Food quality and conspicuousness shape improvements in olfactory discrimination by mice

Abstract

How animals locate nutritious but camouflaged prey items with increasing accuracy is not well understood. Olfactory foraging is common in vertebrates and the nutritional desirability of food should influence the salience of odour cues. We used signal detection analysis to test the effect of nutritional value relative to the conspicuousness of food patches on rates of foraging improvement of wild house mice Mus musculus searching for buried food (preferred peanuts or non-preferred barley). Olfactory cues were arranged to make food patches conspicuous or difficult to distinguish using a novel form of olfactory camouflage. Regardless of food type or abundance, mice searching for conspicuous food patches performed significantly better than mice searching for camouflaged patches. However, food type influenced how mice responded to different levels of conspicuousness. Mice searching for peanuts improved by similar rates regardless of whether food was easy or hard to find, but mice searching for barley showed significant differences, improving rapidly when food was conspicuous but declining in accuracy when food was camouflaged. Our results demonstrate a fundamental tenet of olfactory foraging that nutritional desirability influences rates of improvement in odour discrimination, enabling nutritious but camouflaged prey to be located with increasing efficiency.

1. Introduction

Finding nutritious food with increasing efficiency provides an advantage to foragers, and is an implicit assumption of foraging models based on the optimization of effort [1]. In variable and stochastic environments, animals that can exploit newly available and nutritious foods quickly will gain the maximum value out of a temporary resource. Animals can discriminate between foods differing in nutritional value when faced with a choice [2–5] and vary their nutritional intake depending on physiological needs [6,7]. However, whether animals learn to discriminate cues from nutritionally desirable foods faster than other foods is not clear.

Experimental evidence shows that animals, including gastropods and humans, can detect nutritional differences in similar food types from odour cues alone [8,9]. Indeed, it has been suggested that the olfactory system is used to both detect odour cues in the environment and also respond to metabolic signals associated with nutritional requirements (see review by [10]). In combination, these studies imply that animals use olfactory cues to detect and hone in on particular food types, and that animals would focus attention on locating their most desired foods.

Experience of a highly rewarding food is predicted to enhance a forager's memory of the food odour, enabling faster learning of the cue [11]. The benefit of quickly detecting and finding resources with desirable compounds presumably leads to ongoing selection for greater olfactory sensitivity to those food types [12,13]. For example, male and female meadow voles are more interested in the odours of individuals of the opposite sex that had been fed high protein diets [14], suggesting that animals able to find high-quality food may also gain a reproductive advantage. Olfactory foragers have been shown to be particularly responsive to odours from foods rich in lipids [15], but will also attempt to maximize their dietary protein intake over other nutrients when in protein-limited environments [16].

Exploiting olfactory information to find food within natural environments can be challenging. The relative conspicuousness of an odour, which is a consequence of its concentration and spatial context, will influence the time and effort required to find its source [17–19]. Where food is conspicuous and easily found, the cognitive demands of foraging are low. Regardless of food type, rapid increases in foraging efficiency are to be expected although it is predicted that more desirable food will be the most vulnerable. However, if finding food is challenging, the forager's perception of the value of the food will determine the degree of investment in the more attentionally demanding task, particularly if other food is available [20].

We hypothesized that foraging efficiency using olfactory cues should improve faster when searching for more nutritionally rewarding food than when searching for less rewarding food, particularly when food is difficult to find. As it is impossible to understand how other animals perceive different foods, we used a novel method to control for inherent differences in the conspicuousness of alternate food types that allowed for normal olfactory foraging behaviours within a relatively controlled but ecologically relevant environment. Using wild house mice Mus musculus in large outdoor enclosures, we compared the rate of improvement in finding food using odour when it was conspicuous to when it was hard to find (i.e. camouflaged) and then examined whether animals responded differently to two different food types (peanuts: high quality; barley: lower quality). We camouflaged each food type against a background of its own odour, predicting that a forager is more likely to overcome the effects of olfactory camouflage when the food is nutritionally rewarding than when the food is less rewarding. By comparing rates of improvement in prey discrimination for different foods under challenging conditions (i.e. when food is camouflaged) and easy conditions (i.e. food patches are highly conspicuous), we should see whether food value alone can motivate foraging improvements, and if so, how these improvements may arise.

