- Email: [email protected]
A rule of thumb in mammalian herbivores?
ANIMAL BEHAVIOUR, 1998, 56, 337–345 Article No. 980786
A rule of thumb in mammalian herbivores? MAGNUS AUGNER, FREDERICK D. PROVENZA & JUAN J. VILLALBA
Department of Rangeland Resources, Utah State University (Received 19 May 1997; initial acceptance 24 June 1997; final acceptance 12 December 1997; MS. number: 5549R)
ABSTRACT In two experiments on appetitive learning we conditioned lambs, Ovis aries, to particular concentrations of a flavour by mixing the flavour with an energy-rich food that complemented their energy-poor diet. The lambs were subsequently offered energy-rich food with five different concentrations of the flavour (the concentration to which they were conditioned, two higher concentrations, and two lower concentrations). At these tests, the lambs consistently preferred the weaker flavours. This finding stands in contrast to earlier results on generalization gradients. In a third experiment, similarly designed to the other two, we tested for effects of a strong flavour on the behaviour of lambs when they were offered a novel nutritious food. Half of the lambs were offered unadulterated wheat, and the others strongly flavoured wheat. We found that the flavour in itself was initially aversive. We propose that the lambs’ avoidance of foods with strong flavours may be an expression of a rule of thumb of the type ‘given a choice, avoid food with strong flavours’. Such a rule could be part of a risk-averse foraging strategy displayed by mammalian herbivores, and which could be of particular importance when they encounter unfamiliar foods.
ences made by itself or by accompanying conspecifics. For instance, a plant with a familiar odour may be more readily accepted and sampled than one with an unfamiliar odour (e.g. Provenza 1996). However, when a plant and the sensory stimuli associated with it are truly novel the animal has no previous experience yet still has to make a decision. Mammalian herbivores generalize in a qualitative fashion over cues (sensory stimuli) provided by unpalatable foods (e.g. Launchbaugh & Provenza 1993). However, it is not known if they generalize quantitatively, that is, there are no studies of generalization gradients of food-associated cues in mammalian herbivores. A generalization gradient is defined as the intensity of the behavioural response of an animal as a function of the stimulus, when the stimulus varies (quantitatively) in onedimensional space, for example, in brightness or in flavour concentration (Spence 1937). The intensity of the response is generally strongest to stimuli that are close to the stimulus to which the animal is conditioned; as the stimuli diverge, the responses decrease in intensity. A conditioned generalization gradient could be seen as a learnt behaviour that gives rise to a kind of rule of thumb: ‘react the strongest to familiar stimuli’. Mammalian herbivores often show a high degree of neophobia when first encountering an unfamiliar food (e.g. Provenza et al. 1995; Provenza 1996). Furthermore, a familiar food that is normally readily eaten is sampled
The use of information provided by plants is fundamental to the whole foraging process of herbivores. The plant community provides herbivores with large amounts of (potential) information, in the form of olfactory, gustatory, visual and tactile stimuli. These stimuli can be perceived both before and during feeding. Taken together, there are probably no two plants whose ‘information signatures’ are identical. This large amount of information potentially constitutes a problem for generalist herbivores, and they must be selective as to what fraction of the potential information they evaluate. Consequently, the ability to learn to recognize foods and the ability to generalize over cues are potentially important traits. It has been proposed that familiarity is the major determinant of animals’ responses to foods and flavours (e.g. Kalat 1974; Rozin & Schulkin 1990). Another information-related problem for generalist herbivores is that they may encounter plants with unfamiliar information signatures in new environments as well as in familiar surroundings. When such a novel plant is encountered the herbivore can either (1) ignore it, (2) sample it to some extent, or (3) eat from it in an uninhibited fashion. In this decision, the herbivore may rely on earlier experiCorrespondence and present address: M. Augner, Department of Theoretical Ecology, Lund University, Ecology Building, S-223 62 Lund, Sweden (email: [email protected]). F. D. Provenza and J. J. Villalba are at the Department of Rangeland Resources, Utah State University, Logan UT 84322-5230, U.S.A. 0003–3472/98/080337+09 $30.00/0
1998 The Association for the Study of Animal Behaviour
337
1998 The Association for the Study of Animal Behaviour
338 ANIMAL BEHAVIOUR, 56, 2
carefully if its flavour is changed (Provenza et al. 1993, 1995). We propose that when a herbivore encounters a given food, the whole preingestive information signature of the food is important in its decision whether to feed. In this decision, familiarity is an important factor (e.g. Biquand & Biquand-Guyot 1992; Provenza et al. 1995; Provenza 1996), and the stronger the sensory input, the easier a herbivore can determine whether a food is familiar or unfamiliar. Also, neophobia could be regarded as a kind of rule of thumb: ‘avoid unfamiliar foods’. However, in contrast to learned generalization gradients, neophobia seems to be an innate behaviour in mammalian herbivores. We performed two experiments to ascertain how lambs, Ovis aries, generalize across flavours associated with a beneficial food. The lambs were conditioned to a distinctly flavoured, energy-rich food (ground wheat). After conditioning, we simultaneously offered the lambs the energy-rich food with five different concentrations of the original flavour. Our hypotheses were: (1) lambs that received no special pretreatment would prefer the food with the flavour concentration to which they were conditioned; (2) after a preload with another energy-rich food, the lambs would prefer foods with lower flavour concentrations relative to the concentration to which they were conditioned (and the total intake of the energyrich food would decrease); and (3) after being deprived of energy, the lambs would prefer foods with higher flavour concentrations (and the total intake of the energy-rich food would increase). Hypotheses 2 and 3 imply that the pretreatments caused peak shifts in the generalization gradients (e.g. Leimar & Tuomi, in press), that is, displacements of the maximum response of the average generalization gradient towards lower and higher flavour concentrations, respectively. We performed a third experiment to ascertain the effects of a strong flavour on the intake of a novel beneficial food by lambs. The lambs were provided with a nutritionally balanced food marginal in energy. For a short time each day, half the lambs were offered a distinctly flavoured, energy-rich food (ground wheat), while the other half was offered unadulterated ground wheat. Our hypothesis here (4) was that a strong novel flavour is more unfamiliar than a weak novel flavour. By increasing the degree of neophobia a strong flavour can in itself be aversive. METHODS
Experimental Animals and Environment We performed the experiments at the Green Canyon Ecology Center, Utah State University, Logan, Utah, U.S.A. The lambs were kept outdoors, under a protective roof in individual, adjacent pens measuring 2.43.6 m. The lambs always had access to water, salt blocks and alfalfa pellets ad libitum. Before the experiments the lambs were given an adjustment period of about 3 weeks, during which we gave them 200 g of rolled barley per day. In experiment 1 we used 20 lambs (two of which refused to eat the flavoured wheat); and in experiment 2 we used 24 lambs. These lambs were commercial crossbreed
white-faced lambs, of both sexes, and approximately 3 months old. In experiment 3, we used 20 lambs which constituted a very homogeneous group: they were commercial crossbreed white-faced lambs, all male, approximately 60 days old, and they had been reared together. These lambs had been weaned for about a week when we received them, and they had some experience of solid foods (alfalfa pellets and rolled barley).
Treatments We gave the animals alfalfa pellets ad libitum, and as test food we used pure ground wheat. Alfalfa pellets constitute a nutritionally balanced food, but their energy content is marginal for the rapid growth of lambs (NRC 1985). In contrast, wheat is rich in readily digestible energy (NRC 1985), thus complementing the nutritional merits of alfalfa pellets. In the test food (ground wheat) we mixed flavour substances: it was mixed fresh daily (see Generalization gradients and Novel food, below). The lambs were initially naive to the flavours.
