Introduction

Aquaculture is a key global provider of protein for human consumption, and intensive marine fish farming has grown rapidly since the 1970s and is today established as an important industrial segment in all continents except Antarctica. For many cultured species, the farming process is divided into two distinct phases, the first of which is conducted in hatcheries and land-based facilities, before the fish are transferred to marine sea cages where they enter the second phase of the production known as the on-growing phase. During this phase, the fish are intensively fed and brought up to marketable sizes before they are finally collected, slaughtered, and sold. While growth is normally the main focus during the sea phase of marine fish farming, fish farmers are increasingly starting to introduce measures to ensure acceptable fish welfare (Barreto et al. 2022). This is motivated by a desire to maintain ethically sound animal production and because impaired welfare and stress may lead to reduced product quality and, in the most extreme cases, mortality events (Santurtun et al. 2018). The motivation to monitor and eventually control the welfare and stress levels in commercial fish farms is therefore strong.

The operations and management measures applied when producing fish in sea cages are known to cause stress in fishes (Braithwaite and Ebbesson 2014). This includes common practices such as feeding, counting, grading, and harvesting that may occur throughout the entire on-growing phase at sea (Santurtun et al. 2018). Stress may be particularly high prior to harvesting, as fish are often fasted, and then crowded so they can be pumped or brailed into boats or mobile stunning facilities (Lines and Spence 2012). The effects of stress may range from immediate effects, such as changes in blood hormones including cortisol (Dunlop et al. 2004), to long-term effects such as impaired appetite, tissue inflammation, and increased energy into tissue repair processes (Løvmo et al. 2022), all of which contribute to reducing the efficiency of the fish farm and impacting animal welfare. Given the ongoing expansion of aquaculture, with farming of aquatic animals accounting for ≈ 49% of global fisheries and aquaculture production in 2020 (FAO 2022), the development of methods to assess fish stress within net pens is necessary for effective management, both by improving animal welfare and increasing production efficiency by optimising possible mitigation strategies.

Stress levels of caged fish can be assessed and measured by a variety of techniques, including visual observation, blood sampling, and implanted sensors. Visual observation, typically involving cameras mounted within the pen, offers the ability to infer stress levels by changes in fish behaviour (Martins et al. 2012). Such analyses are often supplemented by automated video analysis techniques (Yang et al. 2021) and are non-invasive, but have the limitation that individual fish behaviours may be obscured by other individuals in dense groups, particularly in turbid water. On the other extreme, blood sampling (e.g., Casanovas et al. 2021) provides a direct measurement of blood constituents associated with stress (such as cortisol) for individual fish, but is very invasive in requiring capturing and handling the fish. Since this sampling procedure may in itself induce stress, blood sampling is limited in scope and can only provide information at distinct point samples throughout the time fish are held in the pen. Sensor implants, extensively used in studies on wild fish for decades (e.g., Cooke et al. 2004; Thorstad et al. 2013), offer a third option to this sampling challenge. They have the potential to provide continuous monitoring of fish in sea cages, both in terms of internal state (e.g., heart rate) and behaviour (e.g., swimming activity) (Brijs et al. 2021). While the deployment of sensor implants is invasive in the sense that the fish must be captured and sedated, and then undergo surgical implantation, studies have suggested that the data obtained is valid if the fish are given sufficient recovery time after surgery (Brijs et al. 2021; Føre et al. 2021; Warren-Myers et al. 2021). Sensor implants have been shown to be effective for monitoring activity and heart rate in Atlantic salmon in tanks (Hvas et al. 2020; Svendsen et al. 2021b; Zrini and Gamperl 2021; Yousaf et al. 2022) and sea cages (Brijs et al. 2018; Føre et al. 2018b; Gamperl et al. 2021; Stockwell et al. 2021; Warren-Myers et al. 2021), suggesting their viability for use in aquaculture. However, only in tank experiments have activity and heart rate been measured simultaneously and while subjecting the fish to experimental stress conditions (Svendsen et al. 2021b). Thus, for sea cages, the ability of sentinel fish carrying such tools to provide proxy information on stress requires further evaluation.

The aim of this study was to assess if sensor implants can be used to measure stress and thus estimate one important welfare indicator for fish in aquaculture pens, and if such measurements can be correlated to physiological factors measured through blood sampling. The experiment was conducted in meso-scale sea cages and consisted of two parts: i) a biologger study that used implanted sensors to monitor activity and heart rates of fish undergoing stress (crowding); and ii) a blood chemistry study that analysed samples of blood constituents (cortisol, glucose, lactate, and chloride) of a second group of fish subject to the same conditions. Consistency between the biologger data and blood chemistry studies in terms of response to stressing would suggest that sensor implants may be used as an alternative or supplement to blood sampling. The specific objectives of this study were to examine (1) how fish activity and heart rate, as measured by sensors, respond to stress initiated by crowding of the type that occurs in aquaculture pens; (2) the consistency in response to stress between sensor measurements and blood samples, and thereby evaluate if sensors can be used in place of blood sampling to monitor stress; and (3) if tagging has adverse effects on fish, in terms of blood chemistry or damage to organs.

Material and methods

Ethics statement

All fish handling, surgery, and blood sampling were conducted in compliance with the Norwegian Animal Welfare Act and approved by the Norwegian Animal Research Authority (FOTS permit no. 23250). Several experimental refinement strategies were applied in the trials, including sampling fish with a knotless dip net, immediately transferring sampled fish to an anaesthetic bath to minimise stress, and continuously irrigating the gills with aerated water containing maintenance anaesthetic and covering the heads of all fish with a moist cloth during surgery.

