of gynodioecious Thymus vulgaris L.

Transcription

1 Journal of Ecology 2004 Temporal variation in sex allocation in hermaphrodites Blackwell Publishing, Ltd. of gynodioecious Thymus vulgaris L. BODIL K. EHLERS and JOHN D. THOMPSON Centre d Ecologie Fonctionelle et Evolutive, CNRS, 1919 Route de Mende, Montpellier cedex 5, France Summary 1 Theory predicts that hermaphrodites adjust sex allocation in relation to the relative fitness contributions of male and female functions. Few studies have, however, focused on temporal variation in sex allocation within individuals. 2 We quantified temporal variation in allocation to male and female function of hermaphrodites of gynodioecious Thymus vulgaris across a single flowering season, in families from four natural populations grown in an experimental garden. 3 Allocation to pollen production and to seed set and germination varied significantly among sample dates, with male function highest when female function was lowest, and a significant negative correlation between pollen production and seed set among families indicated a trade-off between male and female function. 4 Females and hermaphrodites showed gender-specific flowering phenology: female plants achieved peak flowering later and flowered for a significantly shorter period than hermaphrodites. 5 The number of open female flowers was significantly positively correlated with pollen production by hermaphrodites. Hermaphrodite male function was maximized at the time of peak flowering by females in all four populations. 6 These results fit theoretical predictions that across-season variation in the mating environment could select for a temporal variation in hermaphrodite sex allocation, but other factors (e.g. variation in the selfing rate or resource levels) may also contribute to the observed pattern. Key-words: gender, mating environment, phenology, sex allocation Journal of Ecology (2004) Ecological Society Introduction Sex allocation theory predicts that allocation of resources to male and female function reflects the relative fitness contribution obtained by hermaphrodite plants via pollen and seed production (Charnov 1982). Indeed, variation in resource allocation to male and female functions among outcrossing hermaphrodite plants is well documented (e.g. Horovitz 1978; Devlin & Stephenson 1987; Brunet 1992). A range of factors, such as plant size and age, the selfing rate and the mating environment, have been shown, both theoretically and empirically, to be important in shaping such variation (e.g. Lloyd 1976; Brunet 1992; Klinkhammer et al. 1997). The focus of the present study is on temporal variation in sex allocation within individual plants across a single flowering season. This may depend on competition for resources among flowers. Position-dependent effects such as first-produced and/or basal-positioned Correspondence: Bodil K. Ehlers (tel ; bodil.ehlers@cefe.cnrs-mop.fr). flowers closer to the resource pool receiving more resources than later or more distal ones have been demonstrated in Fragaria virginia and were also shown to interact with the size of individual plants (Ashman & Hitchens 2000; Ashman et al. 2001). Ashman & Baker (1992) showed an overall decline in allocation to secondary sexual structures (e.g. sepals and petals) in Sidalcea oregana, but no changes in allocation to primary sexual structures among early- and lateproduced flowers, although there was a trend towards increased maleness in late-season flowers. Theoretical studies predict that temporal variation in sex allocation of hermaphrodites may also arise due to a variation in the mating environment. Brunet & Charlesworth (1995) demonstrated that, for protandrous hermaphrodites with sequential blooming, variation in the availability of ovules could select for a variation in sex allocation among hermaphrodite flowers across a single flowering season (even in the absence of among-flower differences in resource allocation). Flowers with a high probability of pollen transfer should allocate proportionally more of their resources

2 16 B. K. Ehlers & J. D. Thompson to male function than flowers with a lower probability of pollen transfer. When the possibility of pollinating flowers which specialize in female function is high, sex allocation to male function will be further enhanced. The seasonal variation in sex allocation predicted by their model is a female-biased sex allocation in early-produced flowers (where pollen is abundant and competition for ovules is high) and a shift towards male-biased sex allocation in later produced flowers (where the number of available ovules has increased). In gynodioecious populations, where hermaphrodites coexist with females, gender-specific flowering phenology may additionally influence temporal variation in the mating environment across the flowering season. Gender-specific flowering phenology in which females flower less often or over shorter periods than males in order to reduce