Heated rivalries: Phenological variation modifies competition for pollinators among arctic plants

Abstract When plant species compete for pollinators, climate warming may cause directional change in flowering overlap, thereby shifting the strength of pollinator‐mediated plant–plant interactions. Such shifts are likely accentuated in the rapidly warming Arctic. Targeting a plant community in Northeast Greenland, we asked (a) whether the relative phenology of plants is shifting with spatial variation in temperature, (b) whether local plants compete for pollination, and (c) whether shifts in climatic conditions are likely to affect this competition. We first searched for climatic imprints on relative species phenology along an elevational gradient. We then tested for signs of competition with increasing flower densities: reduced pollinator visits, reduced representation of plant species in pollen loads, and reduced seed production. Finally, we evaluated how climate change may affect this competition. Compared to a dominant species, Dryas integrifolia × octopetala, the relative timing of other species shifted along the environmental gradient, with Silene acaulis and Papaver radicatum flowering earlier toward higher elevation. This shift resulted in larger niche overlap, allowing for an increased potential for competition for pollination. Meanwhile, Dryas emerged as a superior competitor by attracting 97.2% of flower visits. Higher Dryas density resulted in reduced insect visits and less pollen of S. acaulis being carried by pollinators, causing reduced seed set by S. acaulis. Our results show that current variation in climate shifts the timing and flowering overlap between dominant and less‐competitive plant species. With climate warming, such shifts in phenology within trophic levels may ultimately affect interactions between them, changing the strength of competition among plants.

Generally, wild plants share their flowering period with many other temporally overlapping plant species (e.g., CaraDonna et al., 2017;Mosquin, 1971). Thus, competition by simultaneously flowering species could hamper the reproduction of a focal species through increased competition for pollinators. For many insectpollinated plants, it is crucial to gain access to pollination services provided by local pollinators. As a consequence, Mosquin (1971) hypothesized that species have evolved differing flowering times to increase their fitness by reducing interspecific competition for pollinators. Especially more abundant and attractive, dominant plant species are expected to reduce the fitness of less attractive species when flowering at the same time (Carvalheiro et al., 2014;Goodell & Parker, 2017;Montero-Castaño, Ortiz-Sánchez, & Vilà, 2016).
At present, climate change is shifting not only the phenology of individual species, but also the relative phenology of interacting species (CaraDonna, Iler, & Inouye, 2014;Høye, Post, Schmidt, Trøjelsgaard, & Forchhammer, 2013;Kudo & Cooper, 2019;Rafferty, Diez, & Bertelsen, 2020). Thus, individual species have to simultaneously adjust both to new abiotic conditions and to changes in the species with which they interact (Burkle & Alarcón, 2011;Kaiser-Bunbury, Muff, Memmott, Müller, & Caflisch, 2010;Saavedra, Rohr, Olesen, & Bascompte, 2016). In terms of pollination, progressing climate change may cause the timing of flowering in competing plant species to slide toward larger or smaller overlap (e.g., CaraDonna et al., 2014;Forrest, Inouye, & Thomson, 2010;Schmidt et al., 2016), as a likely consequence of interaction partners responding to different environmental cues. Consequently, shifts in the relative phenology of flowering may change the strength of intraguild competition and fitness of individual species. Yet, such effects remain poorly studied (but see Giejsztowt, Classen, & Deslippe, 2019;Kehrberger & Holzschuh, 2019). These changes in relative phenology are expected to come with more drastic consequences when the diversity is low (offering less potential for interaction rewiring; Benadi, Hovestadt, Poethke, & Blüthgen, 2014) or the flowering is highly seasonal (resulting in more dramatic mismatches in systems without "background noise" of flowers and pollinators; Pelayo, Soriano, Márquez, & Navarro, 2019).
The Arctic is a region where the above effects are likely accentuated (Høye, Post, Meltofte, Schmidt, & Forchhammer, 2007;Iler, Høye, Inouye, & Schmidt, 2013). Here, the growing season is short, and all flowering thus confined to a relatively short time window.
Moreover, the growing season has shrunk even shorter with progressing climate warming Prevéy et al., 2019).
Since climate change is particularly rapid in the Arctic (due to so called arctic amplification; Kattsov et al., 2005), we may expect to see climate-induced intensification in plant-plant competition for pollinators (Cirtwill, Roslin, Rasmussen, Olesen, & Stouffer, 2018;Post et al., 2009). Mimicking trends seen elsewhere (Cardoso et al., 2020;Powney et al., 2019), several arctic pollinators have also suffered recent population decline (Gillespie et al., 2019;Loboda, Savage, Buddle, Schmidt, & Høye, 2017). Hence, at high latitudes we may expect particularly strong competition for pollinators, and particularly accentuated shifts in interspecific competition for pollination with climate change.
In this study, we focus on a pollination community of the High Arctic. Focusing on six quantitatively dominant plant species and their associated pollinators in the Zackenberg valley of Northeast Greenland, we ask (a) whether the relative phenology of local plants is shifting along temperature gradients, (b) whether local plants compete for pollination, and (c) whether altering climatic conditions are likely to affect this competition in space and time. To this aim, we first examine the climatic imprints on relative phenology of flowering species along an elevational gradient (representing current climatic variability). We then search for signs of intensified competition with increasing flower densities, expressed as reduced visits per flower, reduced representation of plant species in pollen loads, and a reduced proportion of inflorescences producing seeds. The latter analysis was specifically focused on two species identified through our data on pollinator visits: Dryas integrifolia × octopetala Rosaceae as a plant attracting a particularly high proportion of insect visits, and Silene acaulis Caryophyllaceae is a gynodioecious plant with some proportion of individuals being female only. Such individuals will exhibit a particularly high demand for pollination, since self-pollination is-per definition-excluded. After testing for both phenological shifts in flowering time along environmental gradient and competition for pollination among the flowering species, we infer how the changing climate is likely to translate into shifts in the competitive landscape, and ultimately fitness, of local species.