Our experimental design enabled us to control for differences in the conspicuousness of the two food types as well as the nutritional information embedded in odour cues. Comparing the relative rate of improvement in foraging efficiency within each food type when it is conspicuous with when it is camouflaged overcomes issues associated with comparing different food types. While minor differences between the two food types may affect initial search success, they would not affect the rate of learning given our design.

We had two specific hypotheses for the different drivers of behaviour over the short and longer term. The first hypothesis dealt with foraging behaviour on the first night: that foraging efficiency would be primarily influenced by the olfactory conspicuousness of the prey patches. We predicted that when patches containing buried food were highly conspicuous and easy to find (by adding odour cues only to those dishes), mice should have high foraging success and investigate only those dishes containing buried food. Conversely, when all potential foraging patches had odour added, including ones without buried food (i.e. the food patches were ‘camouflaged’ within the grid), we predicted that mice would investigate a high proportion of dishes, leading to low overall foraging success and wasted energy [21]. Our second hypothesis dealt with learning over the two nights: that the rate at which foraging success improved between nights (i.e. learning) would be primarily influenced by food type, particularly when finding food is challenging (i.e. camouflaged). Mice in our study system are protein-limited [22], and have been shown to respond to higher levels of dietary protein with increased breeding rates [23]. We predicted that mice searching for higher protein peanuts should rapidly become more efficient foragers, overcoming the distracting effects of the camouflaging odour more quickly than mice searching for lower protein barley.

2. Material and methods

We recorded the foraging performance of mice over multiple nights searching for buried food within a matrix of empty foraging patches. We manipulated nutritional value by comparing mice searching for peanuts (per 100 g: 2385 kJ, 48 g fat, 25 g protein) with those searching for barley (per 100 g: 1474 kJ, 1.2 g fat, 9.9 g protein). We further manipulated patch quality by altering the number of food pieces within the patches. To manipulate the olfactory conspicuousness of the buried food, we had three treatments. We either added the odours of each food type to the 15 foraging patches that contained buried food (conspicuous), or all 100 of the foraging patches to camouflage the food patches (camouflage), or none of the foraging patches [18]. These three treatments provided a theoretical gradient of food patch detectability that went from a high-level when only the 15 dishes containing buried food had the same food odour added (conspicuous), to mid-level when the food was buried but had no additional odour added, to a low level when the buried food was ‘camouflaged’ within the 100 dishes filled with odour (figure 1). We included the treatment with no odour added to reveal inherent differences in the conspicuousness of peanuts and barley as perceived by mice.

Figure 1.

The three food patch conspicuousness treatments: (a) high (conspicuous)—odour only added to the 15 patches containing food () and plain sand in the remaining 85 patches (), (b) medium—food buried in 15 of the 100 patches (•) and plain sand in the remaining 85 patches (); and (c) low (camouflage)—odour added to all 100 patches () with food buried in 15.

A single mouse was released into one of nine outdoor predator- and mice-proof enclosures (15 × 15 m) that were constructed of two layers of galvanized steel embedded in the ground, 2 m apart and covered in wire netting (5 cm mesh, 2.5 m high) to protect from predators and seed-eating birds. The enclosures approximate home range sizes recorded for wild mice in the locality during the breeding season [24] and support natural vegetation similar to that in surrounding agricultural lands (as per [25]). Each enclosure was divided into a 10 × 10 grid of foraging patches using plastic Petri dishes (Sarstedt, 90 mm diameter, 10 mm depth) filled with sand placed at each grid point (100 dishes per enclosure).

Wild adult mice (more than 11 g) were trapped from agricultural properties surrounding the Mallee Research Station, Walpeup, Victoria, Australia (35°07′ S, 142°01′ E). Small Elliot traps (33 × 10 × 9 cm, Elliot Scientific Equipment, Upwey, Victoria, Australia) were baited with sunflower seeds to prevent mice associating peanut/barley odours (used in treatments) with the trapping experience. Prior to release in enclosures, animals were temporarily housed in standard mouse cages (48 × 26 × 15 cm) for between 1 day and three weeks, and fed an ad libitum diet of rodent pellets, vegetables and water. An even sex-ratio and spread of source locations was maintained within each treatment. To promote foraging motivation, individuals were separated and denied access to food for approximately four hours prior to release into the experimental enclosures.

