From YY, we excavated 34 fruiting males and 15 females on which the first flowers were just opening. Plants were potted using soil from the study site, sand and peat (2 : 1 : 1), and placed in an unheated, pollinator-free glasshouse. Fruiting males were either cross- or self-pollinated (n = 15 and 19, respectively). We emasculated flowers on crossed fruiting males before anthers opened and crossed them using pollen from several male plants. We selfed fruiting males using pollen from the same plant. Females were crossed similarly to fruiting males (n = 15). Other fruiting males and females (both n = 15) were marked in the field as open-pollinated controls. Shortly after flowering, these plants were excavated, established in the glasshouse and treated as the hand-pollinated plants. We watered and fertilized all plants regularly.
Seed set and progeny performance
We harvested fruits from each plant in the five pollination treatments and counted the numbers of seeds and aborted seeds. We compared numbers of seeds and aborted seeds among treatments with ancovas, with the number of flowers per plant as a covariate. The pollination × covariate interactions were not significant (seeds, P = 0.713; abortions, P = 0.508), and they were omitted from final analyses. The covariate was positively related to both traits (seeds and abortions, both P < 0.001).
We weighed individually either 15 seeds or all of the seeds produced from each of 15 families in the five pollination treatments (n = 1111 seeds). Of the 19 selfed fruiting males, five produced between 11 and 14 seeds and four produced too few seeds for the subsequent experiments. We calculated mean seed mass per family and compared treatments using a one-factor anova, with families as replicates.
We assessed seed germination by placing 11–15 seeds from each family in separate Petri dishes on moistened filter paper (n = 1111 seeds). Three dishes per treatment were assigned to each of five shelves in a germination cabinet set at 12-h light (16°C) and 12 h darkness (8°C). Dishes were rearranged within shelves weekly to reduce microenvironmental variation. Seed germination was scored every week for 4 months. Ungerminated seeds had brown embryos and were not viable. We initially compared per cent seed germination using a two-factor random block anova with pollination treatment and cabinet shelves and as fixed and random block factors, respectively. Shelves and the shelves × treatment interaction were not significant (shelves, P = 0.266; interaction, P = 0.337), and they were omitted from the final analysis. We then compared germination among treatments using a one-factor anova.
We planted eight seedlings from 11 selfed and 10 open-pollinated fruiting male families, and 12 families each of crossed fruiting males and crossed and open females (n = 456 seedlings). Seedlings were planted into 12, 64-cell seedling trays (cell volume, 52 cm3) containing sand, loam and peat (1 : 1 : 1). Most trays contained one family of each of the five pollination treatments. Trays were placed in a growth cabinet at 12-h light (20°C) and 12-h darkness (10°C) for 4 weeks; all seedlings survived transplanting. Trays were then transferred to a glasshouse and their positions were rearranged every 10–14 days to minimize micro-environmental variation. Plants were watered regularly and fertilized every month with half-strength liquid fertilizer.
Before seedlings entered summer dormancy, they all produced a single leaf that we measured (n = 456 seedlings). During dormancy, we reduced the frequency of watering to every 14 days. After 4 months, the previous watering and fertilizing regimes were resumed, and resprouting was monitored. At the end of the second growing season prior to the dormancy period, we counted the number of surviving individuals in each family. We then excavated plants and weighed corms and leaves separately after at drying 65°C.
We calculated per cent survival and means for leaf length, and corm and plant mass for each family, and compared treatments using random block anovas with trays as blocks and families as replicates. In all analyses, trays were not significant (all P > 0.70), and they were omitted from further analyses. We then compared treatments using one-factor anovas. We were interested in five nonorthogonal, planned contrasts: (i) selfed vs. crossed fruiting males, to estimate inbreeding depression; (ii) open fruiting males vs. open females, to examine differences in fitness of the sex morphs following natural pollination; (iii) crossed fruiting males vs. crossed females, to examine differences in fitness of the sex morphs when outcrossing rate was equal; (iv) open vs. cross-pollinated fruiting males, to examine effects of inbreeding following natural pollination on fruiting male fitness; and (v) open vs. crossed females, to examine effects of biparental inbreeding following natural pollination on female fitness. We used sequential Bonferroni adjustments to maintain an overall significance level of α = 0.05.
Here and elsewhere, we transformed counts, measurements and masses using natural logarithms, and percentages using arc-sine square-root. Means (±SE) are given. Analyses were computed using jmp (version 5.01a, SAS Institute Inc., NC, USA).
Relative performance and inbreeding depression
We estimated the relative performance (RP) of selfed to crossed fruiting males as: RP=ws/wc, where ws and wc are the performances of selfed and crossed progeny, respectively. RP was calculated for seed production, seed germination, seedling survival and corm mass, the latter included as an indicator of differences in growth rate that may affect future survival and fecundity. Pollination treatments were undertaken on different plants, and RPs were calculated as the ratios of trait means without regard to family, providing population rather than family estimates. Cumulative RP was estimated as the product of the RPs of the above traits. Inbreeding depression was summarized as: δ = 1 – cumulative RP (Charlesworth & Charlesworth, 1987). To correspond with the planned contrasts described in the previous section, we also calculated individual and cumulative RPs for open fruiting males to open females, crossed fruiting males to crossed females, open to crossed fruiting males and open to crossed females. We calculated standard errors for RP estimates using 1000 bootstrap samples randomly resampled from the original data (Chernick, 1999). We tested whether the estimates, or differences between estimates, differed from one or zero, respectively, using methods described above for the mating system analyses.