Indirect effects of a large mammalian herbivore on small mammal populations: Context‐dependent variation across habitat types, mammal species, and seasons

Abstract Multiple consumer species frequently co‐occur in the same landscape and, through effects on surrounding environments, can interact in direct and indirect ways. These interactions can vary in occurrence and importance, and focusing on this variation is critical for understanding the dynamics of interactions among consumers. Large mammalian herbivores are important engineers of ecosystems worldwide, have substantial impacts on vegetation, and can indirectly affect small‐mammal populations. However, the degree to which such indirect effects vary within the same system has received minimal attention. We used a 16‐year‐old exclosure experiment, stratified across a heterogeneous landscape, to evaluate the importance of context‐dependent interactions between tule elk (Cervus canadensis nannodes) and small mammals (deer mice [Peromyscus maniculatus], meadow voles [Microtus californicus], and harvest mice [Reithrodontymys megalotis]) in a coastal grassland in California. Effects of elk on voles varied among habitats and seasons: In open grasslands, elk reduced vole numbers during fall 2013 but not summer 2014; in Lupinus‐dominated grasslands, elk reduced vole numbers during summer 2014 but not fall 2013; and in Baccharis‐dominated grasslands, elk had no effect on vole numbers in either season. Effects of elk on the two mice species also varied among habitats and seasons, but often in different ways from voles and each other. In fall 2013, elk decreased mice abundances in Lupinus‐dominated grasslands, but not in Baccharis‐dominated or open grasslands. In summer 2014, elk decreased the abundance of harvest mice consistently across habitat types. In contrast, elk increased deer‐mice numbers in open grasslands but not other habitats. Within the same heterogenous study system, the influence of elk on small mammals was strongly context‐dependent, varying among habitats, mammal species, and seasons. We hypothesize that such variability is common in nature and that failure to consider it may yield inaccurate findings and limit our understanding of interactions among co‐occurring consumers.

The indirect effects of one plant consumer on another will likely vary greatly in occurrence and importance, depending on a range of temporal and spatial factors in the landscape. Numerous experiments have revealed the importance of context-dependent variation in the outcome of indirect interactions among consumers (Maclean, Goheen, Doak, Palmer, & Young, 2011;Pringle et al., 2007). For example, differences in soil type can influence the nature of indirect interactions between consumer species by altering the susceptibility of a shared host plant to herbivory (Brown et al., 2001;Cobb et al., 1997). Biotic factors, such as the introduction or range expansion of a non-native plant species, can also alter the indirect interactions between consumers in the invaded areas (Pearson, 2009). The consequences of indirect interactions can also depend on climatic conditions that shift over time (Brown et al., 2001;Heske, Brown, & Mistry, 1994;Long, Wambua, Goheen, Palmer, & Pringle, 2017). This variability in outcomes, combined with the rapidly changing global climate (Rosenzweig et al., 2008) and the increasing prevalence of exotic species in ecosystems (Hellmann, Byers, Bierwagen, & Dukes, 2008;Rosenstock, 1996), make long-term experiments of critical importance for examining context-dependent interactions among consumers. Failure to consider such variability may limit our understanding of interactions among co-occurring consumer species (Chamberlain, Bronstein, & Rudgers, 2014).
Host-plant-mediated indirect interactions among consumers have been shown to arise among insects, between mammals and insects (Ohgushi, 2005;Wilkerson, Roche, & Young, 2013), and between large and small mammalian consumers (Parsons, Maron, & Martin, 2013;Smit et al., 2001). Although multiple consumers frequently co-occur across a range of habitat types, few studies have examined if and how such environmental heterogeneity alters the outcome of their interactions. The outcome of an interaction between species may remain fairly consistent across different habitat types, or it may be strong in one habitat and weak, absent, or reversed in another. Thus, stratifying ecological experiments across multiple habitat types is critical for assessing the importance of variability in the outcome of interactions between co-occurring consumers (Foster, Barton, & Lindenmayer, 2014).
In this study, we used a 16-year-old experiment, stratified across three habitat types, to examine the importance of and variation in indirect interactions between tule elk (Cervus canadensis nannodes) and three co-occurring species of small mammals-deer mice (Peromyscus maniculatus), meadow voles (Microtus californicus), and harvest mice (Reithrodontymys megalotis). We addressed three questions: (a) Do elk alter the height and biomass of herbaceous vegetation that can be important to small mammal populations? (b) Do elk alter the densities of small-mammal populations, and do these effects vary among habitat types, small mammal species or over time?
(c) By altering the population density of small mammals, do elk indirectly affect seed predation rates, and do these effects vary among habitat types? Through their activities as herbivores and disturbance agents, we predict that elk will reduce the amount of vegetation and in turn reduce the abundance of small mammals. We further predict that these effects will vary substantially among habitat types, with the influence of elk on small mammals being more pronounced in open grasslands than in shrub-dominated grasslands that provide greater protection from predators. Finally, we predict that, by reducing small-mammal populations, elk will reduce seed predation rates for a dominant nitrogen-fixing shrub.

