Introduction

Organisms are exposed to variation in their environment, so that their fitness is related to how they adapt to the fluctuation regime in their habitats (Levins, 1968; Kolasa & Rollo, 1991; Meyers & Bull, 2002; Vasseur & McCann, 2007; Pearman et al., 2008; Botero et al., 2015). Environmental fluctuation can be decomposed in predictable and unpredictable components. For instance, in a wide range of organisms, an internal circadian clock has been described, allowing anticipating a physiological adjustment to the periodic variation in temperature (Huey & Bennett, 1990; Kern et al., 2015) or light (e.g., Yerushalmi & Green, 2009; Brown, 2016). Even rather irregular fluctuations can still be anticipated as shown by the effects of infochemicals on the development of inducible defenses when predators are present (e.g., Tollrian & Harvell, 1999; Lass & Spaak, 2003), or on triggering physiological switches when a quorum for sexual reproduction has been achieved (Snell et al., 2006; Kubanek & Snell, 2008; Tian et al., 2018). These instances show that fluctuation in predictability needs to be evaluated from the point of view of the focal organism (i.e., subjective unpredictability).

An expected evolutionary response to environmental fluctuations is adaptive tracking, with genetic change driven by selection tracking the environmental change. The resulting pattern is a gradual trait change lagging environmental change. If environmental change is fast enough it may not allow the fixation of optimal genotypes, and would contribute to the maintenance of genetic variance for the traits under selection (Sasaki & Ellner, 1997). Besides adaptive tracking, which does result in adaptation, adaptive responses that are expected to evolve when environmental change is too fast to be tracked are of two kinds: adaptive phenotypic plasticity and bet hedging (Simons, 2011; Botero et al., 2015). Phenotypic plasticity—the capacity of a single genotype to exhibit variable phenotypes in different environments—is expected to occur when organisms have reliable information of future environmental conditions (Furness et al., 2015); so it implies predictive capability. On the contrary, bet hedging involves strategies by which a genotype reduces fitness variance at the expense of a decrease in short-term fitness to spread risk in future uncertain environments (Seger & Brockmann, 1987; Olofsson et al., 2009; Gremer & Venable, 2014). Two main modes of bet hedging have been distinguished: diversified and conservative (Childs et al., 2010). Diversified bet hedging occurs when a single genotype produces a variety of phenotypes in its offspring in advance of future unpredictable conditions and constitutes the canonical mode of bet hedging (Cohen, 1966). Empirical evidence in favor of diversified bet hedging is still scarce (see Simons, 2011) and much focused on plants, especially in relation to seed dormancy in desert annuals (Philippi, 1993a, b; Clauss & Venable, 2000; Evans et al., 2007). Nevertheless, studies on animals do exist (e.g., Philippi et al., 2001, for the duration of diapause in branchiopods; Halkett et al., 2004, for the proportion of sexual vs. asexual forms in aphids). Conservative bet hedging takes place when a genotype produces offspring with a single low-risk phenotype (Philippi & Seger, 1989); that is a phenotype performing well in case of a bad period occurring unexpectedly. Examples of conservative bet hedging have been reported both in plants (e.g., delayed flowering; Simons & Johnson, 2003) and animals (e.g., timing of switching to diapausing egg production in copepods; Hairston & Munns, 1984). Bet-hedging theory focuses commonly on life-history traits. However, bet hedging can involve other traits; e.g., constitutive anti-predatory defenses just in case predators are present in the environment. The outlined bet-hedging scheme focuses on one or more phenotypes controlled by a single genotype (herein, standard bet hedging). Additionally, genetic diversity in the offspring (e.g., via recombination) resulting in phenotypic variation has been also regarded as bet hedging. Thus, sexual reproduction has been recently studied as a case of bet hedging (Li et al., 2017).

Monogonont rotifers are cyclical parthenogens (Fig. 1) dwelling in aquatic habitats with important environmental fluctuations including the more or less predictable alternation of suitable and unsuitable periods. At the (typically annual) growing season onset, the active population starts with females hatched from diapausing eggs. From these stem females, clones of females proliferate asexually (i.e., relatively fast, without incurring the costs of sex). Sexual reproduction is induced by environmental cues that trigger asexual females to produce sexual daughters as some fraction of their offspring (the sexual reproduction ratio or mixis ratio), so that asexual reproduction does not stop (Gilbert, 1974; Pourriot & Snell, 1983; Schröder, 2005). Sexual females produce meiotic haploid eggs that develop into either males or, if fertilized, into diapausing eggs that sink to the sediment. These sexually produced eggs are the only ones able to enter into an extended diapause period and to resist adverse conditions. When suitable conditions are met, a proportion of the survivors of these eggs hatch (diapausing egg hatching fraction), the rest remaining in the sediment egg bank.

