Flower morphological differentiation and plant-pollinator interactions among sympatric Aframomum species (Zingiberaceae) with floral trumpet type in the tropical African rainforest

Background and aims – Diversification in plant-pollinator interactions based on floral diversity is potentially a mechanism of coexistence in angiosperms. However, besides high floral diversity, some genera seemingly exhibit the same floral type in many of their species. This contradicts some expectations of competitive exclusion. We thus tested on a finer flower morphological scale whether five sympatric Aframomum species (61 spp., Zingiberaceae) in southeastern Gabon exhibiting the same general floral type (trumpet) were differentiated, and whether this resulted in different “pollinator niches”. Material and methods – We carried out a detailed survey measuring 18 flower morphological parameters as well as nectar volume (μl) and sugar concentration (% Brix) on five flowers per species and locality. Furthermore, we observed inflorescence phenology and pollinator activity from 8 am to 4 pm for 12 to 50 hours per species and conducted pollinator exclusion experiments. Key results – This study proves fine-scale flower morphological and resource differentiation within the trumpet floral type. Pollination-relevant parts of the flowers, however, remain constant across species. Our pollinator observations reveal the same broad bee pollinator spectrum for all observed simultaneously flowering sympatric species. Conclusion – As we could not detect a pollinator-based differentiation in the studied sympatric Aframomum species we assume that species boundaries developed randomly by genetic drift during geographic isolation in the past. The trumpet floral type and its pollinator guild, however, were maintained due to similar selection pressures in comparable habitats during isolation and are potentially an advantage for increased pollinator attraction through co-flowering.