their cost of reproduction has been reported in several species (Delph 1999). As in many gynodioecious species females compensate for lack of male function by having higher seed set, one could predict that sex allocation by hermaphrodites may respond, not only to the availability of ovules of hermaphrodite flowers in their female phase (as predicted by Brunet & Charlesworth 1995) but also to the availability of ovules on female plants. Thymus vulgaris is a gynodioecious species with a nuclear-cytoplasmic male sterility system (Belhassen et al. 1991). Females carry a cytoplasmic male sterility gene that prevents the production of functional anthers. Male function is restored by the presence of the appropriate nuclear restorer alleles. Variation in both male sterility genes and restorer alleles exists. Mismatches between male sterility genes and nuclear restorer alleles are believed to be important for the high variation in female frequency (6 93%) observed among populations (Belhassen et al. 1989; Manicacci et al. 1996; Manicacci et al. 1997). Newly established populations often have high female frequencies, probably because of the absence of nuclear alleles that restore male fertility of the local cytoplasmic types. When appropriate restorer alleles arrive at the population and increase in frequency, female frequency declines (Couvet et al. 1986; Belhassen et al. 1989; Manicacci et al. 1992; Manicacci et al. 1996). Although hermaphrodites show genetic variation for both pollen (Atlan et al. 1992) and viable seed (Thompson & Tarayre 2000) production, previous studies of different populations that vary widely in sex ratio have shown no evidence that hermaphrodites adjust their male function in relation to female frequency (Manicacci et al. 1998; Thompson & Tarayre 2000). This lack of correlation between hermaphrodite male function and female frequency among populations may be explained by rapid population turnover and the longevity of individual plants preventing an evolutionary response in hermaphrodite male function at the population level (Manicacci et al. 1998). However, if female and hermaphrodite plants show genderspecific flowering phenology that is consistent across populations, this may create a more predictable temporal variation in mating environment that could select for a response in hermaphrodite sex allocation within populations across a flowering season. Hermaphrodite flowers are protandrous, with male and female phases of roughly the same duration (2 3 days). Individual plants flower for up to 6 weeks. As females have a two to five times higher seed set than hermaphrodites in natural populations (Assouad et al. 1978) and after controlled pollination (Thompson & Tarayre 2000), pollen that fertilizes the ovules of a female plant will have a greater chance of successfully siring a seed than pollen fertilizing a hermaphrodite plant. We would thus predict that, across a flowering season, sex allocation in hermaphrodite flowers of T. vulgaris may respond to both variation in the number of hermaphrodite flowers in their female phase and to the flowering phenology of female plants. We quantified the flowering phenology of females and hermaphrodites of T. vulgaris in order to examine whether temporal variation in the mating environment due to gender-specific flowering phenology occurs. We examined how hermaphrodites of similar age and size originating from 29 different families sampled from four populations vary in sex allocation across one flowering season. We followed individual hermaphrodite plants across their entire flowering period and estimated their allocation to male (pollen production) and female (seed production and germination rate) functions at different dates. Based on the flowering phenology of females and hermaphrodites we estimated the relative number of open female flowers and hermaphodite flowers in the male and female phases on each of the dates on which pollen and seed production were quantified. We were thus able to examine whether temporal variation in sex allocation in hermaphrodites is correlated with temporal availability of female flowers through the flowering season. Materials and methods MATERIAL Plants used in this study all originated from populations in and around the St Martin-de-Londres valley 25 km north of Montpellier. The sampling scheme and pollination experiment used to produce these plants is described in Thompson & Tarayre (2000). Briefly, four natural populations with varying female frequency (population 1, 26%; population 2, 62%; population 3, 12%; and population 4, 80%) served as seed sources. Fruits were collected from 10 female plants in each population (F 0 -parents, each giving rise to a maternal lineage) and their seeds germinated and brought to flowering at the CEFE-CNRS experimental garden in Montpellier. One or two F 1 females and hermaphrodites were selected from the offspring of each F 0 female parent and F 2 offspring were produced from selfing, within-family and between-family (within population)

