Systematic analysis of feeding behaviors and their effects on feed efficiency in Pekin ducks

Birds and housing

All ducks were reared in the same house with the same feeding and lighting program. Each house hold 100 feeding machines. Ten machines were grouped together as one pen with barrier for duck management. We randomly selected 3 batches of 3 wk. Male Pekin ducks obtained from the Beijing Golden Star Company (358 birds in total). Each of the 3 batches (batch one: n = 126; batch two: n = 119; batch three: n = 113) was measured from 3 wk of age until 6 wk of age. The intervals between batches was 8 d and measurement was same for each batch. Ducks in each batch were fed ad libitum for 25 d in a single pen. The pen area was 96 m² and contained 10 automatic feeders and a water tank. Illumination was 10 lx for 24 h during the entire period. The pen design is illustrated in Additional file 1: Figure S1.

Feeding behavior data acquisition

We designed duck feeding machine which use RFID as sensor to record each duck. The design of duck feeding machine is similar with Bley and Bessei [12]. Feeding behaviors (entering time, exiting time, feed consumption) and live body were recorded for each feeding for each duck from 19 d to 42 d. Only one duck can enter the machine for each feeding event adjusting barrier according to duck body size. The machine has been extensively used in the duck breeding practices [14]. Each feeding machine was spaced at 0.5 m intervals, and the feeder ID was set from 1 to 10 in linear. The first machine was nearest the gate and the last one was against the barrier. Tiny electronic tags were permanently fastened to the upper neck of each duck until slaughtering, which couldn’t interfere with animal activity. Upon entering a feeder, each duck’s electronic label ID was recognized by the machine. The date and time at beginning and end of each visit was recorded. The central server collected the raw data from each feeder and calculated feed intake and meal duration. At the end of feeding, the ducks with very small body size, without measurement or unusual feeding behavior were removed before analysis of feeding behavior. Any visit that could not be assigned to a specific duck was removed from the data set before analysis. During the whole measuring period, 0.001% of missing record events were observed and these records were removed. Only the ducks with complete records were used for further analysis. Additionally, the records that individual weight were far outweighed other records on the same day were removed, which may be caused by interference from other ducks.

Traits measurement

Based on the distribution of RFI, three subsets were selected for later analysis. Ducks in the bottom 10% of RFI were defined as the low-RFI level, from 45% to 55% were defined as the middle-RFI level and the top 10% RFI individuals were selected as the high-RFI group. Each subgroup had 33 ducks, and these ducks were randomly chosen from the combined data.

Every visit was treated as a valid record if feed intake and meal duration were greater than zero. The intervals between visits were calculated to estimate the meal criterion (Additional file 2: Figure S2). The meal criterion was estimated using the method 1 developed by Howie et al. [15]. For each duck, visits were defined as part of a meal if the interval between visits was lower than the meal criterion. Feeding behavior traits were calculated daily and over the test period for each duck: feed intake, feeding rates, average intervals between meals, number of visits and number of meals. The distribution of records was also analyzed to investigate the feeding tendency of ducks.

Because feeding patterns of animals are affected by the environment and their growth needs, the data has been divided into three stages, based on previous analysis. The first three days were considered as a period of adaptation, the fourth day to the tenth day after the commencement of the study was defined as the pre-growth period, and the eleventh to twentieth day was classified as the growth period.

Data analysis

In formation, level _j was the RFI levels, and the batch _i was the batch effect. E _ij was the residual of the model. Additionally interaction effect between batch and RFI level was not significant in the data.

Descriptive statistics and meal interval estimation

Phenotypic correlation

Comparison among different RFI level

Changes in feeding behavior over time

To further understand the relationship between RFI and feeding behavior, changes in feeding behavior over time were investigated and are illustrated in Fig. 1. During the test period, average DFI increased from 0.07 kg to 0.35 kg (Fig. 1a), and the effect of RFI level was significant from the 3^rd day to the end of test (P < 0.01). The average number of daily visits and meals increased constantly in the early days, then gradually diminished, and finally stabilized (Fig. 1b, c). Obvious differences in daily visits were found among different RFI level from 4 wk. To the end of test, and these differences magnified with age; however, there was no difference in daily meal number. The daily feeding rate increased over the entire test period (Fig. 1d, e); in contrast, meal duration gradually decreased with age. Moreover, between-meal pauses was observed to increase until the 11^th day, then was maintained at a relative stable value, and finally, in the between-meal pauses of low RFI ducks were slightly longer than high RFI animals.