INTRODUCTION
Floral divergence and pollinator specialization has driven angiosperm diversification and is potentially a mechanism of angiosperm coexistence (Givnish 2010). Examples of pollinator-driven divergence include Aquilegia L. (Ranunculaceae) (Bastida et al. 2009), Disa P.J.Bergius (Orchidaceae) (Johnson et al. 1998), and Salvia L. (Lamiaceae) (Claßen-Bockhoff et al. 2004;Wester & Claßen-Bockhoff 2007). Specialised plant-pollinator associations lead to a strict separation in pollen transfer, preventing any admixture. The association of plants with certain pollinators and the exclusion of others is reached through e.g. different spur or tube lengths, floral sizes, colours, odours, and divergences in many other floral characteristics.
We can often predict the pollinators of a flower from its floral biology (e.g. colour, scent, reward, opening time, etc.) (Ollerton & Watts 2000). This set of floral characteristics is called "pollination syndrome". These specialized plantpollinator associations ensure optimization of pollen transfer based on a highly differentiated floral morphology optimizing mechanical fit between plant and pollinator (Armbruster et al. 1994;Johnson et al. 1998;Ley & Claßen-Bockhoff 2009). The plant species is either visited by one or several specific pollinators or by the same set of pollinators as its sympatric neighbours but with pollen deposited in a rather specific place on those pollinators -both guaranteeing a rather unique pollen transfer (Grant 1994). The pollinator's floral spectrum, its pollen transfer efficiency, flight radius, and pattern of floral visitation within and among plant individuals and populations shape the patterns of gene flow within and among plant species influencing their genetic differentiation and fitness, e.g. selfing versus outcrossing and/or hybridisation (Lunau 2004).
In contrast to these conspicuous floral radiations (e.g. Disa: Johnson et al. 1998;Marantaceae: Ley & Claßen-Bockhoff 2011;Costaceae: Ricklefs & Renner 1994, Specht et al. 2001Salvia: Wester & Claßen-Bockhoff 2007) there are also plant species that exhibit morphologically very similar floral types (see selected groups of species within Impatiens: Grey-Wilson 1980;Marantaceae: Ley 2008; Aframomum: Ley & Harris 2014). These contradict our expectations of competitive exclusion (either through different pollinator species or mutual pollen incompatibility), or at least suggest a strong possibility of high rates of interspecific pollen transfer. Interspecific pollen transfer is one of the mechanisms underlying potential competition among plants for pollinators by either the interference with conspecific pollen by stigma clogging and during fertilization and/or the reduction of the amount of pollen transferred between conspecific flowers (Morales & Traveset 2008;Moreira-Hernández & Muchhala 2019). Successful interspecific pollen transfer can lead to the dilution of species boundaries, i.e. hybridisation as seen in some Marantaceae (Ley 2008).
Still, morphologically similar flowers with (near-) identical pollinator niches showing synchronized flowering can be at an advantage when the number of pollinators attracted to an area increases at an accelerating rate with an increasing number of flowers (Tachiki et al. 2010). In this case, facilitation of pollination by different species exceeds the negative influence of interspecific plant competition. In some species with morphologically very similar flowers, the flowers might differ in other than morphological traits such as the quality of their rewards (e.g. odour, nectar composition) leading at least to partially different pollinator spectra and thus further species separation (Urru et al. 2010). Due to the rather uniform appearance of these systems at first sight they remain little investigated regarding their current degree of differentiation and their ecological-evolutionary background.
Such a system of a large recent radiation of the same floral type can be found in the species-rich tropical African genus Aframomum (61 spp., Zingiberaceae; Dhetchuvi 1996;Harris et al. 2000;Auvray et al. 2010;Zakaria 2013;Harris & Wortley 2018). Its species are broad leaved perennial herbs from the rainforest understorey, gaps, edges, and savannas distributed from Senegal to Madagascar (Dhetchuvi 1996). They form large conspicuous flowers of five different floral types (trumpet, apron, open, short tube, collar) with one single floral type dominating (trumpet) (Ley & Harris 2014). Based solely on morphological characters and the concept of "pollination syndromes" (Ollerton & Watts 2000), first hypotheses about their pollinating species have been pronounced (Ley & Harris 2014).
The dominant floral trumpet type in about 60% of all Aframomum species consists of a large pink tubular flower of delicate tissue with a large landing platform and a horizontal to slightly vertical floral entrance with often conspicuous yellow nectar guides (UV patterning is not yet documented) and some variation in size between species (Ley & Harris 2014). Species of this floral type can be found throughout tropical Africa with the centre of species diversity in Cameroon and the Republic of the Congo. Here, many of these species can be found in sympatry in the same habitat (Ley & Harris 2014). This floral type was hypothesized to be pollinated by medium-sized to small bees (Ley & Harris 2014), however, direct field observations are still lacking (except Lock et al. 1977). Due to a lack of genetic studies, there is no evidence of hybridization yet. The fruits of Aframomum are large and conspicuously red in colour with many small black seeds and distributed by monkeys (Lekane Tsobgou 2009;Zakaria 2013).
The aim of this study was to reveal the pollination system of five Aframomum species with trumpet type flowers, growing in sympatry in the landscape mosaic of forest and savanna in southeastern Gabon. We perform a finescale morphological survey of the flowers, direct pollinator observations and pollinator exclusion experiments. For the savanna species A. alboviolaceum (Ridl.) K.Schum., we additionally compared floral morphology and pollinator visitation among different sites.
Specifically, we wanted to test 1) whether species exhibiting the same general floral type (trumpet) were differentiated morphologically in certain characters and/ or their reward and whether this differentiation may place the species into different "pollinator niches". 2) We further wanted to detect the actual pollinator spectrum of each species through direct pollinator observations. Here, we explicitly tested for pollinator sharing (different species using the same pollinator spectrum, e.g. to increase pollinator attraction) versus pollinator exclusion (different species using a differentiated pollinator spectrum, e.g. to prevent interspecific pollen transfer). 3) Moreover, we wanted to establish the importance of pollinators for fruit set in Aframomum. Thus, we conducted pollinator exclusion experiments documenting fruit set in the presence and absence of pollinators.

Study localities and material
The study was conducted in southeastern Gabon, which is characterized by a mosaic of forest and savanna (Walters 2012 At each locality, flowering Aframomum species were identified using available identification keys (Koechlin 1964;Dhetchuvi 1996) and comparing own collections with herbarium specimens of the National Herbarium of Gabon in Libreville (LBV) checked by David Harris (Royal Botanic Garden Edinburgh).

Floral phenology, morphology, and resources
Field work was conducted during the peak flowering of Aframomum species which is between September/ October and December/January (Harris & Wortley 2018). Inflorescence phenology was monitored daily on five to 68 inflorescences of five to 48 individuals per species for 36 days at Bakoumba (15 Oct. -19 Nov. 2015), for five days at Ossélé (10-15 Aug. 2015), and for 27 days at Franceville (25 Nov. -20 Dec. 2015). A total of 18 morphological parameters (Ley & Harris 2014; supplementary file 1 table S2, S3) were measured on five flowers (living material) per species and locality with each flower coming from a different individual. We used a calliper with a precision of ± 0.01 cm. In addition, the amount of nectar (μl) and its sugar concentration (% Brix) were measured in the morning on newly opened bagged flowers using a capillary pipette and an Eclipse refractometer.

