Causes and consequences of prolonged dormancy for an iteroparous geophyte, Silene spaldingii

Authors


*Author to whom correspondence should be addressed: P. Lesica. Fax: +1 406 243 4184. E-mail: peter.lesica@mso.umt.edu.

Summary

  • 1Prolonged dormancy, during which a plant does not produce above-ground shoots during one or more growing seasons, is common in temperate herbaceous species, but its role in life history is poorly understood.
  • 2We analysed stage transitions to determine the consequences of prolonged dormancy for Silene spaldingii, a perennial geophyte of semi-arid grasslands in the Columbia Basin of western North America. We monitored 179 S. spaldingii plants from 1987 through 2005, and analysed demographic rates as a function of plant state and seasonal precipitation.
  • 3Dormant plants had similar survival and growth to vegetative plants, and were more likely to flower in the following year, resulting in a greater reproductive value being associated with the dormant state. Thus, prolonged dormancy is likely to increase fitness in S. spaldingii.
  • 4Both external (precipitation) and internal (stage in previous year) factors were associated with S. spaldingii vital rates. Prolonged dormancy was more likely following flowering than after being vegetative, and following a wetter summer and/or drier fall the previous year.
  • 5Our results suggest that geophytic species may respond to current resource availability in a manner dependent on internal state (previous performance) and suggest that prolonged dormancy plays a role in resource allocation and life history.

Introduction

Prolonged dormancy, failure to produce above-ground shoots during the growing season for one or more consecutive years, has been well documented for orchids (Kull 2002) and observed in members of at least eight plant families (Lesica & Steele 1994), suggesting it is widespread among species that do not maintain growing points at or above the soil surface during the regular dormant season (geophytes). In most cases, 25% or less of a population is dormant at any one time, but in some cases, over half the population can be dormant in some years (Lesica & Steele 1994). The duration of dormancy bouts is usually 1 or 2 years but it can be as long as 5 or more years. The function of prolonged dormancy in plant life history is poorly understood.

Studies of prolonged dormancy suggest an association with environmental stress (Lesica & Steele 1994). Drought is commonly reported to induce prolonged dormancy (Epling & Lewis 1952; Thomas 1981; Tyler & Borchert 2002; Kery et al. 2005). Inundation (Oostermeijer et al. 1992), fire (Abrahamson & Hartnett 1990; Keeley 1991), and herbivory (Morrow & Olfelt 2003) are other factors reported to promote prolonged dormancy. Associations with stressful conditions suggest that prolonged dormancy is at least partly induced by a shortage of resources such as water, nutrients or energy, although it is not clear whether plants avoid stressful conditions through prolonged dormancy or are constrained to be dormant by low resource availability. The occurrence of prolonged dormancy may also depend on a plant's internal state, as indicated by former performance. For example, prolonged dormancy was less likely following flowering in the orchids Ophrys sphegodes (Hutchings 1987) and Cypripedium calceolus (Shefferson et al. 2003). Both Calvo (1990) and Mehrhoff (1989) reported that smaller plants of the orchids Cyclopogon cranioides and Isotria medeoloides were more likely than larger plants to display prolonged dormancy.

At first glance it seems unlikely that prolonged dormancy is an adaptive life-history trait because plants forgo reproduction as well as photosynthesis and the resulting energy gain for 1 or more years. If prolonged dormancy is solely a reflection of resource constraint, dormancy should be followed by increased mortality, reduced size and/or reduced fecundity relative to plants that made leaves in previous years (Shefferson et al. 2003), and a greater frequency of dormancy would reflect lower individual fitness.

Alternatively, plants may benefit from dormancy. Dormant plants may absorb water passively or nutrients from microbial symbionts (Morrow & Olfelt 2003; Hultine et al. 2004) without incurring the cost of producing above-ground structures. Prolonged dormancy could also be a ‘strategy’ for escaping stressful conditions such as above-ground herbivory (Morrow & Olfelt 2003) or drought, similar to the dormant-seed strategy often observed in desert annuals (Went 1949; Schaffer & Gadgil 1975). In these cases dormant plants should have higher survival and/or emerge as larger or more fecund plants than those that made leaves the previous year.

