Random timing of aversive events extends stress responses in fish

Responses to potential stressors are shaped by their predictability. While predictable events may reduce negative effects, the impact of temporal uncertainty remains unclear. This study examined how different degrees of predictability influence stress responses. Nile tilapia (Oreochromis niloticus) were exposed to a confinement stressor under four conditions: predictable (fixed timing), random (variable timing), unpredictable (single unexpected event), and unhandled control. Ventilatory frequency (VF) was used as a non-invasive indicator of physiological activation and measured before and after exposure. All confinement treatments increased VF immediately after exposure. However, response patterns differed. In the predictable condition, VF increased but was not sustained. In contrast, both random and unpredictable conditions produced elevated and prolonged responses, with the strongest effect under complete unpredictability. Partial predictability (random condition), where the stressor was anticipated but not temporally defined, was sufficient to maintain activation. These findings show that temporal uncertainty plays a key role in sustaining stress responses. The results support the concept of emotional plasticity, in which physiological responses are dynamically modulated by environmental predictability, with implications for animal welfare and stress regulation.

The study of motivation, and of the functional systems that enable animals to regulate their interactions with the environment, has long been considered essential for an understanding of welfare and stress in humans and other animal species (Broom, 1981; Broom and Johnson, 1993, 2019). For example, animals have a range of functional systems, and they have to switch between them using brain control of behavioural and physiological mechanisms (Cannon, 1929; Sibly and McFarland, 1974; Dickinson and Balleine, 2002). While welfare is likely to be more positive if they can control their lives, stress and poor welfare result when there is a lack of control and overtaxing of control mechanisms. There are limits to the degree of adaptation to which individuals are capable, and stress is not typified by any one response but is an environmental effect on an individual which may overtax its control systems and result in adverse consequences and eventually reduced fitness (Broom and Johnson, 1993; Broom, 2006; Broom and Johnson, 2019). We consider it conceptually imprecise to use the term stress as a synonym for stimulation or for physiological activation, including adrenal responses, particularly when such responses may ultimately be neutral or beneficial to the individual (Broom and Johnson, 1993; Broom, 2006; Broom and Johnson, 2019). In this paper, we restrict the term stress to conditions that result in adverse effects on the individual, as defined above. Stimuli whose consequences are uncertain will be referred to as potential stressors.

The effects of a wide range of stressors have long been described in vertebrates, but, more recently, the feeling of imminence and the anticipation of a stressor by subjects have been investigated (Koolhaas et al., 2011). Cognitive appraisal of the predictability of a stressor modulates neurobehavioural and physiological stress responses in fish (Cerqueira et al., 2020). However, important aspects remain insufficiently explored, particularly how varying degrees of predictability influence stress responses. Evidence from humans and other animal species indicates that individuals preferentially select conditions that allow anticipation of aversive events (Badia et al., 1979), especially when avoidance is not possible (Bassett and Buchanan-Smith, 2007). Understanding how the degree of predictability shapes stress outcomes is, therefore, critical for refining our interpretation of stress responses. Studies with dogs, sheep, humans, and other species show that unpredictability of disturbing events is aversive (Greiveldinger et al., 2007; Amat et al., 2014; Tanaka and Kawabata, 2021). There can be a substantial preference for predictability, for example, an electric shock was less disturbing when predictable than when presented under an unpredictable regime (Seligman and Meyer, 1970; Badia et al., 1973).

Furthermore, the predictability and controllability of aversive events critically influence both the magnitude and the nature of emotional responses. Predictability of negative or aversive events has been recognised as a central factor in individual susceptibility to pathological anxiety (Seligman and Binik, 1977 apud Grillon, 2002; Zvolensky et al., 2000). The safety-signal hypothesis refers to an individual who can predict threat because it is signalled, while the absence of a signal for threat predicts the absence of danger, that is, safety (Grillon, 2002). However, when aversive events are not signalled, periods of safety are also not signalled, and the organism may be in a state of chronic anxiety (Seligman and Binik, 1977 apud Grillon, 2002).

