The roles of external factors such as predation and food in population regulation have puzzled population biologists for decades (Elton 1949). Food supplementation has been shown to affect positively survival and breeding across vertebrate taxa (see Boutin 1990; for review), and predation is one of the most widely studied processes negatively affecting prey populations (Krebs et al. 2001; Schmitz 2008). The mere risk of predation is known to affect behaviour, morphology and physiology of the prey individuals (Lima & Dill 1990; Lima 1998, 2002). Risk effects arise when a prey alters its behaviour in response to predation risk, and these responses carry survival or reproductive costs (Boonstra et al. 1998; Preisser, Bolnick & Benard 2005; Creel & Christianson 2008).
Predation risk effects on prey fitness are cascade-like processes. A common response of a prey is to avoid contacts with the predator by reducing activity or movement areas. Reduced activity and searching for food restricted to smaller area will lead to a reduced foraging efficiency and consequently may influence individual condition. In addition, predator avoidance by decreased activity during reproductive period may lead to missed encounters with potential mates (Brown 1988) or otherwise disturbed sexual encounters (Ronkainen & Ylönen 1994) and hence to lowered fitness. A lowered body condition can affect the breeding success by causing smaller and/or lower quality litters. It is also possible that the onset of breeding after the non-breeding season may be delayed and even cause a reduction in the number of litters produced during the shortened breeding season (Eccard & Ylönen 2001).
Suppressed or delayed breeding as an adaptive strategy enhancing individual (female’s) reproductive success, as proposed by Ylönen & Ronkainen (1994), has been tested experimentally and its optimality has been modelled in several studies. The main message is that under best breeding season conditions with all individuals in ‘moderate to high reproductive performance’ (Kokko & Ranta 1996), the strategy of not breeding is not optimal regardless of strength of predation pressure. This has also been proven in experimental populations (Mappes, Koskela & Ylönen 1998) under simulated predation risk during boreal summer. However, the theoretical approaches by Kokko & Ranta (1996) and Kokko & Ruxton (2000) leave the possibility open that some additional restrictive conditions are necessary before delayed breeding could be optimal. The same phenomenon of possible additive nature of predation risk with other breeding constraining factors like food limitation or high conspecific density stress was suggested also by Hansson (1995) and Ylönen, Koskela & Mappes (1995). The few studies having verified breeding delay in the field also come from this kind of restrictive conditions, from sub-alpine environment (Fuelling & Halle 2004) and from the latter half of the breeding season with high conspecific densities (Jochym 2011).
In northern areas, winter survival and onset of breeding are constrained by the kind of restrictive conditions described by Kokko & Ranta (1996) and Kokko & Ruxton (2000). Winter season is characterized by diminishing food resources and higher energetic demands for overwintering ground-dwelling animals because of low temperatures combined with high moisture in the sub-nivean space. On the other hand, the long-lasting permanent snow cover protects ground-dwelling small rodents from most predators during winter. Under the snow, the greatest predation risk threat is caused by small mustelids, especially by the least weasel (Mustela nivalis nivalis), which is small enough to hunt voles in their tunnels and cavities of the sub-nivean space. On the snow surface, dangers are more diverse and are caused by generalist mammalian predators and owls. In the sub-nivean space, vole predation is predictable. If there is a hunting small mustelid like the weasel in the area, there probably are also signs of its activity in the form of weasel scent (Erlinge, Sandell & Brinck 1982). On the other hand, lack of weasel odour cues under the snow should reliably indicate a low predation risk. Many of the experiments conducted during the breeding season using predator odour have failed to find effects of experimentally increased risk on breeding or spacing behaviour (Wolff & DavisBorn 1997; Jonsson, Koskela & Mappes 2000; Hellstedt, Kalske & Hanski 2002). However, it is likely that under the snow, the odour remains present for a longer period of time than in open air, because air circulation that increases the evaporation of volatile compounds in the weasel scent (Brinck, Erlinge & Sandell 1983) is practically absent. In addition, the reduced activity of microbial decomposers during cold climate should increase the time when volatile compounds are present.