3 17 Sex allocation in thyme Table 1 Sample sizes used to quantify flowering phenology of hermaphrodites (H) and females (F) in the of F 2 offspring of five to eight families from each of four populations of Thymus vulgaris grown in an experimental field Family Population H F H F H F H F v Total crosses. Due to variation in growth and flowering of these F 1 plants, the number of maternal lineages, hereafter families, derived from a population varied from five to eight (Table 1). For each cross, seeds were germinated and up to 10 seedlings were planted in a single randomized block in the experimental field at the CEFE-CNRS laboratory in Montpellier, to give F 2 plants. The sex ratio values for the F 2 offspring were 26, 43, 46 and 53% (for populations 1 4, respectively). All plants used were 3 years old and grew in the same experimental field. Hence we limited the expression of any variation in phenology and resource allocation that could be caused by ontogenetic or large-scale environmental differences. Any variation among sexes, populations and families thus, at least partly, reflects genetic variation. FLOWERING PHENOLOGY Twice weekly, from the end of March until the beginning of June 2001, all plants (c. 1200, see Table 1 for details) were observed and classified into five phenological classes: pre-flowering, beginning of flowering, maximum flowering, end of flowering, and post-flowering. A plant was considered to be at maximum flowering when more than 50% of its stems bore open flowers, and to have ended its flowering when the last open flower had withered. Thus, for each individual plant data were obtained from its day of onset of flowering, onset of maximum flowering, and end of flowering. POLLEN AND SEED PRODUCTION In each F 2 family from each population, three hermaphrodites were chosen for the study of sex allocation (a total of 87 hermaphrodite plants). Care was taken to choose individuals of similar size in order to reduce any size-dependent effects there might be on sex allocation pattern. To prevent variation in the degree of inbreeding within and between families confounding our results we used only plants produced by a between-family cross. Furthermore, as resource allocation to male and female functions among progeny may depend on the sex of the maternal plant (Atlan et al. 1992; Gigord et al. 1999), all the hermaphrodites we studied had female plants as both their maternal parent (F 1 ) and grandparent (F 0 ). On each of five dates across the flowering period (i.e. days 17, 29, 36, 42 and 48 after the day when the first open flower was observed), four randomly chosen flower buds from each hermaphrodite were collected just prior to anthesis and placed in an eppendorf tube and kept at 5 C until the end of the flowering season. Flower buds were collected from the same individuals at each date and any variation across the season thus represents variation in sex allocation within individuals, not between early and late flowering individuals. Pollen was extracted by adding 300 µl of pure sulphuric acid to each sample for 24 h, to destroy all floral parts except the pollen grains. After adding 1.5 ml of a 2% Triton solution samples were centrifuged for 5 min at 2000 g. The pellet was washed twice with 1.0 ml ethanol, each time centrifuged for 5 min at 2000 g, before drying; 100 µl of counting solution (20% glycerin and 30% saccharose) was then added and samples were kept at 5 C until counting. Immediately before counting, samples were placed in a ultrasonic bath for 2 min to prevent pollen grains from aggregating. Pollen grains in a 1-µL alignment were counted under a microscope using a stage micrometer and the mean of two counts used as an estimate of pollen production. This procedure is that used in previous studies of male gender in Thymus (Atlan et al. 1992; Manicacci et al. 1998). Fruit set of flowers was estimated for three different periods within the flowering season, i.e. beginning, middle and end of the flowering period. On days 17, 36 and 48, four to six flower buds about to undergo anthesis were selected on the same plants from which pollen production was estimated and their bracts marked with paint. Mature, marked fruits were later collected and seed set counted. As for estimates of pollen production, any variation in fruit set across season reflects variation in sex allocation within individuals. Some fruits had fallen to the ground or had lost their marks prior to collection and only a total of 148, 65, 161 and 168 fruits from populations 1 4, respectively, were analysed. All seeds produced from these fruits (125, 86, 226 and 205, respectively) were scarified by automatic shaking for 60 min in eppendorf tubes containing Fointainebleau sand. Families were pooled within each population and sampling date and seeds were sown in plastic trays containing a 1 : 1: 1 sand : humus : garden soil mixture and left to germinate for a period of 3 months. We used pollen production per hermaphrodite flower as an estimate of male function and used seed set per fruit and germination rate as estimates of female function. We also marked flowers produced across the season on one female plant from each family. However, because many fruits were lost or had their paint rubbed