Experimental animals, housing, and husbandry

The experiment was conducted between 5.10.2020 and 4.11.2020 at the LetSea research facility on Dønna, northern Norway (N 66°04′, E 12°29′). The site consists of a floating steel array holding square net pens (5 × 5 m, 8 m deep, 174 m3 volume per pen) and is accessible from land. The pens were equipped with automatic feeders as well as temperature and oxygen sensors. During the course of the experiment, daily average water temperature within the pens decreased from 12 to 9 °C.

A total of 294 Atlantic salmon of the same genetic strain (Salmobreed) were included in the experiment. During the experimental period, fish were fed a standard industrial diet, except on fasting days. Husbandry was carried out by on-site personnel trained in handling experimental fish.

Study design, sampling plan, and timeline

The experiment was conducted over a 31-day period. After an acclimation period, the fish were subjected to crowding on day 15, followed by a recovery period. Data storage tags were used to monitor activity and heart rate throughout the experimental period, supplied with blood samples taken at distinct time points (Fig. 1).

Fig. 1
figure 1

Experiment timeline. Overview of the experiment timeline (not to scale) and the expected stress level in fish during the experiment. Timings of blood sampling events (S1–S5) as well as specific periods for observation of activity (ACT) and heart rate (HR) (B1, B2, L1–L3, R1–R3) that were used for statistical comparison are indicated. Note that the elevations in stress level implied at sampling events S1, S3, and S5 reflect that sampling itself is a stressor

Full size image

Fish were kept in two holding pens of the same size as the experimental pens for > 2 weeks, where they were fasted the last 2 days before the experiment. At the very beginning of the experiment (day 1), blood was taken from 18 fish (sampling S1, Table 1) in the holding pens to provide an indication of blood chemistry, pre-handling and -treatment. Each sampled fish was anaesthetised using Benzoak Vet (70 mg L−1 benzocaine) until loss of consciousness; blood was then immediately sampled from caudal blood vessels using heparinised 5-mL syringes. The fish were hauled from the water in groups of five and sampling was completed within 2 min for each group when the next group of fish had been moved to the sampling bench. The procedure was completed within 5 min. In general, the sample time for each fish was around 10 s. The fish were then killed with a sharp blow to the head.

Table 1 Blood sampling plan. Overview of timing and number of fish sampled for blood analysis (see also Fig. 1)
Full size table

Over a 2-day period (day 1 and 2 of the experiment), a sample of fish from the lightly crowded holding pens was caught at random and implanted with data loggers to monitor activity and heart rate (N = 39 fish) before being distributed into three pens assigned for tagged fish. Each selected fish was anaesthetised using a Benzoak Vet solution (70 mg L−1 until loss of consciousness, and 35 mg L−1 maintenance dosage during surgery). The fish was then tagged with a DST (milli-HRT ACT, Star Oddi LTD; 13 mm × 39.5 mm, 12 g in air, approx. 5.2 cm3 volume) containing a tri-axial accelerometer and an ECG sensor. The tag was surgically implanted into the peritoneal cavity through an incision in the abdomen and anchored to prevent movement relative to the heart by one anterior suture through the peritoneal wall as described in Føre et al. (2021). After successful surgery, the tagged fish was monitored during recovery until it regained consciousness before it was distributed to one of three “tag pens” (13 fish per pen).

Fish not undergoing surgery (co-habitants for the tagged fish) were caught and anaesthetised simultaneously with the tagged group, measured for length and weight, and distributed over the three tag pens. In the same manner, three additional pens were populated with untagged fish designated for blood sampling throughout the experiment (“blood sample pens”). A subsample (N = 219) of these untagged fish was used to assess the median weight and length as well as the condition factor of the experimental population (5.6 ± 0.06 SE kg, 77 ± 0.3 cm and 1.2 ± 0.01, respectively) as fish undergoing surgery could not be weight/measured due to logistical constraints. During the distribution procedure, random fish (16 fish, S2) were chosen and sampled for blood constituents (as described above) to assess the stress levels in the initial phase of the experiment. As these fish had experienced variable durations of light crowding, increased stress levels were expected. At the end of day 2, the three “tag pens” each contained a total of 40 fish consisting of 13 tagged fish as well as 27 untagged co-habitants, while the three “blood sample pens” each contained 46 untagged fish. In total, this procedure took approximately 9 h split over 2 days with fish being crowded for sampling for max. 3 h before switching to the second holding pen each day.

On day 15, after an acclimation period ending with 2 days of fasting, a crowding operation was applied to the fish in all pens by lining up the net and thus raising the bottom and reducing pen volume. During the crowding operation, a “tag pen” and a “blood sample pen” were paired and treated simultaneously as not all six pens could be monitored simultaneously. Crowding was conducted according to standard industry practice in three stages with increasing levels of intensity. The first two stages lasted 1 h and equalled level 1 “goal: no stress, no vigorous activity” and level 2 “acceptable: some fins on surface”, respectively, as described in Fig. 1.1–3 in Noble et al. (2018). The final stage, level 3: “undesirable: some overexcited swimming, some white sides visible”, was limited to 10 min to minimise impacts on fish welfare. Within 5 min after the initiation of crowding at level 1, a sample of fish from the three “blood sample” pens was caught and euthanised for blood sampling (N = 18, 6 fish per blood sample pen, S3). Each of the six experimental pens therefore contained 40 fish during the crowding operation. At the end of the third level crowding stage, another sample for blood measurements was taken (as described above) from the three “blood sample” pens (N = 18, 6 fish per blood sample pen, S4) before nets were released.

