Interacting effects of predation risk and signal patchiness on activity and communication in house mice


Correspondence author. E-mail:


1. Social signalling can be risky when signals are open to exploitation by eavesdropping predators. Unlike other signal modalities, olfactory signals cannot be ‘switched off’ in the presence of an eavesdropping predator, leaving receivers of scent signals at an increased risk of predation long after the signaller has moved on. Yet individuals of some olfactorily communicating species appear unwilling to forego the receiving of signals under an increased risk of predation. Foraging theory predicts that predation risk can operate at multiple spatial scales, however, such that prey behaviour should be sensitive to the broader olfactory environment beyond the risks of a single point source of odour.

2. Here, we use the house mouse Mus domesticus to test whether the spatial distribution and overall level of receiving activity varies with the spatial distribution of conspecific scent signals and the risks posed by an eavesdropping predator, the cat Felis catus. We assessed the mice’s responses to these risks using overall visitation, activity and scent marking rates at conspecific scented locations (in clumped, random or regular distributions) and the surrounding matrix (non-scented) locations with and without a predator cue (cat urine). We then used univariate and bivariate spatial point pattern analyses to assess behavioural responses (activity) to both treatments across a range of spatial scales.

3. Visitation, activity and scent marking rates were not affected by the predator cue or the spatial distribution of scents. But these non-significant results masked a fine scale anti-predatory response. Mouse activity was significantly more clustered at small scales when in the presence of the predator cue; this response held across all spatial distribution treatments. Mice were also sensitive to the predation risks of clumped scents, and dispersed their activity at intermediate scales significantly more when exposed to the predator cue, than in the control scent treatment.

4. These results suggest that olfactorily communicating species use scale-sensitive anti-predatory behavioural changes to compensate for their increased risks of predation when receiving scent signals. We highlight the importance of examining a variety of scales when investigating predator–prey interactions, and discuss the implications of these findings for behaviourally responsive predators and prey.


The signals used by animals confer benefits such as mate attraction or competitor intimidation, but they can also betray a signaller’s location to eavesdropping predators or parasitoids. In the arms race between predator and prey (Dawkins & Krebs 1979), prey species have adopted a wide variety of strategies to reduce their risk of predation when communicating. For species using short-lived signals, such as visual and auditory displays, risk reduction strategies may involve temporal or spatial changes in the location of signalling (Zuk & Kolluru 1998). But this strategy is not available to most territorial mammals, whose olfactory signals (urine, faeces or other secretions deposited directly onto the substrate; hereafter referred to as scent signals) are designed to persist in the environment (Brown & MacDonald 1985) and cannot be ‘switched off’ when a predator is detected. Moreover, because of the small spatial range over which most scent signals can be detected and the high economic costs of marking an entire territory, scent marks are often concentrated around valued resources and in areas where the likelihood of competitor intrusion is high (Gosling 1981). But predators rapidly detect concentrations of scent signals (Sundell et al. 2003; Ylönen et al. 2003), and their hunting success is greatest where scents are concentrated (Banks, Norrdahl & Korpimäki 2000, 2002). Thus, prey species should perceive sites of spatially predictable signals as areas of high predation risk.

Although olfactory communicators cannot reduce signal longevity, they can potentially reduce their risk of predation via other methods. First, prey can avoid scent patches. This precludes species from receiving the many signals which require direct contact (Brown & MacDonald 1985) however, and will subsequently affect territoriality, mating success and aggression (Hurst 1993). Alternatively, large movements will reduce association with a signal, although this approach has high social and energetic costs and it may also increase encounter rates with mobile predators (Anholt & Werner 1995). And while many individuals may opt to avoid predation by temporarily reducing activity (e.g. Jędrzejewski, Rychlik & Jędrzejewska 1993; Downes 2002), prolonged inactivity also increases predation rates due to the subsequent accumulations of odours (Banks et al. 2000, 2002). With predation risks associated with both high and low mobility, Banks et al. (2000) suggested that prey may use spatial behavioural shifts on an intermediate scale to reduce their predation risk.

