Bimodal breeding phenology in the parsley frog Pelodytes punctatus as a bet-hedging strategy in an unpredictable environment despite strong priority effects

When environmental conditions are unpredictable, expressing alternative phenotypes spreads the risk of failure, a mixed strategy called bet-hedging. In the southern part of its range, the parsley frog Pelodytes punctatus breeds both in autumn and in spring. Our aim was to study the breeding phenology and reproductive success associated with the use of those two seasonal niches to understand how this breeding strategy can be maintained. Field surveys revealed that breeding phenology was typically bimodal with a higher breeding effort in autumn. More importantly, in spring, the survival rate of offspring was severely reduced by the presence of autumn tadpoles, indicating a clear priority effect. However, the autumn cohort often failed to survive over winter, in which case spring cohorts were often successful. Based on those results, we constructed a model in which females can allocate a variable portion of eggs to each season and added a priority effect. We conclude that the existence of the two breeding seasons may indeed constitute a bet-hedging strategy.

parsley frog Pelodytes punctatus breeds both in autumn and in spring. Our aim was to study the 23 breeding phenology and reproductive success associated with the use of those two seasonal 24 niches to understand how this breeding strategy can be maintained. Field surveys revealed that 25 breeding phenology was typically bimodal with a higher breeding effort in autumn. More 26 importantly, in spring, the survival rate of offspring was severely reduced by the presence of 27 autumn tadpoles, indicating a clear priority effect. However, the autumn cohort often failed to 28 survive over winter, in which case spring cohorts were often successful. Based on those results, 29 we constructed a model in which females can allocate a variable portion of eggs to each season 30 and added a priority effect. We conclude that the existence of the two breeding seasons may 31 indeed constitute a bet-hedging strategy. 32

INTRODUCTION 34
Breeding phenology is one of the key components of adaptation to temporally variable 35 environments. Temporal dynamics of both the biotic and abiotic environment impose selective 36 constraints on parental development and physiological state (to be able to reproduce) as well as 37 offspring survival (at the various developmental stages until they reach sexual maturity and 38 start to reproduce) (Rand 1973  Gienapp 2014). Most of these studies concern species with a single reproductive peak in the 44 year, which has to match as precisely as possible a seasonal peak of resource availability in 45 order to maximise reproductive success (e.g. caterpillar availability for tits). The exact date of 46 the resource peak may vary from year to year and species usually rely on cues to anticipate it 47 and plastically delay or advance the onset of reproduction every year. However, in some cases 48 reproductive success depends on even more irregular and/or unpredictable conditions. In such 49 situations, species face the risk of complete reproductive failure at any given breeding attempt, 50 a regime that favors the expression of alternative phenotypes to spread the risk (Cohen 1970;51 fitness is highly impacted by low values; hence, traits with lower fitness variation may have 59 this seasonally variable breeding strategy. Do we have a single protracted breeding season or a 133 really bimodal reproduction generated by the coexistence of alternative breeding timing? If so,134 what is the relative importance of autumn versus spring reproduction? What is the survival of 135 offspring produced at the two breeding periods and how is it affected by the presence of 136 conspecifics? Once this basic knowledge is obtained, it can be fed into theoretical models for 137 the evolution of mixed breeding strategies. 138 In this paper, we characterize the breeding phenology (temporal dynamic, relative proportion 139 of each breeding period) of parsley frog in a French Mediterranean area based on results from 140 a 3-year field survey. We monitored the survival of offspring produced in each season to 141 estimate the success of this breeding strategy. We also investigated the factors influencing 142 breeding and tadpole survival, in particular whether there is a priority effects between seasonal 143 cohorts. Finally, using an analytical model adapted from Cohen (Cohen 1966)  included in subsequent analyses (as adults as well as larvae).
We divided the total counts for each amphibian larvae and invertebrate predators captured in 182 each pond by the number of dipnet sweeps taken in each pond. This procedure yielded a crude 183 proxy for density on the basis of catch per unit effort and could therefore be compared across 184 localities. This index could only be estimated in about one third of the breeding events when hatching was 198 successful (i.e. the number of small tadpoles was not null)). 199 The number of tadpoles in a pond was estimated using the mean number of tadpoles caught per 200 dipnet sweep scaled to a sampling surface of 1 m² (we estimated that one dipnet sweep sampled 201 a surface of 0.5 m², taking the dipnet size and the length of the haul into account) and then 202 multiplied by the surface of the pond. This should not be taken as an attempt to estimate 203 precisely the number of tadpoles present in a pond at a given time but as an index of abundance 204 that can be compared between ponds and between breeding events. It was sometimes impossible 205 to follow the larval development and metamorphosis of offspring from a particular breeding event. Indeed, parsley frogs may breed three to four times during each seasonal breeding event. 207 In these cases, the successive sub-cohorts produced are indistinguishable after a few weeks, and 208 we summed the eggs counted in two or three successive visits to evaluate survival from a 209 combination of breeding events within a given season (and within a site). Survival measures 210 should be viewed as an index to assess the differences of reproductive success between seasons 211 as there is no reason to expect any seasonal bias in this index. 212 213