| MATERIAL S AND ME THODS
To examine spatiotemporal patterns of flowering abundance, phenology, and interspecific overlap, we targeted sites along an elevational gradient in a high arctic system: the Zackenberg valley in NE Greenland (74°28′N, 20°34′W). At each site, we monitored the abundance of flowers at weekly intervals. To relate these flower abundances to interspecific competition for pollinators, we monitored multiple stages of the pollination process (Ne'eman, Jürgens, Newstrom-Lloyd, Potts, & Dafni, 2010) among the dominant flowering plant species: insect visitation rates (i.e., pollinator visits per flower and time unit), pollen transport (i.e., the representation of the species in pollen loads carried by pollinating flies), and seed set success (i.e., the proportion of inflorescences setting seed). We then searched for climatic imprints on relative species phenology along the elevational gradient (representing current climatic variation, with elevation as a space-for-time surrogate for climate change; e.g., Benadi et al., 2014;Elmendorf et al., 2015;Hoiss, Krauss, & Steffan-Dewenter, 2015;Kearns, 1992;Körner, 2007).

| Study species
To characterize variation in the phenology of flowering and competition for pollinators among plant species, we counted inflorescences of all flowering species. Six species emerged as quantitatively dominant in the local flora: Cassiope tetragona Ericaceae, D. integrifolia × octopetala Rosaceae, Papaver radicatum Papaveraceae, Salix arctica Salicaceae, Saxifraga oppositifolia Saxifragaceae, or S. acaulis Caryophyllaceae. These species are all abundant and widely distributed across the Arctic (Walker et al., 2005).