Predictability depends on previous experience, which enables individuals to identify the nature of the stressor, its location, and the timing of its occurrence. The occurrence of a potential stressor in nature, for example, the appearance of a predator, might be known and predictable, whereas the exact moment it will occur is uncertain. Even though there is knowledge about a range of stressors, little is known about the impact of the degree of predictability on the stress responses. In this study, the impact on Nile tilapia of different degrees of predictability was investigated. Nile tilapia is the most widely farmed fish in the world (FAO, 2020), and the species has been studied in a range of scientific fields. Previous research in Atlantic salmon has shown that the temporal predictability of feeding schedules can influence behavioural interactions and welfare outcomes, with unpredictable schedules associated with greater aggression and fin damage than predictable ones (e.g., Cañon-Jones et al., 2012). Our findings may extend to other species, highlighting the role of stressor predictability in shaping coping strategies. The response to the treatment was measured by a simple non-invasive technique which quantifies the fish’s ventilatory frequency (VF). VF is mediated by sympathetic pathways involved in the stress response by the release of catecholamines, while elevated plasma cortisol stimulates catabolic metabolism, elevating the need for oxygen uptake and consequently increasing VF. VF is therefore considered a useful indicator of the extent to which fish are disturbed or stressed (Alvarenga and Volpato, 1995; Sager et al., 2000; Queiroz and Magurran, 2005; Barreto and Volpato, 2006).

Animals and holding conditions

Forty-eight juvenile Nile tilapia were used in this study. The fish were previously maintained in indoor 2000-litre tanks (2 fish/L) for at least 30 days before the tests. The tanks were equipped with a system of constant water renewal (around 6 to 6.5 L/h), biological filters, and aeration systems to increase the oxygen input in each tank. The water temperature was maintained between 24°C and 27 °C, pH between 6.2 and 6.7 and ammonia levels < 0.04 ppm. The photoperiod was set from 07:00 am to 07:00 pm, and feeding was provided once a day with commercial ration (Guabi Pirá ® Evolution - fingerlings growth 40% protein) at 3% of fish biomass. In each of the four treatments, we operated with 12 size-matched fish with similar body weights (F = 0.05; p = 0.98; DF = 3): 49.3 g ± 30.3 g (predictable), 52.3 g ± 41.3 g (random), 50.6 g ± 28.3 g (unpredictable) and 47.5 g ± 24.3 g (unhandled). The fish used in our experiment weighed between 19.1 and 168.5 g. Despite this wide variation, all experimental groups contained both smaller and larger subjects – which resulted in a wide standard deviation across all groups, but also resulted in no difference in weight between treatments.

During the tests, the fish were isolated for five consecutive days. Cichlid fish are territorial, so the isolation was necessary to prevent alterations in the ventilatory frequency due to social interaction, rather than to the regime for the predictability of the potential stressor. All fish were returned to the maintenance tanks, where they were with other fish, immediately after the end of the tests. During the five days of the experiment, the fish were fed daily around 11:30 am, when the first round of handling (confinement and measurement of VF) finished.

All the experimental procedures were conducted in accordance with the Brazilian legislation regulated by the National Council for the Control of Animal Experimentation (CONCEA) and Ethical Principles in Animal Research formulated by the Brazilian Society of Science in Laboratory Animals. The procedures were approved by the Bioscience Institute, UNESP, Ethics Committee on Use of Animals (CEUA) protocol number 543.

Experimental design

The potentially stressful experience to which the fish were subjected is described as a potential stressor. For any individual subject, it is often not known whether the experience had negative effects.

The potential stressor was confinement of the fish, performed using an opaque partition which restricted the fish to one side of the aquarium in 10% of the normal water volume for 30 min (adapted from Auperin et al., 1997; see Figure 1). This confinement limited fish movement and was chosen because, although it was disturbing to the fish, there was no substantial suffering caused and no physical injury to the animals. We evaluated VF responses of Nile tilapia submitted to different predictability schedules of a confinement potential stressor (Figure 1). For this purpose, fish were first isolated and then randomly assigned to one of four groups (N = 12), which varied according to the confinement schedule for five consecutive days, as follows:

Predictable treatment: the individual fish were exposed to the potential stressor at the same time every day. Confinement occurred twice every day, always at 08:00 am and 01:00 pm.

Random treatment: fish were exposed to the potential stressor at different times every day. Fish were confined twice every day, at a random time each day (once in the morning and once in the afternoon).

