.425 (SEM = 0.029). Since individuals from random samples were at least 15 m distant, this estimate should not be biased by apomixis. In a diploid population at equilibrium, the selfing rate is equal to s = . In this case, we obtain s = 0.599 (confidence interval calculated from the SEM of F_IS, C.I._0.95 = 0.535–0.653).

Selfing rates were also estimated under an autotetraploidy model. We pooled minor alleles to reduce allelic variation to two classes. The two resulting homozygote classes were of about equal frequency for each of the two loci (Table 4). Estimated selfing rates ranged from 0.86 to 0.93 with an average value of 0.884 (SEM = 0.027).

Long distance population structuring:
The four population samples were between 40 and 4000 m distant. Allele frequencies were significantly different between all samples. The F_ST ranged from 0.004 to 0.070, but there was no correlation with distance (Table 5).

Detecting sexual reproduction:
Seeds collected in the field from 31 plants yielded 360 offspring (the average number per family is 11; the range is 3–39). Most of them were cytogenetically similar to their parent. The genotypes of 358 offspring were obtained. In addition to the 4 3n individuals cited above (hexaploid individuals in reference to DNA content), we found 1 2n offspring (allotetraploid in reference to DNA content) lacking an allele from its parent. This individual was probably the product of sexual reproduction. However, the characterization of this case of sexuality rested on only two loci, and was thus dubious. This uncertainty could be resolved using the RAPD primers. We compared the banding pattern of the parent to 3 of its offspring (2 individuals identical to the parent for microsatellites, plus the unresolved case) and to that of 2 unrelated individuals used as controls. Duplicate PCR amplifications were carried out on the 6 individuals simultaneously. Results (Table 6) were identical in the two series. The three available primers amplified a total of 26 bands of which 20 were polymorphic. The parent and its 2 identical offspring showed the same pattern. They showed 8 differences with the tested offspring. There were between 11 and 14 differences between unrelated individuals. The tested offspring thus appears intermediate between its parent and a random sample of the population. This agrees with interpreting this plant as the product of the fertilization of a reduced ovule by a reduced pollen. Of the 8 different bands observed between the parent and the offspring, 3 were found only in the offspring, suggesting that it resulted from an outcross. Under selfing, one would merely expect some bands of the parent to be missing in the offspring.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 APPENDIX 2 LITERATURE CITED

The original scope of this study was to understand the mating system of an ecologically important species. However, H. diplandra population genetics may be an important system for elucidating the evolutionary forces acting upon apomixis. We can report several important features of this system.

Mating system:
As formerly shown by DUJARDIN 1979 , H. diplandra is an allotetraploid (with 2n = 40). Recent allotetraploidy was suggested by the fact that meiotic diakineses in pollen involve a majority of bivalents with only a few aberrations (DUJARDIN 1979 ). Aberrations are expected if the original genomes contributing to the new tetraploid stock belong to closely related species. In this case, some mispairings can occur between paralogous chromosomes. Our genetic analysis detected some aberrations, such as the presence of four hexaploid offspring in the progeny study and that of individuals (with unrecorded ploidy level) with three alleles at a locus in the population survey. These individuals most probably involve, in the first case, hexaploid individuals (3n) originating from fertilization of unreduced ovules by reduced pollen and, in the second case, either hexaploid zygotes or trisomic individuals (2n + 1) originating from aberrant meiotic reduction.

The study of meiosis in pollen by DUJARDIN 1979 could not record apomixis in H. diplandra, since pollen is meiotically reduced under apomixis. Pollen is required for the endosperm development and is therefore essential for seed maturation. Plants must thus invest in pollen as an agent eliciting the formation of a nongerminal tissue, notwithstanding its additional properties as a gamete. This primary function of pollen is a constraint that suffices to explain why apomictic grasses make pollen. "Pollination" must be distinguished from "fertilization." The evolution of "autopollination" by a plant would thus depend on selective forces that may have little relationship with the trade-offs often considered when opposing auto- and allofertilization. Since it is critical for an apomict to get fertilized, it should invest in pollen for itself, that is, to secure its autopollination and the mutual pollination of the neighboring individuals from the same clone. This suggests some kin-selection component among factors acting on the evolution of pollen dispersal.

Our study provides evidence that H. diplandra is a facultative apomict, with rare events of sexual reproduction. Our progeny survey found evidence of four cases of aberrant fertilization of an unreduced ovule by a reduced pollen (in these cases, possibly originating from the same plant) and one "normal" case of fertilization of a reduced ovule by a reduced pollen (in this case, originating from another plant). Observed cases give only an order of magnitude of the low occurrence of sexuality. It should be noted that some events of sexual reproduction may have gone unnoticed. Population homozygosity masks some cases of outcrossing, and segregation masks some cases of selfing. From our estimates of the selfing rate (s = 0.599) of heterozygosity at each locus (H = 0.615) and of deviation from Hardy-Weinberg proportions (F = 0.425), we could detect a proportion 1 - (1 - )² of fertilizations by selfing and a proportion 1 - (1 - H)² of fertilizations by outcrossing, yielding a weighted expectation P = 0.534. Our observation of only one case of fertilization of a reduced ovule out of 360 seeds (disregarding the 4 hexaploid seeds) shows that caryogamy occurs at a very low rate. Its estimate is = 0.525%, with a wide range of uncertainty.

