Geographic variation in floral traits and the capacity of autonomous selfing across allopatric and sympatric populations of two closely related Centaurium species

Floral traits and the relative contribution of autonomous selfing to total seed set varies geographically and is often driven by the availability and abundance of suitable pollinators and/or the presence of co-flowering relatives. In the latter case, competition for pollinator services and costs of hybridization can select for floral traits that reduce interspecific gene flow and contribute to prezygotic isolation, potentially leading to geographic variation in floral divergence between allopatric and sympatric populations. In this study, we investigated variation in floral traits and its implications on the capacity of autonomous selfing in both allopatric and sympatric populations of two closely related Centaurium species(Gentianaceae) across two distinct geographic regions(UK and mainland Europe). Although the magnitude and direction of floral differentiation varied between regions, sympatric populations were always significantly more divergent in floral traits and the capacity to self autonomously than allopatric populations. These results indicate that mating systems can vary substantially within a species and that the joint occurrence of plant species can have a major impact on floral morphology and capacity of autonomous selfing, most likely as a way to reduce the probability of interspecific interference.


Material and Methods
Study species. Centaurium erythraea and C. littorale are biennial, monocarpic herbs belonging to the Gentianaceae. Centaurium littorale is often found in young, wet dune slacks, prefers moist and saline conditions and is a flood resistant species 37,38 . Its distribution is restricted to coastal areas from southern Scandinavia to the UK, France and Eastern Europe (Fig. 1). In contrast, C. erythraea is tolerant of a wider range of environmental conditions and can be found in a much broader variety of habitats, including coastal dunes, river banks, wood margins and calcareous grasslands. Distributed over most of Europe, its southern and eastern range ends in North Africa and Afghanistan, respectively 39 (Fig. 1). In coastal areas, the distribution of both species frequently overlaps, resulting in sympatric populations at many locations throughout the distribution range of C. littorale 34,[37][38][39] (Fig. 1).
For both species, first year juveniles overwinter as a basal rosette and generally flower in the next growing season. Both species produce numerous flowers that show a comparable floral morphology on one or more flowering stalks. C. erythraea produces showy, pink flowers while C. littorale flowers are often mauve. Once the flowers open, they last for 4 to 5 days 35,36 . Both species flower from the end of June until the beginning of September 33 . Flowers of both species do not produce any nectar and largely share the same pollinator community. Insects reported to visit the flowers of both species include pollen-gathering hoverflies (Diptera, Syrphidae), bees (Hymenoptera, Apidae) and small flies (Empididae-Muscidae) 11,35 . If a flower is successfully pollinated, fruits produce large amounts (241.1 ± 51.7 and 206.5 ± 42.8 seeds per fruit in C. erythraea and C. littorale respectively) of tiny seeds (< 0.01 mg) 35 . When successful pollinator-mediated pollination fails, both species are able to set fruit through autonomous self-pollination. C. erythraea and C. littorale are self-compatible and spontaneous selfing occurs through anther-curling 35 . Because of incomplete reproductive isolation, both species have been shown to hybridise in the field 33,34,36 . Study area and populations. To assess geographical variation in flower morphology and to investigate the impact of interspecific interference on floral traits, 36 populations of C. erythraea and C. littorale were studied across two disjunct geographic regions in Europe. Twenty-one populations were located on mainland Europe (along the coast of Belgium and the Netherlands), whereas fifteen populations were sampled in the UK (Scotland, northern Wales and England) (Fig. 1, Table 1). Continental populations included six allopatric C. erythraea, five allopatric C. littorale populations, and ten sympatric populations. In the UK, six allopatric C. erythraea, five allopatric C. littorale populations, and four sympatric populations were sampled (Table 1). All populations were sampled in coastal areas and dune slacks of the two regions, so that all locations were similar, i.e. dominated by saline and moist conditions. All allopatric populations were distinct entities that occurred at least 500 meters from another population of the same or the other species, whereas in sympatric populations, plants of both species Floral measurements. In each population, variation in floral traits was stable, so we selected 30 plants and of each individual one freshly opened flower was collected during peak flowering. Each of these flowers were opened longitudinally and scanned to obtain a cross-section of the flower that was used for floral measurements. For each flower, six floral traits (total flower length, petal length, petal width, pistil length, stamen length and the degree of herkogamy) were measured using ImageJ software 40 (Fig. 2). Herkogamy was determined as the distance between the pistil and the first anther. Because previous research 35 has shown that these floral traits do not vary within the same plant, harvesting a single flower per plant was sufficient to characterize variation in overall floral morphology among populations and species.
