Prolonged dormancy interacts with senescence for two perennial herbs


Correspondence author. E-mail:


  1. Senescence is characterized by a decline in survival, fecundity and reproductive value with age among adult individuals. Simple age-dependent life cycles progress forward through developmental stages, with each successive stage being characterized by age-specific vital rates. In contrast, size- or stage-based life cycles for perennial plants are more complex and often include stasis and retrogression to previous vegetative or reproductive life stages, indicating possible slowing or even reversing the developmental progress.
  2. Many plants remain in nonemergent, below-ground stages during the growing season (prolonged dormancy), which may affect the process of senescence. Stasis in the dormant stage implies that senescence is interrupted while plants are below-ground.
  3. We explored the underlying assumptions of size- or stage-based life cycle graphs and developed four different demographic models for how prolonged dormancy may mediate the relationship between age and vital rates. We then tested these models using more than 20 years of demographic data on 2 perennial herbs, Astragalus scaphoides and Silene spaldingii.
  4. Results from model fitting suggest that prolonged dormancy interacts with the age dependence of vital rates. The model using true biological age (time since germination) of emergent and dormant plants to estimate vital rates was never the best model for our data. For both species, the model assuming that dormancy resets plants to the same postdormant state experienced earlier in life independent of their predormant age resulted in the best fit, though not for every vital rate.
  5. Older Astragalus plants had declining annual survival probabilities and reproductive value, suggesting senescence. Silene showed the opposite pattern for reproductive value that increased with age, indicating negative senescence.
  6. Synthesis. Using long-term demographic data from two perennial herbs, this study shows mixed evidence for senescence in perennial plants. Our results indicate that prolonged dormancy interacts with the age dependence of vital rates and may sometimes retard the process of senescence.


In evolutionary ecology, senescence refers to a persistent decline in the age-specific vital rates (e.g. survival and fecundity) of an individual due to constraints arising from intrinsic physiological and developmental processes (cf. Charlesworth 1980; Rose 1991; Watkinson 1992). Similarly, reproductive value, which is a function of survival and fecundity, is predicted to decline with age as a sign of senescence (Partridge & Barton 1996). This decline in vital rates with age is expected to be the result of weaker natural selection on performance at later ages (Fisher 1930; Hamilton 1966; Charlesworth 1980). The declining force of natural selection with age may either lead to the accumulation of deleterious mutations that only act later in the life cycle (Medawar 1952; Charlesworth 1980), or antagonistic pleiotropic genes that have favourable effects on vital rates early in life but adverse effects later in life (Williams 1957; Charlesworth 1980; Rose 1982).

In contrast to ageing, which simply refers to the advancement of age over time (cf. Watkinson 1992; Roach 1993) and is a necessity to any life history, the occurrence of senescence has been questioned for many plants (Williams 1957; Watkinson & White 1986; Watkinson 1992; Roach 1993). For a large number of long-lived perennials, vital rates and reproductive value remain constant or even slightly increase with age (Silvertown, Franco & Perez-Ishiwara 2001; García, Dahlgren & Ehrlén 2011), indicating negative senescence (sensu Vaupel et al. 2004), although some long-lived species seem to senesce (Silvertown, Franco & Perez-Ishiwara 2001; Baudisch et al. 2013).

Many herbaceous perennials have dormant stages in which plants may remain below-ground for one or more years without resprouting (reviewed in Lesica & Steele 1994; Shefferson 2009; Reintal et al. 2010). This prolonged dormancy (vegetative dormancy sensu Shefferson 2009) may be beneficial particularly in stochastic environments (Shefferson 2009; Gremer, Crone & Lesica 2012), providing a way to escape unfavourable conditions (Shefferson, Kull & Tali 2005). To date, however, it is not known how prolonged dormancy is related to the processes of plant senescence, although this question is highly relevant to demographic studies of species exhibiting prolonged dormancy. For other taxa, particularly insects, dormancy is sometimes associated with retarded senescence because of changes in stress response or metabolism (Tatar, Chien & Priest 2001; Tatar & Yin 2001).