On day 31, after a 16-day recovery period including a final 2 days of fasting, the experiment was concluded. Experimental pens were emptied one at a time. Fish were lightly crowded, hauled from the water in groups of five and anaesthetised before blood samples as well as length and weight measurements were taken from all tagged fish and a subsample of untagged fish from all pens (N = 59, S5). The procedure was completed in less than 10 min per pen. All 39 tagged fish and a random subsample of 23 untagged fish were then euthanised with a sharp blow to the head and autopsied to study general heart fat deposition (more detailed occurrence of fat deposition was assessed after freezing heart samples and transport to a laboratory), heart fresh weight, and gut condition. Additional data on the amount of inflammation of the external and internal sutures, the level of tag encapsulation with connective tissue and the distance from the tag to the heart were then collected for the tagged fish. Heart deposition was assessed by focussing on signs for arteriosclerosis: fat deposition around the heart in general, as well as in detail on dorsal and ventral side of the bulbus and ventricle of the heart to evaluate possible effects on heart parameters and, in the worst case, mortality following surgery. Heart deposition, gut condition, and tag encapsulation were scored on a scale from 0 to 3: absent/nothing (0), low (1), medium (2), high (3): high values for heart deposition and tag encapsulation indicate potential health issues; a high gut condition value is evidence of recent feeding. Remaining fish not autopsied were anaesthetised and then euthanised with a sharp blow to the head.

Processing of DST data

The DSTs measured acceleration and heart rate. Each datapoint of acceleration was derived from 60 s of observations of acceleration in three dimensions at 1 Hz, while datapoints of heart rate were derived from 15 s of raw ECG data sampled at 100 Hz. Datapoints were stored at varying time intervals throughout the project. In the initial phase of the project (from tagging to day 13), datapoints for activity and heart rate were logged at 30-min intervals, with each logging point producing 1 min of acceleration data at 1 Hz and 1 heart rate value. The storage rate was then increased for the 6 following days (days 13 to 19) to achieve a higher data density around the crowding events: activity and heart rate were sampled every 2 min, each registration featuring 1 min of acceleration at 1 Hz and 1 heart rate value. Thereafter, rates were adjusted to 30-min intervals in the final experimental stages (day 19 to experiment end). The DSTs were programmed such that they stopped measuring acceleration and heart rate at 0400 h UTC on the final day of the experiment (4.11.2020). All ECG data are referenced with a quality index (QI), a proprietary ECG signal quality calculated by an onboard tag algorithm, where the index indicates signal quality from high (QI = 0) to low (QI = 3).

Of the 39 implanted DSTs, five recorded no or incomplete data and were excluded from the analyses. One additional DST provided faulty activity readings for parts of the experiment so activity data from this tag were discarded. Therefore, analysis of the recorded data was done for 33 (activity) and 34 (heart rate) tags. For each DST, accelerometer data were converted from m s−2 to G, followed by high-pass filtering using a cut-off frequency of 0.2 Hz to remove the unwanted low-frequency component associated with gravity effects. The L2 norm was then calculated and averaged over a 30 s window to represent the overall body dynamic activity. Tag data for heart rate were edited to exclude values that were deemed unrealistic for adult Atlantic salmon (< 5 or > 100 BPM), or with low data quality (QI ≥ 2).

Processing of blood sample data

Blood was collected from anaesthetised fish from the caudal blood vessels using heparinised syringes. Blood samples were immediately centrifuged for 5 min at 10,000 rpm, plasma was collected and frozen at − 20 °C until it was shipped to a laboratory where it was stored at − 80 °C until analysis. Plasma lactate and glucose were analysed by the Penta400 Clinical analysator and chloride by the Mettler Toledo AT1000 chloride titrator. Cortisol was assessed by the Tecan cortisol ELISA kit RE52061 (e.g., Magnoni et al. 2019). Data collected from samples taken at S3 and S5 was compromised (a large number of measurements below the minimum detection threshold (1.68 ng mL) and were therefore excluded from the analyses.

Statistical analysis

To determine the effect of crowding on activity and heart rate, mean activity and heart rate were established by averaging collected data for each fish in each of eight observation periods (Fig. 1): two baseline periods, B1: recorded on the day before crowding, B2: recorded over the hour before initiation of crowding; three crowding periods, L1, L2, and L3, corresponding to the three crowding intensity levels; and three periods during recovery, R1: recorded over the hour after ending the crowding, R2: recorded on the day following crowding, and R3: recorded on the final full day of the experiment (day 30). B1, R2, and R3 were measured over a period corresponded to the time of day of the crowding period to avoid potential diurnal effects. For each of the three crowding intensities, activities and heart rates were extracted from the time spanning the treatment, except that data from the first and last minutes of the treatment were removed to exclude any pre- or post-treatment values that might have been included due to clock drift. The removal of heart rate values that were unrealistic (< 5 or > 100 BMP) or had low data quality (QI ≥ 2) sometimes removed all heart rate data within an observation period for certain tags, meaning that the number of individuals providing heart rate data varied according to observation period: N = 28 (B1), 25 (B2), 24 (L1), 18 (L2), 11 (L3), 29 (R1), 31 (R2), and 9 (R3); mean activity was available for all 33 individuals. Effects of observation period (B1–R3) on activity and heart rate were then determined using linear mixed effects models (R function nlme::lme). The pen containing the fish was included as a random effect. Pre-model fitting, activity was normalised by a Box-Cox transformations (geoR::boxcox.fit); heart rate had a near normal distribution and was not transformed. Body temperature was initially included as a covariate given that this varied during the experiment and might have potentially affected activity and/or heart rate. However, body temperature was found to have no significant effect, and its inclusion increased the model AIC values. Therefore, body temperature was omitted from models. After model fitting, pairwise comparisons were made among the time periods (emmeans::emmeans) using a Bonferroni correction.

