Do we truly understand pollination syndromes in Petunia as much as we suppose?

Abstract Petunia is endemic to South America grasslands; member of this genus exhibit variation in flower colour and shape, attracting bees, hawkmoths or hummingbirds. This group of plants is thus an excellent model system for evolutionary studies of diversification associated with pollinator shifts. Our aims were to identify the legitimate pollinator of Petunia secreta, a rare and endemic species, and to assess the importance of floral traits in pollinator attraction in this Petunia species. To determine the legitimate pollinator, field observations were conducted, and all floral visitors were recorded and evaluated. We also measured the nectar volume and sugar concentration. To characterize morphological cues for pollinators, we assessed the ultraviolet (UV)-light response in detached flowers, and characterized the floral pigments and pollen volatile scents for four different Petunia species that present different pollination syndromes. Petunia secreta shares the most recent ancestor with a white hawkmoth-pollinated species, P. axillaris, but presents flavonols and anthocyanin pigments responsible for the pink corolla colour and UV-light responses that are common to bee-pollinated Petunia species. Our study showed that a solitary bee in the genus Pseudagapostemon was the most frequent pollinator of P. secreta, and these bees collect only pollen as a reward. Despite being mainly bee-pollinated, different functional groups of pollinators visit P. secreta. Nectar volume, sugar concentration per flower, morphology and components of pollen scent would appear to be attractive to several different pollinator groups. Notably, the corolla includes a narrow tube with nectar at its base that cannot be reached by Pseudagapostemon, and flowers of P. secreta appear to follow an evolutionary transition, with traits attractive to several functional groups of pollinators. Additionally, the present study shows that differences in the volatiles of pollen scent are relevant for plant mutualistic and antagonist interactions in Petunia species and that pollen scent profile plays a key role in characterizing pollination syndromes.