Life cycle graphs form a basis for demographic models, such as matrix population models, which are used to quantify population dynamics and the fitness of individuals representing different phenotypes (Caswell 2001). These life cycle graphs are based on demographic transitions between different age, size or developmental stage classes from time t to time + 1 (Caswell 2001). Demographic studies of senescence are traditionally based on only age-structured models of the life cycle (but see Caswell & Salguero-Gómez 2013). In these models, age represents the organism's physiological and developmental progress towards maturity and possible senescence later in the life, and sexual or asexual reproduction is the only mechanism that resets this progress (Caswell 2001). However, size- or stage-based classification is commonly used for iteroparous perennials (Caswell 2001), where prolonged dormancy has often been described as a single stage (e.g. Knight 2003; Lesica & Crone 2007; García, Goñi & Guzmán 2010; but see Miller, Antos & Allen 2007, 2012; Jäkäläniemi et al. 2011; Gremer, Crone & Lesica 2012). The construction of such models is either based on biological intuition or an evaluation of which plant characteristics best predict plant performance (e.g. Jäkäläniemi et al. 2011; Gremer, Crone & Lesica 2012). Although these size- or stage-based models are useful for many purposes, they may be of limited value when pursuing questions regarding how dormancy affects senescence (but see Caswell 2012). Size- and stage-based models with a single dormant stage suggest that once dormant individuals emerge, they exhibit the same vital rates as individuals that did not go dormant. These models thus ignore the past history of individuals, although it may be important for plant population dynamics (Ehrlén 2000). More importantly, these models implicitly assume that the process of senescence is not affected by the dormant stage. These assumptions, however, may not be reasonable depending on the costs and benefits of dormancy to individual performance (Miller, Antos & Allen 2007, 2012; Shefferson 2009; Jäkäläniemi et al. 2011; Gremer, Crone & Lesica 2012).

Here, we examine the relationship between prolonged dormancy (hereafter referred to as dormancy) and plant senescence. We start by exploring theoretically the underlying assumptions of size- or stage-based life cycle graphs regarding senescence and develop four different demographic models for how prolonged dormancy may interact with intrinsic processes that eventually may lead to senescence. We then test these assumptions and models empirically using a long-term data set from two perennial herbs each covering a period of >20 years in which we could identify the true age of plants in the study. Using these demographic data and identifying which model best fits the data, we not only examine whether our data set is consistent with the signs of senescence in old age, but also identify which demographic models are most appropriate for life cycles of plants with prolonged dormancy. Our model-based approach provides a novel method to study how dormancy may affect the senescence process in herbaceous perennial plants.

Description of demographic models for plants with dormancy

Demographic studies of perennial plants are often based on life cycle graphs similar to the one depicted in Fig. 1, where dormancy is modelled as a single stage (D in Fig. 1). In the life cycle graph, stasis (individuals staying in the same stage, not transferring to subsequent or previous stages) accounts for dormant periods longer than one time interval and emergence is modelled as several vegetative and reproductive stages (E1–E5 in Fig. 1). Transitions between emergent and dormant stages often imply that plants may return to stages experienced earlier in life (retrogression) and indicate that in theory plants have the potential to transition indefinitely among all or a subset of stages included in the life cycle graph. The occurrence of stasis and retrogression more or less exclude senescence as a possible process for late life performance (Pedersen 1999; Salguero-Gómez & Casper 2010).

Figure 1.

Life cycle graph traditionally adopted in demographic studies of plants with prolonged dormancy (adapted from Jäkäläniemi et al. 2011). Nodes E1–E5 are emergent stages and D is a single dormant stage. Transitions representing reproduction while plants are emergent are not included.

At the same time, demographic models used so far have been based on implicit assumptions on how dormancy affects and is affected by the plant's developmental and physiological state (status). For example, the life cycle graph in Fig. 1 assumes that (i) entering dormancy resets plants to the same state in terms of vital rates (i.e. survival, the probability of remaining dormant and the probability of emerging do not depend on the plant's state before entering dormancy; cf. Jäkäläniemi et al. 2011); (ii) vital rates do not vary throughout the dormant period when dormancy is prolonged over two or more growing seasons; (iii) The performance of emergent plants after a period of dormancy does not depend on the length of the dormant period or the state before entering dormancy.

For the purpose of this study, we developed four alternative demographic models, each of which relaxed some of the above-mentioned assumptions regarding the effects of dormancy. Each model is based on explicit assumptions on how dormancy interacts with age, resulting in alternative predictors of vital rates over time (Table 1). When testing the demographic models, we used these alternative predictors for four conditional, annual vital rates: survival, dormancy conditioned on survival, flowering probability conditioned on emergence and seed production conditioned on emergence. Table 1 exemplifies the calculation of the predictors. Each model corresponds to a life cycle graph that is presented in Fig. 2 together with the associated transition coefficients. Transition coefficients are formally defined in the presentation of each model below, while the vital rates (survival and dormancy conditioned on survival) are expressed in terms of transition coefficients in Table S1 in Supporting Information.

Table 1. Calculation of state variables (x = age (years since germination), = the number of growing seasons as emergent since birth, x’ = years since last period of dormancy, = the number of dormant growing seasons) in four demographic models (true age, sleeping beauty, reboot and active dormancy) used to study patterns of senescence in plants with prolonged dormancy. The table shows an example based on a hypothetical plant life history described by the plant's age and history of emergent (E) and dormant (D) growing seasons
Example plant life historyTrue ageSleeping beautyRebootActive dormancy
AgeE/D x y x x’ y z
1E111 10
2E222 20
3E333 30
4D43  31
5D53  32
6E64 142
7E75 252
8E86 362
9D96  63
10D106  64
11E117 174
12E128 284
13E139 394
Figure 2.

