Study species

We chose eight species, including four invasive species: Leucanthemum vulgare Lam. (oxeye daisy), Centaurea stoebe L. ssp. micranthos (Gugler) Hayek (spotted knapweed; née C. stoebe L.), Leonurus cardiaca L. (common motherwort) and Potentilla recta L. (sulphur cinquefoil), and four native species: Heterotheca villosa (Pursh) Shinners (hairy false golden aster), Gaillardia aristata Pursh (common gaillardia), Agastache urticifolia (Benth.) Kuntze (nettleleaf giant hyssop) and Potentilla arguta Pursh (tall cinquefoil). All seeds were collected from natural grasslands in western Montana. The distributions of these species can overlap, but we selected pairs of target species that generally occur in habitats that differ in soil moisture: C. stoebe and H. villosa occur primarily in more xeric habitats, L. vulgare, P. recta, G. aristata and P. arguta primarily in mesic habitats, and L. cardiaca and A. urticifolia generally in very wet habitats (Table 1).

Experimental design

The experiment was conducted in a glasshouse on the campus of the University of Montana, Missoula, Montana. Glasshouse temperatures were maintained at 15–30 °C, corresponding roughly to natural summer temperatures in the region. Natural light was supplemented by metal halide bulbs and maximum total photosynthetically active radiation during the day reached 1200 μmol · m⁻² s⁻¹. Seeds of all species were sown in plastic trays (54.2 × 27.3 cm in width and 6.5 cm in height) in January 2010. One week after seedling emergence, individual seedlings were transplanted into pots (7 × 7 cm in width and 20.6 cm in height) filled with a 1:1 mixture of top garden soil and sterile silica sand. Forty days after transplanting, before the first round of treatments were applied, the longest leaf of each plant was measured as an estimate of the initial size of each individual. A split plot design was implemented with the first round of treatments as a main factor, and the second round of treatments and species as subfactors. For both the first and second rounds, there were three treatments: inundation, moderate watering and drought. At the end of the first round, a subgroup of plants from each of the early treatments was harvested to obtain information on their initial responses to the three water treatments [early responses (ER)] and as a reference for calculating their relative growth after a subsequent treatment. The remaining plants from each of the three early-treatment groups [early inundation (EI), early moderate watering (EM) and early drought (ED)] were divided into three subgroups, each of which was later exposed to one of the same three treatments [late inundation (LI), late moderate watering (LM) and late drought (LD)] in the second round (Fig. 1). For each treatment combination, ten individuals of each species were used, with one individual per pot, giving 10 individuals × 8 species × 3 early treatments × 3 late treatments + 10 ×8 × 3 (individuals with early treatments only) = 960 pots in total.

Six identical tanks were used to create inundation, drought and moderate water conditions, with two tanks assigned to each treatment. Tanks were 161 × 91.3 cm in width and 8.5 cm in height, lined with heavy plastic and fit with drains to regulate the maximum water depth. For the inundation treatment, the water level was maintained at 7 cm depth above the bottom of the tank, and approximately 10 cm below the surface of the soil in the pots. These pots were watered to saturation from the top every day. There was no standing water in the moderate and drought treatment tanks. In the moderate treatment, pots were watered to saturation every other day, while those in the drought treatment were watered to saturation once or twice per week in an effort to create very dry conditions, but prevent high mortality rates. The first round of treatments lasted for 70 days before plants in each treatment were divided and assigned to one of the three different conditions in the second round or harvested and measured to evaluate their early responses (Fig. 2). The duration of the second round of treatments was another 70 days, and glasshouse and hydrological conditions were as similar as possible to the first round. Thus, plants receiving both rounds of treatments were harvested and measured 140 days after the beginning of the first round. Height and leaf number were measured and then separated into roots and shoots at harvest, dried at 60 °C for 2 days and weighed.