Unpredictable treatment: fish were not handled on days 1–4 and were exposed to the potential stressor only once, on day 5 at 09:00 am.

Unhandled control: there was no confinement of fish. Fish were not handled throughout the experiment. VF was measured at the same pre- and post-confinement time intervals as in the other treatments.

Each fish was observed on five consecutive days. The experiment was conducted in two blocks in two consecutive weeks, each block consisting of 24 animals in total, 6 per treatment, reaching 12 fish per treatment at the end of the experiment.

Procedures

Fish were transferred from the indoor tanks to individual glass aquariums (40 cm × 20 cm × 25 cm; one fish per aquarium), where they were maintained until the end of the experiment (a total of 5 days). All aquaria were equipped with aerators functioning continuously and were cleaned daily, at the end of the day, by removing the faeces on the bottom with a siphon and replacing the volume with treated water. There was no water circulation among the aquaria to avoid the exchange of chemical information among the fish. Aeration was constant, and the space allowance was one 50 g fish within a 20-litre aquarium, i.e. 2.5 g/L, allowing adequate oxygen supply for each individual. In these aquaria, individual fish were submitted to independent treatments (N = 12) as detailed above.

Ventilatory frequencies were measured 5 min before confinement (pre-confinement VF) and at 0 (immediately after the end of confinement), 15-, 35-, and 75-min post-confinement on all 5 days of tests (Figure 1). As explained above, VF was used as a non-invasive indicator of potential stress response in fish (Sager et al., 2000; Barreto and Volpato, 2004; Tzaneva and Perry, 2014). Minimal handling or other measurement-linked disturbance were avoided (Ellis et al., 2004).

The VF was measured by direct observation and was visually determined by counting the required time for 10 opercular beats (as described by Alvarenga and Volpato, 1995). The observer watched from the front of each aquarium: using a hand chronometer to time 10 opercular beats. The number of seconds for 10 opercular beats was recorded, and then the ventilatory frequency (opercular beats per minute) was calculated.

Data analyses

The data (Supplementary Table 1) showed a normal distribution (Kolmogorov-Smirnov; p > 0.05 and were homoscedastic; Levene’s test; p > 0.05). Thus, ventilatory frequencies (and their variations; ΔVF) for day 5 were compared by two-way ANOVA with repeated measures, with treatments as the categorical factor and time of VF as repeated measures. We compared the ventilatory frequencies between the groups before the confinement on day 5 (pre-confinement) and at 0 min after the confinement (post-confinement 0 min). We also evaluated VF variation (ΔVF) at 0, 15, 35 and 75 min after the stressor, using the pre-confinement ventilatory frequency at day 5 as a baseline, meaning ΔVF was calculated on day 5 of the experiment by subtracting VF at 0, 15, 35 and 75 min after a stressor from the respective VF of the pre-stressor condition measured 5 min before starting confinement. We used Fisher’s LSD test for post-hoc comparisons. Comparisons of data with the first day of the experiment were not conducted because fish were disturbed by being introduced into the test aquarium on this day. A p-value criterion of 0.05 was used for all analyses. The software SigmaStat version 3.5 was used for all analyses.

Pre-confinement VF on day 5 did not differ among treatments (Table 1). Immediately after confinement (0 min post-stressor), VF was higher in all treatment groups compared to the unhandled control on day 5. Confinement increased VF in fish in both the unpredictable and predictable treatments on the final test day. Although the absolute differences in operculum movement rate may appear modest, a graded increase was observed across treatments, with a 43% increase in the unpredictable group, compared to 25% in the random and 15% in the predictable group.

When we compared ventilatory frequency changes on day 5 after releasing the fish from the confinement (0 min after the stressor), VF was increased in all confinement treatments but was not in the unhandled control (Figure 2). This effect, however, was higher and maintained over time only in the unpredictable treatment.

By assessing the raw data and not the variation, at point 0 (immediately after confinement), only the unhandled control was different (Figure 3). At 15 minutes after confinement until the end of measures, there was no difference among groups.