Introduction
Pollination syndromes were first defined by Federico Delpino (Fenster et al. 2004) as suites of floral traits associated with particular pollinator groups (Etcheverry and Alemán 2005). Animal pollinators have acted as drivers of floral diversification and plant speciation (van der Niet and Johnson 2012;van der Niet et al. 2014), and the pollinators that most frequently visit and efficiently pollinate the flowers select for a number of floral traits in the long term (Armbruster 2014). In this sense, it is expected that similar suites of floral traits can reflect convergent adaptation to a particular type of pollinator in distantly related taxa (Proctor et al. 1996;Fenster et al. 2004).
Pollinators are agents of directional selection on interlinked floral traits such as colour and volatiles (Yan et al. 2016). Greater effectiveness in insect landing is promoted when visual cues and olfactory signals from the pollen are combined (Lunau 1992). Floral volatiles play an important role in attracting pollinators (Raguso 2001;; scent stimuli are learned more quickly than visual clues in bees (Arenas and Farina 2014) and may differentially attract certain pollinator species (Huber et al. 2005;Klahre et al. 2011).
Petunia is a young lineage in the Solanaceae family; it comprises species from subtropical and temperate South America. These species display flowers with different colours and shapes (Stehmann et al. 2009) and have bees, hawkmoths or hummingbirds as pollinators (Gübitz et al. 2009;Knapp 2010). The genetic architecture of floral syndromes has been studied in the Petunia, revealing genes associated with floral traits, especially volatile emissions and ultraviolet (UV)-light reflectance (Amrad et al. 2016;Sheehan et al. 2016). These features make Petunia a suitable model system for investigating pollinator-driven divergence and a good example of how key floral traits that affect pollinator behaviour can lead to reproductive isolation and adaptation (Gübitz et al. 2009;Fregonezi et al. 2013;Vandenbussche et al. 2016).
When plant species that do not present intrinsic (postpollination) reproductive barriers occur sympatrically, they usually exhibit different floral signals attracting different pollinator species (Huber et al. 2005). Previous studies have suggested that selection for different pollinators is an important force driving floral diversification in Petunia (Fregonezi et al. 2013). In addition, the most recent molecular phylogeny of the genus (Reck-Kortmann et al. 2014) supports two main clades mainly related to differences in the corolla tube length. The first clade includes 11 bee-pollinated species presenting short corolla tubes, pink flowers and blue pollen. The second clade includes three species with long corolla tubes and yellow pollen (Petunia axillaris, P. exserta and P. secreta) with remarkably diverse pollination syndromes and corolla colours. Basal to this long-tube clade arises Petunia occidentalis, which displays traits of typical species included in the first clade.
The long-tube Petunia species exhibit diverse flower morphologies and pollinators. Plants of P. axillaris have white flowers, produce floral scents at night and are moth-pollinated (Galliot et al. 2006;Venail et al. 2010); P. exserta has red flowers, with anthers and stigmas conspicuously exerted from the corolla, and pollination by hummingbirds (Lorenz-Lemke et al. 2006;Stehmann et al. 2009). Petunia secreta has pink flowers, and bees have been suggested as the probable pollinators, based on the flower morphology and some informal observations (Stehmann and Semir 2005).
Several studies made under garden conditions have identified the floral traits that attract pollinators in a few Petunia species. Using molecular tools and comparisons of pollinator behaviour and preferences, these studies showed that in P. axillaris, P. inflata, P. integrifolia and P. exserta the flower morphology, scent emission, nectar composition and UV-light reflectance are involved in the specialization to different pollinators and consequently in species diversification (Hoballah et al. 2007;Venail et al. 2010;Hermann and Kuhlemeier 2011;Klahre et al. 2011;Sheehan et al. 2012Sheehan et al. , 2016Dell'Olivo and Kuhlemeier 2013;Gleiser et al. 2014;Hermann et al. 2015).
Despite these previous studies, the literature on reproductive biology, pollinator attraction and evolutionary aspects of plant animal interactions remains scarce for the majority of Petunia species in natural conditions. For example, although floral scent appears to play an important role in reproductive isolation in Petunia (Verdonk et al. 2005;Hoballah et al. 2007;Klahre et al. 2011;Kessler et al. 2013;Amrad et al. 2016), information on natural populations is scarce regarding scent chemistry across the genus. Data on the pollen scent composition of Petunia are still unavailable, and could further elucidate the plant-pollinator interactions for Petunia species.
Petunia secreta is an interesting species because it belongs to the clade that presents the greatest floral variation in the genus, probably driven by pollinators (Fregonezi et al. 2013). This species, with its pink and nonfragrant corolla (Stehmann and Semir 2005), diverged recently from the large and white-flowered P. axillaris (Reck-Kortmann et al. 2014). Though not being found in exactly the same sites, P. secreta and P. axillaris generally occur in the same geographical region (Turchetto et al. 2015a;Rodrigues et al. 2018).
In this study, we evaluated the floral biology of P. secreta by recording its floral traits (nectar, petal colour and pollen scent); we also made field observations for flower visitors and legitimate pollinators, as well as recorded their foraging behaviours. In addition, we compared the pollen scent and floral features among species showing variation associated with pollination syndromes: P. secreta (possibly bee-adapted), P. axillaris (moth-adapted), P. exserta (hummingbird-adapted) and P. integrifolia (bee-adapted). We wished to establish whether there is any relationship between these traits and the known or predicted pollinators for these species.
We have two main aims in this study: (i) a detailed study of pollination of one species, P. secreta, including description of pollinator-attraction traits, and (ii) a comparative study of pollen scents in four Petunia species with different pollination syndromes. Our questions were as follows: (i) What is the legitimate pollinator of P. secreta? (ii) Does P. secreta offer rewards to its pollinators? (iii) How can its floral attributes affect pollinator attraction? (iv) What is the chemical composition of the pollen volatiles in different Petunia species that present different pollen colour and different pollinators? (v) Can pollen fragrance profiles and nectar provide information for predicting pollinators in P. secreta and be useful to understand diversification in the genus?

Studied species and area
Petunia secreta is annual and blooms from September to December (spring in the South Hemisphere) similarly to other Petunia species. The P. secreta corolla consists of a long tube that flares into a trumpet. It is pink, and the anthers are yellow (Stehmann and Semir 2005). This species is endemic to a low-elevation mountain range in a region known as Serra do Sudeste (Stehmann et al. 2009) and is an endangered species according to IUCN criteria. Two lineages associated with different environments were found by Turchetto et al. (2016). In the Serra do Sudeste, P. secreta co-occurs with P. axillaris, P. exserta and P. integrifolia although each species inhabits different sites.
In the greenhouse, P. secreta flowers remain open for 4 days if not pollinated, and flower senescence is characterized by changes in the colour of the corolla, followed by the gradual wilting of the petals. Anther dehiscence occurs simultaneously with the opening of the flower (within ~30 min), and the anthers are always positioned below the stigma. Petunia secreta is self-compatible, but it cannot spontaneously selfpollinate (Rodrigues et al. 2018).
We carried out the experiments in Caçapava do Sul municipality in the central region of the Rio Grande do Sul Brazilian state (Fig. 1A), ca. 350 m in elevation. During the spring of 2 years (September to December 2014 and 2015), we visited the region to observe pollinators. To minimize the impact on the natural populations, seeds were collected and germinated in growth chambers to obtain plants that were then cultivated in a greenhouse following the protocols of Rodrigues et al. (2018). Various traits were investigated in the cultivated plants, such as nectar (volume and sugar concentration), stigmatic receptivity, flower UV-light response and floral pigments [see Supporting Information- Table S1].

