Indirect and mitigated effects of pulsed resources on the population dynamics of a northern rodent

Authors


Correspondence author. E-mail: nlobo3@uwo.ca

Summary

  1. Pulsed resources have significant effects on population and community dynamics in terrestrial ecosystems. Mast seeding is an important resource pulse in deciduous forests; these boom and bust cycles of seed production generate strong lagged population responses by post-dispersal seed predators such as rodents, which then cascade through multiple trophic levels and regulate population dynamics of their predators and prey. However, similar interactions in another major pulsed system, coniferous forests, are inconsistent, and the effects of interannual variation in conifer seed production on many consumer populations are largely unknown.
  2. We used large-scale manipulation and intensive monitoring to examine the population dynamics of deer mice (Peromyscus maniculatus) in relation to fall seed production by two northern conifers, white spruce (Picea glauca) and subalpine fir (Abies lasiocarpa). Previous studies have shown that spruce seeds are a preferred food source of mice, while fir seeds are generally avoided if other foods are available. Therefore, we expected that there would be a positive relationship between mouse demography and previous spruce seed production, but no effect of fir mast seeding.
  3. Supplementation of a mouse population using spruce seeds indicated that increased fall spruce seed availability can enhance overwinter survival and population densities in the following spring, summer, and fall. However, long-term population monitoring indicated that mouse demography was not positively affected by spruce mast seeding, likely due to strong interspecific competition with the North American red squirrel (Tamiasciurus hudoniscus), a dominant pre-dispersal spruce seed predator.
  4. Conversely, we observed an unexpected delayed effect of fir mast seeding, where increased fall fir seed production did not influence overwinter or spring mouse demography, but instead enhanced summer survival, body masses and pregnancy rates of overwintered adults. This led to increased summer population densities and may have been mediated by population responses of invertebrate post-dispersal seed predators to increased fir seed availability.
  5. Our results indicate that rodent responses to resource pulses in coniferous forests are more complex than in deciduous environments and reveal previously unobserved direct and indirect consumer–resource dynamics that require further examination. This system is ideal for the large-scale, integrative ecosystem studies that ecologists are encouraged to pursue.

Introduction

Resource pulses greatly affect community dynamics in terrestrial ecosystems (Ostfeld & Keesing 2000; Yang et al. 2008, 2010). These brief, irregular events of increased resource availability do not just influence their main consumers, but also have cascading direct and indirect effects at multiple trophic levels (McShea 2000; Ostfeld & Keesing 2000; Yang et al. 2008, 2010). As such, the patterns and mechanisms of resource pulses and their interactions with consumers have emerged as important areas of ecological research, revealing fundamental aspects of ecosystem structure and function, and providing significant insights into the effects of environmental and resource variability on the regulation of behaviours, populations and communities (Yang et al. 2008; Bergeron et al. 2011).

One common terrestrial resource pulse is mast seeding, defined as the intermittent, synchronous production of large seed crops by most reproductive adults in a plant population (Ostfeld & Keesing 2000). The most well-documented hypothesis for the ultimate function of mast seeding is predator satiation (Kelly & Sork 2002). These large fluctuations in seed availability also directly influence the reproduction and survival of many seed predators, often generating strong lagged population responses to the resource pulse (Ostfeld & Keesing 2000; Yang et al. 2010). A prominent example of this phenomenon is the sporadic spring/summer rodent population peaks in deciduous forests in Europe and eastern North America that are directly linked to masting by oak (Quercus spp.), maple (Acer spp.) and beech (Fagus spp.) trees the previous fall (Jensen 1982; Wolff 1996; McCracken, Witham & Hunter 1999; McShea 2000; Ostfeld & Keesing 2000; Falls, Falls & Fryxell 2007). Increased seed availability after a masting event permits sustained breeding into the late fall and winter, enhanced overwinter survival, and earlier onset of breeding the following spring (Jensen 1982; Wolff 1996; McCracken, Witham & Hunter 1999; Ostfeld & Keesing 2000). Subsequent low fall seed production cannot sustain high rodent populations, leading to decreased overwinter survival and densities in the following non-mast years (Jensen 1982; Wolff 1996; McCracken, Witham & Hunter 1999).

While this seed–consumer relationship has been well documented in deciduous forests, similar interactions in another major pulsed system, coniferous forests (Alexander, Shearer & Shepperd 1990; LaMontagne & Boutin 2007), are less clear and the effects of conifer mast seeding on rodent populations appear to be more complex and largely inconsistent. Gashwiler (1979) reported that spring and summer rodent populations did not differ following good and poor fall conifer seed crops; however, population densities were higher in the fall following a mast year, indicating a delayed response to increased seed supply. Other studies have shown inconsistent population peaks (Jameson 1953; Elias, Witham & Hunter 2006), population declines (Stickel & Warbach 1960) and no response (McCracken, Witham & Hunter 1999; Schnurr, Ostfeld & Canham 2002; Boonstra & Krebs 2006) in relation to heavy conifer seed crops. The consumer–resource dynamics of this pulsed system require comprehensive evaluation, as the multidimensional nature of these interactions may strongly depend on factors such as climatic or seasonal ecological limitations, additional food or nutrient availability, and inter- and intraspecific competition.