Open image in new window

Feeding tendency

Fig. 2a summarizes the records of all the feeders. The machine with the highest number of visits is the eighth feeder which has 15,102 visits, and the feeder with the lowest records was the first one, with 4422 visit records. Ducks in the top 10% RFI level recorded 14% of the total records, and visit records of the lowest 10% RFI animals were 11% of the total. To further understand the feeding tendencies of ducks, the distance to the gate was used to define meal positions (Fig. 2b). The results show that animals were concentrated in middle positions (4 to 5 m from gate) for the first two days. After an adaption time, more ducks preferred to eat at 7 ~ 8 m from the gate; however, high RFI ducks frequently visited feeders at positions of 3 to 4 m from the gate, in which frequency was 1.70 per day, higher than 1.57 per day for low RFI animals. The feeder machines visited by high RFI ducks was closer to gate than for low and middle RFI ducks.

Open image in new window

Distribution of daily feeding time

To investigate the daily feeding time of ducks, the feeding records at different time points every day are summarized in Fig. 3. The results showed that duck feeding was mainly concentrated at noon and in the afternoon during the adaptation period, and reached the highest peak between 1200 h at noon and 1600 h in the afternoon. In the pre-growth period, feeding activity was observed during the entire 24 h period, and the number of visits increased significantly from 2300 h at night to 0500 h in the early morning. Compared to the first 10 d, the number of feeding visits was reduced during the growth period, but the tendency (frequent activities from 2300 h at night to 0700 h in the early morning) was similar to the pre-growth period, and was concentrated from midnight to the following morning.

Open image in new window

Meal determination

Feeding behavior consists of different feeding activities, which are referred to feeder visits. Bley and Bessei [12] found that Pekin ducks exhibited an average of 29 visits per day at 4 wk of age, decreasing to fewer than 15 visits per day at 7 wk of age. Basso et al. [11] obtained a similar trend with 27 visits per day at 3 wk, rapidly decreasing to 15 visits, and then stabilizing. In this study, we found that the number of visits was gradually reduced from an average 25 visits per day at 3 wk of age to 14 visits at 4 wk, and then was stable at about 13 visits until the end of the study at 7 wk (Fig. 1b).

Our results are consistent with the two studies above. Actually a meal is not always equal to one visit, but rather contained a few visits and short relaxation pauses called within-meal pauses for animal relaxing or drinking; frequently, meals are separated by non-feeding intervals called between-meal pauses which are much longer than within-meal pause [11]. Therefore, a meal criterion is needed for grouping different feeding events into defined meals [15, 16]. Howie et al. [17] obtained a meal criterion of 1725 s for Pekin ducks by fitting a truncated log-normal model, and Basso et al. [11] found the meal criterion of 1790 s for Mule ducks using same method; however, Drouilhet et al. [18] estimated the meal criterion as 2208 s in their study in which the feeder entrance had a door. In our study, the log distribution of visit intervals is similar to the trends reported by Howie et al. [15] and Basso et al. [11] (Additional file 2: Figure S2); hence, we used Howie’s method for our data set and calculated a meal criterion of 1,132 s for Pekin ducks. This value is, lower than the meal criterion reported by others, which is likely due to the differences in feeder construction and in duck population. We found that meals consisted on average of 1.7 visits and that the number of meal per day had a similar trend in three RFI-level. Basso et al. [11] and Drouilhet et al. [18] found that the number of meals tended to decrease with age; however, between-meal pauses tended to increase. In boilers, Howie et al. [15] also found same trend. Our results showed that the number of meals tended to decrease with age (from 12 meal/d to 7 meal/d) after feeder adaption time and the between-meal pauses tended to increase (from 60 min/d to 2 h/d more). The results are in agreement with previous studies.