| Study sites
We recorded the flowering of plants and flower visitor abundance at 24 study sites (50 m × 50 m each) from late June to early August in 2016. In order to track the effects of the local variation in climate along the elevation gradient on plant-pollinator interactions in a spacefor-time experiment, we chose eight study sites in each of three zones along an elevation gradient (0-60, 60-240, and 240-480 m above sea level, 3 × 8 = 24 sites total; see Figure S1). To characterize temperature conditions across the gradient, we drew on records from another study (Kankaanpää, 2020), with data presented in Figure S2. To minimize the generally large effects of changing plant community on pollination along the elevation gradient (Simanonok & Burkle, 2014), each of the sites represented the same vegetation type, Dryas heath. This vegetation type is abundant and widespread at all elevations considered (Bay, 1998). To avoid effects caused by spatial variation in the timing of snow-melt (Kankaanpää et al., 2018;Kudo & Hirao, 2006), the study sites were established in areas at the same phenophase, that is, when the first flowers opened. The study sites were separated by distances of at least 250 m, a scale over which we assumed few arctic insects to move during their daily foraging. Thus, the sites were considered at least semi-independent in terms of their insect populations. Within each of the 24 study sites, we marked 10 study plots (circular, radius 50 cm) with small flags to locate them later.

| Phenological variation in flower densities and insect visitation
Once a week, we recorded the local, instantaneous density of all flowering plant species at two spatial scales: at the level of the study site and at the level of the study plot. The rate with which different insects visited flowers as a function of contemporary flower densities was scored twice a week by visual observation. During each visit, we walked up to a distance of 2 m from the plot and recorded all visitors on the flowers (thus scoring a snapshot of arthropod abundances present on the flowers upon the observer's arrival). This was, on average, achieved in just some minute per site. The radius of the area for visitor inspection equaled two visual fields of the binoculars used (Ibis, model 10 × 42; Kite), that is 50 cm, with the mark flag held at the center of the circle. Within this area, we recorded all flower visitors at the family level. Flower visitation observations were done mainly between 10:00 and 18:00 and only if the weather conditions were suitable for flower visitors (no rain or heavy wind).

| Pollen transport by flies
To establish the impacts of flower densities on pollen transport (i.e., the representation of pollen from focal flower species in pollen carried by insects), we focused on pollen loads on flies in family Muscidae. This taxon was chosen for being the presumptively most important pollinators in the area (and many other arctic and alpine areas as well; Kearns, 1992;Kevan, 1972;Pont, 1993;Tiusanen, Hebert, Schmidt, & Roslin, 2016), and the numerically dominant fly taxon of the High Arctic (Böcher, Kristensen, Pape, & Vilhelmsen, 2015;Loboda et al., 2017).
With the aim of examining how the pollen loads reflected plant species-specific flower densities and relative flowering phenology, we captured 10 fly individuals at each of the study sites every week.
The flies were individually caught with an insect net while they were basking on vegetation or soil on the study sites. To avoid interfering with local flower visitation patterns, and to prevent secondary contamination by pollen during handling, we explicitly avoided catching insects sitting on flowers. All flies caught were stored individually in ethanol-filled tubes.
To remove the pollen from the flies, we then vortexed the tubes (max rpm for 10 s, Vortex-Genie 2, Scientific Industries, Inc.). To concentrate the pollen in the bottom of the tube, the fly was removed, and the pollen suspension centrifuged at 3,000 g for 3 min (Sigma Laboratory centrifuges, model 4-15C). In order to count and identify the pollen, we evaporated the ethanol and cleaned the dry tube from pollen with the aid of agarose gel (15 ml glycerin, 25 ml water, 0.5 g agar, red food dye, Dr. Oetker), which was then poured onto a microscopy slide. The pollen samples were identified and counted with a microscope (CX41, Olympus). Due to the difficulty of identifying the muscid flies in the field, we stayed with family-level identification.