The contribution of autonomous selfing to total seed production under natural field conditions. To quantify the extent of pollinator-mediated seed set and to assess the contribution of autonomous selfing to total seed production, floral manipulations were performed using methods outlined in Eckert et al. 7 in two allopatric and two sympatric populations per species in each geographic region (UK and the European mainland) (12 sites, 16 populations in total). In each population, we selected and marked 30 similar sized plants throughout the population representing the average size and floral display of that particular population. For each individual, plant height, the number of flowers and the number of flowering stalks was measured. Afterwards, two pollination treatments were applied on eight flowers in total. Four flowers were emasculated immediately after anthesis and before anthers dehisced to obtain an estimate of pollinator-mediated seed production, whereas the other four flowers were left unmanipulated and served as controls. Each of these flowers were exposed to open natural pollination in the field. Because only a few flowers per plant were assigned to the emasculation treatment in this study, we assumed that emasculation did not impact plant attractiveness and pollinator services in this experiment. Moreover, previous studies have indicated that emasculation of flowers in both species did not affect the average life span of a flower nor floral visitation rates 35,41 . These studies also suggest that pollinator behavior on flowers is unlikely to be affected by emasculation, so that all floral visits are of similar quality.
Once fruits were ripe, each marked fruit was harvested and the number of seeds per fruit was counted. For each plant, average fruit-level seed production per treatment was determined by calculating the average number of seeds. With these data, we also calculated for each plant an index of reproductive assurance (RA) as [1 -(seed set of emasculated flowers/seed set of intact flowers)] 7,42 . Data analysis. To test whether overall floral morphology differed between allopatric and sympatric populations and whether these differences were affected by region and species, a multivariate analyses of variance (MANOVA) was applied using the raw data of floral measurements, with region, species and allopatric/sympatric populations and their interactions as fixed effects. The three-way interaction was also included in the model. Subsequently, two-way MANOVAs were conducted for each region separately to investigate in more detail how co-occurrence of the two species affected their overall morphology. In this analysis, species and whether populations were allopatric or sympatric as well as their interaction were used as explanatory variables in the model. Subsequently, univariate analyses of variance were performed per species to see which of the six floral traits differed significantly between population type.
To investigate whether differences in floral traits between the two species were larger in sympatric than in allopatric populations, we calculated for each floral trait the average difference between population means of the two species. The average difference was calculated for all combinations of C. erythraea and C. littorale populations, and done separately for allopatric and sympatric populations. These averages display the magnitude of the difference between the two species in allopatry and sympatry. Finally, the average differences in floral traits between both species were compared between allopatric and sympatric populations. All calculations were done separately for populations on the European mainland and the UK. To compare the average difference in plant height, number of flowers and flowering stalks with differences in floral traits and reproductive assurance, we used the same calculations for average difference as in floral traits, mentioned above. All measurements are represented as means and standard errors.
To test whether the contribution of autonomous selfing to total seed production (i.e., RA) was correlated with any floral traits, we calculated the Pearson correlation coefficient for all combinations of RA and the six floral traits. To test whether the contribution of autonomous selfing to total seed production (i.e., RA) differed between allopatric and sympatric populations and whether these differences were affected by region and species, we used a generalized linear mixed model (GLMM) with the index of reproductive assurance (RA) as dependent variable and region, species, population type (allopatry/sympatry) as fixed effects. Interaction effects were also included in the model and population was included as a random factor. This analysis was conducted in SAS 9.1 (SAS Institute, 2005), with the GLIMMIX procedure and logit link function. All other analyses were performed using SPSS 43 and R 44 . All graphs were made with Sigmaplot (Systat Software, San Jose, CA).

Results
Floral morphology. The results of the MANOVA showed significant (P < 0.05) differences in floral morphology between C. erythraea and C. littorale, between allopatric and sympatric populations and between regions (Table 2). Moreover, interactions between these factors were significant as well, indicating that differences in floral morphology were largely context-dependent. Region had by far the largest effect on overall morphology (Table 2), followed by species and their interaction, indicating that the two species differed significantly in floral morphology depending on the region where they were growing. Interestingly, whether a species was growing in allopatric or sympatric populations also significantly affected floral morphology, and this effect in turn interacted with region and species. Finally, the three-way interaction between region, species and allopatric/sympatric populations was also significant ( Table 2).