Data collection and statistical analyses

We calculated mortality rates for all treatment combinations. A total of 728 individuals survived to the end of the experiment and these were used for further analyses. All five traits: total plant biomass (TM), height (HE), leaf number (LN), shoot mass (SM) and root mass (RM), were used to assess the performance of species in each treatment. For each of the five traits of each species, relative growth (RG) for all early–late treatment combinations was calculated according to the formula:

RG=Y−XX,

where X is the mean trait value of 10 individuals of the early response group, and Y is the mean trait value in the late treatment of 10 individuals that had the same early treatment as the early response group. For example, to calculate the RG of total biomass in late drought for a species with early inundation treatment, X is the mean total mass of individuals harvested after early inundation and Y is the mean value of individuals in the late drought after early inundation. For each species in each treatment combination, RGs of the four traits: height, leaf number, shoot mass and root mass, were combined into a single composite measure of relative growth (RG_C) by averaging the RGs of the four traits. Thus, we used four response variables in evaluating the performance of each species: mortality rate, total biomass, relative growth in biomass (RG_M) and composite relative growth (RG_C).

For any individual or group of species, early inundation and drought treatments can affect their subsequent responses in different ways and/or to different extents. To address this for a given trait, we defined ‘difference variables (‘Diff-’)’, as the difference in the mean trait value from the control (moderate early treatment) due to an early inundation or drought treatment:

Diff‐Y=Y2−Y1,

where Diff-Y was the difference in late performance between individuals with early inundation or drought and those with early moderate treatment for a trait of a species, Y₁ was the mean value of the trait for the species in a late condition after early moderate treatment (control) and Y₂ was its mean trait value in the same late condition after early inundation or drought. For example, to calculate the Diff-TM due to early inundation (EI) for a species in late drought, Y₁ is the mean total biomass of individuals in late drought after the early moderate treatment, and Y₂ is the mean total biomass of individuals in late drought after early inundation.

Mean values of height, leaf number, shoot mass, root mass and total plant biomass were log-transformed to minimize variance heterogeneity. The r² from models for anova or ancova on mortality, the five measured traits, relative growth in biomass and composite relative growth were higher with species as a variable than models with habitat type and exotic status as variables. Therefore, the results of three-way anova or ancova for effects of the first round of treatments, the second round of treatments, species and their interactions for all traits were used for analyses of mean values. The five measured traits were analysed with three-way ancova with the initial size of individuals nested within species as a covariate, and mortality and two measures of relative growth were analysed with three-way anova. All dependent variables were then analysed with one-way anova or ancova to evaluate the effects of the early treatments for each species within each of three late treatments, and differences among the late treatments and among different species (for the two measures of relative growth) within each early treatment. Since there were not sufficient degrees of freedom to analyse the effects of individual species on these variables, we compared the degree of effects of early experiences, differences in mortality, total biomass, relative growth in biomass and composite relative growth (Diff-MR, Diff-TM, Diff-RG_M and Diff-RG_C, respectively) due to early treatments with four-way anova, with the first and second rounds, habitat type, exotic status and their interactions as effects. The LSD method was used for multiple comparisons on mean values for mortality, total biomass, relative growth in mass and composite relative growth and the four Diff- variables among first-round treatments, among second-round treatments and among species (or among different habitat types and between invasive and native species for differences of variables), across all and within each of the other treatments.

Mortality

Across all species and treatments, mean mortality rate in the first round was 8.6%, significantly lower than 20.7%, which occurred in the second round (P = 0.007; Tables 2 and 3; Fig. 3). The overall mortality during late drought was 36.3%, nearly two times higher than in the late inundation and moderate treatments (P < 0.001; Fig. 3, Table 3). Two rounds of drought increased the overall mortality of all species to 47.5%, compared with 28.8% in late drought after early moderate conditions (P = 0.029), while no effects of early inundation on later mortality were found. Species from xeric habitats showed a decrease in mortality in late inundation (P < 0.001) and an increase in mortality in late moderate conditions after early drought (P < 0.05), compared with that after early moderate conditions (Fig. 3). Species from hydric habitats had higher mortality than other species under late moderate conditions after moderate experience (P < 0.01).