However, the issue of scale is poorly understood for predator–prey interactions (Lima 2002). Field studies examining the risk of predation on rodents, for example, often look for behavioural changes at the scale of a home range (e.g. Parsons & Bondrup-Nielsen 1996; Wolff & Davis-Born 1997), or avoidance of specific locations treated with predator odours (e.g. Dickman 1992; Banks, Hughes & Rose 2003). But whereas some studies reveal that behavioural shifts over short distances can significantly reduce an individual’s chances of predation (Brown et al. 1988; Dickman 1992; Korpimäki, Koivunen & Hakkarainen 1996), others report no such behavioural shifts under increased predation risk (e.g. Parsons & Bondrup-Nielsen 1996; Wolff & Davis-Born 1997; Jonsson, Koskela & Mappes 2000). Consequently, it remains difficult to predict at what scale prey behavioural shifts will occur or how best to measure them, although we suggest that doing so at a single inappropriate scale will potentially mask adaptive responses.

House mice (Mus domesticus) are an ideal potential prey species through which to examine scaled responses to predation risk. Mice rely upon scent marks in most aspects of their social lives, including territorial defence, mate choice and individual recognition (Brown 1985). Volatile compounds attract mice to scent marks, but the individual-specific signal component is communicated via non-volatile compounds which require direct contact to be received (Hurst & Beynon 2004). Dominant, territory-holding male mice are the primary scent markers, and their ability to rapidly detect and countermark an intruder’s scent marks is a reliable indicator of their competitive ability to females, subordinates and other territorial males (Rich & Hurst 1998); foreign marks are therefore visited and counter-marked rapidly and repeatedly (Hurst 1989). Scents are also inspected (received) by females and non-dominant males, although their marking rates are lower than those of dominant males (Hurst 1990a,b).

The unique spatial and temporal properties of mouse scent marks that make them susceptible to eavesdropping predators also facilitate the conduct of signalling experiments. Scent marks are easily collected from specific individuals, stored until needed, and then distributed as desired. Previous experiments using these methods have revealed that mice are unwilling to reduce their visitation to individual scents under an increased risk of predation (but see Roberts et al. 2001; Wolff 2004; Pastro & Banks 2006; Hughes, Kelley & Banks 2009). However, it is possible that antipredatory behaviours occurred at a scale not detected by these past experiments. Because the concentration of prey odours affect a predator’s ability to track prey (e.g. Vergassola, Villermaux & Shraiman 2007), spatial heterogeneity in the distribution of scent marks should therefore modulate patterns of prey receiving. However, this remains to be empirically tested.

In this paper, we have three aims. First, to test whether patches of scent are perceived as areas of heightened predation risk; secondly, to examine whether mice alter their behaviour to reduce this risk of predation; and thirdly, to determine the spatial scale at which this response occurs. We manipulated the spatial distribution of scents (clumped, random or regular) in the presence or absence of a predator cue (cat urine added and no cat urine added respectively) to increase perceived predation risk. We hypothesized that clumped scents would be perceived by prey as especially risky, and individuals should generally avoid these clumps in the cat urine treatments. Thus in cat urine treatments, we predicted a reduction in the overall rates of visitation and activity to clumped scents compared to those distributed randomly or regularly. Furthermore, we predicted that activity would shift away from clumped scents in the cat urine treatment, but not away from random and regularly distributed scents.

Materials and methods

Study design

We conducted the study in seven outdoor enclosures at the Mallee Research Station, Walpeup (35°07′S, 142°01′E), Australia, in October and November 2007. Enclosures (internal dimensions: 15 × 15 m) were constructed of two layers of galvanized steel (152·5 × 0·1 cm), 2 m apart and buried to a depth of 50 cm (internal fences) or 80 cm (external fences), and were covered with wire netting (5 cm mesh, 2·5 m high) to provide protection from terrestrial and avian predators. We mowed a 50 cm buffer around the perimeter of each enclosure to encourage mice to remain within the dense vegetation (grasses, primarily Avena spp.).