Nectar traits and stigmatic receptivity
We measured the volume and sugar concentration of nectar from five flowers from each of four individuals. The flower buds were bagged, and nectar volumes were extracted 24 h after opening of the flower with a graded 25 µL volume Hamilton microsyringe (Sigma-Aldrich Co., St. Louis, MO, USA). The sugar concentration was measured with a portable refractometer. Stigma receptivity was tested in five individuals by plunging the stigmatic surface into 100 % hydrogen peroxide P.A. (Merck & Co., Kenilworth, NJ, USA) at 100 %. The tests were performed in four stages using 10 flowers per stage: pre-anthesis floral buds; flowers after anthesis immediately after the opening of the anthers; flowers in which the corolla colour was starting to change (pink to purple, indicating the early flower senescence stage); and flowers with wilted petals. A positive result was observed when oxygen bubbles resulting from stigma-hydrogen peroxide reactions were produced (according to Zeisler 1938).

UV-light response
For the UV-light response experiments, we used flowers from greenhouse-grown plants of P. secreta, P. axillaris, P. exserta and P. inflata. These species represent different floral morphologies and all pollination syndromes described in Petunia. We obtained images of detached flowers with UV light using a Nikon 60 mm 2.8D microlens and a Nikon D7000 SLR camera (Nikon Co., Tokyo, Japan) that was converted to record UV light by replacing the manufacturer's filter with a UV-specific filter that blocked visible and infrared light (Advanced Camera Services Ltd, Watton, UK). A Metz MZ76 flashgun (Metz-Werke GmbH & Co. KG, St. Chandler, AZ, USA) that was modified to produce UV-A light (320-390 nm; Advanced Camera Services Ltd) provided the light source. Images were converted to greyscale in Photoshop CS5 (Adobe Systems Co., San Jose, CA, USA) and, when necessary, exposure was adjusted over the complete image. Flowers were scored either as UV-absorbent or UV-reflective based on comparison with the UV-absorbent positive control, a P. axillaris flower (Sheehan et al. 2016).

Spectrophotometric quantification of flavonols and anthocyanins
We used petals of cultivated individuals of P. secreta, P. axillaris, P. exserta and P. integrifolia growing under the same conditions to quantify the flavonols and anthocyanin floral pigments. For each species, we sampled discs from the corolla limb (8 mm in diameter) of one flower from each of three different individuals, put each disc into 1 mL of extraction buffer (2:1:7 methanol:acetic acid:water) and kept the solution in the dark for 48 h (modified from Ando et al. 1999). A spectrophotometer SpectraMax M4 (Carl Zeiss AG, Oberkochen, Germany) was used to measure the absorption spectra. Flavonols levels are detected at 300-385 nm (Tsimogiannis et al. 2007), whereas anthocyanins are detected at 400-600 nm (Merzlyak et al. 2008).

Pollen scent composition
Plants used for scent collection were selected at random from populations in the field. We collected anthers of P. secreta, P. axillaris, P. exserta and P. integrifolia. For each species, anthers of 10 flowers of different individuals were gathered in the same flowering season (November 2015) in sealed tubes 1 h after anthesis (~11:00 AM).
Volatile compounds from the pollen were determined out using the headspace solid-phase microextraction method (HS-SPME; Supelco Inc., Sigma-Aldrich) and gas chromatography-mass spectrometry (GC/MSD). The peak area of each compound was used for quantification. A 100-µm polydimethylsiloxane divinylbenzene (PDMS/DVB) fibre was used. After 5 min of sample conditioning, the SPME fibre was exposed in the headspace for 30 min, and then immediately inserted into the GC-MS injector port at 250 °C for 5 min. The MSD data were used for compound identification based on comparison of the mass spectra with those from the National Institute of Standards and Technology (https:// www.nist.gov) and their retention indexes with the published data.
Volatile components were first classified into different chemical categories: phenylpropanoids, benzenoids, mono-and sesquiterpenes, nitrogen-containing compounds and aliphatic alcohols , which allowed better visualization of the variation in molecular compounds. Then, we performed a similarity analysis of the volatile organic compounds (VOCs) based both on a presence/absence matrix and on the quantitative measures. We conducted a principal component analysis (PCA) using the prcom function of the package stats in R v.3.5.0.