(a). Experimental procedure

We randomly allocated each mouse to one of twelve distinct treatment combinations (total for the whole experiment: n = 90 mice) (electronic supplementary material (ESM), table S1) and ran between seven and eight replicates of each of the 12 treatments.

Three experimental factors were tested within a full factorial experimental design: (i) food type; peanut (Arachis hypogaea) or pearl barley (Hordeum vulgare) pieces of similar size and weight (mean weight ± s.e. of two pieces of peanut 0.057 ± 0.006 g; mean weight ± s.e. of two pieces of barley 0.063 ± 0.003 g); (ii) food patch value: either two or six pieces of food to negate effects of handling costs or visual cues and randomly allocated to 15 locations within the 10 × 10 grid; (iii) food patch conspicuousness: (a) low or ‘camouflaged’: where food-scented oil was added to all 100 foraging patches, creating an even olfactory background with as little contrast as possible between empty patches and patches with buried food pieces; (b) medium, where pieces of food were buried in 15 out of the 100 foraging patches but no additional odour cues were added; and (c) high or ‘conspicuous’, where food-scented oil was added only to the 15 foraging patches containing buried pieces of food to intensify their olfactory conspicuousness within the 100-patch grid (electronic supplementary material, table S1; figure 1).

We created food odour cues using either peanut- or barley-infused canola oil. Canola oil was regarded as a suitable carrier oil as it readily absorbed the olfactory properties of the nut and grain being used, had very little intrinsic odour and contains no protein [26,27]. To make the olfactory cue, 500 ml of either powdered peanuts or pearl barley were mixed with 500 ml of canola oil, placed over a saucepan of boiling water for 1 h and then sieved to remove solid particles; 500 ml of scented oil filled 115 Petri dishes after being mixed with 6 l of sand. This equates to approximately 4.3 ml of canola oil per Petri dish, containing 1.9 g fat and 335 kJ. We cannot categorically exclude the possibility that mice were attracted to the canola oil, but regard it as unlikely given the strong food odours and almost dry consistency of the sand when the infused oil was thoroughly mixed in. There was no evidence that mice ate the sand. We can also not exclude the possibility that any of the odours used had a unique effect on learning or discrimination, however, it is extremely unlikely given our results.

An individual mouse was released into a randomly allocated enclosure at dusk and remained there for two consecutive nights. We compared each mouse's ability to find the 15 patches with buried food among the 10 × 10 grid per night to calculate the average rate of change in food detection for each treatment over two nights. Previous trials had demonstrated that mice acclimatized to the enclosures and experimental set-up within the first night [28]. Every foraging grid was completely replaced each day while mice were in burrows and food patches allocated to new locations to prevent memory or previous cues being used to find the food. A visit to a foraging patch was recorded if there was any evidence of mouse disturbance on the sand of the foraging patch. This method removed the potential for mice to use tactile cues when searching, rather than scent cues, despite this being extremely unlikely. To reduce the possibility that seed-eating ants could interfere with the food items, the foraging grids were set out just prior to dusk and collected at first light when ants were not active. Care was taken during all stages of the experiment to ensure that there was no contamination between the odours of the two food types.

(b). Analysis

Finding food against a noisy background is a signal detection task where the signal is unclear or unreliable [29,30]. Thus, we used a signal detection framework to reveal differences in the foraging strategies of animals searching for food of differing nutritional value that varied in olfactory conspicuousness [31]. The effect of nutritional desirability on improvements in signal detection within an olfactory system has not been previously examined, despite the relevance to a broad range of foraging interactions.

Three different metrics were calculated to estimate the reward: effort trade-off of finding more food patches at the expense of foraging in a greater number of empty patches; first, a simple measure of foraging efficiency (number of prey dishes visited/total dishes visited) to draw out initial trends in performance for each mouse, followed by two metrics derived from a signal detection framework (foraging accuracy and sensitivity to specificity ratio) that allow for a more precise assessment of discrimination ability [32]. The rate of improvement was inferred from the change over two nights in an individual's performance, using a two-by-two ‘confusion matrix’ (figure 2) to classify the proportion of observed foraging decisions (visits to patches) against the proportion of dishes containing food (positive) or that were empty (negative). The signal detection metrics are comprised of the ratios of different foraging decisions. True positive (TP) decisions result from a correct decision to forage in a patch (Petri dish) that contains food, a false positive (FP) decision results from an incorrect decision to forage in a patch that does not contain food, a true negative (TN) decision results from the correct decision to ignore an empty patch and a false negative (FN) decision results from an incorrect decision to ignore a patch that contains food.

