Broader phenology of pollinator activity and higher plant reproductive success in an urban habitat compared to a rural one

Abstract Urban habitat characteristics create environmental filtering of pollinator communities. They also impact pollinating insect phenology through the presence of an urban heat island and the year‐round availability of floral resources provided by ornamental plants. Here, we monitored the phenology and composition of pollinating insect communities visiting replicates of an experimental plant assemblage comprising two species, with contrasting floral traits: Sinapis alba and Lotus corniculatus, whose flowering periods were artificially extended. Plant assemblage replicates were set up over two consecutive years in two different habitats: rural and densely urbanized, within the same biogeographical region (Ile‐de‐France region, France). The phenology of pollination activity, recorded from the beginning (early March) to the end (early November) of the season, differed between these two habitats. Several pollinator morphogroups (small wild bees, bumblebees, honeybees) were significantly more active on our plant sets in the urban habitat compared to the rural one, especially in early spring and autumn. This resulted in different overall reproductive success of the plant assemblage between the two habitats. Over the course of the season, reproductive success of S. alba was always significantly higher in the urban habitat, while reproductive success of L. corniculatus was significantly higher in the urban habitat only during early flowering. These findings suggest different phenological adaptations to the urban habitat for different groups of pollinators. Overall, results indicate that the broadened activity period of pollinating insects recorded in the urban environment could enhance the pollination function and the reproductive success of plant communities in cities.

Urban habitat characteristics can also impact the phenology of plant-pollinator interactions. Concerning plants, the warmer urban climate (through the presence of an urban heat island, hereafter UHI) may either advance or delay (Jochner & Menzel, 2015;Neil, Landrum, & Wu, 2010) plant flowering phenology. Moreover, the year-round presence of ornamental plants in urban green spaces may extend the availability of floral resources for pollinating insects (Tasker, Reid, Young, Threlfall, & Latty, 2020). Individual species may display various phenological responses, ultimately causing shifts in potential interaction partners and transforming the mutualistic networks (Harrison & Winfree, 2015). Concerning pollinators, the UHI should enable them to be more active throughout the season than in rural habitats. Indeed, some recent studies report a broadening of the flight period of pollinators in the city, whereas pollinator activity tends to peak earlier in spring in seminatural habitats (Harrison, Gibbs, & Winfree, 2018;Leong, Ponisio, Kremen, Thorp, & Roderick, 2016;Luder, Knop, & Menz, 2018;Wray & Elle, 2015).
This extended period of activity may also be supported by the above-mentioned year-round availability of floral resources in cities.
Taken together, these plant and insect phenological changes should strongly impact the pollination function in urban habitats. However, there is concern that plants and pollinators might have different responses to warming, potentially leading to loss of phenological synchrony that would disrupt the pollination networks (Forrest, 2015;Memmott, Craze, Waser, & Price, 2007), although this appears to be dependent on the biodiversity level (Bartomeus et al., 2013). It has been theorized that the local adaptive responses of plant-pollinator networks to UHI effect could be considered as a small-scale model for the larger-scale consequences of global warming (Jochner & Menzel, 2015).
Here, we set up an all-season monitoring of the pollination activity, pollinator assemblage composition, and the resulting pollination function, in an urban-rural paired experimental design encompassing a dense urban habitat (the city of Paris, France) and rural habitats located within the same region (Ile-de-France region, France). In order to standardize our monitoring from the beginning of spring to mid-autumn, and also to simulate potential climate change-induced modifications in the flowering phenology of plants, we used temporal transplants of an experimental plant assemblage (Morton & Rafferty, 2017), comprising two insect-pollinated plant species native to this region. In other words, we brought plants to bloom in and out of their natural flowering period. These plant assemblages, whose flowering phenology was either "advanced" or "delayed" (in contrast to "natural"), thus played the role of plants with shifted phenological patterns. We aimed to investigate whether these out-ofseason floral resources would find matching pollinators in the dense urban and the rural habitats investigated, and what consequences it would have on the reproductive success of the plants. Hence, these controlled plant sets can be considered as "pollinometers" (Theodorou et al., 2017), as measuring their reproductive success could be a proxy to assess the efficiency of the pollination function throughout the season between urban and rural habitats within the same region.
Our hypothesis is that, in the city, pollinator activity would show different phenological patterns that in a rural habitat. This would lead to differences in the efficiency of the pollination function and contrasting plant reproductive success over the time between these two habitats. More precisely, we expect a broadening of the pollinator flight season in the urban habitat, thus leading to more efficient early and/or late pollination, and higher overall plant reproductive success in urban habitats compared to rural ones.
To our knowledge, this is one of the few studies (Rafferty, Caradonna, Burkle, Iler, & Bronstein, 2013) that have associated an all-season monitoring of pollinator activity to the evaluation of the pollination function, through the assessment of plant reproductive success in an urban-rural paired design.