Generalization gradients In these two experiments, the lambs were randomly divided into four groups. Each group was conditioned to a given flavour concentration. We used different conditioning treatments because no experiment of this kind had been done before and we wanted to see if the flavour concentration, as such, had an effect on the outcome of conditioning. In experiment 1, the lambs received ground wheat with either 0.5, 1.0, 1.5 or 2% flavour (dry weight) during conditioning. In experiment 2, these flavour concentrations were 0.25, 0.5, 0.75 and 1.0% flavour (dry weight). As flavour substances, we used ‘green apple feed nectar’ in experiment 1, and dried onion powder in experiment 2. The feed nectar is an artificial flavour developed to stimulate food intake by livestock by giving them seemingly more varied diets (‘Green apple feed nectar (4501)’, Agrimerica Inc., Northbrook, IL, U.S.A.). To humans, this substance has a distinct odour and a weak, sweet taste. We used commercially available dried onion (‘Onion powder white’ (63775); Pacific Foods, Kent, WA, U.S.A.). Onions, Allium sp., occur naturally on rangelands in many parts of western U.S.A., and are readily eaten by sheep (e.g. Cheeke 1991). To the human palate, the onion powder had no taste but a very strong, aromatic odour. Since we cannot be sure that the onion powder did not provide the lambs with any taste, we prefer to call it a flavour rather than an odour. The flavour substances and concentrations used were carefully chosen to be stimulatory neutral, in the sense that after repeated exposures they are inherently neither attractive nor aversive to lambs; lambs often show marked neophobic responses to unfamiliar flavours and/or foods (e.g. Provenza et al. 1993, 1995; Provenza 1996). Our choices were based on earlier extensive preference/aversion tests (single and multiple choice tests) of flavours and flavour concentrations (F. D. Provenza et al., unpublished data). Furthermore, in earlier
AUGNER ET AL.: A FORAGING RULE OF THUMB 339
experiments, neither feed nectar (Early & Provenza, in press) nor dried onion (Villalba & Provenza 1997a) had any negative or positive postingestive effects. In experiment 1, wheat was offered ad libitum for 30 min each morning during conditioning and trials. In experiment 2, we offered the lambs 100 g of wheat for 15 min each morning during conditioning, and during the trials wheat was offered ad libitum for 15 min. Before these experiments, the lambs had 1 week’s experience with eating ground wheat. In trial 1A we simultaneously offered the lambs wheat with five flavour concentrations: the one used during conditioning, two lower concentrations (half and a quarter the concentration), and two higher concentrations (two and four times higher). Trial 1B was performed as trial 1A, but 1 h before the trial we gave the lambs starch solution via gavage. The amount of starch corresponded to half the amount of wheat eaten by each individual on the previous day. Trial 1C was performed as trial 1A, but on the day before this trial the lambs did not receive any wheat, so they were somewhat energy deprived. In trial 1D, we offered the lambs unflavoured wheat ad libitum to ascertain any aversion to wheat. Trials 2A–D were performed as trials 1A–D, except for four changes. As mentioned above, in experiment 2, we (1) changed the flavour substance and (2) halved the flavour concentrations. We also made two more changes. (3) Since the lambs were given only 100 g of wheat during conditioning, they were probably energy deprived in trial 2A. Therefore, on the evening before this trial, we offered the lambs 200 g of rolled barley (a familiar food). (4) One hour and 15 min before trial 2B, we offered all lambs 200 g of rolled barley for 15 min. In all trials, we presented the five choices in separate containers with the conditioned flavour in the middle, the lower concentrations (in descending order) on one side, and the higher concentrations (in ascending order) on the other. On which sides the higher and lower concentrations were presented was decided randomly. During conditioning, the wheat container was placed randomly in the feed box. In both experiments, the schedule was: conditioning (days 1–7); trial A (day 8); conditioning (days 9–14); trial B (day 15); conditioning (days 16–21); trial C (day 22); trial D (day 23).
Novel food In this third experiment, we offered 10 lambs pure, ground wheat, and to 10 lambs we offered ground wheat to which we had added 5% (dry weight) dried onion powder (the same as in experiment 2). The wheat was offered ad libitum for 30 min each morning for 7 days. The lambs were completely naive to wheat, onion powder and to the plastic food containers.
mixed model ANOVA with conditioning group and test food (repeated measure) as sources of variation. When we found no interactive effects between these factors, we pooled the data and performed the tests again with test food as the only source of variation (Fig. 1, Table 1). To test for any changes in preference due to the pretreatments, we used mixed model ANOVA with conditioning group, pretreatment and test food as sources of variation. Pretreatment and test food were repeated measures. Since our computer program for statistical analysis allowed for only 10 repeated measures in each test, we performed these tests as pairwise analyses of the different trials (Table 2). We also tested for effects of the pretreatments on the total intake of test food. For the analyses we used repeated measures ANOVA, where the source of variation was trial (Fig. 2, Table 3).
Novelty We tested for effects of the flavour treatment on the intake of test food during the 7 days (Fig. 3) by using mixed model ANOVA with flavour treatment and test day (repeated measure) as the sources of variation (Table 4). In the ANOVA, we excluded those animals that ate less than 1 g of test food on all days (two in the ‘no flavour’ group, and one in the ‘flavour’ group), since we could not be sure that they ever sampled the wheat. To test for differences in intake between the two groups on separate days we used one-way ANOVAs. The source of variation was flavour treatment. Here, as well, we excluded those animals that ate less than 1 g of test food on all days.
Statistical Methods We used SYSTAT, version 5.02, for all statistical tests, except for tests of normal distribution of data for which we used Wilk–Shapiro’s test (on share-ware ‘ODDJOB’, Dallal 1989) on residuals (Shapiro & Wilk 1965). We used Levene’s test for heteroscedasticity of variances (Levene 1960, cited in Wilkinson 1990, page 313). When either of these tests showed that it would be inappropriate to use normal data, we tried with logarithmic transformed data, otherwise we performed rank transformation (Potvin & Roff 1993; Seaman et al. 1994). Some of the data we used twice. To correct for this we used the Dunn–Sida´k method (Ury 1976, in Sokal & Rohlf 1995). Where the result was close to being nonsignificant we present the corrected value corresponding to P=0.05 in the text. RESULTS
Generalization Gradients Conditioning treatments
Test Design Generalization gradients We tested for effects of the conditioning treatments and the flavour treatment on the intake of test food by using
In the first two trials of both experiments, the different conditioning treatments had no significant effects on subsequent intake (Table 1). In the third trial of experiment 1, there was a (weak) significant effect of the conditioning treatments (Table 1). However, when we
340 ANIMAL BEHAVIOUR, 56, 2
Experiment 1 300
No pretreatment
Energy preload
Energy deprived
200
Amount eaten (g)
100
0
Experiment 2 150
100
50
0
× 1/4 × 1/2
×1
×2
×4
× 1/4 × 1/2 × 1 ×2 ×4 Flavour concentration
× 1/4 × 1/2
×1
×2
×4
Figure 1. Intake of the five test foods. One food had the same flavour concentration as the one to which the lambs were conditioned. Of the other four, two had lower and two had higher flavour concentrations. The data are given as means+SD for all lambs, for each test food. Note that the scales on the Y axes are different for the two experiments. The time allowed to feed was 30 min and 15 min, in experiments 1 (apple) and 2 (onion), respectively.
split the data and tested the different conditioning groups against each other we did not find any logical or consistent effects. When we looked at the graphical representation of the data, we considered there might have been a weak trend for animals conditioned to higher flavour concentrations to have a stronger preference for the foods with lower flavour concentrations, but again this is uncertain. In the third trial of experiment 2, there was both a significant effect of the conditioning treatments and of the interaction Conditioning*Test food (Table 1). Here, the main effect of the conditioning treatment was that the animals in conditioning group 2 (0.5%) behaved differently from the others as they strongly preferred the same flavour concentration as the one to which they were conditioned. Going back to the original data, we found that of the six animals in this group, two ate almost only from this particular food (240.2 and 242.8 g, respectively) while another two ate mainly from this food. This could seem to imply that there was something wrong with the four alternatives, or something very appetitive with this particular food. However, each flavour concentration was mixed in one batch, and since the foods given to conditioning groups 1, 2 and 4 overlap (the food ‘1’ for group 2 is the same as the foods ‘2’ and ‘1/2’ foods for groups 1 and 4, respectively) and there were no such preferences or aversions shown by the animals in groups 1 and 4 we can only conclude that this was a chance effect. Such chance effects will of course be noticed more
in small sample sizes, but working on large animals one is often forced to make do with such. Therefore, in neither experiment could we find any effects of the conditioning treatments, at least no effects that we could make any sense of (either through logic or consistency).
Flavour treatments Contrary to our first hypothesis, the lambs consistently preferred foods with flavour concentrations lower than those used during conditioning (Fig. 1, Table 1). We saw three possible causes. (1) The flavour was inherently deterrent despite earlier results to the contrary. (2) The large amount of energy available in the morning (the lambs ate 200–400 g of wheat) caused an aversion to the flavour associated with the high-energy food. (3) The lambs did not generalize over the flavours as hypothesized. In experiment 2, to test for causes 1 and 2, respectively, we changed the flavour and halved the concentrations, and we offered the lambs less wheat during conditioning. Still, experiment 2 produced the same result as experiment 1 (Fig. 1, Table 1), and we must reject our first hypothesis.
