Habitat partitioning in native Geranium species through reproductive interference

Abstract

Background and Aims

Heterospecific pollen transfer may reduce the fitness of recipient species, a phenomenon known as reproductive interference. A theoretical study has predicted that distributions of species pairs affected by reproductive interference may be syntopic under negligible reproductive interference, sympatric but with partitioning at small spatial scale (i.e. allotopic) under weak interference, or exclusive when reproductive interference is strong. Verifying these predictions is essential for evaluation of the applicability of reproductive interference as a general assembly rule of biological communities. The aim of this study was to test these predictions in two sympatrically distributed wild Geranium species, G. thunbergii and G. wilfordii.

Methods

To measure the effect of reproductive interference, the associations between the relative abundance of the counterpart species and seed set in the focal species, and seed set reduction following mixed pollination, were analysed. The possibility of hybridization with viable offspring was examined by genotyping plants in the field and after mixed pollination. Fertility of putative hybrids was based on their seed set and the proportion of pollen grains with apertural protrusions. A transect study was conducted to examine spatial partitioning, and possible influences of environmental conditions (canopy openness and soil moisture content) on partitioning between the species were analysed.

Key Results

Neither abundance of the counterpart species nor heterospecific pollen deposition significantly affected seed set in the focal species, and hybridization between species was almost symmetrical. Putative hybrids had low fertility. The two species were exclusively distributed at small scale, although environmental conditions were not significantly different between them.

Conclusions

The allotopy of the two species may be maintained by relatively weak reproductive interference through bidirectional hybridization. Re-evaluation of hybridization may allow ongoing or past reproductive interference to be recognized and provide insight into the distributional relationships between the interacting plants.

INTRODUCTION

Understanding the processes governing species coexistence is a primary goal of ecology (Burns and Strauss, 2011). One widely known ecological phenomenon is that closely related species tend to be distributed exclusively, and recent literature has revealed that reproductive interference is a major agent causing exclusive distribution (see Kyogoku, 2015, for a review). Reproductive interference has been defined by Gröning and Hochkirch (2008) as any kind of interspecific interaction during the process of mate acquisition that adversely affects the fitness of at least one of the species involved and is caused by incomplete species recognition. A major mechanism of reproductive interference in plants might be heterospecific pollen deposition ( Nishida et al., 2014; Kyogoku, 2015; Takemori et al., 2019), which can reduce the number of offspring by causing fertilization failure, defective development (due to pollen allelopathy or ovule discounting), or the maternal reproductive investment to be wasted through the production of less-fertile hybrids.

In homogeneous environments, reproductive interference can theoretically rapidly exclude one or the other of a species pair because of its positive frequency dependence and its self-enhancing property through positive feedback (Rebeiro and Spielman, 1986; Kuno, 1992; Takakura et al., 2010). If one of the species pair strongly affects reproduction by the other species by reducing the number of fertile offspring produced by the second species, the first species becomes relatively more abundant in the next generation. As a result, the effect is exerted more frequently than in the last generation, and this positive feedback eventually leads to the exclusion of the second species. Because the distributions of the two species rapidly become exclusive, it can be difficult, in particular when both species are native, to identify reproductive interference as the key driving force behind spatial partitioning observable in the wild in the present (e.g. Briscoe Runquist, 2012). For this reason, the earlier literature on reproductive interference has focused on interactions between native and alien species pairs, in which the effects of reproductive interference can be measured and evaluated as real-time processes; e.g. Lythrum spp. (Brown and Mitchell, 2001), mulberry (Morus spp.) (Burgess et al., 2008), dandelion (Taraxacum spp.) (Takakura et al., 2009, 2011, 2012; Matsumoto et al., 2010; Nishida et al., 2012) and bird’s-eye speedwell (Veronica spp.) (Takakura, 2013) pairs.

With these considerations in mind, we argue that reproductive interference has sometimes been the driving force behind the present observed distributions of not only native/alien species pairs but also native/native species pairs. Here, we focus on interactions between two species of Geranium, G. thunbergii and G. wilfordii, native to Japan. At large scale, the distributions of the two species appear to overlap (Kadota, 2016), but our preliminary field survey indicated that they likely occupy separate habitats. In addition, we have identified a putative hybrid with intermediate morphology between the two species that is characterized by low fertility; the existence of such a hybrid implies that hybridization with less-fertile offspring is a possible mechanism of reproductive interference in this species pair. In a theoretical study, Nishida et al. (2015) predicted that when reproductive interference occurs in closely related species, moderate or high reproductive interference should lead to allopatry, weak reproductive interference should lead to habitat partitioning, and negligible reproductive interference should allow the two species to coexist, subject to the combined effects of differences in environmental suitability and resource competition. Here, strength of reproductive interference denotes the relative degree of the reduction in fertile offspring caused by the interference. According to these predictions, closely related native species pairs that share the same geographical distributions might be sympatric and occupy the same habitat (Rivas, 1964) if the level of reproductive interference is negligible, or, if reproductive interference is weak, they might be sympatric but with small-scale spatial partitioning (i.e. allotopic; Rivas,1964). Verification of these predictions in these two Geranium species would be a first step in evaluating the applicability and universality of reproductive interference as a general assembly rule of wild plant communities.

