Pollinator availability , mating system and variation in flower morphology in a tropical savanna tree

Widely distributed organisms face different ecological scenarios throughout their range, which can potentially lead to micro-evolutionary differentiation at specific localities. Mating systems of animal pollinated plants are supposed to evolve in response to the availability of local pollinators, with consequent changes in flower morphology. We tested the relationship among pollination , mating system, and flower morphology over a large spatial scale in Brazilian savannas using the tree Curatella americana (Dilleniaceae). We compared fruit set with and without pollinators in the field, and analyzed pollen tube growth from selfand cross-pollinated flowers in different populations. Populations with higher natural fruit set also had lower fruit set in bagged flowers, suggesting stronger barriers to self-fertilization. Furthermore, higher levels of autogamy in field experiments were associated with more pollen tubes reaching ovules in self-pollinated flowers. Morphometric studies of floral and leaf traits indicate closer-set reproductive organs, larger stigmas and smaller anthers in populations with more autogamy. We show that the spatial variation in mating system, flower morphology and pollination previously described for herbs also applies to long-lived, perennial tropical trees, thus reemphasizing that mating systems are a population-based attribute that vary according to the ecological scenario where the plants occur.


Introduction
The interactions established between organisms can vary greatly throughout space and time (Thompson 2005).For instance, plants can interact with contrasting assemblages of pollinators throughout their geographical range (Gómez et al. 2014;Herrera 2005;Waser et al. 1996).The resulting pollination interactions can vary from specialist to generalist for both plants and animals, and may have strong implications for the evolution of plant mating systems (Gómez 2002;Dart et al. 2012;Barrett 2013).Historically, plant mating systems were considered an attribute of the species (Brys & Jacquemyn 2011;Rosas-Guerrero et al. 2014), even though their lability between populations was pointed out over 40 years ago (Stebbins 1974).Therefore, the mating system may vary within a species' range due to trade-offs between genetic variability, promoted by cross-pollination, and reproductive assurance from selfpollination, which vary in response to different ecological scenarios throughout space and time (Barrett et al. 1989;Moeller 2006;Opedal et al. 2016).For example, some herb species with mixed mating systems show different levels of autogamy throughout their ranges (Kalisz & Vogler 2003;Moeller 2006).
To rely on a mixed mating system, and oscillate between the two extremes of selfing and outcrossing as the environment demands, may be a consequence of the reproductive assurance hypothesis proposed by Darwin (1876), which was later re-elaborated (Baker & Stebbins 1965;Baker 1967).According to this hypothesis, mechanisms ensuring autogamy should be favoured in areas with low pollinator availability (Lloyd 1992;Herlihy & Eckert 2002).On the other hand, shorter-lived plants experience stronger selection for reproductive assurance (more prone to selfing), as they usually only experience one or few reproduction events early in life (Moeller et al. 2017).
Besides spatially structured processes, past climate dynamics and stability have also affected species distribution and diversity patterns (Svenning & Skov 2007;Cárdenas et al. 2011;Sandel et al. 2011;Kissling et al. 2012), population demography and genetic structure (Grazziotin et al. 2006;Cabanne et al. 2007), and the structure of mutualistic plant-pollinator assemblages (Dalsgaard et al. 2011;2013).Therefore, just as the lack of reproductive partners during range expansion or small population size may currently favour individuals with higher levels of autogamy (Ivey & Carr 2012;Levin 2012;Hargreaves & Eckert 2013;Griffin & Willi 2014), this could have happened over geological time.
Changes in mating systems are usually associated with variation in a group of floral traits (Berg 1960;Armbruster et al. 1999).For example, transitions to higher levels of autogamy are usually related to reduction in flower size and herkogamy (the separation between male and female reproductive organs within a flower -Wyatt 1988), faster flower development time (Armbruster et al. 2002;Mazer et al. 2004), and flowering earlier in the season (Mazer et al. 2004;Martin & Willis 2007).Other flower traits, such as the reduction of rewards and attractiveness in the display (Ornduff 1969;Dudley et al. 2007), are hypothesised to follow the previously mentioned primary modifications.Most research has focused on herb species, which are known to consistently present lower outcrossing rates (Moeller et al. 2017).Therefore, the life history (annual, perennial, etc.) and growth form (herb, vine, tree, etc.) are the main predictors of outcrossing rates among flowering plants (Moeller et al. 2017).To the best of our knowledge, only one studied has described the geographical variation in the mating system and flower morphology for a non-herbaceous species, in this case a vine (Opedal et al. 2016).
In a recent and extensive review of mating system variation among populations, the authors described such scenario as a widespread phenomenon weakly related to phylogeny or pollination mode (Whitehead et al. 2018).They concluded that ecological studies about the processes generating interpopulational variations in mating systems are still needed.With the aim of clarifying how pollinator availability relates to mating system and flower morphology in perennial plants, we studied populations of a widely distributed Neotropical tree Curatella americana (Dilleniaceae) (Ratter et al. 2003).Focusing on populations in three disjunct regions of savanna in Brazil, we tested the following hypotheses: 1. Effective pollinator availability (mean visitation frequency of pollinators per flower) is positively correlated to natural fruit set and inversely related to autogamy; 2. Smaller populations, or those at the distribution edge, will have higher levels of autogamy; 3. Populations with smaller flowers and less herkogamy will have higher levels of autogamy.