Pretreatments In experiment 1, we found some significant effects of the pretreatments when we compared the first trial, A, to the other two (Table 2). The data indicated that the slope
AUGNER ET AL.: A FORAGING RULE OF THUMB 341
Table 1. Differences in intake of the five test foods Trial
Source
Experiment 1 A Between subjects Within subjects B
Between subjects Within subjects
C
Between subjects Within subjects
Experiment 2 A Between subjects Within subjects B
Between subjects Within subjects
C
Between subjects Within subjects
df
F
P
Conditioning (C) Error Test food (T) T*C Error Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error
3 14 4 12 56 3 14 4 12 56 3 14 4 12 56
0.24
0.087
3.18 0.86
0.020 0.58
0.64
0.60
11.6 1.56
<0.001 0.13
4.46
0.021
9.80 1.26
<0.001 0.27
Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error
3 20 4 12 80 3 20 4 12 80 3 20 4 12 80
1.20
0.34
6.33 1.25
<0.001 0.26
1.78
0.18
5.36 0.46
0.001 0.93
5.00
0.010
7.16 2.20
<0.001 0.019
The sources of variation were conditioning treatment and test food. Conditioning: in experiment 1, the lambs received ground wheat with either 0.5, 1.0, 1.5 or 2% apple flavour (dry weight) during conditioning. In experiment 2, these flavour concentrations were 0.25, 0.5, 0.75 and 1.0% onion flavour (dry weight). Test food: the lambs were simultaneously offered wheat with five flavour concentrations: the one used during conditioning, two lower concentrations (half and a quarter the concentration), and two higher concentrations (two and four times higher). A: no pretreatment; B: energy preload. C: energy deprived.
of the test food effect was lower in trial A than in the other two (Fig. 1). This may simply be because the lambs had still not got used to the test and conditioning food (ground wheat), as this difference is also reflected in the total intake of wheat (Fig. 2, Table 3). In the second experiment, the only significant effects that we found occurred when we compared trials A and C (Table 2). These effects are probably due to the same phenomenon as we discuss in the results of the conditioning treatments, above. The lack of effects could simply be that the pretreatments were not severe enough to affect the basic preference pattern (except perhaps in the last trial; Fig. 1, experiment 2, ‘energy deprived’). Consequently, our second and third hypotheses are rejected. We found no indication that the lambs formed any aversion to wheat (Fig. 2, Table 3).
Novelty In the individual intake data in experiment 3, we found a significant interaction between Day and Treatment
(Table 4), and split the data for further tests. We found no significant effect of test day in the ‘no flavour’ group (Table 4), but in the ‘flavour’ group intake increased significantly over the 7-day period (Table 4). On the first day, lambs offered onion-flavoured wheat ate significantly less than lambs given unadulterated wheat (F1,15 =22.5, P<0.001; Fig. 3). We found no significant difference in intake on the second day (F1,15 =4.21, P=0.058), or on the following test days (P>0.10). DISCUSSION Sheep readily learn to recognize foods (e.g. Provenza 1995), and aversions can be formed within hours (Provenza et al. 1994), or even minutes (Conrad et al. 1977), after exposure to unsuitable foods. There are no data on how fast preferences for flavours associated with beneficial foods can be formed in mammalian herbivores: in experiments where sheep have formed such preferences conditioning has lasted 4–16 days (Villalba & Provenza 1997a, b). Familiarity is important to mammalian herbivores, and food intake decreases if the
342 ANIMAL BEHAVIOUR, 56, 2
Table 2. Effects of pretreatments on intake test foods Test
Experiment 1 A versus B
B versus C
A versus C
Experiment 2 A versus B
B versus C
A versus C
Source
df
F
P
Pretreatment (P) P*Conditioning (C) Error P*Test food (T) P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error
1 3 14 4 12 56 1 3 14 4 12 56 1 3 14 4 12 56
16.2 0.53
<0.001 0.67
3.32 0.91
0.017 0.54
1.21 2.06
0.29 0.15
0.46 1.35
0.76 0.22
11.7 1.03
0.004 0.41
2.37 1.13
0.063 0.36
Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error
1 3 14 4 12 56
0.23 1.23
0.64 0.32
1.09 0.50
0.37 0.91
1.58 2.50
0.22 0.089
0.84 0.94
0.505 0.51
5.51 0.63
0.029 0.60
3.82 2.07
0.007 0.028
3 14 4 12 56 1 3 14 4 12 56
The sources of variation were trial pretreatment, conditioning treatment and test food (the latter two are repeated measures). A: no pretreatment; B: energy preload; C: energy deprived. For conditioning and test food, see Table 1. (P value corrected for data used twice: P0.05 =0.0253.) (See Table 1 for effects of conditioning and flavour treatments.) Experiment 1: apple flavour; experiment 2: onion flavour.
food or the flavour is changed (e.g. Biquand & BiquandGuyot 1992; Provenza et al. 1995). However, had familiarity been the only important factor in herbivore foraging decisions, then in our experiments the lambs should have preferred the flavour concentration to which they were conditioned, at the very least during the first trial. Still, they consistently preferred the weaker flavours (Fig. 1, Table 1). Experiments on aversion learning in rats, Rattus norvegicus (Tapper & Halpern 1968; Smith & Theodore 1984; Spector & Grill 1988) and in sheep (Launchbaugh et al. 1993) have yielded results similar to ours: that the animals preferred weaker flavours than the ones to which they were conditioned. One interpretation of this is that the animals learned to associate the flavour with toxicosis and avoided the conditioned flavour in relation to its concentration (i.e. they expressed an intensity gradient). Such an intensity gradient is a logical response following aversion learning, but it is less clear why we found the
same pattern in lambs offered a beneficial food, instead of the lambs expressing the expected bell-shaped generalization gradients. We found no indication of the lambs having formed any aversion to the test food (Fig. 2, Table 3), and the time was too short for the lambs to experience any postingestive responses during the trials. As for the flavours we used, both the substances as such and the concentrations were carefully chosen to be stimulatory neutral (after any initial neophobic response during the first week of conditioning). Experiments yielding bellshaped generalization gradients have typically involved animals conditioned to visual stimuli (e.g. Hanson 1959; Hearst 1968; Marsh 1972; Bowers & Richards 1986). It is quite possible that different learning and generalization mechanisms are acting in conjunction with different senses (Garcia 1989). Mammalian herbivores may prefer weak flavours because they do not learn to associate the flavour with the postingestive responses in a quantitative fashion, but
AUGNER ET AL.: A FORAGING RULE OF THUMB 343
Experiment 1 500
b
b
b
a 400 300
Amount eaten (g)
200 100 0 Experiment 2 300
a
ab
a b
200
100
0
No pretreatment
Energy preload
Energy deprived
No flavour
Figure 2. Total intake of test foods. The data are given as means+SD for all lambs. Different letters above the bars indicate a significant difference between them (see Table 3). Note that the scales on the Y axes are different for the two experiments. The time allowed to feed was 30 min and 15 min, in experiments 1 (apple) and 2 (onion), respectively. Table 3. Effects of pretreatments on the total intake of test foods Trial
Experiment 1 A versus B B versus C C versus D A versus D Experiment 2 A versus B B versus C C versus D A versus D
df
F
P
1,17 1,17 1,17 1,17
17.5 1.52 1.62 6.43
0.001 0.23 0.22 0.021
1,23 1,23 1,23 1,23
0.22 1.32 7.04 0.05
0.64 0.26 0.014 0.82
The source of variation was trial (repeated measure). A: no pretreatment; B: energy preload; C: energy deprived; D: unflavoured wheat. (P value corrected for data used twice; P0.05 =0.0253.) Experiment 1: apple flavour; experiment 2: onion flavour.
instead follow a rule of thumb of the kind ‘given a choice, avoid strong flavours’. If unpalatable food plants often have strong odours and/or taste (Eisner & Grant 1981), then such a rule could be part of a risk-averse foraging strategy. Theoretically, plants should always signal the presence of their defences (Augner 1994). (Here, ‘defence’ means a trait that reduces a plant’s fitness losses caused by herbivory, and which is detrimental to herbivores feeding on the plant; and ‘defence signal’ means any plant cue that herbivores can associate with the defence,
and which can be perceived before ingestion.) In another model, Leimar & Tuomi (in press) showed that for a continuously variable plant defence signal to increase in strength through natural selection, a peak shift in the herbivores’ generalization gradient over the signal is necessary. This implies that if herbivores learn to avoid plant defence signals and avoid plants in relation to the signal strength, then stronger signals would be selected for. Therefore, theory suggests that we should expect defended plants to have strong signals and generalist herbivores to avoid (strong) signals associated with plant defences. These are quite logical premises, and the latter is also implied by results in experiments on taste aversion learning in rats (Tapper & Halpern 1968; Smith & Theodore 1984; Spector & Grill 1988) and in sheep (Launchbaugh et al. 1993). It is harder to find a rationale for why herbivores should avoid beneficial foods with strong flavours. Mammalian herbivores can learn to eat food plants with strong flavours growing in the wild, such as onion, Allium sp. (e.g. Cheeke 1991; Stubbendieck et al. 1992). However, as we show in our third experiment, lambs avoid strong flavours, at least initially, even when they are not given a choice: the flavour added to the wheat was aversive on the first day (Table 4). It is conceivable that if unpalatable (defended) plants generally have stronger effects on the herbivore’s physiological homeostasis than do palatable plants, then to avoid strong flavours can be beneficial even if this means that herbivores temporarily do not utilize palatable plants to the full. Such avoidance behaviour can have evolved to reduce the risk of feeding on detrimental foods, and as a side effect it also influences the feeding preferences of beneficial foods. We do not propose that an avoidance rule of thumb is inflexible, but that it provides a baseline on which other factors are acting; its expression is determined by a herbivore’s nutritional status and its previous experiences (i.e. the familiarity–novelty spectrum: e.g. Rozin & Schulkin 1990; Wang & Provenza 1996). It has been suggested that plant cues, such as odours and flavours, should be deterrent only when they are correlated with some detrimental quality of the plant (e.g. Rhoades 1979; Augner 1994), or if herbivores generalize over such cues from one food to another (e.g. Launchbaugh & Provenza 1993). In such cases, one could say that a plant smells or tastes ‘bad’. Here, we propose that the odour or flavour of a plant can be deterrent because it smells or tastes ‘unfamiliar’. We do not suggest that strong preingestive plant cues and actual plant defences are mutually exclusive, nor that such preingestive cues cannot also be correlated with defences. However, the mechanisms of deterrence would differ. On the one hand, plant cues can be deterrent to herbivores because they are correlated with potentially detrimental effects of some other trait (either a learned or an innate aversion). On the other hand, a plant cue could be deterrent because it is unfamiliar (basically an innate response that can be modified through experience). To conclude, we propose that the behaviour we observed could be an expression of a rule of thumb of the type ‘given a choice, avoid food with strong flavours’, and
344 ANIMAL BEHAVIOUR, 56, 2
Table 4. Effects of flavour treatments and test day on the intake of two test foods Source
Both groups, all 7 test days Between subjects Treatment Within subjects Day Day*Treatment The two groups separately ‘No flavour’ group Day ‘Flavour’ group Day
df
F
P
1,15 6,90 6,90
0.294 1.92 3.16
0.60 0.086 0.007
6,42 6,48
1.24 3.84
0.30 0.003
The sources of variation are treatment and day (repeated measure). Treatment: either unadulterated ground wheat or ground wheat with 5% dried onion powder added. Day: test day.