Figure 2.

A confusion matrix for the results of each night's foraging activity per mouse [32].

(c). Signal detection metrics

• (i) Foraging accuracy—we calculated the accuracy of each mouse on each night from the total number of correct decisions the mouse made, both to visit patches with food and to not visit empty patches, divided by the total number of dishes available ((TP + TN)/100; figure 2). A factorial analysis of variance (ANOVA) including all interactions was conducted with food type, patch value and food conspicuousness as categorical factors, and change between nights as the response variable. The data and residuals met the assumptions of parametric analysis (data: normality—Shapiro–Wilks W test, p = 0.619; homogeneity of variances—Levene's test, p > 0.05).

• (ii) Sensitivity: specificity—the discrimination ability of each mouse per night was characterized by the proportion of food patches visited (sensitivity = TP/(TP + FN)) divided by the proportion of empty patches visited (specificity = FP/(FP + TN)) [32]. This ratio is a measure of the ability of an individual to discriminate food patches within the foraging grid from empty patches. The difference between nights was the response variable in a generalized linear model—traditional linear model (GLM) (normal distribution and identity link function), as the data could not be transformed for homogeneous variances by treatment or food conspicuousness categories (Levene's test p ≤ 0.01) [33,34]. The GLM contained the three categorical factors (food type, patch value and food conspicuousness) in a full factorial arrangement including all interaction terms. Contrast tests were conducted on significant terms.

To look for effects of differing rates of food consumption, we also analysed the amount of food removed from foraging patches in each treatment (food harvest) using a full factorial ANOVA, including all interactions between food type, patch value, food conspicuousness and night as categorical factors. The number of pieces of food eaten per night was the response variable. The data met the assumptions of parametric analysis (residuals: normal (visual assessment); homogeneity of variances—Levene's test, p > 0.05 for prey type). All analyses were performed in JMP 7 statistical software [34].

3. Results

Our camouflaged food treatment was equally effective for both barley and peanut pieces. On the first night, mean foraging efficiency of mice within camouflaged grids was less than expected of a random search (mean foraging efficiency for camouflaged peanut treatments = 0.12 ± 0.02, mean foraging efficiency for camouflaged barley treatments = 0.11 ± 0.02), indicating there was no difference in the efficiency with which the food-infused oil camouflaged the two food types (one-way ANOVA, F_1,28 = 0.24, p = 0.63). Foraging efficiency of mice within camouflaged grids was on average 42% lower for barley and 55% lower for peanuts compared with when the respective food patches were conspicuous (figure 3; night 1 combined treatments mean + s.e.: camouflaged barley: 0.11 ± 0.02, camouflaged peanut: 0.12 ± 0.02, conspicuous barley: 0.19 ± 0.03, conspicuous peanut: 0.27 ± 0.04). Mice foraged most efficiently on the first night when food patches were conspicuous and contained highly nutritious peanut pieces, and least efficiently when food patches were camouflaged and contained barley pieces. On average, individual mice foraging for camouflaged peanuts foraged 83% (±33%) more efficiently on the second night than the first night, whereas mice searching for camouflaged barley were only able to improve by 48% (±22%) between nights.

Figure 3.

The mean nightly foraging efficiency of mice by treatment (±s.e.). Foraging efficiency was measured as the number of food patches visited/total number of patches visited per night. The dotted line at y = 0.15 represents the foraging efficiency of a random forager. Key to treatments: food patch conspicuousness—low (camouflage), medium, high (conspicuous); food type—P (peanut), B (barley); food patch value—six or two pieces of food. The asterisk indicates a significant change between Night 1 and Night 2 (t-tests; p > 0.05).

The greatest improvement in foraging performance between nights was when food patches were conspicuous. However, the type of food being searched for influenced how mice responded to the way in which odour heightened the conspicuousness of or camouflaged the prey. Mice searching for peanuts improved in their ability to find food patches between nights regardless of the odour treatment. By contrast, mice foraging for barley only improved when the food patches were olfactorily conspicuous: they significantly improved their accuracy and specificity between nights, whereas mice foraging for camouflaged barley showed no improvement between nights (figures 4 and 6). Surprisingly, the number of food items per food patch had no effect on rates of improvement for either food type but overall more peanut pieces were harvested than barley (table 1).