Fig. 1
figure 1

Typical life cycle of monogonont rotifers, as occurring in the genus Brachionus (modified from Serra et al., 2018), where the sexual induction signal is a protein secreted by the females (i.e., density-dependent sex initiation; Snell et al., 2006). Sexual propensity for sex initiation is determined by the threshold density to trigger the sexual phase

Full size image

A population of cyclically parthenogens can be envisioned as a collection of annual clones (genetic individuals; genets; Young, 1979; Serra et al., 2004; Gómez & Carvalho, 2000; Stelzer, 2005). If so, a clone would be conceived as a ‘monoecious sexual individual’ that ‘grows’ by asexual reproduction and ‘reproduces’ by sexual reproduction, the latter resulting in diapausing egg production. This makes cyclically parthenogenetic clones and their diapausing eggs similar, for instance, to an annual plant and its seeds. In line with this, initiation of sexual reproduction and diapausing egg hatching can be respectively equaled to initiation of plant bloom and seed germination, to get insight on what monogonont rotifer life cycle tells about adaptation to environmental unpredictability. First, we will dissect the approaches in the study of bet hedging, stressing the potential of rotifers as model organisms. Second, we focus closer on the life cycle of a typical monogonont species and its diapause-related traits (hereafter, DRTs; e.g., timing of sex, patterns of hatching of diapausing, sexually produced eggs) in order to elaborate specific theoretical predictions for adaptation to environmental unpredictability. One of our interests here is to highlight the conflicting selection factors acting on DRTs. Third, we review the results from a model study system: the Brachionus plicatilis Müller, 1786, populations in Eastern Spain. We review in greater depth the studies in this model system because it embraces (a) a metric for the environmental unpredictability, which provides an independent variable in statistical analyses, (b) genetic effects on DRTs inferred from measuring phenotypes, and (c) detection of candidate genes for divergence in DRTs. Additionally, studies from these populations include both natural and laboratory-evolved populations. Fourth, we revisit the database of this study system (analyses in Supplementary file) in order to assess the relationships between within-population genetic diversity in DRTs and environmental unpredictability.

Approaches for hypothesis-testing on adaptation to unpredictability

An important difficulty to test bet hedging is to define how unpredictability should be conceived and assessed (Table 1). An epistemic risk exists of distinguishing the level of unpredictability in a habitat from the organism’s traits, hence incurring circularity. Rather, an educated guess is required to adopt the point of view of the focal organism in order to deduct what it can predict. For instance, if the length of the growing season fluctuates, a rotifer diapausing egg can be assumed to have partial information on whether the season would allow the production of a new cohort of diapausing eggs, as for seeds of annual plants. Fortunately, many model predictions in ecology and evolution can be tested in an ordinal way (e.g., the higher the factor X, the higher the trait Y; Levins, 1968), rather than in a strictly quantitative way, so that a proxy for unpredictability may be enough. As another methodological difficulty, regardless of how it is defined for the focal organism, environmental unpredictability is related to departures from a repeated pattern. Therefore, in order to decompose environmental fluctuations in their predictable or unpredictable components, a long-term time series of data is needed. This is the case of recent studies quantifying unpredictability in aquatic habitats. Franch-Gras et al. (2017a) estimated unpredictability in a set of water bodies in Eastern Spain by means of long time series of satellite data (27 years) on water-surface area and several environmental models. Using the length of periods when the ponds had water (hydroperiod length; hydroperiod being a necessary condition for rotifer habitat suitability), their results showed that the studied ponds cover a wide range of predictability. As other instance, Pinceel et al. (2017) used long-term hydrological variation (85 years) in rock pools to calculate the level of habitat uncertainty as the fraction of inundations long enough for reproduction of a cladoceran species.