We trapped adult male mice (>12 g) around buildings on geographically separated properties (>500 m between properties) and housed them in familiar pairs of individuals from the same trapping location. We held mice for up to 2 weeks in standard mouse cages (48 × 26 × 15 cm), and fed them an ad libitum diet of rodent pellets, sunflower seeds and water. We collected scent marks from captive male mice by placing individuals in clean cages (48 × 26 × 15 cm) lined with clean tiles (5 cm × 5 cm) for several hours, as mice typically scent mark exposed surfaces within their territories and new environments (Hurst 1987). Mice were then returned to their home cage and no adverse behaviour between pairs was noted. Scent marked tiles were individually labelled and immediately frozen in separate, air-tight containers for later use (Drickamer 1982). Mice were held singly for 48 h prior to being used in trials. During this time, we supplemented their diet with sodium fluorescein (0·5%) dyed sunflower seeds to dye the mice’s urine and faeces bright yellow and thereby increase the visibility of scent marking behaviour.

We established a 10 × 10 grid of equidistant locations (1·2 m apart) within each enclosure; each location consisted of a covered sand tray (14·5 × 9·5 × 7·5 cm) which allowed access from either end. We placed clean tiles in the centre of each sand tray at 85 grid locations (matrix locations) and scented tiles collected from 15 different captive animals at the remaining 15 locations (scented locations). We used scents from 15 different individuals to prevent scent matching (Gosling 1990) from reducing the receiving value of newly encountered scents. All scent donors were unknown to the mouse used within a trial. Previous experiments using this grid system showed that mice concentrated their activity at a scale of 2 × 2 grid locations (A. Carthey, unpublished data). Therefore, we distributed the 15 scented tiles in either a clumped (five patches of three scents), random or regular distribution at this scale. We increased the perceived risk of predation from an olfactorily hunting predator, the cat, in the predator cue treatment by distributing nine sticks (1·8 × 15 × 0·2 cm) soaked in cat urine evenly throughout an enclosure. Cat urine was collected from a large number of healthy individuals (mixed sexes and breeds) and frozen immediately. Several individual’s (both sexes) urine was thawed and pooled just prior to its application to the sticks. Control treatments had nine clean sticks placed in the same distribution. To ensure scent freshness, both scent marked and matrix tiles and the control and cat urine treatments were distributed at dusk. Immediately following the distribution of all treatments, we released a single, adult male mouse into an enclosure. We measured mice’s behaviour on the first night in the enclosure as previous experiments had shown that mice rapidly habituated to the experimental arena (J. Bytheway, unpublished data). Mice were randomly allocated to each treatment (n = 10) and were only used once. Treatment order was randomized and treatments were applied to randomly selected enclosures.

Scented and matrix locations were checked the next morning at first light (07:00–08:00). We estimated the mice’s responses to both the spatial distribution of conspecific scents and the predator cue treatment using patterns of visitation (presence/absence of mouse footprints in sand), activity, and scent marking (conspicuous yellow dyed urine and faeces) at scented and matrix locations. Activity was estimated from the proportion of sand surface disturbed, measured using a 10 × 10 grid overlayed on the sand surface). Activity was used as an index of the intensity of activity and hence time spent at a location. We removed the mice from enclosures the following night.

Statistical analyses

Rates of receiving and signalling

We analysed our measures of visitation, activity and scent marking separately using mixed model anovas. We used predator cue (cat urine and control), scent distribution (clumped, random and regular) and location (scented or matrix) as fixed factors, and individual identity nested within spatial distribution as a random factor to account for the repeated measure of location for each individual. All anovas were run in JMP7 (SAS Institute, Cary, NC, USA), and homogeneous variances (Levene’s test), normally distributed residuals (Shapiro Wilks test) and sphericity (Mauchly Criterion) were confirmed for all data.

Spatial patterns of activity

We analysed the mice’s spatial response to treatments using point pattern analyses, Ripley’s K-function (Ripley 1976, 1981), to reveal two types of responses. First, we used univariate analyses to describe how mice distributed their activity throughout the enclosure and the scales at which these patterns occurred. Secondly, we employed bivariate analyses to test whether there was any spatial relationship between the scented tiles we distributed and mouse activity (patterns 1 and 2 respectively, see below), and the scales at which these relationships occurred.