Figure 4.

The change in accuracy between nights by food patch conspicuousness and food type (±s.e.). The significant interaction between food patch conspicuousness and food type (p = 0.03) is a result of differences between the food types. When searching for barley, mice foraging for conspicuous food patches showed rapid increases in accuracy compared with mice searching for barley in medium or camouflage treatments. There was no difference in the change in accuracy between the peanut treatments. Letters indicate significant differences between treatments (Tukey's HSD, α = 0.05).

Figure 6.

Mean changes in specificity (true negatives/(false positives + true negatives)) between nights showing a significant difference between the level of food conspicuousness in the barley treatments (±s.e.). Letters indicate significant differences between treatments.

Table 1.

Results of analyses of signal detection metrics showing that improved foraging performance is attributed to food patch conspicuousness, but that there is also an interaction with food type. Italic values indicate significant results.

factor	d.f.	ANOVA—change in accuracy			GLM—change in sensitivity : specificity
		SS	F	p-value	χ2	p-value
food patch conspicuousness (Cs)	2	0.14	4.85	0.01	9.96	0.007
food type (food)	1	0.04	2.67	0.11	0.35	0.55
Cs × food	2	0.11	3.70	0.03	2.45	0.29
food patch value (value)	1	0.02	1.16	0.28	2.10	0.15
Cs × value	2	0.01	0.38	0.68	3.70	0.16
food × value	1	0.01	0.68	0.41	1.93	0.16
Cs × food × value	2	0.00	0.14	0.87	3.66	0.16
error	81a; 78	1.16

^aDegrees of freedom (d.f.) of error for change in accuracy.

(a). Foraging accuracy

Mice foraged with increasing accuracy when food patches were conspicuous, but only when mice were searching for barley (food conspicuousness × food type: Tukey's HSD, α = 0.05; table 1; figure 4). Individual mice searching for conspicuous food patches improved in accuracy by 11.2% (±3.3%) overnight compared with 9.3% (±4.7%) when searching for food patches that were camouflaged or a 1.3% (±3.7%) decrease when searching for food patches where no odour cue was added (table 1—food patch conspicuousness: Tukey's HSD, α = 0.05). However, these average changes were primarily a consequence of how mice searched for barley, rather than peanuts. Individual mice searching for conspicuous barley improved their accuracy by an average of 14.4% (±4.4%), whereas mice searching for barley in no odour and camouflaged treatments declined in accuracy by 11.5% (±4.9%) and 4.6% (±6.9%), respectively. Food patch value had no effect on changes to accuracy and figure 4 shows combined treatments.

(b). Sensitivity : specificity

Sensitivity : specificity of mouse foraging decisions changed between nights only in response to food conspicuousness (table 1 and figure 5) (Kruskal–Wallis test , p = 0.02). Mice in conspicuous odour treatments improved more than those searching in camouflage treatments or when no odour was added to food patches (individual Wilcoxon rank-sum tests with the significance level adjusted for multiple comparisons; Bonferroni adjustment to α = 0.017; conspicuous odour > camouflaged odour , p = 0.015; conspicuous odour > no odour , p = 0.014; no odour = camouflaged odour , p = 0.880). This result was primarily a consequence of decreases in mean specificity under the camouflaged and no odour treatments (figure 6). There was a significant difference in the change in mean specificity from night 1 to night 2 between the barley conspicuous odour and barley no odour treatments (Tukey's HSD, p < 0.05). Figures 5 and 6 show combined treatments.

Figure 5.

The change in sensitivity : specificity between nights (±s.e.). Conspicuous cues lead to greater between-night improvements than other food patch conspicuousness treatments. Letters indicate significant differences between treatments (Bonferroni adjusted α = 0.017, n = 90).

