| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Animal Nutrition and Health Department, Scottish Agricultural College, Edinburgh, EH9 3JG Scotland
2 Correspondence:
Animal Nutrition and Health Department, Scottish Agricultural College, Bush Estate, Penicuik EH26 0PH, Scotland (phone: 0131 535 3212; fax: 0131 535 3121; E-mail:
m.yeates{at}ed.sac.ac.uk).
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
Key Words: Cows • Feeding Behavior • Food Preferences • Meal Composition • Protein • Synchronization
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
The shortest unit of feeding that can be measured is often a visit to a feeder supplying one feed type only (e.g., Barrio et al., 2000; Bornett et al., 2000). Therefore, diet choice cannot be expressed during a single visit. However, visits are usually clustered into meals (Morgan et al., 2000; Yeates et al., 2001), and meals are therefore the shortest unit in which diet selection can be expressed. The aim of this study was to investigate whether long-term average diet choice is a direct result of cows selecting a consistent diet within meals. We developed a number of hypotheses (see below) on the relationship between long-term average diet choice and diet selection within meals and used large data sets of dairy cow feeding behavior for severe statistical tests of these hypotheses. Our analyses initially investigated feeding behavior in terms of visits, then utilized a novel technique for the analysis of how meals are composed of visits. Our analyses are discussed in relation to short-term feeding behavior in other species.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
Housing and Daily Routine
The experiments took place at the Langhill Dairy Cattle Research Centre (Edinburgh, Scotland). Holstein-Friesian cows were kept in a yard, as described by Tolkamp and Kyriazakis (1997), for the duration of the experiments. They left the yard twice daily during milking, which lasted for up to 60 min (between 0600 and 0800 and between 1600 and 1730). During each of these periods, cows received 0.5 kg of parlor concentrates. On rare occasions, cows also left the yard during short periods for management reasons, such as foot trimming and pregnancy diagnosis. During these times, cows had no access to the feeders.
Foods, Food Dispensers, and Feeding Regime
The yard was equipped with 28 computerized feed dispensers (Insentec B.V., Marknesse, The Netherlands) as described by Tolkamp et al. (1998a). Each feed dispenser consisted of a bin with a capacity of 160 L, mounted on two load cells. Access to the bin was via a pneumatically operated, computer-controlled gate. The gate was equipped with an antenna that responded to a transponder that each cow wore around its neck. Therefore, each cow could be given access to specific feeders. On entry to a feed bin, the cow’s identification number, the weight of the bin, and the time were recorded. When the cow left the feeder, the gate closed and the time was again recorded. The gate remained shut for at least 10 s to allow stabilization of the feed bin prior to reweighing. Time was measured to the nearest second, and weight to the nearest 0.1 kg. Feeders were equipped with yokes during Exp. 1 and Exp. 2 to stop "stealing" of feed by nonexperimental cows housed in the same yard.
Access to the feeders was continuous except during milking and between approximately 0800 and 0930, when feed residues were removed from the bins and fresh mixed feed was supplied. Approximately three-quarters of the daily feed was offered in the morning, with the remaining feed added to the bins during the afternoon milking. The quantities of feeds offered were calculated daily to allow at least 10% orts. Water was available ad libitum from two troughs situated at either end of the row of feed dispensers.
Experiment 1
Foods were a mixture of 70% grass silage and 30% concentrate on a fresh weight basis, and were formulated to have similar nutritional properties except for their protein contents, which were either high (H) or low (L). Table 1
shows that although effective rumen-degradable protein (eRDP; defined by AFRC (1993) as slowly degradable protein plus 80% of quickly degradable protein) were similar, differences in the concentrate components of the feeds resulted in two feeds that differed in yields of metabolizable protein (MP; defined by AFRC (1993) as an estimate of the availability of AA to the cow).
|
). Cow pressure was defined as the number of cows that on average would attempt to eat at each feeder if all cows tried to eat at the same time.
Experiment 2
Foods were a mixture of 70% grass silage and 30% concentrate on a fresh weight basis and were formulated to have similar nutritional properties except for their protein contents, which were either high (H2) or low (L2). Table 1 shows that differences in the concentrate components of the feeds resulted in two feeds that differed in eRDP and MP yield.
Feeding behavior data were collected during 33 wk from 16 cows. All cows had access to 12 feeders and 79,386 visits were recorded. Feeders 7 to 12 contained H2 and could be accessed by the experimental cows only. Feeders 13 to 18 contained L2 and could be accessed by experimental cows and some nonexperimental cows. Cows were recorded to select a diet that consisted of approximately 70% of H2. The access of nonexperimental cows to additional feeders was manipulated such that the cow pressure per experimental feeder (Table 1) was kept approximately equal for both feed types, despite the consistent nonrandom diet choice made by experimental cows (Tolkamp et al., 1998b).
