1. Introduction
Worldwide, cattle contribute approximately 10% of anthropogenic greenhouse gases [
1]. In Europe, 90% of ammonia emissions come from agriculture, and a considerable proportion come from cattle farming [
2]. Ammonia is formed when faeces and urine mix [
3]. Thus, technical measures have been developed to separate faeces and urine [
4], but excreta is still spread over a large area by cattle, resulting in a large emission area, which makes technical separation difficult. An innovative alternative approach would be to reduce the emission area by training cattle to use a latrine. This would also facilitate the separation of excreta and lead to improvements in animal wellbeing through better hoof and udder health, as cows would be less exposed to their excreta [
5,
6,
7,
8].
Reliable latrine use would require that cattle, like other species, learn to control a range of voluntary and reflex responses associated with toileting [
9,
10,
11]. These include the ability to suppress impending voiding, a reflexive-like behaviour [
9,
11], then move to the latrine, a voluntary behaviour [
12,
13], and finally reinitiate voiding at the latrine. Dirksen et al. [
14] reviewed the literature on the toilet training of humans and other animals and reported that cattle have the neurophysiological and cognitive capabilities for toilet training, concluding that toilet training of cattle should be possible by using operant training methods [
14]. It is well known that mammals such as dogs, cats [
15] and humans [
11] are able to learn toileting behaviour.
For latrine (toilet) training, a sequence of new behaviours needs to be learned. An established method for training sequences is to reward a series of connected responses (behavioural chains). Chaining is a method that is often used in animal training to break down complex behavioural sequences into smaller units that can be trained step-wise using forward or backward conditioning [
16,
17,
18]. In this process, each learned behavioural unit of the entire sequence serves as the conditioned reinforcer for the preceding behaviours and as the discriminative stimulus for the next response [
19,
20]. Forward chaining starts with training the first behaviour in a sequence, after which the next behavioural element is added after the first has been mastered, and so on. In backward chaining, training starts with the last behaviour of the sequence, after which the penultimate response is added, and so on [
16,
17,
21]. The available published evidence shows no consistent differences between forward and backward chaining regarding the ease of training animals [
17,
18,
22]. Slocum and Tiger [
17] concluded that considering the high variability between studies, forward and backward chaining show similar effectiveness in teaching new behavioural chains. However, it was acknowledged that there might be differences depending on the particular task being trained. In toilet training in children, both forward and backward chaining have been used. In backward chaining, children are placed on the toilet before showing signs of voiding [
23], whereas in forward chaining, voiding is interrupted by verbal instruction, after which children are moved to the toilet [
24]. Toilet training can be learned via both procedures but appears to be learned faster with forward chaining [
24]. Previous studies have shown that cattle are able to learn some elements of the toileting process [
25,
26], but the training of a full toileting sequence has not yet been achieved.
Several studies have found that learning ability is faster in younger animals than in older animals [
27,
28]. Other advantages of using younger animals include the ease of handling smaller animals and the possibility of utilizing the learned behaviour over their entire lifetime. For these reasons, calves were used in our study. Cattle defecate and urinate relatively infrequently (less than once per hour) [
29,
30,
31], which makes training difficult. To train efficiently, a higher frequency of voiding is needed. It is easy to increase the frequency of urination with diuretics [
25], hence urination was used for our proof-of-concept training procedure.