Analyses for each region separately showed that floral morphology differed significantly between plants of allopatric and sympatric populations, and that this effect was dependent on species (Table S1). Univariate analyses  of variance showed that total length (F 1,268 = 12.41; P = 0.001), petal length (F 1,268 = 59.74; P < 0.001) and petal width (F 1,268 = 69.54; P < 0.001) of C. littorale flowers on the European mainland were significantly smaller in sympatric than in allopatric populations, indicating that flowers of C. littorale in sympatric populations were in general smaller than those in allopatric populations (Fig. 3, Table S2). In contrast, total length (F 1,369 = 37.76; P < 0.001), petal length (F 1,369 = 51.37; P < 0.001) and petal width (F 1,369 = 37.53; P < 0.001) were significantly larger in sympatric populations of C. erythraea than in allopatric conditions, indicating that flowers of C. erythraea are larger under sympatric conditions (Fig. 3, Table S2). The univariate analyses of variance further showed that in allopatric populations, C. erythraea flowers showed a higher degree of herkogamy than flowers of C. littorale, and this distinction significantly increased in sympatric populations with a significant increase of herkogamy in C. erythraea (F 1,369 = 79.20; P < 0.001) and a decrease in C. littorale (F 1,268 = 21.69; P < 0.001) (Fig. 3, Table S2). Results for populations in the UK showed that petal length (F 1,240 = 40.17; P < 0.001) and width (F 1,240 = 49.80; P < 0.001) of the flowers were significantly larger in sympatric than in allopatric populations for C. erythraea populations, while C. littorale flowers showed a significant decrease from allopatric to sympatric flowers for petal width (F 1,243 = 44.25; P < 0.001) (Fig. 3, Table S2). Despite this, sympatric populations of C. erythraea were significantly smaller in total length (F 1,240 = 37.13; P < 0.001) than allopatric populations and total length (F 1,243 = 7.59; P = 0.006) of C. littorale increased from sympatric to allopatric populations. Pistil length did only differ between sympatric and allopatric populations of C. erythraea, with sympatric populations being significantly larger (F 1,240 = 12.03; P = 0.001), whereas stamen length increased in sympatric populations of C. erythraea (F 1,240 = 18.19; P < 0.001) and C. littorale (F 1,243 = 5.26; P = 0.023) (Fig. 3, Table S2). The level of herkogamy, on the other hand, decreased significantly in sympatric populations of C. erythraea (F 1,240 = 18.86; P < 0.001), and increased slightly in sympatric populations of C. littorale (F 1,243 = 5.47; P = 0.020).
In both regions, the difference in herkogamy between the two species was substantially larger in sympatric populations than in allopatric populations (Table 3, Fig. 4). On the European mainland, the difference in herkogamy between the two species increased from 0.38 mm ± 0.08 mm in allopatric populations to 0.82 mm ± 0.06 mm in sympatric populations. The same effect was seen in the UK, where the difference in herkogamy between the species increased from 0.19 mm ± 0.04 mm in allopatry to 0.37 mm ± 0.01 mm in sympatry. This indicates that the co-occurrence of the two species magnifies the difference in herkogamy (Fig. 4). On the European mainland, this difference is primarily driven by differences in stamen length, which increased from 0.04 mm ± 0.12 mm in allopatry to 0.33 mm ± 0.08 mm in sympatry. In the UK, on the other hand, differences in stamen length were similar between allopatric and sympatric populations, but variation in pistil length was mainly responsible for the variation in herkogamy and increased from 0.03 mm ± 0.06 mm in allopatry to 0.14 mm ± 0.03 mm in sympatry (Fig. 4).
Plant height, number of flowers and flowering stalks did also differ between allopatric and sympatric conditions (Table S3). On the European mainland, the difference in plant height increased from 9.31 cm ± 2.15 cm in allopatric populations to 13.65 cm ± 3.55 cm in sympatric populations, indicating that the two species tend to be more different in height when they co-occur. The average difference in the number of flowers and flowering stalks decreased from allopatric to sympatric populations, indicating that the two species tended to be more similar for these characters under sympatric conditions. In the UK, differences in these three characters decreased from allopatric populations to sympatric populations, indicating that the plants were more similar in sympatry.