| Experimental sites
Experiments were conducted over two consecutive years in four (2018) and six (2017) locations in grasslands located in dense urban habitat and forest-dominated seminatural habitat-hereafter referred to as "rural." All sites were located in the same biogeographical region: the Ile-de France region that encompasses a large diversity of habitats, from the city of Paris (largest city in France) to seminatural and rural habitats (INSEE, 2015).
In 2017, urban experimental sites were located in downtown  Figure 1). On the other hand, rural sites were set up in grasslands mostly surrounded by forests. These grasslands are not harvested and do not receive any chemical inputs. The SEF site is part of a forest biosphere reserve, while the two CEREEP sites are located in a large experimental ecology field station encompassing seminatural forests and grasslands. For this reason, all these sites can be considered as "seminatural," despite their potential proximity to discontinuous suburban areas.

| Experimental setting
In each experimental site, two 1.6 × 1.2 m plots were set up side by side in a grassland area, each containing one of the two focal plant species (the Brassicaceae Sinapis alba and the Fabaceae Lotus corniculatus). Sinapis alba L. is an annual forb that grows along roads, in wastelands or near crops, and is considered naturalized in the Ile-de-France region (Lombard, 2001). It is an obligate outcrossing species (Cheng, Williams, & Zhang, 2012), the fruits of which are siliques containing up to eight seeds (Jauzein & Nawrot, 2011). On the other hand, Lotus corniculatus L. is a perennial, nitrogen-fixing plant widespread in grasslands and disturbed habitats (Jones & Turkington, 1986). Native to the Ile-de-France region (CBNBP, 2016;Jauzein & Nawrot, 2011), this strictly entomophilous species (Pellissier, Muratet, Verfaillie, & Machon, 2012;Stephenson, 1984) bears cylindrical pods containing up to 30 seeds. No spontaneous L. corniculatus or S. alba conspecifics were found in a 100 m radius around either urban or rural site.
Although they both bear yellow flowers, these two plant species were chosen for their contrasting floral morphologies, in order to attract a diverse range of pollinators: S. alba has flat corollas with floral rewards accessible to pollinators with short mouthparts (Fontaine, Dajoz, Meriguet, & Loreau, 2006;Geslin, Gauzens, Thébault, & Dajoz, 2013;Jones & Turkington, 1986) In each plot, 20 pots containing one plant of the same species were buried in four rows of five, each plant being spaced from others by 25 cm in all directions. We kept plants in their plastic pots to prevent competition for soil resources. A plastic tag was planted in each pot to individually number each plant, and all plants were watered regularly. The plots were regularly weeded to avoid interference from spontaneous plants.
Since the objective was to maintain a regular floral cover throughout the study period, and since the full flowering stage of both plant species did not exceed three weeks (V. Zaninotto, pers. obs.), the plants were renewed regularly in each plot, on the same day for all experimental sites. For both species, blooming plants were exposed to pollinators for about 20 days, before being  (Jauzein & Nawrot, 2011), some of the flowering rounds were set before, and after, that period. Plants that were artificially brought to bloom during these rounds can be described as temporal transplants. Three phases were thus defined in the experiment (see Table 2 for S. alba, September to November for L. corniculatus) flowering periods.

| Monitoring
During these periods, the plant-pollinator interactions on these plant plots were regularly monitored. Twice a week, all locations were monitored on the same day, in alternating order.