Changes in blood sample concentrations (cortisol, glucose, lactate, and chloride) among the five blood sample events (S1–S5; Table 1, Fig. 1) were analysed using Kruskal–Wallis tests (rstatix::kruskal_test) and pairwise comparisons between sample times were made using Wilcox tests (rstatix::wilcox_test) with a Bonferroni correction due to a lack of homogeneity of variance among groups (tested using car:: leveneTest). Here, only data from untagged fish were included. To assess the potential effect of tagging on blood chemistry, samples of untagged fish collected at S5 were compared to the simultaneously collected samples from tagged fish using t-tests (stats::t test) after establishing homogeneity of variance among groups for glucose, lactate, and chloride; differences in cortisol between the untagged and tagged fish were not analysed using t-tests due to there being too many cortisol measurements below the minimum detection threshold. The propensity for cortisol measurements to be below the minimum detection threshold (1.68 ng ml−1; untagged N = 49; tagged N = 19) was examined using a test of equal proportions (stats::prop.test).

Autopsy data of tagged and untagged fish sampled at the end of the experiment were compared to assess potential impacts of tagging on fish condition, based on body length, body weight, condition factor, heart weight, and heart fat deposition. Comparisons were conducted using Fisher tests (stats::fisher.test) and Mann–Whitney tests. For tagged fish, the effect of tag encapsulation, distance between tag and heart, and heart fat deposition on quality (QI-value) of recorded data was assessed using a Mann–Whitney test, a correlation test, and a Kruskal–Wallis test, respectively. In addition, possible effects of suture inflammation on activity were tested comparing activity of fish without inflammation to fish with (any) inflammation for the eight observation periods using a Mann Whitney U-test (stats:: Wilcox_test).

Results

Mortality throughout the experiment was low. All tagged fish survived surgery, one tagged fish died shortly after the crowding operation on day 15, and two more tagged fish died during the final sampling event on day 31. No mortality occurred among non-tagged fish.

Activity and heart rate

Immediately after tagging and for a period lasting 2 to 3 days, activity and heart rate were elevated and more variable compared to subsequent days: this period lasted for days 1 and 2 for activity, and for days 1, 2, and 3 for heart rate (Fig. 2). Following this, activity did not show a long-term trend. In contrast, heart rate showed a slight long-term decline: for instance, average daytime and night-time heart rates were 47.5 and 39.1 BPM, respectively, on day 4, falling to 36.2 BPM (daytime) and 30.7 BPM (night-time) on the final full day of the experiment (day 30). This decline concurred with a decline in average fish body temperature from ≈ 11.6 °C (day 4) to ≈ 9.0 °C (day 30).

Fig. 2
figure 2

Long-term trend in activity (a) and heart rate (b) across the experiment. The blue bar indicates the day of crowding. The dashed red line superimposed on the heart rate plot indicates the average fish body temperature. ACT N = 33, HR N = 34

Full size image

Both activity and heart rate showed clear circadian rhythms, with increased activity and heart rate during daytime (Fig. 3): mean activity = 0.028 G (daytime, 06:00–1900 h, excluding tagging and stressing days) and 0.020 G (night-time, 1900–06:00 h); mean heart rate = 42.9 BPM (daytime) and 33.7 BPM (nighttime). Activity began to increase from ≈ 06:00 h and reached a maximum at ≈ 08:00 h, before declining to reach a stable minimum at ≈ 19:00 h. Heart rate showed a similar diurnal pattern, except that the rate of the decline in heart rate throughout the day was slower. The diurnal pattern was less consistent for heart rate than for activity. For instance, the fish with the highest heart rate (average heart rate ≈ 60–70 BMP) did not show a circadian pattern.

Fig. 3
figure 3

Circadian trends in activity (a) and heart rate (b) throughout the experiment. Thin lines show averages for individual tags; the thick black line shows the average across all tags. NB: tagging days (days 1 and 2) and crowding day (day 15) have been excluded. ACT N = 33, HR N = 34

Full size image

Activity varied according to observation period (Fig. 4a). Average activity measured during the two initial crowding levels (median G = 0.079 (L1) and 0.093 (L2)) was significantly higher than baseline levels measured on the day before and the hour before crowding (median G = 0.030 (B1) and 0.037 (B2)). The final level of crowding resulted in the highest average activity measured during the trial (median G = 0.159 (L3)). Activity was still significantly elevated compared to baseline values (median G = 0.049 (R1)) during the hour post crowding but returned to baseline levels by the following day (median G = 0.025 (R2)) and remained low until the final full day of the experiment on day 30 (median G = 0.031 (R3); Fig. 2). The difference in activity between the three experiment pens was small (see Table SI 1 for summary of the mixed model).