| Proportion of inflorescences setting seed
To estimate the impact of competition on the seed set of plants, we chose two abundant and widely distributed species: D. integrifolia × octopetala, henceforth Dryas for brevity, and S. acaulis. Of these, Dryas was selected as a particularly dominant species in the plant community with a need for pollinators for optimal seed set (Tiusanen et al., 2016) and S. acaulis is a plant with a particularly high demand for pollen transport services: S. acaulis is gynodioecious, with some plants being hermaphrodites and others female only (Kevan, 1972;Shykoff, 1992). Thus, in S. acaulis, we used female-only plants with an obligate need for pollinators to score whether they got successfully pollinated or not. For this purpose, we targeted a subset of study sites with sufficient abundances (>500 flowers per site) of these flowering species (14 and 10 sites for Dryas and S. acaulis, respectively). To keep track of spatiotemporal variation in the seed production, we marked 10 flowers per species (female-only individuals of S. acaulis) at each targeted site each week. All the marked flowers were recently opened (no more than 24 hr old). During the first visit to the study sites, we also found Dryas flowers, which were already senescent (at 14 of the sites) and withered (at 13 of the sites). At each of these sites, we marked 10 old and 10 withered flowers (in addition to freshly opened flowers). Reflecting the average flowering time of Dryas (M. Tiusanen, personal observation, June-July, 2016), we subsequently treated them as if they had first opened 2 or 6 days, respectively, before our visit.
To resolve the effect of pollinator availability on the seed set of Dryas, and to probe for differences in seed set over time, we excluded pollinators from accessing some of the flowers. We did so by covering 10 unopened buds with plastic cups (Iisi, 0.25 L, Nupik International). To minimize the effects of the treatment on temperature and moisture, the bottom of which had been replaced with a mesh (mesh size 0.3 mm × 0.3 mm, Yleistylli, pehmeä, Eurokangas).
Wherever possible, we chose flowers on the same tussocks as the flowers monitored for seed set in the presence of pollinators. For S. acaulis, we relied on the self-sterility of the female-only individuals targeted (see above). Seed set by such individuals reflects successful pollen transport from another individual. Given the spatial distribution of plants in the study system, it will almost invariably require an insect vector.
At the end of the season, we investigated seed set success of Dryas and S. acaulis (i.e., whether the flower had produced seeds or not) of the marked flowers, and-for Dryas-of the flowers from which pollinators had been excluded. Since Dryas emerged as the most attractive species (see Section 3) we used the difference of the date of mean flowering between Dryas and the focal species (C. tetragona, P. radicatum, S. arctica, S. oppositifolia, or S. acaulis) as the response variable and elevation as an explanatory variable. Sites with observations of a species flowering only on 1 day were excluded from the species-specific analyses, as offering records with disproportionately low precision.

| Relative phenology of flowering species
Notably, the key interest here relates to the interaction "plant species" × "overall annual phenology." A significant interaction will reveal differential responses in different species, and the slope estimates will indicate the extent to which flowering in different species gets compressed or spread aside by variation in the relative earliness of the year. In other words, the interaction will quantify the impact of climate on the overlap in timing among competing species. To account for environmental variation causing site-to-site differences in flowering time, we used study plot as a random effect. The model was fitted with package lme4 (Bates et al., 2017) in R (The R Core Team, 2016).
To further examine the potential for competition between the plant species, we characterized the site-specific temporal niche overlap in flowering between Dryas and other species with Schoener's index (SI, Schoener, 1970). Sites with observations of a species flowering only on one day were excluded from the species-specific analyses, as offering records with disproportionately low precision.

| Visitation rate as a function of flower densities
If with more flowers of the most attractive species (

| Pollen transport by flies
Since To account for site-to-site differences in pollen loads, we included study site as a random effect. Since the dependent variable was a proportion of events, we assumed a logit-link function and binomially distributed errors. The presence or absence (1/0) of S. acaulis pollen was then modeled by an equivalent GLMM. The models were fitted package lme4 (Bates et al., 2017) in R (The R Core Team, 2016).
Because monitoring of study sites V, W, and X (see Figure S1) did not start from the beginning of the season, they were excluded from the analyses.

