Appendix 2- Fecundity components

Statistical analyses

Between-year and among-populations variation in fecundity components (see distribution used bellow) were analyzed using Generalized Mixed Model with individuals as a random factor. Differences between levels of significant factors were estimated using contrasts corrected by Bonferroni sequential tests (Rice, 1989).

The four different fecundity components measured (fruit set, seeds/flower, seed weight and seed germination) were separately fitted to General or Generalized Linear Mixed Models. We analyzed the fecundity components separately to check whether floral traits and robbing frequency could have contrasting influences on different fecundity components. These models were conducted separately for each population and included year as fixed factor; corolla length, nectar volume, lip size and robbing frequency as continuous, predictor variables; the interactions between all these factors and covariates; and individual plant as random factor. Due to the nature of the data, we used: 1) Binomial distributions and logit link for fruit set and seed germination; 2) Poisson distribution and log link for seed set; and 3) Gaussian distribution and identity as link for seed weight. For the analysis of seed germination, seed weight was included as an additional predictor variable. None of the binomial or poisson models showed overdispersion, since in all of them the ratio of residual deviance to residual degrees of freedom was < 1.2 (Zuur et al. 2009). All predictor variables showed VIF values smaller than 3 and were, therefore, included in the analyses (Zuur et al. 2009). Prior to the analyses we specified sum-to-zero contrasts to ensure that the drop1() function performed Type III tests. We used Akaike information criterion (AIC) to select the best models amongst the full model and possible subsets of variables. All the analyses were conducted in R (version 2.15.3; R Development Core Team 2008).

Results

Two fecundity components (fruit set and seeds/fruit) varied non-additively among populations and between years (population × year: χ22 =16.79, P = 0.0002; χ22 =17.07, P = 0.0002; χ22 =8.92, P = 0.012, respectively). In contrast, the other two components (seed weight and seed germination) did not differ significantly among populations, between years, or with their interaction (all P > 0.53). In 2011, the hawkmoth population was the population with the highest, and the short-tongued population the population with the lowest, fruit set, seed set and total fecundity in 2011; whereas this pattern was reversed in 2012.

Hawkmoth population

Fruit set increased with corolla length but only when flowers contained small volumes of nectar, while the fruit set of flowers with abundant nectar was not affected by corolla length (Table S1A; FigS1A). The number of seeds per fruit tended to decrease with nectar robbing, but only in 2011, when it was much more frequent (Table S1A; Fig S1B). In contrast, seed germination increased with nectar robbing (Table S1A; Fig S1C). Seed weight was not affected significantly by any of the predictor variables or their combinations (all P ≥ 0.32).

Short-tongue population

Fruit set decreased with corolla length in 2012, but not in 2011; and the interaction between corolla length and nectar volume was marginally significant (Table S1B; Fig. S2A). In 2011, when nectar robbing was more frequent, it resulted in increased fruit set (Table S1B; FigS2B). None of the predictor variables or their combinations had significant effects on the other three fecundity components (all P ≥ 0.23), with the exception of a significant effect of seed weight on seed germination (Table S1B; Fig. S2C).

Mixed population

Fruit set increased with corolla length and decreased with nectar volume, but only in 2011 years (Table S1C; Fig. S3A and C). It also increased slightly with robbing frequency in both years (Table S1C; Fig. S3B). Seeds/fruit and seed weight were not significantly affected by any of the predictor variables or their combinations (all P ≥ 0.15). Seed germination increased marginally significantly with corolla length (Table S1C; Fig. S3D), and significantly with robbing frequency (Table S1C; Fig. S3E) and seed weight (Table S1C; Fig. S3F).

Table S1 Results of Generalized Linear Models relating four predictor variables (corolla length, nectar volume, lip size and robbing frequency) and their one-way interactions, to three components of plant fecundity (fruit set, seed set, and seed germination). Separate models were fitted to each dependent variable at each population. The significance (based on likelihood ratio tests; LRT) of the variables included in the best model (selected using AIC) is given. All the variables included in interactions were also present in the fitted models, but LRT are not calculated for continuous variables when they appear in significant interactions with fixed factors.

Population / Variable / Fruit set / Seeds/fruit / Seed germination
A) Hawkmoth population / Year / χ 12= 6.72, P = 0.0003
Corolla length / χ 12= 6.26, P = 0.012
Nectar volume / χ 12= 3.56, P = 0.059
Robbing frequency / χ 12= 8.26, P = 0.004
Robbing frequency × year / χ 12= 3.07, P = 0.090
Corolla length × nectar volume / χ 12= 4.76, P = 0.029
B) Short-tongued population / Year / χ 12= 17.38, P < 0.0001
Nectar volume / χ 12= 2.73, P = 0.099
Corolla length × year / χ 12= 13.82, P = 0.0002
Robbing frequency × year / χ 12= 13.40, P = 0.0002
Corolla length × nectar volume / χ 12= 3.42, P = 0.064
Seed weight / χ 12= 19.24, P < 0.0001
C) Mixed population / Year / χ 12= 8.01, P = 0.005
Corolla length / χ 12= 3.28, P = 0.070
Robbing frequency / χ 12= 4.58, P = 0.032 / χ 12= 6.66, P = 0.010
Corolla length × year / χ 12= 3.94, P = 0.047
Nectar volume × year / χ 12= 7.17, P = 0.007
Seed weight / χ 12= 10.49, P = 0.001

Fig. S1. Partial residual plots showing the effects of corolla length and nectar volume on fruit set (A), and robbing frequency on seeds /fruit (B) and seed germination (C) in the hawkmoth population. Several lines are used, when significant interactions were detected, to show estimated effects of one predictor variable (X-axis) for either different years or different levels of other continuous variable (nectar volume).
Fig. S2. Partial residual plots showing the effects of corolla length and nectar volume on fruit set (A), robbing frequency on fruit set (B) and seed weight on seed germination (C) in the short-tongued population. Several lines are used, when significant interactions were detected, to show estimated effects of one predictor variable (X-axis) for either different years or different levels of other continuous variable (nectar volume).
Fig. S3. Partial residual plots showing the effects of corolla length (A), robbing frequency (B) and nectar volume (C) on fruit set, and corolla length (D), robbing frequency (E) and seed weight (F) on seed germination, in the mixed population. Several lines are used, when significant interactions were detected, to show estimated effects of one predictor variable (X-axis) for either different years or different levels of other continuous variable (nectar volume).

References

R Development Core Team. 2008. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. ISBN 3-900051-07-0, URL http://www.Rproject. Org

Rice WR (1989) Analyzing tables of statistical tests. Evolution 43: 223-225.

Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM (2009) Mixed effects models and extensions in ecology with R. Springer, New York.