Observation of floral visitors
We observed floral visitors to flowers of P. secreta in the field at Pedra do Segredo (Fig. 1A). We selected this site because the population comprises several individuals with several flowers per individual each season, which is not always the case in P. secreta (Turchetto et al. 2016;Rodrigues et al. 2018). Initially, we patrolled the population and gathered information about all possible visitors from 08:00 AM to 07:00 PM for 2 days. As visitors were not observed after 6:00 PM and P. secreta reflects UV light (see Results), we restricted the observations for this study to daytime only. Subsequently, diurnal visitors were recorded, and the observations took place over 24 days (12 days in 2014; 12 days in 2015) of which eight were cloudy or windy days and only 2 days were partially rainy from 08:00 AM to 06:00 PM without interruption, and for 35 different flowers overall. It is important to note that during the spring in this region, the sun rises at ~6:00 AM, but due to the landscape and vegetation, it touches the P. secreta plants at this site only after 8:00 AM and remains until 6:00 PM.
The behaviour of visitors was directly observed, and photographs and videos were taken using a Nikon D3200 SLR camera with a Nikon DX AF-S Nikkor 18-55 mm microlens (Nikon Co.) positioned 3 m from the flowers to reduce any interference due to observer presence. We recorded the number and taxonomic group of visitors and their behaviour during the visit, the landing site on the flower, contact with pollen, the position of pollen grains on the pollinator body, the ability to touch the stigma, visit duration, floral resource type collected and number of visited flowers.
We recorded the total number of visitors per individual flower and the number of flowers visited by each kind of visitor. The frequency of visits was analysed by dividing the number of visits made by each visitor by the total number of visits or pollinations during all observations per year.
We classified animals as visitors or pollinators based on their behaviour and likelihood to conduct effective pollination. We considered as legitimate pollinators of P. secreta only those floral visitors that had contacted the anthers and stigma for long enough to transfer the pollen. Floral visitors were divided into four functional groups according to Fenster et al. (2004), namely, longtongued bees, short-tongued bees, hummingbirds and hawkmoths. Insects considered as potential pollinators were collected and preserved in 70 % ethanol for taxonomic identification and deposited in the Science and Technology Museum, Pontifical Catholic University of Rio Grande do Sul (Porto Alegre, Rio Grande do Sul, Brazil).

Floral biology
Petunia secreta bloomed from September to December in both years. In the field, flower opening occurred only during the daytime; the flowers remained open for ~2 days if pollen contacted the stigma, but became senescent after 4 days in the absence of pollination. Anther dehiscence took place simultaneously with flower opening.

Floral traits
In P. secreta, the nectar was secreted at the base of the corolla (Fig. 2D), the total volume ranged from 4 to 20 μL per flower (mean 8 μL) and the total sugar concentration varied from 16 to 26 % per flower (mean 21.5 %) across the 20 flowers measured [see Supporting Information- Table S2]. The stigma receptivity tests revealed that the stigma surface was active during all stages, suggesting that the P. secreta stigma is receptive before flower opening until withering of the petals.
Flowers of the four Petunia species differed in appearance under visible ( Fig. 2A) and UV light (Fig. 2B). UV light revealed that, as expected for flowers pollinated by moths, P. axillaris petals absorbed UV light (dark colour); the petals of P. inflata and P. secreta reflected UV light (light colour), a trait that is associated with bee-pollinated flowers. The petals of bird-pollinated P. exserta also reflected UV light (light colour).
Petunia secreta, P. axillaris, P. exserta and P. integrifolia (Fig. 2C) showed the presence of flavonols within a 302-340 nm wavelength range, with P. axillaris and P. secreta exhibiting higher values of absorbance at 1.19 and 1.15, respectively, whereas P. exserta and P. integrifolia demonstrated values of 0.38 and 0.58, respectively. Anthocyanin peaks appeared in P. exserta, P. secreta and P. integrifolia within a 524-538 nm wavelength range; absorbance values were species-specific at 0.56, 0.36 and 0.10, respectively. Petunia axillaris did not present any peaks within the anthocyanin range.

Pollen aroma compounds
Gas chromatographic measurement from pollen scents revealed 63 biologically active compounds, mainly aliphatic compounds (35), benzenoids (16), cyclic compounds (5), terpenoids (4), organic compounds (1) and nitrogen compounds (2). Only three compounds (1-butanol, 3-methyl, 2-butanone, 3-hydroxy and phenylethyl alcohol) were found in all four species, indicating that the scent profiles differ among the four Petunia species [see Supporting Information- Table S3]. Petunia exserta presented the highest number of exclusive compounds (23), whereas P. secreta showed the lowest number of compounds (only one exclusive), and the greatest similarity in composition was observed between P. integrifolia and P. secreta. Petunia secreta and P. axillaris were quantitatively most similar, and the volatile emissions of P. exserta differed the most from those of the other species (Fig. 3).
Thirty-one compounds were detected in the pollen of P. axillaris with methylbenzoate, 2,3-butanediol, 3,7-dimethyldecane and phenylethyl alcohol being the major compounds. Petunia integrifolia pollen contained 12 compounds, with the most abundant being isobutyl phthalate, cis-caryophyllene, 1-butanol, 3-methyl and ethanol. The major constituents of the P. exserta pollen aroma were toluene, 2-pyrrolidinemethanol, 1-methyl 3-amino-5-tert-butylpyrazole and 1-octanol, with a further 34 compounds being detected. Finally, P. secreta presented 15 VOCs in pollen, with methylbenzoate, 2,3-butanediol, ethanol, 1-butanol, 3-methyl and isoeugenol being the most frequent pollen aroma compounds [see Supporting Information- Table S3]. Petunia secreta shared 14 of its pollen VOCs with at least one of the three other analysed species, of which 12 were previously reported to elicit positive behaviour in bees, four were reported to attract hawkmoths and two have not been evaluated to date (Table 1).