Pollinator observations
Pollinators were observed in one-hour intervals for 12 to 50 hours per Aframomum species and locality between 8 am and 4 pm. Visiting insects were termed pollinators when they entered the tubular flower and came into contact with the reproductive organs (thecae and stylar cavity). The number and respective length of visits of a pollinator to a flower was noted. At the end of each observation period at a locality, a few individuals of each pollinator species were collected for morphological measurements with Optika (size of the head, and thorax & abdomen). Bees were identified by Connal Eardley (Pretoria, South Africa). Due to a scarcity of flowering individuals no pollinators could be observed on A. longipetiolatum Koechlin.

Breeding system and fruit set
Natural fruiting was documented on five open and five bagged (pollinator excluded) inflorescences per species (one inflorescence per individual). Fruit set was determined after about one month by dividing the number of fruits produced per inflorescence by the number of flowers produced on each inflorescence multiplied by 100.

Statistical analysis
Morphological dissimilarity among flowers of different Aframomum species was studied via a principal component analysis (PCA) using 18 quantitative morphological parameters and two nectar traits applying the method prcomp in R v.3.3.2 (R Core Team 2016). To show which traits are behind the group differentiation differences among PCA groups were tested for significance using linear mixed models (R package lme4; Bates et al. 2011) followed by a post-hoc Tukey test (Tukey 1957). Before analyses, traits were visually inspected for normal distribution and log or exponential transformed whenever necessary using PCA groups as fixed and the respective morphological parameter as random factor. To evaluate whether there was a correlation between nectar sugar concentration and nectar volume, a linear regression assay was performed in PAST (Hammer et al. 2001).

Floral morphology and resources
Based on 18 quantitative flower morphological parameters, nectar volume, and sugar concentration the five studied species with trumpet flowers were divided into three groups using principal component analysis (PCA: A, B, and C) (fig. 2) (table 1, supplementary file 1 table S2). The first two axes explained 37% and 24% of the total variance in the data, respectively (supplementary file 1 table S4). PC1 represented traits of floral size (length of calyx, labellum, petals, filament, and style) and PC2 dimensions of inner organs such as stylar head and thecae and their relative position to each other (supplementary file 1 table S5). The PCA grouping was supported by statistically significant differences in floral measurements among groups (table 1, fig. 3C). There was a negative correlation between nectar volume and sugar concentration (linear regression test, p value = 1.16 × 10 -8 < 0.001). Yellow nectar guides on the labellum at the floral entrance were present in all species ( fig. 1).

Inflorescence and flower longevity
Inflorescences of the five Aframomum species produced one (A. subsericeum) to up to 20 (A. hirsutum) flowers (supplementary file 1 table S2). The life of an Aframomum flower lasted, independent of species, one day from about 8 am till the afternoon. Only in the savanna species, the rim of the floral tube wilted a bit earlier in the afternoon at around 2 pm instead of after sunset as in the forest species. One to three flowers per inflorescence opened every or every second to third day (supplementary file 1 tables S6-S8). Thus, the length of flowering of an inflorescence varied between species lasting from 1 to 3 days to 3 to 4 weeks (supplementary file 1 tables S6-S8). Through simultaneous blooming of several inflorescences per species and population there were always > 5 flowers open per species at our study localities (supplementary file 1 tables S6-S8).

Plant-pollinator interaction
The study of the pollinator community of the five Aframomum species revealed about 11 different pollinator species altogether. This included eight species from two hymenopteran orders (Apidae, Halictidae), a fly (Diptera), and two species of butterflies (table 2, for specimens refer    Bee pollinators would land on the upper labellum tip of the flower which is horizontally arranged and then crawl into the floral tube. Bees would stay inside the floral tube for half a minute to up to about four minutes. Also, the butterflies would crawl inside the floral tube to reach the nectar.

Fruit set in the presence and lack of pollinators
Fruits started to appear about three weeks after flowering. This initial fruit appearance was used to determine fruit set. Natural fruit set varied from 3% (A. hirsutum (Pop2)) to 78% (A. alboviolaceum at Ossélé) (table 4). It was inversely related to the number of flowers produced per inflorescence. There was no correlation between visitation rate per flower and fruit set (supplementary file 1 fig. S2). In the pollinator exclusion experiments no fruits were observed in any of the species (table 4).