| Fruit set and seed set measurement
At the end of each 20-day floral round, and in all experimental sites, five S. alba and three L. corniculatus plants were randomly selected to estimate fruit set, while two plants of each species were selected to estimate seed set. Control plants were also grown in order to estimate the selfing rate and resulting fruit set and seed set of the two focal species. In 2017, one control plant per species and per round was kept in the greenhouse during its entire flowering period. In 2018, one control plant per species, per locality, and per round was set up in the field in an insect-proof mesh cage and then brought back to the greenhouse at the end of each round. This allowed to determine a selfing rate and resulting fruit set and seed set of the two species under greenhouse conditions and under natural conditions. After each round, both plants exposed to insect visitation and control plants were kept in the insect-proof greenhouse for an additional two weeks to allow for fruit development.
We used different methods of estimating fruit set for each plant species. For S. alba, all flower peduncles are still visible after the flowers have wilted, even in the absence of fruit. Therefore, on each insect-exposed plant and each control plant, we were able to count the number of fruits produced by 10 contiguous flowers on the stem from a random starting position. For L. corniculatus, fruit set was not assessed in 2017 as the fruit set estimator used at the time was not appropriate for this plant species. In 2018, it was estimated by counting all fruits produced on each plant. For this same species, to take into account size differences among plants (and thus size-related differences in floral display), plant aboveground biomass was collected individually, dried in an oven for 48 hr at 60°C, and weighted (scale precision: 1 × 10 −4 g). For both species, seed set was estimated by counting the number of seeds contained in three fruits (when present), randomly picked on each selected insect-exposed plant and each control plant.

| Data analysis
All data analysis was performed using R software (R Core Team, 2019, version 3.6.1). First, visitation rates were analyzed by constructing generalized mixed effect models with the "glmmTMB" function ("glmmTMB" package, Brooks et al., 2017), which deals well with zero-inflated data. The response variable was "Visitation rate," defined as the number of pollinator visits per plant and per 5-min observation session, with a negative binomial distribution to account for overdispersion. Fixed effects were the habitat ("rural" or "urban"), the flowering period ("advanced", "normal", and "delayed"), and their interaction, as well as the flower display size of the plant, cloud cover, the relative temperature (measured temperature relative to expected seasonal temperatures), and the year. The experimental site was included as a random effect. This model was replicated for the different morphological groups of pollinators and both plant species.
We also built generalized mixed effect models to analyze fruit set estimators. For S. alba, the response variable was the proportion of flowers that gave fruits, with a binomial distribution. Fixed effects were the habitat, the flowering period ("advanced", "normal", and "delayed") and their interaction, as well as the year; the experimental site was again included as a random effect. For L. corniculatus, the response variable was the total number of fruits on the plant, with a Poisson distribution. Fixed effects were the habitat, the flowering period, and their interaction, as well as the aboveground dry mass of the plant, with the experimental site as a random effect. The same types of models were used for both plant species to analyze seed set estimators, with the number of seeds per fruit as the response variable following a Poisson distribution.
For all models, we evaluated the contribution of each factor to the model via type III Wald chi-square tests ("ANOVA" function in "car" package, Fox & Weisberg, 2019) and performed model selection based on the AIC ("step" function, "backward" method). We also verified the absence of multicollinearity between the predictors ("check_collinearity" function from "performance" package, Lüdecke, Makowski, & Waggoner, 2020). We compared visitation rates, fruit set, and seed set between habitats within flowering periods through post hoc Tukey's tests with the "emmeans" and "contrast" functions ("emmeans" package, Lenth, 2020).
In addition, nonparametric Wilcoxon tests were carried out to compare fruit set estimators of control plants kept under insect-proof conditions with plants exposed to pollinators, in order to estimate the rates of self-fertilization of the two plant species. Eventually, the control plants of both species did not produce enough fruits to be able to estimate their seed set.

