Plants Are the Drivers of Geographic Variation of Floral Scents in a Highly Specialized Pollination Mutualism: a Study of Ficus Hirta in China

Background Floral volatiles play an important role in pollinator attraction. This is particularly true in obligate brood site pollination mutualisms. The plants generally produce inconspicuous owers and depend on odours to attract to their inorescences specialised pollinators that breed in their oral structures. Little is known about the processes shaping the micro-evolution of these oral odours. Here, we investigate geographic variation of oral odour in an obligate host-specic brood site pollination mutualism where plant and pollinator genetic structures are different, Ficus hirta and its specialised pollinators. Results We evidence progressive geographic divergence of oral odours. The pattern of variation ts plant genetic structure but differs from pollinating insect structuring into species and populations. In our study system, the evolution of receptive oral odour presents a pattern that is not distinguishable from neutral drift that is not canalised by the insects. We propose that this pattern characterises obligate brood site pollination mutualisms in which pollinators are host specic and dispersal is limited. Insects with their short generation times and large population sizes track variation in host receptive inorescence odours. Plants are the drivers and insects the followers. Strict sense plant-insect co-evolution is not involved. In contrast, stabilizing selection may be at work in more dispersive brood site pollination mutualisms, while pollinators may mediate local interspecic plant oral odour convergence when plant species share local pollinators. of resources used by the pollinator offspring and in terms of the organs emitting oral odours, their olfactive signalling seems to follow a common set of evolutionary rules. We propose a general framework to investigate the evolution of oral odours within plant species involved in obligate brood site pollination mutualisms. When pollinators are host specic, plants are the drivers of the evolution of oral odours, and pollinating insects, with their large population sizes and short generation times, are the followers. The presence and the intensity of vegan [63]. We relative of by gs (semiquantitative Data standardized prior the analysis. Two-dimensional plots constructed using the function algorithm. Pairwise Bray–Curtis dissimilarity which ranges between and stress test overall signicantly we carried permutational analysis of variance tests on a The with the after data standardization were using the after [64], test dissimilarities of contributions of compound to the Bray-Curtis dissimilarity. compounds most to the locations.

Ficus provide a highly diversi ed brood site pollination mutualism for investigating factors affecting the evolution of signalling. Floral odours produced by receptive gs are generally species speci c [31,32] and play a central role in pollinator attraction [33]. Nevertheless, only limited data is available on geographic variation in g oral odours, and in these cases, plant and insect present share a same spatial pattern of genetic structure [34,35]. To establish whether plant or insect drives the evolution of oral odours, it is necessary to analyse spatial variation of oral odours in a system in which insect and plant present contrasted spatial genetic structures. This is the case for Ficus hirta and its pollinators.
Ficus hirta Vahl is a widely distributed shrub growing throughout continental South-East Asia from the Himalayan foothills to Java. It presents a pattern of spatial genetic structure suggesting genetic isolation by distance without genetic discontinuity across continental South-East Asia [29,36]. It is pollinated by a set of parapatric wasp species forming the species complex of Valisia javana sensu lato [29]. In China, F. hirta is pollinated by sp1 in the south-east and the south, from Fujian province to Guangxi province, while it is pollinated by sp2 westwards in Yunnan province. Throughout continental south-eastern to southern China, over more than 1000 km, sp1 forms a single population, while on Hainan island, 20 km off the coast, it is pollinated by a different population of sp1 [29,37]. The contrasted genetic structure between pollinators and the host g allows addressing the question of how variation in oral odour is determined. If the insects are driving the selection for receptive g odour variation then we expect to observe two or three groups of receptive g odours: one in Yunnan, one in south and south-east China and the same or a different one in Hainan island depending on the speed of evolution. Alternatively, if variation in receptive odours is driven by plant spatial genetic structure, then we predict a simple pattern of geographic differentiation by distance. Finally, if there is ongoing stabilizing selection then we predict no geographic variation in receptive g odour. We investigated geographic variation of Ficus hirta receptive g odours in China to answer these questions.