Study sites
We studied ten populations of the Neotropical savanna tree Curatella americana L. (Fig. 1), regarding pollinators, floral biology and morphology, mating system, and pollen tube growth under controlled pollination.Populations 5-10 (Fig. 1) were in the large continuous area of savanna (Cerrado) located in Central Brazil, hereafter called Southern populations.Populations 1-3 were in the disjunct area of Roraima state, hereafter referred to as Northern populations and population 4 was in a fragment of savanna surrounded by the Amazon forest, hereafter called Santarém.Populations 9 and 10 (Jatai and Caldas Novas; Fig. 1) belong to the Southern region, located at one of the southern edges of the C. americana distribution in South America, but are still connected to the large continuous area of the distribution.These two populations therefore experience different scenarios when compared to the disjunct areas, as they are edge populations with much greater area.Such difference is important to note when considering the findings of this study (see Results).
Floral biology and autogamy were studied in all populations, but morphology for the purpose of morphometric analysis was studied in five populations from all the regions (2, 3, 4, 5 and 6; Fig. 1).The closest populations were 100 km apart and the furthest were more than 2,700 km apart from each other.

Floral biology
Floral biology measurements sensitive to dehydration were carried out in situ or, in the case of herkogamy, from pictures of the flowers in standard position using a fixed holder connected to the camera (Canon EOS Kiss X5).Time, sequence, and duration of anthesis, flower longevity, stigma receptivity, odour emission (presence of osmophores), and pollen availability (time of pollen presentation) were observed and described following Dafni et al. (2005).Receptivity was assessed by dropping hydrogen peroxidase onto the stigmas and checking for bubbles.The location of osmophores was visually assessed after immersing flowers in a solution of neutral red for five minutes.
Flowers were collected from five populations (2, 3, 4, 5 and 6; Fig. 1) and fixed in 70 % alcohol for linear or size measurements in the lab.The length of stamens (total and anther) and size of pistil (total length -ovary+style+stigmaand stigma area) were measured using a digital calliper.Measurements of shape of petal, sepal, ovary, and leaf (as a control) were obtained from pictures and analysed using geometric morphometrics (Bookstein 1991).Landmarks and semi-landmarks were placed on scanned structures using the software TPS Dig (Rohlf 2010).Semi-landmarks were used to define the borders of the structures when a homologous landmark was not available (Fig. S1  average of visitation rate for the observed flowers was then used to predict natural fruit set (NF) and autogamy (AU) in each area.During the other twenty minutes of observation we recorded visitors' behaviour on flowers and observed their potential as pollinators by looking for contact between insects and reproductive structures.
To check for possibile self-pollen deposition, we also bagged flowers one day before anthesis and then collected one stigma per flower as soon as the flower opened and another stigma four hours later.Each flower had two equal stigmas, which made the test perfectly paired.The stigmas were placed directly on a slide with stained glycerine jelly, covered with a cover slip, and pollen grains were counted under light microscopy.The difference in the number of pollen grains was tested by a paired t test.
To test for possible accidental pollen deposition during manipulation, some stigmas were manipulated and the second stigma was removed 10 minutes later.Since the number of pollen grains on stigmas was not different between these two treatments, we assumed that accidental deposition had a minor effect and would not change the experimental results.