| Pollinator visit frequencies
In 2017 As we could expect, our models reveal that pollinator visitation rates to a plant are strongly associated with the floral display of that plant, but also with the relative temperature at the time of observation. On both plant species, we also observed differences between the two habitats that varied throughout the season, as evidenced by the terms "Habitat," "Flowering Period," and their interaction (Table 3).
Overall, pollinator visits on S. alba were not restricted to the nat- Therefore, there was no significant difference between the overall visitation rates in the two habitats (t = −1.63, df = 16,830, p = .10).
Finally, during the delayed flowering period, since there was a surge Note: For each term, chi-square and p-value of the type III Wald chi-square tests are presented, as well as the estimates of each coefficient (±SE) ("urb." = urban; "adv." = advanced; "del." = delayed; rural habitat, normal flowering period, and year 2017 were taken as references).
in domestic honeybees' visits in the urban habitat (Figures 3 and   4d)-while syrphid fly visitation rates increased substantially in both habitats (Table 3; Figure 4c)-overall visitation rates again became significantly higher (t = −5.07, df = 16,830, p < .0001) in the urban habitat than in the rural one.
Pollinator visits on L. corniculatus were more restricted by the natural flowering period of the plant (Table 3; (Figure 5d).

| Plant reproductive success
In the urban habitat, fruit set rate of S. alba remained elevated for the three experimental phases and was not restricted to the natural flowering time (Table 4; Figure 6a). In particular, the fruit set rate during the advanced period was already as high as than during delayed period, 68 ± 6.0%). In contrast, rural fruit set rates were always significantly lower (Figure 6a; df = 352, advanced period: t = −10.2, p < .0001; natural period: t = −6.81, p < .0001; delayed period, t = −4.22, p < .0001), but they seemed to slowly increase throughout the season (advanced period, 32 ± 4.0%; natural period, 37 ± 5.0%; delayed period, 49 ± 7.0%). Last, mean fruit set rates of the control plants were significantly lower than mean fruit set rate of plants exposed to pollinators (6.0 ± 2.0% and 3.0 ± 1.0% of fruit set for 2017 and 2018 controls, respectively; no significant difference between these two values, whereas mean overall fruit set of plants exposed to pollinators in both years was 55 ± 2.0% SE, W = 1702 p = 1.4e−13). This indicates that differences in fruit set between habitats are not due to higher selfing rates because of differences in other biotic or abiotic characteristics of the environment. Not only was the fruit set rate of S. alba higher in the urban habitat, but the fruits also contained more seeds during each of the periods studied (Figure 6b; df = 259, for each period: t = −2.56, p = .011). This resulted in an overall higher reproductive success of the plant in the urban habitat than in the rural habitat.

F I G U R E 4 Predicted pollinator visitation rates (number of visits per 5-min session) on
For L. corniculatus, fruit set showed the same dynamics as pollinator visit frequencies and was consequently higher during the natural flowering period in both habitats (Table 4; Figure 6c). However, this fruit set was significantly higher in the urban than in the rural habitat during the advanced-flowering period (t = −4.07, df = 68, p = .0001), though well below the value during the natural flowering time. Last, mean fruit set rates of the two types of control plants were significantly lower than mean fruit set rate of plants exposed to pollinators (no fruit has ever been observed on 2017 or 2018 controls, whereas mean overall fruit set of plants exposed to pollinators in both years was 3.11 ± 0.35 fruits per gram of dry aboveground biomass, W = 260 p = 1.5e−14). This also strongly suggests that differences in fruit set between habitats are not due to higher selfing rates because of differences in other biotic or abiotic characteristics of the environment. As for seed set rates, there seemed to be no difference between the two habitats throughout the season ( Figure 6d). In contrast, L. corniculatus visits were predominantly carried out by bumblebees, whose visitation rates were more limited to the natural flowering period of the plant. As a result, fruit set rates, and by extension reproductive success of the plant, were more restricted to this period than in S. alba. This can be associated with a higher degree of specialization in the pollination ecology of L. corniculatus, with deep and hard-to-reach floral resources that are nevertheless accessible to bumblebees (Figure 2).
Since these pollinators seem to be determinant to the pollination of L. corniculatus (Fontaine et al., 2006;Jones & Turkington, 1986), the synchronization between the natural flowering period of this plant and the activity of the bumblebees was expected. Before that, during the advanced blossom, we registered greater bumblebee activity in urban sites, though visits remained scarce in both habitats. This may explain why plants achieved a better fruit set in urban sites at that time, which could result in a better reproductive success.
Bee pollinating peak was reached in early spring in the forest (natural) habitat, whereas it was delayed to mid-summer in the city (late summer in arable land for Leong et al., 2016). Both Harrison et al. Note: For each term, chi-square and p-value of the type III Wald chi-square tests are presented, as well as the estimates of each coefficient (±SE) ("urb." = urban; "adv." = advanced; "del." = delayed; rural habitat, normal flowering period, and year 2017 were taken as references).