Fig. 4
figure 4

Effects of crowding on activity and heart rate. Activity (a) and heart rate (b) measured on the day before crowding (B1), the hour before crowding (B2), during the three increasing levels of crowding (L1, L2, L3), during recovery in the hour post crowding (R1), the day after crowding (R2), and on the final full day of the experiment (R3). Lower-case letters indicate results from pair-wise comparisons, identifying statistically identical groups (p < 0.05). The horizontal black line in the box denotes the median, with boxes representing quartiles and extending from quartile 1 to quartile 3, and the whiskers extended from quartile 1 – 1.5 × interquartile range to quartile 3 + 1.5 × interquartile range. Mixed model result details can be found in Tables SI 1 and 2

Full size image

Heart rate was less dependent on observation period (Fig. 4b). Average heart rate measured during the initial two crowding levels (median = 57.4 (L1) and 44.0 BPM (L2)) was similar to the baseline heart rate measured on the day before and the hour before crowding (median = 44.6 BPM (B1) and 47.4 BPM (B2)). During the final crowding level, however, average heart rate declined (median = 19.0 BPM (L3)) such that it was significantly lower than before crowding or during the initial crowding level). One hour after crowding, heart rate increased to a level similar to the first crowding stage and slightly above pre-crowding levels (median = 59.0 BPM (R1)). On the next day, 24 h after crowding, heart rate returned to baseline levels and stayed there for the rest of the experiment (median = 51.9 BPM (R2) and 37.2 BMP (R3); Fig. 2). Similar to activity, there was only a negligible difference in heart rate between pens (see Table SI 2 for summary of the mixed model).

Blood chemistry

Cortisol, glucose, and lactate levels showed a similar response to crowding (Fig. 5), being significantly higher when measured at the end of the crowding period (S4) relative to when measured immediately before (S3). Glucose levels declined such that levels measured on the final day of the experiment (S5) were similar to those before the crowding period (S3); lactate also decline to a lesser extent. However, the increase in level with an increase in crowding that was found when comparing levels measured on S4 with those measured on S3 was not consistently reflected in the initial samples taken when fish were relaxed (S1) and crowded (S2) at the beginning of the experiment. In contrast to cortisol, glucose, and lactate, there was little variation in chloride levels across the experiment: there was no evidence of a change in chloride levels with a change in crowding, and, contrary to glucose and lactose, chloride levels on the final day of the experiment (S5) were significantly higher than during earlier sample times.

Fig. 5
figure 5

Effects of crowding on blood chemistry. Blood chemistry results of untagged fish measuring cortisol (a), glucose (b), lactate (c), and chloride (d) levels before initial collection of experimental fish (S1), during initial crowding for fish collection (S2), at the very beginning of the crowding treatment (S3), at the end of the final, intensive crowding period (S4) and after recovery at the end of the experiment (S5). Lower-case letters indicate results from post hoc comparisons, identifying statistically identical groups (p < 0.05). The horizontal black line in the box denotes the median, with boxes representing quartiles and extending from quartile 1 to quartile 3, and the whiskers extended from quartile 1 – 1.5 × interquartile range to quartile 3 + 1.5 × interquartile range. Dots represent values of individual fish. Numbers of samples included in the analysis are indicated above each box. Cortisol samples taken at S3 and S5 had too few samples above the detection limit to be included in the statistical analysis (indicated by an asterisk)

Full size image

Possible effects of tagging on fish

No difference in blood chemistry between the tagged and untagged fish was found on the final day of the experiment (S5) for glucose, lactate, and chloride (Fig. 6). However, a significantly higher proportion of tagged fish had cortisol levels above the minimum threshold (Pearson’s χ2 = 8.64, p = 0.002).

Fig. 6
figure 6

Effects of tagging on blood chemistry. Blood chemistry results comparing tagged and untagged fish measuring glucose (a), lactate (b), and chloride (c) levels at the end of the experiment (S5, day 31). The horizontal black line in the box denotes the median, with boxes extending from quartile 1 to quartile 3, and the whiskers extended from quartile 1 – 1.5 × interquartile range to quartile 3 + 1.5 × interquartile range

Full size image

Autopsies of the 39 tagged fish showed that direct effects of surgery were present, including some longer-term effects on fish health (indicated by examining the heart). While one-third of the tagged fish (N = 13) showed no inflammation of the external suture, about half showed low (N = 9) or medium (N = 12) inflammation, and five fish showed high inflammation levels. Inflammation of the internal suture was less prominent: 90% had no (N = 11) or low (N = 24) inflammation; the remainder showed moderate inflammation (N = 3); and one fish had high inflammation. The presence of inflammation had no impact on activity data (see Tables SI 3 and 4). Tag encapsulation occurred in all tagged fish, though mostly to a low extent (N = 34) with only four fish having moderate encapsulation and one fish having high encapsulation.

Autopsies furthermore showed similar physiological condition between tagged and untagged fish (Table 2). No significant difference existed in fat deposition around the heart between tagged and untagged fish. However, gut condition was significantly higher for untagged fish, indicating stronger signs of recent feeding compared to tagged fish. Untagged fish were also larger and fatter (median body mass = 6.7 kg, median length = 79.8 cm, median condition factor = 1.33, N = 58) than tagged fish (median body mass = 5.6 kg, median length = 77 cm, median condition factor = 1.21). Hearts of untagged fish (median = 5.8 g, N = 23) were also significantly heavier than those of tagged fish (median = 4.8 g, N = 39) though both made up 0.1% of body weight. This difference in heart weight was mainly due to the difference in fish size.