Understanding the role of prolonged dormancy in plant life history and how this role varies among different species and environments may allow a better understanding of resource allocation in geophytes. Prolonged dormancy may play a role in cyclical reproduction (Lesica 1997; Crone & Lesica 2004) as well as survival in semi-arid environments. By comparing vital rates of vegetative and dormant plants, we can begin to evaluate the role of prolonged dormancy in plant life history. In this paper, we examine the function of prolonged dormancy in the life history of Silene spaldingii, an herbaceous perennial geophyte of semi-arid grasslands. We ask whether there are costs or benefits to prolonged dormancy in terms of growth, survival and reproduction. We use demographic models to infer whether dormancy, on average, results in a net cost or benefit for S. spaldingii relative to producing leaves in years when plants do not flower. We then evaluate the effect of precipitation and internal state (stage in previous year) on annual probabilities of being dormant, vegetative, or flowering.

Methods

study species

Silene spaldingii (Caryophyllaceae) is an iteroparous, perennial herb with one or few vegetative or flowering stems arising from a caudex surmounting a long taproot without rhizomes or other means of vegetative propagation (Hitchcock & Maguire 1947). The number of flowers per plant is set prior to initiation of flowering. 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, non-reproductive plants produce one or two sterile stems 1–10 cm high without rosettes. A plant's flowering stage is larger than its vegetative stage, which is larger than the rosette (Fig. 1). All rosettes are new recruits, but fall germinants may first appear in the mid-July census as vegetative plants (P. Lesica, personal observations). Rosettes transitioned only to the rosette or dormant stages. Otherwise, plants alternate between vegetative, flowering and dormant states throughout their lives. Vegetative and flowering plants emerge in mid- to late-May and senesce in September. Flowers bloom in July and set seed in August. Most flowering plants produce at least one fruit, even in the absence of pollinators (Lesica 1993). Seeds can germinate with as little as 4 weeks of cold treatment, so some germination may occur in fall as well as in spring (Lesica 1993). Seeds lack a hardened seed coat (Hitchcock & Maguire 1947), so a long-term seed bank is unlikely. Prolonged dormancy of germinated plants is common, with bouts of dormancy usually lasting 1–2 years (Lesica 1997). Silene spaldingii is endemic to the Palouse region of south-east Washington and adjacent Oregon and Idaho, and is disjunct in north-west Montana and British Columbia (Hitchcock & Maguire 1947). The climate of S. spaldingii habitat is semi-arid; mean July and January temperatures for our Montana study site were 19 °C and –5 °C, respectively (NOAA 2005). Mean (± SD) annual precipitation was 373 (± 93) mm.

Figure 1.

Above-ground stages of Silene spaldingii. From left to right: rosette, vegetative, flowering.

field methods

We monitored a total of 179 Silene spaldingii plants on Dancing Prairie Preserve, 6 km north of Eureka in Lincoln County, Montana, annually for 19 years with an initial sample of 100 plants in 1987. Four permanent belt transects were established within the population on the north end of the preserve (Lesica 1987). We recorded the fate of all individuals mapped in 1-m wide belt transects in mid-July when flowers were present but most plants had not yet fully senesced. Each year, we recorded the number of stems and stage (rosette, vegetative or flowering) of above-ground plants. In 1995 we estimated detection probability by reading transects twice, once in late May and once in July. All 45 flowering plants (detection rate 1.0), but only 40 of 47 vegetative plants (detection rate 0.85), were detected in May and again in July.

data analysis

Descriptive statistics

We calculated several descriptors of S. spaldingii population dynamics over time as a companion to the results of capture-recapture analyses. Bouts of dormancy can be unambiguously recorded only when they are both preceded and succeeded by a visible stage within the study period (Lesica & Steele 1994; Kery & Gregg 2004). We used these unambiguous bouts to calculate the number of successive years for which plants were dormant. For these bouts, we calculated whether plants emerged from dormancy in a smaller or larger stage class than that in which they entered dormancy. We also tabulated the number of unambiguously dormant plants in each year. Most of these bouts of dormancy lasted 1 or 2 years (see Results), so we could identify a representative sample of dormant plants with high confidence by excluding the first 3 years of the study (to distinguish new recruits from previously dormant plants), and the last 3 years (to distinguish dormancy from death). We analysed transition patterns among stage classes using these middle 13 years of data, for comparison with the capture-recapture analyses.