4 18 B. K. Ehlers & J. D. Thompson off, it was not possible to present data on female seed set. Manicacci (1993) showed that females do not vary their seed set across flowering season, provided they are not pollen limited, and Thompson & Tarayre (2000) showed that seed set of females shows little variation relative to that by hermaphrodites. ESTIMATION OF THE NUMBER OF OPEN FEMALE FLOWERS AND TOTAL POLLEN PRODUCTION In order to determine whether pollen production of hermaphrodites was correlated with the number of available flowers in the female phase we estimated (a) the number of open female flowers on female plants and (b) the number of hermaphrodite flowers in the male and female phase, on each of the five dates on which pollen production was estimated. We assumed that all plants produce the same maximum number of flowers at maximum flowering and that the number of flowers produced by a plant can be described as a linearly increasing function from its first day of flowering to its date of maximum flowering and as a linearly decreasing function from then to the end of flowering. To estimate the total number of open flowers on female plants from each population on a given day we summed the number of flowers produced by all the individual female plants within that population. As individual plants flower continually across a season and male and female phases within an hermaphrodite flower last the same length of time, we assumed that at any given time the number of flowers in the female phase was half the total number of open hermaphrodite flowers (calculated as described above). Furthermore, as hermaphrodite individuals produce two to five times fewer seeds than females we reasoned that one female-phase hermaphrodite flower is of less value than a flower on a female plant, and therefore divided the number of the former by three to give approximately equivalent values. These estimates do not describe the absolute number of flowers open on a given date, but provide an estimate of their relative number for comparison among dates within a population. The number of male-phase flowers (the other half of the open hermaphrodite flowers on each of the sampling dates) was multiplied by the pollen grains produced per flower to give a relative estimate of the total pollen production in a population. STATISTICAL ANALYSIS The effects of sex, population and family, and the interactive effects of sex and population or family, on flowering phenology were analysed by a mixed model ANOVA with families nested within populations, using PROC GLM (SAS 1999). To further examine the effect of family and sex, a mixed-model ANOVA was performed for each population separately. In both analyses the effect of family and the interaction between sex and family were specified as random effects with denominator degrees of freedom calculated using Satterthwaite s approximation (SAS 1999). F-ratios were calculated using type III sums of squares. Plots of residual error values were inspected visually in order to verify homogeneity of variances. The effect of population on pollen and seed production (all dates combined) were analysed by one-way ANOVA using PROC GLM (SAS 1999). Variation in pollen production and seed production among populations and dates were analysed by repeated measures ANOVA with the PROC GLM (SAS 1999) using the REPEATED statement, which takes into account measures taken on the same plants but at different dates. Due to missing data within families at some dates these analyses were performed using the family means for each date. Hence, the family time interaction could not be calculated. Type III sums of squares were calculated in all ANOVAs. Differences in seed germination among populations and dates were analysed with a contingency tables test using the computer program JMP (SAS 1995). As the low number of families did not meet the criteria for normal distribution of data, we used a non-parametric Spearman correlation to examine the association between pollen production and seed set across families. Results FLOWERING PHENOLOGY Analysis of variance for all populations combined (Table 2) revealed significant population sex interactions on Table 2 Mixed-model ANOVA of timing (first, maximum and end) and length of flowering in hermaphrodites and females from four populations of Thymus vulgaris First flowering Maximum flowering End of flowering Length of flowering Source d.f. MS F P MS F P MS F P MS F P Sex < < < Population < < < Family (population) < < < Sex population < Sex 2004 family British (population) Ecological Error Society, * * , *Error degrees of freedom were 1047.