| Study system
We performed this study on Tomales Point, a 1,030 ha peninsula that is part of Point Reyes National Seashore, 65 km northwest of San Francisco, California, USA. The vegetation of Tomales Point is a mosaic of shrub-dominated coastal grassland and scrub, interrupted by steep canyons containing dense riparian shrubs (Lathrop and Gogan (1985). Three distinct habitat types occur within our 300-ha study area: Baccharis-dominated grasslands, Lupinus-dominated grasslands, and open grasslands. Open grasslands occur on the Kehoe soil formation (derived from Cretaceous granitic parent rock; (Kashiwagi, 1985) and are dominated by herbaceous species and largely devoid of shrubs (Johnson & Cushman, 2007). Baccharisdominated grasslands occur on a subvariant of the Kehoe formation (Kashiwagi, 1985) and are characterized by herbaceous-dominated patches mixed with dense stands of Baccharis pilularis (Asteraceace), a long-lived native shrub. Lupinus-dominated grasslands are located on a mix of soil formations, either completely in Sirdrak sand (derived from a Quaternary dune sandstone parent rock) or a mixture of Sirdrak sand and Kehoe variant (Kashiwagi, 1985). The latter soils are extremely well-drained, resulting in much drier conditions in the Wildlife Refuge (Lathrop & Gogan 1985). Following a period of rapid population growth during the first two decades after reintroduction, elk at Tomales Point reached a population size of approximately 450 individuals in 1998. Since then, the herd size has fluctuated between 400 and 600 animals, although in 2014 numbers declined to below 300, possibly due to drought (D. Press, unpublished data).
The diet of tule elk at this site consists primarily of herbaceous forbs and grasses, but they also consume shrub foliage during the winter months when there is less herbaceous vegetation available (Gogan & Barrett 1995;Johnson & Cushman, 2007).
California meadow voles, active both day and night, consume grasses, forbs, and soft seeds. They create runways through grassy areasand breed when grasses are fresh and sprouting (Cudworth & Koprowski, 2010). Deer mice, primarily nocturnal animals, are omnivorous, eating seeds, insects, fungi, and herbaceous plants (Jameson & Peeters, 2004). Harvest mice are also nocturnal and primarily granivorous. Typically found in grassy habitats, they often use the same runways made by voles (Webster & Jones, 1982), but their populations can negatively respond to high population densities of voles (Heske, Ostfeld, & Lidicker, 1984). The relative abundances of these rodents across the three habitats were not known prior to this study.

| Elk exclosure experiment
This study centers around an elk exclosure experiment on Tomales Within each of the three habitat types, there are four pairs of plots, with plots within pairs randomly assigned fencing to exclude elk or left unfenced to serve as controls. The control and exclosure plots within a pair are adjacent to one another, separated by a 3-m wide buffer to reduce edge effects. The fencing that surrounds each exclosure plot is 2.5-m tall and effectively excludes elk (Johnson & Cushman, 2007). Smaller herbivores at the site, such as hares (Lepus californicus) and the small-mammal species mentioned previously, are able to pass easily through the fence. Black-tailed deer (Odocoileus

| Small-mammal populations
To assess the influence of elk on small-mammal populations, we placed 25 Sherman live traps (8 × 9 × 23 cm, H. B. Sherman Traps, Inc., Tallahassee, FL) in a 20 × 20 m grid centered within each of the 24 plots of the experiment (traps were not placed along the edges of the plots to avoid edge effects). We trapped each exclosure at the same time as its neighboring control plot. During three-night trapping sessions, we sampled 4-6 plots per session until all 24 plots had been sampled (6-7 weeks; 1,800 trap-nights). The first round of trapping occurred between 26 October and 14 December 2013, whereas the second round took place between 13 June and 9 August 2014. By trapping in two different seasons over a year, we were able to test whether the effect of elk on rodents varied over two time periods that differed in rainfall, length of day, food availability, life cycle stages of the elk and rodent population, and other factors.
In each plot, we pre-baited (without setting) all traps for one night to acclimate the animals to the traps. We then trapped in each plot pair for three consecutive nights. For each captured animal, we recorded its trap location, species, and weight before releasing it at the site of capture. Peromyscus and Reithrodontomys individuals were fitted with a numbered Monel ear tag (National Band and Tag Company, Newport, KY, USA). Ear tagging allowed us to identify individuals that were caught on multiple nights, so their behavior would not influence our estimates of relative abundance. Due to low capture rates early in the study, we did not ear tag or record weights of voles. Small mammals in this study were treated according to guidelines from the American Society of Mammalogists (Sikes & Gannon, 2011), and the methods were approved by the National Park Service (permit #PORE-2013-SCI-0028) and Sonoma State University IACUC.