Table 2 Statistical results regarding the effects of tagging. Evaluation of the effects of tagging on fish health (n = 39 tagged and 23 untagged fish) as well as potential impacts on data quality (n = 34) based on autopsy data
Full size table

A post facto assessment of the potential influence of tag encapsulation, distance from the tag to the heart (between 1 and 4 cm, with most tags secured at 2 cm distance) and heart fat deposition on heart rate data quality (abundance of data with quality index = 3), showed no statistical effect (Table 2).

Discussion

The present study successfully measured activity and heart rate in Atlantic salmon in meso-scale sea cages over 31 days using data storage tags, during which fish were subjected to crowding events. Activity was a valid proxy for increased stress levels during crowding, showing a response that was consistent with sampled blood glucose, lactate, and cortisol. In contrast, heart rate decreased during peak crowding before returning to baseline levels, suggesting that it may not be a useful metric for measuring acute stress when used alone. Data logging was useful for monitoring both long-term and circadian trends in activity and heart rate.

Tagging had some effect on fish condition in the form of increased mortality and lack of growth compared to the untagged group. In addition, suture inflammation and tag encapsulation were present in some fish, although blood parameters did not differ between groups. In conclusion, data storage tags are a promising tool for welfare monitoring in aquaculture, yet more research is needed to improve precision of the used methods as well as to minimise impacts of tagging on fish welfare and thus ensure validity of the tagged group as proxy for their untagged conspecifics.

Effect of crowding on activity

Crowding resulted in increased activity, with maximum values measured during the highest crowding intensity being 30 times greater than baseline activity (day before crowding). While there was a tendency for increased activity during the second crowding level compared to the initial one, these differences were not significant. This pattern is very similar to measurements of Føre et al. (2018b) who recorded activity of Atlantic salmon in sea cages over 4 months and three delousing events that included two crowding stages ahead of a treatment and found a sevenfold increase comparing baseline activity to the main crowding and treatment phase. However, similar to the lower crowding stages in the present study, Føre et al. (2018b) found no consistent difference in activity when comparing different crowding intensities. This observation suggests two interpretations: that activity is too variable among individuals for nuanced differences between crowding stages to be observed, or that the fish are not sensitive to differences in crowding at lower densities. It is, however, clear that level 3 crowding conducted in this study caused a strong reaction in the fish, emphasising why crowding to this level is not recommended from a fish welfare perspective (Noble et al. 2018). In contrast, crowding experiments in tanks conducted by Svendsen et al. (2021b) did not result in a similar immediate increase in activity during crowding. Instead, values increased 2.4 h after the stress event, and stayed elevated for 16.2 h. However, the crowding applied in that study was realised by reducing water levels in the tanks, a process that may have provoked a different behaviour than crowding using nets in open sea water. This harmonises with the findings of Yousaf et al. (2022) who in a similar crowding experiment observed that fish were staying motionless on the bottom of the tank while water levels were low, before showing increased activity (though not formally measured) afterwards when free movement was possible. It is therefore possible that the two stressors are not perceived equally by the fish and thus cause different reactions.

Regarding the length of impact, the present study confirmed patterns observed by Føre et al. (2018b) and Svendsen et al. (2021b) in that crowding did not have a lasting effect on activity levels but returned to baseline values within 24 h after the crowding trials. In conclusion, activity measurements can be used to track high-impact stress events such as crowding but may not be suitable to discern possibly nuanced reactions within the observed event.

Effect of crowding on heart rate

Heart rate did not follow expectations of increased values during the crowding procedure, instead showing no measurable difference during the initial crowding stages followed by a significant reduction during the final crowding stage. After crowding, heart rate was slightly elevated compared to baseline values an hour after the treatment before returning to baseline values the day after. In comparison, most studies measuring heart rate of Atlantic salmon and trout during crowding observed a pronounced increase (Brijs et al. 2018; Svendsen et al. 2021b; Warren-Myers et al. 2021). Generally, these studies recorded that heart rate remained elevated after the stress event for 24 h (Brijs et al. 2018; Hvas et al. 2020; Svendsen et al. 2021b), though Warren-Myers et al. (2021) reported a post crowding heart rate below pre-crowding levels. One other study has observed a similar drop in heart rate during a crowding treatment (Yousaf et al. 2022), an effect that was attributed to fish resting motionless on the tank floor while water levels were low. As fish in this study displayed agitated swimming typical for intense crowding to level 3, this can be excluded as an explanation. A possible explanation for the drop in heart rate observed in our study can be found in the relationship between heart rate and stroke volume (and oxygen extraction). Fish have an extreme capacity for modulation of stroke volume, thus reducing the need to adjust heart rate when regulating oxygen consumption (Thorarensen et al. 1996). As such, heart rate may be responsible for less than 50% of variation in oxygen consumption, a number that moreover appears to decrease when fish are exposed to increasingly stressful procedures (Thorarensen et al. 1996). It may thus be that salmon in our study did increase oxygen consumption as a reaction to the stressful situation; however, the adjustment was done via an increase in stroke volume of such a size that heart rate could in fact be lowered. Without any way to measure stroke volume, it is not possible to decide with finality.