Results
Overall Odour Pro le Across nine locations, a total of 45 receptive g odour samples were collected and analysed (Fig. 1). Thirty eight different volatile organic compounds (VOCs) were detected and identi ed in the odours emitted by the receptive gs ( Table 1). The identi ed scent compounds included 3 fatty acid derivatives, 6 monoterpenes and 24 sesquiterpenes, while 5 compounds remained unidenti ed. Odours from locations pollinated by sp1 were mainly composed of a few sesquiterpenes while at XTBG, pollinated by sp2, the odours contained markedly higher quantities of monoterpenes, mainly (E) -β-ocimene but also linalool ( Table 1). Compounds that were present throughout all locations accounted for 84-95% of local emissions, depending on location. VOC emissions varied among locations and gs in south-eastern locations emitted higher quantities of volatiles than average (general ANOVA, P < 0.001, locations Ning, Sui and Sha different from mean, respectively p < 0.05, p < 0.05 and p < 0.01, all other locations not signi cantly different from mean). Table 1 Occurrence and relative proportion (% mean ± SD) of volatile compounds from three classes, and total amount, detected in the bouquets of scents emitted b * compound identi cation con rmed by comparison of mass spectra and RI with those of authentic standards; N= number of individuals sampled; O = number of individuals in which that compounds was found; RI = Kowat retention index; n.d. = compound not detected; in bold compounds that represent more than 5% in the average bouquet of scents in at least one site.

Geographic variation in oral scents
The scatterplot obtained by NMDS ordination based on the Bray-Curtis dissimilarity index (stress = 0.172) is shown in Fig. 2. Overall there was signi cant variation among locations in the relative proportions of the different compound in odours emitted by receptive gs (PERMANOVA, F 8,45 = 5.0198, P = 0.001).
The combined results of pairwise comparisons (Table 2) and the NMDS plot suggest that samples of oral odours can be grouped into three geographic clusters, namely a south-eastern cluster (3 locations: Ning, Sha and Sui), a southern cluster (4 location: SCBG, DHS, Nan and Wan) and a south-western cluster (XTBG). The assignment of location Ding to cluster is ambiguous as it is not signi cantly different from location Ning and from location Wan in the PERMANOVA analysis, and it has an intermediate position between the south-eastern and the southern group in the NMDS plot. Analysis of the similarity percentage (simper) reveals that the quantities of 4 to 6 compounds explained more than 30% of the dissimilarity between locations and relative quantities of both main and minor compounds, involving four to six compounds, explained more than 30% of the dissimilarity between locations. A second set of correlations were examined using on a single odour composition value for each location in order to avoid potential pseudo-replication problems. There was a signi cant correlation between chemical distance and geographic distance when including all the locations (Mantel statistic r = 0.3423, p = 0.028) and when removing location XTBG (Mantel statistic r: 0.3525, p = 0.043). When including only southern and south-eastern continental locations (6 data points) the test became marginally non-signi cant (Mantel statistic r = 0.45, p = 0.072).

