Patterns in ecology are the products of current factors interacting with history. Nevertheless, few studies have attempted to disentangle the contribution of historical and current factors, such as climate change and pollinator identity and behaviour, on plant reproduction. Here, we attempted to separate the relative importance of current and historical processes on geographical patterns of the mating system of the tree species Curatella americana (Dilleniaceae). Specifically, we asked: 1) How do Quaternary and current climate affect plant mating system? 2) How does current pollinator abundance and diversity relate to plant mating system? 3) How does mating system relate to fruit/seed quantity and quality in Curatella americana? We recorded pollinators (richness, frequency and body size) and performed pollination tests in ten populations of C. americana spread over 3,000 km in the Brazilian savannah. The frequency of self-pollination in the absence of pollinators was strongly influenced by historical climatic instability and not by present-day pollinators. In contrast, seed set from hand-cross and natural pollination were affected by pollinators (especially large bees) and temperature, indicating the importance of current factors on out-cross pollination. Two populations at the Southern edge of the species’ distribution showed high level of hand-cross-pollination and high flower visitation by large bees, but also a high level of autogamy resulting from recent colonization. Our results indicate that historical instability in climate has favoured  autogamy, most likely as a reproductive insurance strategy facilitating colonization and population maintenance over time, while pollinators are currently modulating the level of cross-pollination.

Study sites and species

We studied ten populations of Curatella americana in three disjunct areas of savannah. Vegetation physiognomies are very similar among sites, but in general plant species diversity decreases northwards (Ratter et al. 2003, Bridgewater et al. 2004). We observed animal pollinators and performed experiments on C. americana at all the studied sites. The species flowers from June to September in Central Brazil, mid-August to early October in Pará state, and October and November at Roraima state. Flowers are white, pentamerous and grouped into dense inflorescences, and each flower stays receptive for three to five hours for one single day (see Rech et al. 2018 for more details). 

Mating system

In order to study the reproductive system of C. americana in situ we applied the following pollination tests: hand-cross-pollination, hand-self-pollination, autogamous self-pollination and natural pollination. All pollination tests were performed with flowers previously bagged using cloth insect exclusion bags, except for natural pollination, which involved counting and tagging flowers exposed to flower visitors. In order to mitigate possible differences related to resource allocation we always performed the pollination tests on the same branch (considered as a functional unit). The number of tested flowers was always higher than 20 flowers per individual and a mean of 15 different individuals per test per population. In two of the studied areas (Nova Xavantina and Caldas Novas) we chose 12 individuals and compared the fruit weight from self (n = 107) and cross (n = 102) pollinated flowers, which may represent seed quality (Coomes & Grubb 2003).

Flower visitation and pollination

For all populations we recorded daily flower visitors (species richness and abundance) from anthesis until the end of visitation. In order to quantify visitation, we counted all visits to an observable (and counted) set of flowers for ten minutes each half an hour for at least 20 hours (120 x ten minute sessions) in each population. All the visitors touching anthers and/or stigmas were considered and scored as potential pollinators. After observing behaviour, flight distance and pollinator size, we grouped the pollinators into two categories: 1) Large-sized bees, and 2) Others, which includes bees the same size or smaller than Apis mellifera, beetles, flies and wasps. We separated pollinators according to size because flight range correlates with body size (Gathmann & Tscharntke 2002, Araújo et al. 2004, Greenleaf et al. 2007). Based on this premise, we expected a higher level of cross-pollination by large-sized bees.

Statistical analysis

To test for differences in fruit set related to the mating system and the regions, we used a Generalised Linear Mixed Model assuming a binomial distribution. The fixed factors were region, pollination experiment treatment, and the interaction between them. The random factors were the individuals nested within sites and these nested within regions. Our response variable was the production of a fruit from each flower. We performed the models with all fixed factor combinations and only a fixed intercept (Null Model), always keeping the random factor. For the fruit weight comparison we used pollination treatment (self- and cross-pollination) as predictors and generated models using individuals as random factors. All the alternative models were built removing factors or interactions between factors from the full model. A null model using only
178 the intercept was also considered. In order to compare the generated models we used the Akaike Information Criterion – AIC (Burnham & Anderson 2004). All tests and models were performed in the R environment (R Core Team 2018). For each studied site, we modelled the climate changes since Last Glacial Maximum (LGM) by estimating the mean annual temperature (MAT_LGM) and annual precipitation (MAP_LGM) at each location for 21ky, according to the Community Climate System Model (CCSM) (Gent et al. 2011). We also extracted the current values of temperature (MAT_Current) and precipitation (MAP_Current) from the Global Climate Data (Worldclim 1.4 -http://www.worldclim.org/). For each site, we calculated the anomalies and velocities of change in temperature (MAT_Velocity_21) and precipitation (MAP_Velocity_21), as the long-term average over the last 21ky. Both climate anomaly and velocity are measures of climate stability (or climate change), but they are calculated in two different ways. Whereas climate anomaly simply is the difference in climatic conditions between two time periods (today and 21,000 years ago), climate velocity integrates macroclimatic shifts (i.e. anomalies) with local spatial topoclimate gradients. Velocity is calculated by dividing the rate of climate change through time (i.e. anomaly) by the local rate of climate change across space (Sandel et al. 2011). All calculations are based on a 2.5 minutes geographical resolution. We then estimated the effect of climate and pollinator activity on pollination mode. Due to the modest sample size of populations (n = 10) and some predictor variables being strongly correlated (i.e. r ≥ 0.6; Table S2), we took the following modelling approach. First, we modelled the effect of climate on pollination mode using current and past climate predictors, identifying minimum adequate models (MAMs) using the approach outlined in Diniz-Filho et al. (2008). As the temperature and precipitation anomalies used as a measure of past climate stability were strongly correlated, we modelled the effect of temperature and precipitation anomaly separately. The effect of past climate stability was also tested using modelled temperature and precipitation velocity instead of anomaly, giving qualitatively the same results (not shown). Second, we tested whether the four pollinator variables (pollinator richness, visitation frequency, and proportion of large bee visitation calculated both with and without the exotic honey bee) were significantly related to pollination mode. To do this we used single correlation tests using traditional non-spatial correlation analysis and correcting the degrees of freedom using Dutilleul’s (1993) method (Table 1), followed by models testing whether each of these pollinator activity variables may have other or additional effects from climate. We examined this by again following the approach of Diniz-Filho et al. (2008) to identify MAMs, but this time only considering climate variables included in the above-identified MAMs and each of the four pollinator variables. For all analyses, MAP, MAP anomaly, MAP velocity and MAT velocity were Log10-transformed, pollination visitation frequency was square root transformed, and all proportional measures (i.e., pollination mode variables and large bee predictors) were arcsine-square root transformed. All other variables were left untransformed. All analyses were conducted using the software Spatial Analysis in Macroecology, SAM 4.0 (Rangel et al. 2010)