| Vegetation structure
To determine the effect of elk on vegetation structure, we quantified the height and biomass of accumulated dead herbaceous vegetation, which often formed a mat-like layer of "thatch" in the plots.
To measure thatch height, we divided each plot into four quarters and, between February and April 2013, measured thatch height in two shrub-free patches closest to the center of each quarter. In September 2014, we quantified the amount of thatch biomass on the surface of each plot by collecting, drying, and weighing four 0.25 m 2 samples per plot. The decision to take only four biomass samples per plot was drive by our desire to minimize destructive sampling and the observation that the effect of elk on biomass was consistent and great enough that additional samples were not necessary. We collected the samples from shrub-free patches nearest to each of the four corners of a 20 × 20 m area of each plot used for small-mammal sampling. We dried the samples at 60°C for 48 hr and weighed them immediately upon removal from the drying oven.

| Seed predation rates
To test the hypothesis that large mammals indirectly affect rates of seed predation by small mammals, we quantified seed-removal rates in all plots of the experiment during November of 2014. As a bioassay, we used the seeds of bush lupine (Lupinus arboreous), a shrub that dominates the Tomales Point study and whose seeds are regularly consumed by small mammals (Maron & Simms, 2001;Pitts & Barbour, 1979). This shrub drops most of its seeds in late summer when pods dry out and burst, and continues to drop seeds throughout the winter as the remaining seeds in the pods are dislodged by wind, rain, and other disturbances. We placed two seed depots 5 m from either side of the center of each plot in the experiment. Each depot consisted of a round plastic food storage container (12 × 20 cm) with a 2.5 cm diameter hole cut into the side to allow small mammals to enter. The hole was 2.5 cm above the bottom of the container to ensure that seeds would not easily roll out of the depot. Each container had a lid to protect the seeds from wind and disturbance by larger animals. We placed 25 lupine seeds in each container and assumed that all missing seeds had been consumed by small mammals. This assumption was supported by frequent observations of rodent feces in depots with missing seeds. After 18 days in the field, we checked the depots to quantify seed removal. We pooled the number of seeds removed from the two depots in each plot, to avoid pseudo-replication. We occasionally found insects and isopods in the depots, but we saw no evidence that they were able to remove or consume the seeds. All seeds used in the seed predation experiment were collected within 20 km of the study site at Point Reyes National Seashore.

| Statistical analyses
We analyzed data on small-mammal abundances, mean weights of captured deer and harvest mice in summer 2014, seed-removal rates, aboveground thatch biomass, and thatch height using linear mixed models, with elk (present, excluded), habitat type (Baccharisdominated, Lupinus-dominated, open grasslands), and their interaction as fixed effects and plot pair (1-12) nested within habitat type as a random effect. For small-mammal abundances, we also nested trap (1-25) within plot pair as a random effect, included moon phase during each trap session as a covariate and tested for an interaction between elk and moon phase. We recorded moon phase as the percent illumination according to the Astronomical Observations Department, U.S. Naval Observatory, Washington, D.C. For meadow vole abundances, we included season (fall, summer) as a fixed effect to allow for comparison across the two sampling sessions, to capture variation due to any temporal factors such as rainfall or different points in the reproductive cycle of the voles. We measured seed predation rates as the number of bush lupine seeds removed from the depots in each plot over 18 days. In these models, we initially included the following interaction terms: elk × habitat type, elk × season, elk × moon phase, elk × moon phase × habitat type, and elk × habitat type × season. If any fixed effects other than elk, habitat type, season, or the elk × habitat type interaction yielded p-values exceeding 0.15, we removed them from the model, beginning with the higher-order interactions (Crawley, 2014). We used the Kenward-Roger method (Kenward & Roger 1997) to estimate error degrees of freedom, which is widely recognized as the most rigorous approach when using linear mixed models (Kenward & Roger 1997;Schaalje, McBride & Fellingham 2002;Bolker et al. 2009). To ensure that assumptions for linear mixed models were met, we visually assessed all model residuals for approximate normality and checked for homoscedasticity of residual plots. If habitat type or any interaction terms were significant in our models, we followed up with Tukey multiple comparison tests to assess differences among means.
When evaluating small mammal abundance, we used the minimum numbers of mice (as determined by ear tagging over the three nights) and meadow voles (as determined by maximum number captured in one night) known to be active in each plot during each 3-day trapping session (Slade & Blair, 2000). For 2013, we combined capture data for deer mice and harvest mice, as we did not always differentiate between these species. In 2014, we recorded harvest mice and deer mice separately and tested to determine if responses to elk differed across mouse species. Only one animal tagged in fall 2013 was recaptured in summer 2014. Since mouse species were not differentiated in 2013, we did not compare their abundances across seasons.