In conclusion, heart rate did not function as a credible proxy to predict acute stress in salmon in this study and contrasts as such with other studies. Together with the variability seen in the literature, the results from this study underline the need for more research before heart rate can be used as a reliable tool in aquaculture stress and welfare monitoring.

Effect of crowding on blood chemistry

During the experimental crowding (S4), the fish responded with a significant stress response. This was again confirmed with elevated glucose and lactate levels indicating that the need for energy approached maximum aerobic capacity, as shown in Svendsen et al. (2021b). However, compared to normal resting plasma cortisol levels that are in the range < 20 ng ml−1 (Olsen et al. 2002), the relatively high initial cortisol levels in many of the fish sampled at S1 suggest that they, too, had experienced some stress already at the time of the initial sampling. While sampling was conducted fast after very short and light crowding, activity surrounding the pens may have caused this increase in stress markers (Wendelaar Bonga 1997; Iversen 2013). Fish sampled at S2 showed cortisol and glucose values similar to the experimentally crowded fish at S4, indicating that extended crowding (up to 3 h) at very light levels as experienced during distribution into the experimental pens caused stress similar to short and intense crowding (Svendsen et al. 2021b).

For the chloride levels, we observed no significant changes that indicated osmoregulatory failure in the fish, and levels were similar to observations made in previous studies on adult salmon.

Recovery after surgery and long-term trends across the experiment

Activity and heart rate were elevated for 2 to 3 days following tag implantation. This is similar to observations made in other tagging studies with salmonids where typical recovery times until normalisation of heart rate are reported to be between 3 and 4 days (Brijs et al. 2018; Føre et al. 2021; Warren-Myers et al. 2021). While heart rate stabilised in general, it showed a subtle decline in both day- and nighttime values over the entire duration of the 31-day experiment. A similar extended decline in heart rate past an initial drop observed in the first days has been described in other experiments with Atlantic salmon: Hvas et al. (2020) reported a pronounced decrease in heart rate in the first week following surgery, and a less severe decline in the second week before stabilising for the remaining 11 weeks of the trial; Yousaf et al. (2022) reported a gradual decline until 10 days after surgery following an initial drop in heart rate after 2 days; and Zrini and Gamperl (2021) observed stabilisation of heart rate 14–22 days post-surgery. The authors interpreted the consistent decline as a sign of continued recovery from stress during tag implantation. Whereas Hvas’ and Yousaf’s trials were conducted in tanks with controlled conditions including water temperature, Zrini and Gamperl (2021) conducted their experiments in tanks with ambient temperature sea water where they observed a reduction in daytime heart rate of approx. 3.0 BPM per degree centigrade water temperature.

The current study was conducted in open sea water pens where the fish experienced a decline in temperature from 12 to 9 °C over the course of the experiment. Heart rate thus declined with a decrease in temperature (4.5 BPM per 1 °C decrease of water temperature), similar to the results of previous studies with Atlantic salmon in sea cages where heart rate was tracked over several months over a range of temperatures (approx. 4.3 BPM per 1 °C; Gamperl et al. 2021), and observations of Chinook (Onchorhynchus tshawytscha) and Sockeye salmon (Onchorhynchus nerka) during biologger-assisted studies in rivers with a temperature gradient (Prystay et al. 2019; Twardek et al. 2021). These observations are consistent with the knowledge that elevated temperature leads to elevated metabolic rates and tissue oxygen demand, requiring increased cardiac activity (Eliason and Anttila 2017). The observed decline in heart rate may thus not necessarily indicate continued recovery from stress, but rather imply an adjustment of heart rate due to the decline in temperature. This is supported by the fact that activity did not show a similar extended recovery despite this having been observed in other studies (e.g., Føre et al. (2021) reported impacts on activity for up to 10 days in some fish).

Circadian effects

A clear circadian rhythm was established in both activity and heart rate within 24 h after logger implantation and lasted for the entire duration of the experiment. Peak values in the morning around 8:00 h coincided with increasing light levels (sunrise between 5:15 and 6:17 h in the experimental time period) and the beginning of regular maintenance and husbandry activities at the farm site (between 7:00 and 8:00 h). Circadian rhythms with higher values during daytime for activity and heart rate are often recorded for Atlantic salmon, both in sea cages and in tanks (Føre et al. 2018b; Stockwell et al. 2021; Warren-Myers et al. 2021; Zrini and Gamperl 2021), and have also been described for other salmonid species (e.g., Brijs et al. 2018; Prystay et al. 2019; Twardek et al. 2021). This natural pattern may be supported by feeding taking place during daytime hours and a general higher activity around the fish. However, opposing observations exist where activity (though not heart rate) of Atlantic salmon in tank experiments peaked during nighttime (Føre et al. 2021).

Both mean daily and nightly heart rates of 40 and 33 BPM, respectively, were within the range of results from similar studies where heart rate spanned 26 to 63 BPM during the day and 23 to 55 BPM during the night (Hvas et al. 2020; Føre et al. 2021; Gamperl et al. 2021; Warren-Myers et al. 2021; Zrini and Gamperl 2021). Fish in these studies ranged in weight from an average 1.2 kg to 3.6 kg, and were thus considerably lighter than the tagged individuals in our study (5.5 kg). However, in general, weight appears to have less impact on heart rate than temperature, with the highest heart rates reported from the warmest waters (Gamperl et al. 2021) and the lowest from studies with the coldest water temperatures (Føre et al. 2021).