200
X
SD No flavour
X
SD Flavour
NS NS
150 Amount eaten (g)
NS
NS 100 NS NS 50
0
*
1
2
3
4 Test day
5
6
7
Figure 3. Intake of test foods during a 7-day experiment on lambs. The data are given as mean values +SD. The animals were divided into two groups; one group was offered unadulterated ground wheat, the other was offered ground wheat with 5% (dry weight) dried onion powder added. In the ‘no flavour’ group, N=8; in the ‘flavour’ group, N=9; (three animals were excluded because they ‘ate’ less than 1 g of test food on all 7 test days). *P<0.001.
which is part of a risk-averse foraging strategy of mammalian herbivores. In nature, for such a rule to evolve it is necessary that strong flavours are associated with detrimental plant substances often enough for it to be economical for herbivores to follow the rule. Unfortunately, there are currently no relevant data on this topic. Acknowledgments We thank C. Cheney, T. Fagerstro ¨ m, J. Garcia and P. Lundberg for valuable comments. This paper is published with the approval of the Director, Utah Agricultural Experiment Station, Utah State University, Logan, as Journal Paper number 5007. M.A. was supported by the Swedish Natural Science Research Council, and by the Crafoord Foundation. References Augner, M. 1994. Should a plant always signal its defence against herbivores? Oikos, 70, 322–332.
Biquand, S. & Biquand-Guyot, V. 1992. The influence of peers, lineage and environment on food selection of the criollo goat (Capra hircus). Applied Animal Behavior Science, 34, 231–245. Bowers, R. L. & Richards, R. W. 1986. Generalization gradients following training on a symbolic matching-to-sample task. Animal Learning and Behavior, 14, 51–56. Cheeke, P. R. 1991. Applied Animal Nutrition: Feeds and Feeding. New York: McMillan. Conrad, H. R., Baile, C. A. & Mayer, J. 1977. Changing meal patterns and suppression of feed intake with increasing amounts of dietary nonprotein nitrogen in ruminants. Journal of Dairy Science, 60, 1725–1733. Dallal, G. E. 1989. ODDJOB: a collection of miscellaneous statistical techniques. American Statistician, 43, 270. Early, D. M. & Provenza, F. D. In press. Food flavor and nutritional characteristics alter dynamics of food preferences in lambs. Journal of Animal Science. Eisner, T. & Grant, R. P. 1981. Toxicity, odor aversion and ‘‘olfactory aposematism’’. Science, 213, 476. Garcia, J. 1989. Food for Tolman: cognition and cathexis in concert. In: Aversion, Avoidance and Anxiety (Ed. by T. Archer & L. Nilsson), pp. 45–85. Hillsdale, New Jersey: L. Erlbaum.
AUGNER ET AL.: A FORAGING RULE OF THUMB 345
Hanson, H. M. 1959. Effects of discrimination training on stimulus generalization. Journal of Experimental Psychology, 58, 321–334. Hearst, E. 1968. Discrimination learning as the summation of excitation and inhibition. Science, 162, 1303–1306. Kalat, J. W. 1974. Taste salience depends on novelty, not concentration, in taste-aversion learning in the rat. Journal of Comparative and Physiological Psychology, 86, 47–50. Launchbaugh, K. L. & Provenza, F. D. 1993. Can plants practice mimicry to avoid grazing by mammalian herbivores? Oikos, 66, 501–504. Launchbaugh, K. L., Provenza, F. D. & Burritt, E. A. 1993. How herbivores track variable environments: response to variability of phytotoxins. Journal of Chemical Ecology, 19, 1047–1056. Leimar, O. & Tuomi, J. In press. Synergistic selection and graded traits. Evolutionary Ecology. Levene, H. 1960. Robust tests for equality of variance. In: Contributions to Probability and Statistics (Ed. by I. Olkin), pp. 278–292. Palo Alto: Standard University Press. Marsh, G. 1972. Predictions of the peak shift in pigeons from gradients of excitation and inhibition. Journal of Comparative and Physiological Psychology, 81, 262–266. NRC. 1985. Nutrient Requirements of Sheep. Washington, D.C.: National Academy Press. Potvin, C. & Roff, D. A. 1993. Distribution-free and robust statistical methods: viable alternatives to parametric statistics. Ecology, 74, 1617–1628. Provenza, F. D. 1995. Postingestive feedback as an elementary determinant of food preference and intake in ruminants. Journal of Range Management, 48, 2–17. Provenza, F. D. 1996. Acquired aversions as the basis for varied diets of ruminants foraging on rangelands. Journal of Animal Science, 74, 2010–2020. Provenza, F. D., Lynch, J. J. & Nolan, J. V. 1993. The relative importance of mother and toxicosis in the selection of foods by lambs. Journal of Chemical Ecology, 19, 313–323. Provenza, F. D., Lynch, J. J., Burritt, E. A. & Scott, C. D. 1994. How goats learn to distinguish between novel foods that differ in postingestive consequences. Journal of Chemical Ecology, 20, 609– 624. Provenza, F. D., Lynch, J. J. & Cheney, C. D. 1995. Effects of a flavor and food restriction on the intake of novel foods by sheep. Applied Animal Behavior Science, 43, 83–93.
Rhoades, D. F. 1979. Evolution of plant chemical defense against herbivores. In: Herbivores: Their Interaction with Plant Secondary Metabolites (Ed. by G. A. Rosenthal & D. F. Janzen), pp. 3–54. New York: Academic Press. Rozin, P. N. & Schulkin, J. 1990. Food selection. In: Neurobiology of Food and Fluid Intake. Handbook of Behavioral Neurobiology, Vol. 10 (Ed. by E.M. Stricker), pp. 297–328. New York: Plenum. Seaman, J. W., Walls, S. C., Wise, S. E. & Jaeger, R. G. 1994. Caveat emptor: rank transform methods and interaction. Trends in Ecology and Evolution, 9, 261–263. Shapiro, S. S. & Wilk, M. B. 1965. An analysis of variance test for normality (complete samples). Biometrika, 52, 591–611. Smith, D. V. & Theodore, R. M. 1984. Conditioned taste aversions: generalization to taste mixtures. Physiology and Behavior, 32, 983–989. Sokal, R. R. & Rohlf, F. J. 1995. Biometry. 3rd edn. New York: W. H. Freeman. Spector, A. C. & Grill, H. J. 1988. Differences in the taste quality of maltose and sucrose in rats: issues involving the generalization of conditioned taste aversions. Chemical Senses, 13, 95–113. Spence, K. W. 1937. The differential response in animals to stimuli varying within a single dimension. Psychological Review, 44, 430– 444. Stubbendieck, J., Hatch, S. L. & Butterfield, C. H. 1992. North American Range Plants. 4th edn. Lincoln: University of Nebraska Press. Tapper, D. N. & Halpern, B. P. 1968. Taste stimuli: a behavioral categorization. Science, 161, 708–710. Ury, H. K. 1976. A comparison of four procedures for multiple comparisons among means (pairwise contrasts) for arbitrary sample sizes. Technometrics, 18, 89–97. Villalba, J. J. & Provenza, F. D. 1997a. Preference for wheat straw by lambs conditioned with intraruminal infusions of starch. British Journal of Nutrition, 77, 287–297. Villalba, J. J. & Provenza, F. D. 1997b. Preference for flavoured foods by lambs conditioned with intraruminal administrations of nitrogen. British Journal of Nutrition, 78, 545–561. Wang, J. & Provenza, F. D. 1996. Food preference and acceptance of novel foods by lambs depend on the composition of the basal diet. Journal of Animal Science, 74, 2349–2354. Wilkinson, L. 1990. SYSTAT: The System for Statistics. Evanston: Systat.