Experiment 3
The feeds were a mixture of 80% grass silage and 20% concentrate on a fresh weight basis and were formulated to have similar nutritional properties except for their protein contents, which were either high (H3) or low (L3). Table 1 shows that differences in the concentrate components of the feeds resulted in two feeds that differed in eRDP and MP yield.
Feeding behavior data were collected from 24 cows during wk 4 and 5 of the experiment, as described by Tolkamp et al. (1998b). All cows had access to 28 feed dispensers. Feeders 8 to 14 and 22 to 28 contained H3. Feeders 1 to 7 and 15 to 21 contained L3. In addition, two control groups of six cows each had access to feeders supplying H3 and L3, respectively. A total of 20,971 visits to the feeders by choice cows were analyzed. Cows were recorded to select a diet that consisted of approximately 70% of H3. This led to a cow pressure per experimental feeder (Table 1) that was approximately twice as high at feeders supplying H3 compared to those supplying L3.
Visit-Based Analyses
The visit was the shortest unit in which intake was recorded. Therefore, the feeding behavior of individual animals in each experiment was first analyzed in terms of visits. We reasoned that if animals were eating randomly from the two feeds (as concluded from earlier analysis of daily intakes during Exp. 1), then they could be expected to visit feeders supplying H (H feeders) as frequently as those supplying L (L feeders), and to eat an amount per visit that was independent of the feed type consumed (hypothesis 1). Rejection of this hypothesis for Exp. 1 would cast serious doubts on earlier conclusions (Tolkamp et al., 1998c) that animals in this experiment ate at random. However, animals with a long-term diet choice that was different from random (e.g., Exp. 2 and Exp. 3) could achieve this by visiting feeders supplying a given feed more frequently, or by consuming more than the average amount when visiting feeders supplying a given feed, or some combination of these. As far as we know, an analysis of the short-term feeding behavior of choice-fed animals has never been published. However, Chambers et al. (1995) offered locusts a diet choice and monitored their feeding behavior in a study where diet choice was achieved mainly by more frequent visits to one feed type. We have taken this finding as the basis of our second hypothesis—that animals would achieve a nonrandom diet choice mainly by visiting feeders supplying a given feed type more frequently, rather than regulating their intake per visit depending on the feed type visited (hypothesis 2).
To facilitate comparisons between experiments, the following visit characteristics were calculated for each experiment: the intake, number of visits, and visit duration at H feeders as a proportion of the total intake, total number of visits, and total visit duration, respectively. If animals ate at random, then visit characteristics were expected to be unrelated to the type of feed being consumed. Therefore, the calculated proportion of each of the above visit characteristics would be 0.5. Proportions were calculated from the average individual cow visit characteristics and t-tests used to determine if the calculated proportions differed significantly from 0.5.
The following visit characteristics were calculated to allow comparisons between feed types within experiments: the mean intake per visit, mean duration per visit, and mean intake rate during visits to feeders supplying H and L. This was achieved by fitting the following model (using average intake per visit as an example): Yij = µ + Ci + Fj + eij, where Yij = average intake per visit for cow i when eating at feeder j; µ = average intake per visit; Ci = effect of cow i; Fj = effect of feeder j; and eij = error term. One-way ANOVA of the mean intake per visit for each feeder, with cow effect accounted for, was used to test for an affect of the feed type that each feeder supplied on the mean intake per visit.
Meal-Based Analyses
A meal criterion is an estimate of the longest nonfeeding interval between visits to a feeder that can be considered part of a meal. Calculation of a meal criterion allows the clustering of recorded visits into meals. The lengths of intervals between visits to the feeders were calculated for each cow as the time between the end of one visit and the start of the next. All loge-transformed interval lengths between visits to the feeder greater than zero were utilized to estimate individual meal criterion using the method of Yeates et al. (2001) (see Appendix A). After calculation of individual meal criteria, visits were clustered into meals for each cow. This resulted in meals that were composed of between one and more than 10 visits. For each observed meal, the proportion of visits to H feeders was calculated. For instance, the proportion of visits to H feeders in a two-visit meal could be 0, 0.50, or 1 and in a three-visit meal could be 0, 0.33, 0.67, or 1.