Life cycle graphs with predictors of plant performance (x = age (years since germination), = the number of growing seasons as emergent since birth, x’ = years since last period of dormancy, = the number of dormant growing seasons) and transition coefficients (p= probability of being emergent, b= probability of entering dormancy, a= probability of leaving dormancy, d= probability of remaining dormant) corresponding to the (a) true age, (b) sleeping beauty, (c) reboot, (d) active dormancy, and (e) age-independent models. Transitions representing reproduction are not included. Numbers within nodes in (d) denote plant age (z). Grey nodes are emergent stages, while black nodes are dormant stages. See text for further details.

True Age Model

In the true age model, we relax the assumption that dormancy resets plants to the same state. The basic assumption is that dormancy does not interfere with developmental or other processes that may lead to senescence. The true age model assumes that age (x) – or time since germination (hereafter birth) together with the state of the plant at age x (i.e. whether the plant is dormant (Dx) or emergent (Ex)) describe the state and performance of the plant during the next time interval (Fig. 2a). Thus, prior history (i.e. the sequence of emergent and dormant periods up to age x) does not affect performance. This could be the case, for instance, for one of our study species, Silene spaldingii, where dormant individuals have similar survival and growth to vegetative individuals (Lesica & Crone 2007).

Under the true age model, transition coefficients are defined as follows (cf. Fig. 2a): p= P(Ex+1|Ex) is the probability of being emergent at age + 1 conditioned on emergence at age x, b= P(Dx+1|Ex) is the probability of entering dormancy at age + 1, a= P(Ex+1|Dx) is the probability of leaving dormancy at age + 1, and d= P(Dx+1|Dx) is the probability of remaining dormant at age + 1.

For example, the plant in Table 1 has been dormant for a total of 4 growing seasons by the age of 13 years old, and the state variable (x) simply corresponds with the true biological age of the plant because dormancy does not interfere with developmental processes that may lead to senescence. In this model, a persistent decline in annual survival probability or reproduction with x at large values of x, for both emergent and dormant states, is consistent with senescence.

Sleeping Beauty Model

In the sleeping beauty model, we relax the assumptions that dormancy resets plants to the same developmental and physiological state (status), and that the performance of emergent plants after a period of dormancy does not depend on the state before entering dormancy. The basic assumption of this model is that dormancy temporarily halts developmental or other processes that may lead to senescence. Thus, the number of growing seasons that the plant has been emergent since birth (y) and whether the plant is dormant (Dy) or emergent (Ey) after y emergent growing seasons together describe the state of the plant and determine the plant's performance during the next time interval (Fig. 2b).

Under the sleeping beauty model, transition coefficients are defined as follows (cf. Fig. 2b): p= P(Ey+1|Ey) is the probability of being emergent in the next growing season (entering emergent state Ey+1) conditioned on being in emergent state Ey, b= P(Dy|Ey) is the probability of entering dormancy after y emergent growing seasons, a= P(Ey+1|Dy) and d= P(Dy|Dy) are the probabilities of leaving dormancy and of remaining dormant when the dormant period was entered after emergent growing seasons.

In Table 1, the plant is emergent for the first three years and is then dormant for the following two years. The state variable (y) of the sleeping beauty model differs from the true age model as it remains constant for ages 3–5 (= 3) and consequently, only the number of emergent years matters in this model. For instance, plant performance might positively depend on resources and biomass accumulated during the emergent years (y), while dormancy as such would neither imply any metabolic maintenance costs of below-ground organs nor resource benefits mediated by, for example root-colonizing mycorrhizal fungal symbionts (see Shefferson 2009; Jäkäläniemi et al. 2011 for discussion). In contrast, a persistent decline in the performance of emergent plants with increasing y is consistent with senescence, and dormancy just delays (in terms of age) the onset and progress of this decline.

Reboot Model

In the reboot model, we retain the assumption that entering dormancy resets plants to the same state experienced earlier in life in terms of vital rates. However, we relax the assumption that performance in the emergent state does not vary with age. This model thus assumes that dormancy resets developmental or other processes that may lead to senescence to the same state after dormancy. A sign of rebooting might be, for instance, the increased flowering probability of Ophrys sphegodes individuals after dormancy (Hutchings 1987). Age (x) describes the state of the plant in the predormant period before plants first enter dormancy, while time since last period of dormancy (x′) describes the state of the emergent plant in postdormant periods (Fig. 2c). In Table 1, the reboot model assumes that the plant first enters dormancy at the age of = 3, and after dormancy periods the plant is reset to the stage x′ = 1.