In this study, we conducted a field survey to investigate the effects of reproductive interference between G. thunbergii and G. wilfordii on seed set in the two species, together with hand pollination experiments to examine the effect of heterospecific pollen deposition on seed set. We also examined whether genotypic variations in the putative hybrid and in its postulated parent species support its identification as a hybrid. Then, referring to the genotype results, we estimated the frequency of hybridization in offspring resulting from artificial pollination with mixed (conspecific and heterospecific) pollen of G. thunbergii and G. wilfordii. In addition, we compared seed set and the proportion of pollen grains with apertural protrusions between the pure species and the putative hybrids to evaluate the relative fertility of the putative hybrids.

Finally, at one study site, we measured the exact locations of individual plants along a transect to examine the degree of spatial partitioning between the two Geranium species. We also examined canopy openness and soil moisture content at each plant location to evaluate whether these environmental factors might have influenced any observed partitioning.

On the basis of the results obtained by these investigations, we try to answer the following questions in this paper: Can reproductive interference be recognized between the two species? If present, does it occur through hybridization only by reducing seed set (i.e. non-viable offspring) or through the production of viable but less fertile offspring? How strong is the possible reproductive interference and how might it and the presence of putative hybrids affect distributional relationships between the two species? Lastly, is it possible to recognize ongoing or past reproductive interference between wild plant species?

MATERIALS AND METHODS

Study system

Geranium thunbergii and G. wilfordii are widely distributed in eastern Asia, including in the Russian Far East, China, Korea and Japan (Kadota, 2016). These two species and G. tripatrium form a monophyletic clade according to a molecular phylogenetic study of native Geranium species in Japan (Wakasugi et al., 2017). Both species are reported to have 2n = 28 chromosomes (Kadota, 2016). Geranium thunbergii is a common species in a wide variety of environments from cultivated fields to mountain slopes, whereas G. wilfordii is relatively uncommon and restricted to mountain slopes. The two species are morphologically similar with respect to the leaves on the upper part of the stem (mainly three-lobed) and flower size (1–1.5 cm across), but they can be discriminated by close inspection of the leaves on the lower stem, sepals and pedicels: G. thunbergii has five-lobed (rarely three-lobed) leaves on the lower stem and erect glandular hairs densely distributed on the leaves, sepals and pedicels, whereas G. wilfordii has three-lobed leaves on the lower stems and appressed hairs on the leaves, sepals and pedicels (Kadota, 2016). In G. thunbergii, flowers may be dark pink or white, although at our study sites all of the flowers were white or whitish pink. Flowers of G. wilfordii are always pale pink. In the study area, the flowering periods of the two species overlap from late August to early November (S. Nishida, pers. obs. in 2008–10). They share the same kinds of pollinators, including bees and syrphids, and we have confirmed that these pollinators do not consistently visit flowers of the same species; rather, they often (39.9 % of 158 visits) move between different species, including between the two study species and Persicaria sp. (S. and T. Nishida, pers. obs. in 2017). According to Kandori (2002), G. thunbergii is self-compatible but self-pollinated flowers produce significantly fewer seeds than outcrossed flowers. There is no information on self-compatibility in G. wilfordii. Both species had protandrous flowers at the study sites, but their stigmas opened while their pollen was still viable, which suggests that self-crossing can occur (S. Nishida, pers. obs. in 2009). The seeds of both species are dispersed explosively by the bursting of the seed pod, and their dispersion abilities are likely similar. Clonal reproduction has never been reported in these two species to our knowledge. The putative hybrids are morphologically intermediate between the two pure species: they have three-lobed leaves on the lower stems, whitish pink to pink flower petals, and sparsely pubescent sepals and pedicels with erect but non-glandular hairs.

Study sites

All of our surveys and experiments were carried out at three study sites on the lower slopes of Mt Yokoyama, Shiga Prefecture, central western Honshu Island, Japan: Suganami (SGA; 35°35′58″ N, 136°13′48″ E), Sugino (SGI; 35°34′20″ N, 136°15′46″ E), and Tsuchikura (TSU; 35°35′37″ N, 136°18′2″ E) (Supplementary Data Fig. S1). SGA and SGI are about 4.5 km apart, as are SGI and TSU, but SGA and TSU are separated by about 6.8 km. All three study sites are in a partly open forest edge along a stream and a road, and they are characterized by a similar combination of environments, except that the roads at SGA and SGI are about 3 m wide and paved, whereas the road at TSU is about 1.5 m wide and unpaved. The deciduous forest surrounding each site is dominated by Quercus serrata with some patches of planted Cryptomeria trees. We conducted the field survey at SGA and SGI. At TSU, we conducted the hand-pollination and mapping experiments, and also measured canopy openness and soil moisture. Thus, we did not perform any hand pollination, mapping and measurement activities that might disturb natural conditions at the field survey sites. In addition, we used TSU for the mapping study because it was the largest of the three study sites.