Capture-recapture analyses

If plants are not visible when the study period begins or when it ends, dormant plants cannot be distinguished from pre-germinant or dead plants. In the past decade, capture-recapture methods have become standard for estimating survival of dormant plants (Shefferson et al. 2001; Slade et al. 2003; Kery & Gregg 2004). Capture-recapture methods have long been used to account for differences in detectability of individuals in animal populations over time (e.g. Lebreton et al. 1992; Williams et al. 2002). These methods are described in detail elsewhere (see Nichols 1992; Cooch & White 2004, for non-technical introductions to capture-recapture methods; and Alexander et al. 1997; Shefferson et al. 2001; Kery & Gregg 2004, for application to analysis of prolonged dormancy in plants). However, more recently, Kery et al. (2005) showed that vital rates are not uniquely estimable for many, possibly most, models of prolonged dormancy in plants. In light of these results, we present capture-recapture analysis as a complement to our descriptive analysis, and a potentially interesting comparison for this long-term data set. Specifically, we estimated vital rates for S. spaldingii using multistate models in Program MARK (White & Burnham 1999), which estimate the survival and transition rates among stage classes by simultaneously solving for maximum likelihood values of survival for all stage classes, and transition probabilities among stage classes. We did not attempt goodness-of-fit tests such as comparing observed frequencies to model predictions or bootstrapped data (see, e.g., White & Burnham 1999; Williams et al. 2002), because, to the best of our knowledge, no such methods exist for multistate capture-recapture models.

For this analysis we divided plants between the three visible stage classes (flowering, vegetative and rosette), and an unobservable dormant stage. We fixed detection probabilities at 1 for flowering plants, 0.85 for vegetative plants and rosettes, and 0 for dormant plants (see ‘Study species’, above). Note that this slightly changes the definition of dormancy, relative to our descriptive analysis, which pooled vegetative plants that died back early in the season with dormant plants. We tested whether vital rates differed among states and over time by comparing transition models for S. spaldingi with the full model defined by:

  • inline image

where R indicates rosettes, D indicates dormant plants, V indicates vegetative plants, F indicates flowering plants, sY indicates survival of plants in stage class Y, pXY indicates transition probabilities from stage class Y to stage class X, conditioned on survival, and t indicates that a different probability was estimated for each transition period (pair of years) during the study. Note that this matrix defines rosettes to be new recruits only (no stage can regress to being a rosette), and does not include transitions from rosettes to flowering plants (we never observed this transition in the field). We observed rosettes in only 8 years, so we set all transitions from rosettes to 0 in the 10 transition periods with no dormant plants (18 × 9 + 8 × 3 = 186 parameters).

We tested whether vital rates differed among years by comparing the full model with a reduced model with parameters held constant over all years. We tested the specific hypotheses that dormant and vegetative plants had identical survival by setting sV,t = sD,t for all t, and that dormant and vegetative plants had identical probabilities of flowering the next year, by setting pFV,t = pFD,t for all t. To test each hypothesis, we compared each reduced model with the full model using likelihood ratio tests.

Demographic analysis

We evaluated the consequences of dormancy by calculating the reproductive value of each class relative to a value of 1.0 for the rosette class. Reproductive values indicate the relative contribution of an individual plant in each stage class to future generations, and maximization of reproductive value often equates to maximization of fitness (Caswell 1989, p. 176). Reproductive values were calculated from deterministic matrix population models. These models used survival and stage transition rates from the stage-dependent, time-invariant, capture-recapture model, fitted using Program MARK. We estimated recruitment in two ways; both assumed that Silene spaldingii has no persistent seed bank (see ‘Study species’, above). First, we solved for the per capita recruitment rate (new recruits per flowering plant) that led to stable population size (population growth rate of 1). Based on results of Lesica (1997), we assumed that half the new recruits appeared as rosettes and half as vegetative plants. Secondly, directly following Lesica (1997), we calculated the ratio of new recruits in each year to flowering plants in the previous year, excluding the first 3 years of the study, when previously dormant plants cannot be distinguished from new recruits. For matrices calculated using both estimates of recruitment, we calculated confidence limits for reproductive value by parametric bootstrapping, i.e. independently sampling vital rates over beta distributions defined by the mean and 95% profile likelihood confidence limits for each rate, and calculating the 95% limits of reproductive values from 1000 matrices. Both estimates of recruitment led to similar (±0.1) estimates of reproductive values, and nearly identical (±0.01) differences between the reproductive values of dormant and vegetative plants. Therefore, we present results calculated using only the first method.