| Seed set by inflorescences
To resolve temporal patterns in seed set, we used GLMMs of the fraction of Dryas and S. acaulis inflorescences, respectively, produc-  Meanwhile, compared to Dryas, the relative phenology of P. showed no patterns in relation to the phenology of Dryas along the elevation gradient.
As a consequence of shifts in phenology, the niche overlap with Dryas showed variation along the elevation gradient (Figure 2b).

| Arctic plants compete for pollination
Of a total of 2,287 observed flower visitations, a highly dispropor-  Figure 6). found that pollinators in the target area is in short supply, as evidenced by decreasing visitation rates with higher flower densities ( Figure 3) and by decreasing seed set with the exclusion of pollinators ( Figure 6). As a result, plants compete for pollinators-which is evidenced in decreasing visitation rates with a higher abundance of the more-attractive Dryas (Figures 1 and 3), in a declining representation of the less-competitive S. acaulis in pollen loads ( Figure 4) and in patterns of seed set ( Figure 5). Finally, this competition is likely affected by the shifting climate along the elevation gradient which modifies the intensity of the competition through changes in phenological overlap between the flowering species ( Figure 2).

| D ISCUSS I ON
Below, we will examine each of these findings in turn.

| Elevation affects relative flowering phenology and niche overlap
During the arctic summer, flowering advances quickly, with the peak flowering period at Zackenberg lasting less than 3 weeks. Within such a short summer, it is crucial for plants to reproduce efficiently. Thus, Day of year interactions observed along elevational gradients appear to arise through changes in the respective species pools (Adedoja et al., 2018;Brittain, Kremen, & Klein, 2013;Lara-Romero et al., 2019;Maglianesi, Blüthgen, Böhning-Gaese, & Schleuning, 2015;Partida-Lara et al., 2018;Simanonok & Burkle, 2014), with communities at higher elevations being characterized by less species and more generalized interactions (e.g., Hoiss et al., 2015;Ramos-Jiliberto et al., 2010). In this study, we tried to minimize the effects of changes in plant diversity and composition (Simanonok & Burkle, 2014) by choosing all our study sites within a single, standardized vegetation type (Dryas heath). Unavoidably, flower abundances and plant diversity proved slightly lower with higher elevation (Figures S3 and S4), with a major change being the gradual disappearance of C. tetragona and Vaccinium uliginosum with increasing elevation. However, as neither species was particularly attractive to pollinators (see Figure 1), we are confident that variation in their abundances will have little if any effects on the patterns observed. Thus, the patterns detected seem more reflective of climatic impacts on the relative phenologies of a constant set of species than of climatic impacts on community composition as such.
In terms of the exact climatic cline occurring along our focal elevational gradient, we notice a surprising feature. Contrasting with general biogeographic patterns and with our own initial expectations, in our study system mean temperatures tend to increase with increasing elevation (0.4°C/100 m; Figure S2). This is due to two phenomena: a shading effect of the neigh-  Høye et al. (2013) showed that during the last two decades, the range of temperature variation at Zackenberg has been substantial, and large enough to change the relative flowering of Dryas and S. acaulis. In earlier (and warmer) years, the flowering of S. acaulis in "late" sites has become relatively earlier compared to Dryas flowering, whereas the opposite has been true for "early" sites. Thus, how the timing of the species will shift does depend on the warming experienced. For all we can tell, further warming is likely to shift the flowering of S. acaulis and Dryas closer together and increase their niche overlap (to the right in Figure 2a,b).

| Arctic plants compete for pollinators
In  (Holzschuh et al., 2011) or invasive plants (Brown et al., 2002;Goodell & Parker, 2017). While highly attractive plants may also prop up local pollinator densities to the benefit of less attractive plants (Ghazoul, 2006;Losapio et al., 2019), the current findings show how the monopolization of pollen transport may strongly outweigh such benefits during the condensed flowering season of the Arctic.