Field survey: effects of counterpart frequency on seed set in the focal species

If reproductive interference results from ovule usurpation by heterospecific pollen, reproductive performance, e.g. seed set, of the focal species should decrease as the counterpart species immediately surrounding the focal species becomes more abundant. To quantify the effect of the relative abundance of the counterpart species on the reproductive success of each Geranium species, we conducted a field survey at SGI in October 2008 and at SGA in October 2009. We arbitrarily chose five localities at SGI and four at SGA, and counted individuals of each Geranium species within a circle with 5-m radius at each locality. No recent studies have measured the effective range of reproductive interference in relatives of Geranium, but a study of Taraxacum estimated that 50 % or more of flowers of one species within 1.57 or 5.31 m of another species individual (depending on the habitat) could receive interspecific pollen (Takakura et al., 2011). Considering that Geranium plants are roughly two to three times larger than Taraxacum plants, and because there were few circles with a 2-m radius that contained both Geranium species, we adopted a 5-m radius to measure the effects of reproductive interference caused by interspecific pollen transfer. We later confirmed that most pollinators visiting a Geranium plant made their next visit within 5 m from that plant (96.2 % of 168 visits); moreover, pollinator species, which include bees and syrphids, do not differ greatly between Geranium and Taraxacum (S. and T. Nishida and A. Naiki, pers. obs. for Geranium in 2017 and Taraxacum in 2018).

We used the ratio of individuals of the counterpart species to the total number of Geranium individuals within the 5-m radius as the relative abundance of the counterpart species, after first confirming that the numbers of individuals and flowers were moderately well correlated (R² = 0.847 for G. thunbergii; R² = 0.484 for G. wilfordii, based on S. and T. Nishida, pers. obs. in 2008). We did not include putative hybrids in the calculation of this ratio, because our preliminary observations suggested that putative hybrids produce considerably less pollen than pure species; thus, including them as counterpart individuals might produce misleading results for the effect of the counterpart species on the reproductive success of the focal species. We then arbitrarily selected up to seven individual plants from near the centre of each locality (to avoid evaluating the effect of pollination occurring outside of the 5-m radius), and collected up to five fruits from each plant (we used a smaller number of plants or fruits if fewer than seven plants or five fruits were found). We brought the fruits back to our laboratory and counted the number of normally developed and undeveloped seeds in each fruit. Each fruit in G. thunbergii and G. wilfordii has five ovules, and developed seeds are easily discriminated from undeveloped seeds by their size and hardness.

We used a generalized linear mixed model (GLMM; Wolfinger and O’Connell, 1993) with a binomial error structure and a logit link function to test the effect statistically. In this analysis, the response variable was normal seed development, and the explanatory variables were relative abundance of the counterpart species and the total number of individuals of both species. The total number of individuals was included because if interactions for abiotic resources or pollinators between Geranium individuals were competitive and stronger than competitive interactions with other species, a negative effect of this variable would be detected. Besides this analysis, additional analyses were conducted in which density of the counterpart species or that of the focal species were included as the explanatory variables to determine whether density, rather than frequency, affected seed set. We analysed the data from each study site (SGA, SGI) both independently and inclusively (i.e. for both of the sites together), because the data for G. thunbergii at SGA were insufficient for an independent analysis at that site. Individual plants were nested within locality and then incorporated as a random effect in the independent analyses, whereas they were first nested within locality and then nested within study site before being incorporated as a random effect in the inclusive analyses. All analyses were conducted with R version 3.5.2 software (R Core Team 2018). As a more intuitive measure, we also calculated the proportion of all ovules that were normally developed as the seed set at each locality.

Examination of the effect of heterospecific pollen deposition on seed set

To measure the degree to which seed set was reduced because pollen of the counterpart species was deposited on the flowers, we conducted a reciprocal hand pollination experiment at TSU. In September 2009, we arbitrarily selected 30 individuals of each of the two species and assigned one flower of each individual to one or the other of the following two treatments: a mixed pollination treatment or a control treatment (conspecific pollination only). In the mixed pollination treatment, conspecific and heterospecific pollen grains were applied at the same time by seizing two stamens, one from each of the two species, with forceps and gently applying the pollen grains on the stamens to the centre of a recipient’s stigma. In the control treatment, pollination was carried out in the same way but with a single conspecific stamen. Although we did not count the exact number of pollen grains applied in each treatment, we tried to apply an equal number (about 20) of pollen grains from each stamen to each recipient stigma. Before and after the pollen application, we kept each of the recipient flowers covered with a small polyester bag to prevent unintended pollination. A few days after the pollination, we removed the bag because the flower pedicels are not strong enough to withstand rain with the cover on. When we removed the covers, we confirmed that the stigmas were closed and that the fertilization stage of the flowers was over. About 2 weeks after the pollination, we collected the resultant fruits, brought them back to the laboratory, and counted the developed seeds in each fruit.

We used a GLMM with a binomial error structure and a logit link function to analyse the results; treatment was used as the explanatory variable, and normal seed development was used as the response variable. Individual recipient plants were incorporated as a random effect. The analyses were conducted with R version 3.5.2 software (R Core Team, 2018). We also pooled all ovules in each treatment and calculated the proportion that developed normally as a more intuitive measure of the seed set in each treatment.