Given the general lack of understanding of how rodent populations are affected by high interannual variation in conifer seed production, and the significance of their population fluctuations for structuring bird and insect population dynamics (McShea 2000; Ostfeld & Keesing 2000), long-term data sets on annual conifer seed production and intensive rodent trapping can provide valuable insights into this important interaction in forest ecosystems. The effect of conifer masting on consumer population dynamics, and its implications for the predator satiation hypothesis and the evolution of masting, have recently been documented for North American (Tamiasciurus hudsonicus) and Eurasian (Sciurus vulgaris) red squirrels, two dominant pre-dispersal seed predators (Boutin et al. 2006; Wauters et al. 2008; Fletcher et al. 2010; Di Pierro et al. 2011; Archibald et al. 2012). However, the efficacy of this hypothesis for post-dispersal seed predators is unclear.

Here, we report on experimental and observational studies examining the population dynamics of deer mice (Peromyscus maniculatus) in relation to seed production by two northern conifers. It was our goal to examine how the characteristics of both the resource and the pulse event influence major consumers. Deer mice are highly granivorous, with seeds comprising a significant component of their diet during most of the year (Jameson 1952). White spruce (Picea glauca) and subalpine fir (Abies lasiocarpa) are dominant masting conifer species in western Canada, and previous studies have shown that spruce seeds are a high-quality food source to rodents, while fir seeds are not. Mice consume (Radvanyi 1970; Lobo, Duong & Millar 2009) and cache (Lobo 2013) large quantities of spruce seeds available to them and can maintain body condition on diets restricted to these seeds (Lobo & Millar 2011). Conversely, fir seeds contain low nutritional value and high concentrations of plant secondary compounds (PSCs) (Lobo & Millar 2011; Rubino et al. 2012) and are avoided by mice if other food options are available (Lobo, Duong & Millar 2009). Mice must also alter their food intake and digestion patterns to persist on fir seed-diets (Lobo & Millar 2011). As such, we hypothesized that there would be a positive relationship between mouse populations and previous spruce seed production, but no effect of fir mast seeding. We tested this hypothesis in two ways: (i) a large-scale food addition experiment using spruce seeds to simulate a masting event and (ii) examining spruce and fir seed production, and mouse population dynamics at a long-term study site over a 10-year period.

We supplemented a long-term deer mouse population-monitoring grid with an excess of white spruce seeds during the typical seed rain period in a single non-mast year. This population was then compared with a control population from a nearby long-term monitoring grid for several years prior to, and one year after, the supplementation event. The two grids have historically shown similar annual demographical patterns (Kalcounis-Rueppell, Millar & Herdman 2002). Therefore, we predicted that the population dynamics of the two grids would be similar in most years prior to seed supplementation, but we would observe higher overwinter survival and spring/summer breeding and population densities on the treatment grid after supplementation.

When examining the long-term demography of deer mice in response to natural conifer seed production, we predicted that the increased food supply from fall spruce masting would enhance the overwinter survival of mice and allow for earlier and increased breeding in the spring and summer following the masting event. These factors should generate increased mouse densities in the spring and summer following spruce masting, relative to years of low seed production. We also expected that there would be no relationship between population vital rates and previous fir seed production.

Materials and methods

Study Area and Species

This study was conducted from 2002 to 2011 in the Kananaskis Valley in SW Alberta, Canada, a 4200-km2 multiuse area located in the front ranges of the Canadian Rocky Mountains. Two long-term mouse population-monitoring grids were used; the control (50°45·99′ N, 115°08·62′ W) and treatment (50º47·16′ N, 115º09·52′ W) grids were approximately 3 km apart, and both located in optimum deer mouse habitat (Millar, Innes & Loewen 1985). The main tree species on both grids were white spruce, subalpine fir and lodgepole pine (Pinus contorta), with spruce trees most prevalent. White spruce and subalpine fir are fall masting species (Alexander, Shearer & Shepperd 1990; Nienstaedt & Zasada 1990), while lodgepole pine produces low, consistent annual seed crops but in predominately serotinous cones (Despain 2001).

The most abundant rodent species in the study area is the deer mouse (43% of all captures across major habitat types; Millar, Innes & Loewen 1985). Population densities of mice in the Kananaskis Valley are low (1979–1997: spring density = 8·63 ± 0·79, fall density = 9·26 ± 1·00; Millar & McAdam 2001) and show annual cycles, with densities typically increasing over the summer and declining over winter (Millar & McAdam 2001). Breeding is constrained and highly seasonal; overwintered (OW) adults initiate reproduction in the spring, and females can produce multiple litters in a summer, but breeding by young-of-the-year (YY) is rare (Millar & McAdam 2001).

Deer Mouse Trapping and Individual Data Collection

Mice were trapped at each grid 2·14 ± 0·02 times per week during the breeding season (May-September) using Longworth live traps (one trap per station with 20 m spacing; control grid = 42 stations, treatment grid = 33 stations) baited with sunflower seeds and oats. The control grid was trapped from 2002 to 2011 with an estimated effective trapped area of 1·68 ha, and the treatment grid from 2004 to 2011 with an estimated effective trapped area of 1·40 ha (inclusive boundary strip method; Stickel 1954). Each mouse was tagged (Monel #1 ear tags affixed to each ear) at first capture. Upon each capture, tag number, mass (nearest 0·5 g, using a Pesola spring balance), sex, age and breeding condition were recorded. Age was assigned as overwintered adult (OW) or (YY), based on size and pelage colour. Breeding condition of males was classified as scrotal or non-scrotal, and females as perforate or non-perforate, pregnant, lactating or pregnant and lactating. A total of 23 442 traps were set during the study, with 5548 total mouse captures, and 892 individual mice trapped during the trapping sessions.