A rule of thumb in mammalian herbivores? MAGNUS AUGNER, FREDERICK D. PROVENZA & JUAN J. VILLALBA
Department of Rangeland Resources, Utah State University (Received 19 May 1997; initial acceptance 24 June 1997; final acceptance 12 December 1997; MS. number: 5549R)
ABSTRACT In two experiments on appetitive learning we conditioned lambs, Ovis aries, to particular concentrations of a flavour by mixing the flavour with an energy-rich food that complemented their energy-poor diet. The lambs were subsequently offered energy-rich food with five different concentrations of the flavour (the concentration to which they were conditioned, two higher concentrations, and two lower concentrations). At these tests, the lambs consistently preferred the weaker flavours. This finding stands in contrast to earlier results on generalization gradients. In a third experiment, similarly designed to the other two, we tested for effects of a strong flavour on the behaviour of lambs when they were offered a novel nutritious food. Half of the lambs were offered unadulterated wheat, and the others strongly flavoured wheat. We found that the flavour in itself was initially aversive. We propose that the lambs’ avoidance of foods with strong flavours may be an expression of a rule of thumb of the type ‘given a choice, avoid food with strong flavours’. Such a rule could be part of a risk-averse foraging strategy displayed by mammalian herbivores, and which could be of particular importance when they encounter unfamiliar foods.
ences made by itself or by accompanying conspecifics. For instance, a plant with a familiar odour may be more readily accepted and sampled than one with an unfamiliar odour (e.g. Provenza 1996). However, when a plant and the sensory stimuli associated with it are truly novel the animal has no previous experience yet still has to make a decision. Mammalian herbivores generalize in a qualitative fashion over cues (sensory stimuli) provided by unpalatable foods (e.g. Launchbaugh & Provenza 1993). However, it is not known if they generalize quantitatively, that is, there are no studies of generalization gradients of food-associated cues in mammalian herbivores. A generalization gradient is defined as the intensity of the behavioural response of an animal as a function of the stimulus, when the stimulus varies (quantitatively) in onedimensional space, for example, in brightness or in flavour concentration (Spence 1937). The intensity of the response is generally strongest to stimuli that are close to the stimulus to which the animal is conditioned; as the stimuli diverge, the responses decrease in intensity. A conditioned generalization gradient could be seen as a learnt behaviour that gives rise to a kind of rule of thumb: ‘react the strongest to familiar stimuli’. Mammalian herbivores often show a high degree of neophobia when first encountering an unfamiliar food (e.g. Provenza et al. 1995; Provenza 1996). Furthermore, a familiar food that is normally readily eaten is sampled
The use of information provided by plants is fundamental to the whole foraging process of herbivores. The plant community provides herbivores with large amounts of (potential) information, in the form of olfactory, gustatory, visual and tactile stimuli. These stimuli can be perceived both before and during feeding. Taken together, there are probably no two plants whose ‘information signatures’ are identical. This large amount of information potentially constitutes a problem for generalist herbivores, and they must be selective as to what fraction of the potential information they evaluate. Consequently, the ability to learn to recognize foods and the ability to generalize over cues are potentially important traits. It has been proposed that familiarity is the major determinant of animals’ responses to foods and flavours (e.g. Kalat 1974; Rozin & Schulkin 1990). Another information-related problem for generalist herbivores is that they may encounter plants with unfamiliar information signatures in new environments as well as in familiar surroundings. When such a novel plant is encountered the herbivore can either (1) ignore it, (2) sample it to some extent, or (3) eat from it in an uninhibited fashion. In this decision, the herbivore may rely on earlier experiCorrespondence and present address: M. Augner, Department of Theoretical Ecology, Lund University, Ecology Building, S-223 62 Lund, Sweden (email: [email protected]). F. D. Provenza and J. J. Villalba are at the Department of Rangeland Resources, Utah State University, Logan UT 84322-5230, U.S.A. 0003–3472/98/080337+09 $30.00/0
1998 The Association for the Study of Animal Behaviour
337
1998 The Association for the Study of Animal Behaviour
338 ANIMAL BEHAVIOUR, 56, 2
carefully if its flavour is changed (Provenza et al. 1993, 1995). We propose that when a herbivore encounters a given food, the whole preingestive information signature of the food is important in its decision whether to feed. In this decision, familiarity is an important factor (e.g. Biquand & Biquand-Guyot 1992; Provenza et al. 1995; Provenza 1996), and the stronger the sensory input, the easier a herbivore can determine whether a food is familiar or unfamiliar. Also, neophobia could be regarded as a kind of rule of thumb: ‘avoid unfamiliar foods’. However, in contrast to learned generalization gradients, neophobia seems to be an innate behaviour in mammalian herbivores. We performed two experiments to ascertain how lambs, Ovis aries, generalize across flavours associated with a beneficial food. The lambs were conditioned to a distinctly flavoured, energy-rich food (ground wheat). After conditioning, we simultaneously offered the lambs the energy-rich food with five different concentrations of the original flavour. Our hypotheses were: (1) lambs that received no special pretreatment would prefer the food with the flavour concentration to which they were conditioned; (2) after a preload with another energy-rich food, the lambs would prefer foods with lower flavour concentrations relative to the concentration to which they were conditioned (and the total intake of the energyrich food would decrease); and (3) after being deprived of energy, the lambs would prefer foods with higher flavour concentrations (and the total intake of the energy-rich food would increase). Hypotheses 2 and 3 imply that the pretreatments caused peak shifts in the generalization gradients (e.g. Leimar & Tuomi, in press), that is, displacements of the maximum response of the average generalization gradient towards lower and higher flavour concentrations, respectively. We performed a third experiment to ascertain the effects of a strong flavour on the intake of a novel beneficial food by lambs. The lambs were provided with a nutritionally balanced food marginal in energy. For a short time each day, half the lambs were offered a distinctly flavoured, energy-rich food (ground wheat), while the other half was offered unadulterated ground wheat. Our hypothesis here (4) was that a strong novel flavour is more unfamiliar than a weak novel flavour. By increasing the degree of neophobia a strong flavour can in itself be aversive. METHODS
Experimental Animals and Environment We performed the experiments at the Green Canyon Ecology Center, Utah State University, Logan, Utah, U.S.A. The lambs were kept outdoors, under a protective roof in individual, adjacent pens measuring 2.43.6 m. The lambs always had access to water, salt blocks and alfalfa pellets ad libitum. Before the experiments the lambs were given an adjustment period of about 3 weeks, during which we gave them 200 g of rolled barley per day. In experiment 1 we used 20 lambs (two of which refused to eat the flavoured wheat); and in experiment 2 we used 24 lambs. These lambs were commercial crossbreed
white-faced lambs, of both sexes, and approximately 3 months old. In experiment 3, we used 20 lambs which constituted a very homogeneous group: they were commercial crossbreed white-faced lambs, all male, approximately 60 days old, and they had been reared together. These lambs had been weaned for about a week when we received them, and they had some experience of solid foods (alfalfa pellets and rolled barley).
Treatments We gave the animals alfalfa pellets ad libitum, and as test food we used pure ground wheat. Alfalfa pellets constitute a nutritionally balanced food, but their energy content is marginal for the rapid growth of lambs (NRC 1985). In contrast, wheat is rich in readily digestible energy (NRC 1985), thus complementing the nutritional merits of alfalfa pellets. In the test food (ground wheat) we mixed flavour substances: it was mixed fresh daily (see Generalization gradients and Novel food, below). The lambs were initially naive to the flavours.