Table 1 Features in the study of bet-hedging strategies
Full size table

For many species, uncertainty can occur at several time scales. In a recent review, Gilbert (2017a) (1) compiled the evidence for rotifer non-genetic polymorphisms (e.g., spine development in many brachionid species, sexual female production, diapausing egg types, etc.); (2) analyzed if these polymorphisms are due to adaptive plasticity (each genotype produces a single environment-dependent phenotype) or not (each genotype produces more than a single phenotype after controlling for the environment); and (3) associated the former to adaptation to predictable environments and the latter to adaptation to unpredictable environments (diversified bet hedging). Interestingly, Gilbert’s review shows the role of the time scale on adaptive responses. For instance, maternal-age induced spine development in B. calyciflorus Pallas, 1766—which produces daughters with a range of spine lengths and is considered a case of bet hedging—can be interpreted as an anti-predator defense when predator density fluctuates significantly at short time scales, close to the Brachionus lifespan. A similar time scale seems to be relevant for the production of diapausing asexual eggs in Synchaeta—in response to uncertainty in food availability—or cruciform-to-campanulate female transition in Asplanchna—related to uncertainty of prey type. Contrastingly, in many monogononts, the optimization of patterns of sexual female production seems to be better related to fluctuations on a larger scale; i.e., the clonal lifespan and the length of the growing season. We elaborate on the latter below.

Organisms might face certain types of unpredictability successfully by bet heading in multiple traits (Venable & Brown, 1988; Buoro & Carlson, 2014). For instance, when facing uncertain variation in the suitability of a pond, monogononts could hedge their bets at least on two non-exclusive alternatives: (1) diapausing egg hatching or not, and (2) sexual reproduction initiation or not. Therefore, tests on bet hedging yielding negative results when focusing on a single trait may not be conclusive, as bet hedging might occur in other traits with equivalent results. This makes advisable to study a suit of related traits.

The organism features predicted by bet-hedging theory could be recorded at different biological levels. They can be recorded phenotypically without considering the organism genotype. However, if the aim is to test for local adaptation in relation to the degree of environmental unpredictability, controlling for the genotype is needed in order to discard an environmental-induced response (i.e., plasticity). Approaches, as those inspired by using formal quantitative genetics, can take into account the genotype, even if it is not directly measured. Rotifers offer an easy way to correlate phenotypes and genotypes due to their clonal proliferation. An additional approach is also possible by measuring divergence in specific genes affecting traits that might bet hedge, and studying allelic variation in relation to environmental unpredictability. Although promising in the future, this approach is, however, currently a raw approximation due to the complexity of gene function assessment, particularly in organisms that are not genetic models.

In some studies invoking or testing bet hedging, the focus is on a single biological unit (a clade, a population, a species, etc.) that displays a singular feature—for instance, an anti-predator constitutive defense, and this feature is interpreted to be relevant in relation to unpredictability—for the previous instance, predation pressure is thought to be fluctuating in an unpredictable way. The cases revised in Gilbert (2017a) are good instances of this. Alternatively, comparing biological units whose environments differ in their unpredictability can test bet hedging (Simons, 2009, 2011; Graham et al., 2014). If the entities compared are phylogenetically distant, drawing conclusions can be difficult because divergence may affect several features with no relevance in relation to uncertainty. Species and genera with high within-taxon diversification offer an opportunity for this approach.

In order to relate organism features to environmental unpredictability, both observational and experimental (evolutionary) approaches are in principle possible; saying, correlating environmental uncertainty to bet-hedging traits—even if these traits are measured in lab controlled conditions to eliminate maternal and other phenotypic plasticity effects—and recording evolution of putatively bet-hedging traits in experimental environments. The former approach has the well-known drawback of confounding factors due to concomitant variations in the environment. The latter approach might be affected by an experimental setup weakly related to the conditions in the wild with the risk that the adaptation hypothesized did occur in the wild. Ideally, a combination of both approaches would increase the confidence in testing the hypothesis of interest. Nevertheless, in both approaches, laboratory assays might fail in reproducing environmental factors that affect the trait in the field (see for instance Cooper & Kaplan, 1982). Interestingly, rotifers have a set of features that make them suitable models in experimental evolution studies (i.e., rapid adaptive response), as stressed in recent reviews (Fussmann, 2011; Declerck & Papakostas, 2017).

Trade-offs in the adaptation to environmental unpredictability of the monogonont life cycle