In a strongly seasonal environment, overwintering rodent populations decline from the autumn population highs to next spring. Local populations may go also extinct, which has been shown in both experimental field populations (Aars & Ims 2002) and in enclosure experiments (Eccard et al. 2011a). Survival over the non-breeding season, gaining energy for physiological changes needed for hormonal changes during maturation after winter (Galea & McEwen 1999) and optimal timing of first reproduction are major determinants of individual fitness (Roff 1992; Eccard & Ylönen 2001). When to start breeding is the key question for iteroparous income breeders such as the bank vole (Myodes glareolus). Individuals may gain significant fitness benefits by starting to breed earlier than competing conspecifics, if the successfully weaned offspring are among first ones to start the next cohort breeding (Millar & Gyug 1981; Ergon 2007).
Overwinter survival, individual condition measured as weight change from the early winter weight to spring weight, as well as onset of breeding, have been shown to be strongly improved by food supplementation (Boutin 1990; Ylönen & Eccard 2004; Von Blanckenhagen, Eccard & Ylönen 2007). Studies on predation risk effects in winter are rare (see Krebs et al. 1995), but even if information would exist, it would be confounded by more plausible explanation of food limitation during the harsh non-breeding season (Creel & Christianson 2008). However, both predation risk and food can limit populations as suggested by Krebs et al. (1995) in their experimental case of the snowshoe hare (Lepus americanus).
Studying direct predation impact experimentally during winter is not possible. Thus, we conducted for the first time an experiment in 20 large outdoor enclosures to test the effects of either long-lasting predation risk (weasel odour under the snow) or continuous high quality food supplementation and their interaction for overwintering success and onset of breeding in bank voles. More importantly, we measured direct fitness effects of fear of predation on the population scale during the snowy period of winter when vole populations decline strongly or even go locally extinct (Aars & Ims 2002). Predation risk is not expected to cause direct mortality effects (cf. Preisser, Bolnick & Benard 2005), but could affect condition and even survival by indirect interactions with decreased activity and foraging, increased stress and vigilance and predator facilitation if risk under snow drives individuals on the snow and exposed to direct predation. We hypothesized that:
- 1 Vole condition and survival are negatively affected by risk of predation and positively by food supplementation.
- 2 Food supplementation advances breeding and predation risk delays it.
As we predict opposite effects of food and predation risk, an open question of the strength of these factors for individual performance remains in populations where both treatments run simultaneously.
Materials and methods
The experiment was carried out in 20 enclosures of 0·25 ha (50 m × 50 m) in size, fenced with metal sheets that reach 0·5 m below ground and c. 0·7–1 m above the ground preventing voles from escaping. Enclosures were built into old fields where vegetation was mainly tall grassland with some Salix spp. bushes. Bank voles commonly live in this kind of habitat in the absence of competition from larger species. In each enclosure, there were 25 Ugglan special multiple capture traps (Ugglan special, Grahnab ab, Hillerstorp, Sweden) in 10-m intervals in a 5 × 5 grid. The traps were covered with bottomless metal chimneys that allowed food and predation risk manipulation under the snow and trapping in snowy conditions.
The bank vole is small, common boreal microtine rodent. The breeding season of bank voles usually begins at the end of April and lasts until September. During the non-breeding season, bank voles are assumed to overwinter in small aggregations in winter nests (West & Dublin 1984; Ylönen & Viitala 1985). During the breeding season, female bank voles are strictly territorial and need an exclusive territory to breed (Bujalska 1985). Breeding male territories overlap with several female territories (Ims 1987).
The least weasel is a small mustelid specialized on hunting small mammals. Weasels maintain a kind of foraging territory during the breeding season (Erlinge 1974, 1975), but there is little information of weasel movements or spacing behaviour during winter (Nyholm 1959), but weasels hunt both on and under the snow. Populations of bank voles, like other small vole species fluctuate cyclically in boreal landscapes (e.g. Hanski et al. 2001). Weasel populations follow the vole cycles with an approximately half-year time lag (Korpimäki, Nordahl & Rinta-Jaskari 1991).