The contribution of autonomous selfing to total seed production. Seed set differed significantly (F 1,902 = 760.99; P < 0.001) between emasculated and open pollinated flowers (Fig. 5). Intact open pollinated flowers generally produced more seeds per fruit (on average 248.3 ± 4.1 seeds) compared to emasculated flowers that only produced on average 106.4 (± 4.7) seeds per fruit. This resulted in overall reproductive assurance value of RA = 0.65 ± 0.02, and this capacity to self autonomously was significantly related to the degree of herkogamy (r = − 0.708; P = 0.049, N = 8), but not to the other floral traits. Furthermore, the extent to which autonomous selfing contributed to total seed production, depended significantly on species (F 1,212 = 7.20; P = 0.008) and was in general higher for C. littorale (RA = 0.68 ± 0.03) than for C. erythraea (RA = 0.60 ± 0.03). The species specific contribution of autonomous selfing to total seed production also interacted significantly with geographic region (F 1,212 = 14.77; P = 0.0002), with C. erythraea showing an increase in the contribution of autonomous selfing in the UK (RA = 0.67 ± 0.03) compared to the European mainland (RA = 0.48 ± 0.06), whereas the opposite pattern that was found in C. littorale. The contribution of autonomous selfing in C. littorale increased from 0.64 (± 0.04) in the UK to 0.76 (± 0.03) on the European mainland. We also found a significant interaction between species and whether the population is allopatric or sympatric (F 1,212 = 8.51; P = 0.004). The contribution of autonomous selfing to total seed production was lower in sympatric (RA = 0.56 ± 0.05) than in allopatric (RA = 0.66 ± 0.04) populations of C. erythraea, while C. littorale showed the opposite pattern and had a higher contribution of autonomous selfing in sympatric (RA = 0.74 ± 0.03) than in allopatric populations (RA = 0.62 ± 0.04).

Discussion
Geographic variation in floral morphology and the capacity of autonomous selfing. We found substantial geographic variation in floral traits in the two closely related Centaurium species studied here. Flowers  Table 3. Magnitude of differences and standard errors in morphological characters between allopatric and sympatric populations of C. erythraea and C. littorale sampled on the European mainland and the UK. of both species were on average larger on the European mainland than in the UK. The level of herkogamy was largest on the European mainland for C. erythraea whereas flowers of C. littorale showed higher levels of herkogamy in the UK. These results confirm earlier findings of Ubsdell 36 and Brys and Jacquemyn 35 , who already showed that the flower morphology of these two Centaurium species differed between these two geographic regions. The observed shifts in the level of herkogamy between allopatric and sympatric populations also translated into different capacities to self spontaneously. Interestingly, autogamy in sympatry was achieved by different means in mainland vs island populations. In the UK, reduced herkogamy was realized by an increase in pistil length, whereas on the mainland higher autonomous selfing was obtained through a reduction in stamen length. These results therefore suggest that character displacement occurred independently on the mainland and island. Although the spatial separation of anthers and stigmas was quite small in both study species (ranging between 0.06 and 5.59 mm), it had a significant impact on a plants' capacity to reproduce autonomously (Fig. 6), which varied between 0 to 99.7% seed set realized via autonomous selfing. Herkogamy and flower size are plastic traits that can vary among sites, and some studies have shown that this can be correlated with drought stress [45][46][47] . In dryer regions, plants tend more to self-fertilization with faster flowering time, shorter period of anthesis, smaller flowers and lower herkogamy. Based on monthly weather data obtained from weather stations nearby the study populations, we can conclude that, apart from some minor differences in temperature, the two regions are characterized by similar climatic conditions, most likely as a result of their close proximity to the sea. Although we cannot completely exclude such an effect as explanatory factor, the studied populations were all taken from similar coastal habitats and even if a drought-related effect would take place, differences among allopatric and sympatric populations are still visible.