Cone Production Estimation

We estimated fall cone production by counting the number of cones visible through binoculars on one side of the top 3 m (approximate, based on subjective evaluation) of each of 11 white spruce and eight subalpine fir trees, each year in late August between 2001 and 2011 (see LaMontagne, Peters & Boutin 2005). The height of each tree was approximately 10–12 m (J. Millar unpublished). To maintain consistency in data collection, the same observer (JSM) performed the cone counts each year, standing in the same spot each year approximately 20–30 m from the base of each tree. The same trees, scattered across the valley, were used each year to obtain a representative estimate of the annual variability in cone production in our study area. The number of cones on each tree was ln(+ 1) transformed, and these transformed values were averaged among all trees of the same species within each year to derive the cone index for that year (Boutin et al. 2006; Appendix S1: Fig. S1). We also measured seed rain at three study sites from 2006 to 2010 (Appendix S1: Fig. S2); see Appendix S1 for further details on this and the use of cone abundance indices.

Seed Supplementation Experiment

No manipulations were conducted on either grid between May 2004 and early September 2010. In late fall 2010 (September 19, 2010), we supplemented the treatment grid with 150 kg of white spruce seeds; seeds were broadcast by hand in one late afternoon period, over the entire trapping area as well as a few metres beyond the periphery. The density of seeds supplemented was approximately 10·71 g m−2; this amount is almost double the lower threshold density at which rodents cease active foraging for spruce seeds (5·60 ± 0·61 g m−2; Lobo 2013; Appendix S2). However, based on estimates of spruce seed mass, the number of spruce seeds per cone, and cone counts during mast years, the amount of seeds produced by spruce trees (prior to pre-dispersal seed predation) in a mast year would be sufficient to produce the supplemented seed density (see ‘Population responses to white spruce masting’ section in Discussion). Spruce cone production was low in 2010 (Appendix S1: Fig. S1), so no significant effects of background food availability were expected.

Estimation of Demographic Parameters

The abundance of deer mice was estimated using Pollock's Robust Design Model implemented in Program mark (White & Burnham 1999; Appendix S3). This was carried out for the full population each year, as well as OW and YY mice separately. Goodness-of-fit testing did not indicate any significant deviations from equal catchability and equal survival assumptions of basic open mark–recapture models (Appendix S3). Population densities were calculated by dividing abundance estimates by the effective trapped area and were categorized by spring (prior to mid-May), summer (late May to mid-August) and fall (late August onward) seasons. Annual summer and winter population growth rates were calculated as the intrinsic rate of increase week−1 (Appendix S3). As female deer mice are highly philopatric but emigration rates of males are high (Teferi & Millar 1994), we estimated overwinter survival as the proportion of tagged fall resident YY females that were also trapped the following spring. Excluding males reduced confounding effects of emigration on overwinter survival estimates. Overwinter immigration was estimated as the proportion of spring residents that were not part of the resident population the previous fall.

Body masses of females were excluded from the data set to avoid potential effects of undetected pregnancies. We averaged multiple body mass recordings of each male and these values were used to calculate annual average spring, summer and fall masses of OW males. For each year, we also recorded the mass at first capture of YY males and estimated individual growth rates of juvenile males (Appendix S3).

We estimated the initiation of breeding, average parturition date and length of the breeding season for each year (Appendix S3). We also calculated the annual proportions of scrotal OW males in the spring, summer, and fall, as well as the annual proportions of resident OW and YY pregnant females, and the annual proportions of resident OW females that had multiple litters.

Statistical Analyses

We screened all data for outliers and deviations from normality prior to statistical analyses. Unless otherwise stated, analyses were performed using r version 2.14.2 (R Foundation for Statistical Computing, Vienna, Austria) and spss version 16.0 (SPSS Inc., Chicago, IL, USA), α was set to 0·05, and values are presented as means ± SE.

Effects of Seed Supplementation on Mouse Demography

The effects of supplementation on deer mouse population, breeding and body mass dynamics were examined by comparing the treatment and control grids in each year before (2004–2010) and after (2011) the supplementation event (late fall 2010). As only one treatment and one control grid were used, we could not conduct statistical comparisons of average population densities and growth rates between grids. However, parametric and nonparametric statistics could be used to test the effects of supplementation on other population, breeding and body mass parameters. We evaluated whether supplementation affected average spring, summer and fall mouse population densities using a density effect ratio (see Boonstra & Krebs 2006), comparing the annual average densities on the treatment grid to those on the control grid in each year. Cormack–Jolly–Seber models (CJS; Appendix S3), implemented in Program mark, were used to evaluate whether the summer survival of OW mice varied between grids in each year before and after the late fall 2010 supplementation event. The probability of survival between trapping sessions and the encounter probability were modelled as a function of grid and time, for a total of 15 models in the candidate model set each year. Model comparison and selection was performed using Akaike's Information Criterion corrected for small sample sizes (AICc; Hurwich & Tsai 1989). See Appendix S3 for further details on model selection using AICc.

Effects of Natural Mast Seeding on Mouse Demography

Only data from the control grid (2002–2011) were used to examine the effects of conifer masting on deer mice. Proportion data (overwinter survival, proportions of breeding males and females) were arcsine square root transformed, and all other population, body mass and breeding data were ln (+ 1) transformed prior to statistical analyses. We used the coefficient of variation (CV) to describe variability in average cone production among years.