(c). Food harvest

The number of food pieces harvested (i.e. removed and assumed to have been consumed) from food patches varied by food type, night and patch value (ESM, table S2). On average, mice removed more peanut pieces than barley pieces but only increased significantly in the number of pieces harvested per night when searching for peanuts. Similar amounts of both food types were harvested on the first night (ESM, figure S1). Overall, there was a 57.1% increase in the number of food pieces harvested on the second night relative to the first (average pieces of prey removed—night 1: 16.1 ± 1.6, night 2: 25.3 ± 2.2; F-ratio = 29.5, d.f. = 1, p < 0.001) and on average 143.3% more pieces were removed when patches contained six prey pieces than when they contained two pieces of food (average pieces of prey removed: two piece patches: 12.0 ± 1.0, six piece patches: 29.2 ± 2.2; F-ratio = 20.9, d.f. = 1, p < 0.001). There was also a significant interaction between patch value and night, with significantly more prey taken on night 2 from six-piece food patch treatments, but not from two-piece food patch treatments. Surprisingly, the different odour treatments had no effect on the number of food pieces that mice harvested.

4. Discussion

We found that the nutritional value of food in combination with the ease or difficulty of finding it affects the olfactory foraging strategies of a wild mammalian forager. Mice were able to find peanuts, a highly nutritious food, with increasing efficiency and accuracy regardless of whether they were easy or hard to find, but they responded differently to barley, the less nutritious food. The conspicuousness of barley influenced the rate at which wild mice improved in the olfactory discrimination task. Mice demonstrated rapid and significant improvements between nights when barley was easy to find. However, when barley was camouflaged, mice failed to improve in their ability to find food, performing worse on the second night in some cases. Surprisingly, patch value had no effect on the learning rates for either food type.

As we hypothesized, olfactory conspicuousness was the driver of foraging efficiency over the first night. Mice initially foraged most efficiently on the first night when food was conspicuous, because discriminating food patches from empty patches was an easy task regardless of food type. Camouflaging the food patches, by spreading food odour throughout the grid, initially reduced foraging efficiency relative to the other odour treatments for both foods, demonstrating that the mice perceived peanuts and barley as equally camouflaged on the first night.

Our hypothesis that food type drove improvements in foraging efficiency over the longer term was also supported, but only when comparing the responses within each food type to different levels of search difficulty. While mice searching for conspicuous food patches showed the greatest improvements between nights irrespective of food type, mice searching for peanuts, the food item higher in fat and protein, improved their foraging performance on the second night to the same extent regardless of how hard it was to find the peanuts. By contrast, mice searching for the less nutritionally beneficial barley did not improve their foraging accuracy when the food was camouflaged but did improve significantly when the food was conspicuous. However, there was no significant difference in the overall rate of improvement of mice searching for peanuts compared with barley.

These results suggest that nutritional qualities can affect the response of a forager to food based on its conspicuousness. Within a night, mice learnt to exploit some olfactory property of the nutritionally desirable peanuts to overcome the olfactory ‘camouflage’ and perform similarly to mice searching for peanuts that were easy to find. However, they did not or could not do the same when searching for camouflaged barley and found it more difficult to discriminate food patches from empty patches. Mice searching for barley that was highly conspicuous rapidly increased in their ability to find food compared with mice searching for camouflaged barley. The apparent interaction between the rate of foraging improvement, conspicuousness of the food and nutritional value is indicative of a cognitive mechanism akin to selective attention being used [35].

Selective attention is used by foragers when searching for cryptic food in order to focus the brain's relatively limited attentional capacity on cues that have proven to be reliable [36]. It is particularly pertinent to understanding olfactory search behaviour within complex odour environments [37]. For example, salticid spiders used selective attention to improve detection of both food and mates when either were olfactorily cryptic [38]. Our results for mice go further to suggest that the relative nutritional value of food influences whether selective attention is used to overcome difficult discrimination tasks, affecting the olfactory salience of food cues. Salient odours are learnt more rapidly than less salient odours, either as a consequence of the physical properties of the odour or the perceptual characteristics of the forager or a combination of both [39]. Our results support the notion that the mice were able to use selective attention when searching for camouflaged peanuts, but not for barley.

The lack of improvement by mice searching for camouflaged barley relative to those searching for conspicuous barley was not a consequence of a lack of interest in the food item, as demonstrated by the amount of barley harvested. Mice searching for conspicuous barley patches demonstrated the greatest improvement in foraging efficiency between nights of all the treatments.