To conclude, H. diplandra seeds can have one of three origins. They originate either from apomixis with an 0.9947 frequency, from self-fertilization with an 0.0031 frequency, or from outcrossing with an 0.00210 frequency. These are gross estimates showing that propagation through apomixis is dominant, yet remaining compatible with high levels of polymorphism. It should be noted that the rate of apomixis was determined at the zygotic stage through the analysis of germination, while the rate of outcrossing was inferred from the observation of fully grown individuals, that is, after selection may have differentially affected seeds of different origins in the wild.

These conclusions are based on the assumption that H. diplandra is an allotetraploid. This is a reasonable assumption, yet it still requires independent evidence. If this grass were an autotetraploid, our simulations indicate that the selfing rate would be higher (F = 0.884) and the frequency of the major allele quite high (range 0.43–0.74). Another uncertainty affecting our estimates under an allotetraploid model results from the difficulty of assessing the exact proportion of aberrant individuals (trisomic or aneuploid ones). The occurrence of three-banded individuals implies that some two-banded individuals are actually hexaploid or trisomic. Our numerical estimates are therefore biased, although probably very slightly, and this should not affect the qualitative conclusions of this study.

Short range distribution and the dispersal of seeds:
Our study of structured samples shows that neighboring individuals often belong to the same apomictic clone, while individuals 15 m distant are unrelated. We can suppose that apomictic clones occur in patches, the survival of which is inversely related to the rate of sexual reproduction. The extent of these patches in space would depend on two parameters: seed survival and seed dispersal.

Seed survival would depend on competition for space. Ecological observations have shown that the aerial parts of individuals occur in tufts, which are separated by soil surfaces deprived of organic nitrogen by the recycling activity of roots (ABBADIE et al. 1992 ). The success of germination over base soil would be low. Another possibility is that germination might occur within the parental tuft. Since sexuality occurs at a rate of 0.5%, a clone could theoretically survive 200 generations by successively replacing itself. We do not know the longevity of H. diplandra zygotes. It could be several years, meaning that a genotype could survive many centuries. In this study, we collected only one sheath per tuft, thus ignoring possible sources of genotypic heterogeneity within tufts and possibly underestimating microscale diversity. Tufts composed of different species have been observed (E. LEPROVOST, personal communication).

Seed dispersal could theoretically be estimated from the extent of clones, both in space and in number of individuals. Isolation-by-distance models assuming that dispersal distances are normally distributed are available (MALECOT 1966 ); however, these assumptions are probably unrealistic (FELSENSTEIN 1975 ). Since savanna grasses are nearly evenly distributed over space, and since genetic markers allow one to record the dispersal of seeds from the same apomictic clones, H. diplandra provides an interesting system for inspiring and validating new theoretical models. For example, combining the spatial extent of clones with inbreeding level in sexual reproduction events would allow one to infer the range of pollen dispersal. This question may be of primordial importance for the evolution of apomixis vs. sexuality and for the evolution of pollination within a clone. In savanna grasses, the spatial structuralization of clonal patches that can affect communal "autopollination" may induce coevolution between apomixis and selfing. Our study does not answer these questions, which are currently being examined.

Long range distribution and gene flow:
The five samples of the density survey, together with the four random samples of the population genetics survey, give a consistent picture of the organization of H. diplandra populations, which are evenly spread over the savanna, approximately following a Poisson distribution. Allele frequencies were significantly different between all samples, but these differences showed no correlation with distance. Samples collected 50–500 m away were more divergent than some samples collected 4 km away. Such structuring properties pertain to the island model rather than to the isolation-by-distance model. This suggests that the population is evenly distributed, with local fluctuations from average frequencies. Fluctuations could result from the small size of samples. Samples were collected before H. diplandra was known to be an apomict. These large areas involved thousands of individuals, although they were actually comprised of a much smaller number of clones. This suggests that genetic structuring involves two levels. At a local scale, individuals are clustered into clones, and areas 5000 m² (the approximate size of our random sample sites) constitute genetically differentiated neighborhoods. At the scale of the reserve, gene flow is sufficient for mixing the alleles between neighborhoods.