In cyclically parthenogenetic rotifers, the recorded sexual reproduction ratios are extremely variable, but commonly are below 30% and almost never reach 100% (e.g., Snell & Boyer, 1988; Carmona et al., 1994, 1995, 2009; Gilbert, 2002; Gilbert & Schröder, 2004, 2007; Gilbert & Diéguez, 2010). This partial investment in sexual reproduction has been interpreted as a diversified bet-hedging strategy, which allows the maintenance of both reproductive modes within a clone (Gilbert, 2003; Fussmann et al., 2007). Hence, diapausing eggs allowing for long-term persistence will be produced concurrently with clonal proliferation by asexual reproduction, as long as the environment remains suitable. Additionally, most studied monogononts do show a population phase with exclusive asexual reproduction before sexual reproduction is induced by the cue of high population density. The involvement of a signal suggests that the optimal timing for sex initiation is partially predictable. Otherwise, the expectation would be for a constant investment in sex. A signal inherent to the population dynamics also suggests that the optimal timing of sex is not periodically constant. However, a notorious exception occurs in a few species where sex initiation has been related to photoperiod (Notommata copeus Ehrenberg, 1834, N. codonella Harring & Myers, 1924, Trichocerca rattus Müller, 1776, and B. rubens Ehrenberg, 1838; Laderman & Gutman, 1963; Pourriot, 1963; Pourriot & Clément 1975; Pourriot et al., 1986). It would be interesting to address whether these species have relatively constant growing seasons in relation to the annual cycle, as suggested by Pourriot et al. (1986) for N. copeus.

When adaptation of DRTs is analyzed, a set of contrasting selective pressures needs to be considered (Table 2), as they usually imply trade-offs affecting the optimal strategy in relation to the unpredictability level. In species in which investment in sexual reproduction is density-dependent, early sex is expected to be the result of a low population density threshold for sex initiation (high propensity for sex), and to evolve as a response to (1) ephemerality or (2) unpredictability of the end of the growing season (Serra & King, 1999; Spencer et al., 2001; see below). In fact, the latter is the consequence of avoiding the fatal effect of an unexpected short growing season (an ephemeral one), when the clone fails to produce diapausing eggs. Early sex incurs the cost of a decreased rate of current, parthenogenetic proliferation. This decrease opens an opportunity for population/genotype invasion by competitors during the current growing season (Montero-Pau & Serra, 2011), and diminishes the rate of male-sexual female encounters (Gilbert, 1963; Snell & Garman, 1986). As a result of these compromises on fitness, the expectation is that populations dwelling in more unpredictable environments will have higher propensity for sex, and will initiate sexual reproduction early in the growing season (Fig. 2).

Table 2 Bet hedging and associated effects in DRTs for cyclically parthenogenetic rotifers with a density-dependent sexual phase
Full size table
Fig. 2
figure 2

Three hypothetical habitats (H1, H2, and H3) with different levels of unpredictability in the length of the growing season. Dark green horizontal bars are periods always suitable; light green horizontal bars are periods when the habitat may become unpredictably unsuitable. If a symmetric distribution of the end of the growing season is assumed, the three habitats have the same average length of the growing season. Predictions are that (1) in the ideal, completely predictable habitat, parthenogenetic proliferation should continue until (short time before) the habitat will become unsuitable, and then all reproduction should be sexual (sexual reproduction ratio: 100%); (2) As the length of the growing season increases in unpredictability (H1 → H2 → H3), sexual reproduction should start earlier in order to neutralize the effects of the shortest growing season, but the investment in sexual reproduction (sexual reproduction ratio) should be lower in order to exploit long growing seasons. This scheme assumes: (1) a constant environment while habitat is suitable (no time-dependence or density-dependence of egg production rate or of egg quality), (2) neglects differences in genetic composition of the eggs produced at different times, and (3) does not consider the role played by the fraction of diapausing eggs that hatch at the onset of the growing season to counterbalance a failure in producing a new cohort of diapausing eggs

Full size image

In some species or clones, stem females are invariably asexual (Gilbert, 2007). The effect of population density on sexual female production is partially blocked during the first generations after the stem female, and this sex-blocking effect gradually disappears in the subsequent generations. This blocking of sex causes a delay in sexual reproduction. The suggested advantage is to allow clonal proliferation for clones hatching late in the growing season, when population density is already high (Serra et al., 2005). However, it might be maladaptive since it may cause failure to produce new diapausing eggs if the growing season is short. Two observations in line with this (revised in Gilbert, 2017a) are (a) sexual females hatched from diapausing eggs in desert-pool Hexarthra (Schröder et al., 2007), and (b) very low threshold density for sex initiation in B. calyciflorus from an ephemeral pond in Patagonia (Gilbert & Diéguez, 2010). In these two examples, the growing season may be dramatically short. Experimental evolution under different hydroperiod regimes (permanent vs. ephemeral) showed consistent results (Smith & Snell, 2012), with earlier timing of sex under the ephemeral selective regime than under the permanent one.

As stated above, intermediate fractions of sexual daughters are frequently observed. The expectation is that, if the end of the growing season is highly unpredictable, the ratio of sexual reproduction should evolve to be relatively low, as a way to exploit the occasional occurrence of long growing seasons (Fig. 2). Computer model simulations focused on the production of diapausing stages support this expectation (Spencer et al., 2001).