Ripley’s K is a cumulative frequency distribution of observations at a given point-to-point distance, and the function K(r) is the expected number of points in a circle of a given radius r from an arbitrarily chosen point, divided by the intensity of the pattern. Ripley’s estimator of the K-function is:


where r is the radius of a circle centred on a point in the pattern, n is the number of points in region A with an area |A|, dij is the distance between the ith and jth points, ωij is a weight function that corrects for edge effects and is the proportion of the area of a circle centred at the ith point with radius dij that lies within the study region, and Id is an indicator function which is 1 if the distance dij between points i and j is ≤r, otherwise Id = 0. This formula can be applied to both univariate and bivariate point patterns; however, because the K-function is difficult to interpret visually, a square root transformation of K(r) called the L-function (Besag 1977) is used instead:


For a univariate point pattern, L11(r) > 0 indicates that points are aggregated at distances up to r, while L11(r) < 0 indicates they are dispersed. For our univariate analyses of activity, activity is aggregated when L11(r) > 0 and dispersed when L11(r) < 0. For a bivariate point pattern, L12(r) > 0 indicates that there are on average more points of pattern 2 (activity) within a given distance (r) of points of pattern 1 (scented locations) than would be expected under independence (i.e. clustering up to distance r). Conversely, L12(r) < 0 indicates repulsion of the two patterns up to distance r. Therefore, L12(r) > 0 indicates clumping of activity around scented locations up to distance r, while L12(r) < 0 indicates that there is less activity around scented locations up to distance r (Diggle 2003; Wiegand & Moloney 2004).

Weighted K-functions are traditionally used to produce a single test statistic for comparing spatial patterns across replicated treatments, where the weight is the number of points in an individual replicate divided by the total number of points across all replicates (Diggle 2003). However, we were interested in examining differences between treatments at each individual distance of r. Additionally, our individuals were subjected to strict treatments of an underlying process (Diggle 2003) and the number of points of each pattern was similar across individuals and treatments (Table 1). We therefore calculated an average L-function for each treatment at each distance by averaging the L-functions across individuals within a treatment. Five individuals who visited fewer than 10 locations were excluded from the analyses because of the measure’s sensitivity to small sample sizes (Wiegand & Moloney 2004); the final sample size did not vary between treatments (χ2 = 0·118, P = 0·94). We calculated the significance (at the 95% level) of any deviance of the observed patterns from a null model of complete spatial randomness (CSR) by comparing the observed distribution function to the confidence envelopes generated by 999 Monte Carlo simulations of the CSR null (Besag 1977; Diggle 2003). For completeness, we compared our average L-functions and 95% confidence envelopes for each treatment against the corresponding weighted functions; all comparisons were significantly correlated (< 0·001, R2 > 0·92 for all comparisons). The results presented are therefore our averaged functions. All spatial analyses were performed using the grid based estimator in Programita (Wiegand & Moloney 2004).

Table 1.   The results of mixed model anovas on the effect of the predator cue (cat urine or control) and the spatial distribution of conspecific scents (clumped, random or regular) on receiving (visitation and activity) and signalling (scent marking) across locations (matrix or scented)
 Predator cue 1, 530·2720·604
 Scent distribution2, 530·6620·520
 Predator cue × scent distribution2, 530·2670·767
 Location1, 53186·556<0·001
 Predator cue × location1, 530·1960·660
 Scent distribution × location2, 531·7560·183
 Predator cue × scent distribution × location2, 530·1210·886
 Predator cue 1, 530·1330·717
 Scent distribution2, 530·2440·785
 Predator cue × scent distribution2, 530·2240·800
 Location1, 5364·956<0·001
 Predator cue × location1, 534·0010·050
 Scent distribution × location2, 531·1030·339
 Predator cue × scent distribution × location2, 531·4230·250
Scent marking
 Predator cue1, 530·0010·980
 Scent distribution2, 530·4800·622
 Predator cue × scent distribution2, 534·0080·024
 Location1, 5313·332<0·001
 Predator cue × location1, 530·4810·491
 Scent distribution × location2, 531·3200·276
 Predator cue × scent distribution ×    location2, 530·1040·901

We tested for treatment effects of predator cue and the scent distribution on L-functions at each distance using two factor anova with the factors predator cue, scent distribution and their interaction. Univariate and bivariate patterns were analysed separately.