Discussion
The results of the present study provide new insights into the possible drivers of micro-evolution of ower signaling in highly specialized mutualistic plantpollinator interactions. The geographic pattern of increasing receptive g odour differentiation with distance is analogous to the pattern of genetic isolation by distance exhibited by the plant. It is strikingly different from the pattern of genetic structuring of the pollinating insects into species and populations [29].
Hence, there was no evidence in favour of the stabilising selection predicted by theoretical models [30], and no evidence in favour of strict sense plant-insect co-evolution for oral odour composition and its perception by the insects. These results demonstrate that, in F. hirta, the plant is the driver of oral odour evolution.
Structuring occurs at much larger spatial scales in more dispersive systems such as the one represented by F. racemosa. In this species, the plant is structured into large gene pools, covering huge surfaces, and presenting no or almost no spatial genetic structure. Each gene pool is pollinated by a wasp species forming a single population [38]. In that situation, no differences in receptive g odours were observed between two locations, 900 km apart, corresponding to a same gene pool [34]. Large gene ow could limit geographic variation in oral odour either by limiting drift or by limiting adaptive differentiation of oral odours. A direct consequence of the lack of geographic differentiation of oral odours is that pollinators drifting in the wind above the canopy and dispersing over large distances [39] will still recognise receptive gs: long distance plant gene ow facilitates long distance insect colonisation and reciprocally in a feedback loop. The opposite may be true for the lower dispersal system of F. hirta. Differentiation of local oral odours may select for reduced pollinator dispersal, and this will result in stronger spatial genetic structure in the plants, in a self-reinforced process. Such self-reinforced processes may be at the origin of the divergent dispersal strategies of g-pollinating wasps. Pollinators of monoecious Ficus species generally disperse by drifting in the wind above the canopy while pollinators of the spatially more structured dioecious Ficus species, including F. hirta, are generally not observed in the aerial plankton [39].
The evolutionary stability observed in the monoecious F. racemosa could be present in other systems such as some species of yuccas. Indeed, Yucca lamentosa oral odours do not vary across its range [40,41]. Both Y. lamentosa and its pollinator Tegeticula yuccasella presents limited genetic differentiation across their ranges suggesting strong gene ow, reminiscent of the situation in F. racemosa [40,42]. A broader study of receptive oral odour variation among Yucca species is required to uncover patterns. Indeed, the oral odours of Y. elata are undistinguishable of those of Y. lamentosa while other Yucca species produce distinctive odours [43]. However, Yucca moth and pollinating g wasp population dynamics are not comparable. Individual Yucca moths may survive in the soil in diapause for 30 years [44], while g wasps only survive one or two days outside gs [45]. This short life span leads to local pollinating wasp population extinctions during climatic accidents. The distinctive genetic signature of such dynamics is lack of spatial genetic structure combined with small effective population size in species presenting very large populations [46].
Ficus septica in the Philippines and Taiwan provides complementary information on patterns of geographic variation [35]. Ficus septica is structured into at least three gene pools (Taiwan / Luzon-Negros / Mindanao) and is pollinated by a different black coloured wasp species belonging to the Ceratosolen bisulcatus species group in each of the three regions. Nevertheless, a fourth wasp species, belonging to the same species group, C. jucundus, has colonised the whole region, bridging the odour differences [35]. This observation suggests that, given su cient time, receptive g odour differentiation within hostspecies does not preclude wasps from expanding their range.
Geographic variation in oral odours has been investigated throughout numerous populations in the facultative brood site pollination mutualism between Lithophragma spp. and Greya moths. As in Ficus hirta, oral odours varied among locations within species, and the difference in oral odours increased with geographic distance (Fig. 4 in [47]). Further, odour distance between populations of two sympatric but not syntopic clades of Lithophragma did not correlate with geographic distance, demonstrating that there was no concerted odour evolution between the two clades [47]. These results suggest that in this example too, the plants are the drivers of oral odour evolution and the insects are the followers, despite strong spatial genetic structure in the Greya pollinators [48].
Models of mutualism predict the occurrence of stabilising selection, especially when individuals of one species need to interact with many mutualistic individuals of another species [49]. Lithophragma spp. individuals, and even more Ficus spp. trees, need to interact with many individuals of their insect pollinator species, but the evolution of their oral odours do not conform to prediction. However, because of the short generation time of pollinators comparatively to the plants, the theoretical models do not apply. We suggest that in such systems, local populations of pollinators may rapidly track the slow evolution of local oral odours.
In the case of plant-species sharing pollinators, a simple prediction could be local convergence of oral odours between the different plant species associated with a same set of insects. This has been shown in cases of plants sharing pollinators in Cycads [50], in Ficus [51,52] and in Glochidion [53,54]. We may further predict that plant species involved in different pollination rings in different localities may evolve different attractive odours among localities as suggested by results on Cycads [50]. Results on Cycads suggest within pollinator-species evolution of divergent responses to oral odours, adjusted to the odours produced locally by their hosts [50], as suggested above for C. jucundus. In such systems, the pollinators are the mediators of the selection for odour convergence among co-occurring plant species.