In order to utilize the methodology of analysis described below, it was essential to have the largest possible number of meals (i.e., pooling of information from individuals was required). We therefore investigated first if there was any evidence to suggest that the meal data from the individual cows should not be pooled. To that end, the individual SD of the proportion of visits to the H feeder in a meal were determined. Using a Kolmogorov-Smirnov normality test (Zar, 1996), the frequency distribution of the coefficient of variation was tested to determine if it differed from a normal distribution. This was not the case in any of the experiments (P > 0.15). Therefore, the meal data were pooled within experiments to give a frequency distribution of meal composition.
Subsequently, we tested whether the observed frequency distribution of meal compositions differed from a hypothetical frequency distribution of bout compositions, predicted using probability theory as described by Zar (1996). For each experiment, a frequency distribution of bout compositions was calculated with the assumption that each bout would consist of visits that were drawn randomly from the observed population of visits in that experiment. Therefore, the probability of visiting a feeder was assumed to be related only to the feed type it supplied, with all feeders that supplied the same feed type being equally likely to be visited. For instance, consider an experiment in which the proportion of visits to H feeders is 0.5. The probability that a randomly drawn visit from that population would be a visit to an H feeder would therefore be equal to the probability that the visit would be to an L feeder (i.e., 0.5). Thus, the probability that a bout consisting of two randomly drawn visits would contain both visits to the same feed type would be 0.5 x 0.5 = 0.25. Consequently, the probability that a bout would consist of one visit to each of the feed types would be 2 x 0.5 x 0.5 = 0.5. Similarly, for a bout consisting of three visits, the probability of all visits being to the same feed type would be 0.5 x 0.5 x 0.5 = 0.125. Likewise, the probability of two visits to one feed type and one to the other feed type would be 3 x 0.5 x 0.5 x 0.5 = 0.375. The same process can be repeated for bouts consisting of four or any other number of visits. In this manner, a frequency distribution of bout composition was constructed that could be expected for animals that did not regulate diet choice during a meal.
The same procedure can be followed for an experiment in which the proportion of observed visits to H feeders is, for example, 0.7. The probabilities are then 0.7 and 0.3 that a randomly drawn visit is to an H or L feeder, respectively. For a two-visit bout, the proportion of visits to an H feeder can be 0, 0.5, or 1. The probability of each of these occurring can then be calculated as 0.3 x 0.3 = 0.09; 2 x 0.3 x 0.7 = 0.42; and 0.7 x 0.7 = 0.49, respectively. Similarly, for a three-visit bout, the proportion of visits to an H feeder can be 0, 0.33, 0.67, or 1. The probability of each of these occurring can then be calculated as 0.3 x 0.3 x 0.3 = 0.027; 3 x 0.3 x 0.3 x 0.7 = 0.189; 3 x 0.3 x 0.7 x 0.7 = 0.441; and 0.7 x 0.7 x 0.7 = 0.343, respectively. In this manner, we again constructed a frequency distribution of bout composition that could be expected for animals that did not regulate diet choice during a meal.
For each experiment, observed frequency distributions of meal composition were compared with predicted bout composition to test for evidence of animals regulating diet choice within a meal. To avoid increasingly complicated calculations for very rarely occurring meals consisting of large numbers of visits, the above procedure was repeated when increasing the number of visits per meal until at least 90% of all the meals in an experiment were included in the comparison.
We used this approach to test the following hypotheses. If animals ate randomly from the two feeds (as expected for Exp. 1), then we would not expect the observed frequency distribution of meals to deviate from the predicted bout composition (hypothesis 3). Alternatively, if animals regulate a nonrandom diet choice on a meal basis, then the observed frequency distribution of meals would be expected to deviate from the predicted bout composition. More specifically, we hypothesized that meals with a composition close to the long-term average diet choice were expected to occur much more frequently than predicted by randomly composed bouts. In contrast, we expected meals with a composition far from the long-term diet choice to occur much less frequently than predicted by randomly composed bouts (hypothesis 4). We used 
2 analyses to test for differences between recorded meal and predicted bout compositions. To avoid the disturbing effects on the 2 analysis of classes with very low numbers of observations (Zar, 1996), frequencies were grouped using the same 11 classes that were arranged symmetrically about a diet choice of 0.5 in all experiments.