Pollen tubes
Pollen tubes were observed to identify possible mechanisms operating at the pistil level, related to the different pollination treatments in the studied populations.In order to check e pollen tube growth through the style, we collected ten hand-pollinated pistils (five cross and five self-pollinated) from ten individuals using exactly the same protocol in each population.The pistils were fixed in FAA (Formalin-Acetic acid-Alcohol) + ethanol 50 %, 24 hours after hand pollination.The preparation followed the technique proposed by Martin (1959) and pistils were stained with a solution of aniline blue.Since pistils were covered by trichomes, we pre-treated them in a solution of sodium hypochlorite (10 %) and distilled water to facilitate trichome removal.Even after such treatment it was still very difficult to count individual pollen tubes, therefore, results were assessed qualitatively through visual comparisons of the ten slides from each population under fluorescence microscopy.

Morphometric analysis
We separated the form of all the landmark configurations into shape and size using geometric morphometrics (Bookstein 1991).Size was measured as centroid size (CS), the square root of the sum of the squared distances of each landmark from the centroid, or gravity centre, of the landmark configuration.Size variation among populations and regions was tested using mixed models, comparing a set of three models: one with region, another with population, and a null model with only the intercept as a fixed effect.
Individual was included as a random effect in all models.Models were compared using AICc (Burnham & Anderson 2002).
To measure shape, all configurations were scaled to unit CS, and superimposed by a generalized least squares (GLS) Procrustes procedure.Given the presence of semilandmarks, a sliding procedure minimizing the GLS residuals was used, as the exact location of the semi-landmarks along the structure outline was arbitrary.A mean shape was calculated and the differences between its landmarks and the landmarks of each individual were the residuals of the GLS procedure.To test for differences in shape among the regions, we ran a Procrustes Analysis of Variance (ANOVA) using a hierarchically nested design, with replications (repeated measurement of the landmarks from leaves, petals, etc.) nested within individuals, populations, and regions.Differences in shape among individuals, populations, and regions were visualized by doing a Principal Components Analysis on the superimposed shapes.All geometric morphometrics analysis were done using the geomorph package in R (Adams & Otarola-Castillo 2013).
For the univariate flower morphology measurements (number of stamens, stigma area, anther size, gynaecium, and androecia length) we used a one-way ANOVA, and when results were significant, the means were further differentiated using a Tukey multiple comparisons test (at p ≤0.05).All analyses were carried out using the R software environment (R Development Core Team 2014).

Floral biology
Basic flower biology was observed in the individuals where the experiments were performed in all populations.In total, about 68,000 flowers were manipulated and at least 680,000 were observed.Curatella americana flowered from June to September in the Southern Cerrado (populations 5-10; Fig. 1), from September to October in Santarém (population 4), and from October to November in the Northern Cerrado (populations 1-3).Individual trees massflowered, producing all of their flowers within a short period of two weeks.Flowers are pentamerous, actinomorphic and open (plate type), grouped in panicles, white to the human eye, and UV absorbing (Fig. 2