Under the reboot model, transition coefficients are defined as follows (cf. Fig. 2c): p= P(Ex+1|Ex) is the probability of being emergent at age + 1 conditioned on emergence at age x and no dormant periods before age x, bP(D|Ex) is the probability of entering dormancy for the first time at age + 1, a = math formula is the probability of leaving dormancy, d = P(D|D) is the probability of remaining dormant, math formula and math formula are the probabilities of emergence and dormancy x′ + 1 time units as last dormant period conditioned on emergence at x′.

The reboot model is the only framework considered here where transitions among mature stages form loops that allow plants to return to stages experienced earlier in life (i.e. rejuvenation). This feature deviates from the concept of a unidirectional life cycle where the production of new individuals is the only process that resets development. In the reboot model, dormancy may repeatedly reset the mature plant to the same stage. Even though performance may decline over predormant and postdormant stages (i.e. with increasing x and x′), the reboot model opens a possibility for plants to slow-down or even escape senescence by repeated rebooting through dormancy.

Active Dormancy Model

In the active dormancy model, dormancy may interact with developmental and physiological processes that may lead to senescence either by speeding up or slowing down such processes. All assumptions concerning dormancy in traditional size- or stage-based life cycle models for this type of plants are to some degree relaxed. Dormancy does not reset plants to the same state, the performance of emergent plants after a period of dormancy may depend on the state before entering dormancy and the length of the dormant period, and plants may senesce while dormant. Senescence during dormant periods may occur due to respiratory or other physiological costs (Gremer, Crone & Lesica 2012). This model assumes that the number of dormant (z) as well as emergent (xz) growing seasons experienced until age x, and whether the plant is dormant (Dy,z) or emergent (Ey,z) together describe the state of the plant at age x and determine the plant's performance during the next time interval (Fig. 2d). In Table 1, the plant has spent nine years as emergent and four years as dormant over its whole life span of 13 years. In contrast to the true age and sleeping beauty models, the active dormancy model incorporates accumulated effects of both dormant and emergent periods.

Under the active dormancy model, transition coefficients are defined as follows (cf. Fig. 2d): py,z = P(Ey+1,z|Ey,z) is the probability of being emergent at stage + 1, z conditioned on emergence at stage y,z, by,z = P(Dy+1,z|Ey,z) is the probability of entering dormancy when emergent at stage y,z, ay,z = P(Ey+1,z|Dy,z)is the probability of leaving dormancy when dormant at stage y,z and dy,z = P(Dy+1,z|Dy,z) is the probability of remaining dormant at stage y,+ 1 conditioned on dormancy at stage y,z.

Materials and methods

Study Species

Astragalus scaphoides (Fabaceae) is an iteroparous, perennial herb with a taproot surmounted by a branched caudex. Rhizomes and other means of vegetative propagation are absent. Reproductive individuals are 20–50 cm high and have an inflorescence composed of one or more racemes (Barneby 1964; Crone & Lesica 2006). Once established, plants alternate between vegetative, flowering and dormant states throughout their lives (see Fig. S1). If plants emerge above-ground, they initiate growth in April, flower from late May to mid-June, and die back to perennating roots in July. Using data from Lesica (1995), expected average life span conditional on reaching the flowering stage is estimated to be 21 years, (Ehrlén & Lehtilä 2002), and median age to first reproduction is 3 years (Lesica 1995). High flower and seed production have a tendency to occur in alternate years (Lesica 1995; Crone & Lesica 2004, 2006; Crone, Polansky & Lesica 2005). The proportion of plants displaying prolonged dormancy ranged from 1% to 23% with a mean of 10% between 1987 and 1992 (Lesica 1995) and a mean of 14% between 1989–2007 (Gremer, Crone & Lesica 2012). In our population, about 70% and 17% of prolonged dormancy bouts were for 1 or 2 growing seasons, respectively (Lesica & Steele 1994; Gremer 2010). Associations with mycorrhizal partners seem weak (E. Crone, H. Addy & M. Rilling, unpublished).