Why should a savanna grass become an apomict?
In a social plant like H. diplandra, apomixis is unlikely to be an "escape from sterility" as suggested by DARLINGTON 1939 , since fertilization opportunities are frequent. It could be viewed as an escape from some disadvantages of sexuality: either because segregation breaks up locally adapted combinations of genes or because it induces inbreeding. Considerable attention has been devoted to the interaction between inbreeding depression and mating systems in evolution. Although they probably interact a great deal, they seem to have the potential to generate many different evolutionary outcomes. To briefly overview some key factors that have been considered for the resolution of this problem, the evolution of selfing may involve overdominance (HOLSINGER 1988 ; UYENOYAMA and WALKER 1991A ), the distribution of lethal mutations in the genome (UYENOYAMA and WALKER 1991B ), and the opposite effects of strongly and weakly selected mutations (CHARLESWORTH et al. 1990 , CHARLESWORTH et al. 1993 ). Several models predict either intermediate levels of autogamy (HOLSINGER 1988 ) or extreme selfing or outcrossing strategies depending on population history (LANDE and SCHEMSKE 1985 ). In the case of H. diplandra, we observe a substantial level of selfing (60%). Yet most of this apparent selfing is probably due to outcrossing between neighboring individuals from the same clone. Thus even though apomixis could be a way of delaying reproduction, it does not prevent inbreeding when sexuality eventually occurs.

Another possibility is that apomixis allows the preservation of given genotypes. While our study was underway, LATA et al. 1999 showed that two kinds of "ecotypes" of H. diplandra coexist in the Lamto reserve, differing in nitrogen metabolism. One ecotype has a high nitrate reductase activity, assimilates nitrogen as nitrate, and stimulates the activity of nitrifying soil bacteria. The other has a low nitrate reductase activity, assimilates nitrogen as ammonium, and inhibits nitrifying soil bacteria. These two ecotypes belong to the same species but develop diverging, coherent strategies of nitrogen assimilation, and they tend to occur in clusters. Apomixis in H. diplandra could be a way of retaining locally adapted genotypes and escaping the hazards of fertilization by gametes from other ecotypes.

	ACKNOWLEDGMENTS

We thank R. Vuattoux, the staff of the Lamto Station, and Abidjan University for field facilities; the Centre National de la Recherche Scientifique (CNRS) Evolution des Systèmes Végétaux laboratory in Orsay for greenhouse facilities; the CNRS Institut des Sciences Végétales at Gif and the AEB laboratory in Amiens for the cytometry; S. Yakovlev and O. Panaud for discussion; M. Cobb for help in the preparation of the manuscript; and anonymous referees for fruitful comments. This research was supported by the CNRS Unité Mixte de Recherche 7635 and the CNRS Institut Fédératif de Recherche 3.

Manuscript received March 30, 1999; Accepted for publication June 2, 2000.

	APPENDIX 1

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 LITERATURE CITED

	APPENDIX 2

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 2 LITERATURE CITED

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION APPENDIX 1 APPENDIX 1 LITERATURE CITED

ABBADIE, L., A. MARIOTTI, and J. C. MENAUT, 1992  Independence of savanna grasses from soil organic matter for their nitrogen supply. Ecology 73:608-613.

ASKER, S., and L. JERLING, 1992 Apomixis in plants. CRC Press, Boca Raton, FL.

BARRÉ, M., 1995 Diversité génétique du genre Tripsacum et évolution des complexes agamiques. Ph.D. dissertation, Institut National Agronomique Paris-Grignon, France.

BROWN, S. C., C. BERGOUNIOUX, S. TALLET and D. MARIE, 1991 Flow cytometry of nuclei for ploidy and cell cycle analysis, pp. 326–345 in A Laboratory Guide for Cellular and Molecular Plant Biology, edited by I. NEGRUTIU and G. GHARTI-CHHETRI. Birkhaüser-Verlag, Basel, Switzerland.

CHARLESWORTH, D., M. T. MORGAN, and B. CHARLESWORTH, 1990  Inbreeding depression, genetic load, and the evolution of outcrossing rates in a multilocus system with no linkage. Evolution 44:1469-1489.

CHARLESWORTH, D., M. T. MORGAN, and B. CHARLESWORTH, 1993  Mutation accumulation in finite outbreeding and inbreeding populations. Genet. Res. 61:39-56.

DARLINGTON, C. D., 1939 The Evolution of Genetic Systems. Cambridge University Press, Cambridge, UK.

DE WET, J. M. J., 1967  Diploid-tetraploid-haploid cycles and the origin of variability in Dichantium agamospecies. Evolution 22:394-397.

DOYLE, J. J. and J. L. DOYLE, 1987  A rapid DNA isolation procedure for small quantities of fresh leaf tissue. Phytochem. Bull. 19:11-15.

DUJARDIN, M., 1979  Additional chromosome numbers and meiotic behaviour in tropical African grasses from western Zaïre. Can. J. Bot. 57:864-876.

