Bumble Bee Selection Of Mimulus Guttatus Flowers: The Effects Of Pollen Quality And Reward Depletion

Ecology, Dec, 1999 by Alastair W. Robertson, Claire Mountjoy, Brian E. Faulkner, Matthew V. Roberts, Mark R. Macnair

Outdoor plot trials 1985-1991

The nine plots that were set outdoors to test for visitor preferences for viable pollen production yielded inconsistent results (Table 2). In some plots, bees were clearly discriminating in favor of plants that produced better quality pollen (e.g., plot 86-1, Fig. 2). In other plots, however, there was an insignificant relationship between viable pollen production and visitation rate, based either on the mean number of flowers visited, or visits per flower per hour. Such inconsistency could be either the result of an effect that truly varies in magnitude, or the result of Type II sampling error. We therefore carried out a meta-analysis of these correlations, and these comparisons, using only the most common bee species for each plot, are presented at the bottom of Table 2. For each combination of dependent (visits per flower per hour or flowers visited at a plant) and independent variables (viable pollen number or proportion viable), a combined estimate of the overall effect size (p) and its statistical sign ificance is presented along with a test of the heterogeneity of the plot correlations (Q). These analyses reveal that, overall, there is a highly significant positive relationship between both measures of pollen production and the two measures of visitation rate. However, the results also show that the relationship between mean numbers of flowers visited per plant and the proportion of viable pollen varied significantly among plots. The results suggest that although there is a moderately strong relationship between this parameter and pollen production, its strength varies significantly among plots and that the variation is not due simply to Type II sampling error. Visits per flower per hour, which was measured only on later plots, showed an overall positive, significant and consistent relationship with both aspects of pollen production.

Pairwise choice tests

When individual workers of Bombus pratorum were presented with a pair of genotypes of Mimulus, 21 of the 26 bees discriminated significantly against one of the pair presented (binomial tests with an expected proportion of 50% visits to each genotype). To examine the relative importance of the four predictors of pollinator preference, a logistic regression was performed with pollinator preference as the dependent variable and the measures of pollen and nectar production as independents (Table 3). The results for individual bees were pooled to obtain independent data points. This analysis was performed with S-PLUS (Mathsoft 1997) using a binomial distribution and a logit link function. The full model was highly significant and explained 40% of the variance in the choices bees made. A stepwise removal procedure using Aikake Information Criterion (AIC) (Aikake 1974) did not remove any of the predictors from the final model, suggesting they all have unique predictive power. The number of viable pollen grains was the single most important predictor of bumble bee behavior (explaining 20% of the variance in pollinator behavior). Nectar standing crop was positively associated with preference but explained only 3% of the variance. The proportion of pollen that was viable, and the total pollen production were negatively correlated with preference, after the other two variables had been entered. This suggests that bumble bees preferentially visit genotypes that have high numbers of viable pollen but an otherwise low total pollen production. Our unpublished data show that the viable grains in a mixed pollen type genotype tend to be larger than pollen produced in flowers that have only viable pollen.