Experimental voles were bred in captivity and housed individually after weaning age (20 days) in standard mouse cages with wood shavings and dry hay as bedding material in the Konnevesi Research Station’s animal house. Water and food in the form of rodent pellets (LabFor) were available ad libitum. Light regime in the laboratory was 18L: 6D during summer and 12L: 12D from September onwards. At the beginning of the experiment, voles were 2–5 months old and marked individually with ear tags (Hasco Tag Company, Dayton, KY, USA, type 1005-1). Five female and five male bank voles were released into each enclosure during the second week of October. Weight and age distribution of the voles was equalized for all enclosure populations. No siblings or half-siblings were released into the same enclosure. Three weeks before releasing, the vole cages were brought to outdoor shelter to let voles acclimatize to the natural day length and temperatures.
It is commonly known that late autumn mortality before onset of permanent snow cover is high, especially because of avian predation before the onset of snow cover. Furthermore, increased stress levels of captive bred animals to the nature (Eccard, Jokinen & Ylönen 2011b) can increase mortality considerably. To compensate for this early mortality, we trapped enclosures on the 15th of November to assess vole post-release survival. Male survival from the release was lower than female, which was expected based on an earlier experiment (Klemme, Ylönen & Eccard 2008). At the end of this trapping session, vole densities in the enclosures were standardized by relocating voles from higher densities to lower-density enclosures, to give equal starting densities of three to four females and three males in each enclosure.
Food supplementation treatment
Food supplementation in the form of protein muffins was added biweekly in half (10) of the enclosures. Protein muffins were made of 4·5 g of hera80® protein and 15 g sunflower seed. These were mixed into a paper muffin mould and attached to a solid form with gelatin liquid. Ingredients for the muffins were chosen based on earlier studies, so that they included both plant (Ylönen & Eccard 2004) and animal protein (Von Blanckenhagen, Eccard & Ylönen 2007). Muffins were spread in 12 (every second) trap chimneys of the food supplementation enclosures. Food supplementation started in mid-November and lasted until the second week of April. Supplemental food was added to the enclosures 11 times, and altogether 132 protein muffins were added per enclosure. Each time new muffins were added, and the possible remnants from the previous supplementation were removed. As the food remnants were scarce, because of moderate food supplementation rate c. 2·5 g day−1 including husk with initial vole density, we did not weigh the remnants. Therefore, no direct consumption by voles was estimated. To give some reference value, bank voles living in rodent cage consumed 2·7 g of high protein and 3·5 g of low protein rodent pellets (S. Sipari, M. Haapakoski, I. Klemme, J. Sundell & H. Ylönen, unpublished) in climate-adjustable laboratory at 8 °C.
Predator risk treatment
Predator odour was made from c. 1·2 kg of weasel latrine beddings containing in average 330 g faeces that were collected from cages of a weasel colony housed at the Konnevesi Research Station. Faeces of many individuals and both sexes were mixed with 5 L of water and with c. 28 L of dry wood shavings in a 30-L bucket. The bucket was sealed with an air-tight lid and stored overnight at temperature of 4 °C. This was performed to get the smell and moisture evenly distributed into the wood shavings. The mixture was spread the following day. For all biweekly spreading of weasel odour, the mixture was prepared fresh. As control treatment, we used clean wood shavings mixed with water in same proportion than predator smell to make its consistency similar to that in predator treatment enclosures. One cup (c. 0·1 L) of the predator odour and control odour was spread biweekly in each trapping chimney in half of the enclosures. The weasel odour liquid was prepared with the same amounts of faeces and water like in summer studies where the odour then was sprayed to the vole habitat (e.g. Mappes, Koskela & Ylönen 1998). However, because of winter temperatures and avoiding freezing of ground in trap chimneys, we choose to use the mix of wood shavings and weasel scent. The amount of manipulated wood shavings for each trap chimney, probably, was as treatment weaker than the summer sprayings, but remained longer in confined sub-nivean space.
For all treatment populations, the avian predation risk caused by resident pygmy owls (Glaucidium passerinum) and Ural owls (Strix uralensis) was similar. Both owl species were seen, or hunting signs of them on snow cover were frequently found throughout the winter. Presence of pygmy owls was also evident in fresh caches from November in the nest boxes in the vicinity of the enclosure area.