Explanatory variables 214
Explanatory variables for the breeding probability and breeding effort are the season, depth of 215 the pond as well as the presence of conspecific and inter-specific competitors (larvae of anuran 216 species) and predators (invertebrates and adult newts) in the pond. Except for the depth of the 217 pond, all those explanatory variables were also applied to explain the success (offspring 218 survival) of breeding events. We summed the density of competitors and similarly the density 219 of predators despite the differences in competitive performance and predation pressure of the 220 various species toward parsley frog tadpoles. 221 To assess the potential impact of predation and competition on survival rates, we evaluated the 222 mean density of predators and competitors encountered by parsley frog tadpoles during their 223 larval development. More precisely, data from literature indicates that only small tadpoles (<12 224 mm snout-vent length) have lower survival due to predation by aquatic invertebrates (Tejedo 225 1993). Above this size, the predators will only injure them or even fail to catch them. Larvae 226 laid in autumn reached this limit size in about 3 months, whereas only 1.5 month is necessary 227 for larvae laid in spring (personal observation). Thus, we used the mean density of predators 228 and competitors over a period of 3 months after spawning date for autumn tadpoles and 1.5 229 months for spring tadpoles. (1966), on the optimal reproduction strategy of an annual plant whose seeds can either 253 germinate or remain dormant. In our case, there are two strategies: autumn breeding with initial 254 success (i.e. the ability of offspring to persist until spring) depending on the environmental 255 conditions, and spring breeding with success depending mainly on the presence of autumn tadpoles, hence on the initial success of autumn breeding (as suggested by the results on success 257 of autumn and spring breeding events, see below). 258 Let c be the proportion of eggs laid in autumn (thus 1-c in spring)we assume, in agreement 259 with our data (see results), that c represents a fixed strategy, i.e. the frogs cannot predict failure 260 in advance to avoid laying in autumn, nor can they avoid laying eggs in spring when an autumn 261 cohort is present. As mentioned above, the autumn cohort is assumed to succeed or fail, at 262 random, with probability q and 1-q respectively. When it succeeds, a fraction s1 of the offspring 263 survive to reproductive age. The spring cohort completely fails whenever the autumn cohort 264 has survived in a pond (a reasonable simplification based on our survival rates estimates, see 265 below), otherwise a proportion s2 of spring tadpoles survive. Overall, the expected number of 266 offspring reaching sexual maturity is s 1 when the autumn cohort doesn't fail and 267 (1 − ) s 2 when it does. 268 If each frog reproduced only during one year, the optimal strategy would maximize the 269 geometric mean of the annual reproductive outcome (Dempster 1955) which is 270 Or, equivalently 272 However, reproductive life lasts more than one year in frogs (say, n years), which in itself is a 274 way to spread the risk of failure among successive cohorts of offspringan uncertainty remains 275 however, for each frog, on how many (k) of the n breeding years will not allow the autumn 276 cohort to survive. For each individual, k is distributed binomially with probability 1-q so that 277 We explored numerically the selection gradients in order to find potential ESS using 287 Mathematica (Wolfram Research Inc. 2018) based on the following parameter combinations. 288 We set survival probabilities based on our estimates of survival from egg to metamorphose: s1 289 = 0.047 (estimated among breeding events producing offspring that survived until spring) and 290 s2 = 0.038 (in the absence of autumn tadpoles). We assumed that survival and fecundity were 291 equal for both seasonal cohorts for the rest of the life cycle. We set the number of reproductive 292 years n = 3 to 5, according to a study of age structure of a breeding population in Spain (  The number of egg masses was higher in autumn than in spring (23.0 ± 4.0 egg masses per 321 breeding event in autumn and 13.7 ± 2.4 in spring; χ ² 1=9.25, p-value=0.002, Fig. 2 , Table 2  322 and Annex 2). As a result, autumn breeding contributed slightly more than spring breeding to 323 the production of egg masses (57% versus 42.9%). 324 325