Silene spaldingii (Caryophyllaceae) is an iteroparous, perennial, geophytic herb with vegetative or flowering stem(s) arising from a caudex surmounting a long taproot (Hitchcock & Maguire 1947). The species reproduces only by seed and reproductive individuals are 20–40 cm high with 3–20 flowers borne in a branched, terminal inflorescence. Plants in their first or second year usually form only a rosette, while older, nonreproductive plants produce 1 or 2 sterile stems 1–10 cm high without rosettes. Rosettes transition only to the rosette or dormant stages, otherwise plants alternate between vegetative, flowering and dormant states throughout their lives (Lesica & Crone 2007; Fig. S1). A total of 23 of 100 individuals recorded on the first census 25 years ago were still alive in 2011, indicating that this perennial herb is relatively long-lived (average life span not known). Vegetative and flowering plants emerge in mid- to late-May and wither in September. Flowers bloom in July and set seed in August, the first reproduction usually occurs during the first four years. Seeds germinate in spring but some germination may occur in fall (Lesica 1993). Bouts of prolonged dormancy, in which plants remain below-ground during the growing season, are mainly for one (77%) or two (18%) seasons (Lesica & Steele 1994; Lesica 1997; Lesica & Crone 2007). Silene has probably no associations with mycorrhizal fungi (P. Lesica, pers observation).

Study Sites and Field Methods

We followed the fate of Astragalus individuals (note that plant size was not explicitly measured) in a population in southern Beaverhead County, Montana (45°06.4'N, 113°02.8'W, altitude 1920 m a.s.l.) in a sagebrush (Artemisia tridentata) steppe. Two permanent one-metre wide belt transects were established at the study site in 1986 following methods outlined in Lesica (1987, 1995). A total of 897 plants were monitored, of which 515 were born during the study period and included in analyses.

We followed the fate of Silene individuals at The Nature Conservancy's Dancing Prairie Preserve in Lincoln County, Montana, USA (46°56.8'N, 115°04.5'W, altitude 825 m a.s.l.), where Silene occurs in a mesic Festuca scabrella dominated grassland at the bottom of shallow swales and on cool slope exposures with relatively deep soil (Lesica 1997). We monitored a total of 180 Silene individuals annually for 25 years, of which 76 were born during the study period and included in analyses. We recorded the fate of all individuals mapped in 1-meter wide, permanent belt transects in mid-July when flowers were present following methods outlined by Lesica (1987, 1997).

Prolonged dormancy can be inferred by following the fate of marked or mapped individuals for numerous consecutive years. Unambiguous bouts of dormancy were recorded only when they were both preceded and succeeded by a visible stage within the study period (Lesica & Steele 1994; Kery & Gregg 2004; Lesica & Crone 2007). Because dormancy bouts for both species tend to be one or two years, we excluded the data from the first and last two years of each study to be able to distinguish between dormancy and recruitment at the beginning of the study, and between dormancy and mortality at the end of the study (Lesica 1995). Our analyses include only plants that emerged for the first time after 1988 for Astragalus and after 1989 for Silene so that they could be aged exactly.

Data Analyses

From the population size (N), we calculated the finite rate of increase λ = Nt+1/Nt for successive years t and + 1, and the long-term population growth rate as the geometric mean of annual growth rates. The study population of Astragalus has remained rather stable over the years; the geometric mean finite rate of increase (λgeom) was 1.05 in 1990–1999 (= 170–290) and 1.02 in 2003–2010 (= nearly 300 individuals in 2010). The Silene population remained stable in 1987–1994 with λgeom = 1.03 (N = 100–125), but has since then declined with λgeom = 0.92 from 1995 to 2011 (= 41 in 2011).

Fitting the alternative demographic models and model selection

To empirically test the assumptions of the four alternative demographic models (the true age model, the sleeping beauty model, the reboot model, the active dormancy model) with data, we calculated four separate, annual vital rates: survival; the probability of dormancy, conditioned on survival; the probability of flowering, conditioned on emergence (where emergence is the opposite of dormancy); and the number of seed pods produced, conditioned on flowering for both study species. We fit each of the four models to the annual vital rates using GLMMs (R Development Core Team 2011) with linear and quadratic terms for age (defined differently for each model, see Table 1). For the reboot model, we estimated separate parameters for age before and after the first bout of dormancy (with age always reset to 1 after each bout of dormancy). For the active dormancy model, we separately estimated the number of dormant growing seasons since ‘birth’ (z in Table 1) as well as the interaction between these terms. Because dormant plants are identified as those that re-emerge, by definition dormant plants have perfect survival. In addition, we included an age-independent model which had constant vital rates assuming that age or prior history do not affect plant performance and consequently, plants do not senesce (see Fig. 2e). In other words, this age-independent model is a simplified version of the stage-based model commonly used in demographic studies for plant species with prolonged dormancy (Fig. 1). All models included random effects of individual plant and year. Models for survival and dormancy used a binomial distribution with a logit link function, while models for the number of pods used a Poisson distribution with a log link function. We then compared the fit of these alternative models to the data using an Akaike Information Criterion (AICs, Burnham & Anderson 2002).