Generalization gradients In these two experiments, the lambs were randomly divided into four groups. Each group was conditioned to a given flavour concentration. We used different conditioning treatments because no experiment of this kind had been done before and we wanted to see if the flavour concentration, as such, had an effect on the outcome of conditioning. In experiment 1, the lambs received ground wheat with either 0.5, 1.0, 1.5 or 2% flavour (dry weight) during conditioning. In experiment 2, these flavour concentrations were 0.25, 0.5, 0.75 and 1.0% flavour (dry weight). As flavour substances, we used ‘green apple feed nectar’ in experiment 1, and dried onion powder in experiment 2. The feed nectar is an artificial flavour developed to stimulate food intake by livestock by giving them seemingly more varied diets (‘Green apple feed nectar (4501)’, Agrimerica Inc., Northbrook, IL, U.S.A.). To humans, this substance has a distinct odour and a weak, sweet taste. We used commercially available dried onion (‘Onion powder white’ (63775); Pacific Foods, Kent, WA, U.S.A.). Onions, Allium sp., occur naturally on rangelands in many parts of western U.S.A., and are readily eaten by sheep (e.g. Cheeke 1991). To the human palate, the onion powder had no taste but a very strong, aromatic odour. Since we cannot be sure that the onion powder did not provide the lambs with any taste, we prefer to call it a flavour rather than an odour. The flavour substances and concentrations used were carefully chosen to be stimulatory neutral, in the sense that after repeated exposures they are inherently neither attractive nor aversive to lambs; lambs often show marked neophobic responses to unfamiliar flavours and/or foods (e.g. Provenza et al. 1993, 1995; Provenza 1996). Our choices were based on earlier extensive preference/aversion tests (single and multiple choice tests) of flavours and flavour concentrations (F. D. Provenza et al., unpublished data). Furthermore, in earlier
AUGNER ET AL.: A FORAGING RULE OF THUMB 339
experiments, neither feed nectar (Early & Provenza, in press) nor dried onion (Villalba & Provenza 1997a) had any negative or positive postingestive effects. In experiment 1, wheat was offered ad libitum for 30 min each morning during conditioning and trials. In experiment 2, we offered the lambs 100 g of wheat for 15 min each morning during conditioning, and during the trials wheat was offered ad libitum for 15 min. Before these experiments, the lambs had 1 week’s experience with eating ground wheat. In trial 1A we simultaneously offered the lambs wheat with five flavour concentrations: the one used during conditioning, two lower concentrations (half and a quarter the concentration), and two higher concentrations (two and four times higher). Trial 1B was performed as trial 1A, but 1 h before the trial we gave the lambs starch solution via gavage. The amount of starch corresponded to half the amount of wheat eaten by each individual on the previous day. Trial 1C was performed as trial 1A, but on the day before this trial the lambs did not receive any wheat, so they were somewhat energy deprived. In trial 1D, we offered the lambs unflavoured wheat ad libitum to ascertain any aversion to wheat. Trials 2A–D were performed as trials 1A–D, except for four changes. As mentioned above, in experiment 2, we (1) changed the flavour substance and (2) halved the flavour concentrations. We also made two more changes. (3) Since the lambs were given only 100 g of wheat during conditioning, they were probably energy deprived in trial 2A. Therefore, on the evening before this trial, we offered the lambs 200 g of rolled barley (a familiar food). (4) One hour and 15 min before trial 2B, we offered all lambs 200 g of rolled barley for 15 min. In all trials, we presented the five choices in separate containers with the conditioned flavour in the middle, the lower concentrations (in descending order) on one side, and the higher concentrations (in ascending order) on the other. On which sides the higher and lower concentrations were presented was decided randomly. During conditioning, the wheat container was placed randomly in the feed box. In both experiments, the schedule was: conditioning (days 1–7); trial A (day 8); conditioning (days 9–14); trial B (day 15); conditioning (days 16–21); trial C (day 22); trial D (day 23).
Novel food In this third experiment, we offered 10 lambs pure, ground wheat, and to 10 lambs we offered ground wheat to which we had added 5% (dry weight) dried onion powder (the same as in experiment 2). The wheat was offered ad libitum for 30 min each morning for 7 days. The lambs were completely naive to wheat, onion powder and to the plastic food containers.
mixed model ANOVA with conditioning group and test food (repeated measure) as sources of variation. When we found no interactive effects between these factors, we pooled the data and performed the tests again with test food as the only source of variation (Fig. 1, Table 1). To test for any changes in preference due to the pretreatments, we used mixed model ANOVA with conditioning group, pretreatment and test food as sources of variation. Pretreatment and test food were repeated measures. Since our computer program for statistical analysis allowed for only 10 repeated measures in each test, we performed these tests as pairwise analyses of the different trials (Table 2). We also tested for effects of the pretreatments on the total intake of test food. For the analyses we used repeated measures ANOVA, where the source of variation was trial (Fig. 2, Table 3).
Novelty We tested for effects of the flavour treatment on the intake of test food during the 7 days (Fig. 3) by using mixed model ANOVA with flavour treatment and test day (repeated measure) as the sources of variation (Table 4). In the ANOVA, we excluded those animals that ate less than 1 g of test food on all days (two in the ‘no flavour’ group, and one in the ‘flavour’ group), since we could not be sure that they ever sampled the wheat. To test for differences in intake between the two groups on separate days we used one-way ANOVAs. The source of variation was flavour treatment. Here, as well, we excluded those animals that ate less than 1 g of test food on all days.
Statistical Methods We used SYSTAT, version 5.02, for all statistical tests, except for tests of normal distribution of data for which we used Wilk–Shapiro’s test (on share-ware ‘ODDJOB’, Dallal 1989) on residuals (Shapiro & Wilk 1965). We used Levene’s test for heteroscedasticity of variances (Levene 1960, cited in Wilkinson 1990, page 313). When either of these tests showed that it would be inappropriate to use normal data, we tried with logarithmic transformed data, otherwise we performed rank transformation (Potvin & Roff 1993; Seaman et al. 1994). Some of the data we used twice. To correct for this we used the Dunn–Sida´k method (Ury 1976, in Sokal & Rohlf 1995). Where the result was close to being nonsignificant we present the corrected value corresponding to P=0.05 in the text. RESULTS
Generalization Gradients Conditioning treatments
Test Design Generalization gradients We tested for effects of the conditioning treatments and the flavour treatment on the intake of test food by using
In the first two trials of both experiments, the different conditioning treatments had no significant effects on subsequent intake (Table 1). In the third trial of experiment 1, there was a (weak) significant effect of the conditioning treatments (Table 1). However, when we
340 ANIMAL BEHAVIOUR, 56, 2
Experiment 1 300
No pretreatment
Energy preload
Energy deprived
200
Amount eaten (g)
100
0
Experiment 2 150
100
50
0
× 1/4 × 1/2
×1
×2
×4
× 1/4 × 1/2 × 1 ×2 ×4 Flavour concentration
× 1/4 × 1/2
×1
×2
×4
Figure 1. Intake of the five test foods. One food had the same flavour concentration as the one to which the lambs were conditioned. Of the other four, two had lower and two had higher flavour concentrations. The data are given as means+SD for all lambs, for each test food. Note that the scales on the Y axes are different for the two experiments. The time allowed to feed was 30 min and 15 min, in experiments 1 (apple) and 2 (onion), respectively.
split the data and tested the different conditioning groups against each other we did not find any logical or consistent effects. When we looked at the graphical representation of the data, we considered there might have been a weak trend for animals conditioned to higher flavour concentrations to have a stronger preference for the foods with lower flavour concentrations, but again this is uncertain. In the third trial of experiment 2, there was both a significant effect of the conditioning treatments and of the interaction Conditioning*Test food (Table 1). Here, the main effect of the conditioning treatment was that the animals in conditioning group 2 (0.5%) behaved differently from the others as they strongly preferred the same flavour concentration as the one to which they were conditioned. Going back to the original data, we found that of the six animals in this group, two ate almost only from this particular food (240.2 and 242.8 g, respectively) while another two ate mainly from this food. This could seem to imply that there was something wrong with the four alternatives, or something very appetitive with this particular food. However, each flavour concentration was mixed in one batch, and since the foods given to conditioning groups 1, 2 and 4 overlap (the food ‘1’ for group 2 is the same as the foods ‘2’ and ‘1/2’ foods for groups 1 and 4, respectively) and there were no such preferences or aversions shown by the animals in groups 1 and 4 we can only conclude that this was a chance effect. Such chance effects will of course be noticed more
in small sample sizes, but working on large animals one is often forced to make do with such. Therefore, in neither experiment could we find any effects of the conditioning treatments, at least no effects that we could make any sense of (either through logic or consistency).
Flavour treatments Contrary to our first hypothesis, the lambs consistently preferred foods with flavour concentrations lower than those used during conditioning (Fig. 1, Table 1). We saw three possible causes. (1) The flavour was inherently deterrent despite earlier results to the contrary. (2) The large amount of energy available in the morning (the lambs ate 200–400 g of wheat) caused an aversion to the flavour associated with the high-energy food. (3) The lambs did not generalize over the flavours as hypothesized. In experiment 2, to test for causes 1 and 2, respectively, we changed the flavour and halved the concentrations, and we offered the lambs less wheat during conditioning. Still, experiment 2 produced the same result as experiment 1 (Fig. 1, Table 1), and we must reject our first hypothesis.
Pretreatments In experiment 1, we found some significant effects of the pretreatments when we compared the first trial, A, to the other two (Table 2). The data indicated that the slope
AUGNER ET AL.: A FORAGING RULE OF THUMB 341
Table 1. Differences in intake of the five test foods Trial
Source
Experiment 1 A Between subjects Within subjects B
Between subjects Within subjects
C
Between subjects Within subjects
Experiment 2 A Between subjects Within subjects B
Between subjects Within subjects
C
Between subjects Within subjects
df
F
P
Conditioning (C) Error Test food (T) T*C Error Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error
3 14 4 12 56 3 14 4 12 56 3 14 4 12 56
0.24
0.087
3.18 0.86
0.020 0.58
0.64
0.60
11.6 1.56
<0.001 0.13
4.46
0.021
9.80 1.26
<0.001 0.27
Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error Conditioning Error Test food T*C Error
3 20 4 12 80 3 20 4 12 80 3 20 4 12 80
1.20
0.34
6.33 1.25
<0.001 0.26
1.78
0.18
5.36 0.46
0.001 0.93
5.00
0.010
7.16 2.20
<0.001 0.019
The sources of variation were conditioning treatment and test food. Conditioning: in experiment 1, the lambs received ground wheat with either 0.5, 1.0, 1.5 or 2% apple flavour (dry weight) during conditioning. In experiment 2, these flavour concentrations were 0.25, 0.5, 0.75 and 1.0% onion flavour (dry weight). Test food: the lambs were simultaneously offered wheat with five flavour concentrations: the one used during conditioning, two lower concentrations (half and a quarter the concentration), and two higher concentrations (two and four times higher). A: no pretreatment; B: energy preload. C: energy deprived.
of the test food effect was lower in trial A than in the other two (Fig. 1). This may simply be because the lambs had still not got used to the test and conditioning food (ground wheat), as this difference is also reflected in the total intake of wheat (Fig. 2, Table 3). In the second experiment, the only significant effects that we found occurred when we compared trials A and C (Table 2). These effects are probably due to the same phenomenon as we discuss in the results of the conditioning treatments, above. The lack of effects could simply be that the pretreatments were not severe enough to affect the basic preference pattern (except perhaps in the last trial; Fig. 1, experiment 2, ‘energy deprived’). Consequently, our second and third hypotheses are rejected. We found no indication that the lambs formed any aversion to wheat (Fig. 2, Table 3).