For practical and ethical reasons, it is not recommended to use diuretics for prolonged periods. Thus, to reduce the use of diuretics, the voluntary (non-voiding) behavioural elements of the toileting sequence were trained first. Voluntary responses were trained with a combination of backward and forward chaining (elements indicated in black in
Figure 1). The voluntary elements included moving to the latrine in response to an externally applied signal, waiting in front of the reward dispenser (signalled by a flashing light), reward delivery and consumption, and exiting from the latrine. Thus, we sought to bring each element under stimulus control [
32], e.g., movement to the latrine was triggered by a vibration signal and ultimately rewarded by food presentation. Subsequently, reflexive voiding responses were introduced into the chain in a manner akin to total task training [
33] (elements indicated in green in
Figure 1). The training of the reflexive responses involved pairing the vibration signal with the initiation of urination. An ideal outcome for the study would be initiation of movement to the latrine, firstly, in response to the externally applied stimulus (vibration) and subsequently to other behaviours occurring at the time of initiation of urination (e.g., internal stimuli associated with a full bladder). There is ample evidence in the literature from rats, pigeons and humans that initiation of behaviours can be readily transferred from one external stimulus to a novel one when the new signal occurs shortly before the first-learned stimulus (e.g., [
34,
35,
36]). To our knowledge, transfer of control from an exteroceptive stimulus to an internal reflexive excretory signal in cattle has not previously been studied. Interestingly, evidence for the transfer of control between stimuli is provided by two different types of behavioural responses. In one case, experimental subjects orient towards the novel stimulus before the first-learned signal is presented [
35] or they may orientate towards the reward location [
37]. Attention to the reward location (sometimes called goal tracking) is more likely when the novel stimulus occurs in close spatial and temporal proximity to reward delivery [
37,
38]. When training cattle to void, Whistance et al. [
26] reported that the animals orientated towards the person delivering the reward, an example of goal tracking. It was anticipated that the triggering of movement to the latrine by urination intention behaviours (and/or vibration signals) would interrupt urination and be followed by latrine entry and the reinitiation of urination (
Figure 1). Thus, we expected that reflexive and operant responses could be combined to train toileting in cattle. The aim of this study was to establish a procedure for toilet training cattle using chaining methods, first by conditioning the key voluntary behaviours in the chain and then by incorporating the reflexive responses.
2. Materials and Methods
The study took place at the Experimental Facility for Cattle at the Leibniz Institute for Farm Animal Biology (FBN) in Dummerstorf, Germany. All procedures involving animal handling and treatment were approved by the Committee for Animal Use and Care of the Ministry of Agriculture of Mecklenburg-Western Pomerania, Germany (file reference: 7221.3-1.1-002/18).
2.1. Animals and Housing
Ten female German Holstein calves were purchased at an age of 30 d (range: 14–40 d) from a local commercial dairy farm, whereupon they were housed as a group in a straw-bedded pen (6.9 m × 5.7 m) and fed milk on-demand from an automatic feeder up to an age 71 d. From the 12th day after arrival at the FBN, each calf was halter trained to make later handling easier. The mean age at the start of latrine training was 91 d.
2.2. Experimental Facility
The training took place in a 10 × 10 m testing arena (
Figure 2). The floor of the arena was covered with rubber mats. Pens and gates were constructed from steel piping (50 mm diameter). Visual contact between calves was possible except when in the designated latrine. The latrine was shielded by a solid wooden wall. The arena comprised two identical experimental training areas (one on the left, the second on the right,
Figure 2). Each side contained a waiting area, a start box, an alley comprising four segments and a latrine with a bowl into which liquid rewards could be delivered (SUEVIA Mod. 20, SUEVIA HAIGES GmbH, Kirchheim/Neckar, Germany) mounted at a height of 0.6 m. Distinctive visual stimuli indicating the latrine/reward area were provided by black and yellow tape affixed to the pipework neighbouring the reward bowl. The black and yellow tape was chosen because cattle can distinguish yellow particularly well from various shades of grey [
39]. The waiting area (2 m × 4 m) was connected to the start box (0.9 × 2 m), where animals were prepared prior to training (feeding, administration of diuretic). The start box was connected to the latrine (1.5 × 1 m + 2 × 2 m) via segments 3 (1.5 × 3.5 m), 2 (1.5 × 5 m) and 1 (1.5 × 2 m). Segment 4 (1.5 × 3.5 m/6 m) was also used in the training procedure. The segments were separated by manually operated gates. A remotely controlled pneumatic gate separated the latrine from segment 1 (left and right sides were operated simultaneously but could be closed individually).