Total biomass

For plants harvested after the first round, the mean total biomass of all species was significantly lower under drought than under moderate and inundation conditions (P < 0.001; Fig. 4). Compared with moderate conditions, drought reduced the biomass of five of the eight species (the exceptions were G. aristata, P. arguta and L. cardiac; P < 0.05), whereas inundation only reduced the biomass of A. urticifolia (P = 0.02), indicating the treatment of inundation was much less stressful than drought at the early stage.

For plants that received both rounds of treatments, there were highly significant effects of both the first- and second-round treatments on all measured variables, with strong interactions of species with the first- and second-round treatments (Table 2 and Table S1 in Supporting Information). However, there were few significant interactions between the first- and second-round treatments. Across all species and all late treatments, the mean biomass after early drought was 0.31 g, 43.5% lower than after early moderate conditions (0.54 g, P < 0.001), but there was no effect of early inundation (P = 0.221; Table 3 and Fig. 4). For individual species, early drought decreased final biomass in many more cases than did early inundation (Fig. 4).

The effects of late treatments on biomass were more species specific than those of early treatments, as indicated by greater significance of second-round × species interactions than first-round × species interactions (Table 2). Overall, the mean biomass after late drought was significantly lower than after late inundation and moderate conditions (P < 0.05, Table 3). Biomass was greatest under continuous (both rounds of) moderate conditions (F = 6.20; d.f. = 2, 172; P = 0.003; Table 3), especially for P. recta and C. stoebe (Fig. 4). After early inundation, negative responses of biomass to late drought were found for more species than after any other early treatment (P < 0.05), whereas early moderate or drought treatments resulted in a negative response to late inundation for only C. stoebe (P < 0.01 and P = 0.01 for early moderate and drought experience, respectively).

Relative growth

For both relative growth in biomass (RG_M) and the composite measure of relative growth (RG_C, mean relative growth of all the other four traits), and across all species, the effects of the second round of treatments were much stronger than the first round (Table 2). Both measures of relative growth, RG_M and RG_C, were significantly lower under late drought than in more favourable late conditions (P < 0.01; Table 3 and Figs S2 and 5). For the effects of the first-round treatments, early drought improved relative biomass growth in late inundation only when compared to the early moderate treatment (+99.7% vs. +42.4%, P = 0.037; Table 3 and Fig. S2a). Across all late conditions, both early inundation and drought increased the mean composite measure of growth, RG_C, by 29.7% and 32.4%, respectively, relative to +11.3% of the control (P = 0.017 and P = 0.006; Table 3).

Significant differences in growth among species were found between the two fastest growing species (P. arguta and H. villosa) and the other species (Figs S2b and 5). There were also differences in growth between L. vulgare and P. recta versus the remaining species for composite relative growth (Fig. 5). For individual species, differences in relative growth among the first round treatments were found only for species from hydric and xeric habitats (Fig. S2c; P < 0.05), and there was little difference among the second-round treatments (Fig. S2d). For the composite measure of relative growth, interactions between species and treatment were much stronger for the first than for the second round (Table 2). Species from wet and xeric habitats were more likely to show improved RG_C than species from mesic habitats (Fig. 5). After early inundation, C. stoebe's RG_C in late inundation and drought was +15.9% and +20.1%, higher than the control [−26.9% (P = 0.065) and −38.9% (P = 0.006), respectively]. RG_Cs of L. cardiaca and A. urticifolia were +47.0% and −4.1%, respectively, across all late treatments, 217.5% and 91.8% higher than controls (−40.0% and −49.7%, P < 0.05). Early inundation reduced the composite relative growth of P. recta in the late moderate treatment (P = 0.032). Similarly, effects of early drought were positive for A. urticifolia in all late treatments, as it increased the average RG_C from −49.7% to −8.6% (P < 0.05). However, early drought also had some adverse effects: it decreased RG_C of P. recta in late drought by 257.8% (from −11.90% to −42.60%; P = 0.043).