Analysis of temporal variation

We used canonical discriminant function analysis (PROC DISCRIM, SAS Institute 2004) to relate S. spaldingii transition rates to previous stage and precipitation. Discriminant analysis determines functions of independent predictor variables that separate groups as well as possible (Manly 1986). In this case we sought to separate dormant, vegetative and flowering plants. We did not include newly recruited plants because evaluation of the effects of ‘previous stage’ is problematic for these plants. As in our descriptive analyses, we omitted data from the first 3 and last 3 years of the study to account for nearly all plants not appearing above ground that were dead or not yet recruited, rather than dormant (Lesica 1997). Independent variables were monthly precipitation and stage class during the year preceding July when the dependent variables were recorded. We inferred interannual differences in precipitation from the local weather station in Eureka, MT, USA (NOAA 2005), and tabulated total monthly precipitation for the year preceding plant emergence each May. In addition, we included dummy variables identifying each plant, to account for repeated measures of individuals. We used this analysis as a descriptive tool because our predictor variables were intercorrelated (see ‘Results’), and it is not possible to separate causal relationships among multiple correlated predictor variables.

Results

On average, S. spaldingii plants spent approximately one-third of growing seasons in each post-recruitment stage: flowering, vegetative, and dormant (Fig. 2). Bouts of prolonged dormancy were observed to last up to 6 years; however, bouts of 1 and 2 years accounted for 76% and 16% of all bouts, respectively (n = 371). There was no evidence that prolonged dormancy affected the size of S. spaldingii plants. Plants were recorded as being in the same stage class immediately before and after prolonged dormancy in 58% of recorded bouts. They achieved a larger stage class following dormancy in 20% of bouts and a smaller stage class in 22% of bouts (n = 371, χ2 = 0.26, P = 0.61).

Figure 2.

Number of Silene spaldingii plants in four stage classes from 1990 through 2002. Data are from the descriptive analysis, including only unambiguously dormant plants. The first 3 and last 3-year periods are not included because plants not appearing above ground in this period may have been dormant (last 3 years) or not yet recruited (first 3 years).

During the middle 13 years of the study, plants transitioned among mature stage classes at the following frequencies:inline image

where D, V and F indicate dormant, vegetative and flowering plants, respectively. It is obvious from this matrix that our descriptive analysis only identifies plants that survive dormant periods (survival of dormant plants, the sum of transitions from dormancy, is 1). Interestingly, the probability of flowering, conditioned on survival, was higher for dormant than vegetative plants. During this period, we observed 431 transitions from the vegetative state, of which 99 were followed by flowering. Correcting for survival of vegetative plants, 99 were followed by flowering out of 371 that survived (431 × 0.86 = 371, leading to a conditional flowering probability of 0.28). We observed 465 transitions from dormancy, of which 173 were followed by flowering (173/465 > 99/371, χ2 = 10.4, d.f. = 1, P ≈ 0.001).

Although survival of dormant plants cannot be measured directly, rates can be bounded based on vital rates of vegetative and dormant plants, for comparison with capture-recapture estimates. At one (unrealistic) extreme, we assumed perfect survival of both flowering and vegetative plants, i.e. we assumed all vegetative and flowering plants that appeared to die actually entered dormancy, and only dormant plants died. Ninety-three per cent of flowering plants appeared to survive, leaving 7% that could enter dormancy rather than dying; 86% of vegetative plants appeared to survive, leaving 14% that could enter dormancy before dying. We converted these percentages to mortality rates for dormant plants by adjusting for the proportion of plants in each stage class: 409 flowering plant-years (31%), 431 vegetative plant-years (33%), and 465 dormant plant-years (36%) → (0.07 × 0.31 + 0.14 × 0.33)/0.36  0.19 probability that plants become dormant then die.