Rates of receiving and signalling

Mice were strongly attracted to conspecific scents, and both visitation and activity were significantly higher at scented than matrix locations (Table 1, Fig. 1a,b). In contrast to our first prediction, that overall rates of visitation and activity would be lower at clumped than randomly or regularly distributed scents in the presence of cat urine, neither of these measures varied with the predator cue treatment or the spatial distribution of scents, nor was there an interaction between these factors (Table 1). Pairwise tests to examine a marginally significant interaction between predator cue treatment and location for activity (Table 1) simply highlighted the significantly higher level of activity at scented than control locations in both predator cue treatments (Tukey’s HSD α < 0·05).

Figure 1.

 The effect of the predator cue (cat urine or control) and the spatial distribution of conspecific scents (clumped, random or regular) treatments on the mean (a) proportion of locations visited, (b) activity at visited locations and (c) proportion of visited locations that were scent marked.

Consistent with previous field studies (Wolff 2004; Hughes et al. 2009), overall scent marking rates were higher at scented than matrix locations and scent marking was unaffected by the addition of cat urine (Table 1; Fig. 1c). In cat urine treatments, there was a tendency for mice to deposit fewer scent marks at scented locations when these locations were clumped or regularly distributed (Fig. 1c). In contrast, rates of scent marking at both scented and matrix locations were higher in cat urine treatments compared to control treatments when scent locations were randomly distributed (Fig. 1c). Consequently, there was a significant interaction between the predator cue and the spatial distribution of scents treatments (Table 1). Within the cat urine treatment, pairwise tests revealed higher rates of mouse scent marking when scented tiles were randomly distributed than when scents were clumped or regularly distributed. Scent marking rates were also higher in the cat urine treatment than the control treatment when scents were randomly distributed; these differences, however, were not significant after adjusting for multiple comparisons (Tukey’s HSD α > 0·05).

Spatial patterns of activity

Whereas there were no effects of predator cue or scent distribution on overall patterns of activity, the point pattern analyses revealed significant spatial shifts in behaviour with both treatments. All univariate L-functions were >0 at short distances (r = 1 and 2; Fig. 2), indicating that mice clustered their activity at small scales. But mice in the cat urine treatment significantly increased this clustering of activity at distances up to r = 4; this response was consistent across all three spatial distributions of scented locations (Table 2; Fig. 2). Further, the degree of clustering declined rapidly with distance in the control treatment: activity was distributed randomly at distances of r = 3, 4 and 5 when scents were in random, regular and clumped distributions respectively. In contrast, the spatial patterning of activity did not become random or regular until distances of r = 6 (clumped) and 7 (random and regular) in the cat urine treatment.

Figure 2.

 Univariate L-functions describing the spatial distribution of activity when scents were distributed in (a) clumped, (b) random or (c) regular spatial distribution in the cat urine (solid line) and control (dashed line) treatments. Error bars are standard errors. Grey shaded areas are the 95% confidence intervals calculated from the Monte Carlo simulations of the CSR null of both cat urine and control treatments, as these were highly correlated (R2 > 0·985). Where L-functions are >0 there is significantly more activity is distances r than expected from independence, while there is significantly less activity at distances r where L-functions are <0*Indicate distances at which there was an overall significant predator cue treatment effect (P < 0·05); see Table 2 for details.