Fine-scale differentiation within the uniform trumpet type flowers
The detailed morphological investigation confirmed a large size range with three specific groups within the apparent uniform trumpet type flowers from different species of Aframomum (this study but compare also Ley & Harris 2014 for a larger species range). The flowers of the savanna species were by far the smallest. However, in all cases the relative position of inner organs that determine pollination efficiency was rather similar (i.e. the distance between pollen sac and labellum and also the width of the arch formed by the lateral appendices of the anther), suggesting equal-sized effective pollinators for all species. The smaller corolla size in the savanna species could represent an adaptation towards elevated desiccation in the open savanna habitat just as found in leaves (Tomlinson et al. 2013). However, the overall floral type with its delicate tissue remained the same. Instead, we observed an earlier wilting in the afternoon. This, however, was at the end of the daily height of insect activity (Ley & Claßen-Bockhoff 2009). Thus, we hypothesize that the effect of strong selection forces for desiccation tolerance in the savanna on floral morphology might be dampened by the short overall flowering time of an Aframomum flower (max. 1 day), their phylogenetic constraints (ancestral floral type: trumpet type; Ley & Harris 2014) and the high fitness component of successful pollinator interactions achieved by the floral trumpet type (see also further on).
The detected uniform floral morphology (except for size) in the investigated trumpet type flowers leaves only floral resources as potential source of floral divergence in Aframomum (see e.g. Silva et al. 2020). The nectar sugar concentration influences the viscosity of nectar (Kim et al. 2011) which can limit feeding to specific pollinators. In the investigated Aframomum species, the nectar sugar concentration is in the range of bee pollinated species (Roubik et al. 1995;Perret et al. 2001;Ley & Claßen-Bockhoff 2009). This suggests the same specific pollinator group and not a divergence in pollinator spectrum among species. The high intraspecific variation in nectar sugar concentration found in the savanna species (30-37%) might be tied to variations in microclimate and soil (Cruden 1976;Herrera et al. 2006;Farkas et al. 2012  (~37%) (Zajácz et al. 2006). A further potential species differentiation might lie in the amino acid composition of the nectar ). However, this needs further investigation in Aframomum. Thus, so far, based solely on the fine scale floral morphology and the nectar resource (volume and sugar concentration) of the five species with trumpet type flowers, we might expect the same pollinator spectrum for all investigated species.

