Phenological shifts drive biodiversity loss in plant–pollinator networks

Plant–pollinator interactions are key for ecosystem maintenance and world crop production, and their occurrence depends on the synchronization of life-cycle events among interacting species. Phenological shifts observed for plant and pollinator species increase the risk of phenological mismatches, threatening community stability. However, the magnitudes and directions of phenological shifts present a high variability, both among communities and among species of the same community. Community–wide consequences of these different responses have not been explored. Additionally, variability in phenological and topological traits of species can affect their persistence probability under phenological changes. We explored the consequences of several scenarios of plant–pollinator phenological mismatches for community stability. We also assessed whether species attributes can predict species persistence under phenological mismatch. To this end, we used a dynamic model for plant–pollinator networks. The model incorporates active and latent life-cycle states of species and phenological dynamics regulating life-cycle transitions. Interaction structure and species phenologies were extracted from eight empirical plant–pollinator networks sampled at three locations during different periods. We found that for all networks and all scenarios, species persistence decreased with increasing magnitude of the phenological shift, for both advancements and delays in flowering phenologies. Changes in persistence depended on the scenario and the network being tested. However, all networks exhibited the lowest species persistence when the mean of the expected shift was equivalent to its standard deviation and this shift was greater than two weeks. Conversely, the highest species persistences occurred when earlier-flowering plants exhibited stronger shifts. Phenophase duration was the most important attribute as a driver of plant persistence. For pollinator persistence, species degree was the most important attribute, followed by phenophase duration. Our findings highlight the importance of phenologies on the stability and robustness of mutualistic networks. Author summary Plant-pollinator interactions involve a great number of species and are essential for the functioning of natural and agricultural systems. These interactions are facing a great number of threats. In both plants and pollinators, life-cycle events including flowering and adult emergence are triggered by environmental cues such as temperature and snowmelt. Climate change has the potential to alter the timing of these events. These phenological shifts generate mismatches in the timing of interacting species. Thus, plants and their pollinators may not match in time and/or space, leaving flowers unpollinated and disrupting pollinator feeding. Given that natural communities are composed of multiple species interacting in complex ways, experimentally assessing the effects of this kind of perturbation is difficult. To tackle this challenge, we simulated different scenarios of phenological shifts for several empirical communities. Our results indicate that strong shifts in the timing of life-cycle events may represent a greater risk of community collapse. Likewise, plants with short blooming periods and pollinators with short activity periods or high specialization face a greater risk of extinction.

Interactions between plants and their pollinators are a fundamental component of 2 production and biodiversity in terrestrial ecosystems. The majority of angiosperm 3 plants (87.5%) depend of animal pollination for reproduction [1], and thousands of 4 animal species obtain food and other resources produced by flowers [2]. On the other 5 hand, pollination plays an important role in world crop production [3]. Plant-pollinator 6 interactions can occur only if flowers and active pollinators (e.g. adult insects) overlap 7 in space and time, and cannot be realized if any of the mutualists is either absent or 8 present only in a latent state such as seeds or larvae. Thus, an adequate synchronization 9 between the life cycles of plants and pollinators is critical for the realization of 10 mutualistic interactions. 11 Long-term studies indicate that the phenologies of many organisms are shifting [4][5][6][7]. 12 One of most frequently observed phenomena has been the shifts in the occurrence of 13 plant flowering [8], possibly due to changes in average temperatures [8,9]. Because plant 14 and animal phenologies are likely to respond to different climatic cues [8,10,11], 15 phenological mismatches are increasingly likely [8,12,13], with uncertain effects on the 16 strength and maintenance of mutualistic interactions, species abundance and persistence 17 and ecosystem functioning [13,14]. In spite of our increasing understanding of the 18 population-level consequences of plant-pollinator mismatches, it is largely unknown 19 how complex ecological communities will respond to such phenological mismatches. 20 Furthermore, the magnitudes and directions of phenological shifts has been shown to 21 vary both among different communities and among species of the same community [9]. 22 Thus, there is a dearth of studies exploring the community-wide consequences of 23 plant-pollinator phenological mismatches. 24 The dynamics of multispecies networks experiencing phenological shifts involve 25 broad temporal scales, broad enough to make empirical research unfeasible. Thus, to 26 study these networks using mathematical models is a reasonable alternative. 27 Appropriate models should represent the long-term as well as the phenological 28 dynamics of multiple interacting species [15,16]. In addition, life-cycle transitions play a 29 major role in phenological dynamics and we see as necessary to represent them in the 30 model. Yet, most of the modeling efforts conducted so far to understand the 31 consequences of phenological shifts on community dynamics have excluded these 32 desirable properties. For example, some studies [17,18] have considered only single 33 species systems. In [19] they relied on a static analysis to study multispecies networks. 34 In [20,21], the adopted approach is dynamical but it considers a single pair of 35 interacting species. None of them consider life-cycle transitions. 36 Here we assess the consequences of plant-pollinator phenological mismatches for 37 community stability using mathematical modeling.  [25,26] Population dynamics 58 We use a dynamic model as a set of coupled ODEs. This system is a particular case of a 59 more general integro-differential model published in [27]. Each plant species is  available to pollinators and immature seeds and larvae are produced (Fig. 1).