In a predictable environment, the hatching of all viable diapausing eggs would be optimal as far as the resulting clones would have enough time to produce during the growing season a number of diapausing eggs higher than the survivorship of the diapausing eggs remained dormant during the same period (see García-Roger et al., 2006 for the selective role of sediment conditions). This expectation neglects the effect of the quality of the diapausing eggs (e.g., the effect of their genetic diversity). In this context, the model that Cohen (1966) developed for annual plants could be applied to monogononts (García-Roger et al., 2014). With the additional assumption that cues for diapausing egg hatching would anticipate good conditions correctly most of the time—otherwise, population persistence would be difficult, then the expectation is that the higher the level of unpredictability, the lower the hatching fraction of diapausing eggs, the leftover fraction remaining for future chances. Nonetheless, realized patterns for diapausing egg hatching can be strongly conditioned by environmental constraints. With respect to habitat type, diapausing eggs in the sediments of shallow ponds, if compared to deep lakes, might be exposed to extreme environmental conditions (e.g., desiccation, compacting pressure while buried in the sediment, high exposure to radiation, damages produced by salt crystals when the pond dries out, predators, etc.), which may accelerate deterioration processes (García-Roger et al., 2006). However, this does not imply higher success in deep lakes, as the microenvironment might make it difficult for diapausing eggs to leave diapause and to return to the active phase. For instance, burial and lack of hatching stimuli in deep lakes would cause low hatching rates, except after re-suspension of the sediments, whereas diapausing eggs in shallow-water sediments are more likely to experience conditions conducive to hatching (Gilbert, 2017b).

Persistence in habitats with unpredictable growing season length could be in principle achieved even if a genotype does not show a bet-hedging strategy (standard meaning; see Introduction) for DRTs, but produces genetically diverse offspring for DRTs (i.e., high genetic variation for DRTs; low maternal effect for DRTs). This could be favored by sex and recombination. Fluctuating selection caused by successive short and long growing seasons could also work to maintain genetic variation in DRTs (Sasaki & Ellner, 1997). The expectation would be that the higher the unpredictability, the higher the frequency of sexual reproduction—in order to promote recombination—and critically the higher the genetic variation for DRTs. However, a high cost of selection (Haldane, 1957) would affect this mechanism causing a disadvantage if compared with standard bet hedging.

A model system for studying bet hedging: B. plicatilis populations in Eastern Spain

The relationship between DRTs and environmental unpredictability has been studied in a rotifer model system consisting of B. plicatilis (sensu stricto) populations dwelling in inland ponds of Eastern Spain. The B. plicatilis species complex has been studied in this region using population approaches for over two decades, and there is information available on these populations (Gómez & Carvalho, 2000; Gómez et al., 2002, 2007; Campillo et al., 2009, 2011a, b; Ciros-Pérez et al., 2001; Ortells et al., 2003; Montero-Pau et al., 2011; Gabaldón et al., 2013, 2015a, b, c, 2017; Serra & Carmona, 1993; Serra & King, 1999; Serra et al., 1998, 2004, 2011; Aparici et al., 1998, 2001, 2002; Carmona et al., 1995, 2009; Gabaldón & Carmona, 2015).

The study region is an endorheic territory constituted by inland ponds located in an area of approximately 800 km2 (38°55.4′ to 38°41.803′N and 1°47.32′ to 1°24.26′W) and characterized by a high level of seasonality and temporal unpredictability at several timescales, typical of the Mediterranean region (Blondel et al., 2010). The climate is semiarid, with an average annual rainfall of ca. 343 mm and mean temperature of approx. 14°C. Dry periods occur between June and September—when temperatures may exceed 40°C—and most precipitation occurs as heavy rains in spring and autumn. The set of ponds and lakes includes water bodies with a wide range of sizes (water-surface area from 0.013 to 119 ha), characterized by being shallow (maximum depth ca. 1 m), non-permanent, and brackish or saline (salinity ranging from 1.8 to 54 g l−1). The biological communities confined in this kind of non-permanent ponds are expected to be adapted and strongly reliant on patterns of pond inundation.