Genotyping putative hybrids and offspring of the mixed pollination treatments

To confirm the status of the putative hybrids between G. thunbergii and G. wilfordii, we observed allozyme variations in samples from several plants. We employed allozyme observations to characterize the plants genetically because no other molecular markers for discriminating the two species and their hybrids were known until recently (Wakasugi et al., 2017). At SGI on 21 October 2008, we arbitrarily selected eight G. thunbergii individuals, nine G. wilfordii individuals and eight putative hybrids, and sampled one leaf from each individual. The sampled putative hybrids had three-lobed leaves on the lower stem (as in G. wilfordii), erect hairs on the sepals and pedicels (as in G. thunbergii, but less densely distributed), and pinkish petals (like G. wilfordii). To minimize the sampling of genetically contaminated plants, we selected G. thunbergii and G. wilfordii individuals for sampling from localities where only a single species was present. We put the leaf samples into an insulated, cooled container immediately after collection for transport to the laboratory, where we placed them in a deep freezer (about −80 °C) until analysis. Three days after the sampling, we conducted an electrophoresis experiment, basically following Shiraishi (1988), using vertical polyacrylamide slab gels and a discontinuous buffer, and observed the results. For observation, we used the staining procedures of Richardson et al. (1986) and Shiraishi (1988) with minor modifications. We examined 11 enzymes: 6-phosphogluconate dehydrogenase (6PGD, E.C. 1.1.1.44), alanine aminopeptidase (AAP, E.C. 3.4.11.1), aspartate aminotransferase (AAT, E.C. 2.6.1.1), alcohol dehydrogenase (ADH, E.C. 1.1.1.1), diaphorase (DIA, E.C. 1.6.99.–), glutamate dehydrogenase (GDH, E.C. 1.4.1.2), glucose-6-phosphate isomerase (GPI, E.C. 5.3.1.9), leucine aminopeptidase (LAP, E.C. 3.4.11.1), malate dehydrogenase (MDH, E.C. 1.1.1.37), phosphoglucomutase (PGM, 5.4.2.2) and shikimate dehydrogenase (SKD, E.C. 1.1.1.25).

To examine the possibility of hybridization more directly and to measure the frequency of hybridization, we also observed allozyme variations in offspring resulting from the mixed pollination. In March 2010, we sowed the seeds obtained from the hand pollination experiment conducted in September 2009. After the seeds germinated and the seedling emerged, we sampled one leaf from each seedling and placed them in a deep freezer (about −80 °C) until analysis. From the samples suitable for analysis, we arbitrarily selected 37 seedlings from 20 parent individuals for G. thunbergii and 20 seedlings from 12 G. wilfordii parents. (Although we had planned to analyse an equal number of samples from each parent species, the condition of some of the G. wilfordii samples deteriorated while in the freezer.) As a control, we collected one leaf of each species from offspring resulting from the conspecific pollination experiment (one sample from each species). Using the same procedures as described above, we observed the allozyme variations among the samples at the loci of two enzymes, AAT and DIA. First we confirmed that the offspring from the conspecific pollination had the same banding patterns as those of the corresponding pure species collected from SGI, and then we compared the genotypes of the offspring from the mixed pollination and conspecific pollination experiments. We considered offspring with a banding pattern that combined the banding patterns of both pure species at AAT and DIA loci to be hybrids. Three of the 57 samples had an obscure banding pattern at AAT or DIA loci and were excluded from the results. We then used Fisher’s exact test in the R version 3.5.2 software package (R Core Team, 2018) to analyse statistically the frequency of heterozygous banding patterns between samples from seedlings with G. thunbergii mothers and samples from ones with G. wilfordii mothers.

Investigation of the fertility of the putative hybrids

To infer a possible mechanism of reproductive interference between G. thunbergii and G. wilfordii, we investigated the fertility of the putative hybrids by examining their seed set and the proportion of pollen grains with apertural protrusions.

To examine seed set, we arbitrarily collected 55 individual Geranium plants at TSU that had flowers and fruits in October 2018. We brought them back to the laboratory, checked the leaf shape on the lower stems and the pubescence of the sepals to identify them as from one or the other species or a putative hybrid, and examined seed development in one of their fruits. The identification revealed that the samples included 13, 18 and 24 individuals of G. thunbergii, G. wilfordii and the putative hybrids, respectively.

In a preliminary study, we were unable to obtain long pollen tubes even on culture media with a range of sucrose concentrations (1–15 %), but we could still observe apertural protrusions as reported for viable pollen grains in Geranium by Weber (1996). Therefore, we examined whether the individual pollen grains had apertural protrusions or not to evaluate the fitness of the plants with respect to their male function. For this examination, we prepared glass slides by covering them with an ~0.5-mm-thick layer of culture medium containing 15 % sucrose and 1.5 % agarose. We used the same 55 individuals used to check the seed set, picked one anther from each flower with forceps, gently put it on one of the glass slides, and then placed the slide in an aluminium foil-covered plastic zipper bag. The slides were placed at ~23 °C for 24 h and then the pollen grains were checked for apertural protrusions. We examined 15 pollen grains (or a smaller number of them if fewer pollen grains were found) for each individual, which resulted in 808 pollen grains observed.

We conducted likelihood ratio tests to determine whether differences in the plant type (pure species or putative hybrid) was associated with differences in seed set and the proportion of pollen grains with apertural protrusions. A binomial error structure and a logit link function were applied with the plant type as a fixed effect and individual as a random effect. In the seed set analysis, seed development was the response variable; in the pollen grain analysis, apertural protrusion of pollen grains was the response variable. We compared the null model (with only the random effect) with the model including the fixed effect (plant type), and defined the association to be significant if the P value was <0.05. The analyses were conducted with R version 3.5.2 software (R Core Team, 2018). As more intuitive measures, we also calculated the proportions of normally developed ovules and pollen grains with apertural protrusions in the pure G. thunbergii plants, the pure G. wilfordii plants and the putative hybrids.