Novelty In the individual intake data in experiment 3, we found a significant interaction between Day and Treatment
(Table 4), and split the data for further tests. We found no significant effect of test day in the ‘no flavour’ group (Table 4), but in the ‘flavour’ group intake increased significantly over the 7-day period (Table 4). On the first day, lambs offered onion-flavoured wheat ate significantly less than lambs given unadulterated wheat (F1,15 =22.5, P<0.001; Fig. 3). We found no significant difference in intake on the second day (F1,15 =4.21, P=0.058), or on the following test days (P>0.10). DISCUSSION Sheep readily learn to recognize foods (e.g. Provenza 1995), and aversions can be formed within hours (Provenza et al. 1994), or even minutes (Conrad et al. 1977), after exposure to unsuitable foods. There are no data on how fast preferences for flavours associated with beneficial foods can be formed in mammalian herbivores: in experiments where sheep have formed such preferences conditioning has lasted 4–16 days (Villalba & Provenza 1997a, b). Familiarity is important to mammalian herbivores, and food intake decreases if the
342 ANIMAL BEHAVIOUR, 56, 2
Table 2. Effects of pretreatments on intake test foods Test
Experiment 1 A versus B
B versus C
A versus C
Experiment 2 A versus B
B versus C
A versus C
Source
df
F
P
Pretreatment (P) P*Conditioning (C) Error P*Test food (T) P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error
1 3 14 4 12 56 1 3 14 4 12 56 1 3 14 4 12 56
16.2 0.53
<0.001 0.67
3.32 0.91
0.017 0.54
1.21 2.06
0.29 0.15
0.46 1.35
0.76 0.22
11.7 1.03
0.004 0.41
2.37 1.13
0.063 0.36
Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error Pretreatment P*Conditioning Error P*Test food P*T*C Error
1 3 14 4 12 56
0.23 1.23
0.64 0.32
1.09 0.50
0.37 0.91
1.58 2.50
0.22 0.089
0.84 0.94
0.505 0.51
5.51 0.63
0.029 0.60
3.82 2.07
0.007 0.028
3 14 4 12 56 1 3 14 4 12 56
The sources of variation were trial pretreatment, conditioning treatment and test food (the latter two are repeated measures). A: no pretreatment; B: energy preload; C: energy deprived. For conditioning and test food, see Table 1. (P value corrected for data used twice: P0.05 =0.0253.) (See Table 1 for effects of conditioning and flavour treatments.) Experiment 1: apple flavour; experiment 2: onion flavour.
food or the flavour is changed (e.g. Biquand & BiquandGuyot 1992; Provenza et al. 1995). However, had familiarity been the only important factor in herbivore foraging decisions, then in our experiments the lambs should have preferred the flavour concentration to which they were conditioned, at the very least during the first trial. Still, they consistently preferred the weaker flavours (Fig. 1, Table 1). Experiments on aversion learning in rats, Rattus norvegicus (Tapper & Halpern 1968; Smith & Theodore 1984; Spector & Grill 1988) and in sheep (Launchbaugh et al. 1993) have yielded results similar to ours: that the animals preferred weaker flavours than the ones to which they were conditioned. One interpretation of this is that the animals learned to associate the flavour with toxicosis and avoided the conditioned flavour in relation to its concentration (i.e. they expressed an intensity gradient). Such an intensity gradient is a logical response following aversion learning, but it is less clear why we found the
same pattern in lambs offered a beneficial food, instead of the lambs expressing the expected bell-shaped generalization gradients. We found no indication of the lambs having formed any aversion to the test food (Fig. 2, Table 3), and the time was too short for the lambs to experience any postingestive responses during the trials. As for the flavours we used, both the substances as such and the concentrations were carefully chosen to be stimulatory neutral (after any initial neophobic response during the first week of conditioning). Experiments yielding bellshaped generalization gradients have typically involved animals conditioned to visual stimuli (e.g. Hanson 1959; Hearst 1968; Marsh 1972; Bowers & Richards 1986). It is quite possible that different learning and generalization mechanisms are acting in conjunction with different senses (Garcia 1989). Mammalian herbivores may prefer weak flavours because they do not learn to associate the flavour with the postingestive responses in a quantitative fashion, but
AUGNER ET AL.: A FORAGING RULE OF THUMB 343
Experiment 1 500
b
b
b
a 400 300
Amount eaten (g)
200 100 0 Experiment 2 300
a
ab
a b
200
100
0
No pretreatment
Energy preload
Energy deprived
No flavour
Figure 2. Total intake of test foods. The data are given as means+SD for all lambs. Different letters above the bars indicate a significant difference between them (see Table 3). Note that the scales on the Y axes are different for the two experiments. The time allowed to feed was 30 min and 15 min, in experiments 1 (apple) and 2 (onion), respectively. Table 3. Effects of pretreatments on the total intake of test foods Trial
Experiment 1 A versus B B versus C C versus D A versus D Experiment 2 A versus B B versus C C versus D A versus D
df
F
P
1,17 1,17 1,17 1,17
17.5 1.52 1.62 6.43
0.001 0.23 0.22 0.021
1,23 1,23 1,23 1,23
0.22 1.32 7.04 0.05
0.64 0.26 0.014 0.82
The source of variation was trial (repeated measure). A: no pretreatment; B: energy preload; C: energy deprived; D: unflavoured wheat. (P value corrected for data used twice; P0.05 =0.0253.) Experiment 1: apple flavour; experiment 2: onion flavour.
instead follow a rule of thumb of the kind ‘given a choice, avoid strong flavours’. If unpalatable food plants often have strong odours and/or taste (Eisner & Grant 1981), then such a rule could be part of a risk-averse foraging strategy. Theoretically, plants should always signal the presence of their defences (Augner 1994). (Here, ‘defence’ means a trait that reduces a plant’s fitness losses caused by herbivory, and which is detrimental to herbivores feeding on the plant; and ‘defence signal’ means any plant cue that herbivores can associate with the defence,
and which can be perceived before ingestion.) In another model, Leimar & Tuomi (in press) showed that for a continuously variable plant defence signal to increase in strength through natural selection, a peak shift in the herbivores’ generalization gradient over the signal is necessary. This implies that if herbivores learn to avoid plant defence signals and avoid plants in relation to the signal strength, then stronger signals would be selected for. Therefore, theory suggests that we should expect defended plants to have strong signals and generalist herbivores to avoid (strong) signals associated with plant defences. These are quite logical premises, and the latter is also implied by results in experiments on taste aversion learning in rats (Tapper & Halpern 1968; Smith & Theodore 1984; Spector & Grill 1988) and in sheep (Launchbaugh et al. 1993). It is harder to find a rationale for why herbivores should avoid beneficial foods with strong flavours. Mammalian herbivores can learn to eat food plants with strong flavours growing in the wild, such as onion, Allium sp. (e.g. Cheeke 1991; Stubbendieck et al. 1992). However, as we show in our third experiment, lambs avoid strong flavours, at least initially, even when they are not given a choice: the flavour added to the wheat was aversive on the first day (Table 4). It is conceivable that if unpalatable (defended) plants generally have stronger effects on the herbivore’s physiological homeostasis than do palatable plants, then to avoid strong flavours can be beneficial even if this means that herbivores temporarily do not utilize palatable plants to the full. Such avoidance behaviour can have evolved to reduce the risk of feeding on detrimental foods, and as a side effect it also influences the feeding preferences of beneficial foods. We do not propose that an avoidance rule of thumb is inflexible, but that it provides a baseline on which other factors are acting; its expression is determined by a herbivore’s nutritional status and its previous experiences (i.e. the familiarity–novelty spectrum: e.g. Rozin & Schulkin 1990; Wang & Provenza 1996). It has been suggested that plant cues, such as odours and flavours, should be deterrent only when they are correlated with some detrimental quality of the plant (e.g. Rhoades 1979; Augner 1994), or if herbivores generalize over such cues from one food to another (e.g. Launchbaugh & Provenza 1993). In such cases, one could say that a plant smells or tastes ‘bad’. Here, we propose that the odour or flavour of a plant can be deterrent because it smells or tastes ‘unfamiliar’. We do not suggest that strong preingestive plant cues and actual plant defences are mutually exclusive, nor that such preingestive cues cannot also be correlated with defences. However, the mechanisms of deterrence would differ. On the one hand, plant cues can be deterrent to herbivores because they are correlated with potentially detrimental effects of some other trait (either a learned or an innate aversion). On the other hand, a plant cue could be deterrent because it is unfamiliar (basically an innate response that can be modified through experience). To conclude, we propose that the behaviour we observed could be an expression of a rule of thumb of the type ‘given a choice, avoid food with strong flavours’, and
344 ANIMAL BEHAVIOUR, 56, 2
Table 4. Effects of flavour treatments and test day on the intake of two test foods Source
Both groups, all 7 test days Between subjects Treatment Within subjects Day Day*Treatment The two groups separately ‘No flavour’ group Day ‘Flavour’ group Day
df
F
P
1,15 6,90 6,90
0.294 1.92 3.16
0.60 0.086 0.007
6,42 6,48
1.24 3.84
0.30 0.003
The sources of variation are treatment and day (repeated measure). Treatment: either unadulterated ground wheat or ground wheat with 5% dried onion powder added. Day: test day.