Pollinator and visitor observations
Visitors of three different functional groups were observed on flowers of P. secreta (Table 2) during the 225 h of observations [see Supporting Information- Table S4]. Bee species of two genera with short tongues, one genus of long-tongued bees and one unidentified hummingbird species were recorded foraging and visiting P. secreta. We counted 51 visits in total of which 39 (76 %) were by species of Halictidae; Pseudagapostemon sp. bees were the most frequent visitors. Most pollinator visits occurred during the afternoon from 1:00 PM to 5:00 PM, with the peak occurring between 1:00 PM and 3:00 PM. There was a low frequency of visits to P. secreta during the first few hours of the day [see Supporting Information- Fig. S1].
Bees belonging to the Pseudagapostemon genus (Fig. 1B) landed directly on the anthers and collected pollen exclusively. These individuals approached the flowers, flew away and approached again several times until they landed. We counted 38 visits, most of them occurring in 2015, in the same season in which we observed more flowers of P. secreta. Individuals of Pseudagapostemon sp. displayed a behaviour and body structure that fulfilled all our criteria for consideration as a legitimate pollinator of P. secreta. We observed 22 visits of Pseudagapostemon sp. individuals that landed directly on the reproductive structures (Fig. 1C), with the front legs scraping the anthers and transferring pollen to the scopa in the abdomen (dense set of hair or bristles specialized for pollen adherence) and to the hind tibia [see Supporting Information-Movie S1], always positioned on the flower with the abdomen and legs in front of the stigma. These individuals took, on average, <2 min to collect pollen.
Pseudagapostemon sp. individuals were observed on P. secreta flowers only when pollen was present; therefore, each flower received a maximum of two visits. In the presence of pollen, bees removed all pollen, and in its absence, bees did not land. Differences in anther height (Fig. 1D) influenced bee behaviour during pollen collection, making the insect stand in different directions and slip on the stigma surface to completely remove the pollen. There was no standard time of day for visits by Pseudagapostemon to P. secreta flowers; visits were spread from 10:30 AM to 6:00 PM. Visits occurred on sunny days, and none occurred on rainy, windy or cloudy days [see Supporting Information- Table S4].
Four male individuals of Lanthanomelissa clementis were seen only once at dusk, in a group using the flower as a dormitory, and remained inside the flower until the following morning. One individual of Xylocopa sp. that was seen on 1 day in October 2015 visited the flowers several times, cut out a piece of the corolla and took it away. The unidentified hummingbirds were observed Table 1. Chemical composition of pollen scent in Petunia secreta compared to the other three Petunia species. Colours reflect maximum emission (% in relation to total emission): blue (0%); green (<5%); yellow (>5% and <10%); orange (>10% and <20%); red (>20%). We considered a floral scent compound attractive when it was reported in the literature as eliciting positive bee and/or hawkmoth behaviour. secr -Petunia secreta; inte -P. integrifolia; axil -P. axillaris; exse -P. exserta; ✓-compound described as attractive for; NA -not available; References: 1 - Braunschmid et al., 2017;2 -Carril, 2014;3 -Cordeiro et al., 2017;4 -Dötterl and Vereecken, 2010;5 -Dudareva and Pichersky, 2006;6 -Filella et al., 2011;7 -Goodrich et al., 2006;8 -Haverkamp et al., 2016;9 -Hetherington-Rauth and Ramírez, 2016;10 -Hoballah et al., 2007;11 -Johnson et al., 2005;12 -Knudsen and Tollsten, 1993;13 -Knudsen et al., 2006;14 -Levin et al., 2001;15 -Pham-Delègue et al., 1992;16 -Raguso, 2004;17 -Raguso and Light, 2003;18 -Raguso et al., 1996;19 -Teichert et al., 2012;20 -Vega et al., 2009;21 -Wadhams et al., 1994;22 -Wiens et al., 2008;23 -Williams and Whitten, 1983;24 -Wright et al., 2008 once in October 2015 and four times in November 2015; on each visit, the bird introduced its bill into the flower on average for ~3 s, and restricted itself to one flower per visit. We were not able to take photos that would allow identification of the species of hummingbirds, and we were also not able to verify effective pollination by these birds. We think the birds are unlikely to transfer much pollen, which would be confined to the beak and not the feathers, and due to the flower morphology, which hides the anthers inside the tube. However, it is possible that the birds shake the flower and pollen could be deposited on the stigma, promoting self-pollination by a secondary pollinator.