Generalized bee pollinator-sharing in trumpet type flowers
In accordance with the lack of pollination-relevant differentiation in fine scale floral morphology and resources, we observed mainly the same pollinator species (mainly bees and two small butterflies) for all our studied co-flowering Aframomum species. However, to get to the full spectrum of bee pollinators for each Aframomum species, it will still be necessary to include many more sites per species as shown by the different bee pollination spectra by site in A. alboviolaceum.
Bee pollination in the trumpet type was already hypothesized by Ley & Harris (2014). The purple colour, tubular shape, the lengthening of the labellum as a landing platform, the yellow nectar guide, and the rather high concentration of sugar present in the flowers of Aframomum are ideal floral traits for bees and some small butterflies (Brisson et al. 1994;Herrera et al. 2006;Ley & Claßen-Bockhoff 2009;Ley & Harris 2014).
This strategy of different sympatric plant species attracting the same pollinators is termed "pollinator sharing" (Macior 1971). An increased local floral display is reached through the simultaneous inter-and intraspecific flowering of sparsely, thus cost-effectively, flowering individuals (see also Moeller 2004). This simultaneous flowering has proven in other plant species to attract in total a greater and more stable pollinator community to a given area through the provision of a continuously rich food resource and thereby increases the individual rate of successful pollination (Gottsberger 1989;Tachiki et al. 2010). Similar pollination patterns have been observed elsewhere in the tropics (Schemske 1981;Ley & Claßen-Bockhoff 2009;Wang et al. 2016) and might also play a role in these sparsely flowering sympatric Aframomum species. Additionally, adjacent habitats (here: savanna and forest) can potentially contribute to a reciprocally-enriched pollinator community (Schüepp et al. 2012;Stanley & Stout 2014).
All observed pollinating bees belong to the group of longue-tongued bees (Brisson et al. 1994;Eardley et al. 2010). The bending of the floral tube and the delicate tissue seem to exclude birds and other larger animals from visiting these flowers -at least during our observations. Other smaller insects have also been observed visiting the open accessible trumpet type flowers, however, due to their small size it is unlikely that they are relevant pollinators because on their way down into the floral tube to get to the nectar they do not come into contact with the reproductive organs. The distance between the labellum on which they walk and the thecae is simply too wide, which prevents them from coming into contact with the pollen (unless they feed on the reproductive organs and thereby come into contact with the pollen).
Using the same pollinators across trumpet type species opens the strong possibility of high rates of interspecific pollen tansfer. Still, there are constant morphological differences between species of Aframomum and longaccepted species concepts in the genus (Harris & Wortley 2018). We therefore assume that species are largely genetically incompatible. This incompatibility might have developed randomly by genetic drift during geographic isolation in the past (Maley 1996;Ley & Harris 2014;Couvreur et al. 2020). However, detailed experiments on cross species compatibility have still to be conducted (Wang et al. 2016). Also, the current summary of several different bee species under "small black insects Ossélé" and "small black insects Bakoumba" still hold a potential of a partial differentiation of pollinators between species.
The different composition of bee pollinator spectra by site might mirror differences in available local habitats as  breeding and feeding site for the different bee species (Viana et al. 2012). Apis mellifera was the only species observed at all three study localities in all habitats and on all studied species. This bee is widespread and known as a generalist pollinator (Fohouo et al. 2010;Hagen & Kraemer 2010;Giannini et al. 2015). The very high frequency of Apis mellifera at Bakoumba and Franceville can additionally be explained by the local presence of bee hives (Nzigou Doubindou pers. obs.). It needs to be checked whether Apis mellifera totally replaces a more diverse bee community at Franceville.
The visitation frequencies differed among species of pollinators and localities and thus pollinator preference and pollen transfer efficiency need further testing (Silva et al. 2020). However, both frequent and rare visitors might be effective pollinators, together contributing to the reproductive success of the species (Schemske 1981;Moeller 2004). The bee visitation frequency was highest in flowers of the savanna. This might be related to the higher solar radiation in this habitat -as bees are ectotherms (Hagen & Kraemer 2010). At localities or in years of low visitation frequency, the large number of flowers per inflorescence as found in A. hirsutum might be an advantage as it yields a long flowering period (an individual flower lasts a single day only) and thereby increases the probability of effective pollination.

Breeding system and fructification
All studied Aframomum species are xenogamous, thus they need pollinators to produce fruits as shown by our pollinator exclusion experiments. Xenogamy increases the likelihood of cross-fertilisation by which genetic diversity within a species is maintained and/or increased (Bawa 1990;Brisson et al. 1994;Ley & Claßen-Bockhoff 2009). Further tests of self and cross-species compatibility in Aframomum are needed to establish whether fruits can arise from selfing, probably geitonogamy, and whether pollinator sharing might facilitate hybridisation. The observed spatial isolation of style head and thecae through their respective relative position to each other in the flower rather contradicts the potential for autogamy (see Ley & Harris 2014). The potential for geitonogamy is reduced through the sparse flowering of an inflorescence (rarely more than one simultaneously open flower): geitonogamy would only be possible as a result of pollen transfer across adjacent clonal individuals.
Self-incompatibility might be one potential explanation for the detected low natural fruit set of 3-8% in the richflowering A. hirsutum, in contrast to a rather high fruit set of > 60% in all other sparsely flowering species (except A. longipetiolatum -no data) (compare Sutherland & Delph 1984;Ley & Claßen-Bockhoff 2013). However, the low fruit set in A. hirsutum could also be an effect of resource limitation. This hypothesis builds on the idea that the production of the large and thus energetically costly fruits in Aframomum is restricted by available resources in favour of e.g. genetically "advantageous" (i.e. outcrossed) fruits (Sutherland 1986;Horvitz & Schemske 1988). Currently, it seems as if in the studied Aframomum species a given inflorescence cannot bear more than one to three large fruits -but more specific data is still needed to prove or reject this idea.

SUPPLEMENTARY FILE
Supplementary file 1 -Information on the specimens of Aframomum collected (coll.) at the three study sites in Gabon (table S1); average morphological measurements (cm) of five sympatric Aframomum species (table S2); average morphological measurements (cm) of Aframomum alboviolaceum at three different localities (table S3); PCA statistics of morphological measurements (tables S4, S5); results of statistical differentiation between PCA groups, phenological data (tables S6, S7, S8); information on voucher specimens of pollinators and relationships of fruit set (table S9)