Testing the Relationship Between Proportion of Visits to, and Intake from, H Feeders
In theory, an animal that consumes a meal with a proportion of visits to an H feeder that is far from the long-term proportion of H selected in the diet can "compensate" by consuming different amounts of feed depending on the feed type it is eating. For instance, an animal that selects 2/3 H in the long-term can still select 2/3 H during a meal, even if the proportion of visits to the H feeders during that meal is only 1/2. This can be achieved by consuming twice as much feed during visits to the H as opposed to the L feeders. Therefore, any sign that the proportion of H consumed during a meal deviates systematically from the proportion of visits to feeders supplying H during the meal, would be evidence for diet choice on a meal basis. We investigated this possibility by first calculating the expected diet choice, assuming that this was not regulated within a meal. Subsequently, differences between expected and observed diet choice were calculated and regressed on the proportion of visits to H feeders. The null hypothesis was that animals would attempt to regulate diet choice within a meal by compensating in this way (hypothesis 5). A significant negative regression coefficient would provide evidence for diet choice on a meal basis, whereas a nonsignificant regression coefficient would falsify our hypothesis.
|
Results and Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
shows the consistent nature of group mean diet choice in the three experiments. During the 4 wk of Exp. 1, diet choice remained close to 500 g of H/kg of intake (i.e., cows apparently ate a random diet). In contrast, diet choice was consistently around 700 g of H/kg of intake during the 33 wk of Exp. 2, and the 2 wk of Exp. 3 (i.e., clearly different from random intake). Table 2 gives a summary of the daily intake, number of visits, and visit duration of the cows in each of the three experiments. It can be seen that the duration of visits differed considerably between Exp. 3 and those of Exp. 1 and Exp. 2. Part of these differences is probably related to differences in feed composition between experiments. The higher silage to concentrate ratio in Exp. 3 compared to the other experiments was associated with a higher average daily visit duration in this experiment. This is in agreement with observations by Friggens et al. (1998). The lower number of daily visits in Exp. 1 and Exp. 2 compared to Exp. 3, was very likely related to the presence of yokes, which have been found to significantly affect recorded visit characteristics, though not to affect the total daily intake (Tolkamp et al., 2000). The total daily DM intake was similar for cows in all three experiments (Table 2).
|
|
shows that no aspect of the recorded visit characteristics from Exp. 1 (expressed as a proportion) differed from 0.5 (i.e., none differed from random). This is in agreement with the visit characteristics presented in Table 4 which also shows that animals in this experiment did not consume a diet that differed from random. Therefore, as animals in Exp. 1 consumed a diet at random from the two feeds, we cannot reject hypothesis 1. This confirms previous analysis of the diet choice for these cows based on daily intakes (Tolkamp et al. 1998c). Table 4shows that within all experiments, feeding rate was unaffected by the feed being consumed. Therefore, animals show no evidence of regulating choice by altering the rate at which they consume the feeds. However, when comparing the feeding rate across experiments, the lower rate in Exp. 3 is evident. This is likely to be related to the higher silage to concentrate ratio in Exp. 3, compared to Exp. 1 and Exp. 2 (Tolkamp et al., 2002). Table 3 shows that, in contrast to Exp. 1, all recorded visit characteristics from Exp. 2 and Exp. 3 differed from that expected if animals visited feeders at random (P < 0.001). Mean visit duration and consequently intake per visit (Table 4) were seen to be higher (P < 0.001) when visiting H feeders in Exp. 3, with the same (nonsignificant) trend recorded in Exp. 2 (P = 0.055). However, the recorded intake per visit to H feeders in Exp. 3 was only 23% higher than the intake per visit during visits to L feeders. If a diet choice of 0.73 H (Table 3) was achieved by having higher intake per visit at H compared to L feeders then the intake per visit at feeders supplying H would need to be 170% higher than at the L feeders. Clearly, the higher intake per visit at H feeders contributed relatively little to regulating diet choice. Therefore, we cannot reject hypothesis 2, as cows appear to select their diet mainly by visiting H feeders more frequently than L feeders.
|
|
shows the recorded frequency distribution of the loge-transformed interval lengths between visits, pooled across cows, and the fit of the model of Yeates et al. (2001). The observations consist of three populations; there is a peak at long intervals, and two overlapping peaks at shorter interval lengths. The peak at long interval lengths is associated with the long intervals between meals (Tolkamp et al., 1998a), whereas the other peaks represent within-meal intervals when cows do, or do not, visit the drinker (Tolkamp and Kyriazakis, 1999). The meal criterion is estimated at the point where these distributions cross. Figure 2 shows there is little overlap between the final two populations, therefore the meal criterion can be reliably determined as few intervals will be assigned to the wrong population (Yeates et al., 2001). Table 5 gives the average meal criteria for the cows in each of the experiments. The estimated meal criteria were similar for cows in all three experiments and were comparable with estimated meal criteria found by Tolkamp et al. (2002). The frequency distribution of interval lengths (Figure 2) from Exp. 3 differs from the other experiments as it contains a smaller proportion of long intervals. This is a result of the higher daily number of visits in this experiment (Table 2), and consequently, there is a lower proportion of between meal intervals.