Mapping the two species and analysing their distributional relationships

To investigate small-scale spatial relationships between G. thunbergii and G. wilfordii plants, we mapped their distributions along a transect at TSU on 27 September 2009. We established a 600-m-long transect and recorded the distance from the starting point of the transect to each point where a line drawn from an individual plant crossed the transect at a right angle for all plants. We then used a DLE 150 laser distance estimator (Bosch, Germany) to record the distance from the point on the transect to the plant. In total, we located 499 individuals along the transect.

After mapping the data onto a coordinate grid, we evaluated the correlation between the distributions of the species by calculating Morisita’s index of correlation of interspecific distributions (Morisita, 1959), R_δ, as follows:

$${\displaystyle \begin{array}{l}{\displaystyle {\delta }_x={\displaystyle \underset{i=1}{{\displaystyle \sum ^q}}}{n}_x(n_x-1)/N_x(N_x-1)}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\delta }_y={\displaystyle \underset{i=1}{{\displaystyle \sum ^q}}}{n}_y(n_y-1)/N_y(N_y-1)}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle C_{\delta }=2{\displaystyle \underset{i=1}{{\displaystyle \sum ^q}}}{n}_x{n}_y/({\delta }_x+{\delta }_y)N_x{N}_y}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle W_{\delta }=2/({\delta }_x+{\delta }_y)q}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle R_{\delta }=C_{\delta }-W_{\delta }}\end{array}}$$

where n_x and n_y are the number of individuals of species x and y, respectively, in the ith quadrat (i = 1, 2, ..., q) and N_x and N_y are the total number of individuals of the species x and y, respectively. This index is a measure of the interspecific correlation and similarity between communities, and it is less influenced by average density than other measures, such as Pearson’s product-moment correlation coefficient, which tend to be biased towards smaller values when densities of the two species are low (Morisita, 1959). We first divided the plot (the whole mapped area) into 3 m × 3 m quadrats and counted the numbers of G. thunbergii and G. wilfordii individuals within each quadrat. Then, using all quadrats containing at least one individual of either species, we calculated R_δ from the observation data as the correlation index of the observed distributions. We next conducted a permutation test to evaluate quantitatively the degree to which the distributions of the two species were exclusive. In this test, we calculated the R_δ index 10 000 times using randomized populations to obtain the probability density distribution of the index for the null hypothesis that the distributions of the two species were independent. The randomized populations were made by randomly rearranging the number of individuals of either species (e.g. G. thunbergii) among the quadrats. Note that the intraspecific distribution properties, such as the mean and standard deviation of the number of individuals per quadrat, remained the same as the observation even after the randomization. The value of Morisita’s R_δ can range from −1 to +1, where −1 indicates that the distributions of two species are completely exclusive, +1 indicates that their distributions are completely syntopic, and 0 indicates that the two species are randomly distributed. We compared the index value for the observed distributions with the probability density distribution of the index to assess the statistical significance of the observed R_δ value. We followed the same procedures to calculate R_δ for quadrats of 1 × 1, 2 × 2, 4 × 4, and 5 × 5 m² to examine the dependency of its value on the spatial scale.

Examination of environmental conditions associated with the two species

To investigate whether there were any environmental conditions that might cause spatial partitioning, we measured canopy openness above and soil moisture content below each individual plant.

We defined the degree of canopy openness as the percentage of unobstructed sky above the plant and used it as an estimate of light availability. At TSU on 2 October 2010, we arbitrarily selected ~20 individuals each of G. thunbergii and G. wilfordii. We set up a Ricoh GX200 camera (Ricoh, Japan) with a fish-eye lens on a tripod next to each plant. The camera was set 1 m above the ground with its top pointing north and its lens facing upwards. Two pictures with a 72-dpi resolution were taken next to each individual plant. We then used hemispherical photograph analysis software to calculate the canopy openness (the area of open sky relative to the total image area) in the two images obtained next to each plant (Canop-On2; http://takenaka-akio.org/etc/canopon2/index.html) and used the mean value as the canopy openness value for that plant.

On 1 September (dry weather conditions) and 1 November (humid weather conditions) 2012, we measured the soil moisture content at TSU, where we had measured canopy openness in 2010. We arbitrarily selected 20 individuals of each species (not necessarily the same individuals that were used to measure canopy openness) and measured the moisture content of the soil below each plant at 10 cm depth using a Hydrosense soil water measurement system (Campbell Scientific, Canberra, Australia).

To evaluate the influence of these two environmental parameters on spatial partitioning of the species, we used a Wilcoxon rank sum test with continuity correction to analyse the association between the differential occurrence of each species and canopy openness or soil moisture content. The analyses were conducted with R version 3.5.2 software (R Core Team, 2018).

RESULTS

Field survey: effects of counterpart species frequency on seed set in the focal species

Seed set of the two species did not decrease significantly as the relative abundance of the counterpart species increased at either SGI or SGA (Fig. 1, Table 1). This result indicates that the presence of the counterpart species did not affect seed set in either species. Both species consistently achieved relatively high seed sets, which suggests that pollen limitation was not a critical determinant of seed set. The association between the total number of individuals of both species and seed set in each was significantly positive for G. thunbergii at SGA but otherwise non-significant (Table 1). This result suggests that density-dependent factors such as resource competition and competition for pollinator services among the Geranium individuals did not adversely affect seed set in either species. This association with seed set was non-significant even when we included density of the counterpart species or that of the focal species as an explanatory variable (Supplementary Data Table S1).