Changes due to early treatments

Significant effects of early, late treatments and habitat type were found for all Diff- values except mortality (Tables 4, S3 and Fig. S3). There were significant interactions between late treatments and invasiveness for both Diff-RG_M and Diff-RG_C (Tables 4 and S3). For composite relative growth, there were significant interactions between early treatment and habitat type, late treatments and habitat type, and early treatments and invasiveness. Among the early treatments, we found a significant difference between effects of inundation and drought only for Diff-TM. Early inundation resulted in an average increase in biomass (Diff-TM = 0.06) across all species and all late treatments, significantly higher than the average decrease in biomass due to early drought (Diff-TM =ucodep>−0.24, P = 0.031).

Comparing species from different habitats in the late treatments, hydric species showed an increase in relative growth after early inundation relative to the control, with a Diff-RG_M = 1.02 and Diff-RG_C = 0.66 for L. cardiaca and A. urticifolia, respectively, which was higher than those of xeric and mesic species [averaged Diff-RG_M = 0.06 (P < 0.05) and Diff-RG_C = −0.01 (P < 0.001); Table S3]. Xeric species demonstrated a greater increase in composite relative growth after the early drought treatment (average of 0.32) than after early inundation (average of 0.02) across all late treatments (P = 0.030). For species from mesic habitats, differences in composite growth due to early inundation were higher for late inundation and drought treatments than for late moderate conditions (0.13 and 0.29 vs. −0.35, P < 0.001 in both cases). For invasive versus native species across all late treatments, the beneficial effect of early inundation on composite growth was greater for invaders than for natives (0.44 vs. 0.04, P < 0.001), but the beneficial effect of early drought was greater for natives (0.41 vs. 0.10, P = 0.003).

Effects of early treatments on later performance

Improved tolerance for environmental stress after previous exposure to stress has been called the ‘priming effect’ (Tanou, Fotopoulos & Molassiotis 2012) or ‘stress memory’ (Walter et al. 2013) and has been documented for drought and inundation in previous studies (Li et al. 2011; Onate, Blanc & Munne-Bosch 2011). But there has been little discussion of how the occurrence of these two stresses at different points in time during plant growth might influence performance (Miao, Zou & Breshears 2009). Our results are the first experimental evidence that either kind of experience early in the life can improve a plant's performance to both forms of stress at a later stage. This is consistent with the theoretical and empirical studies on synergetic effect of multiple stressors (Folt et al. 1999; Coors & Meester 2008; Crain, Kroeker & Halpern 2008; Piggott, Townsend & Matthaei 2015). Thus, as philosopher Fredrick Nietzsche said, ‘that which does not kill us, makes us stronger’ may apply to plants.

Part of the explanation for improved performance under second-round inundation or drought after either early inundation or drought may be that some of the mechanisms for tolerance to drought and inundation are similar (Lopez & Kursar 2003). For example, an ethylene-response-factor-like gene Sub1A that confers submergence tolerance to rice (Xu et al. 2006) also improves survival when rice is exposed to rapid dehydration (Fukao, Yeung & Bailey-Serres 2011). The submergence-inducible gene Sub1A may endow plants with the ability to overcome prolonged submergence, oxidative damage and rapid dehydration that often occur during recovering from submergence, as well as water deficit caused by drought (Fukao, Yeung & Bailey-Serres 2011). ABA, the effects of which can be augmented by Sub1A, is a key signalling molecule that coordinates water balance, expression of stress-inducible genes and metabolic adjustment under water-deficient conditions (Zhu 2002; Seki et al. 2007). Dehydration is also a component of a transient submergence event (Setter, Bhekasut & Greenway 2010), and exogenous ABA increases survival of oxygen deprivation in plants (Hwang & Vantoai 1991; Ellis, Dennis & Peacock 1999). Therefore, the levels of ABA may also act as a positive regulator for submergence tolerance as well as for drought tolerance. The shared mechanisms for tolerance to inundation and drought stress also observed in physiological and anatomical traits (Parolin et al. 2010). For example, both inundation and drought can induce stomatal closure (Parolin 2001; Elcan & Pezeshki 2002), reduce leaf gas exchange and decrease leaf transpiration and carbon assimilation (Parolin 2001; Baraloto et al. 2007). Adventitious roots, aerenchyma and leathery xeromorphic leaves of trees in Amazonian floodplains are commonly recognized as adaptation to inundation (Parolin 2001), but they may also alleviate the effects of drought.