We used information-theoretic methods (Burnham & Anderson 2002) to evaluate 14 plausible linear models explaining annual variation in population densities and growth rates, and overwinter survival (response variables) of deer mice. The simplest model was a random walk, y = a + εt, where εt is a normally distributed random variable with mean zero. In alternate models, we included the previous fall's spruce, fir and total cone indices, and previous fall/spring population density as covariates (explanatory variables), as well as various biologically relevant combinations of these factors. We used the CV and the s-index (standard deviation of log10(+ 1); Henttonen, McGuire & Hansson 1985) to describe variability in average spring and fall population densities among years.

We used CJS models, implemented in Program MARK, to evaluate the relationship between summer survival of OW mice and the previous fall's spruce, fir and total cone indices. The probability of survival between trapping sessions was modelled as a function of these covariates and time, while the encounter probability was modelled only as a function of time. Our candidate model set contained 28 models.

Backwards stepwise linear regression models were used to test for the effects of the previous fall's spruce and fir cone production on all body mass and breeding parameters. The α-to-leave was set to 0·10. Simple linear regression models were also used to examine the relationships between these parameters and previous total cone production.

Results

Effects of Seed Supplementation on Mouse Demography

Supplementation of the treatment grid using spruce seeds in late fall 2010 resulted in enhanced survival and population densities of deer mice. Mouse populations on the treatment and control grids fluctuated in synchrony prior to supplementation (2004–2010; R = 0·55, n = 21, P = 0·01), with treatment densities typically lower than, or similar to, densities on the control grid (spring density ratio = 0·64 ± 0·04, summer density ratio = 0·77 ± 0·10, fall density ratio = 0·93 ± 0·15). However, this trend was reversed after the supplementation event, with densities on the treatment grid consistently higher than the control grid (spring 2011 density ratio = 1·44, summer 2011 density ratio = 2·38, fall 2011 density ratio = 1·63; Fig. 1). Consequently, the long-term population fluctuations on the two grids were no longer correlated once the post-supplementation densities were included in the analysis (R = 0·26, n = 24, P = 0·23). We observed no difference in overwinter immigration between grids in any year of the study (grid × year interaction: χ26 = 6·37, P = 0·38; grid main effect: χ21 = 0·25, P = 0·62), but supplementation resulted in increased overwinter survival. Loglinear analysis indicated that differences in survival depended on both grid and year (χ26 = 14·71, P = 0·02), where no difference between grids was observed in all years prior to supplementation (all  0·09), but overwinter survival was highest on the treatment grid after supplementation (2011; P = 0·01; Fig. 2).

Figure 1.

Long-term white spruce (open bars) and subalpine fir (closed bars) cone production, and deer mouse population densities on the control (open circle/dashed line) and treatment (closed circle/solid line) grids in the Kananaskis Valley, Alberta. Densities are presented as spring, summer and fall means ± SE. No manipulations were conducted at either grid from spring 2004 – early fall 2010. The black arrow indicates when the treatment grid was supplemented with white spruce seeds (late fall 2010).

Figure 2.

Overwinter survival(± SE) of deer mice on the control and treatment grids. No manipulations were conducted at either grid from spring 2004 to early fall 2010. The black arrow indicates when the treatment grid was supplemented with white spruce seeds (late fall 2010). The number of tagged fall resident young-of-the-year (YY) females is shown. Within each winter, an asterisk denotes that the grids were significantly different ( 0·05) from each other.

The OW mouse population on the treatment grid declined at a considerably slower rate than the control grid in the summer after supplementation, but this trend was also observed in most summers (five of seven) prior to supplementation (Appendix S4: Fig. S3b). CJS analyses indicated that the combined weight of all models in which summer survival between trapping sessions was dependent on grid was only 0·34 in 2004, but was high (range = 0·62–0·97) in all other years prior to supplementation, and 1·00 in the summer after supplementation (Appendix S4: Table S2). Overall, survival of OW mice on the treatment grid was typically higher than mice on the control grid both before and after supplementation, and it is unclear to what degree the enhanced survival in the supplemented OW population was influenced by a grid effect.

Young-of-the-year (YY) mice were typically found in lower or similar densities on the treatment grid compared with the control grid in most years prior to the late fall 2010 seed supplementation (summer density ratio = 0·53 ± 0·13, fall density ratio = 0·88 ± 0·19; Appendix S4: Fig. S4). However, the YY density on the treatment grid was almost double that of the control grid (density ratio = 1·90) in the summer after supplementation, but this disparity did not persist into the fall (density ratio = 1·20; Appendix S4: Fig. S4). Both YY populations grew over each summer during the study, but the rate of population growth was higher on the control grid after supplementation, while the opposite trend was observed in most years prior to supplementation (Appendix S4: Fig. S3c). See Fig. S5 (Appendix S4) for annual population densities of OW and YY mice on both grids.

Supplementation did not advance the timing of breeding, nor did it enhance breeding by OW or YY individuals (Appendix S4). Furthermore, supplementation did not affect the body masses of OW mice or the mass at first capture and growth rates of YY males (Appendix S4).

Effects of Natural Mast Seeding on Mouse Demography

Average cone production of both white spruce (CV = 1·34) and subalpine fir (CV = 1·29) varied considerably among years, satisfying the criterion widely used for identifying mast seeding species (CV > 1; LaMontagne & Boutin 2009). White spruce masted in 2001 and 2003 and subalpine fir in 2001 and 2002; cone production was not ‘all-or-nothing’, with both species producing bumper crops periodically (Fig. 1 and Appendix S1: Fig. S1). Spruce and fir cone indices were not significantly correlated (R = 0·55, n = 11, P = 0·09).