Fig. 1.

Relationships between the relative abundance of the counterpart Geranium species (compared with the total number of Geranium spp. individuals) and seed set of G. thunbergii (A) and G. wilfordii (B) at each locality at SGA (filled symbols) and SGI (open symbols). Vertical bars show 95% confidence intervals.

Table 1.

GLMM analysis results for seed set in G. thunbergii and G. wilfordii plants in relation to the relative abundance of the counterpart species (ratio of individuals of the counterpart species to the total number of Geranium individuals within a 5-m radius) and to the total number of Geranium individuals within the 5-m radius

Site and year	Species	Number of flowers (individuals)	Relative abundance of the counterpart species			Total number of Geranium individuals
			Coefficient ± s.e.	Z	P	Coefficient ± s.e.	Z	P
SGI, 2008	G. thunbergii	30 (5)
SGI, 2008	G. wilfordii	109 (19)	1.464 ± 2.691	0.544	0.586	−0.065 ± 0.040	−1.627	0.104
SGA, 2009	G. thunbergii	69 (23)	2.853 ± 2.826	1.010	0.313	0.011 ± 0.006	2.014	0.044
SGA, 2009	G. wilfordii	63 (21)	18.875 ± 9.060	2.083	0.037	0.013 ± 0.015	0.915	0.360
SGA and SGI	G. thunbergii	99 (28)	2.747 ± 1.810	1.518	0.129	0.011 ± 0.006	1.837	0.066
SGA and SGI	G. wilfordii	172 (40)	−0.851 ± 1.646	−0.517	0.605	0.013 ± 0.012	1.088	0.276

Lowest two rows show the results of analyses of combined data from SGA and SGI, in which individuals were treated as a random effect nested within locality and study site.

Effect of heterospecific pollen deposition on seed set

Although seed set was slightly reduced following mixed pollination compared with the control (Fig. 2), the reduction was not significant in either species (GLMM, coefficient of mixed pollination = −0.082, Z = −0.198, P = 0.843 for G. thunbergii; coefficient = −0.379, Z = −1.199, P = 0.231 for G. wilfordii). These results suggest that heterospecific pollen had only a marginal effect on seed set in both Geranium species.

Fig. 2.

Comparison of seed set in G. thunbergii (A) and G. wilfordii (B) between mixed pollination (pollination by both conspecific and heterospecific pollen) and conspecific pollination treatments. Thin vertical bars show 95% confidence intervals.

Allozyme variations in putative hybrids and mixed pollination offspring

Our examination of allozyme variations showed that plants with intermediate morphology between the two species were most likely true hybrids.

Of the 11 enzyme systems examined, nine could be scored for 13 putative loci: 6Pgdh, Aap, Aat-1, Aat-2, Aat-3, Adh-1, Adh-2, Dia-1, Dia-2, Gdh, Gpi, Lap and Pgm. Among these, ten loci were not informative: the banding patterns of five (Aap, Adh-1, Adh-2, Dia-1 and Gdh) were monomorphic in all samples, and those of the other five (6Pgdh, Aat-3, Gpi, Lap and Pgm) were polymorphic in at least one of the two species. The other three loci, Aat-1, Aat-2 and Dia-3, were informative and could be used to recognize the status of the plants with intermediate morphology. At Dia-3, the banding patterns were monomorphic within each species but different between the two species, and in the plants with intermediate morphology a heterozygous banding pattern was observed. At Aat-1 and Aat-2, similar differences in the banding patterns were observed in the two species and in the putative hybrids, although because the bands in the two parent species were close to each other, the bands of the putative hybrids appeared continuous.

In offspring resulting from mixed pollination, five of 35 (14.3 %) seedlings from G. thunbergii mothers and three of 19 (15.8%) seedlings from G. wilfordii mothers had banding patterns heterozygous for the two species at Aat-1, Aat-2 and Dia-3. Frequencies of hybrids following the mixed pollination were not significantly different between the two species (Fisher’s exact test, P = 1.00, odds ratio = 1.123). These results suggest that hybridization between the two species was basically symmetrical.

Fertility of the putative hybrids

Seed set and the proportion of pollen grains with apertural protrusions were significantly lower in the putative hybrids than they were in the pure species (Fig. 3). The effects of plant type (pure species or putative hybrid) on both seed set and the proportion of pollen grains with apertural protrusions were significant (likelihood test: χ² = 17.127, P < 0.001 for seed set; χ² = 19.697, P < 0.001 for apertural protrusions of pollen grains).

Fig. 3.

Comparison of (A) seed set (proportion of ovules that developed into normal seeds) and (B) proportion of pollen grains with apertural protrusions in G. thunbergii, G. wilfordii and the putative hybrids between G. thunbergii and G. wilfordii. Thin vertical bars show 95% confidence intervals.