These extreme assumptions led to the following transition matrix:inline image

This sets an absolute minimum survival of dormant plants at 0.81. Note that dormant plants would be more likely than vegetative plants to flower in the next year, even under these extreme assumptions.

More realistically, we assumed that flowering plants had high but not perfect survival (0.99), and F → D and V → D transitions occur in the same proportion (0.27/0.36) as observed in our empirical matrix. If 99% of flowering plants survived, 6% must have transitioned to dormancy then death. Under these assumptions, the unobservable transition from vegetative plants to dormancy then death would be (0.27/0.36) × 0.06 = 0.045. We used these rates to calculate mortality of dormant plants as above: (0.06 × 0.31 + 0.045 × 0.33)/0.36 ≈ 0.093 mortality of dormant plants. This would lead to the transition matrix:inline image

Under these reasonable assumptions, dormant and vegetative plants would have the same survival (0.91), and dormant plants would be more likely than vegetative plants to flower in the next year.

Capture-recapture analyses corroborated these patterns. Flowering S. spaldingii plants had the highest average annual survival (c. 0.99), followed by vegetative plants (0.90), dormant plants (0.88) and rosettes (0.86) (Tables 1 and 2). Confidence limits calculated by likelihood profiling of the time-invariant model correspond well to the range of possible transition rates calculated above (Table 2). Survival did not differ significantly between vegetative and dormant plants (Table 1), but dormant plants were significantly more likely than vegetative plants to flower the next year (Fig. 3, Table 1), and plants tended to enter dormancy after flowering. Vegetative plants were most likely to remain vegetative (Fig. 3). Rosettes always became dormant for at least 1 year before moving to a larger size class. Mean reproductive value, a correlate of future fitness of plants in each life stage, was 1.11 (95% confidence limits 1.01–1.84) for dormant plants, but only 0.97 (0.83–1.62) for vegetative plants (parametric bootstrap two-tailed P = 0.07). Reproductive value was highest in flowering plants (average = 1.40, 95% limits = 1.24–2.27).

Table 1.  Model selection statistics for S. spaldingii vital rates. We used likelihood ratio tests to compare reduced models with the full state- and time-dependent model, with degrees of freedom equal to the difference in the number of parameters (K) between the full stage- and time-dependent model and each reduced model. A significant difference indicates that a model is worse than the full model at fitting the data, e.g. P = 0.015 for ‘V and D same flowering’ indicates flowering probability differed significantly between V (vegetative) and D (dormant) plants
Vital rates–2 ln(L)KAICcχ2P
  • Akaike's ‘An Information Criterion’, provided for comparison with other capture-recapture studies.

Stage- and time-dependent4138.01864510.0
Stage-dependent only4611.5 124635.5473.5< 0.001
V and D same survival4159.21684495.2 21.20.270
V and D same flowering4171.41684507.4 33.40.015
Table 2.  Parameter estimates for transition rates, with 95% confidence limits calculated by likelihood profiling. ‘Average’ parameters refer to estimates from the time-invariant model, not the average of annual transition rates. See Methods for meaning of transition rate symbols
Year*sRsDsVsFpRDpRVpDVpDFpVDpVFpFDpFV
  • *

    First year of annual transition, e.g. 1987 refers to transitions from 1987 to 1988.