A series of studies in nine ponds inhabited by B. plicatilis have been performed adopting a population approach—i.e., accounting for within-population variation (Franch-Gras et al., 2017a, b). By using long-term time series of water-surface area (see above), Franch-Gras et al. (2017a) have shown that the nine ponds embrace a wide range of unpredictability (metrics following Colwell, 1974). By sampling the egg bank of B. plicatilis from the sediments, Franch-Gras et al. (2017b) founded 270 clones of this species (30 clones per pond). Then, the propensity for sex of each clone was estimated under laboratory conditions, controlling for environmental and maternal effects. Estimates of this DRT were significantly correlated to the level of environmental unpredictability estimated for the pond where the clone was isolated. Results (Fig. 3a) showed the expected pattern of a higher propensity for sex (i.e., earlier initiation of sexual reproduction) in the less predictable environments (Table 2). As unpredictability correlated with the average annual hydroperiod length over ponds, Franch-Gras et al. (2017b) modeled the propensity for sex as due to these two environmental features, and found that unpredictability had a higher explanatory power than hydroperiod length. In contrast, hatching fractions from diapausing eggs from the studied populations—produced and hatched under standardized laboratory conditions—were not significantly correlated to either unpredictability level or hydroperiod length (Fig. 3b).

Fig. 3
figure 3

Relationship between diapause-related traits (DRTs) and environmental predictability in clones of Brachionus plicatilis isolated in the field. a DRT: propensity for sex as indicated by the population density for sex initiation and pond predictability. The population density at first male appearance is a standard proxy of sex initiation (e.g., Carmona et al., 2009; Smith & Snell, 2012). b DRT: diapausing egg hatching fraction. Values in both plots are estimated population means (± 1SE) of each DRT. Dashed lines indicate least-square regression fitting. ‘ns’ stands for non-significant. Data obtained from Franch-Gras et al. (2017b)

Full size image

In a parallel study by Tarazona et al. (2017), the evolutionary dynamics of both the propensity for sex and the diapausing egg hatching fraction were monitored in laboratory populations subjected to two contrasting environmental regimes (treatments: predictable vs. unpredictable) through seven growing cycles interrupted by periods of habitat unsuitability. That is, growing cycles simulate growing seasons, so that a new growing season started after diapausing egg hatching. The predictable regime was simulated in the lab with growing cycles of constant length, whereas the length varied randomly in the growing cycles of the unpredictable regime. Worth to note, both regimes (i.e., either with constant or random length of their growing cycles) imply a fluctuation (suitable vs. unsuitable periods). The two treatments were constrained to have identical average length in their growing cycles. Six multiclonal, genetically diverse, laboratory populations were generated by mixing females from each of the 30 clonal lines per natural population funded by Franch-Gras et al. (2017b). Three out of the six laboratory populations (working as evolutionary replicates) were randomly assigned to each experimental treatment. Laboratory populations showed rapid adaptation (< 77 days) to the predictability of their environments, displaying a divergent response in the DRTs measured in standard conditions, as in Franch-Gras et al. (2017b). Populations subject to the unpredictable regime evolved a higher propensity to sex (i.e., earlier initiation of sexual reproduction) than populations under the predictable regime (Fig. 4a), in agreement with the conservative bet-hedging strategy pointed out by the results of the study in natural populations (Franch-Gras et al., 2017b). In addition, populations in the unpredictable regime also presented lower hatching fractions than those in the predictable one (Fig. 4b), suggesting diversified bet-hedging strategies. Thus, in an unpredictable environment, a higher fraction of the diapausing eggs does not hatch at the beginning of the growing period, so that some diapausing eggs survive in the sediment until the next growing period.

Fig. 4
figure 4

Evolution of diapause-related traits (DRTs) in B. plicatilis laboratory populations along growing cycles in two contrasting fluctuating environments with respect to the length of the growing season: predictable (regular fluctuation pattern) and unpredictable (random fluctuation pattern). a DRT: Propensity for sex as indicated by the population density for sex initiation and selective growing cycle. b DRT: diapausing egg hatching fraction. Values in both plots are estimated population means (± 1SE) of each DRT. Asterisks indicate significance between selective regimes at the corresponding growing cycle. Data obtained from Tarazona et al. (2017)