Putative pollinator as suggested by morphological traits
We investigated the floral traits and the plant-pollinator interactions of P. secreta. Petunia secreta presents a set of floral features such as pink corolla, diluted nectar and volatiles in pollen that distinguish it from the other Petunia species. We found that P. secreta is mainly pollinated by Halictidae, a functional group of short-tongued bees. We cannot say that other insects or birds never promote pollination. The pollen scent and flower colour are likely adaptations that attract bees. However, the narrowness of the corolla tube and the nectar characteristics seem to be adaptations for other pollinators, possibly hummingbirds or some kind of Lepidoptera, although we rarely saw hummingbirds and never saw any Lepidoptera visiting P. secreta.
Interspecific differences in the UV-light floral response are found among the Petunia species that are indicative of their pollination syndromes. Petunia secreta has a pink corolla that reflects UV light, traits mainly present in bee-pollinated species (Papiorek et al. 2016) and appears to reflect the ancestral state of the Petunia genus (Reck-Kortmann et al. 2014), represented here by P. inflata. However, P. secreta does not have all the typical floral features known for the bee-pollination syndrome, such as unscented flowers, blue pollen, a low volume of nectar and a wide and short corolla tube. Conversely, P. secreta shares several traits with P. axillaris, such as the long and narrow corolla morphology that limits nectar access by large insects, similar amounts of flavonoids in flowers, moderate nectar resources and odour emitted from its yellow pollen comprising compounds that have previously been described as attractive to diverse insect species. Of note, P. secreta does not emit a floral scent at dusk, which is the main characteristic that attracts hawkmoths in P. axillaris (Venail et al. 2010;Klahre et al. 2011).
The P. secreta reflectance peak spectrum suggests that the corolla is attractive to bees. However, petals of P. secreta are also within the range of vision of hummingbirds, which perceive colour over wavelengths ranging from 300 to 600 nm (Cronk and Ojeda 2008). Nevertheless, the visual display differs by corolla colour and reflectance among the species analysed here, and suggests that P. inflata, P. secreta and P. exserta are visually more adapted to diurnal pollinators and P. axillaris to nocturnal pollinators. The contrast in floral colour suggests that P. secreta and P. axillaris are adapted to different pollinator assemblages, and the similarity in their pollen odours possibly reflects their shared evolutionary relationships (Reck-Kortmann et al. 2014). Thus, the pink colour could be associated with an increase in detection by the bees and with a decrease in detection by nocturnal moths (Venail et al. 2010).
Nectar volume varies among Petunia species. The average volume and sugar concentration of P. secreta nectar are lower than those observed in P. axillaris subsp. axillaris (Gleiser et al. 2014), but the volume is much higher than that reported for P. integrifolia (Gübitz et al. 2009). This result shows that P. secreta produces Table 2. Visitation and pollination frequency of different functional groups as observed in Petunia secreta per year. ND-not determined; % visitation corresponds to the frequency of views without pollen transfer to stigma; % pollination corresponds to the frequency visits with pollen transfer to stigma; -no views or pollen not transferred. a modest amount of nectar; despite this, the nectar does not act as a reward for the most frequent pollinator (Pseudagapostemon sp.) but is occasionally used by hummingbirds. The sugar concentration in nectar from P. secreta matches what has been proposed for bird-pollinated flowers (Baker 1975;Proctor et al. 1996). Birds that effectively introduce their bill into flowers can promote pollen transference when visiting flowers searching for available nectar (Maruyama et al. 2013). The characteristics of nectar from P. exserta flowers are not known, but they present several traits related to hummingbird pollination such as their bright red corollas (Gübitz et al. 2009), backward-folding petal limbs and reproductive structures exerted from the corolla, which improves contact with the bird's head and facilitates pollen transfer (Lorenz-Lemke et al. 2006;Stehmann et al. 2009). Petunia secreta lacks all these morphological traits of P. exserta.

Floral visitors
The colour and scent are equally important to hawkmoths in foraging decisions among flowers with different morphologies (Glover 2011). Flowers of P. axillaris support this observation (Venail et al. 2010;Klahre et al. 2011) despite possessing a nectar volume and sugar concentration close to the optimal amounts reported not only for hawkmoths (Gleiser et al. 2014) but also for several types of bees (Kim et al. 2011). The loss of flower odour and gain of visible colour in P. exserta is likely related to the greater trend towards bird pollination compared to P. axillaris (Kessler et al. 2013;Amrad et al. 2016), and we think this could also be the case in P. secreta. Some shared traits present in all long corolla tube species could correspond to shared ancestral features and may not be related to the most important pollinator.