|
|
gives a summary of the meal characteristics for the three experiments. The mean number of meals each day was remarkably similar between experiments and across cows, considering the large difference in mean number of visits per day (Table 2). This is in agreement with Tolkamp et al. (2000), who found the meal to be a more biologically relevant unit of intake than the visit. Intake per meal was also similar between experiments, however meal duration was greater in Exp. 3 compared to Exp. 1 and Exp. 2. This was a consequence of both the decreased intake rate (Table 4), which was probably linked to the high silage to concentrate ratio, and the greater number of visits per meal in this experiment. Therefore, as meal duration is a summation of time spent visiting the feeders and the duration of within-meal intervals, the increased number of visits leads to an increased number of intervals, and hence contributed to a greater meal duration.
The recorded long-term consistency in diet choice (Figure 1) could be a result of animals regulating diet choice within meals by composing many meals with a diet choice that approximates the long-term diet choice. Alternatively, the long-term consistency in diet choice could be the result of "averaging out" over multiple meals with a wide range of compositions. If the meal is the unit of diet choice regulation, then during Exp. 2 and Exp. 3, we would expect to see many meals with a composition of approximately two-thirds of the visits to the H feeders. Animals would be expected to avoid eating meals consisting of one feed type only, as the diet choice resulting from such meals is far from the long-term average. However, as cows in Exp. 1 ate a random diet, we would expect to see in this experiment a frequency distribution of meal and of bout composition that did not differ.
Meal-Based Analyses—Experiment 1
Figure 3a compares the distribution of recorded meal compositions (up to nine visits per meal) and predicted bouts. There is a remarkable similarity between the frequency distribution of observed meal and predicted bout compositions. However, contrary to our expectations, cows had a higher frequency of single-feed meals than predicted from the frequency distribution of bout compositions (Figure 3a). Associated with this, the observed frequency of meals with a composition similar to the long-term average was lower than expected. This resulted in a significant difference between the frequency distribution of recorded meals and predicted bouts (2 = 104, df = 10, P < 0.001). Prior analysis of diet choice based on daily intakes (Tolkamp et al., 1998c) and the visit-based analysis (shown above) strongly suggested that cows in this experiment ate randomly. However the results from the meal-based analysis (Figure 3a) did not agree with this. In principle, at least two explanations could account for this apparently nonrandom behavior. The cows could have been regulating their diet choice, despite avoiding meals with a composition similar to the long-term average diet composition. However, the evidence discussed above showed that this was not the case. Alternatively, this discrepancy could be the result of an error in the assumptions used to predict bout composition. Therefore, we examined the feeding data from these cows more closely to determine what might have caused this result.
|
was predicted on the assumption that the probability of a cow visiting a given feeder was constant. For Exp. 1, the probability of visiting each feed type was 0.5 (Table 3), and there were six feeders per feed type. Our predictions were based therefore, on the assumption that the likelihood of a cow visiting a feeder was 0.5/6 = 0.083 for each feeder. This likelihood was assumed to be independent of the feeder the cow visited previously. The records showed, however, that this was not the case. The probability of a cow revisiting the feeder it had just left was 0.23 (i.e., much higher than expected from our assumption of randomness). Revisits were observed to a similar extent for H and for L feeders. The propensity of cows to revisit the same feeder was affected by time of the day. The highest incidence of revisits to the same feeder within a meal were recorded during periods of high feeding activity (e.g., immediately after fresh feed supply) and the lowest during quiet periods (i.e., at night).
Observations of cows in our experimental facility have shown that cows may be disturbed by other cows when visiting a feeder. Cows were seen to leave the feeder momentarily and then re-enter the same feeder, where they continued to feed. Therefore, it seems likely that the higher probability of cows revisiting the same feeder is related to this disturbance of feeding animals. Since revisits to the same feeder imply another visit to the same feed type, this higher likelihood can be expected to result in a higher frequency of single-feed meals than if the probability of visiting all feeders is assumed to be equal. At the same time, this would decrease the relative frequency of meals with mixed composition, as observed in our data (Figure 3a).
We recalculated the expected frequency distribution of bout compositions taking into account this higher probability of revisits to the same feeder (see Appendix B). Figure 3b shows the adjusted frequency distribution of bout composition. Differences between recorded meal and predicted bout compositions, as seen in Figure 3a, have largely disappeared and are no longer significant (2 = 15, df = 10, P > 0.1). Therefore we cannot reject hypothesis 3, because when the higher probability of revisits to the same feeder was taken into account, the bouts predicted on the basis of random behavior did not differ from the actual meals composed by cows in Exp. 1.