| Small-mammal abundance
This study involved 843 small-mammal captures over a total of 3,600 trap-nights. In the fall of 2013, we captured and tagged 266 mice (including deer mice and harvest mice). In summer 2014, we tagged 143 mice, of which 109 were deer mice (76%) and 34 were harvest mice (24% ; Table 1). We captured meadow voles 80 times in 2013 and 136 times in 2014, with a minimum of 37 and 62 unique individuals in each year, respectively. We rarely trapped shrews (Sorex spp., 10 occasions) or the Point Reyes jumping mouse (Zapus trinotatus orarius, one occasion) and thus did not include these species in our analyses.

| D ISCUSS I ON
Studies by Heske and Campbell (1991), Keesing (1998), andPedersen et al. (2014) also failed to find a link between rodent abundance and vegetative cover, possibly because factors other than refuge from predation were more important drivers of small-mammal abundance in those systems. While elk reduced meadow vole abundance in open grasslands, they increased deer mice numbers in this habitat. A similar outcome was documented by Bueno et al. (2011), who found that vole numbers sharply declined in areas grazed by cattle, while those for deer mice increased. We considered the possibility that interactions between voles and mice might be affecting their abundances, but found no evidence of this in our system. Meadow voles (Cudworth & Koprowski, 2010) and harvest mice (Webster & Jones, 1982) are especially associated with grassy habitats, but deer mice F I G U R E 2 Mean numbers of individual (a) harvest mice (REME, Reithrodontomys megalotis) and (b)  occur in a wider range of habitats (Jameson & Peeters, 2004) and may be less dependent on the protection from predators offered by herbaceous plants. Food availability can be a more important driver of rodent numbers than the amount of protective vegetation (Keesing & Young, 2014), and elk may sometimes facilitate growth of food plants preferred by deer mice, despite reducing overall biomass of herbaceous vegetation (Arsenault & Owen-Smith, 2002).
Vole populations can rapidly increase when there is an abundance of fresh vegetation (Cudworth & Koprowski, 2010), and this may explain why their numbers were greater in the summer of 2014 than the fall of 2013 (Figure 3). Winter rains prompt new plant growth in this grassland, and our fall sample was taken just prior to the rainy season, when conditions were exceptionally dry. Both sampling periods occurred during a persistent drought in California, which probably increased competition among rodents for limited food resources.
We also found that voles were only seasonally affected by elk in  showed less preference between patches. While this effect occurred during both wet and dry seasons, it was stronger during the dry season. The variation in these results illustrates the importance of considering the context of consumer interactions and points to potential limitations in the predictive ability of unstratified experiments. Here, we tested for an elk × habitat interaction and showed that habitat characteristics can significantly alter the effect of elk on small mammal abundance.
Small mammals can alter their host communities through seed predation (Dangremond et al., 2010;Maron & Simms, 2001), raising the possibility that large mammals might indirectly affect plant communities through their effects on the abundance of granivorous small mammals. However, we did not detect an effect of elk on seed predation rates, even though they commonly altered rodent abundances. Our results contrasted with those of Smit et al. (2001) who found that elk, deer, and wild sheep reduced seed predation rates, presumably because they also reduced the abundance of wood mice (Apodemus sylvaticus) and field voles (Microtus agrestis). Seed predation from our depots was often "all-or-nothing," with some depots emptied quickly and others apparently unvisited for weeks. Seedremoval rates may vary with vegetative cover (Dutra et al., 2011;Mattos, Orrock, & Watling, 2013) and among seed species (Orrock & Damschen, 2005). It is likely that local conditions surrounding each depot, such as shrub cover or the availability of alternate food sources (including Lupinus seeds from nearby plants), were more important determinants of seed-removal rates than rodent abundance over the entire plot and may explain the greater seed-removal rates observed in open grassland habitat.
Our 16-year-old exclosure experiment has shown that the reintroduction of a large mammalian herbivore has had substantial indirect effects on small-mammal populations. By stratifying the experiment across three habitat types and two seasons, we have shown that elk have strongly context-dependent effects on two of the three small-mammal species. Although numerous studies have evaluated the interactions between co-occurring consumers, fewer have examined their indirect interactions, and fewer still have examined the context dependency of these interactions across different habitat types, seasons, and mammal species. We hypothesize that habitat-specific variability in consumer interactions is common in nature and that failure to focus on it may limit our understanding of interactions among co-occurring consumer species.

ACK N OWLED G M ENTS
Dave Press of the Point Reyes National Seashore provided invaluable logistical support throughout this entire project. Special thanks