Average spring population densities on the control grid ranged from 3·57 to 10·42 ha−1, while fall densities ranged from 7·44 to 18·45 ha−1 (Fig. 1). Population variability among years was low, but similar, in the spring (CV = 0·31, s-index = 0·15) and fall (CV = 0·38, s-index = 0·16). Variability in fall YY densities was similar to the full population (CV = 0·37, s-index = 0·16), while variability in fall OW densities was considerably higher (CV = 0·70, s-index = 0·29). Still, the overall degree of population fluctuation among years was quite low (s-index < 0·50; Henttonen, McGuire & Hansson 1985) for the full, OW and YY populations. See Fig. S5 (Appendix S4) for annual population densities of OW and YY mice.

Annual variation in average spring and fall population densities were not explained by the previous fall's white spruce, subalpine fir or total cone production (Table 1). They were also not explained by population densities in the preceding fall or spring, respectively (Table 1). However, the most parsimonious model to explain the annual variation in average summer densities contained the previous fall's fir cone production (Table 1); mouse populations were higher in the summers following heavy fir cone crops (Fig. 3a). We observed the same pattern when considering OW (Fig. 3b) and YY (Fig. 3c) densities separately (Table 1). Annual variation in summer population growth was not explained by cone production during the previous fall, but was inversely related to the preceding spring population density (β = −0·07 ± 0·03, 95% CI(β)  = −0·13, −0·01; Table 1). Winter population growth and overwinter survival were not related to any of the covariates (Table 1).

Table 1. Evaluation of the parsimony of linear models examining the annual variation in average deer mouse density (N), population growth (r) and overwinter survival (Φ) relative to previous population density (Nt-1) and white spruce (St-1), subalpine fir (Ft-1) and total (Tt-1) cone indices. Only the most supported models are shown; the complete candidate model sets for each population parameter are presented in Appendix S4: Tables S7-S11. εt refers to a normally distributed random variable with mean zero, and K is the number of estimable parameters in each model. Model results are provided for the full population, as well as overwintered (OW) and young-of-the-year (YY) mice separately. The best-supported models, based on AICc scores, are in bold
Model K Full populationOW populationYY population
AICcΔAICcAICcΔAICcAICcΔAICc
Population density
Spring
  N = a + ε t 3 11·31 0·00
  N = a + bS t-1  + ε t 417·366·05
Summer
  N = a + bF t-1  + ε t 4 8·39 0·00 3·29 0·00 8·25 0·00
  N = a + bT t-1  + ε t 411·062·67–1·272·0215·517·26
Fall
  N = a + ε t 3 15·17 0·00 18·70 0·00 14·75 0·00
  N = a + bF t-1  + ε t 419·604·4324·495·7918·944·19
Population growth
Winter
  r = a + ε t 3 −55·36 0·00
  r = a + bN t-1  + ε t 4−41·0614·30
Summer
  r = a + ε t 3−32·922·67 −25·42 0·00
  r = a + bN t-1  + ε t 4 −35·59 0·00 −21·55 3·87
Overwinter survival
  Φ = a + ε t 3 12·18 0·00
  Φ = a + bF t-1 t 419·247·06
Figure 3.

Average summer deer mouse population densities on the control grid for the (a) full, (b) overwintered (OW) and (c) young-of-the-year (YY) populations in relation to the previous fall's subalpine fir cone production, the best predictor in models examining annual variation in average summer population densities between 2002 and 2011.

In the most parsimonious CJS model (AICc weight = 0·61), summer survival of OW mice was dependent on both the previous fall's white spruce and subalpine fir cone production (Appendix S4: Table S3). The probability of survival between trapping sessions was inversely related to previous spruce cone production (β = −0·12 ± 0·04, 95% CI(β) = −0·19, −0·05), suggesting lower survival of OW mice after heavy spruce cone crops and positively related to previous fir cone production (β = 0·10 ± 0·06, 95% CI(β) = 0·00, 0·21), although the confidence interval overlapped zero. The second-ranked model also had substantial support (ΔAICc = 1·54, AICc weight = 0·28; Appendix S4: Table S3), but only contained a negative relationship between summer survival and the previous fall's spruce cone index (β = −0·08 ± 0·03, 95% CI(β)  = −0·14, −0·03). Using the summer population growth rate of OW mice as an index of survival from spring to fall, annual variation in survival over the entire breeding season was not explained by spring population density or the previous fall's cone production (Table 1).

Average spring and fall masses of OW males were not related to the previous fall's conifer cone production (Appendix S4: Table S4). However, the final backwards stepwise regression model explaining annual variation in average summer masses of OW males contained a positive relationship with the previous fall's fir cone index (β = 0·02 ± 0·01, 95% CI(β)  = 0·00, 0·05) and was nearly statistically significant (R2 = 0·33, n = 10, P = 0·08); this trend suggests that OW males were heavier in the summers after heavy fir cone crops. The average mass at first capture and growth rate of YY males were not associated with the previous fall's conifer cone indices (Appendix S4: Table S4).