Possible effects of tagging on fish

Tagging did have a negative effect on fish survival and growth. While there was no mortality among cohabitants, three of the tagged fish died during the experimental period, resulting in a mortality rate of 7.6%. This was, however, slightly lower than the expected average of ≈ 12% for a 31-day sea cage experiment (Macaulay et al. 2021). The tagged fish did not gain any weight or length in comparison to the average population weight and length taken at the start of the experiment, and were 16% lighter and 3% shorter than the subgroup of cohabitants assessed for weight and length at the end of the experimental period. Intestinal condition was scored consistently higher for the untagged cohabitants than for tagged fish, indicating a lack of appetite in tagged fish and thus offering a potential explanation for the lack of growth. Possible reasons behind this may be found in the presence of inflammation of the external suture for some of the fish, although more than half of the tagged fish had good wound healing but still lacked substantial growth. Moreover, the presence of inflammation had no impact on activity of the tagged fish, indicating that other factors than wound healing could have been more critical. One such contributing factor may have been the sheer presence of the tag in the abdominal cavity. While the tags had a low volume (approx. 5.2 cm3) compared to the fish (approx. 5,600 cm3 at 5.6 kg), they may still have displaced organs and thus affected appetite and/or food processing (Macaulay et al. 2021). Other studies report similar reduced growth with tagged fish being 18% lighter and 4% shorter (Hvas et al. 2020) or 20% lighter and 8% shorter (Warren-Myers et al. 2021) after experimental periods of 13 weeks and 5 months, respectively.

Studies show salmonids reacting to starvation with a lowered metabolism resulting in lower activity and lowered baseline heart rate or cardiac output. However, as this does seem not to include swimming activity or cardiac performance in reaction to stimuli (Höjesjö et al. 1999; Petersen et al. 2011), it may not serve as an explanation for the unexpectedly reduced heart rate during crowding.

Blood parameters chloride, glucose, and lactate were not influenced by tag implantation. Chloride levels were relatively stable and suggested no major leakage around the wound. Likewise, low variation in glucose and lactate pointed in the same direction. However, the fact that prevalence of measurable cortisol levels at the end of the experimental period was higher in tagged fish than untagged individuals may indicate an effect of the surgery and/or the presence of the tag. Thus, similar to the lack of growth, possible impacts on fish behaviour during crowding cannot be conclusively excluded.

Arteriosclerosis of the coronary artery is relatively common in farmed Atlantic salmon and may vary from small deposition to major lesions. In severe cases it is possible that blood flow is reduced leading to impaired swimming performance (Ekström et al. 2018) and more seriously reduced stress tolerance, myocardial necrosis, and mortality following stress and handling (Poppe et al. 2007). There was no difference in fat deposition on the bulbus or ventricle of the heart between tagged and untagged fish. Furthermore, there were no parameters of heart function that could be related to arteriosclerosis, and neither could a correlation between heart rate and arteriosclerosis be found.

Degree of tag encapsulation or tag placement had no impact on heart rate data. As tag placement in terms of distance and position is critical for good heart rate data quality (Brijs et al. 2018), these results confirm that the surgical implantation was adequate despite the fish being considerably larger than other experimental fish used for reference of tag placement.

Conclusion and future work

Activity measurements can be used to track high-impact stress events but may not be suitable to discern possibly nuanced reactions to lower impacts. Heart rate did not function as a credible proxy to predict stress in this study. The results thus underline challenges observed in previous work around the use of heart rate as stress indicator. While heart rate was measured reliably, e.g., in showing circadian rhythms, the translation of its meaning into a proxy for stress needs further work. Current technology to measure stroke volume in free swimming fish requires very extensive (60–90 min) surgery (Brijs et al. 2019) or is limited to laboratory settings in, e.g., swim tunnels (Brodeur et al. 1999, 2001). The emergence of miniaturised biosensors, however, may offer the opportunity to delve deeper into such questions. Recently, an implant measuring photoplethysmograms (PPG) for estimation of arterial blood oxygen saturation (i.e., pulse oximetry) was introduced for Atlantic salmon (Svendsen et al. 2021a; Svendsen et al. 2023). This technique relies on the change in perfusion with the cardiac cycle and signal morphology is in mammals related to, e.g., blood pressure and cardiac output (Park et al. 2022). If similar considerations can be made using PPGs from Atlantic salmon, online cardiac output evaluations may be possible in the future.

In conclusion, this study emphasises that implantation of biologgers provides useful data, but may impact the welfare of the tagged fish and thus compromise some of the collected data. The extent of the impacts and thus the quality and validity of the collected data may depend on many factors including fish condition at the start of the experiment, recovery times after surgery, experimental duration, and potential stressors encountered. Our results highlight the importance of a thorough assessment of potential impacts of tagging on the observed fish. Future studies may benefit from even more detailed assessments, including, e.g., an extended necropsy as well as additional inflammation markers.

While tag-based monitoring of stress is not without its difficulties, studies such as this provide a wealth of information on salmon behaviour and physiology (Brijs et al. 2021; Macaulay et al. 2021). Such knowledge is needed to develop and validate novel tools that can support less invasive monitoring technologies based on, e.g., computer vision or acoustic observation (Barreto et al. 2022). This will bring the industry another step closer to its goal of precision fish farming where a direct feedback from non-invasive monitoring can guide operations to improve health and welfare on farms (Føre et al. 2018a).