The role of pollen scent in relation to pollinator affinities
There is little information on pollen scents for most plants. However, it is known that bees are able to discriminate between pollen odours in biologically realistic concentrations, which suggests that the pollen odour may attract these pollinators (Cook et al. 2005;Ruedenauer et al. 2015).
The VOCs described here from Petunia pollen are generally present in flower bouquets , and some of them are among the most frequently observed . Each Petunia species emits a characteristic mixture of volatiles with distinct compounds and different total amounts that are compatible with the pollination syndrome.
We found that plant-pollinator interactions in P. secreta cannot be interpreted as a bee-pollination syndrome based only on UV-light response and corolla colour, and pollen volatiles can have an important effect on the legitimate pollinator, Pseudagapostemum sp., since VOCs in this species are associated with responses by bee antennae in other angiosperms that allow the bees to detect pollen from a distance before landing (Dötterl et al. 2005;Dötterl and Vereecken 2010). Indeed, different sets of volatile pollen compounds are shared between P. secreta and P. integrifolia, both of which display a corolla colour and UV reflectance related to bee pollination, and some compounds are found at high levels that are unusual in floral aroma and are associated with bee attraction (Goodrich et al. 2006).
In P. axillaris, the most abundant compound in the pollen scent profile is one of the three most frequent endogenous VOCs (Negre et al. 2003), and several others associated with bee pollination are also present in similar proportions to those observed in P. secreta pollen. The similarity between these two species may be explained based on their evolutionary proximity and because diurnal secondary pollinators in P. axillaris were observed (Gübitz et al. 2009).
The pollen scent profile in P. exserta presents the highest number of exclusive compounds, many of which are related to defence against herbivores. Plants with exposed pollen like P. exserta produce anti-herbivore deterrents in pollen (Dafni et al. 2000) and have specific floral bouquets to deter florivores and nectar robbers and simultaneously attract pollinators and antagonists Kessler et al. 2015).

Legitimate pollinators of Petunia secreta
Based on general floral colour and shape, P. secreta was described as a bee-pollinated species (Stehmann and Semir 2005), and our findings support this assertion.
Pseudagapostemon sp. bees can be attracted to P. secreta by corolla colour and UV reflectance, but pollen volatiles also appear to play a role because bees land only on flowers with at least one intact anther. The pollen aroma may be involved in specific pollinator attraction at short distances. Pseudagapostemon sp. behaviour is consistent with the view that certain pollen compounds constitute a stimulus to bees landing. The critical nature of the floral scent in the foraging behaviour of host-specialized solitary bees has been demonstrated in honeybees that associate scent and pollen (Arenas and Farina 2012).
Although P. secreta flowers secrete nectar, bees cannot reach the bottom of the corolla tube to collect it. Field observations (data not shown) and previously published measurements (Stehmann and Semir 2005;Turchetto et al. 2016) indicate that the distance between the point at which the filaments fuse to the corolla and the deeper portion of the tube where the nectar accumulates is small (ca. 2 cm), and the tube along this region is slender (ca. 2-4 mm in diameter). These measurements suggest that it is impossible for individuals of Pseudagapostemon sp. to gather the nectar in contrast to Callonychium individuals that can obtain nectar from flowers of P. integrifolia (Wittmann et al. 1990). The body length of Pseudagapostemon sp. is ca. 5-11 mm (Michener 2007), but the length of the proboscis is not known; however, even if the tongue is as long as the body, it would still not cover the distance of 20 mm required to reach the nectar in P. secreta. Furthermore, Pseudagapostemon sp. individuals did not act as nectar collectors in any of the observed visits.
Bees visited the flowers, but each visit was restricted to a single flower, and the asynchronous mass flowering in P. secreta may reduce the frequency of geitonogamy since this species produces only one flower per time per individual and just a few individuals make up the plant patches at the studied site (Turchetto et al. 2016). However, Pseudogapostemon sp. appeared to be responsible for the most pollination events, and its pollen collection behaviour likely promotes a high frequency of self-pollination (D. M. Rodrigues et al., unpubl. data). Petunia secreta is self-compatible (Rodrigues et al. 2018) and shows a high genetic diversity compared to other congeneric species (Turchetto et al. 2016), probably due to secondary pollinators that promote cross-fertilization.
We found that short-tongued bees mainly pollinate P. secreta plants; however, the field observations and some floral cues do not allow us to rule out hummingbirds and other insects as occasional pollinators.