5 19 Sex allocation in thyme all phenological variables except end of flowering, indicating that the effect of sex on flowering phenology varied among populations. Differences between the two sexual phenotypes also depended on family, as shown by the significant sex family interaction for all variables. There were significant differences between the two sex phenotypes for all phenological variables, and population of origin had a significant effect on the timing of flowering. The significant population sex effect is due to the fact that although hermaphrodites always achieve maximum flowering sooner and end flowering later, and thus flower for a significantly longer period than female plants, the degree of difference between the two sexes vary between populations (Fig. 1). Analysis of variance for each population separately confirmed that the response is the same for all populations: a significant effect of sex on the timing of maximum and end of flowering, and the length of flowering, but not on its Fig. 1 Population mean values (+ SE) for number of days to first, maximum and end of flowering and for the length of flowering (in days) for hermaphrodites (closed bars) and females (open bars) in the F 2 offspring from four populations of Thymus vulgaris grown in the same field. Times are measured as the number of days since the first open flower was observed in a population. onset (Table 3). Although the peak flowering period of hermaphrodites starts before that of females (around day 25 vs. day 36) it lasts longer and encompasses that of females. In all populations (and families), hermaphrodites flower for a longer period than females (6 vs. 3 4 weeks, Fig. 1, Table 3). POLLEN AND SEED PRODUCTION Analysis of variance of pollen production per flower across all dates combined showed no effect of population (F 3,101 = 0.96, P > 0.1) but a significant effect of family within populations (F 23,101 = 1.83, P < 0.05). Repeated measures analysis of variance showed no effect of population and revealed a significant effect of time on pollen production per flower (Table 4). Thus pollen production showed similar variation across the flowering season in all four populations (Fig. 2a). After a sequential Bonferonni correction for multiple (four) comparisons, there was no significant variation in pollen production per flower between days 17 and 29 (F 1,16 = 0.84, P > 0.1), a significant increase in pollen production between days 29 and 36 (F 1,16 = 7.71, P < 0.05), a small but significant decrease between days 36 and 42 (F 1,16 = 6.87, P < 0.05), and a highly significant decrease between days 42 and 48 (F 1,16 = 9.18, P < 0.01). On day 48 pollen production per flower was reduced by 83%, 74%, 69%, and 71% of that on day 36 in populations 1 4, respectively (Fig. 2a). Total pollen production (number of hermaphrodite flowers in male phase multiplied by pollen per flower at each sampling date; Fig. 3, bars) showed more marked temporal variation, with a 100% increase in all populations around day 36 relative to days 17 and 48. This temporal variation parallels the number of open female flowers (Fig. 3, line). A positive correlation between total pollen production and number of open female flowers was found across all populations (pairwise correlation: r = 0.58, n = 20, P < 0.01), and within populations 2, 3 and 4 (Spearman rank correlation similar for each of the three populations: r s = 0.9, P < 0.05, n = 5) but not in population 1 (r s = 0.77, P = 0.1, n = 5). Overall ANOVA showed no significant effect of population (F 3,108 = 2.07, P = 0.09) or family within populations (F 23,108 = 1.48, P = 0.08) on seed production. Repeated measures analysis of variance showed no effect of population but a significant effect of time on seed set (Table 4). In contrast to pollen production per flower, seed set per fruit on hermaphrodite plants was maximal at the beginning of the flowering season, showed a significant decrease at peak flowering (F 1,20 = 28.44, P < ) and a significant increase (F 1,20 = 7.15, P < 0.05) between days 36 and 48, with similar patterns in all populations (Fig. 2b). Seed germination patterns were similar to those for seed production, although the late season increase was 2 more marked (Fig. 2c). Except for population 1 ( χ 2123, = 2.71, P > 0.1), all populations showed a significant