Plants may have the ability to compensate for harsh conditions early in their lives through increased growth later, just as plants can compensate for herbivory, sometimes performing as well as or better than plants that were not grazed (McNaughton 1979; Paige & Whitham 1987; Agrawal 2000). Damage results in lower immediate growth, but sometimes promotes faster growth afterwards, enabling damaged plants to catch up. An increase in relative growth rate after damage or prior extreme experience can also be an allometric effect: the relative growth rate of a plant decreases as it grows. An event that makes or keeps plants smaller can therefore increase a plant's relative growth rates afterwards.

Metaplasticity associated with species’ habitats

Our species from hydric, mesic or xeric habitats responded to the experience of early inundation or drought in ways that were generally consistent with the hydrology of the habitats in which they occur. For example, gas exchange rates in flood-tolerant species usually recover rapidly following a inundation event (Kozlowski 1982; Elcan & Pezeshki 2002). Here, species from hydric habitats had the greatest increase in relative growth after early experience with inundation, although early drought was not worse for them than for other species. Similarly, species from xeric habitats showed greater tolerance for drought after exposure to early drought than did other species, consistent with other studies of drought priming (Onate, Blanc & Munne-Bosch 2011; Walter et al. 2011; Wang et al. 2014). Enhanced performance by species from both more extreme habitats later under the same or different stressful conditions, relative to species from more moderate habitats, suggests the hypothesis that plants may evolve a general tolerance for various forms of stress.

Repeated exposure can lead to the evolution of a more stress-tolerant genotype (Chapin, Autumn & Pugnaire 1993), which may show reduced performance under more favourable conditions. We did not find a significant decrease in relative growth under later moderate conditions for species from hydric or xeric habitats after early drought or inundation. Evidence for such a cost was only found in one species from a mesic environment (P. recta) after exposure to inundation. This could be because (i) a cost of acclimation to early stress may be more likely to occur in species without a history of adaptation to extreme environments; or (ii) that the cost of responses to early stress may be small and difficult to detect (Sultan & Spencer 2002; Weijschedé et al. 2006; Auld, Agrawal & Relyea 2010); or (iii) the decreased relative growth of P. recta in late drought after early drought may simply be due to damage from a longer period of continuous drought, which had more adverse effects than inundation.

Long-term adaptation to benign habitats may lead to the evolution of genotypes with reduced ability to tolerate extreme events. According to this hypothesis, species from mesic habitats, especially P. recta, are less able to cope with more extreme environmental events. The ability of a species to develop metaplasticity in response to prior extreme experience may be closely linked to its tolerance for such experience. It is to be expected that the habitat in which a species occurs plays an important role in determining metaplasticity in response to early experience. This could be a useful focus for future studies on variation in plasticity and metaplasticity.