| Does competition for pollination explain seasonal patterns in seed set?
We note that the patterns in seed set here attributed to competition are also patterns in time, since high densities of Dryas are concentrated to a given period in the early summer ( Figure 5a). Thus, other factors changing in time could potentially contribute to the patterns observed. Yet, we note considerations providing evidence for a competitive effect beyond temporal variation in plant performance.
In the current data, we found a general decline in seed set of Dryas with the advance of the summer, and this effect extended both to flowers visited by pollinators and to flowers from which pollinators were excluded ( Figure 6). Seed set in Dryas was highest for flowers open before the flowering peak, when pollinators are active but competition by other flowers (including other Dryas flowers) is still limited (see Figure 5). This pattern might be an adaptation to the naturally short growing season of the Arctic, where investment in early seeds may be safer than seed set by late flowers, as the risk of failure increases with the proximity of the fall frost period (a risk demonstrated in an Alpine system by Kudo (1993). Meanwhile, despite the overall decline of seed set in late flowers, the fraction of seed set attributable to pollinator presence increased toward the late season ( Figure 6). This pattern may potentially arise from plants investing more in producing outcrossed seeds (see, e.g., Free, 1966;Orford, Murray, Vaughan, & Memmott, 2016), thereby challenging the vagaries of the arctic fall.
Meanwhile, seed set of S. acaulis showed an opposite temporal pattern with increasing seed set toward the end of the season, suggesting that the seasonal patterns in environmental conditions are not driving the patterns of seed set of plants in the area.

| Competition changes with a warmer climate
On top of the competition for pollinators caused by Dryas, we observed shifts in the relative timing and niche overlaps of flowering species. However, they are by no means the only climate-driven changes observed in the system. Two changes in particular may have increased overall competition among plants for pollinators: First, in years of early phenology, the flowering season of plants is shorter and compressed toward the early season, that is there is shorter time for pollination (see Høye et al., 2013;Prevéy et al., 2019;Schmidt et al., 2016) and-given less compression of the timing of pollinator activity )-less pollinators per individual flower. This will increase the overall demand for pollen transport during this condensed flowering period, with no corresponding increase in pollinator availability. Second, changes in the pollinator community may be affecting the general availability of pollination services, and the competitive context of the plants. Interestingly, the pollen limitation during the peak flowering of Dryas may already have been accentuated by climate-related effects: the abundance of the important muscid flies at our study area has decreased by 80% during the past 20 years (Loboda et al., 2017). Overall, the mismatch between flowering of plants and pollinator flight season at Zackenberg has apparently increased, as mainly caused by a shortening flowering period with no matching reduction in the insect activity period Schmidt et al., 2016). The shortening of flowering season and increasing mismatch between pollinators could leave late-emerging insects devoid of relevant resources to reproduce, thus contributing to changing of community composition (Gillespie et al., 2019) and to the decline of both muscid flies and other important pollinators Loboda et al., 2017).

| CON CLUS IONS
Progressing climate change may not only change the distribution of species, but also the distribution of their biotic interactions through changes in species phenologies. When species at the same trophic level compete for interaction partners, a shift in their abundances or relative timing may cause indirect effects through shifts in the strength of competition. In this paper, we reveal suggestive evidence for such a signature of climate change on biotic interactions in a rapidly warming Arctic: We find that during the short arctic summer, plants seem to compete for access to pollinators. The relative intensity of this competition is dictated by their phenological overlap, which is modified by climate. Recent climate change may already have aggravated these patterns-and will likely change competitive interactions further over the next few decades of rapid warming.
Our findings suggest a worrying imprint of climate change, through a shift in the indirect interactions of arctic species. These findings point to shifts in the biotic interactions among arctic species, and urges a novel focus on shifts in horizontal interactions within trophic layers, as driven by ongoing climate change.

ACK N OWLED G EM ENTS
Aarhus University is thanked for providing access to the Zackenberg Research Station, where the Logistics team and the BioBasis team created an excellent working environment. We

DATA AVA I L A B I L I T Y S TAT E M E N T
The complete data set is available at Dryad: https://doi.org/10.5061/ dryad.nzs7h 44pm.