As the active dormancy model is described as the number of emergent (y) and dormant (z) growing seasons (Table 1), we transformed the model to be expressed in terms of true age (x) and the number of dormant growing seasons experienced (y) by replacing y with xz. For both species, we here present three vital rates (survival, dormancy conditioned on survival, seed pod production conditioned on emergence) derived from the fitted active dormancy model as a function of age and dormancy (Fig. 3).

Figure 3.

The relationships of the true age (x) of an individual and years of life spent in dormancy (z) for different annual vital rates; survival of emergent plants (a, e), probability that an emergent (b, f) or dormant (c, g) plant is dormant during the next growing season conditioned on survival, and expected number of seed pods conditioned on emergence (d, h) for Astragalus scaphoides and Silene spaldingii. These relationships result from substituting the number of emergent growing seasons (y) with xz as a predictor in the active dormancy model after fitting the model to demographic data. White region indicates impossible combinations of age and years in dormancy. The number of seed pods is divided by 10 in (h).

Calculation of age-specific reproductive value

To calculate reproductive values for the four demographic models (Fig. 2), we first calculated transition coefficients from the fixed effect coefficients of the GLMMs described previously. Calculations were based on the relationships between the response variables in the GLMMs and transition coefficients given in Table S1. We followed the approach in Cochran & Ellner (1992) and used the transition coefficients of each model to calculate the following age-based vital rates for ages 1–20; age-specific survival probability (note that this is not the same as the survival function; cf. Charlesworth 1980, p. 3 and p. 9), age-specific probability of dormancy and expected seed pod production conditioned on survival (Fig. S2–3). We estimated 95% confidence intervals for all vital rates from parametric bootstrap samples obtained by repeating the calculations for the fixed effect coefficients 1000 times. Simulated coefficients were randomly drawn from a multinomial distribution with parameters equal to the GLMM estimates of fixed effects, their standard errors and correlations.

For each demographic model, we then calculated age-specific reproductive value (Partridge & Barton 1996) for both species based on the following assumptions and procedures: Maximum age for both species was set to 80 years to produce robust predictions (using e.g. the mean life expectancy would produce biased estimates, while using the maximum age > 80 years would not virtually change the current predictions). For the reboot model, age-specific survival and reproduction for age classes beyond the observed data were estimated assuming that the maximum predormant age was 20 years. This approach was, however, not possible for the other three models that do not contain loops where dormancy resets plants to an earlier stage. For these models, age-specific survival was estimated by extrapolating the survival function (Charlesworth 1980), while seed pod production was assumed to be constant after the age of 25 as we had no data on older individuals. Population growth rate was estimated as λgeom from 1990 to 2010 for Astragalus and from 1987 to 2011 for Silene. Recruitment per seed pod was then estimated from Euler's equation (Stearns 1992) assuming the number of recruits per seed pod to be independent of the age of the mother plant. As this assumption is neutral with respect to senescence, we do not unintentionally introduce any artefactual age-specific patterns to the reproductive values.


Alternative Demographic Models

Of all the demographic models for both species, the model using true biological age to estimate annual vital rates was least successful (Tables 2 and 3), indicating that dormancy interacts with age dependence of vital rates. In Astragalus, the reboot model was most successful as it provided the best fit for annual survival and dormancy rates and the number of seed pods conditioned on flowering, but the active dormancy model also fit well to annual dormancy rate (Table 2). The sleeping beauty model best estimated annual flowering probability (Table 2).

Table 2. Model selection for age-specific annual vital rates (e.g. survival from year t to year + 1) for Astragalus scaphoides. In flowering/emergence ≡ 1 – dormancy, both conditional on survival. AIC = model fit, models fitting equally well (dAIC < 2) are in bold
SurvivalTrue age61803
Sleeping beauty61801
Reboot 9 1791
Active dormancy91803
Age independent41806
Dormancy|SurvivalTrue age62284
Sleeping beauty62290
Reboot 9 2281
Active dormancy 6 2281
Age independent42294
Flowering|Emergence/survivalTrue age51281
Sleeping beauty 5 1269
Active dormancy81274
Age independent31314
Seed pods|FloweringTrue age51480
Sleeping beauty51473
Reboot 8 1431
Active dormancy81470
Age independent31481
Table 3. Model selection for age-specific annual vital rates (e.g. survival from year t to year + 1) for Silene spaldingii. In flowering/emergence ≡ 1 – dormancy, both conditional on survival. AIC = model fit, models fitting equally well (dAIC < 2) are in bold
SurvivalTrue age6325
Sleeping beauty6326
Reboot 9 323
Active dormancy9330
Age independent 4 323
Dormancy|SurvivalTrue age6523
Sleeping beauty6526
Reboot 9 520
Active dormancy9525
Age independent6523
Flowering|Emergence/survivalTrue age5368
Sleeping beauty5367
Reboot 9 364
Active dormancy8368
Age independent 3 365
Seed pods|Flowering True age 4 197
Sleeping beauty4199
Active dormancy 7 196
Age independent2221

In Silene, survival was best described by the age independent model, followed by the reboot model (Table 3). The reboot model provided the best fit for annual dormancy rate and flowering frequency, although the age-independent model also fit well for flowering (Table 3). The active dormancy model estimated best the number of seed pods conditioned on flowering, but the true age model fit nearly equally well (Table 3).