The ways by which different species adjust their later plasticity in response to prior experience may play an important role in ecotypic differentiation (Bradshaw & Hardwick 1989; Palacio-López & Gianoli 2011). Metaplasticity may be as important as genetic differentiation for the expansion of a species (Williams & Black 1993; Parker, Rodriguez & Loik 2003) and allow species to respond to a wide range of environments and tolerate novel stresses during invasion. After expansion, the maintenance of a high degree of plasticity throughout a plant's lifetime may be very costly. Assuming that the costs of plasticity increase with the degree of plasticity (DeWitt, Sih & Wilson 1998; Givnish 2002; Auld, Agrawal & Relyea 2010), natural selection in response to environmental heterogeneity may favour greater variability in the plasticity of a specific trait (but see (Pigliucci, Murren & Schlichting 2006; Crispo 2007). The complexity and high operational costs of plasticity suggest the involvement of higher-order control and coordination (Novoplansky 2009). Such flexibility in plasticity may allow plants to produce adaptive plastic responses rapidly when needed, while minimizing costs of maintaining broad plasticity to the degree possible. Flexibility in plasticity may therefore be as important for species’ acclimation to temporal heterogeneity in environmental conditions as the degree of plasticity at one point in time.

Metaplasticity of invasive versus native species

Our results show that plastic responses can be changed by previous environmental experiences, and suggest that the ability of species to regulate their plasticity is related to the habitats in which they occur, and whether they are invasive or native species. Invasive species have been hypothesized to be more plastic than natives, but we found little evidence for this. Contrary to some predictions (Marshall & Jain 1968; Daehler 2003) and some empirical results (Davidson et al. 2011), our four invasive species were not different than their four native counterparts in their responses to water conditions, as found in other studies (Brock, Weinig & Galen 2005; Funk 2008; Palacio-López & Gianoli 2011). Plasticity is highly trait dependent, and a high degree of plasticity in one trait may be associated with low plasticity in other traits (Valladares et al. 2000). This suggests that integrated measures of plant performance may be more important for understanding the role of plasticity in invasion than quantifying the degree of plasticity (or metaplasticity) in individual traits (Hulme 2008). Based on our composite measure of relative growth, the invasive species benefitted from early inundation more than native species did, whereas native species benefited more from early drought. Comparison of a larger number of native and invasive species is needed to draw stronger conclusions.

Adaptive metaplasticity versus constraints and damage

Since A.D. Bradshaw's landmark paper over a half century ago (Bradshaw 1965), there have been extensive discussions and debates concerning the relevant traits, measurement, evolutionary and ecological significance of phenotypic plasticity, as well its costs and mechanisms. It may be premature to theorize about these same issues with respect to the higher-order phenomenon of metaplasticity (Novoplansky 2009). Still, the current study suggests some theoretical considerations:

1. It is important to distinguish between adaptive behaviours and those that are inevitable results of plant growth form, unavoidable constraints or damage. If a plant suffers from phosphorus deficiency, its leaves will be purple, and it may have difficulty dealing with additional stresses. It does not seem useful to interpret this as adaptive plasticity/metaplasticity. Similarly, plant growth is allometric, and a plant's relative growth rate decreases with size. Therefore, any treatment that reduces plant growth but has no other negative effects will result in a higher relative growth under better environmental conditions later when compared to plants that did not suffer from reduced growth earlier (Weiner 2004). This is better interpreted in terms of allometry than as metaplasticity.
2. When should plants evolve metaplasticity rather than simple plasticity? Assuming that there is a cost of plasticity and that this cost increases with the breadth of plastic responses of which the plant is capable, we would expect metaplasticity to evolve when specific environmental conditions are predictably associated with specific later environmental conditions. Under such a scenario, plants could reduce the cost of future broad plasticity by maintaining more limited plasticity for the predicted environment. The life span of the plant relative to the temporal variation in its environment becomes important here. We would expect metaplasticity to evolve when the lifespan of a plant is long enough for it to experience different environmental conditions during its life, but not if its lifespan covers many such changes in different directions (unless the metaplasticity is temporary). Thus, metaplasticity would be expected to evolve in short-lived plants, but not those with the very shortest lifespan, for example desert annuals.