Annual variation in the timing of breeding and length of the breeding season were also not explained by the previous fall's cone production (Appendix S4: Table S4). However, annual variation in the proportion of pregnant resident OW females was related to previous conifer seed production (final multiple stepwise regression model including both spruce and fir cone indices: R2 = 0·74, n = 10, = 0·01), with most of the variation explained by the previous fall's fir cone production (β = 0·21 ± 0·05, 95% CI(β) = 0·10, 0·33; partial R2 = 0·72, n = 10, = 0·003), but an inverse relationship with previous spruce cone production (β = −0·10 ± 0·03, 95% CI(β) = −0·18, −0·03; partial R2 = 0·37, = 10, = 0·02). No other breeding parameters were associated with previous cone production by either conifer species (Appendix S4: Table S4).

Discussion

Deer mouse demography was not positively affected by white spruce mast seeding, despite all contrary indications from individual-level experiments (Lobo, Duong & Millar 2009; Lobo & Millar 2011; Lobo 2013) and our simulated spruce masting event. Conversely, we observed an unexpected delayed effect of subalpine fir mast seeding, where increased fall fir seed production did not influence overwinter or spring mouse demography, but instead enhanced summer survival, body masses and pregnancy rates of OW adults. This led to increased summer population densities (full, OW, and YY) in the year following fir mast seeding. Previous total cone production was not a consistent strong predictor of annual variation in population, body mass or breeding parameters, highlighting the particular significance of fir seeds in this consumer–resource pulse system.

Effects of Seed Supplementation on Mouse Demography

Examining the long-term dynamics of the control and treatment populations (2004–2011) allowed us to compare the two populations post-supplementation (2011) within the context of their long-term relationship, to identify the effects of seed availability on both the density and the mechanisms driving population responses (see Appendix S5 for supplementary discussion of grid effects in population comparisons). Our results indicate that increased availability of spruce seeds to mice during a fall masting event has the ability to enhance populations in the following spring, summer and fall. In eight years of population monitoring, densities on the treatment grid were only consistently higher than the control grid after supplementation with spruce seeds. The biggest impact of increased seed availability was on overwinter survival, which did not differ between the two grids in any year prior to supplementation but was significantly enhanced on the treatment grid after supplementation.

Reproduction by deer mice is energetically demanding and highly seasonal in our study area (Millar 1979), and the initiation (Desjardins 2002) and cessation (Tabacaru et al. 2010) of their breeding season appears to be related to the availability of animal protein. While the crude protein content of spruce seeds used in this experiment was high (Appendix S2), the low digestibility and amino acid content of plant protein, compared with animal protein, can reduce the amount of assimilable protein available to rodents (Robbins 1993); this may reduce the efficacy of spruce seeds in extending or advancing the breeding season of deer mice in our population. Furthermore, supplementation, and consequently the highest seed availability, occurred in the late fall, after mice had already ceased breeding. Seeds would have been cached and consumed over the winter (Barry 1976), so their availability at the end of winter and in the early spring may not have been sufficient to support earlier and increased reproductive activity by these income breeders (Millar 1979). The timing of supplementation may also explain why spring body masses were not enhanced after seed addition, because a single late fall supplementation event would not provide the same continuous excess of food as in prolonged winter supplementation experiments that have led to increased body masses of rodents (see Appendix S5).

Although individual breeding patterns were not affected by spruce seed supplementation, we still observed more YY mice on the treatment grid than the control grid in the summer after supplementation. Supplemented females did not breed earlier or produce more litters than control females, but the higher OW female density on the treatment grid would nevertheless lead to more YY being produced, which is the main contributor to summer and fall populations. Thus, even if the only effect of increased fall spruce seed availability on mice is to improve overwinter and summer survival, masting still has the potential to significantly enhance summer densities through overall increased juvenile production. See Appendix S5 for supplementary discussion on YY population dynamics.

Population Responses to Natural White Spruce Mast Seeding

We expected to observe a lagged positive relationship between white spruce mast seeding and deer mouse population densities, given the results of previous (Lobo, Duong & Millar 2009; Lobo & Millar 2011; Lobo 2013) and current (seed supplementation) experimental studies. However, mice are not the only vertebrate seed predators in our study area. The North American red squirrel is the dominant pre-dispersal seed predator in our study area and feeds primarily on spruce seeds, with individuals annually harvesting and hoarding several thousands of spruce cones prior to cone opening (Smith 1968; Fletcher et al. 2010; Donald & Boutin 2011). Although a large proportion of cones produced on a squirrel's territory can escape hoarding, the degree of cone harvesting is highly variable among individuals and territories, with many spruce trees experiencing complete predation even in mast years (Fletcher et al. 2010; Archibald et al. 2012). Given the efficacy of red squirrels as a pre-dispersal seed predator and their strong influence on the evolution of white spruce mast seeding (LaMontagne & Boutin 2007; Fletcher et al. 2010; Archibald et al. 2012), we hypothesize that pre-dispersal cone harvesting by squirrels prevented spruce masting from effectively enhancing the food supply available to mice in the fall and winter, thereby mitigating mouse population responses to heavy cone crops.