Do we truly understand pollination syndromes in
Petunia as much as we suppose?
In an evolutionary context, our findings suggest that the interaction with Pseudagapostemon sp. bees has minimized the nectar volume and concentration of P. secreta compared to other sympatric Petunia species, while maximizing pollen scent emission, corolla colour and UV-light reflectance to improve the attraction of shorttongued bees, all characteristics that are ancestral conditions of the genus. However, different functional groups of pollinators can play a role in the reproductive success of P. secreta. Oligolectic bees were observed most frequently pollinating P. secreta plants, but hummingbird pollination also seems to occur.
Evolutionary shifts from one pollination syndrome to another often involve particular flower colour transitions (Wessinger and Rausher 2012). The major determinant of flower colour variation between P. integrifolia and P. axillaris that has caused major shifts in pollination is the ANTHOCYANIN2 gene (Hoballah et al. 2007), with gene inactivation promoting the change in corolla colour from pink to white. Moreover, traits such as scent emission, flower architecture and rewards can be clustered and allow rapid switching between pollination syndromes in response to changes in pollinator availability (Hermann et al. 2013), as in P. axillaris and P. exserta.
Studies focusing on the traits involved in host finding by oligolectic bees concluded that visual and olfactory cues are used when bees search for food (Burger et al. 2010;Milet-Pinheiro et al. 2012;Carvalho et al. 2014). Petunia secreta may be visited by different functional groups, but the relative selective pressures that they exert may be different. The corolla colour, as well as the lack of corolla scent at dusk and a low sugar concentration in nectar, can prevent visits by nocturnal pollinators, especially hawkmoths. The pollen odour may attract some pollinators and possibly be inconspicuous to other insects. The absorbance spectrum in petals of P. secreta suggests the ability to attract bees; however, flowers of P. secreta can be easily detected by hummingbirds and it is known that species with non-red flowers are occasionally hummingbird-pollinated, especially at sites where their preferential pollinators are found at low densities (Cronk and Ojeda 2008).
Flower and pollinator features contribute to restricting pollination to individuals of the same species and enhance reproductive isolation in a variety of plant species (Scopece et al. 2014;Breitkopf et al. 2015). It can be argued that classifying flowers as belong to a single specialized pollination syndrome may mask the importance of 'secondary' or 'tertiary' pollinators as drivers of particular floral traits (Queiroz et al. 2015;Cronk and Yang 2016) especially in Petunia, because in this genus we can observe natural hybrids between P. axillaris and P. exserta Turchetto et al. 2015b) that have different pollinators: hawkmoths and bees visit and pollinate P. axillaris (Gübitz et al. 2009) and P. secreta presents different putative pollen vectors. All these species occur in the same geographical area.
Moreover, the validity of pollination syndromes has been widely questioned (Rosas-Guerrero et al. 2014;Gong et al. 2015). Sympatric species mainly depend on specific floral traits to establish relatively strict but not absolute pollinator specificity, and pollinator sharing could be rather common ) and would explain the interrelationships we observed among the Petunia species. When adaptation to a slightly effective pollinator requires minimum loss of fitness compared to a more effective pollinator, plant species may exhibit specialized traits for secondary pollinators (Aigner 2001).

Conclusions
The variation in colour, nectar and pollen scent of different co-occurring species of Petunia can provide information on the specific signals that guide pollinators and may contribute to reproductive isolation. This is a preliminary report on variation in pollen scent from different wild Petunia species and the first field observations of the pollination ecology of P. secreta. Combined, these data suggest that P. secreta exhibits a set of traits that enable these plants to be effectively pollinated by solitary bees despite preserving some shared traits with its cousins that are pollinated by other kinds of animals. Moreover, pollen scents may have evolved in conjunction with the sensory capabilities of different visitors rather than the specific group of pollinators seen to visit representative Petunia species with documented pollination syndromes. A number of these characteristics, especially the colour of petals and abundant amounts of some compounds in pollen, represent a reversion to the ancestral condition in the Petunia genus since they are shared with other bee-pollinated species. Despite this, we cannot rule out the possibility that other floral traits (pink long tubular flowers, and nectar volume and sugar concentration) could attract other functional groups of pollinators (probably hummingbirds) and could constitute an evolutionary shift in the pollination system in progress.
The collection, isolation, identification and bioassay of the pollinator attractants from Petunia deserve further attention in order to investigate the potential interaction between olfactory and other signals in Petunia species and detect which are the most important compounds in mutualistic interactions. Additionally, olfactory experiments are required to test how the bee's behaviour differs between unique blends and the overall quantity of volatile emissions.

Supporting Information
The following additional information is available in the online version of this article- Table S1. Biological sources for different analyses and comparisons. Table S2. Nectar volume and sugar concentration. Table S3. List of pollen volatile organic compounds in four Petunia species. Table S4. Visitor and pollinator records per year and flower. Figure S1.