In Figure 4, the two regimes which received the potential stressor for the five consecutive days are isolated from the unpredictable for a better understanding of the different degrees of predictability of the potential stressor – one treatment with full predictability, every day at the same time, and another with less predictability, every day but at random times. The pattern of change in the random treatment was an increase in the VF, which was then sustained. This pattern was not observed in the predictable treatment fish that were also stressed for 5 consecutive days.

We found that ventilatory frequency in Nile tilapia, an indicator of potential stress, varied according to the degree of predictability of the stressors. Small differences in the timing of potential stressors were sufficient to alter the magnitude and duration of the response. Among the treatment groups, the random schedule appeared to have the greatest negative impact on fish welfare, as VF remained elevated even after five consecutive days of exposure, suggesting a sustained state of alertness. In contrast, in the predictable group, VF was also increased, but the response was not maintained over time, indicating that predictability can mitigate the stress response.

The predictable group of fish were confined at the same time twice a day, in the morning and the afternoon. It is possible that the response was conditioned by the presence of the experimenter, increasing the predictability for this group by associative learning. Meanwhile, for the random group, seeing the experimenter may have made them aware that the stressor was coming, but not exactly when, thereby increasing uncertainty and anticipatory response. If this assumption is correct, it may have been that, if the predictable group was set up as the last one to receive the stressor (same time every day), the fish would show the same sustained response that the random group showed due to the incomplete predictability of the stressor. Degrees of predictability have been discussed as a key component impacting different types of anxious responding as well as anxiety disorders (Zvolensky et al., 2000).

The use of VF as an indicator of stress has been previously criticised as insensitive (Barreto and Volpato, 2004). We acknowledge that VF represents a single physiological marker and was not validated here against endocrine or biochemical indicators (e.g., cortisol or glucose), which constitutes a limitation of the study. However, VF was sensitive enough to detect differences in the degree of potential stressor predictability. Within the context of the study, VF appeared to reflect subtle response differences, indicating potential utility in repeated-measures experimental designs. For the present study, we used a standard time of confinement min and found that VF was increased for all confinement treatments. Thus, VF was sufficient to detect a response to potential stress, as such an increase has been considered a response to metabolic demand imposed by a potentially stressful condition (Barreto and Volpato, 2004).

Fish from all groups with the potential stressor protocol significantly increased VF at time 0 min after confinement, but only fish in predictable and unpredictable treatments increased VF when comparing pre- and post-confinement VFs. This indicates a greater response in these two treatments after confinement. Despite this, such a high VF increase in post-confinement was sustained over time in the unpredictable treatment, as expected. In addition, it is worth highlighting the fact that the pre-confinement VF was equal before the disturbance, even on day 5 of the experiment. It is also important to consider that individual behavioural traits can influence how stimuli are perceived by fish and modulate their stressor response (Millot et al., 2014), which could not be assessed here due to the absence of behavioural measurements.

The higher increase in operculum movements observed under unpredictable (43%), random (25%) and predictable (15%) schedules likely reflects activation of respiratory and autonomic systems. The graded response across treatments suggests that increasing uncertainty progressively limits the animal’s ability to anticipate and downregulate arousal, thereby sustaining physiological activation. Even moderate but persistent elevations of such responses can increase energetic costs and contribute to poor welfare over time.

Considering the random treatment, one possible explanation for this response is that the imminence of the disturbing stimulus, without temporal certainty, may have increased uncertainty associated with the stressor, a condition that has been linked to anxiety-like responses and prolonged physiological activation in vertebrates (e.g. Cho et al., 2009; Zvolensky et al., 2000). Thus, two potential stressors - anticipation and confinement itself - may have interacted in the random treatment, maintaining the response over time (Cho et al., 2009). In contrast, the higher and sustained VF increase in the unpredictable treatment likely reflects coping with a completely unexpected confinement event. In this condition, both the timing and the occurrence of the stressor were unknown, increasing the impact of the aversive experience.

We emphasised the subtle difference between predictable and random treatments. While these differences may appear small, they suggest that minor variations in predictability schedules can sustain stress response, which could be detrimental to the welfare over time. Within an animal welfare framework, stress is defined as exposure to environmental challenges that overtax coping mechanisms and reduce fitness, and sustained physiological activation under such conditions is considered part of poor welfare (Broom and Johnson, 2019). However, our study was limited to a five-day experimental period, and longer-term studies are required to determine whether these patterns persist or escalate under chronic exposure.