The results from Exp. 1, where cows ate at random, were therefore used to test the underlying assumptions used to predict bout composition. This led to the development of a refined methodology to calculate bout composition, taking into account revisits to the same feeder. We used this methodology to predict the bout compositions for Exp. 2 and Exp. 3 (see below). Our hypotheses could then be tested with confidence in the assumptions used to predict bout composition.
Meal-Based Analyses—Experiment 2
Figure 3c compares the distribution of recorded meal compositions (up to 10 visits per meal) and predicted bouts. The remarkable similarity between recorded meal and predicted bout compositions was again evident. The observations are clearly skewed toward meals with a proportion of visits to the H feeders that reflects the long-term diet choice (Table 3). However, there are fewer such meals than predicted by the composition of bouts. Again, there is a corresponding excess of single-feed meals. This difference between the frequency distribution of observed meals and predicted bouts was significant (2 = 1,030, df = 10, P < 0.001). Examination of the feeding patterns of these cows revealed an elevated probability of a revisit to the same feeder, as observed in Exp. 1. The expected meal compositions were therefore adjusted using the methodology developed with data from Exp. 1, using the proportion of revisits found in Exp. 2 (0.32). Figure 3d gives the recorded meal and adjusted predicted bout compositions. The differences between recorded meal and predicted bout compositions seen in Figure 3c have almost disappeared, but are still significant (2 = 137, df = 10, P < 0.001). Figure 3d shows that this significant difference was largely the result of differences between recorded meal and predicted bout compositions for meals and bouts composed of visits to one feed type only. The frequencies of meals and bouts with a composition similar to the long-term average diet selection were approximately equal. If cows regulated diet choice within meals, we expected to see a very high frequency of meals with a composition similar to the long-term average diet choice. As this was not the case, the analysis provides no evidence that animals regulate diet choice on a meal basis. Therefore, we must reject hypothesis 4 on the basis of Exp. 2.
Meal-Based Analyses—Experiment 3
Figure 3e compares the distribution of recorded meal compositions (up to 19 visits per meal) and predicted bouts. The discrepancy between recorded meals and predicted bouts can be seen; again, there are many more single-feed meals than expected. This resulted in a significant difference between the frequency distribution of recorded meals and predicted bouts (2 = 1,365, df = 10, P < 0.001). Also, examination of recorded feeding behavior showed that revisits occurred more than would be expected on the basis of randomness. The expected meal compositions were therefore adjusted using the method developed in Exp. 1, using the proportion of revisits found in Exp. 3 (0.27). Figure 3f gives the frequency distribution of recorded meal and adjusted predicted bout compositions. The difference between these frequency distributions is still considerable even after correcting for revisits (2 = 317, df = 10, P < 0.001). Therefore, there is no evidence that cows in Exp. 3 attempted to compose many meals with a composition that reflects the long-term diet composition; indeed, there are more meals whose composition is further from the long-term diet composition than predicted. Therefore, we must reject hypothesis 4 on the basis of Exp. 3 as well.
The Relationship Between Proportion of Visits to, and Intake From, H Feeders
Figure 4 shows the observed average diet choice per meal in relation to the proportion of visits to feeders supplying H in the meal. The graphs also show the long-term average diet choice. If animals attempted to regulate diet choice within a meal by adjusting their intake per visit, then the observations would have clustered around this long-term average. It is evident that this did not happen in any of the experiments. Instead, the observations are clustered around the expectations for animals that do not regulate diet choice on a meal basis. Indeed, regression analysis confirmed that differences between observed and expected diet choice were not affected by the proportion of visits to H feeders in the meal (P > 0.6, 0.8, 0.5, for Exp. 1, Exp. 2, and Exp. 3, respectively). Therefore, the analysis provides no evidence that cows regulate diet choice within a meal. Thus, we must reject hypothesis 5.
|
|
General Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
There usually is variation in diet selection between individuals (Tolkamp and Kyriazakis, 1997; Atwood et al., 2001). For analyses using probability theory, however, large data sets are essential in order to produce reliable results (i.e., pooling of individual data is frequently required). In theory, pooling of diet choice information obtained with individuals could lead to erroneous conclusions. For instance, some cows could have often attempted to select a diet within a meal that was similar to the long-term average diet choice whereas others could have typically eaten meals consisting of one feed type only. Such different feeding strategies would result in non-normal frequency distributions of individual SD’s of meal average proportion of visits to feeders supplying H. In such cases, the pooling of data obtained with individuals could have led to inappropriate rejection of our hypotheses. However, in all our experiments these distributions were normal, which shows that cows did not have greatly differing behavioral strategies. We are, therefore, confident that our results are not affected by data pooling across individuals.