6 20 Table 3 Mixed model ANOVA of timing (first, maximum and end) of flowering and length of flowering in hermaphrodites and females in each of four B. populations K. Ehlers of & Thymus vulgaris J. D. Thompson Population 1 Population 2 Population 3 Population 4 Source d.f. MS F P d.f. MS F P d.f. MS F P d.f. MS F P (a) First flowering Sex Family < < Sex family Error (b) Maximum flowering Sex < < < < Family < Sex family Error (c) End of flowering Sex < < < < Family Sex family Error (d) Length of flowering Sex < < < < Family Sex family Error Fig. 2 Allocation to male (pollen production) and female (seed set and germination rate) function of hermaphrodites across the flowering season in F 2 offspring from four populations of Thymus vulgaris grown in a uniform field. (a) Mean (+ SE) pollen production per flower. (b) Mean seed set (+ SE) per flower. (c) Germination rate. Dates are number of days after the first flower opened. 2 effect of date on seed germination ( χ 284, = 7.05, P < for population 2, χ 2130, = 9.98, P < 0.01 for population 2 3, and χ 2, 203 = 10.43, P < 0.01 for population 4). We found a significant negative correlation between seed set and pollen production across families pooled over populations (Spearman rho = 0.48, P < 0.01, n = 26, Fig. 4), as previously observed in T. vulgaris (Atlan et al. 1992), suggesting that there may be a genetically based trade-off between male and female function. For the three days on which we estimated

7 21 Sex allocation in thyme Table 4 Repeated measures ANOVA for (a) pollen production and (b) seed set in families from four populations of Thymus vulgaris grown in a uniform garden Pollen production Seed set Source d.f. MS F P d.f. MS F P Population Error Time < Time population Error (Time) Discussion Fig. 3 Relative total pollen production by hermaphrodites in a population (histograms), relative number of open flowers on female plants (solid line) and relative number of open hermaphrodite flowers in female phase (dotted line) scaled to the value of a female flower on a female plant (see text) across one flowering season in the F 2 offspring from four populations of Thymus vulgaris grown in a uniform field. seed set, a significant negative correlation was observed on day 17 (Spearman rho = 0.58, P < 0.05, n = 25) and day 36 (Spearman rho = 0.45, P < 0.05, n = 26) but not on day 48 (Spearman rho = 0.18, P > 0.1, n = 21). The principal result of this study is that hermaphrodites show temporal variation in sex allocation across a flowering season in plants originating from each of four different populations. This variation is due to variation in sex allocation among flowers produced at different times on the same plant and indicates that the relative gender of hermaphrodites varies in time across a single season. In addition, gender-specific flowering phenology results in large temporal variation in the availability of ovules, with the number of open female flowers positively correlated with pollen production by hermaphrodites. Several different factors may contribute to this temporal variation in sex allocation. On a per flower basis, our results agree with the prediction of Brunet & Charlesworth (1995) that relative female function is higher in early-produced flowers (day 17) and that relative male function is higher in later produced flowers (day 36). However, female function predominates again at the end of the flowering season (day 48), due to a decrease in pollen production per flower and an increase in seed germination rates (Fig. 2). Although care should be taken in interpretation, the correlation between pollen production and open female flowers (Fig. 3) fits the prediction that malebiased sex allocation is favoured when the probability of pollen transfer to ovules is high. Such temporal variation of allocation in hermaphrodites may thus have an adaptive basis by maximizing male function and minimizing female function when female plants are at peak flowering. The same temporal allocation pattern was found in hermaphrodites in all four populations, supporting the possibility that it may have been selected for. The temporal variation in hermaphrodite sex allocation might be expected to be greatest in the populations with the highest female frequency as these would also hold the highest number of available female ovules. Significant correlations between pollen production and open female flowers were indeed found for the three populations with the highest female frequency in the F 2 generation (i.e. populations 2, 3 and 4, with female frequencies of 43%, 46% and 53%, respectively), although in the field population 3 had the lowest estimated female frequency (12%) of the four. Sex ratios may, however, vary over years within populations, and this