Average (time invariant)0.860.880.900.990.950.000.350.390.290.250.500.24
(0.53–1.00)(0.82–1.00)(0.82–0.94)(0.88–1.00)(0.76–1.00)(0.00–0.00)(0.29–0.42)(0.34–0.45)(0.21–0.36)(0.21–0.31)(0.40–0.55)(0.20–0.30)
1987NA0.450.720.96NANA0.450.450.480.000.790.13
 (0.00–1.00)(0.17–1.00)(0.71–1.00)  (0.00–0.55)(0.00–0.55)(0.00–0.96)(0.00–0.61)(0.57–0.92)(0.03–0.34)
1988NA0.910.870.93NANA0.110.720.240.520.420.22
 (0.64–1.00)(0.54–1.00)(0.46–1.00)  (0.01–0.31)(0.44–0.91)(0.00–0.63)(0.20–0.83)(0.00–0.83)(0.01–0.75)
1989NA0.880.900.90NANA0.000.390.730.210.710.08
 (0.24–1.00)(0.56–1.00)(0.66–1.00)  (0.00–0.28)(0.08–0.83)(0.43–0.90)(0.07–0.49)(0.52–0.83)(0.03–0.21)
1990NA0.801.000.94NANA0.290.440.470.160.330.36
 (0.57–1.00)(0.73–1.00)(0.81–1.00)  (0.16–0.45)(0.30–0.58)(0.09–0.83)(0.01–0.55)(0.15–0.55)(0.18–0.58)
1991NA1.000.890.98NANA0.230.140.450.160.480.44
 (0.66–1.00)(0.75–1.00)(0.89–1.00)  (0.06–0.45)(0.02–0.32)(0.26–0.65)(0.06–0.32)(0.29–0.67)(0.26–0.63)
19921.001.000.971.000.000.000.340.460.090.710.630.20
(1.00–1.00)(0.81–1.00)(0.86–1.00)(0.89–1.00)(0.00–0.11)(0.00–0.26)(0.20–0.50)(0.32–0.61)(0.00–0.23)(0.54–0.86)(0.33–0.87)(0.04–0.51)
19930.791.000.900.981.000.000.080.240.580.080.830.09
(0.58–0.95)(0.79–1.00)(0.76–1.00)(0.92–1.00)(0.86–1.00)(0.00–0.14)(0.00–0.29)(0.08–0.45)(0.36–0.76)(0.01–0.22)(0.70–0.92)(0.03–0.21)
19940.001.000.891.000.310.020.400.570.230.470.350.11
(0.00–1.00)(0.92–1.00)(0.69–1.00)(0.84–1.00)(0.00–0.98)(0.00–0.69)(0.29–0.51)(0.47–0.67)(0.03–0.49)(0.29–0.71)(0.10–0.65)(0.01–0.41)
19950.001.000.960.870.290.410.450.140.040.390.230.44
(0.00–0.04)(1.00–1.00)(0.87–1.00)(0.78–1.00)(0.00–0.59)(0.00–0.71)(0.08–0.84)(0.00–0.48)(0.00–0.18)(0.24–0.54)(0.12–0.40)(0.29–0.58)
19960.441.000.870.950.450.450.210.790.330.390.260.12
(0.00–1.00)(0.49–1.00)(0.74–1.00)(0.84–1.00)(0.00–0.55)(0.00–0.55)(0.12–0.33)(0.53–0.99)(0.18–0.52)(0.23–0.54)(0.13–0.46)(0.04–0.25)
19970.760.940.851.001.000.000.390.480.520.060.540.18
(0.26–1.00)(0.53–1.00)(0.59–1.00)(0.86–1.00)(0.47–1.00)(0.00–0.00)(0.19–0.61)(0.27–0.67)(0.18–0.74)(0.00–0.25)(0.38–0.67)(0.09–0.32)
19980.070.950.980.910.450.510.710.090.140.080.300.55
(0.00–1.00)(0.75–1.00)(0.79–1.00)(0.69–1.00)(0.00–1.00)(0.00–1.00)(0.50–0.90)(0.02–0.20)(0.00–0.38)(0.02–0.23)(0.09–0.55)(0.32–0.79)
19990.000.770.781.000.420.390.200.350.160.070.230.47
(0.00–0.92)(0.34–1.00)(0.65–0.95)(0.98–1.00)(0.00–0.61)(0.00–0.58)(0.00–0.56)(0.13–0.70)(0.03–0.35)(0.02–0.16)(0.00–0.57)(0.17–0.82)
2000NA1.000.960.89NANA0.440.000.350.110.580.33
 (0.36–1.00)(0.83–1.00)(0.58–1.00)  (0.13–0.78)(0.00–0.16)(0.17–0.54)(0.04–0.22)(0.22–0.86)(0.09–0.69)
2001NA1.000.791.00NANA0.490.290.020.320.000.32
 (0.65–1.00)(0.61–1.00)(1.00–1.00)  (0.26–0.72)(0.13–0.51)(0.00–0.34)(0.16–0.52)(0.00–0.27)(0.07–0.72)
2002NA1.000.740.93NANA0.000.510.300.260.170.42
 (0.16–1.00)(0.49–1.00)(0.67–1.00)  (0.00–0.46)(0.07–1.00)(0.00–0.65)(0.09–0.52)(0.00–0.51)(0.14–0.70)
2003NA0.480.840.99NANA0.640.000.210.400.310.18
 (0.14–1.00)(0.71–1.00)(0.89–1.00)  (0.10–0.95)(0.00–0.27)(0.08–0.56)(0.14–0.56)(0.16–0.55)(0.05–0.32)
2004NA1.000.940.70NANA0.280.480.010.410.040.20
 (0.70–1.00)(0.65–1.00)(0.43–1.00)  (0.02–0.51)(0.17–0.77)(0.00–0.35)(0.16–0.70)(0.00–0.58)(0.03–0.50)
Figure 3.