Conclusion
While brood site pollination mutualisms are highly diversi ed in terms of resources used by the pollinator offspring and in terms of the organs emitting oral odours, their olfactive signalling seems to follow a common set of evolutionary rules. We propose a general framework to investigate the evolution of oral odours within plant species involved in obligate brood site pollination mutualisms. When pollinators are host speci c, plants are the drivers of the evolution of oral odours, and pollinating insects, with their large population sizes and short generation times, are the followers. The presence and the intensity of geographic differentiation of oral odours will depend on spatial genetic structure of the host-plant. When plant species share pollinators, their oral odours converge, through a mimicry process among plants, mediated by the pollinators. In all cases, pollinators have the potential to expand their range by evolving the capacity to recognise oral odours of their host species from new locations. This framework will allow testing predictions on oral odour evolution.

Study system and collection sites
In a previous study samples collected from all the sites investigated here had been included in a broader genetic study of F. hirta and its pollinating wasps throughout China to Java. All the plants were shown to belong to a single species presenting clinal genetic variation while the pollinators belonged to a single species group [29]. Reference herbarium samples for that study were deposited at IBSC under numbers 817854-817899. FK formally identi ed the specimens as Ficus hirta, by comparing live plants from locations SCBG, XTBG and the voucher specimens collected by YH throughout the sampling range, with descriptions and with reference herbarium samples, mainly at P, identi ed by EJH Corner and/or CC Berg. While F. triloba and F. hirta may sometimes be tricky to distinguish in herbarium material, they are easily distinguished in the eld. In this study, sample identi cation in the eld was done either by XD and HY or by XD and FK.
Ficus hirta Vahl. (section Ficus) is a shrub or small tree approximately 1-3 m high. Figs are produced year-round [55]. Figs develop asynchronously within the tree, and a few plants are su cient to produce pollinators throughout the year [55,56]. The production of receptive gs peaks in May-June [55]. In June-July 2019, we collected oral odours from receptive gs in 9 locations distributed across China, with 3 south-eastern locations (Ning, Sha and Sui), 5 southern locations including 2 in Hainan, and 1 south-western location (XTBG) in South Yunnan. (Table 3, Fig. 1). Collections were made on wild-growing plants and we attempted to sample at least 5 individuals per location. All the odour samples collected came from the same season.

Floral Odour Collection
We used the head-space technique following methods initially developed for Silene [57] and that have been successfully used in several Ficus species [34,51,58,59]. As the size of receptive gs varied geographically [60], in order to collect su cient quantities of odour for the analysis, the number of gs used in each bag was adjusted according to g diameter: for south-eastern locations 13 ± 4, for southern locations 17 ± 4, and for the south-western location 19 ± 10. Odour collection was performed under natural light between 10:00 am and 5:00 pm, corresponding to the insects' period of maximum activity during our eld season.
Receptive gs were enclosed in a polyethylene terephtalate (Nalophan®, Kalle Nalo GmbH, Wursthüllen, Germany) bag for 30 min. Then, air was pulled out of the bag ( ow rate: 200 mLmin − 1 ) through a Chomatoprobe lter ( lled with 1.5 mg of Carbotrap 20-40 and 1.5 mg of Tenax 60-80) in which the volatile organic compounds (VOCs) were trapped. Each collection lasted 5 min. Because Ficus hirta gs are small, to increase the quantity of odour trapped, we repeated the above operation three times for each bag. In parallel, for every collection we made a 'blank' extraction from a bag that contained no g, using the same protocol. One microlitre of a solution of internal standards (n-Nonane and n-Dodecane, 110 ng/µl of each) was added to each lter, before odour extraction, so that we could control for VOC loss during storage and transport, and estimate the total amount of VOCs emitted by gs. The samples were stored at -20 °C until VOC analysis. Ruhr, Germany). The instrumentation and temperature programs were as follows. First, the lters were desorbed splitless with a temperature of 250 °C on the CIS trap cooled at -80 °C by liquid nitrogen. Then, the CIS trap was heated to 250 °C with a 1:4 split ratio to inject the compounds in the column. Oven temperature was held at 40 °C for 3 minutes, increased from 40 °C to 210 °C at a rate of 5 °C/min and from 220 to 250 °C at 10 °C/min, and nally held for 2 min. The temperature of the transfer line and the ion source of the mass spectrometer were 250 °C and 200 °C respectively. The acquisition was from 38 m/z to 350 m/z, and the ionization energy is 70 eV. The FID was heated to 250 °C. The Xcalibur™ software (Thermo Scienti c™, Milan, Italy) was used for data processing. Retention times of a series of n-alkanes (Alcanes standard solution, 04070, Sigma Aldrich®) were used to convert retention times into retention index. VOCs were identi ed based on matching of the mass spectra with the NIST 98 MS and Adams 2007 libraries, and on con rmation by comparison of their retention index (RI) with libraries and published data [61]. Identi cation of some compounds was con rmed by comparison of both mass spectra and RI with those of authentic standards (see Table 2).