The reward comprised 140 mL of a glucose-based electrolyte drink for calves referred to as Milkilyt (Milkivit, Trouw Nutrition Deutschland GmbH, Burgheim, Germany), which was mixed with water according to the manufacturer’s recommendation (50 g electrolyte and 1 L warm water) and molasses (50 mL per 1 L electrolyte mixture). The reward was continuously mixed and at a constant temperature (40 °C) in a bain-marie situated on the outside of the latrine. The reward was dispensed into the drinking bowl by a pump (which also emitted an audible signal). A blue flashing light (CO BL 70 2F, Compro® Electronic GmbH, Vechta, Germany) and an audible noise generator (RoLP, Fulleon Ltd, Cwmbran, South Wales, UK) were installed directly above the drinking bowl. Cattle can hear sounds between a range of 23–37 kHz [
40]. The auditory signal used in this experiment had a frequency of 970 Hz. Furthermore, cattle have been trained to approach a feed source after an auditory signal [
41]. It has been shown that cattle can distinguish yellow, pink, red, violet, blue, and green from shades of grey [
39]. Thus, it was anticipated that the calves would readily react to the acoustic and visual stimuli and approach the reward bowl, which was subsequently confirmed. The reward dispensers and the visual and auditory signals on both sides of the arena were activated remotely and simultaneously by the experimenter.
From Phase 1.2 (see chapter 2.3.1.) onwards, the animals were fitted with remote-controlled vibration collars (Dogwell DW998N, TZLong Store by Amazon, Guagzhou City, China). All remote-controlled instruments were activated from a single handheld device.
The training sessions were recorded using four video cameras (AXIS M1124, Axis Communications AB, Lund, Sweden) positioned at the front and back on each side of the arena (
Figure 2) and a microphone (Sennheiser MKE600; Sennheiser Electronic GmbH & Co., KG, Wedemark, Germany) positioned on the ceiling in the middle of the arena. The video and audio recordings were stored directly on a computer using Media Recorder 4.0 (Noldus Information Technology, Wageningen, The Netherlands).
2.3. Experimental Procedure
The calves were assigned to five training pairs matched by birth date. Within each pair, animals were assigned randomly to either the training treatment (test) or the yoked-control treatment (controls). Animals within a pair were identified by identical halter colours (blue, black, turquoise, red, and brown). Test calves were assigned randomly to either the left or right side of the training arena, and the controls were assigned to the side not occupied by the test animals. Thus, blue, turquoise and brown test calves were trained on the left and black and red on the right.
At the start of training, the calves assigned to each side were habituated as a group to the training area, and the liquid food reward for one session with a duration of one hour. During this session, the animals had access to the latrine and segments 1–3. A portion of the reward was already in the bowl, and it was refilled whenever it was fully consumed. Calves that did not drink were directed to the bowl using the halter and encouraged to drink by guiding the muzzle towards the reward. During the various phases of training, the calves had access to different areas of the testing arena (see below).
The training was divided into three main parts (training voluntary behaviours inside and outside of the latrine, and training reflexive responses). Each part consisted of two or more phases. The animals were trained once daily (called a session). During a session there were several trials, each beginning with the delivery of an exteroceptive stimulus (audible/vibration signal) and ending with reward delivery, or, in cases where there was no reaction to the signal, 10 s after the signal had been activated.
On training days (Monday to Friday), the calves were kept together in their respective waiting areas. The test and control calves were tested in a random starting order, except that the same pairs were not tested first or last on two consecutive training days. During all training phases, the control calves received all signals (e.g., vibration, sound, reward) at the same time as the test calves that were unrelated to their own behaviour. During the study, the experimenter was positioned outside of the training arena on the side occupied by the test calf to remotely activate the pneumatic gate, the various stimuli and the reward system with the handheld device. A second person was located on the other side, so the conditions were the same for both calves in a pair.
2.3.1. Training Voluntary Behavioural Responses in the Latrine
The test and the control calves were confined to their respective latrine areas. Each pair of calves was trained once a day in a 30 min session, which was reduced to a 20 min session from the second week onward, as the animals’ reactions to the various stimuli decreased significantly after 20 min. The arenas were hosed down with fresh water after training each pair.