Table 2. anova tests of the effect of the predator cue (cat urine or control) and the spatial distribution of conspecific scents (clumped, random or regular) on L-functions describing the univariate distribution of activity and the bivariate relationship between scented locations and activity at distances (r) 1–10
FactorsActivityActivity around scented locations
  1. d.f. predation risk = 1, scent distribution = 2, predation risk × scent distribution = 2, error = 49.

r = 1
 Predator cue0·2804·1500·0470·0000·0340·855
 Scent distribution0·0450·3300·7200·30114·496<0·001
 Predator cue × scent distribution0·0170·1270·8810·0492·3540·106
r = 2
 Predator cue0·86210·0220·0030·0000·0290·866
 Scent distribution0·0740·4340·6500·0541·7410·186
 Predator cue × scent distribution0·0500·2910·7490·0551·7940·177
r = 3
 Predator cue1·0708·9730·0040·0010·0270·871
 Scent distribution0·0640·2660·7670·0811·4500·245
 Predator cue × scent distribution0·0720·3000·7420·0821·4690·240
r = 4
 Predator cue0·8396·6920·0130·0020·680·796
 Scent distribution0·0920·3670·6950·0180·2610·771
 Predator cue × scent distribution0·1470·5840·5610·0640·8990·414
r = 5
 Predator cue0·4403·5480·0660·0401·0760·305
 Scent distribution0·0180·0740·9290·0140·1870·830
 Predator cue × scent distribution0·0490·1990·8200·1982·6770·079
r = 6
 Predator cue0·3973·9390·0530·0531·3720·247
 Scent distribution0·0890·4420·6450·0170·2200·803
 Predator cue × scent distribution0·0820·4070·6680·2853·6860·032
r = 7
 Predator cue0·2823·9520·0520·0090·2710·605
 Scent distribution0·0390·2750·7610·0140·2420·786
 Predator cue × scent distribution0·0330·2340·7920·2113·5440·037
r = 8
 Predator cue0·0922·7660·1030·0020·1230·727
 Scent distribution0·0400·6060·5500·0371·0510·357
 Predator cue × scent distribution0·0100·1540·8580·0371·0390·362
r = 9
 Predator cue0·0382·4570·1230·0020·2520·618
 Scent distribution0·0591·8870·1620·0231·8860·163
 Predator cue × scent distribution0·0130·4170·6620·0211·7030·193
r = 10
 Predator cue0·0201·4680·2310·0020·6770·415
 Scent distribution0·0662·2900·1120·0192·3070·110
 Predator cue × scent distribution0·0180·6580·5220·0111·3880·259

Bivariate analyses of the spatial relationships between mouse activity and scented locations revealed activity patterns which varied with both the predator cue and the spatial distribution of scents (Fig. 3). L-functions followed a U-shaped trajectory with distance in all treatments. Activity was significantly clustered around scented locations when these were themselves clumped (r = 1, Table 2, Fig. 3a). Activity was also clustered in the cat urine treatment at r = 1 when scents were randomly distributed, but was dispersed away from scented locations at all other distances. In accordance with our predictions, the predator cue treatment differentially affected the relationship between activity and clumped scented locations (Table 2, Fig. 3a): there was significantly less activity at distances of r = 6 and 7 from scented locations in the cat urine treatment than in the control treatment. A pictorial example of the effects of the spatial distribution of scents and predation cue on activity patterns clearly illustrates these results (Fig. 4).

Figure 3.

 Bivariate L-functions describing the spatial distribution of activity around scented locations in (a) clumped, (b) random and (c) regular spatial distributions in the cat urine (solid line) and control (dashed line) treatments. Error bars are standard errors. Grey shaded areas are the 95% confidence intervals as detailed in Fig. 2. L-functions >0 indicate that there was on average more activity within a given distance r of scented locations than would be expected under independence (i.e. clustering), while L-functions <0 indicate repulsion between the scented locations and activity up to distances r. *Indicate distances at which there was a significant scent distribution treatment effect, and † a significant interaction between predator cue and scent distribution (P < 0·05); see Table 2 for details.

Figure 4.

 A visual representation of the effects of predator cues and scent distribution on the spatial patterning of mouse activity. Mouse activity is plotted within enclosures relative to scented locations (15 white dots; the 85 matrix locations have not been included for clarity) that were arranged in a clumped (a, b), random (c, d) or regular (e, f) distribution. Mice significantly increased the clustering of their activity (indicated by large patches of orange and red colours) when in the presence of a predator cue (cat urine treatment; figs a, c and e). Mouse activity was dispersed more evenly throughout the enclosure (indicated by more, smaller patches of orange and red) when the predator cue was absent (control treatment; figs b, d and f). Because visitation and activity did not differ significantly between treatments, all figures were created using the same number of visited scented and matrix locations, and the activity values at these locations were the same to create both figures. Thus, the only differences lie in the spatial patterning of activity. The distribution of activity conforms to the mean univariate and bivariate L-functions derived for each treatment. Contour lines were created using the contour graph function in jmp7 (SAS Institute).