Annual Vital Rates as a Function of Age and Dormancy

Vital rates derived from the active dormancy model revealed that annual survival was generally high (80–90%) for Astragalus, but slightly declined among old plants (Fig. 3a). Annual survival also declined with the number of years spent in dormancy earlier in life (Fig. 3a). Dormancy rate increased with age only for plants that had spent some years below-ground, whereas the opposite was true for plants that did not spend many years below-ground (Fig. 3b,c). Seed pod production conditioned on emergence generally increased with age, but declined with dormancy (Fig. 3d).

In Silene, annual survival somewhat decreased with age and dormancy (Fig. 3e). Dormancy rate increased with age and the number of years spent in dormancy (Fig. 3f,g). Similar to Astragalus, seed pod production conditioned on emergence increased with age for Silene, but declined with dormancy (Fig. 3h).

Age-Specific Reproductive Values

For Astragalus, all four demographic models (true age, sleeping beauty, reboot and active dormancy) predicted a decline in reproductive value late in life (Fig. 4), indicating senescence. However, all models also produced large confidence intervals for reproductive value (Fig. 4), meaning that the estimated values contain high uncertainty particularly for older individuals. Unlike Astragalus, the demographic models predicted reproductive value to increase with age for Silene (Fig. 4), indicating negative senescence that is often described as a decline in mortality and an increase in fecundity of adult individuals with age.

Figure 4.

Age-specific reproductive value of Astragalus scaphoides and Silene spaldingii calculated for 4 different demographic models from GLMMs, and 95% confidence intervals calculated from a parametric bootstrap with 1000 iterations (see methods for details). For the reboot model of Silene, confidence intervals increase exponentially and therefore only the first 15 years are presented.


Size- or stage-classified demographic models containing stasis and retrogression can implicitly overlook senescence processes (Salguero-Gómez & Casper 2010). Here, we empirically examined the underlying assumptions of such models using four different demographic models for how prolonged dormancy may interact with senescence. The fitted models to the long-term demographic data of two herbaceous perennial species, Astragalus scaphoides and Silene spaldingii, revealed that the model using true biological age (time since germination) was not successful in estimating the effects of dormancy on vital rates in this study. This may not be that surprising, given that dormant plants do not have access to photosynthates as do emergent plants, and hence, the differential dormancy history of two plants of the same age is likely to cause differences in their performances (Shefferson 2009). From this perspective, it is a greater surprise that the sleeping beauty model, which assumes dormancy as an inactive stage to slow senescence, was not more successful. The sleeping beauty model fit best in one case only (flowering probability in Silene), and one might therefore question whether the nonemergent plants actually are ‘dormant’ in the strict sense of the word.

The reboot model which assumes that dormancy resets individuals (independent of their age) to a certain state experienced earlier in life (state 1 in the present study; Table 1) provided the best fit for five of eight vital rate tests. This finding suggests that senescence is retarded or interrupted by dormancy. Negligible or even retarded senescence during dormancy has been previously reported at least for insects as a consequence of altered stress response or metabolism (Tatar, Chien & Priest 2001; Tatar & Yin 2001). The same might be true also for perennial plants as several studies have suggested that dormancy may buffer individual plants from stress encountered above-ground (Morrow & Olfelt 2003; Shefferson, Kull & Tali 2005; Shefferson 2009; Shefferson et al. 2012; Gremer & Sala in press). While metabolism in dormant plants has not been directly quantified, a reduction in metabolic rates during dormancy may delay the process of senescence. In a study on orchids, Hutchings (1987) suggested that longer periods of time spent in dormancy seemed to ‘reinvigorate’ plants, and he documented increased flowering rates in years following dormancy. Together with our results, these studies suggest that prolonged dormancy may slow down the process of senescence in perennial herbs. On the other hand, the active dormancy model, which allows senescence to be either sped up or slowed down during dormancy, was the best model in one case (seed pods in Silene) and fit well also in another case (dormancy in Astragalus), proposing that dormancy does not always retard senescence in plants. It should be noticed that the fits of the demographic models used here are slightly biased towards young individuals as they are more abundant than old individuals in the data sets for both study species. The scarcity of old individuals is a common problem in studies of senescence for long-lived species (Roach 2004) and may also affect the generality of conclusions. It might also be possible that the suggested relationship between dormancy and senescence is purely an artefact caused by, for instance, environmental factors not controlled here.

