The transition from outcrossing to self-fertilization is one of the most common evolutionary trends in plants1. Reproductive assurance, where self-fertilization ensures seed production when pollinators and/or potential mates are scarce, is the most long-standing and most widely accepted explanation for the evolution of selfing2,3,4,5,6,7,8, but there have been few experimental tests of this hypothesis. Moreover, many apparently adaptive floral mechanisms that ensure the autonomous production of selfed seed might use ovules that would have otherwise been outcrossed. This seed discounting is costly if selfed offspring are less viable than their outcrossed counterparts, as often happens. The fertility benefit of reproductive assurance has never been examined in the light of seed discounting. Here we combine experimental measures of reproductive assurance with marker-gene estimates of self-fertilization, seed discounting and inbreeding depression to show that, during 2 years in 10 Ontario populations of Aquilegia canadensis (Ranunculaceae), reproductive assurance through self-fertilization increases seed production, but this benefit is greatly outweighed by severe seed discounting and inbreeding depression.
Experimental emasculations were performed in 9 populations in 1998, and 12 in 1999. Removing anthers does not have any confounding effects in terms of either damage to flowers or shortened floral lifespan17,25. Emasculation does not greatly reduce the attractiveness of flowers to pollinators, although some decrease in outcross pollination as a by-product of emasculation does not alter the interpretation of our results. To minimize confounding variation in seed production and selfing among flowers owing to plant size and flower-position effects, we used pairs of plants with the same numbers of flowers, and focal flowers in the same position in the flowering sequence. In each population we used 40 pairs of plants in 1998, and 50–100 in 1999. Droughts reduced sample sizes below that required to estimate self-fertilization for 6 populations in 1998 and 2 in 1999. We include data only from populations for which we reliably estimated both seed production and selfing. For each population, estimates of mating parameters were based on progeny arrays of 10–30 seeds collected from each of 9–57 (mean 30) plants per treatment and assayed for two polymorphic allozyme loci21. The proportion of seeds produced through selfing (s) and the parental inbreeding coefficient (F) were estimated for the population as a whole and for both intact and emasculated flowers separately with the use of maximum likelihood28, and s.e.m. values were derived by bootstrapping (as described in ref. 21).