200
X
SD No flavour
X
SD Flavour
NS NS
150 Amount eaten (g)
NS
NS 100 NS NS 50
0
*
1
2
3
4 Test day
5
6
7
Figure 3. Intake of test foods during a 7-day experiment on lambs. The data are given as mean values +SD. The animals were divided into two groups; one group was offered unadulterated ground wheat, the other was offered ground wheat with 5% (dry weight) dried onion powder added. In the ‘no flavour’ group, N=8; in the ‘flavour’ group, N=9; (three animals were excluded because they ‘ate’ less than 1 g of test food on all 7 test days). *P<0.001.
which is part of a risk-averse foraging strategy of mammalian herbivores. In nature, for such a rule to evolve it is necessary that strong flavours are associated with detrimental plant substances often enough for it to be economical for herbivores to follow the rule. Unfortunately, there are currently no relevant data on this topic. Acknowledgments We thank C. Cheney, T. Fagerstro ¨ m, J. Garcia and P. Lundberg for valuable comments. This paper is published with the approval of the Director, Utah Agricultural Experiment Station, Utah State University, Logan, as Journal Paper number 5007. M.A. was supported by the Swedish Natural Science Research Council, and by the Crafoord Foundation. References Augner, M. 1994. Should a plant always signal its defence against herbivores? Oikos, 70, 322–332.
Biquand, S. & Biquand-Guyot, V. 1992. The influence of peers, lineage and environment on food selection of the criollo goat (Capra hircus). Applied Animal Behavior Science, 34, 231–245. Bowers, R. L. & Richards, R. W. 1986. Generalization gradients following training on a symbolic matching-to-sample task. Animal Learning and Behavior, 14, 51–56. Cheeke, P. R. 1991. Applied Animal Nutrition: Feeds and Feeding. New York: McMillan. Conrad, H. R., Baile, C. A. & Mayer, J. 1977. Changing meal patterns and suppression of feed intake with increasing amounts of dietary nonprotein nitrogen in ruminants. Journal of Dairy Science, 60, 1725–1733. Dallal, G. E. 1989. ODDJOB: a collection of miscellaneous statistical techniques. American Statistician, 43, 270. Early, D. M. & Provenza, F. D. In press. Food flavor and nutritional characteristics alter dynamics of food preferences in lambs. Journal of Animal Science. Eisner, T. & Grant, R. P. 1981. Toxicity, odor aversion and ‘‘olfactory aposematism’’. Science, 213, 476. Garcia, J. 1989. Food for Tolman: cognition and cathexis in concert. In: Aversion, Avoidance and Anxiety (Ed. by T. Archer & L. Nilsson), pp. 45–85. Hillsdale, New Jersey: L. Erlbaum.
AUGNER ET AL.: A FORAGING RULE OF THUMB 345
Hanson, H. M. 1959. Effects of discrimination training on stimulus generalization. Journal of Experimental Psychology, 58, 321–334. Hearst, E. 1968. Discrimination learning as the summation of excitation and inhibition. Science, 162, 1303–1306. Kalat, J. W. 1974. Taste salience depends on novelty, not concentration, in taste-aversion learning in the rat. Journal of Comparative and Physiological Psychology, 86, 47–50. Launchbaugh, K. L. & Provenza, F. D. 1993. Can plants practice mimicry to avoid grazing by mammalian herbivores? Oikos, 66, 501–504. Launchbaugh, K. L., Provenza, F. D. & Burritt, E. A. 1993. How herbivores track variable environments: response to variability of phytotoxins. Journal of Chemical Ecology, 19, 1047–1056. Leimar, O. & Tuomi, J. In press. Synergistic selection and graded traits. Evolutionary Ecology. Levene, H. 1960. Robust tests for equality of variance. In: Contributions to Probability and Statistics (Ed. by I. Olkin), pp. 278–292. Palo Alto: Standard University Press. Marsh, G. 1972. Predictions of the peak shift in pigeons from gradients of excitation and inhibition. Journal of Comparative and Physiological Psychology, 81, 262–266. NRC. 1985. Nutrient Requirements of Sheep. Washington, D.C.: National Academy Press. Potvin, C. & Roff, D. A. 1993. Distribution-free and robust statistical methods: viable alternatives to parametric statistics. Ecology, 74, 1617–1628. Provenza, F. D. 1995. Postingestive feedback as an elementary determinant of food preference and intake in ruminants. Journal of Range Management, 48, 2–17. Provenza, F. D. 1996. Acquired aversions as the basis for varied diets of ruminants foraging on rangelands. Journal of Animal Science, 74, 2010–2020. Provenza, F. D., Lynch, J. J. & Nolan, J. V. 1993. The relative importance of mother and toxicosis in the selection of foods by lambs. Journal of Chemical Ecology, 19, 313–323. Provenza, F. D., Lynch, J. J., Burritt, E. A. & Scott, C. D. 1994. How goats learn to distinguish between novel foods that differ in postingestive consequences. Journal of Chemical Ecology, 20, 609– 624. Provenza, F. D., Lynch, J. J. & Cheney, C. D. 1995. Effects of a flavor and food restriction on the intake of novel foods by sheep. Applied Animal Behavior Science, 43, 83–93.
Rhoades, D. F. 1979. Evolution of plant chemical defense against herbivores. In: Herbivores: Their Interaction with Plant Secondary Metabolites (Ed. by G. A. Rosenthal & D. F. Janzen), pp. 3–54. New York: Academic Press. Rozin, P. N. & Schulkin, J. 1990. Food selection. In: Neurobiology of Food and Fluid Intake. Handbook of Behavioral Neurobiology, Vol. 10 (Ed. by E.M. Stricker), pp. 297–328. New York: Plenum. Seaman, J. W., Walls, S. C., Wise, S. E. & Jaeger, R. G. 1994. Caveat emptor: rank transform methods and interaction. Trends in Ecology and Evolution, 9, 261–263. Shapiro, S. S. & Wilk, M. B. 1965. An analysis of variance test for normality (complete samples). Biometrika, 52, 591–611. Smith, D. V. & Theodore, R. M. 1984. Conditioned taste aversions: generalization to taste mixtures. Physiology and Behavior, 32, 983–989. Sokal, R. R. & Rohlf, F. J. 1995. Biometry. 3rd edn. New York: W. H. Freeman. Spector, A. C. & Grill, H. J. 1988. Differences in the taste quality of maltose and sucrose in rats: issues involving the generalization of conditioned taste aversions. Chemical Senses, 13, 95–113. Spence, K. W. 1937. The differential response in animals to stimuli varying within a single dimension. Psychological Review, 44, 430– 444. Stubbendieck, J., Hatch, S. L. & Butterfield, C. H. 1992. North American Range Plants. 4th edn. Lincoln: University of Nebraska Press. Tapper, D. N. & Halpern, B. P. 1968. Taste stimuli: a behavioral categorization. Science, 161, 708–710. Ury, H. K. 1976. A comparison of four procedures for multiple comparisons among means (pairwise contrasts) for arbitrary sample sizes. Technometrics, 18, 89–97. Villalba, J. J. & Provenza, F. D. 1997a. Preference for wheat straw by lambs conditioned with intraruminal infusions of starch. British Journal of Nutrition, 77, 287–297. Villalba, J. J. & Provenza, F. D. 1997b. Preference for flavoured foods by lambs conditioned with intraruminal administrations of nitrogen. British Journal of Nutrition, 78, 545–561. Wang, J. & Provenza, F. D. 1996. Food preference and acceptance of novel foods by lambs depend on the composition of the basal diet. Journal of Animal Science, 74, 2349–2354. Wilkinson, L. 1990. SYSTAT: The System for Statistics. Evanston: Systat.