Full size image

This set of studies showed that unpredictability acts as an efficient selective pressure on rotifer populations in shaping DRTs. On one hand, the findings from these studies are in line with the increasingly recognized capability of rotifers for fast adaptive evolution (e.g., Fussmann et al., 2003; Becks & Agrawal, 2010, 2012; Scheuerl & Stelzer, 2013; Declerck et al., 2015; Haafke et al., 2016; Walczyńska et al., 2017), and with other studies showing that rotifer populations can adapt locally to environmental unpredictability (Schröder, 2005; Campillo et al., 2011a). The low efficiency of a high dispersal capability to result in gene flow (De Meester et al., 2002) has likely a role in the latter. On the other hand, these studies highlight the importance of adaptation to unpredictability. Moreover, these findings demonstrate empirically the existence of bet-hedging strategies in B. plicatilis regarding DRTs. Nevertheless, despite that intermediate hatching fractions were found in all populations, as well as among-population genetic differentiation in relation to this trait, correlational evidence for the expected relationship between unpredictability and hatching fraction of diapausing eggs was not found in natural populations (Franch-Gras et al., 2017b). This could be due to a hypothetical adaptive tracking (Bell & Collins, 2008) of that fraction, so that a row of good years and a row of bad years would, respectively, select for high and low hatching fractions. If so, bet hedging would not have evolved and the sampled genotypes would reflect rather the effect of this adaptive tracking. Actually, results reported by Tarazona et al. (2017) suggest that, in addition to bet-hedging evolution, DRTs track the conditions in the last growing cycle (i.e., as short growing cycles select for a high propensity for sex and a low hatching fraction). Another possible explanation might be that conditions in the sediment affect hatching in an overwhelming way. For instance, if the conditions allowing hatching are spatially variable, they could introduce a noise for the evolution of the optimal hatching fraction. In other terms, due to a large effect of the environmental variance, the heritability of this trait in the field would be very low. Gilbert (2017b) has emphasized the role of the sediment conditions in shaping diapausing egg hatching. In both hypothetical explanations, bet hedging through the propensity for sex would evolve more easily in the wild in order to face unpredictability, and this bet hedging might fulfill the requirements for genotype persistence.

Genome-wide genetic divergence was estimated and analyzed in the same clones on which Franch-Gras et al. (2017b) estimated propensity for sex and diapausing egg hatching fraction (Franch-Gras et al., 2018). These authors used a genome draft assembly obtained from one of the clones, and then genotyping by sequencing was applied to detect single nucleotide polymorphisms in 270 clones from 9 populations (SNPs). The role of diversifying selection (i.e., that acting differently in different localities) on the SNPs was studied by using two methods. One tested whether FST fixation index for each of the 4,543 SNPs genotyped was larger than expected by chance (BayeScan v. 2.1; Foll & Gaggiotti, 2008). This method does not use a priori information on the environment of the populations. It allowed identifying 12 SNPs that are putatively under diversifying selection (i.e., higher FST than expected by random), 11 of them located within 12 genes (Franch-Gras et al., 2018). The other method (BAYENV; Coop et al. 2010) was used to test for significant correlations between (1) genetic differentiation in SNPs and (2) other measured features of the population. The latter included DRTs (propensity for sex and diapausing egg hatching fraction), and three environmental properties (unpredictability level, average hydroperiod length, and due to its implications for B. plicatilis growth—salinity). Consistent with the fact that BAYENV uses specific a priori information, it yielded a higher number of SNPs candidate to be under selection than BayeScan. The number of SNPs significantly correlated with the studied features of populations ranged from 34 to 39, depending on each feature (Franch-Gras et al., 2018). Interestingly, six SNPs (associated to six genes) were correlated to both the propensity for sex and the environmental unpredictability, but with no other population characteristic included in this study. Only a single SNP correlated simultaneously with the diapausing egg hatching fraction and the environmental unpredictability, but with no other population characteristic included in this study. The genes including SNPs identified under selection by BayeScan or BAYENV had a broad range of gene ontologies (Franch-Gras et al., 2018). Genes associated to sex propensity and environmental predictability were enriched for genes involved in binding functions and those genes associated to hatching fraction were enriched in signal transducer activity. No enrichment was found in those genes associated to hydroperiod length.

The study system used in this set of papers can be used to test additional predictions included in Table 2 (i.e., bet-hedging effects on the hatching fraction; relationship between sexual reproduction ratio and unpredictability). This research is in progress in the Laboratory of Evolutionary Ecology at the University of Valencia.

While not intended for this aim, the databases collected by Franch-Gras et al. (2017b) and Tarazona et al. (2017) can be revisited in order to assess for correlations between environmental unpredictability and genetic variation in DRTs (see Supplementary file). The expectation is higher genetic variance for DRTs in populations with higher unpredictability (Table 2). Such relationship is not supported by our analysis on field populations, but suggested for both propensity for sex and diapausing egg hatching fraction in evolutionary experiments (Fig. S1). Admittedly, the disagreement between both approaches could be due to low statistical power, which would be further diminished in more uncontrolled conditions (i.e., in natural populations). Alternatively, the disagreement might be due to the different selection costs acting on natural and laboratory populations, being higher in the former due to the trade-offs acting in a complex environment. If so, the disagreement stresses that, even if experimental evolution shows an evolutionary potential and can unmask causal factors, other factors acting in the wild could counterbalance the selective strength of the focal factor.