The partial predictability represented here by the random treatment may have allowed the fish to learn over the 5 test days that confinement would occur, but they were not able to anticipate the exact time that it would appear. Accordingly, we conclude that the lack of predictability may elicit the strongest response, which is in line with the idea that animals are less impacted by predictable stressors (Tsuda et al., 1989; Koolhaas et al., 2011; Galhardo et al., 2011). The partial predictability made coping more difficult, raising an anticipatory state. This study, therefore, highlights the importance of a high predictability schedule, and those small differences can have a detrimental effect by sustaining the stress response.

Evidence from other aquaculture species suggests that unpredictable feeding regimes are associated with poorer welfare outcomes. For example, in Atlantic salmon, unpredictability has been linked to increased fin damage and altered social structure (Cañon-Jones et al., 2012). While different welfare indicators were measured in the present study, the consistent association between unpredictability and adverse outcomes supports our interpretation.

Fish were housed individually throughout the experiment, a condition that may itself represent a stressor in cichlids. This design choice represented a trade-off, as social housing introduces additional confounds related to dominance hierarchies. Social isolation was therefore used to minimise variability unrelated to predictability, but its potential contribution to baseline stress should be considered when interpreting the results.

At present, many practices involving fish species neglect consideration of their capacity for suffering. Fish species have been included in the discussions about their interest in not suffering, as complex beings with high levels of cognitive ability. This is because of the large body of evidence showing they are: able to feel pain (Sneddon, 2003, 2004; Braithwaite and Ebbesson, 2014), can form map representation (Braithwaite et al., 1996; Braithwaitede Perera, 2006), experience emotions (Broom, 2016; Tatemoto et al., 2021), and need a complex environment with stimulation to maintain an acceptable level of welfare (Tatemoto et al., 2021). Additionally, one of the frontiers for cognition complexity is the ability to perceive and recognise a reflected mirror image as self. This approach, which is considered a valuable behavioural assay for assessing whether an individual has the capacity for self-recognition, has been demonstrated in fish (Kohda et al., 2019, 2022).

An emerging body of studies emphasises that changes in stress perception and coping are not only reflected at the physiological level but also may involve dynamic molecular and epigenetic regulation of stress-responsive pathways (e.g., Tatemoto et al., 2023). Epigenetic mechanisms, such as DNA methylation and histone modifications, can mediate environmentally induced alterations in gene expression without altering the underlying DNA sequence (Guerrero-Bosagna and Skinner, 2011; Skinner and Guerrero-Bosagna, 2014). Some findings highlight how such environmentally responsive epigenetic changes may persist across the lifespan, interact with gene regulatory networks, and be at least partially reversible, suggesting a basis for emotional plasticity that bridges external predictability and internal physiological states across contexts and time (Allen and Guerrero-Bosagna, 2025). Integrating measures such as stress-related gene expression (e.g., HSP70) and epigenetic marks alongside physiological indicators in future work would therefore help unpack how predictability modulates emotional plasticity and stress resilience.

To the best of our knowledge, this is the first study showing differences in the VF in response to the degree of predictability. The VF is a helpful and non-invasive tool to assess potential stress in fish, especially relevant in repeated-measures experimental designs. Additionally, here we show that VF varies according to small differences in the treatment schedule, increasing the validity and relevance of this indicator. Nonetheless, its use should be interpreted within the context of its limitations and ideally complemented by additional indicators in future work.

We expected that predictability of a potential stressor would mitigate the response, whereas unpredictability would both intensify and prolong it. Interestingly, we found that partial predictability, i.e. when the stressor is known to exist but its timing is uncertain, also sustains the response, highlighting the importance of two distinct components of a stressor: awareness and timing. When the timing of a potential stressor cannot be predicted, negative responses are likely to persist longer, increasing their potential impact on welfare, as reflected here by sustained changes in ventilatory frequency.

Although this study focused on a single non-invasive physiological indicator over a limited time frame, these findings are highly relevant for the management of aversive events in animals and are likely to apply across species, emphasizing the need to consider both awareness and temporal predictability when evaluating stress and designing welfare interventions.