Two signal detection metrics, foraging accuracy and sensitivity : specificity, revealed how the mice changed their foraging strategies between nights. Foraging accuracy measured the proportion of correct decisions each mouse made, either to forage in or ignore a patch. Sensitivity indicates the number of correct decisions to forage as a function of all available food patches, and specificity indicates number of incorrect decisions to forage as a function of all the empty patches. Together, they provide a measure of how easily the mice were able to perceive food under the different treatments. Mice improved from fewer investigations of empty patches on the second night, i.e. an increase in TN decisions, rather than finding more patches containing food. As predicted, these improvements were most evident when food patches were conspicuous, as mice quickly learnt the obvious food : cue association. By contrast, when food patches were camouflaged or had no odour added, mice continued to sample empty patches on night 2, causing mean specificity (the rate of TN decisions) to remain constant or decrease. And mice searching for barley under these conditions (camouflaged or no odour added) sampled far more empty patches on the second night than the first, demonstrating that they were motivated to find food but could not resolve useful cues to improve their performance. This behavioural response may also be a consequence of state-dependent factors, indicating that animals foraging on lower value foods are willing to search more thoroughly to ensure they maximize their intake of resources and take more risks to locate food [40].

Surprisingly, the number of food pieces had no influence on the rate at which mice improved in finding food patches, indicating that food availability per se did not motivate foraging decisions. While this result seems contrary to optimal foraging models (e.g. [20]) and studies of rodent foraging behaviour that suggest larger seed caches are easier to find (e.g. [41]), it supports the conclusion that the intrinsic nutritional properties of the food types motivated the changes to foraging efficiency. Studies of foraging decisions of bumblebees have found similar responses, and attributed the preference for quality over quantity to perceptual biases towards attributes of food cues that indicate food quality [42]. Bateson et al. [43] found that hummingbirds maximized net energy gain, rather than quality over quantity, and suggest that animals are more sensitive to particular attributes of food cues (i.e. quality or quantity) when the variation around that attribute is consequential for energy maximization. Our results suggest that in nutrient-limited systems, foragers may have perceptual biases to exploit opportunities to rapidly harvest high-quality food.

Improvements in prey detection are generally related to the difficulty of the task. In visual systems, prey crypticity, background complexity and competing attentional requirements have been described as relevant to the rate at which prey detection improves [44]. Crypticity within olfactory systems is more complicated to assess than for visual systems, but behavioural assays using relevant predators or foragers are useful for determining the difficulty of a task within a relevant environmental context (e.g. [45]). Our results show that the nutritional desirability of food can motivate improvements in foraging performance under conditions that can be considered olfactorily cryptic. Mice searching for peanuts learnt to discriminate camouflaged food and disregard distracting background odours to perform as effectively as mice searching for highly conspicuous food.

Being able to accelerate the rate at which certain foods are found has direct ecological consequences. Rapid learning of rewarding olfactory cues likely allows foragers to exploit novel food sources (e.g. [46]), respond rapidly to pulses of food, such as heavy seedfall events [47] and seasonal nesting events [48]. Olfactory learning by alien predators may also be a factor in the rapid decline of small, vulnerable populations, for example, following a reintroduction [49].

In conclusion, our findings confirm a fundamental aspect of olfactory foraging behaviour: that nutritional desirability affects rates of learning and improvements in foraging efficiency when cues are hard to discriminate, and provide further insight into cognitive mechanisms such as selective attention likely to preclude the formation of search images. This result helps to explain the factors that influence how quickly olfactory search images develop or dissolve [50]. Olfactory foragers' ability to refine their detection of hard-to-find but desirable prey is integral to odour-mediated foraging interactions. Because nutritional value influences the olfactory salience of food cues, it is likely that physiological and cognitive processes of olfactory foragers are tuned to finding rewarding foods. Whether an individual's nutritional status or broader environmental conditions influence selective attention towards particular food types remains a topic for further investigation.

Data accessibility

The supporting data for this paper have been deposited within the Sydney eScholarship Repository (https://ses.library.usyd.edu.au).

Authors' contributions

C.J.P. and P.B.B. conceived and designed the experiments. C.J.P. performed the experiments and analysed the data. C.J.P. and P.B.B. wrote the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This study was funded by ARC Discovery grant DP0881455 awarded to P.B.B. and conducted in accordance with UNSW Animal Ethics Approval 05/97A. It is dedicated to the late D.C. Price, whose contribution was invaluable to all aspects of the study.