Overall, the experiment was conducted in 20 enclosures: five enclosures with predator odour and without food supplementation (P+F−), five with food supplementation (P−F+), five with predator odour and food supplementation (P+ F+) and five enclosures with control odour and without food supplementation (P−F−). We had to exclude one enclosure (P−F+) from the analysis because the fence was damaged and field voles (Microtus agrestis) invaded the enclosure.
Voles were trapped over one to two nights including two to three trap checks starting in February 21st, March 13th, March 30th, April 18th, May 9th and May 30th to follow vole survival, body mass and breeding condition. During trapping, traps were baited with a fifty–fifty mixture of sunflower seeds and oats. When traps were set, supplemental food was removed from the trap chimneys. Traps were emptied of the seeds and food, and predation treatments were always applied after trapping session. During each trapping session, voles were weighted with a Pesola scale to the nearest 0·5 g and their breeding status was recorded using similar method than Ergon, Lambin & Stenseth (2001), that is, females were defined as mature and capable for breeding when their vagina was perforated and males were regarded mature when their testes were in a scrotal position.
For data analysis, we used mixed effects models and library nlme (Pinheiro et al. 2010) in r (R Development Core Team 2010). Model selection was conducted by choosing the best model from the set of predefined models based on AIC (see Tables S1–S5, Supporting Information). Maximum likelihood method was used for comparing models and restricted maximum likelihood (REML) in the final model to obtain the model estimates. Based on initial evaluation (Zuur et al. 2009) and to account for pseudoreplication, enclosure was used as a random factor in all analysis. Predation risk, food supplementation and sex with all two-way interactions were used as fixed factors in the full model in the analysis.
Only individuals larger than 15 g in February were included in the analysis of the onset of breeding. Three female voles, one from P+F+ and two from P−F+, were lighter than 15 g in February and were excluded from the onset of breeding analysis. The onset of breeding was measured as the number of days after observing the first mature individual (open vagina in females, scrotal testes in males) in any enclosure. Vole condition was measured as a weight change from November to February. Weight change, although it does not measure directly condition, can be used as proxy for condition because it can be manipulated with supplemental food and it clearly affects breeding success after winter (Ylönen & Eccard 2004). All gravid females in February (n = 4) were excluded from condition analysis.
We analysed onset of breeding also by the date when 50% or more of individuals in the each enclosure were mature to be sure that our result is not based on the outliers in the data (Ergon et al. 2011). A two-way anova was used in this analysis with food, predation and the interaction as fixed factors. The interaction model was the full model, and all combinations until null models were run. Best model was chosen by lowest AIC. Both sexes were analysed first together, and then females and males were analysed separately.
For survival analysis, we used r (R Development Core Team 2010) package rmark (Laake & Rexstad 2008) to construct models for program mark (White & Burnham 1999). Survival analysis of different treatments was compiled by Cormac-Jolly-Seber (CJS) models with all the enclosures pooled, and it was conducted only with the founder voles. The CJS models allow an independent estimation of apparent survival (Φ) and recapture probabilities (P). We fitted survival and recapture models as functions of predation risk, food supplementation and time with their two-way interaction and sex without interaction (see all the candidate models from electronic supplementary material). We were studying the model fit/overdispersion of the most complex model with program RELEASE (Burnham et al.1987) test 2 and test 3. Models with ΔAICc values <2 (Burnham & Anderson 2002) were selected for model averaging to get survival and recapture probability estimates and standard errors.
We also monitored overall vole density in enclosure. The expectation was a general decline of adult densities. However, because of snow cover in some places reaching the height of the fence, and due wind-formed temporal snow bridges at the fences, there was possibility for some immigration from outside to the enclosures. This was only the case in the one food treatment enclosure, which attracted eight immigrant bank voles between November and February. These individuals may be seen behind the mid-winter increase in adult densities in the food treatment, but they were excluded in all other analyses. Density was analysed from capture data with robust design model in R (R Development Core Team 2010) with package Rcapture (Baillargeon & Rivest 2007). Chao m0 model was used to get the density estimates for each trapping session for each enclosure. In the density analysis, we used log-transformed densities. Trapping session within each enclosure was used as random factor to take account for the temporal pseudoreplication (Crawley 2007), and time was added as fixed variable. We analysed both sexes pooled, that is, total number of voles estimated to be alive in the enclosure.