Analyses of diet choice during short-term feeding behavior may also inform us of the relevant time scale over which diet choice is regulated and nutrient supply is synchronized. Recently, there has been a lot of interest in the synchronization of energy and protein supply (Sinclair et al., 1993; Forbes and Shariatmadari, 1994; Witt et al., 1999b). Since animals with access to two feeds can select when to eat and which feed to consume, such analyses may reveal how important the animal perceives synchronization to be. In addition, better knowledge of the time scale of diet choice may enable us to deduce likely mechanisms of nutrient monitoring, and hence the way diets are selected.
Our analyses of the short-term feeding behavior of cows that do select a consistent nonrandom diet over prolonged periods of time provided no evidence that animals attempted to reach a diet composition within a meal that is consistent with their longer-term average preferences. This might be related to the nutritional dimension of the feeds that cows monitor, and therefore use to select a diet. From previous analyses, it was concluded that cows very likely select for eRDP rather than for MP (Tolkamp et al., 1998c). In the experiments we analyzed here, the concentration of eRDP differed between H2 and L2, and between H3 and L3, but was similar for H1 and L1. This coincides with when cows did and did not select a nonrandom diet. Therefore, although the feeds in Exp. 1 differed in a nutrient dimension (Table 1), this did not affect the cows’ feeding behavior. However, the cows in Exp. 2 and 3 may have selected a diet based on the eRDP supply of the feeds.
Effective rumen-degradable protein is a measure of the protein that is available for microbial degradation (McDonald et al., 2001). Such degradation results in ammonia release in the rumen (Parker et al., 1995). Different protein sources are degraded by microbes at different rates (Parker et al., 1995), resulting in the ammonia in rumen fluid peaking at around 1 to 6 h after feeding (Church, 1969; Al-Rabbat et al., 1970; Nicholson et al., 1992). Therefore, regulating diet choice by monitoring differences in eRDP between feeds may not be possible within the meal. For this to occur, feedback from the feed consumed in meals, which are typically 30 to 50 min in duration (Table 5), may be insufficient to allow the diet composition to be altered within such a short time frame. However, animals can learn the nutritional consequences of feeding behavior (Collier and Johnson, 1990; Forbes, 1995; Provenza, 1995). Therefore, animals may regulate their protein intake within a meal to ensure synchronization of the protein to energy ratio in the subsequent intermeal interval (i.e., the next few hours). However, we found no evidence that animals attempt to synchronize their diet over this time span.
In contrast to monogastrics, when ruminants eat a meal, they add feed to a rumen that contains the remnants of previous meals. The immediate metabolic consequences of the addition of a single meal may therefore be affected by the animal’s previous nutritional history. Therefore, the relevant time period of regular diet choice may be greater than a single meal. Indeed, Tolkamp and Kyriazakis (1997) found that some cows selected a lower-than-average proportion of H feed during peak (i.e., maximum) competition, compared with nonpeak periods. This is in agreement with Harb et al. (1985), who found that subordinate cows altered their feeding patterns such that they achieved a daily intake similar to dominant cows. Subordinate cows may therefore have been compensating at nonpeak times for a low protein intake during the peak periods, when access to H feeders may have been limited. The differences in intake patterns between animals, as observed by Tolkamp and Kyriazakis (1997), were likely due to differences in cow pressure per feeder between H and L feeders. Changes in cow pressure can greatly influence the feeding behavior of cows (Elizalde, 1993; Elizalde and Mayne, 1993; Tolkamp et al., 2000). This situation was similar to that of Exp. 3, where cow pressure per feeder was higher at H feeders than at L feeders. Therefore attempts to compensate for the effects of cow pressure might explain the differences in meal composition seen between Exp. 2 and Exp. 3. Subordinate cows in Exp. 3 may have had limited access to H feeders during peak intake periods. They may have compensated for this at nonpeak periods by eating meals consisting predominantly of H. This could then have led to the elevated level of single-feed meals seen in Figure 3f. This suggests that cows may have attempted to regulate diet choice during longer time spans (e.g., within a day).