72
Environmental favorability for germination, flowering and pollinator recruitment are dynamics of immature seed biomass is given by: Seed production is proportional to visitation rate (ϕ) of flowers with biomass F T i by 77 pollinators with biomass A T j . The second term represents immature seed mortality.

78
Mature seed dynamics is given by Germination phenology is governed by function f S i (t) and was located in the same 80 temporal position for all plant species. The beginning of germination was fixed to 6 81 weeks after the end of flowering of the latest species. The temporal duration of the 82 germination period was set to 8 weeks for all plant species. The second term represents 83 mature seed mortality. Biomass density growth rate of adult plants is given by: where plant biomass production due to seed germination is limited by intra-and 85 inter-specific competition for space (term within parenthesis). The last term is plant 86 mortality rate. Dynamics of flower biomass density is given by: where the first term represents the increase in floral resources, which is limited by flower 94 biomass. The second term represents resource consumption by pollinators. Resource that has the same structure that equation (1) for seeds. Mature larvae follow the Biomass density dynamics of adults insects is governed by: Beddington-DeAngelis-like functional response: where R(j) is the set of plant species visited by pollinator j and C(i) is the set of pollinator species that visit plant i. Finally, the between-years dynamics of seeds is governed by where Y is the length of the year. Equivalently, inter-annual dynamics of larvae is  S1 Table). State-variables were forced to zero 115 whenever their value decreased below its given extinction threshold (See S1 Table). At 116 the same time, a species was considered extinct when all their states-variables fell down 117 to zero. For each replicate, the parameter values were randomly drawn from uniform 118 distributions, centered in values taken from biologically plausible estimates or based on 119 available literature. These values are shown in S2 Table. For each replicate, a transient 120 simulation was run until the system reached an asymptotic oscillatory behavior, which 121 can be considered analogous to the steady behavior in classical autonomous dynamic 122 models. The number of species persisting at the end of this transient dynamics was 123 recorded and used as the initial species richness for the post-transient phase.

124
Experimental design 125 We evaluated the effects of temporal mismatch between flowering and pollinator activity 126 on the long-term dynamics of plant-pollinator communities. To this end, we conducted 127 three experiments. In all cases, the center of the flowering period of each species was 128 shifted an amount of time that was randomly drawn from a normal distribution, which 129 we call TSD (short for "temporal shift distribution").  observed in plants by [9]. 2. Increasing the standard deviation by 25 equally-spaced 139 levels, from -6 to 6 weeks, with mean of TSD set to 0. This allows evaluating the effects 140 attributable to changes in the variability of phenological shifts among species, even if 141 the community as a whole tends to keep the central position of their phenologies. These 142 treatments, termed TSD-sd, emulate phenological shifts observed in plants by [9]. 3. . Species persistence was recorded for the whole set of species as well as 155 for plants and pollinators separately. In this and subsequent experiments, we performed 156 a non-metric multi-dimensional scaling (NMDS) analysis to evaluate differences in our 157 results among networks. We also tested, with a Mantel test, the null hypothesis that the 158 community persistence differ between networks from different locations but do not differ 159 between networks sampled at different years in the same location.  For all networks and all choices of TSD, species persistence decreased with increasing 220 magnitude of the phenological shift, for both advancements and delays in flowering 221 phenologies (Fig. 2). However, in most cases, changes in persistence did not depend  Analyzing species persistence separately for plants and pollinators offers additional 236 insights: while pollinator response followed the above pattern for community persistence, 237 for plant response curves were flatter and on average higher than for pollinators, with  Phenophase duration was the most important attribute as a driver of plant species 251 persistence, while species degree was the most important attribute for pollinators, 252 followed by phenophase duration (Fig. 3). The only exception to this pattern was the case is best seen in Zackenberg while the second one is more apparent in Nahuel Huapi 261 and Villavicencio (Fig. 3).

262
The magnitude of the phenological shift had little influence in determining which timing of interacting species is already at some community optimum. This temporal 275 arrangement could be attributable to co-evolutionary processes. However, this optimum 276 could involve either plants appearing before their pollinators or vice-versa.

277
It may seem intuitive that this optimal arrangement involves perfect temporal 278 matching among interacting species. However, this is not necessarily the case. In 279 principle the optimal could involve either plants appearing before their pollinators or 280 vice-versa. In the match/mismatch hypothesis [36], it is proposed that a resource 281 population appearing earlier in the season favors the success of their consumers.