Pollination and pollen tubes
Flowers of C. americana were visited by bees, flies, and beetles (Fig. 2, Tab. 1).Due to visitor behaviour (contact with sexual organs), mobility, and flight ranges observed in the field, large-sized bees (larger than Apis mellifera) were considered the main pollinators (for more details see AR Rech et al. unpubl. res.).The pollinator visitation probability was highest in the Southern populations (Tab.2).Fruit set under natural pollination and visitation rate of main pollinators were positively correlated (R2 = 0.49, P1,8 = 0.02, Tab. 2).On the other hand, the relationship between autogamy and pollinator availability was negative and significant only when the two Southern edge populations were not considered (R2 = 0.18, P1,8 = 0.61 for all populations and R2 = 0.65, P1,8 = 0.02 for the restricted analysis).Similarly, the fruit set under natural pollination only negatively correlated to the level of autogamy when the same two populations were not considered (r = -0.29,P8 = 0.40 for all populations and r = -0.78,P6 = 0.02 for the restricted analysis).The two populations from the Southern edge had similar natural pollination fruit sets as the other populations in Central Brazil, but autogamy levels were like the populations from the North.Since the two Southern edge populations face a special biogeographical scenario, we ran the analysis with and without these populations and compared the results (see Materials and methods, and Discussion).The amount of pollen on stigmas increased from 52 pollen grains at the beginning to 141 pollen grains per stigma at the end of the anthesis in bagged flowers (T8 = -4.96,p< 0.001), indicating the potential for spontaneous self-pollination in all populations.Nevertheless, the level of autogamy varied considerably among populations and individuals (Tab.2).These differences in fruit set can be directly related to the way pollen tubes grew in the different treatments.Higher amounts of pollen tubes grew regularly and continuously through the styles of cross-pollinated flowers.Pollen tubes in self-pollinated flowers grew irregularly (strong callose reactions) and the difference between treatments (self-and cross-pollination) was clearly more contrasting in the Southern region than in the North (Fig. 3A-D).Most of the pollen tubes in selfpollinated flowers stopped growing in the first two thirds of the style.Therefore, we only saw pollen tubes in the self-pollination treatment reaching the ovules 24 hours after hand pollination in flowers from the Northern region (Fig. 3E-F).

Morphometrics
The size (measured as centroid size) of petals, sepals, and leaves did not differ between regions, however, the linear lengths of anthers and gynoecium, and stigma area, varied significantly (Tab.3 Fig.4).For floral structures (petal and sepal), the null model including only individual size variation regardless of population and region was as likely as the more complex models.For leaves, the best model included differences among populations, but only the model with regions did not perform well, indicating a more idiosyncratic population-level variation in size instead of a geographical pattern.The number of stamens did not vary among regions (Tab.3).Anthers were larger in the Southern region, while stigma area was larger in the North.Santarém was similar to Southern populations regarding anther size, and to the Northern populations regarding stigma area (Tab.3, Fig. 4).
On the other hand, there were regional differences in shape for all traits considered, even though the proportion of variation among regions varied widely among structures (Tab 4).Although for the ovary, region was the main correlate of shape variation, for the other structures it had a minor role (Tab.4 , Fig. 5).Population-level variation in shape within regions had a minor contribution to all structures, whereas individual variation was always high.Given the major role of regions in ovary shape variation, a clear geographic gradient was observed (Fig. 4), where northern populations presented the two stigmas close together, while in southern populations stigmas were further apart.Santarém, geographically located between populations from the other regions, had an intermediate ovary shape (Fig. 4 and Fig. S2 in supplementary material).For the other structures (leaves, petals, and sepals), no clear patterns of geographic variation were present (Fig. S2 in supplementary material).