Final remarks

Theory of adaptation to environmental unpredictability is well developed in general terms (Cohen, 1966; Hairston & Munns, 1984; Seger & Brockmann 1987; Philippi & Seger 1989; Ellner, 1997; Hairston 1998; Lenormand et al., 2009; Simons, 2011), but specific hypotheses need to be derived from it in order to make predictions, namely bet hedging, testable (Childs et al., 2010; Simons, 2011). This and, particularly, practical and conceptual difficulties in estimating predictability have implications for the scarce empirical evidence supporting theory. Consistent with the unpredictability in their environments, monogonont rotifers show a range of traits susceptible to evolve bet-hedging strategies. Of most interest, these animals, even when belonging to the same species, inhabit sites embracing a broad variation in the degree of environmental predictability, so that they can be used to assess whether bet hedging has actually evolved. This assessment provides post hoc, sound explanations for traits and variations in traits (e.g., Gilbert, 2017a). The works summarized above also show that monogonont populations can be used to formulate a priori predictions and to test them, after developing a metric to quantify the environmental unpredictability of their habitats. As rotifers are small sized, measurements in controlled laboratory conditions can be performed for naturally occurring genotypes, and evolution can be studied experimentally. Rapid growth and short generation times allow to study rapid evolutionary responses (Fussmann, 2011; Declerck & Papakostas, 2017; Tarazona et al., 2017) and to get comprehensive time series of data over many generations within short experimental times (e.g., Serra et al., 2011).

By combining field and laboratory populations, on one side, and demographic, evolutionary, and molecular ecological research, on the other, the studies performed using B. plicatilis populations from Eastern Spain have shown that environmental unpredictability in growing season length acts as an effective selective pressure shaping diapause-related traits. These studies fit in an increasingly used paradigm in ecological and evolutionary research, where genes map on phenotypes that map on environments. That first mapping is still very tentative, due to the lack of information and complexity of the gene function, particularly for organisms that are not genetic models.

Besides the agreement with bet-hedging theory found in natural populations of B. plicatilis, some expectations were not met. This is not surprising if a single bet-hedging trait could fulfill the adaptation required to face unpredictability, and it stresses the importance of measuring several traits with similar adaptive value. Moreover, heritability—the fuel for selection and evolution—for some traits might be very low in the field, due to high micro-environmental variance. This might be the case for the hatching fraction of diapausing egg, where conditions experienced by the eggs in the sediment are expected to be very variable. Hence, caution is required before inferring potential for evolution when heritability is measured in laboratory conditions, where most of the environmental variance has been removed. The studies on bet hedging in B. plicatilis showed contrasting results between correlational and experimental evidence. We advocate for experimental evolution approaches in order to isolate the effects of a single selective factor, namely unpredictability regime, as they incorporate complexity not considered in mathematical models, helping to bridge theory and the natural world. However, we advocate also to complement the experimental results with hypothesis-driven field observations.

Increased genetic variance in DRTs seems not to act as a bet-hedging strategy for unpredictability in the length of the growing season. Nevertheless, adaptation to unpredictability by bet hedging seems to act simultaneously to adaptive tracking. It is quite unavoidable that a coincidental row of long growing seasons would not select for the optimal phenotype for such a row, given the fast evolutionary capability of these animals, rather than for a variable regime, then erasing the long-term optimization achievable by bet hedging. Thus, the expectable mixture of bet hedging and adaptive tracking helps to interpret the results revised here. At this regard, it can be argued that the fast evolution to a high propensity for sex observed in an experimental regime of varying length of simulated growing seasons could be, in fact, determined by a single, randomly occurring short growing season. Actually, in a conservative bet hedging, this is what is expected: the effect of a bad condition in shaping a trait. Here, the critical observation is that the high propensity for sex did not revert during the simulated long growing seasons in the experimental evolution approach.

Organisms face adaptation constrained by a suite of trade-offs. In the case of rotifers, when this suite is considered, additional hypotheses arise and call for their testing; for instance, whether unpredictability relates to the proportion of sexual daughters once sex has initiated, or affects sex-blocking effect; these are pending studies. Additionally, the toolbox of genetic and molecular methods available for non-model organisms in genetics is increasing and becoming cheaper, so that population studies using these methods are practicable and affordable. New problems and new techniques will broaden the avenue of using rotifers as model organisms in testing evolutionary and ecological theories, as for bet-hedging theory addressed in this paper.