Declining annual survival rates and reproductive values among old Astragalus individuals are potential signs of plant senescence. Similar to Astragalus, many perennial plant species do express signs of senescence. Based on the review of survivorship patterns for 44 herbaceous and 21 woody species, most species (55%) showed a decline or the lowest survival at older ages, although an asymptotic decline seemed to be more common for herbaceous than woody species (Silvertown, Franco & Perez-Ishiwara 2001). However, in contrast to Astragalus, Silene did not show signs of senescence in the present study in terms of reproductive value that for this species increased with age, indicating negative senescence. Silene does not seem to be exceptional as a number of studies have reported vital rates to increase with age to a certain point in the plant life cycle, then reach a plateau and remain constant the rest of the life (e.g. Ehlers & Olesen 2004; Horvitz & Tuljapurkar 2008), and some studies have reported even a continued increase in vital rates with age (e.g. García, Dahlgren & Ehrlén 2011).

It has been suggested that prolonged dormancy in plants may be a symptom of senescence (Shefferson 2009). Dormancy in Astragalus and Silene did in fact increase with age among plants that spent more years below-ground, supporting the idea of senescence. However, this obviously is not the whole story because dormancy is common also among young Astragalus and Silene plants (Fig. 3) despite the fact that our assumption of perfect survival for dormant individuals may have underestimated the probability of dormancy for emergent plants. We therefore conclude that dormancy interferes with age dependence in vital rates. However, based on the present results, no such general conclusions can be drawn about the possible interdependences of dormancy and senescence. For both species, dormancy was associated with reduced annual survival and seed pod production, which might be related to a poor physiological status of individuals (Shefferson 2009), but could also reflect other physiological costs associated with dormancy. For instance, Gremer, Sala & Crone (2010) suggested that plants may remobilize structural carbon into nonstructural carbohydrates while dormant, which may be costly.

Our analyses revealed that dormancy interacts with age dependence of vital rates, especially through rebooting. This rebooting is not surprising as life histories generally include single-celled ‘bottlenecks’ that start a life of any individual. For plants, rebooting can physiologically occur from meristems that maintain their functionality and may remain indeterminate regardless of plant age, enabling the development of a new module (Peñuelas & Munné-Bosch 2010). However, this is not sufficient to explain why spending one or two growing seasons as dormant reboots plants, while dormancy during winter does not. A possible mechanistic explanation for rebooting might be the one suggested by Lesica & Crone (2007). They speculated that the performance of Silene individuals might be limited by nutrients (phosphorous or nitrogen) and that plants might increase the reserves of the limiting nutrient during dormancy. Therefore, if successive emergent growing seasons exhaust reserves (possibly causing senescence), while dormancy refills them, one arrives at a mechanism that corresponds well with rebooting. To better understand the function of prolonged dormancy and its potential connections to senescence (if any), it would be necessary to quantify metabolic processes (e.g. mycorrhizal transport) occurring during dormancy in individual plants.

In the present study, we have focused on senescence, which is an age-based process, and have therefore examined annual vital rates in relation to different age-based demographic models. However, the vital rates and demographic fates of perennial plants are often better predicted by the individual's size rather than their age (Lacey 1986; Caswell 2001), although the age-independent model fits best only for the annual survival of Silene. For studies on senescence, age is obviously essential (Charlesworth 1980; Roach 2004). As current demographic tools and techniques enable the modelling of vital rates as a function of multiple variables (Ellner & Rees 2006; Caswell 2012), the size and age of individuals together with environmental variables can be simultaneously included in a demographic model to address eco-evolutionary questions (see e.g. Childs et al. 2003). The modelling framework developed here provides a novel way to combine theory and long-term data sets of age- and state-dependent vital rates of long-lived perennial plants and can be used to further explore their contrasting senescence patterns. Our approach differs from those referred to above because instead of asking which plant variables predict best or covary with plant performance, we start with explicit assumptions about how prolonged dormancy interferes with senescence. For each set of assumptions, we construct a life cycle model that includes the transitions predicted. For instance, the assumption of no interference between dormancy and senescence is addressed in the true age model, and retrogression is possible only in the reboot model. Finally, we fit the models to long-term demographic data to evaluate their alternative assumptions of dormancy and senescence. This modelling approach may well provide new insights into plant senescence as well as prolonged dormancy and could be utilized in comparative studies on senescence when long-term demographic data of plants accumulate.


We thank the numerous field assistants for data collection over the years, Kari Sivertsen for preparing the figures, and Richard Shefferson and three reviewers for their constructive comments. J.T. thanks White Rabbit and Jefferson Airplane, S.R. thanks the Emil Aaltonen Foundation for the support during the project.