The ability to respond appropriately to spatial and temporal heterogeneity in predation risk underlies theories of optimal foraging, habitat selection and the evolution of communication systems. Mice in our experiment modulated their activity according to heterogeneity in risk; not only in response to the presence of predator scents, but also in response to the risks associated with concentrations of conspecific signals. They invoked fine-scale behavioural changes which clustered their activity at small scales to maintain receiving of conspecific signals, while minimizing movement activity throughout the matrix which would likely increase predation rates (Norrdahl & Korpimäki 1998). Small shifts in the level and location of activity are common anti-predator mechanisms employed by a wide range of taxa (e.g. Dickman 1992; Hedrick & Dill 1993; Jędrzejewski et al. 1993; Skelly 1994; Rohr & Madison 2001), as such shifts can significantly increase survival (e.g. Dickman 1992; Banks et al. 2000; Rohr & Madison 2001; Downes 2002). Mice also appeared to perceive patches of clumped scents as areas of elevated predation risk, and dispersed their activity away from risky scent patches accordingly in the predator cue (cat urine) treatment but not in the control treatment. These spatial shifts in behaviour reveal a mechanism which reconciles earlier observations that some olfactorily communicating species were unwilling to forego communication (Wolff 2004; Pastro & Banks 2006; Hughes et al. 2009), in spite of the risks of predation when doing so.

The perceived social benefits of receiving a signal affects the level of risk olfactory communicators are willing to take (Hughes et al. 2009). Mice in this experiment were exposed to the scents of 15 unknown male donors, so the motivation to visit scents should have been high. This level of territorial intrusion probably overestimates that typically experienced in the wild, although densities within a territory do easily reach this number (Hurst 1987; Sutherland & Singleton 2003). Had all scents come from the same individual, we possibly would have found lower levels of visitation and activity at conspecific scents when in the presence of the predator cue (Hughes et al. 2009), and corresponding changes in the spatial distribution of activity. Therefore, it appears as although the maintenance of signal receiving in these species reflects a trade-off between the social costs of not doing so (e.g. Hurst 1993), and costs associated with the behavioural shifts required to counterbalance the increased risk of predation, such as reduced access to foraging resources. Importantly, all the field-based experiments that have failed to detect an avoidance of conspecific odours under an increased risk of predation have emphasized the social costs that avoidance would entail (Wolff 2004; Pastro & Banks 2006).

The importance of scale is frequently emphasized for prey anti-predator behavioural strategies (e.g. Lima & Dill 1990; Lima 1998), but rarely is it examined over more than a limited range. By examining the spatial patterning of activity, we were able to show scale dependent activity under perceived predation risk; isolating our analysis to only one scale would have missed this pattern. It is therefore possible that earlier field studies which did not demonstrate an effect of predator odour on prey behaviour (e.g. Wolff & Davis-Born 1997; Jonsson et al. 2000), may have missed important small-scale behavioural changes. Similar changes in spatial patterns, but not rates, of activity in response to increased costs have recently been observed in other fields. Dunn & Whittingham (2007), for example, showed that female tree swallows that had been experimentally handicapped, maintained high levels of extra-pair matings despite their handicap. To compensate for the additional costs in mate searching, they altered the spatial distribution of males with whom they mated, and the young of handicapped females were more likely to have fathers living close to female nests than the young of control females. Dunn and Whittingham similarly concluded that such spatial trade-offs in response to increased costs would have been missed by only considering the proportion of young sired by extra-pair matings.