Mean annual life cycle for Silene spaldingii at Dancing Prairie. R = rosette, D = dormant, V = vegetative, F = flowering (reproductive). Magnitude of the transition rates is indicated by the thickness of the arrow and the number next to it. Recruitment includes transitions from F to R (0.17) and F to V (0.17). Transition rates were calculated from the time-invariant capture-recapture model.

Survival and transition probabilities differed significantly among years in our capture-recapture analysis (Tables 1 and 2). Discriminant analysis identified two multivariate functions that distinguished conditions associated with a S. spaldingii plant being dormant, vegetative or reproductive (Fig. 4). Prolonged dormancy was associated with having just flowered and with high summer (July, August, September) and low, non-growing season (especially October) precipitation the previous year (Table 3).

Figure 4.

Discriminant functions separating dormant (D), vegetative (V) and flowering (F) plants. Thick crosses indicate means for each group, ±2 standard errors. Extended lines indicate the spread of c. 50% of observations in each group. The strongest correlates of each factor are identified on each axis (see Table 2).

Table 3.   Correlations between independent variables and canonical factor scores resulting from discriminant function analysis. Significant r-values are in bold; P < 0.05 when r > 0.27 (n = 52)
VariableFactor 1Factor 2
Previous stage
 Flowering0.1330.301
 Vegetative0.063–0.222
 Dormant–0.188–0.083
Precipitation in preceding year
 July0.2720.224
 August0.2040.428
 September0.1790.293
 October0.3890.139
 November–0.2520.070
 December–0.200–0.090
 January–0.1600.033
 February–0.068–0.122
 March–0.2130.079
 April–0.0130.098
 May–0.0580.104
 June–0.2440.022

Discussion

Overall, prolonged dormancy appeared to be advantageous for S. spaldingii. Compared with vegetative plants, dormant plants were not more likely to die, and they were more likely to flower the following year. The difference in reproductive value between dormant and vegetative plants corroborates this result. The fact that we detected an advantage of prolonged dormancy compared with being vegetative suggests that it serves an important function in the life history of this plant rather than being solely a passive response to resource depletion.

Whether prolonged dormancy is an advantage or a constraint appears to differ among species as well as across environments. Prolonged dormancy does not always appear to be an advantage for orchids. Hutchings (1987) followed the fate of a population of the orchid Ophrys sphegodes in chalk grassland and reported lower survival following dormancy compared with the vegetative state by truncating the first and last 2 years of data to allow distinguishing dormancy from mortality. Similarly, Shefferson et al. (2003) reported a cost of dormancy for the orchid Cypripedium calceolus in wet meadows; survival and transitions to flowering of survivors for dormant plants were lower than for vegetative plants, although this finding may be an artifact of smaller plants being more likely to display prolonged dormancy (Shefferson 2006). Prolonged dormancy was also associated with decreased adult survival in Neotinea ustulata, an orchid of moist meadows (Shefferson & Tali 2007). However, Shefferson et al. (2005) provide evidence that prolonged dormancy may help mesic-site orchids survive defoliation or shading. Kery & Gregg (2004) studied one population of the orchid Cypripedium reginae in a wetland and one in a forest. At the wetland site dormant plants were less likely to flower the following year than were plants in the vegetative state, but there was no difference between dormant and vegetative plants at the presumably drier forest site. Resources limiting reproduction and the potential costs of transpiration may differ between these habitats and the semi-arid grasslands inhabited by S. spaldingii.