Part 1 started with magazine training (Phase 1.1) to establish an association between an audible signal and the reward (S6,
Figure 1). The audible signal emanated from the reward location to draw the calf’s attention to the latrine, regardless of the direction in which it was looking. When the test calf was oriented away from the bowl, the audible signal was sounded five times in a row at a frequency of 970 Hz and a rate of 0.8 Hz (250 ms on/1 s off), followed by reward delivery. A trial was correct if the calf moved to the reward bowl after the activation of a stimulus. The next trial was initiated only after the test calf had oriented away from the bowl after reward consumption. The number of trials per training session for each pair was determined by the behaviour of the test calf. In the second training step (Phase 1.2), the audible signal was replaced by a vibration signal on a collar, as a vibration signal could be delivered without influencing the calves not currently being trained (S3,
Figure 1). The transition to the vibration signal was achieved using the stimulus-fading procedure [
34], in which the vibration signal was initially presented 1 s prior to the audible signal, followed by gradually increasing the delay to the audible signal, and finally presenting the vibration signal alone (Phase 1.3). After five correct trials in a row, i.e., moving towards the reward bowl within 2 s of the audible signal (Phase 1.1), after the vibration signal but before the audible signal (Phase 1.2) or within a maximum of three activations of the vibration signal (each of 1.5 s duration at intervals of approximately 1.5 s, Phase 1.3) the learning criterion for each phase was attained.
2.3.2. Training Voluntary Behavioural Responses Outside of the Latrine
The goal of the next step (Phase 2.1) was to train the calves to leave the latrine after each reward delivery so that it would be possible to subsequently train latrine entry for urination. In detail, we trained the calves to leave the latrine (R1,
Figure 1) after reward consumption, and to re-approach (R3,
Figure 1) and re-enter the latrine (R4,
Figure 1) after a further vibration signal was given (S3,
Figure 1). This phase was similar to Phase 1.3, but segments 1 to 3 of the arena were also available (
Figure 2). The vibration signal was given when the calf stopped walking in segments 1 to 3, ideally with its body oriented in the direction of the latrine to facilitate immediate movement towards it. The learning criterion was five consecutive correct responses (approach towards and entry into the latrine) to the activation of the vibration signal. Test calves that did not meet this criterion were moved to the next phase after eight sessions.
In Part 3 (training of urination), the calves were expected to learn to wait in the latrine until the completion of urination before receiving any reward. The duration of urination in cattle is usually less than 30 s [
42,
43,
44]. Thus, the purpose of this next step (Phase 2.2) was to prepare the calves for a delay between latrine entry and reward delivery. To provide a signal that a reward was imminent, a blue flashing light above the dispenser (S5,
Figure 1) was activated for the duration of the delay. The delay duration was increased stepwise from 2 s to 30 s (latencies of 2, 4, 7, 11, 15, 20, 25, and 30 over successive days). The calf was required to stand calmly in front of the bowl during the delay; otherwise, the reward was not delivered. All four segments and the latrine were available to the calves in this Phase.
2.3.3. Incorporation of Reflexive Responses into the Chain
The goal of Part 3 was to introduce reflexive responses into the previously trained behavioural chain first in the latrine area only (Phase 3.1) and then in the alley connected to the latrine (Phase 3.2).
Five minutes before the start of a training session, while standing in the start box, the test and control calves were administered 1.3 mL (1.5 mL in the second week) of diuretic (Diuren, Wirtschaftsgenossenschaft deutscher Tierärzte (WDT), Garbsen, Germany), into the jugular vein to increase urination frequency. The training in Part 3 covered periods of two to three days of training with one session per day, interspersed with two to three days without training to prevent undue injury to the jugular vein. In contrast to the previous phases, the number of training sessions in Part 3 was predetermined, as the total number of days on which diuretics could be administered was limited to ten by the Committee for Animal Use and Care. Thus, Phase 3.1 comprised three sessions, and Phase 3.2 comprised seven sessions.
In Phase 3.1, the calves were locked in the latrine, and the vibration signal was activated at the first sign of urination, e.g., raising the tail, spreading the hind legs and/or arching the back. A reward was delivered if i) the calf interrupted urination, approached the reward bowl, and reinitiated urination while waiting in front of the bowl or ii) the calf oriented towards the reward bowl during urination. Otherwise, no reward was delivered.
Waiting in front of the bowl was accompanied by the blue flashing light (waiting signal). It became clear that the reinitiation of urination did not occur regularly within 30 s after approaching the reward bowl. Therefore, this phase was discontinued after three sessions, and the flashing light (see above) was not used in Phase 3.2.