Because concentrated patches of scents attract olfactorily hunting predators, we predicted that mice would perceive clumped scents as patches of increased predation risk, and disperse their activity away from clumped scents at small scales. However, mice in the presence of the predator cue did not respond as predicted. Instead, the overriding antipredator response in all spatial distribution treatments was to increase the clustering of activity. This reduced the number of activity patches across all treatments, such that activity was significantly more dispersed at larger scales (r = 6 and 7) in the clumped, predator cue treatment than in the control treatment. This suggests that mice perceived an increased risk of predation at the enclosure level, rather than at clusters of scents within the enclosure. There are a number of possible explanations for this pattern. First, it is possible that mice did not regard the 2 × 2 clumping of scent marks as sufficiently large a risk to warrant avoidance, or that visiting such patches does not expose individuals to any higher risk of predation. However, a previous study showed that olfactorily communicating voles perceive individual scent marks located many metres apart as sites of increased predation risk (Hughes, Korpimäki & Banks in press), and the more dense marks in this experiment should have been easily identified as risky. Furthermore, an increase in predation rates in areas where scents have been experimentally added (Koivula & Korpimäki 2001) is also indicative of the inherent risk of associating with scents.

Secondly, and more probably, it is possible that mice were responding to risk of predation at the enclosure scale because it is at this larger scale that their predators (in this case cats) search for prey. Although many predator species are likely to be attracted to concentrations of prey scents, exactly how they use this information to make foraging decisions is poorly known. Most models of predator hunting behaviour do not include exploitation of prey signals, but assume predators encounter prey randomly (e.g. Viswanathan et al. 1999; Higgins & Strauss 2004; Ruxton 2005). Yet recent foraging theory predicts that information on patch quality should play an important role in predator movement decisions (Stephens 2007). Predators may restrict their search to small areas around scent patches or, more probably, they might use scent patches as a general cue to prey, before increasing their search on a slightly larger scale (Tinbergen, Impekoven & Franck 1967). Under this scenario, prey will be most at risk at this larger scale because it constitutes the predator’s perception of a prey patch (Schmidt & Brown 1996). Moreover, the concentration of activity into fewer areas suggests that mice perceived movement to be more risky than association with a scent patch once a predator (or their cue) had been detected, possibly because many predators use their acute vision and hearing during later stages of the predator–prey encounter (Conover 2007).

Nevertheless, scent patches will still increase predation risks for prey if they concentrate scents to within the predator’s threshold of detection above background levels. For prey, our results provide a rationale to maintain multiple scent patches under an increased risk of predation if this would reduce the initial likelihood of predators detecting the prey cues and decrease their ability to discern a prey patch. Territory defence would not necessarily suffer from marking multiple sites because different spatial distributions of scents did not affect receiving rates, notwithstanding the effect of the size of the home range and the distribution of resources on marking costs. Furthermore, multiple scent patches should dilute the risks associated with any one patch and create an even distribution of prey cues. An even distribution of prey (and prey cues) at a large scale would also reduce the ability for non-randomly searching predators to track the system (Fauchald 1999; Nachman 2006), and it may disguise differences between the densities of adjacent patches.

The anti-predator strategies of prey that we found suggests that predators need to respond with a dual strategy to find prey; one involving a wide search tactic and another involving attraction to scent patches at the smallest scale, akin to area-restricted searching tactics (Tinbergen et al. 1967). Mice maintained high mobility in the absence of predator cues, and such behaviours should facilitate predator success on a larger scale (i.e. across several scent patches) simply due to random encounter. But risk averse mice in the presence of the predator cue reduced their mobility and receiving behaviour was more localized which would reduce the success of predators solely foraging at this larger scale. To counter this, predators should then focus searches around specific scent patches and rely upon prey visual and/or auditory cues to detect prey at close range (Österholm 1964; Fitzgerald & Turner 2000). Refinement of these predictions for predator and prey strategies requires further research into the scale at which predators track prey (Fauchald 1999), the predator’s ability to discern fresh, individual scent patches from old cues, and their relationship to prey density.


The authors wish to thank the Mallee Research Station for their assistance and access to mouse enclosures, and the Yagoona RSPCA and Paddington Cat Hospital for cat urine. The authors also thank K. Hughes, R. Hughes, A. Lothian and C. Price for their help in the field, and to S. Laffan for statistical advice. This research was conducted with the permission of the UNSW Animal Care and Ethics Committee.