ELLSTRAND, N. C. and M. L. ROOSE, 1987  Patterns of genotypic diversity in clonal plant species. Am. J. Bot. 74:123-131.

FELSENSTEIN, J., 1975  A pain in the torus: some difficulties with models of isolation by distance. Am. Nat. 109:359-368.

GOLDBLATT, P., and D. E. JOHNSON, 1981 Index to Plant Chromosome Numbers. Monographs in Systematic Botany From the Missouri Botanical Garden, Vol. 5. MBG Press, St. Louis, MO.

HALDANE, J. B. S., 1930  Theoretical genetics of autopolyploids. J. Genet. 22:359-372.

HOLSINGER, K. E., 1988  Inbreeding depression doesn't matter: the genetic basis of mating-system evolution. Evolution 42:1235-1244.

HUTCHINSON, J., and J. M. DALZIEL, 1972 Flora of West Tropical Africa.Vol. III, Part 2. Crown Agents for Overseas Governments and Administrations, Millbank, London.

KAMMACHER, P., G. ANOMA, E. ADJANOHOUN, and L. AKE ASSI, 1973  Nombres chromosomiques de Graminées de Côte d'Ivoire. Candollea 28:191-217.

LANDE, R. and D. W. SCHEMSKE, 1985  The evolution of self-fertilization and inbreeding depression in plants. I. Genetic models. Evolution 39:24-40.

LATA, J.-C., J. DURAND, R. LENSI, and L. ABBADIE, 1999  Stable coexistence of contrasted nitrification statuses in a wet tropical savanna ecosystem. Funct. Ecol. 13:762-768.

MALECOT, G., 1966 Probabilités et hérédité. Presses Universitaires de France, Paris.

MAYNARD-SMITH, J., 1978 The Evolution of Sex. Cambridge University Press, Cambridge, United Kingdom.

MENAUT, J. C. and J. CÉSAR, 1979  Structure and primary productivity of Lamto savannas, Ivory Coast. Ecology 60:1197-1210.

MENKEN, S. B. J., E. SMIT, and H. C. M. DEN NIJS, 1995  Genetical population structure in plants: gene flow between diploid sexual and triploid asexual dandelions (Taraxacum section Ruderalia). Evolution 49:1108-1118.

NOGLER, G. A., 1984 Gametophytic apomixis, pp. 475–518 in Embryology of Angiosperms, edited by B. M. JOHRI. Springer Verlag, Berlin/Heidelberg.

PERNÈS, J., Y. SAVIDAN, and R. RENÉ-CHAUME, 1975  Panicum: structures génétiques du complexe des "Maximae" et organisation de ses populations naturelles en relation avec la spéciation. Boisseria 24:383-402.

RAYMOND, M. and F. ROUSSET, 1995  GENEPOP (version 1.2): a population genetics software for exact tests and ecumenicism. J. Hered. 86:248-249[Free Full Text].

RICHARDS, A. J., 1996  Genetic variability in obligate apomicts of the genus Taraxacum.. Folia Geobot. Phytotaxon. 31:405-414.

SAVIDAN, Y. and J. PERNÈS, 1982  Diploid-tetraploid-dihaploid cycles and the evolution of Panicum maximum Jacq. Evolution 36:596-600.

UYENOYAMA, M. and D. M. WALKER, 1991a  Coevolution of self-fertilization and inbreeding depression. II. Symmetric overdominance in viability. Theor. Popul. Biol. 40:47-75[Medline].

UYENOYAMA, M. and D. M. WALKER, 1991b  Coevolution of self-fertilization and inbreeding depression. III. Homozygous lethal mutations at multiple loci. Theor. Popul. Biol. 40:173-209[Medline].

VIELLE CALZADA, J. P., C. F. CRANE, and D. M. STELLY, 1996  Apomixis: the asexual revolution. Science 274:1322-1323[Free Full Text].

WIDEN, B., N. CRONBERG, and M. WIDEN, 1994  Genotypic diversity, molecular markers and spatial distribution of genets in clonal plants, a literature survey. Folia Geobot. Phytotaxon. 29:245-263.

YONEZAWA, K., 1997  Effective population size of plant species propagating with a mixed sexual and asexual reproduction system. Genet. Res. 70:251-258.

This article has been cited by other articles:

				F. A. Matsen and J. Wakeley Convergence to the Island-Model Coalescent Process in Populations With Restricted Migration Genetics, January 1, 2006; 172(1): 701 - 708. [Abstract] [Full Text] [PDF]

				L. K. M. Garnier, J. Durand, and I. Dajoz Limited seed dispersal and microspatial population structure of an agamospermous grass of West African savannahs, Hyparrhenia diplandra (Poaceae) Am. J. Botany, November 1, 2002; 89(11): 1785 - 1791. [Abstract] [Full Text] [PDF]