8 22 B. K. Ehlers & J. D. Thompson Fig. 4 Plot of family means of seed and pollen production of hermaphrodites from five to eight families from four populations. Closed diamonds, population 1; closed squares, population 2; closed triangles, population 3; open circles, population 4. may constrain variation in sex allocation in response to variation in sex ratio among populations. A second factor that may influence temporal variation in sex allocation is variation in the selfing rate. Selfing rate in hermaphrodites increases in populations with increased female frequency in several gynodioecious species (Sun & Ganders 1986; McCauley & Taylor 1997), including T. vulgaris (Valdeyron et al. 1977; Thompson & Tarayre 2000), and it is possible that there may also be temporal variation in the selfing rate of hermaphrodites over the season as female flowering increases. In T. vulgaris, pollination is mainly by honeybees that visit many open flowers on the same plant. Thus, levels of geitonogamous pollination may be very high at peak flowering when the frequency of female flowers is high and when hundreds of flowers may be open on a single hermaphrodite plant. As selfing causes a decline in the germination rate of seeds produced by hermaphrodites of T. vulgaris (Assouad et al. 1978; Thompson & Tarayre 2000) the seriously reduced germination of seeds in the middle of the flowering season could be a consequence of inbreeding depression. The germination rate in some populations is reduced by almost 50% at peak flowering relative to the beginning and end of the flowering period. If increased selfing were to explain this result it would require an inbreeding depression on germination rate of up to 50% assuming complete selfing at peak flowering and complete outcrossing at the beginning and end of flowering. Even higher levels of inbreeding depression would be needed if intermediate selfing rates are assumed. Such high levels of inbreeding depression would be surprising (Husband & Schemske 1996) and have not been reported for this species (Thompson & Tarayre 2000). Thus, variation in the selfing rate alone is unlikely to explain the decrease in seed germination at peak flowering. In the case of an increased selfing rate, sex allocation theory predicts a female-biased sex allocation (e.g. Charlesworth & Charlesworth 1981; De Jong et al. 1999), whereas we observed an increase in male function when selfing was potentially highest. Although selfing rate may vary across the season, the observed allocation towards increased male function is probably not a response to a potentially increased level of selfing at peak flowering. Finally, the reduced female function in hermaphrodites at peak flowering could be a consequence of resource limitation. It is possible that high flower production causes a trade-off between number of flowers and reproductive investment in individual flowers. Mazer & Schick (1991) found that investment in female gametes in Raphanus is indeed more sensitive to resource limitation than male gamete production. Although not a new result (see Atlan et al. 1992), we found a significant negative correlation between seed set and pollen production among families, indicating that trade-offs between male and female function may influence the evolution of resource allocation patterns in hermaphrodites of T. vulgaris. To our knowledge, this is one of the first studies to demonstrate significant within-individual variation in allocation to male and female function across a single flowering season. Our data fit the predictions made by Brunet & Charlesworth (1995) that hermaphrodites should increase their allocation to male function when availability of ovules increases. Deciding whether the pattern is shaped by variation in the mating environment, selfing rate or resource depletion at the time of peak flowering, however, requires comparative studies of hermaphrodite and gynodioecious species. Acknowledgements We are grateful to M. Maistre, who counted the seeds, C. Collin for help with plant cultivation, to B. Husband, S. Maurice and T. Bataillon for discussion, and to T. Bataillon, Pierre-Olivier Cheptou and anonymous reviewers for useful comments on earlier versions of the manuscript. The Danish National Science Foundation (SNF) and the CNRS provided financial support.