Demographic transitions in S. spaldingii depended on environmental factors in a complex interdependent manner (Table 1, Fig. 3) that is also consistent with a resource-based explanation for prolonged dormancy. The discriminant function analysis showed that increased frequency of prolonged dormancy relative to production of above-ground parts was associated with higher precipitation in the previous summer and lower precipitation the previous fall and winter (Fig. 3). The association between low fall and winter precipitation and subsequent prolonged dormancy supports a simple resource-based hypothesis: many plants respond to dry soil conditions by not producing above-ground organs, but, in response to above-average fall and winter precipitation, they obtain resources and flower the following summer. The strong positive association between prolonged dormancy and wet summers is more difficult to explain but may be caused by increased seed production with concomitant resource depletion induced by high summer precipitation.

Our results suggest that the frequency of prolonged dormancy in S. spaldingii was determined not only by immediate external resource supplies (precipitation) but also by previous resource allocation (previous stage). Previous studies of prolonged dormancy suggested that it was associated with immediate resource supply such as precipitation (Epling & Lewis 1952; Thomas 1981; Tyler & Borchert 2002; Kery et al. 2005), increased light and nutrients caused by fire (Abrahamson & Hartnett 1990; Keeley 1991; Lesica 1999) or loss of biomass due to herbivory (Morrow & Olfelt 2003). Our study also demonstrated an association with precipitation; however, we found that previous stage was important as well. Rosettes always entered dormancy before entering a larger stage. The most frequent stage transitions for larger plants were (i) flowering to dormant, (ii) vegetative to vegetative, and (iii) dormant to flowering (Fig. 3). We speculate that plants tending to remain vegetative lack enough energy (i.e. are too small) to flower. Plants alternating between reproduction and dormancy may be less energy-limited, but their ability to flower and set seed is determined by some other resource. A plant that has adequate energy to flower but has exhausted its limiting resource to the point of being unable to flower the following year will remain dormant. If it can obtain enough of the limiting resource during dormancy it will emerge and flower again. Nitrogen and phosphorus are likely candidates for this resource because they are often limiting in semi-arid ecosystems (Risser 1985), and both are required for flower and fruit production (Reekie 1997; Epstein & Bloom 2005).

Alternatively the function of prolonged dormancy in geophyte life history may be similar to that of the seed stage in desert annuals (Went 1949; Schaffer & Gadgil 1975; Mulroy & Rundel 1977). Both strategies would allow plants to avoid stressful above-ground environments. Producing leaves may result in a net loss of energy for some S. spaldingii plants in years when low soil moisture makes early senescence unavoidable. Prolonged dormancy would be an advantage over being vegetative in such years.

Regardless of the mechanism, our study provides evidence that prolonged dormancy plays a role in S. spaldingii resource allocation rather than simply being a symptom of resource depletion. The fact that flowering is more likely following the dormant stage than after the vegetative stage indicates that dormant plants sometimes gain or conserve critical resources better than those producing above-ground structures. Experiments examining the energy, water and nutrient budgets of dormant plants are needed to shed light on the mechanisms making prolonged dormancy an advantage in some populations. Future research directed at the mechanisms providing an advantage to dormant plants will shed more light on how and under what circumstances prolonged dormancy influences resource allocation and life history.

Acknowledgements

Bee Hall and Maria Mantas helped with data collection. Matt Kauffman, Bill Morris, Pamela Kittelson, Anna Sala, Michael Hutchings, Helen Alexander and an anonymous reviewer provided helpful comments. Research was supported by funding from The Nature Conservancy and National Science Foundation grant DEB 02–36427 to E.C.

Ancillary