Other evidence, however, suggests that the relevant time span for obtaining an appropriate mixture of nutrients may be longer than a day. For instance, cows that were given alternate access to high- and low-protein feed for 3 d at a time did not show any decline in intake or milk yield, while such a decline was evident after 1 wk for cows with access to low-protein feed only (Tolkamp and Kyriazakis, 1997). This suggests that lactating cows can tolerate an asynchronous diet for several days, but not for a week. Work with much smaller animal species (e.g., chickens) strongly suggests that animals can tolerate asynchronous supplies of energy and protein for periods of more than 1 d (Forbes and Shariatmadari, 1996). It must be noted that experiments such as these provide evidence of an animal’s ability to tolerate nutritional imbalances (Bigelow and Houpt, 1988). However, these do not suggest the time period over which animals regulate diet composition when they are given free choice. However, if such small and metabolically more intensive species (Kyriazakis et al., 1999) can tolerate a diet that is far from their long-term diet choice for such periods of time, then the conclusion that for cows the relevant time span exceeds a meal may not be surprising.
Although observations on the effects of synchronicity of nutrient supply are not all consistent (i.e., Nia et al., 1995; Shabi et al., 1998; Kim et al., 1999b), the conclusion of a review (Chamberlain and Choung, 1995) of such work for dairy cows concludes that animals are very flexible and can overcome considerable time spans with an asynchrony of nutrient supply. Our work seems to agree with this conclusion, as we found no evidence that cows attempted to select a diet similar to the long-term average within a meal. Ruminants have evolved to graze in heterogeneous environments, therefore they may be expected to be adapted to some asynchrony in the supply of nutrients. Perhaps an asynchronous nutrient supply can be tolerated for periods up to a few days. Our present analysis does not suggest what the most relevant time scale is, except that is must be longer than a meal. In addition, time may not be the dimension over which animals regulate diet choice. Other dimensions that animals may use include monitoring the deviation of nutrient supply from a desired level (i.e., some measure of nutrient intake), or the number of randomly composed meals that will be tolerated before the diet composition is corrected back to within physiologically determined margins. Therefore, as animals are quite flexible, they might not react to an asynchronous nutrient supply as long as the selected diet has a protein content within certain physiological margins.
We consider the type of data presented in this study as potentially suitable for performing analyses in order to answer such questions. For instance, the effects of the diet composition during the preceding meal, or during a longer given time span, on the diet selected during the subsequent meal can be analyzed. Therefore, large data sets of choice-fed animals allow further investigation of the relationships between short-term meal composition and long-term diet selection. This may provide insight into the relevance for animals of nutrient supply synchronization.
|
Implications |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
|
Appendix A |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
and three- (b) population models (for full description see Yeates et al., 2001) were fitted to the interval lengths between visits to feeders, for individual cows. These models are appropriate for cows that do (Model b) or do not (Model a) drink during meals (Yeates et al., 2001). Model a consists of a Gaussian and a Weibull (Yeates et al., 2001) distribution, whereas Model b consists of two Gaussians and a one Weibull distribution. These models have the following probability density functions (pdf; Everitt, 1998):
 | (a) |
 | (b) |
wheret = loge(interval length in seconds); 
= standard deviations of the Gaussian distribution; µ = mean (median) of the Gaussian distribution;c = shape parameter of the Weibull distribution; 
= scale parameter of the Weibull distribution; p = proportion of intervals in first population; q = proportion of intervals in the second population; and subscripts 1, 2, and 3 indicate the first, second, and third populations, respectively.
Individual meal criteria were estimated from the parameters of either Model a or b as the point at which the pdf of the final two populations crossed. Which of the two models was most appropriate for a given cow was decided after examination of the statistical and visual fit of the models, as described by Yeates et al. (2001).
|
Appendix B |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
Appendix Table
| ||||||||||||||||||
For instance, the probability of a visit to an H feeder is p = 0.7 only for first visit in a meal, but the probabilities are pu = 0.5425 and pu + (1 - u) = 0.7675 if the current visit follows a visit to an L or an H feeder, respectively. Similarly, the probability of a visit to an L feeder is (1 - p) = 0.3 only for first visit in a meal, but the probabilities are (1 - p)u = 0.2325 and (1 - p)u + (1 - u) = 0.4575, if the current visit follows a visit to an H or an L feeder, respectively. This set of probabilities allows the calculation of the likelihood of any combination of visits in a bout. For instance, the probability that a bout of three visits consists first of visits to two L feeders and subsequently to an H feeder is 0.3 x 0.4575 x 0.5425 = 0.0745. This methodology was used to derive appropriate probabilities for each experiment. These probabilities were then used to calculate the corrected frequency distribution of bout composition.
|
Footnotes |
|---|
Received for publication February 21, 2002. Accepted for publication June 25, 2002.
|
Literature Cited |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionGeneral DiscussionImplicationsAppendix AAppendix BLiterature Cited |
|---|