Seed set of the putative hybrids was about 65 and 69 % of seed set in pure G. thunbergii and G. wilfordii, respectively. The actual seed set in the putative hybrids was likely lower than the value we obtained because we counted the number of seeds in fruits, but Geranium plants sometimes shed fruits in which no seeds have developed. In our preliminary survey (S. and T. Nishida, pers. obs., SGI in 2008), we roughly checked a total of about 1000 pedicels on 28 individuals (three pure G. thunbergii plants, ten pure G. wilfordii plants and 15 putative hybrids) and counted the number of pedicels that bore fruits. We found that, compared with the plants of the pure species, the putative hybrids retained fruits only about 49.8 % as frequently.

Pollen grains from putative hybrids had apertural protrusions only 55.1 and 54.8 % as frequently as pollen grains from pure G. thunbergii and G. wilfordii, respectively. Furthermore, we sometimes found that anthers of the putative hybrids had far fewer pollen grains than anthers of the pure species, although these observations are only anecdotal.

Distributional relationships between the two species

Along the transect at TSU, each Geranium species seemed to be distributed in patches consisting mostly of the same species (Fig. 4). Putative hybrids sometimes occurred around patch edges, between patches of different species, although they were also found in areas with only one of the two species.

Fig. 4.

Distributions of G. thunbergii (circles), G. wilfordii (diamonds) and putative hybrids (crosses) along the transect at TSU.

Morisita’s R_δ index for the observed distributions based on 3 m × 3 m quadrats was −0.764; this value indicates that the distributions of the two species were mainly exclusive. The middle 99 % of the probability density distributions obtained by the permutation for quadrats of this size under the null hypothesis (that the two species were randomly distributed) ranged from −0.278 to 0.1823. Because R_δ for the observed distribution was much smaller than this range, the observed distributions of the two species were significantly exclusive (P < 0.01). Morisita’s R_δ indices calculated using quadrats of different sizes were also significantly smaller than the middle 99 % of the probability density distributions obtained under the null hypothesis (data not shown). Therefore, the distributions of the two species were exclusive regardless of the scale used.

Examination of environmental conditions associated with the two species

The frequency distributions of G. thunbergii and G. wilfordii with respect to canopy openness (Fig. 5) mostly overlapped, and canopy openness where individuals of the two species grew varied greatly. The distributions of the two species with respect to canopy openness did not differ significantly (Wilcoxon rank sum tests with continuity correction, W = 242, P = 0.597).

Fig. 5.

Frequency histogram of canopy openness above G. thunbergii and G. wilfordii individuals.

The frequency distributions with respect to the soil moisture content under the plants, especially the results for November, suggest that G. wilfordii might grow on wetter soil than G. thunbergii (Fig. 6), but statistical testing did not support this interpretation; the distributions did not differ significantly between the two species (Wilcoxon rank sum tests with continuity correction: September distributions, W = 166, P = 0.363; November distributions, W = 169.5, P = 0.416).

Fig. 6.

Frequency distributions of G. thunbergii and G. wilfordii individuals in relation to soil moisture content in (A) September and (B) November 2012.

DISCUSSION

Whereas neither relative abundance of the counterpart species nor mixed pollination significantly reduced seed set (Figs 1 and 2, Table 1), G. thunbergii and G. wilfordii most likely produced hybrids with significantly lower fertility compared with that of the pure species (Fig. 3). These results suggest that the mechanism of reproductive interference between the two Geranium species is not a direct reduction of seed set by the counterpart species pollen but hybridization that produces viable but less-fertile offspring. The hand-pollination experiments and the genotype investigation of offspring resulting from hand pollination revealed that reproductive interference in this species pair is bidirectional and almost symmetrical. Our distributional relationship results suggest that the two species are mostly exclusively distributed at small spatial scale. These exclusive distributions, however, could not be explained by environmental conditions, represented by canopy openness or soil moisture content (Figs 5 and 6), nor by a reduction of seed set due to a density-dependent factor such as resource competition between the species (Table 1, Supplementary Data Table S1). A founder effect together with a short seed dispersal distance might result in some exclusively distributed populations, but it might not be possible for them to maintain such exclusiveness in the long run without some adverse interaction between species, because the distributions of most unrelated wild plants are mingled regardless of their dispersal distances. Given these observations, reproductive interference due to hybridization may be the major factor allowing these two species to maintain exclusive distributions.

Hybridization has previously been proposed to be one of the main mechanisms of reproductive interference in plants (Mitchell et al., 2009). Levin et al. (1996) have suggested that the growth rate of a population may be slowed by hybridization, regardless of whether the hybrid seeds develop, if hybrid seeds are produced at the expense of conspecific seeds. More recently, hybridization has been reported to affect distributional relationships between native and alien plant species, e.g. in Taraxacum, because of seed abortion (Nishida et al., 2014), and in Morus, as a result of the production of hybrid rather than conspecific offspring (Burgess et al., 2008). Future studies should examine more examples of hybridization to evaluate its contribution to the observed distributional relationships between closely related plant species.