Data analysis
Only VOCs that appeared in at least two different odour samples were retained to determine odour pro les. From this VOC set, we calculated the emission rate and the relative composition of each odour pro le. The emission rates were the sum of emission rates of all VOCs detected in a given sample, calculated as ng/ g/hour. Relative odour composition was the relative contribution of each VOC to the odour pro le, expressed as a percentage.
All statistical analyses were performed with R version 3.5.1 [62]. Emission rate variation among locations were analyzed globally in an ANOVA and testing for deviations from mean value. Divergence in chemical pro les across locations was estimated with non-metric multidimensional scaling (NMDS) in two dimensions, based on a Bray-Curtis similarity matrix, using the package vegan [63]. We used the relative proportions of all the compounds emitted by gs (semiquantitative data). Data were standardized prior to the analysis. Two-dimensional plots were constructed using the "metaMDS" function algorithm.
Pairwise distance between individuals for relative proportions of VOCs was calculated using the Bray-Curtis dissimilarity index, which ranges between 0 and 1. A stress value is given, indicating how well the particular con guration represents the distance matrix (stress values < 0.2 are desirable). To test if the overall variation in chemical composition between groups was signi cantly different, we carried out permutational multivariate analysis of variance tests (PERMANOVA) based on a Bray-Curtis distance matrix. The chemical distance matrices were calculated with the function "vegdist" after data standardization with "decostand" function, and PERMANOVA were performed using the function adonis in the vegan package [63].We performed pairwise comparisons after detecting signi cant interactions with PERMANOVA with the "pair-wise.perm.manova" function in the RVAideMemoire package [64], and we used the false discovery rate method for multiple test p-value correction. Similarity percentage, simper [65], was used to identify the compounds that contributed most to dissimilarities among locations. The simper function performs pairwise comparisons of locations and nds the average contributions of each compound to the average overall Bray-Curtis dissimilarity. The function displays the compounds that contribute most to the differences between locations.
To investigate potential relationships between chemical distance and geographic distance, we performed Mantel tests. We used the chemical matrices generated above. The Mantel test requires that the matrices being tested have the same samples, so we calculated geographic distances using our GPS dataset that were represented in the oral dataset. Geographic distances were calculated using our GPS data. Mantel tests (with 99 999 random iterations) were performed for the entire data set and for data subsets. In a second step, a reduced data set was constituted with a single value per location by averaging across samples the mean peak area of each compound. The mean peak area of each compound for all the samples of a location then became the consensus sample used in all further analyses [47]. This method has been used as a drastic way to avoid the risk of pseudo-replication associated with using several data points from a single location as independent points [47]. was under the direction of the co-authors belonging to the staff of these gardens. All the other sampling sites were not privately owned or protected, and eld sampling did not involve protected species. Therefore, sampling was not subject to authorisation.

Consent of publication
Not applicable.

Availability of data and material
All data generated or analysed during this study are included in this published article.

Competing interests
The authors declare that they have no competing interests. Author's contributions XD collected samples, analysed the data and wrote the manuscript; BB analysed the data; YQP and YC helped collecting samples; HY, KF and MP organized the work and wrote manuscript; all authors have read and approved the manuscript