Floral biology, pollination, and pollen tubes
The mass-flowering pattern of C. americana was already reported for another Neotropical Dilleniaceae species -Davilla kunthii (Rech et al. 2011).Apart from flower colour (white in C. americana and yellow in D. kunthii) and the closing sepals of D. kunthii, both species have similar floral biology, pollination, and mixed mating systems (Rech et al. 2011).The open flower morphology with exposed anthers makes it easy for visitors to contact anthers and stigmas in a single visit, however, many flower visitors hardly moved between individuals, making them unlikely to be significant pollen vectors (Ollerton et al. 2007).Pollen as the only reward and the white UV absorbing pattern found in the petals, indicates a flower that is attractive to bees (Lunau et al. 2011), which was clearly confirmed by visitation frequency (AR Rech et al. unpubl. res.).
Some populations presented very low levels of natural fruit production, as well as low availability of effective pollinators.We interpreted the negative relationship between fruit set under natural pollination and autogamy as an assurance strategy selecting autogamy in areas with low pollinator availability, herein, the Northern region and populations at the Southern edge of the Brazilian Cerrado (Baker 1967;Cheptou 2012).Two populations at the Southern edge of the C. americana distribution (Jatai and Caldas Novas) presented natural fruit sets that were equal to the populations in Central Brazil, but the levels of autogamy were similar to the populations in the North.Palynological evidence shows that the Southern edge of Brazilian savannas presented very dry, treeless environments from 18,000 to 6,000 bp (Salgado-Labouriau et al. 1997).Only from 5,000 bp onwards did trees, such as C. americana, colonize the area.Thus, the level of autogamy in the area may still be a consequence of the colonization process.Indeed, genetic data suggests a recent range expansion for C. americana in Brazilian savannas (Canuto 2011).Also, changes in mating systems may be flexible, as seen in artificial selection experiments (Levin 2012).Furthermore, two generations of strong induced directional selection were enough to elevate the autogamy level from 4 % to 56 % in Phlox drummondii (Bixby & Levin 1996).Therefore, past climate and colonization events related to expansion of Neotropical Savannas could underpin the distribution and consequently the contemporary autogamy rate found in C. americana (Canuto 2011;AR Rech et al. unpublished res.).
Curatella americana has a lifespan of over 200 years (Costa 2013), with a slower process of population replacement when compared to short-lived herbs such as P. drummondii.Therefore, if there is an ecological filter preventing colonization by animal-pollination dependent individuals, the establishment of a predominantly autogamous population could be very quick, as is the case for some invasive plant species (Ollerton et al. 2012).Nonetheless, recovery of higher rates of self-incompatibility when population size increases and pollinators come back into the environment can be slower, specially for long lived species like C. americana.It is important to consider that populations are normally not exclusively self-pollinated, so the unfavourable effects of a pure strategy may be reduced in mixed mating systems (Goodwillie et al. 2005).We also cannot rule out the possible effect of inbreeding depression caused by selfing at edge populations, which leads to less visitation and should be the focus of future genetic studies.

Morphology
Most of the traits studied here had a latitudinal gradient of variation.Moreover, individuals from Santarém frequently presented intermediate shape and size compared to Northern and Southern populations and were more similar to one region or another depending on the structure considered.This seems to be related to a similar palaeoecological history as the Northern region, and a pollinator availability similar to the one found in the Southern region.All the savannas studied herein, currently within isolated regions, were probably connected in the past and the Northern region was probably the first to disconnect from the others (Mayle & Power 2008).At the regional level, floral and vegetative traits varied similarly but with different intensities.The synchronized variation in floral and vegetative traits in plants with generalized pollination systems (open flower morphology and many unrelated species of pollinators) is one of the expectations to "Berg's rule" (Berg 1960;Fenster 1991;Conner & Sterling 1995;Armbruster et al. 1999).As floral traits derive ultimately from the vegetative ones, decoupled patterns of variation are not expected unless clearly different regimes of selection drive the evolution of different groups of traits in different ways (Berg 1960;Armbruster et al. 1999).However, the variation pattern found in the ovary structure is much clearer than in the other structures.Therefore, we interpreted the progressive segregation of stigmas from north to south as a strategy selected to increase genetic diversity by touching different pollinators or pollinators' body parts during visits.
However, contrary to the expectation, not all traits varied in the same direction as an integrated unit or "pleiades" (Berg 1960).Stigma area was larger and there was less herkogamy in the North.These two traits may be related to female reproductive insurance, as both increase the chance a flower will be self-pollinated (see review in Levin 2012).The same pattern was also reported in Nicotiana glauca after it was introduced to parts of the world where there were no suitable pollinators present, reinforcing the role pollinators have in mediating the phenotypic variation of flowers (Ollerton et al. 2012).On the other hand, anthers were larger in the Southern populations.Considering that anther size and pollen production are positively correlated (Harder & Thompson 1989;Philipp et al. 1990), and pollen removal can be related to seed set (Broyles & Wyatt 1990), there seems to be a pressure on the male component (pollen production) of fitness in areas with higher levels of crosspollination (Lloyd 1984).
Finally, we showed that floral structures did not have an integrated pattern of variation across regions and populations.Conversely, different flower traits may respond independently to the pressures imposed by pollinator availability and the importance of autogamy in the different populations (Rosas-Guerrero et al. 2010;Opedal et al. 2016).These divergent responses fulfil the function of the flower by maximizing reproduction, with flexibility to adapt to different scenarios.