Breeding success 326
Hatching success (i.e. the percentage of breeding events producing at least one larvae) was 327 higher in autumn than in spring (68.4% and 43.8% respectively, χ²1=11.12, p-value= 0.001, 328 success were explained by interspecific competition (the density of other amphibian larvae) or 337 by predation (density of potential invertebrate predators or number of adult newts, see Table  338 2). 339 Survival rates are represented in Figure 3 and Annex 3. The survival rates from egg to 340 metamorph were similarly low (autumn: 2.24 % ± 0.61 and spring: 1.97 % ± 0.73, Table 2 This may be due to the lack of information about predation during the first year of survey which 406 reduced our statistical power or to the fact that causative factors are numerous and more 407 complex to identify in the field. However, other studies reported no effect of predation on 408 tadpole survival (Hartel et al. 2007) or even a positive effect (Barandun & Reyer 1997), 409 probably due to predator-induced phenotypic plasticity. Nevertheless, our results suggest that 410 the predation pressure is probably not a stronger constraint in one season than in the other. 411 Spring tadpoles should be exposed to more competitors during their development than autumn 412 tadpoles since the majority of amphibian species in the local community breed in March and 413 April. Nevertheless, we found no effect of interspecific competition on survival for any of the 414 two seasonal tadpole cohorts. This seems surprising since parsley frog is a poor competitor as we observed in several occasions that large autumn tadpoles were eating freshly laid eggs of 429 their own species, which could partly explain the lower hatching rate of spring eggs in presence 430 of autumn tadpoles. Moreover, (Tejedo 1991) previously described how parsley frog tadpoles 431 preys on Epidalea calamita eggs. In this latter study, predatory tadpoles were exclusively old 432 tadpoles and they could cause a loss of 50 to 100% of the eggs. Oophagy has also been 433 demonstrated to be responsible for interspecific priority effects between Scaphiosus couchii 434 and Bufo speciosus (Dayton & Fitzgerald 2005). Intraspecific oophagy has been described in 435 some anuran species (Summers 1999 there is no significant difference in cohort survival (the probability to produce at least one 518 metamorph) between spring and autumn, in spite of a slightly higher risk of drought (and hence 519 complete disparition of the cohort) in autumn. Density-dependence (on which we have no 520 information) might partly explain why autumn cohorts do as well as spring cohorts in spite of 521 higher drought risk. Additionnally, our measures of breeding success are very rough because 522 counting precisely the number of larvae from each cohort in the ponds over the course of the 523 season is extremely difficult. There is thus still much to learn to fully understand the advantages 524 and disadvantages of autumn and spring strategies in this species. 525 Lastly, as explained above, we still don't know if individual females usually breed once a year 526 (either spring or autumn) or several times a year (potentially spring and autumn of the same 527 year). Capture-mark-recapture of adults and larvae would alleviate some of these limitations but would be highly challenging. However, our results remain valid for a large range of 529 parameters, and these uncertainties should not affect our conclusion that the breeding strategy 530 of parsley frogs in southern France constitute an original example of bet-hedging strategy 531 driven by high environmental stochasticity and large inter-cohort priority effect.