282
Extrapolating to our study, this suggests that at the optimal temporal arrangement, 283 host plants should present earlier flowering phenophases than their pollinators. By 284 contrast, Fagan et al. [21] found that a higher plant population growth is obtained when 285 pollinators recruit shortly before plant flowering. Likewise, [37] published a similar 286 result for trophic networks. Unfortunately, their approaches considered only two and 287 three species respectively and therefore they are not directly applicable to our networks. 288 The first reason lies on the difficulty in quantifying basal phenophase shifts (i.e. those 289 occurring naturally in the field before any data manipulation) among multiple 290 interacting species. The second one is that our raw data consists of interaction records 291 and not of independent plant and pollinator phenophases. to be relatively less sensitive to these disruptions, which is largely attributable to the 298 higher persistence of pollinators found for this scenario. Note that our most adverse 299 scenario TSD-msd was the only one used in [28] and [19], which opens the question 300 about the outcomes of their approaches under the rest of scenarios studied here.

301
The decrease in species persistence caused by phenological shifts observed in our 302 results agrees with previous studies, particularly [19], who also found that species were 303 affected more severely as the shifts grow larger. That study reported that, whenever 304 species phenophases are estimated in the same way we did, observing that changing the inter-annual variability in the magnitudes of phenological 312 shifts strongly affects species persistence, even though the mean value of the shifts over 313 all species was zero. A similar result was previously reported from experimental 314 manipulations of life-history events in competing tadpoles [38].

315
Regarding the underlying causes of species persistence, our analysis revealed that 316 phenophase duration in plants, along with both phenophase duration and degree of 317 pollinators, are key attributes determining species persistence in the long run. In 318 agreement with our results, Fagan et al. [21] found that longer flowering phenophases 319 favored plant persistence, while Memmott et al. [19] found that higher connectivity of 320 pollinators led to higher pollinator persistence. Our analysis replicated both previous 321 results with a single modeling approach.

322
In the datasets used in our study, species degree and the length of their phenophases 323 are positively correlated (see S3 Table). However, these correlations are stronger for 324 pollinators than for plants. In the context of our model, we can provide an explanation 325 for the importance of phenophase duration and why degree is specially important for 326 pollinators to overcame phenological shifts. Phenophase length may increase persistence 327 probability of species because a longer phenophase implies more opportunities for  This is because plants can sustain flower abundance through time (see first term of Eqn. 335 (4)). In contrast, if pollinators lengthen their activity period reproductive rate may drop 336 due to decreasing abundance of reproductive adults (Eqn. (8)). For this reason, to 337 maintain larval production, pollinators need to combine high connectivity with long 338 phenophases to compensate for decreases in abundance. This can explain the strong 339 correlation observed between degree and phenophase length in pollinators.

340
Note that degree and abundance are usually directly related in both plants and 341 pollinator species in a given network [39]. Furthermore, the presence of low-abundance 342 species tends to be underestimated since detecting them requires a greater sampling 343 effort [40]. Consequently, the degree and phenophase duration of specialist (i.e. rare) 344 species are likely to be underestimated [40][41][42]. Based on our findings, this imply that 345 ultra-specialist species are more scarce and therefore their persistence probability could 346 be higher than suggested by recorded data.

347
In this study we have considered a rather wide gradient of phenological shifts for our 348 analyses. However, empirical estimations show an average advance of 11.5 days in 120 349 years (roughly 1 day per decade) for appearance of pollinators, and 9.5 days for plant 350 flowering in America [13]. Likewise, estimates for European pollinators suggest an 351 average phenological advancement of 6 days in 60 years (1 day per decade) [43]. Such predict an abrupt increase in extinction rates.

357
Current empirical evidence supports that the timing of flowering and pollinator 358 emergence are being disrupted, presumably due to climate changes [7,44]. These 359 observed shifts in phenological events could lead to temporal mismatches between 360 flowering seasons of species and the periods of pollinator activity (as suggested in [8]), 361 although direct empirical evidence of this kind of interaction disruption is elusive up to 362 date [45,46]. For example, [31] and [47] showed that visitation frequency, associated to 363 pollinator abundances, decreased along with observed flowering advancements.

364
Likewise, [48] showed that early flowering produced lower pollination success and 365 consequent lower reproductive success for plants. Schenk et al. [49] found that even 366 small mismatches between plants and pollinators affect reproduction, activity and 367 survival of pollinators. However, the consequences of phenological shifts and the 368 interaction disruptions remains to be understood at the community and ecosystem 369 levels. Our study contributes to filling a gap in the understanding of community level 370 effects of shifting phenologies in plant-pollinator interactions [50]. However, although 371 we identified consistent responses from the analysis of some perturbed plant-pollinator 372 systems, our results show considerable differences among the studied networks. This 373 underlines the fact that network studies based on a single community should be 374 interpreted with caution since revealing general patterns requires a comparative analysis 375 among networks. Our findings, in agreement to observed in previous work [51], highlight 376 the importance of phenologies on the stability and robustness of mutualistic 377 communities. Future research should consider studying some real phenomena not 378 included in our model. For example, the shortening of the pollination season already 379 observed in some real life systems [43,52]. Additionally, between-community variability 380 in responses to phenological shifts poses the questions about which are the chief 381 determinants of these differences, species traits or connectivity patterns.