Differences in floral and mating system traits between the two regions may, on the other hand, also be driven by differences in pollinator communities. Work in a variety of systems has indeed shown that flowers tend to become smaller and evolve towards selfing at the edge of their distribution ranges due to pollen limitation, unstable populations and/or changing environments 2,29,48,49 . This trend was visible for C. erythraea, but less for C. littorale. It is therefore tempting to suggest that the observed differences in floral traits and capacity for autonomous selfing between the two species could be the result of differences in pollinator communities and abundances between the two regions. Given that the distribution of C. littorale is largely restricted to the coastal areas of Western Europe (Fig. 1), this species may have adapted its floral morphology to the specific pollinator communities that characterize coastal areas, and which are likely to be quite similar throughout most of its distribution area. C. erythraea, on the other hand, shows a much larger distribution area, which encompasses a large part of the Mediterranean and Continental Europe. It can therefore be assumed that this species is exposed to more diverse pollinator communities across its distribution area. C. erythraea populations in the UK also represent the marginal habitat of this species, while C. littorale can be found on even more northern coastlines in Europe (Fig. 1).
On the European mainland, a much larger number of insect species generally visits C. erythraea than C. littorale, and the number of visits per insect species is significantly higher in C. erythraea than in C. littorale 41 . Assuming that the abundance of some of these insects declines with increasing latitude, differences in pollinator communities between continental and British populations or differences in the abundance of specific pollinators may have triggered the evolution towards selfing in northern populations of C. erythraea. Given the large geographic barrier between the European mainland and the British Isles, differences in selection pressures on floral traits between both regions are most likely not counterbalanced by gene flow, leading to significant genetic and morphological divergence between the two regions. However, to unambiguously show that differences in pollinator communities underlie the observed differences in floral morphology and mating systems between regions, more detailed research on the pollinator communities is needed. This includes comparative pollinator surveys over the geographic gradient to test for differences and field experiments where plants are introduced from a different region to look into the effect of the local and foreign pollinator community.
Floral divergence between allopatric and sympatric populations. In addition to pollinator-mediated floral divergence, differences in floral traits may result from inter-specific competition for pollinator services (i.e., reproductive character displacement) and/or selection against heterospecific mating and low fitness of the resulting hybrids (i.e., reinforcement) 19,23,29,33 . Under these conditions, it can be predicted that sympatric populations should display greater divergence in reproductive traits than their congeners in allopatric populations 21 . In some cases, the evolution to selfing will be favoured when a species co-occurs with a more outcrossing sister species, as the reduced herkogamy and an associated increase in the capacity to self autonomously will safeguard the species from heterospecific pollen transfer and hybrid formation. Previous work in a sympatric population of both Centaurium species in Belgium has shown that when the more selfing C. littorale is flowering in close proximity with the more outcrossing C. erythraea, the former is exposed to a significantly higher risk of heterospecific mating 33 . Moreover, the same research has shown that hybridization in Centaurium was costly 33 . The observed differences in floral traits between allopatric and sympatric populations may therefore be the result of selection towards higher autonomous selfing in order to reduce interspecific gene flow.
A geographic mosaic of floral traits and mating systems. In case of the studied Centaurium species, we hypothesize that when C. erythraea colonized the British Isles from the Continent after the last Ice Age, a shift to a more selfing mating system may have occurred in order to ensure sufficient reproductive assurance. When it was exposed to C. littorale, mating systems evolved further to selfing to reduce interspecific mating. On the European mainland, on the other hand, C. erythraea is characterized by a predominantly outcrossing mating system. When coming into contact with more selfing C. littorale, divergence in mating systems became reinforced so that C. littorale became more selfing in sympatric populations. Given the much more restricted distribution of C. littorale and its close affinity to coastal habitats (Fig. 1), it is reasonable to assume that C. littorale did not shift to higher selfing when it re-colonised the UK.

Conclusion
In this study, we confirmed earlier findings that the two Centaurium species displayed substantial variation in floral morphology across different geographic regions. This variation was also reflected in the contribution of autonomous selfing to total seed production, which largely paralleled the geographical variation in floral traits. Overall, these results reinforce the idea that floral traits related to mating systems can vary substantially depending on the ecological context in which the population occurs. Moreover, our study demonstrates that the joint occurrence of closely related plant species can have an important impact on floral morphology and the capacity of autonomous selfing, and suggests that reinforcement may be an important factor driving the evolution and maintenance of mating systems in sympatric populations. However, to fully understand the drivers of mating system variation in closely related plant species and to demonstrate that the evolution towards increased selfing reduces hybridization, reciprocal transplant experiments between allopatric and sympatric populations and across the studied geographic regions should be performed in the future and the fitness of their progeny assessed.