9 23 Sex allocation in thyme References Ashman, T.L. & Baker, I. (1992) Variation in floral sex allocation with time of season and currency. Ecology, 73, Ashman, T.L. & Hitchens, M.S. (2000) Dissecting the causes of variation in intra-inflorescence allocation in a sexually polymorphic species, Fragaria virginiana (Roseaceae). American Journal of Botany, 87, Ashman, T.L., Pacyna, J., Diefenderfer, C. & Leftwich, T. (2001) Size dependent sex allocation in a gynodioecious wild strawberry: the effects of sex morph and inflorescence architecture. International Journal of Plant Science, 162, Assouad, M.W., Dommée, B., Lumaret, R. & Valdeyron, G. (1978) Reproductive capacities in the sexual forms of the gynodioecious species Thymus vulgaris L. Biology Journal of the Linnean Society, 77, Atlan, A., Gouyon, P.-H., Fournial, T., Pomente, D. & Couvet, D. (1992) Sex allocation in an hermaphroditic plant: the case of gynodioecy in Thymus vulgaris L. Journal of Evolutionary Biology, 5, Belhassen, E., Dommée, B., Atlan, A., Gouyon, P.-H., Pomente, D., Assouad, M.W. &. Couvet, D. (1991) Complex determination of male sterility in Thymus vulgaris L. genetic and molecular analysis. Theoretical and Applied Genetics, 82, Belhassen, E., Trabaud, L., Couvet, D. & Gouyon. P.-H. (1989) An example of non-equillibrium processes: gynodioecy of Thymus vulgaris L. in burned habitats. Evolution, 43, Brunet, J. (1992) Sex allocation in hermaphroditic plants. Trends in Ecology and Evolution 7, Brunet, J. & Charlesworth, D. (1995) Floral sex allocation in sequentially blooming plants. Evolution, 49, Charlesworth, D. & Charlesworth, B. (1981) Allocation of resources to male and female function in hermaphrodites. Biology Journal of the Linnean Society, 19, Charnov, E.L. (1982) The Theory of Sex Allocation. Princeton University Press, Princeton, New. Jersey. Couvet, D., Bonnemaison, F. & Gouyon, P.H. (1986) The maintenance of females among hermaphrodites: the importance of nuclear cytoplasmic interactions. Heredity, 57, De Jong, T.J., Klinkhammer, P.G.L. & Rademaker, M.C.J. (1999) How geitonogamous selfing affects sex allocation in hermaphrodite plants. Journal of Evolutionary Biology, 12, Delph, L.F. (1999) Sexual dimorphism in life history. Gender and Sexual Dimorphism in Flowering Plants (eds M.A. Geber, T.E. Dawson & L.F. Delph), pp Springer- Verlag, Berlin. Devlin, B. & Stephenson, A.G. (1987) Sexual variations among plants of perfect-flowered species. American Naturalist, 130, Gigord, L., Lavigne, C.J., Shykoff, J. & Atlan, A. (1999) Evidence for the effects of restorer genes on male and female reproductive functions of hermaphrodites in the gynodioecious species Thymus vulgaris L. Journal of Evolutionary Biology, 12, Horovitz, A. (1978) Is the hermaphrodite flowering plant of equisexual? American Journal of Botany, 65, Husband, B.C. & Schemske, D.W. (1996) Evolution of the magnitude and timing of inbreeding depression in plants. Evolution, 50, Klinkhammer, P.G.L., de Jong, T.J. & Metz, H. (1997) Sex and size in cosexual plants. Trends in Ecology and Evolution, 12, Lloyd, D.G. (1976) The transmission of genes via pollen and ovules in gynodioecious angiosperms. Theoretical Population Biology, 9, Manicacci, D. (1993) Evolution et Maintien de la Gynodioecie: Allocation Sexuelle et Structuration Spatiale Du Polymorphisme Nucléo-Cytoplasmique Etude Théorique et Approches Expérimentales Dans le Genre Thymus. Université des Sciences et Techniques du Languedoc, Montpellier, France. Manicacci, D., Atlan, A. & Couvet, D. (1997) Spatial structure of nuclear factors involved in sex determination in the gynodioecious Thymus vulgaris L. Journal of Evolutionary Biology, 10, Manicacci, D., Atlan, A., Elena-Rossello, J.A. & Couvet, D. (1998) Gynodioecy and reproductive trait variation in three Thymus species (Lamiaceae). International Journal of Plant Science, 159, Manicacci, D., Couvet, D., Belhassen, E., Gouyon. P.-H. & Atlan. A. (1996) Founder effects and sex ratio in the gynodioecious Thymus vulgaris L. Molecular Ecology, 5, Manicacci, D., Olivieri, I., Perrot, V., Atlan, A., Gouyon, P.H., Prosperi, J.M. et al. (1992) Landscape ecology: population genetics at the metapopulation level. Landscape Ecology, 6, Mazer, S.J. & Schick, C.T. (1991) Constancy of population parameters for life history and floral traits in Raphanus sativus L. II. effects of planting density on phenotype and heritability estimates. Evolution, 45, McCauley, D.E. & Taylor, D.R. (1997) Local population structure and sex ratio evolution in gynodioecious plants. American Naturalist, 150, SAS Institute (1995) JMP Users Guide, 3rd edn. Cary, North Carolina. SAS Institute (1999) SAS/STAT Users Guide, 8th edn. Cary, North Carolina. Sun, M. & Ganders, F.R. (1986) Female frequencies in gynodioecious populations correlated with selfing rates in hermaphrodites. American Journal of Botany, 73, Thompson, J.D. & Tarayre, M. (2000) Exploring the genetic basis and proximate causes of female fertility advantage in gynodioecious Thymus vulgaris. Evolution, 54, Valdeyron, G.B., Dommée, B. & Vernet, P. (1977) Selffertilisation in male-fertile plants of gynodioecious species: Thymus vulgaris L. Heredity, 39, Received 20 June 2002 revision accepted 14 October 2003