Most of the recent research on white spruce–red squirrel interactions has been conducted in the spruce-dominated forests of the Yukon, Canada (e.g. Boutin et al. 2006; LaMontagne & Boutin 2007; Fletcher et al. 2010; Donald & Boutin 2011; Archibald et al. 2012). While we have not extensively examined the behaviour of squirrels in our study area, considerably south of the Yukon, indirect evidence suggests that cone harvesting plays a major role in limiting the amount of spruce seed rain produced. First, we found considerably less variation in average spruce seed rain density (CV = 1·01) compared with average cone production (CV = 1·72) during the same period (2006–2010; Appendix S1); this discrepancy is likely the result of significant cone harvesting prior to opening. Second, the highest spruce seed rain density recorded from any of our seed traps was 1·85 g m−2, which occurred during the fall/winter of 2007, a bumper cone crop year (Fig. 1 and Appendix S1). Based on previous estimates of spruce seed mass (2·0 mg; Greene & Johnson 1994) and the number of sound spruce seeds per cone (48; Beaulieu, Deslauriers & Daoust 1998), this is roughly the equivalent of seed rain from 19·27 cones m−2. However, individual white spruce trees can produce between 8000 and 12 000 cones in a mast year (Nienstaedt & Zasada 1990), and our average cone count (top 3 m of the tree) in 2007 was 99·00 ± 34·15 tree−1 (range = 4–342); therefore, at least 9·50 ± 3·28 g of seed tree−1 would have been produced if no cones were harvested prior to opening. Seed production of this magnitude should have led to significantly higher seed rain density measurements if pre-dispersal cone harvesting was not extensive.

A key assumption of our initial hypothesis that spruce masting would positively affect mouse populations was that a sufficient amount of the seeds produced during a heavy cone crop would actually end up on the ground for mice to actively forage for them and cache them, to use the seeds as a major food resource during the fall and winter. In a previous study (Lobo 2013), we determined that the lower threshold density at which rodents categorically cease active foraging (both consumption and caching) for spruce seeds was 5·60 ± 0·61 g m−2; this value is three times higher than the highest seed rain density recorded from our seed traps (Appendix S1: Fig. S2). However, the amount of spruce seeds produced tree−1 in a mast year, prior to cone harvesting, would have easily surpassed this lower threshold density and would have been sufficient to produce seed rain at the density of seeds supplemented on the treatment grid; this suggests that a sufficient amount of seeds would have dispersed during the seed rain period in a mast year to effectively regulate mouse populations if pre-dispersal cone harvesting was less severe. These results indicate that white spruce seed production can provide a fall and winter food supply that meets or exceeds the energy requirements of deer mice and that spruce masting has the potential to enhance mouse populations, but under natural conditions, population responses to variable spruce seed production are not observed because of intervention by red squirrels. This hypothesis can be further tested in the field through long-term, intensive trapping and removal of squirrels from study areas.

White spruce–red squirrel interactions may have also indirectly contributed to the decreased survival and pregnancy rates of OW mice observed in the summers following heavy spruce cone crops (see Appendix S5 for supplementary discussion on negative relationships between mouse demography and spruce seed production). Squirrel populations peak following spruce mast seeding (Smith 1968; Boutin et al. 2006), which can lead to enhanced populations of generalist predators, and consequently increased predation on alternative prey such as deer mice (see Ostfeld & Keesing 2000). However, these effects were relatively minor in our study, as no other demographical parameters, including population density, were negatively related to previous spruce cone production.

Population Responses to Subalpine fir Mast Seeding

The strong relationship that we observed between subalpine fir seed production and deer mouse densities was unexpected. However, the population response to fall fir mast seeding was delayed, as no effects on overwinter survival or spring body mass, breeding and density were observed. Instead, survival, body masses and pregnancy rates of OW mice were higher in the summer following a heavy cone crop, leading to increased summer OW and YY densities (see Appendix S5 for supplementary discussion on fall population dynamics). One possible explanation for this delayed response is that mice avoided fir seeds in the fall and winter, but consumed them in the spring, after the PSCs in the seeds degraded. However, this is unlikely, given that PSCs in seeds are resistant to degradation over time (e.g. Shimada 2001), and the low nutritional value of fir seeds (Lobo & Millar 2011) is probably insufficient to enhance survival and breeding in rodents. It is more likely that the delayed effect of subalpine fir masting on deer mouse populations was mediated by population responses of invertebrate seed predators.

Invertebrate pre-dispersal predators of subalpine cones (e.g. Dioryctria abietivorella, Earoymia spp., Megastigmus spp.) are common and have been well studied, but are unlikely to have produced the delayed population response that we observed. First, the adults are volant, so they would mostly be inaccessible to mice. Second, most species overwinter as larvae either on the ground or within seeds on the ground and would be accessible to mice in the fall/winter, causing an overwinter and spring population effect (Kulhavy, Schenk & Hudson 1976; Hedlin et al. 1980). Instead, population responses of terrestrial invertebrate post-dispersal seed predators are more likely to have mediated the enhanced summer survival, breeding and densities of mice that we observed, although the identity and impacts of these species are largely unknown. Based on reports of terrestrial invertebrate conifer seed predators (see Nystrand & Granström 2000; Lundgren 2009) and invertebrates caught during pitfall trapping in our study areas (N. Lobo & J. Millar unpublished), the most likely candidates are carabid beetles.