Feeding tendency at different RFI levels

Feeding tendency is possibly affected by environmental factors, especially under a wide range of farming conditions. These changes may be due to management mechanisms or system structures. For example, sward structure and herbage quality affected cattle’s grazing behavior [19]. Another study showed that the average time cattle spent lying increased with increasing number of concentrate feeders [20]. Funston et al. [21] found different group composition affected cattle feeding tendencies. In ducks, Drouilhet et al. [18] used a feeder with a door closing behind the animals, and reported that this affected the visit intervals. In our study, most ducks tended to eat at the inside feeders which are far away from the gate. However, the innermost feeder was not the best feeding position, since the feeding machine slightly near the center have the highest recorded number of visits (Fig. 2a). This suggests that ducks spontaneously move away from the gate where animals are susceptible to environmental interference. In contrast, the innermost machine is limited by proximity to the fence and this could greatly reduce activity space for ducks. Our results also show that high RFI ducks tend to eat near the gate, where high RFI ducks were more susceptible to outside interface (Fig. 2b), and this always occurs during the entire period. Some studies indicate that selection for feed efficiency might reduce behavioral reactivity [22], and evidence reveals that animals with better feed efficiency have lower fearfulness. Genetic research with mammals and poultry showed that low RFI animals were less active [23, 24]. Drouilhet, et al. [18] reported that selection for low RFI produces mule ducks that are slightly better adapted to human activity, in agreement with a previous study [25]. Compare to their results, our study suggests that high RFI individuals have lower adaptive ability to the new feeding environment. However, there is not enough evidence to determine the clear reason for this effect. Moreover, we found that feeding time points for adaptation during the first 3 d were different than at other time periods (Fig. 3). Ducks mainly were eating at noon and early afternoon during first 3 d while adapting to the feeders. After adaptation, ducks returned to middle of the night eating habits. These behaviors may be improved by feeder structure modification and selection for adaption response.

Feeding behavior at different RFI levels

The relationship between feeding behavior and feeding efficiency depends on population and species, and there are obvious differences in feeding behavior between mammals and poultry selected for feed efficiency. Studies with cattle reported that high RFI cattle consume more feed, but with similar feeding rates, compared to low RFI animals [26, 27]. For pigs, Young et al. [28] found that low RFI animals had shorter feeding times with a lower DFI compared to the high RFI pigs, and that the feeding rates of low RFI animals were significantly faster than the control group. A similar result was obtained in a hen study; Braastad and Katle [23] found that low RFI birds had shorter feeding times than high RFI birds. Drouilhet et al. [18] found that low RFI mule ducks had significantly lower DFI than high RFI ducks, but no differences were significant between high and low RFI lines for other feeding behavior traits. In this study, we found there was a significant phenotypic correlation between feed efficiency and feeding behaviors except for BMP (Table 2). It indicated that the feeding frequency has a limited impact on feed efficiency, total feeding time and daily feed intake are more critical. The Low RFI ducks spent less feeding time and had a lower daily feed intake than High RFI ducks. However these correlative relationships were not observed in different RFI-level sub-populations (Table 4). Therefore, selection for RFI may change the association between feeding behavior and other traits.

Our results were different with those of Drouilhet, et al. [18]; this discrepancy could be caused by feeder structure and selection of meal criterion. Additionally, when compared day by day, ducks in the three different RFI levels had a similar number of daily visits at 4 wk of age; however, after 10 d of the test period, low RFI ducks had a significantly lower visit number than high RFI ducks, and these differences were magnified with age (Fig. 1b). Over the entire test period, these changes were too microscopic to achieve statistical significance. Our results gave a possibility that selection for RFI might change the unknown mechanisms that underlie control of feeding behavior, but more genetic evidences were needed to support the hypothesis. This showed that selection for feeding behavior is a valuable approach for trait improvement.