Conclusions
Trees have longer lives, and, therefore, slower individual substitution rates than most herbs and smaller woody plants.However, herein we found a similar pattern of variation in the mating systems and flower morphology of a tropical tree, as previously demonstrated for herb and a vine species, indicating colonization filters and pollinator availability as important drivers of flower morphology and mating system (Opedal et al. 2016).The results discussed here re-emphasize mating systems as a property of populations and not of species (Whitehead et al. 2018).Even though, in general, longer-lived organisms seem to have higher outcrossing rates, pollinator availability may also help explain local reproductive strategies and interpopulation variations in mating system.

Figure 1 .
Figure 1.Map of the original area of Brazilian Cerrado (Neotropical Savanna) and indications of the studied populations.Populations 1, 2 and 3 are included in the region called Northern Cerrado, 4 is in Santarém, and 5-10 are in the Southern Cerrado region.

Figure 3 .
Figure 3. Pollen tube growth in flowers of Curatella americana after different pollination tests.A-B.Cross and self-pollination in one population from Southern Cerrado region (Cuiabá).C. Self-pollination in one population from the Northern region (Boa Vista).D. Self-incompatiblity reaction (black arrow) in one self-pollinated flower from Southern Cerrado region (Manso).E. pollen tube growing through the base of the style and F. reaching the ovule (white arrow) at the base of ovary in one population from the Northern Cerrado region (Boa Vista).

Figure 5 .
Figure 5. Principal component analysis (PCA) using geometric morphometric coordinates for ovary shape of Curatella americana.Each dot is an individual measure projected according to the first two axes of the PCA.Colours indicate regions: orange is Santarém, red is Northern Cerrado and blue is Southern Cerrado.Different shapes indicate different populations within each region.

Figure 4 .
Figure 4. Comparison (ANOVA) of floral traits of Curatella americana from different regions in the Brazilian Cerrado (Neotropical savanna).

Table 1 .
Flower visitors registered in the flowers of Curatella americana in the ten studied populations in Brazil.Curculionid Beetles were also observed but not quantified as the stay for the entire observation period without moving from the observed flower.

Table 2 .
Average proportion of fruit for each of the ten studied populations of Curatella americana in Brazil.Autogamy refers to bagged inflorescences with no pollinator access while natural pollination is the fruit set in exposed flowers and cross-pollination (cross) is the supplemented hand cross-pollination treatment.Single flower visitation (SFV) refers to the mean number of visits by large-sized bees to a single flower in each population per day.Population numbers follow Figure1.a Many fruits were destroyed by herbivores and could not be counted.

Table 3 .
Analysis of Variance (ANOVA) for counted and linearly measured floral traits of Curatella americana from ten populations of three disjunct regions of Brazilian savanna (Southern Cerrado, Santarém and Northern Cerrado).

Table 4 .
Nested Procrustes ANOVA for shape of floral and leaf traits of C. americana in the three geographical regions studied (Southern Cerrado, Santarém, and Northern Cerrado).DF: Degree of freedom, MS: Mean Square Sum, p: p value.