In Phase 3.2, the calves had access to the latrine and segment 1 for the first six training sessions and access to segments 2 and 3 in the final (7th) session. Starting with an ‘intention to urinate’ response by a calf (either in the latrine or the alley), six variants of a behavioural chain were possible. Each variant indicated a different level of training success (as shown in
Figure 3). If the calf was in the alley (segment 1–3) when it showed the intention to urinate, then sequences (Seq) 1–5 were possible. Situations in which a calf was already in the latrine at the beginning of a urination event were denoted by Seq6+. In all sequences, when calves entered the latrine after receiving a vibration signal, they were confined for up to 5 min (by closure of the pneumatic gate) to increase the probability of urination in the latrine. The control calves were confined in the latrine simultaneously or as soon as they were in the latrine. The behaviour of the control calves was also assigned to one of the six sequences if the appropriate criteria were met.
Seq1+ and Seq3+ were deemed ‘correct’ responses, as urination in the alley was totally withheld (Seq1+) or interrupted (Seq3+), followed by walking to the latrine, the reinitiation of urination in the latrine and the delivery of a reward.
Seq2-, Seq4-, and Seq5- were not rewarded. In Seq2-, urination was withheld in response to the vibration signal. In Seq4-, urination was stopped either because of interruption or because the bladder was emptied. Both sequences included movement into the latrine but not the reinitiation of urination and were thus designated as an ‘incomplete’ response. In Seq5-, the calves did not inhibit or interrupt urination (in response to the vibration signal) or walk to the latrine (thus being designated a ‘false’ response).
In Seq6+, the calves were already situated in the latrine when urination intention behaviours occurred. Thus, the tactile signal was not activated, and completed urination events were rewarded. For Seq6+, however, it was not clear if the animal had previously entered the latrine in anticipation of urination or if it was situated in the latrine for other reasons (e.g., to be in close proximity to the reward site). Thus, it was unclear whether the responses in Seq6+ demonstrated control of urination or not (thus, being designated an ‘ambiguous’ response).
2.4. Data Analysis
All videos were analysed by a single person using The Observer® XT13 software (Noldus Information Technology, Wageningen, The Netherlands). Due to technical problems, there were no videos available for the black pair for session 1 of Phase 3.1 and session 5 of Phase 3.2.
The following events were coded for the test and control calves: (a) start of the trial with the trial number and location of the calf, (b) end of the trial with the trial number and the result (correct/incorrect), (c) reward delivery (yes/no), (d) duration of the flashing light, (e) confinement in latrine. Events (a) and (b) were coded for Parts 1 and 2 only, and event (e) was coded for Part 3 only. The reason for this was that in Part 3, no trials were defined that started with an exteroceptive stimulus, instead trials started with the intention to urinate. Similarly, events (a) and (b) were not coded in Part 3. To take into account the time and duration of confinement in the latrine, event (e) was coded in Part 3 only. The behaviours recorded in Part 3 are shown in an ethogram (
Table 1). Intraobserver reliability was examined by re-coding 25% of the videos from Phase 3.2 by the original observer [
45,
46] because this was the phase including the complete behavioural sequences. Cohen’s kappa was calculated with the routine included in The Observer® XT13 software with
κ = 0.91 for the frequency, with regard to the sequence of events over all observed behaviours. Cohen’s kappa for duration, with regard to the sequence of events over all observed behaviours, was also
κ = 0.91 [
47]. Cohen’s kappa values between 0.81 and 1.00 are regarded as almost perfect [
48].
Due to the small sample size, some of the data are presented descriptively. The calculation of mean values was performed first for each calf and then across all calves in each session or phase.
Differences in the number of correct trials during voluntary behaviour training in Part 1 between the test and control calves were analysed using the Kruskal-Wallis test in Jamovi (Version 1.1.9.0,
https://www.jamovi.org) which is based on R. The duration of urination by the test calves in the different behavioural sequences within or outside the latrine was analysed using the Friedman test in Jamovi (Version 1.1.9.0,
https://www.jamovi.org). Differences in the duration (total and individual events) and frequency of urination between the test and control calves were analysed using one-way ANOVA according to the MIXED procedure in SAS (Version 9.4, SAS Institute Inc., Cary, NC). In the MIXED procedure, the calf was included as a repeated factor, and the treatment (test/control) was the fixed effect.