TA B L E 4 Summary
(2018) and Wray and Elle (2015)  F I G U R E 6 Reproductive success estimators, per flowering period, and for the two plant species: (a) and (b), respectively, fruit set and seed set rates of S. alba; (c and d), respectively, fruit set and seed set rates of L. corniculatus. Bars represent mean value ± SE. Stars represent significance levels from Tukey's post hoc tests At the community level, the broad urban phenological pattern could also be a consequence of an environmental filtering of pollinator species in favor of generalist species (Geslin et al., 2013;Wray & Elle, 2015). As these generalist species have particularly broad phenologies, the resulting urban assemblage would have a longer flight period. Besides, the environmental filter in the city could lead to the replacement of species by others whose traits better match the phenology of plants in the urban habitat (Banaszak-Cibicka et al., 2018). Previous studies showed this seems to be the case with small bee species of the genus Lasioglossum (Geslin et al., 2016). Species replacement may also be artificially enhanced by the introduction of managed honeybee colonies in the city, as the different pollinator morphogroups investigated showed variable responses to increased apiary densities (Ropars, Dajoz, Fontaine, Muratet, & Geslin, 2019).
It is very unlikely that the observed differences in fruit set and seed set between habitats were due to varying selfing rates between localities or between habitats. Furthermore, S. alba is an obligate outcrossing species (Olsson, 1960) and such is also the case for L. corniculatus (Ollerton & Lack, 1998). On the other hand, Pellissier et al. (2012), while monitoring reproductive success of Lotus corniculatus along an urbanization gradient, observed a greater fruit set in suburban areas than in dense urban sites. Since their study was conducted during the natural flowering period of the plant, this result is not at odds with our present work: Here, we did not observe a significant difference in fruit set between urban and rural sites during the natural flowering period of the plant.
Overall, our results suggest that a flowering phenology broadening might be beneficial to S. alba, as elevated reproductive success rates were not limited to its natural flowering time. This positive impact of phenological broadening on reproductive success was less strong for L. corniculatus, even though it was detected in the urban habitat where bumblebees displayed a longer, and especially an earlier starting flight period. Besides, this plant species needs strict photoperiod conditions to initiate flowering (Steiner, 2002).
Here, we witnessed different phenological patterns of pollinator activity between an urban and a rural habitat. Still, we cannot explain the underlying causes of these differences with our experimental setting alone. The two habitats also differ by several factors, among them by their temperature. Indeed, through meteorological data we detected an UHI effect, with a mean difference of about 2°C in daily minimal temperatures between our urban and rural habitats ( Figure S1). Adaptation to UHI might contribute to shape the We found differences in pollinator assemblages visiting the two focal plant species used in our experimental design. This emphasizes the need to carry such experimental approaches on several plant species with contrasting floral morphologies. Indeed, it is necessary to monitor the responses of a large range of pollinator groups-with various mouthparts morphologies enabling them to visit contrasted corolla shapes-in order to assess the response of the pollination function to plant phenological changes within and among environments. Here, the assemblage of pollinators we witnessed was shaped by our choice of plant models and also by the timing of observation. Hence, it is not an exhaustive survey of all the pollinators present in the two habitats studied.
The high temporal resolution of our experiment, with two weekly monitoring sessions spanned over several months, together with the high level of maintenance required to set up the plant assemblage in the different localities surveyed, made it difficult to multiply geographical replicates. Overall, we are aware that future research on the response of the pollination function to phenological and habitat changes would benefit from a wider set of geographical replicates, especially if they could encompass an urban-rural gradient (Fisogni et al., 2020).

ACK N OWLED G M ENTS
The authors wish to thank the Jardin des Plantes of the MNHN

DATA AVA I L A B I L I T Y S TAT E M E N T
All data are archived in the publicly accessible repository Zenodo, within the "iEES-Paris OpenData" community: https://doi.