Nishida et al. (2015) have predicted that under weak reproductive interference a species pair should show partitioning at small spatial scale (allotopy), whereas under negligible reproductive interference syntopic coexistence might be possible. Here we showed that at small spatial scale at our study site, G. thunbergii and G. wilfordii were distributed exclusively, a distributional relationship that can be considered allotopic. Our results are thus consistent with the predicted distributions under weak reproductive interference. It is difficult, however, to evaluate the strength of reproductive interference between these two Geranium species, especially because we have no data on a comparable group of plants. Nevertheless, some data are available on reproductive interference in three pairs of related plants from which we can infer the impact of reproductive interference on distributional relationships. (1) Polyploid and monoecious populations of Mercurialis annua have been displaced by diploid and dioecious populations through hybridization that results in non-viable or sterile offspring (the estimated proportion of hybrids with a polyploid mother is about 50–70 %, and about 0.02 % with a diploid mother, when equal numbers of individuals from each parent population are planted together; calculated using data from Buggs and Pannell, 2006). (2) In dandelions (Taraxacum spp.), where the native T. japonicum has been displaced by the alien T. officinale, average seed set of the native species was reduced by about 34.4 % following mixed pollination compared with seed set following conspecific pollination (Matsumoto et al., 2010). Hybrids are produced between these dandelions, but hybridization producing viable offspring seems to be rare (Takemori et al., 2019). (3) In Veronica, where the native variety, V. persica var. lilacina, has been displaced by V. persica var. persica (no hybrids between them have been found), the fruiting success of the native variety was reduced by 43.0 % in a transplantation experiment and by 32.3 % in a mixed pollination experiment (averaged values following mixed pollination were compared with those following conspecific-only pollination; Takakura, 2013). These three cases are examples of ‘strong’ reproductive interference, in terms of the theoretical predictions by Nishida et al. (2015), because each pair was allopatrically distributed. In our study, seed set reduction following mixed pollination was not significant in either G. thunbergii or G. wilfordii, but the estimated percentage of hybrid offspring following mixed pollination was around 15 % in both species. The species in the examples above are not Geranium relatives; moreover, the methodologies used to evaluate reproductive success were quite different from the methodology of this study. Despite this lack of comparability, if we venture to compare the results of those studies with our results, we find that the impact of reproductive interference in our Geranium species might be less than half of its impact on the plants in these examples. These examples also differ from our species because those studies reported that in each studied pair, reproductive interference was strongly asymmetrical, whereas we found that reproductive interference between G. thunbergii and G. wilfordii was bidirectional and almost symmetrical. The relatively low impact of this bidirectional reproductive interference may function to maintain the allotopic distributions of these Geranium species, instead of one of the species pair being replaced by the other. We need more examples to compare with our results, especially ones that show distributional relationships similar to that in our results, to test the plausibility of our interpretation that the observed distributional relationships are associated with the strength of the reproductive interference.

Although we do not know how long these two Geranium species have maintained the observed allotopic distributional relationship, we suppose that this distributional relationship has been maintained in part through the production of hybrids. According to Buggs (2007), the low fitness of hybrids can minimize spatial mixing of the parental taxa and lead to narrow hybrid zones. The presence of the hybrids with low fertility may relax the reproductive interference between the two Geranium species, because the hybrids located between patches of each species prevent the direct transfer of heterospecific pollen. Instead, reproductive interference may occur from the hybrids to the parent species, because the hybrids also produce fertile pollen, though not in large amounts. We observed some variation in the hybrid morphology, so it is possible that further introgression may occur by backcrossing between the hybrids and their parent species. Future studies are needed to further investigate the effects of hybridization on the population dynamics of the parent species.

Our finding of small-scale spatial partitioning in Geranium species suggests that the focal scale at which relationships between reproductive interference and the spatial distributions of species are examined needs to be reconsidered. Most plant ecologists would probably describe the distributions of G. thunbergii and G. wilfordii in our study area as sympatric, whereas we showed that at small spatial scale their distributions are significantly exclusive. To recognize reproductive interference between native organisms, the distributional relationships between the involved species must be evaluated in terms of their effective fertilization range and their environmental fitness.

Conclusions

Our findings suggest that ongoing reproductive interference can be recognized even between native wild plant species if its intensity is not very strong and their distributional relationships are examined at an appropriate scale. However, when reproductive interference is intensive or unidirectional, it would still be difficult to identify interference as the key driving force behind spatial partitioning observable at present in the wild, because it would rapidly lead to exclusive distributions. We do not yet know how to detect the ‘ghost of reproductive interference past’ (analogous to the ‘ghost of competition past’; Connell, 1980) between wild plant species or how to prove that past reproductive interference was responsible for exclusive distributions observed in the present. Nevertheless, we have some hope that it may be possible to discover some trace of the ‘ghost’, especially if hybrids still exist. Hybrids of some species pairs are known to occur in a region where only one member of the pair is distributed at present (e.g. Iris, Anderson 1936; Quercus, Tucker 1968; and Arisaema, Kakishima 2012). Unless the hybrids were introduced to the region by a long-distance dispersal mechanism, this phenomenon may suggest that both parent species used to be present in the region but that one of them has since been excluded, leaving behind only the other species and the hybrids. Where such a discrepancy between a hybrid zone and the parents’ distributions is observed, an investigation of the distributional relationships of the parent species and the hybrid, accompanied by a pollination experiment and a model simulation of the effect of both reproductive interference and dispersal on the distributions, might be able to demonstrate past reproductive interference between the species.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Figure S1: map of study sites. Table S1: GLMM analysis results for seed set in G. thunbergii and G. wilfordii plants.

FUNDING

This work was supported by KAKENHI grants 22570088 and 26440211 to S.N., grant 19770023 to K.-I.T. and grant 20657005 to T.N. from the Japan Society for the Promotion of Science and the Ministry of Education, Culture, Sports, Science and Technology, Japan.