Carabid beetles have been historically reported as obligate predators in the literature, but their granivorous nature is underestimated, with adults and larvae of several genera (e.g. Agonum, Amara, Calathus, Harpalus, Pterostichus, Stenolophus) consuming seeds as part of their diet to varying degrees (see Tooley & Brust 2002; Lundgren 2009; Kotze et al. 2011). Species in the genera Amara and Harpalus are the most granivorous (Tooley & Brust 2002), and some consume large amounts of conifer seeds (Dick & Johnson 1958; Johnson, Lawrence & Ellis 1966; Nystrand & Granström 2000; references therein), but many are only prevalent in open habitats, weedy patches and agricultural areas (Lindroth 1968; Tooley & Brust 2002). However, some of the species that are known to consume conifer seeds (e.g. Harpalus cautus) are abundant in forested areas in western Canada (Hatch 1958). Other major post-dispersal conifer seed predators in forested areas include species in the genera Pterostichus (Dick & Johnson 1958; Johnson, Lawrence & Ellis 1966; Nystrand & Granström 2000; Côté, Ferron & Gagnon 2005) and Calathus (Nystrand & Granström 2000).

Previous research has shown that population growth of deer mice may be linked to arthropod abundance (Simard & Fryxell 2003), and we suggest that summer mouse populations in our study may have been enhanced by increased spring and summer populations of carabid larvae and adults following subalpine fir masting. Reproduction, and consequently populations, of carabid beetles are food limited (Lövei & Sunderland 1996; Toft & Bilde 2002; Kotze et al. 2011), and many native carabid species in Alberta (the province where this study was conducted), including forest-dwelling granivorous species (e.g. Pterostichus adstrictus), are spring breeders (Niemelä, Spence & Spence 1992). These species may increase reproductive output in response to the enhanced supply of fir seeds available to them in the spring following a masting event. Given that arthropods are a preferred food source of deer mice (Bellocq & Smith 1994), and animal protein is critical for their reproduction (McAdam & Millar 1999; Desjardins 2002; Tabacaru et al. 2010) and that carabid beetles are common in our study area (rank 3·33 ± 0·67 of 31, based on total numbers of individuals from insect families caught during pitfall trapping at three of our study sites in the summer of 2009; N. Lobo and J. Millar unpublished), increased spring/summer populations of carabid larvae and adults could enhance summer survival, body mass and breeding of mice. Most research on invertebrate predation of conifer seeds has focused on pre-dispersal cone and seed predators, but post-dispersal granivory by invertebrates may play an important role in influencing rodent population fluctuations in coniferous forests, which could have further direct and indirect implications for interactions at other trophic levels (Ostfeld & Keesing 2000). Further examination of the granivorous nature of carabid beetles in coniferous forests, and the relationship between their population fluctuations and conifer masting, would provide significant insights into the validity of this hypothesis.

Conclusion

Overall, rodent population responses to mast seeding appear to be more complex in northern coniferous forests than in temperate deciduous environments and may be influenced by significant pre-dispersal seed predation, indirect/delayed effects on food availability and constraints on population growth and variability (see Appendix S5 for supplementary discussion on external constraints on northern rodent populations). Rodents are major conifer seed predators and have the potential to significantly reduce seed survival and recruitment (Radvanyi 1970; Peters et al. 2004), but our results suggest that the predator satiation hypothesis for the evolution of mast seeding is unlikely to apply to conifer–rodent interactions. The characteristic booms and busts of rodent populations in response to deciduous seed crops were replaced by delayed, low-amplitude fluctuations in our study. Mouse populations in this study were also considerably lower than those often documented in deciduous forests (e.g. Wolff 1996; McCracken, Witham & Hunter 1999; McShea 2000; Falls, Falls & Fryxell 2007).While rodent population responses to deciduous masting have important direct effects on their predators and prey (McShea 2000; Ostfeld & Keesing 2000), similar ecosystem interactions may be more subtle, indirect and convoluted in northern coniferous forests.

Pre-dispersal cone harvesting by specialist red squirrels has significantly influenced the evolution of mast seeding by white spruce trees (Fletcher et al. 2010; Archibald et al. 2012) and appears to prevent spruce masting from influencing generalist rodent populations in our study area. However, in white spruce forests without red squirrels, it is unknown to what degree mast seeding is utilized to satiate major seed predators such as rodents. Given the large resource investment required for mast seeding (Kelly & Sork 2002) and the efficacy of PSCs in deterring seed predation by rodents, masting may be less prominent as a defence strategy in areas without red squirrels, with spruce trees in these areas relying more on PSCs. The geographical mosaic of co-evolution (see Benkman, Holimon & Smith 2001) between white spruce and its major vertebrate seed predators is an exciting avenue for future research.

The particular but indirect importance of fir seeds, an unpalatable food source, to rodent population dynamics highlights the significance of multitrophic community dynamics in this system. Large-scale field manipulations and intensive long-term demographical studies of other pre- and post-dispersal seed predators are required to further elucidate the key interactions regulating rodent populations in coniferous forests that have been proposed in this study and to evaluate the importance of specific pulsed resources on interactions between multiple consumers. This system is ideal for the large-scale, integrative, multidisciplinary ecosystem studies that ecologists are encouraged to pursue (e.g. Krebs, Boutin & Boonstra 2001).

Acknowledgements

We are grateful to the BC Ministry of Forests and Range Tree Seed Centre for donating the white spruce seeds used in this study, the graduate students and research assistants who helped collect data in the field, and L. Zanette and two anonymous reviewers for comments on previous versions of the manuscript. Funding was provided by the Natural Sciences and Engineering Research Council of Canada. All procedures were approved by Animal Use Protocols from the University of Western Ontario Animal Use Subcommittee (protocol nos. 2004-026, 2008-001) and annual Research Permits and Collection Licenses from Alberta Sustainable Resource Development Fish & Wildlife Division.

Ancillary